Title: Method of managing a herd comprising a plurality of animals using an animal management system. Field of the invention The present invention is directed at a method of providing decision support data for making fertility management action decisions with respect to at least one animal from a herd comprising of a plurality of animals by training an autonomous learning data processing model and to a computer program product. Background In order to maintain or extend production capacity of a herd of animals in a farm, a supporting reproduction program is important in order to maintain the average age sufficiently young. Regardless of whether the herd is kept for milk or meat production, the regular supply of young animals is important. Furthermore, independent of the need for young animals in the herd, a reproduction program is also important in order to uphold the milk yield of an animal. Mammals only produce milk after they have given birth. For example, dairy cows must give birth to one calf per year in order to continue producing milk. Similarly, goats and sheep likewise need to give birth in order to start their lactation cycle. Once started, the lactation cycle can be maintained for a prolonged duration by frequent (e.g. daily) milking. However, at some point, e.g. for a cow within three to six months after giving birth, insemination is desired to maintain the animal in production. For this, the farmer may choose between natural insemination using a bull, or artificial insemination. The latter has advantages in terms of control over the reproduction program. However, for a herd of many animals, the planning can be challenging for a farmer. Current reproduction management strategies aim at supporting farmers in their decision on which cows to inseminate. Existing decision support systems commonly are simple rule-based methods. For example, such systems may advise the farmer to perform insemination if a cow is past a first number of days in milk, and to advise against insemination if a cow is beyond a second number of days in milk. Some generally available decision support systems aim at identifying superior reproduction strategies to be used on a farm, and are able to provide a general advice on a certain strategy (e.g. use a hormone protocol combined with estrous detection or start inseminating at day 80 after parturition). However, farm efficiency is based on much more complex data such as optimizing milk yield, resource usage, and balancing workload. None of the systems or methods available provides for a total farming solution that enables to support a farmer in fertility management related decisions. For example, none of these systems and methods focusses on the decision that the farmers needs to make when he observes a cow in heat: whether to inseminate the cow, whether to wait for the next cycle, or whether not to inseminate the cow at all. Summary of the invention It is an object of the present invention to overcome the disadvantages of the prior art, and to provide a means for supporting a farmer or user responsible for the management of a herd, in making fertility management related decisions, such as insemination decisions. To this end, there is provided herewith a method of managing a herd comprising a plurality of animals using an animal management system, by providing decision support data for making fertility management action decisions with respect to at least one animal from the herd; the animal management system comprising an animal management server and one or more receivers for receiving a wireless data signal from a plurality of electronic sensor tags, wherein each electronic sensor tag is associated with one of the animals of the herd, and wherein each electronic sensor tag includes at least one sensor for providing a sensor signal indicative of an animal related parameter, for including the sensor signal in the wireless data signal, wherein the method comprises: obtaining, from the animal management system, one or more decision trigger signals, wherein the decision trigger signals at least include a heat probability signal indicative of an above average probability that the at least one animal is in heat; obtaining, from the animal management system, fertility management history data for the at least one animal, the fertility management history data being indicative of at least one earlier taken fertility management action decision or indicative of a birth; obtaining a herd status indicative of a momentary status of the herd, wherein the herd status is determined based on animal management data obtained from the animal management system; wherein the method further comprises a step of: determining, by the animal management system, a fertility management action decision with respect to the at least one animal based on the one or more decision trigger signals, the fertility management history data and the herd status, wherein the action decision is determined using a trained autonomous learning data processing model. The term ‘fertility management action decision’ may be defined as any decision to take action or decisive action taken to manage the fertility status of an animal. The term ‘fertility status’ can be defined as the physiological status of the animal with respect to its fertility, such as whether or not the cow is pregnant, or the development stages of fertility. The term ‘fertility management action decision’ may thus relate to at least one of: a desired moment of insemination, a pregnancy check, a group change for the at least one animal, a decision to dry off, a decision to do not breed, or a decision to take no action with respect to the animal. According to a further aspect, the invention relates to a method of training an autonomous learning data processing model, for use in a method according to any one or more of the preceding claims, for providing decision support data for making fertility management action decisions with respect to at least one animal from a herd comprising a plurality of animals, wherein the fertility management action decisions relate to at least one of: a desired moment of insemination, a pregnancy check, a group change for the at least one animal, a feed change, a decision to dry off, a decision to do not breed, or a decision to take no action with respect to the animal; wherein the autonomous learning data processing model is a reinforcement learning data processing model, and wherein the training method comprises the steps of: obtaining as first training data, from an animal management system or a user, one or more decision trigger signals, wherein the decision trigger signals at least include a heat probability signal indicative of an above average probability that the at least one animal is in heat; obtaining as second training data, from a user or from the animal management system, fertility management history data for the at least one animal, the fertility management history data being indicative of at least one earlier taken fertility management action decision or indicative of a birth; and obtaining as third training data, a herd status indicative of a momentary status of the herd, wherein the herd status is determined based on animal management data obtained from the animal management system; determining for use as fourth training data, based on the one or more animal management data, a status score for the herd status, wherein the status score represents a reward or a penalty dependent on whether or not the animal management data is indicative of a future expected achievement of one or more objectives; and training the autonomous learning data processing model to autonomously perform a step of determining a fertility management action decision with respect to the at least one animal based on future input data such as to achieve a desired status score, wherein the future input data comprises one or more future decision trigger signals, and wherein the training is performed based on at least the first, second, third and fourth training data. The training method of the present invention enables to provide an autonomous learning data processing model that allows to provide, based on input comprising decision trigger signals, decision support data for making fertility management action decisions with respect to at least one animal from a herd comprising a plurality of animals, wherein the fertility management action decisions relate to at least one of: a desired moment of insemination, a pregnancy check, a group change for the at least one animal, a feed change, a decision to dry off, a decision to do not breed, or a decision to take no action with respect to the animal. These decision trigger signal at least include a heat probability signal that may be obtained from data obtainable from animal tags. In addition, other decision trigger signals may additionally or alternatively be used as input to the autonomous learning data processing model. The training is performed in a first phase by observing the decisions taken by a farmer for a herd. This may be the same herd as the herd containing the animals for which eventually the autonomous learning data processing model will later perform the determination of the desired moment of insemination. However, the training herd may also be a different herd. The training method is performed by obtaining data from an animal management system that enables to determine a momentary herd status in order to monitor the future expected effects of the insemination decision on e.g. milk yield or available resources on the farm. The training method evaluates the herd status, and calculates, for each potential fertility management action decision, a score representing a reward or penalty of the respective action decision. An increase in milk yield may for example receive a positive reward, whereas any undesired results such as a cumulation of projected parturitions in a short time, or a shortage of certain resources (e.g. young animal pens, available personnel) will receive a penalty (i.e. negative reward). All this data, hereinabove referred to as first, second, third and fourth training data, is used as input to the training method. The autonomous learning data processing model, on the basis thereof, will be trained to decide which action decision to take. These action decisions include any one or more of the following: decide on a moment of insemination, decide to perform a pregnancy check, decide on a group change for the at least one animal, decide on a feed change (e.g. to feed the animal with concentrates or to provide standard feed), a decision to dry off, a decision to do not breed, or a decision to take no action with respect to the animal. This enables to provide a system on the basis of this autonomous learning data processing model that signal an above average heat probability, and takes action by notifying the farmer and advising him or her on which fertility management action decision is advisable – in terms of the chance of success and in terms of his planning of the herd management. The term fertility management action decision indicates the potential decisions that a farmer may take in order to manage an animals fertility and pregnancy status, such as to maintain a desired yield. The context thereof is therefore dependent on the needs of the farm. In herds of animals that are kept for products such as milk, wool, eggs, etcetera, the yield will be the amount of the product to be produced. Herds that are kept for meat production, for breeding of the animals themselves, or for other purposes, the yield may be the number of animals themselves. The term fertility management action decision thus relates to any action decisions that may be taken to keep up the production, e.g. such as to keep a cow in milk, or to produce a sufficient amount of pigs. In some embodiments, the heat probability signal is obtained from a further data processing model based on one or more sensor signals from an electronic sensor tag worn by the at least one animal of the herd. For example, in some of these embodiments, trained neural networks are able to calculate and provide a heat probability value on the basis of information obtained from animal worn tags, such as neck tags, ear tags, or tail tags. In other or further embodiments, another data processing model, e.g. a random forest model, may at its output provide a Boolean signal indicative of whether or not an animal is in heat. In yet further embodiments, the heat probability signal is determined based on animal status data obtained from the animal management system. This data may likewise be analyzed to determine a heat probability useable as decision trigger signal. In other or further embodiments, the heat probability signal is obtained as input from a user. Here, the farmer may observe an animal to be in heat and provide this data as input to the training method or autonomous learning data processing model. The decision trigger data may additionally or alternatively include other data as well. For example, in some embodiments, the one or more decision trigger signals further include one or more of a group comprising: heat history data of the at least one animal, milk yield history data of the at least one animal or the herd, age data of the at least one animal, momentary number of days that the at least one animal is in milk, pregnancy status of one or more animals of the herd. The additional decision trigger signals will improve the outcome of the data processing model. Triggers may for example relate to certain combinations of the above, including an animal being in heat. For example, animals being in heat while the milk yield of the herd is decreasing. In some embodiments, the animal management data includes one or more of: resource data or herd data, wherein the resource data comprises data indicative of available resources for management of the herd and wherein the herd data comprises data indicative of characteristics of the herd, such as milk yield or herd size. This data may be used in order to evaluate the herd status and calculate a score. As may be appreciated, an objective may be to maximize milk yield, whereas a different objective may be to maximize milk yield for a herd size manageable with a certain amount of available resources. In this last case, maximizing milk yield in fact becomes optimizing milk yield given the available resources, and the status score may be penalized if a projected herd status indicates a herd size that is too large for the amount of available resources. In some embodiments, the status score is rewarded or penalized dependent on the herd status being indicative of respectively an increase or decrease in milk yield from the herd. Basically, a higher milk yield is desirable, so this may be rewarded. A lower milk yield is not desired, and may be penalized. Therefore, in some of these embodiments, the status score is rewarded or penalized proportional to the increase or decrease in milk yield or meat production. Furthermore, in other or further embodiments, the status score is penalized if the resource data or the herd status is indicative of a resource bottleneck, such as a shortage in pens for offspring, a cumulation of simultaneous parturitions, or a shortage of labor capacity. As explained, under the expectation of a resource bottleneck, an increase of milk yield may not be desired if this means that the number of born calves may be larger than the number of available calving pens. In some embodiments, the status score is penalized if, after an insemination of the at least one animal, a pregnancy status of the at least one animal does not change. This may happen in case the moment of insemination is incorrect due to the animal being insufficiently fertile at that moment. So, the machine learning data processing model in this way will be trained to determine the moment of insemination to be optimal in terms of output. In some or further embodiments, the one or more decision trigger signals further include one or more of a group comprising: heat history data of the at least one animal, milk yield history data of the at least one animal or the herd, age data of the at least one animal, momentary number of days that the at least one animal is in milk, pregnancy status of one or more animals of the herd. These data may advantageously be used to obtain a more sophisticated and optimized decision on desired moment of insemination of the at least one animal. Furthermore, in some embodiments, the step of determining a desired moment of insemination comprises: comparing, for each of a plurality of future moments after receipt of the one or more decision trigger signals, an effect on the status score at a time after the respective moment: - in response to an insemination decision at the respective moment as a first alternative, and/or - a decision to postpone insemination at the respective moment as a second alternative. The method may thereby review the impact of various alternatives on the status score, and for example determine which timing decision relating to the timing of insemination which yields the most favorable impact on the status score. For example, the machine learning data processing model may thereby improve to choose those timing decisions providing the steepest change towards a maximum status score. In some embodiments, the fertility management history data includes data indicative of a pregnancy of the at least one animal, and the autonomous learning data processing model is trained to, upon receipt during the pregnancy of a decision triggers signal indicative of an above average probability that the at least one animal is in heat, determine the fertility management action decision to be at least one of: do nothing, perform a pregnancy check or perform an insemination. In some embodiments, the method is performed during a first training phase and a second training phase, and wherein during the second training phase the method further comprises the autonomous learning data processing model to perform a step of determining an advisable moment of insemination of the at least one animal based on the first training data by evaluating an expected herd status and an associated status score. During this second training phase, the machine learning data processing model that has already been trained during the first training phase, may further be trained by applying it as a decision support system wherein a farmer eventually takes the decision on whether or not to follow the suggested decision with respect to the moment of insemination. This second phase may be preceding a third training phase wherein the machine learning data processing model decides autonomously when to inseminate the at least one animal. In accordance with a second aspect, there is provided a method providing decision support data for making fertility management action decisions with respect to at least one animal from a herd comprising a plurality of animals by using an autonomous learning data processing model which is trained using a method according to any one or more of the preceding claims, wherein the fertility management action decisions include at least one of: determining a desired moment of insemination, deciding to perform a pregnancy check, deciding on a group change for the at least one animal, deciding on a feed change, a decision to dry off, a decision to do not breed, or a decision to take no action with respect to the animal; wherein the method comprises: obtaining, from an animal management system, one or more decision trigger signals, wherein the decision trigger signals at least include a heat probability signal indicative of an above average probability that the at least one animal is in heat; obtaining a herd status indicative of a momentary status of the herd, wherein the herd status is determined based on animal management data obtained from the animal management system; determining, by the autonomous learning data processing model, the decision support data for making a fertility management action decision for the at least one animal based on the one or more decision trigger signals; monitoring, after the fertility management action decision has been carried out, the herd status; and determine, based on the one or more animal management data, a status score for the herd status, wherein the status score represents a reward or a penalty dependent on whether or not the animal management data is indicative of a future expected achievement of one or more objectives. In accordance with a third aspect, there is provided an autonomous learning data processing model, wherein the model is a reinforcement learning data processing model, and wherein the model is trained using a method in accordance with the first aspect, for enabling the steps of: obtaining, from an animal management system, one or more decision trigger signals, wherein the decision trigger signals at least include a heat probability signal indicative of an above average probability that the at least one animal is in heat; obtaining a herd status indicative of a momentary status of the herd, wherein the herd status is determined based on animal management data obtained from the animal management system; determining, by the autonomous learning data processing model and based on the one or more decision trigger signals, decision support data for making a fertility management action decision with respect to the at least one animal from the herd, wherein the fertility management action decisions relate to at least one of: a desired moment of insemination, a pregnancy check, a group change for the at least one animal, a feed change, a decision to dry off, a decision to do not breed, or a decision to take no action with respect to the animal; monitoring, after the fertility management action decision has been carried out, the herd status; and determine, based on the one or more animal management data, a momentary status score for the herd status, wherein the status score represents reward or penalty dependent on whether or not the animal management data is indicative of a future expected achievement of one or more objectives. In accordance with a fourth aspect, there is provided a computer program product for use with an animal management system comprising a controller and a memory, wherein the computer program product comprises instructions which, when loaded into the memory and executed by the controller, cause the animal management system to carry out a method of training an autonomous learning data processing model in accordance with the first aspect. Brief description of the drawings The invention will further be elucidated by description of some specific embodiments thereof, making reference to the attached drawings. The detailed description provides examples of possible implementations of the invention, but is not to be regarded as describing the only embodiments falling under the scope. The scope of the invention is defined in the claims, and the description is to be regarded as illustrative without being restrictive on the invention. In the drawings: Figure 1 schematically illustrates a system wherein the present invention may be applied; Figure 2 schematically illustrates a machine learning data processing model and (at least some phases of) a method of training thereof in accordance with an embodiment; Figure 3 schematically illustrates a neural network that may be used in a method of training thereof in accordance with an embodiment; Figure 4 schematically illustrates a machine learning data processing model in (at least some phases of) a method of training thereof by reinforced learning, in accordance with an embodiment; Figure 5 schematically illustrates a method in accordance with some embodiments of the invention; Figure 6 schematically illustrates a label for use in a system in accordance with an embodiment; Figure 7 shows a graph indicating how the method is applied in order to support decision making in relation to fertility management. Detailed description Figure 1 schematically illustrates an animal management system 3 for managing a herd of cows 1. The animal management system 3 consists of an animal management server 85 and a database 88. The database 88 maybe a local or remotely accessible database or data storage facility. The animal management system is based on the use of animal worn labels 21. Each of the cows 1 wears a label 21 including one or more sensors from monitoring animal related data such as health data of each of the cows, a position of each of the cows 1, behavior of the cows 1, their local proximity to other facilities, and the like. The labels 21 transmit a wireless signal 80 to the base stations 81 of the system 3. The base stations 81, which serve to provide receivers to the system 3, receive the wireless signals and provide these to the management animal server 85 where the signals are analyzed in order to retrieve the desired data. The system 3 in figure 1 is schematically illustrated to indicate some of the features of such systems, and the skilled person is aware of other animal management systems or similar animal management systems wherein the method of the present invention may be applied. A label 21 is schematically illustrated in figure 6. The label 21 for example may comprise a processor 22 that may be operatively connected with a memory 23, a communication unit 24 connected to an antenna 25, and one or more sensors 26. The sensors 26 present in the label 21 may be of any type suitable to measure or sense body parameters in order to provide sensor signals indicative of any of: heat stress, social hierarchy, calving prediction from tags (ear/neck/tail/leg/bolus), milk yield from milk meter, milk composition (fat, protein, somatic cell score, mastitis, metabolic status, stress level) from milk analysis, lameness, low condition, feed intake from computer vision. At least some of the sensors 26 in label 21 provide sensor signals that are indicative of body parameters that enable to detect whether or not the animal 1 may be in heat, i.e. that there is an above average probability that the animal 1 is in heat. This information will be used by the system 3, as explained below, to obtain or provide a decision trigger signal for performing a method in accordance with the invention. The decision trigger signal thus likewise is at least is indicative of an above average probability of the animal 1 being in heat. The decision trigger signal may for example be generated by the system 3 based on the animal management data available in data storage 88 or obtained directly form the sensors 26. The sensors 26, however, may provide additional sensor signals indicative of body parameters that may be used as auxiliary data for performing the method of the invention. Back to figure 1, in the illustrated situation 10 a number of cows 1 are present in the field. Their labels 21 transmit wireless signal to the base station 81, indicating their position for example. Based on the data received via base station 81, it may be determined for example which of the animals 1 has an above average probability of being in heat. For example, if one of the cows 1 is more frequently near a bull, or exhibits increased activity or an increased heart rate, the cow 1 may be in heat. Any of these indicators may be used to determine the above average probability of the cow 1 being in heat. The system 3 enables to determine other animal management data that may be used as decision trigger signals. For example, a milking arrangement 28 that is located in barn 5 may transmit wireless signals 80 to a based station 81 inside the barn 5, which indicate the amount of milk measured by milking meter 7 of the arrangement during milking of the cow 1’. The milking arrangement 28 consists of a milking container 9 that receives milk obtained from cow 1’. Cow 1’ wears a milking unit 8 with teat cups which draw milk from the cow 1’ that is received by milk meter 7, and thereafter provided to the container 9. Milk meter 7 measures the amount of milk produced by the cow 1’ and transmits the wireless signal 80 to base station 81. Because cow 1’ also wears a label 21 (not visible in figure 1), another wireless signal 80 indicates the current location of cow 1’ in the barn 5. For example this information may be applied to associate the amount of milk measured by milk meter 7 with the identified cow 1’ near the milking arrangement 28. This information can be stored as animal management data in data storage 88 for animal management. Other information that can be obtained from the system 3 in figure 1, for example may relate to the use status of calving pens. On the left side of barn 5 in figure 1, a cow 1 with label 21 is illustrated that just has given birth to a calf 2. Calf 2 in the meantime has been provided with a label 21’ that is to be worn by calves having a proper size. The presence of cow 1 in the designated area of the barn 5 for calving, or its proximity to the calf 2, may indicate that the cow 1 has just given birth to the calf 2. This information can be used to trigger a period that indicates the prohibition for the cow to be inseminated. For example, the cow 1 may be ignored for a while in order to recover from the birth of calf 2. Other facilities may be present in the environment 10 that are manageable by the system 3. For example the presence of cows 1 near a feeding arrangement or their social behavior in the herd can be monitored, which may be indicative of an increased heat probability. By receiving all this data from the herd, and optionally by further receiving information via other input means from a farmer managing the herd, the animal management server 85 that registers all this data in the data storage 88 is able to produce momentary data and history data of the herd which can be used in the method for training an autonomous learning data processing model 40 to provide decision support data of fertility management related action decisions, as will be explained below. To this end, in accordance with the invention there is proposed a multiphase training method for training the autonomous learning data processing model. In the examples below, the autonomous learning data processing model includes a decision model 45 followed by a neural network consisting of three hidden layers 57, 58 and 59. As may be appreciated, any other machine learning data processing model suitable for being used in an reinforced learning environment may be applied to perform the steps of the present invention. The invention is not limited to a particular neural network or other type of machine learning data processing model, and neither to a specific number of hidden layers of the neural network. In accordance with the present invention, the system 3 including the autonomous learning data processing model 40, may be used to provide decision support data of fertility management related action decisions. For example, in a preferred embodiment the system 3 with autonomous learning data processing model 40 may be used to determine a preferred or optimal moment of insemination. However, in addition to this or in alternative implementations, the decision support data of fertility management related action decisions may relate to any of: a post calving check, a hormone treatment, a decision not to breed a certain animal (e.g. due to a health condition or age), a culling decision, determining a desired moment of insemination, deciding to perform a pregnancy check, a decision to perform an abortion check (i.e. a pregnancy check to confirm the potential loss of a calve), deciding on a group change for the at least one animal, deciding on a feed change (e.g. to feed concentrates or to feed standard feed), a decision to dry off, a decision to do not breed , or a decision to take no action with respect to the animal), such as to achieve a desired goals or objective (e.g. maximize milk yield, optimize costs versus yield, preventing resource shortages or capacity issues. The invention as claimed could relate to a system that advises on each of these actions decisions, or to only some of these action decisions or even only one of these. In the below examples, an embodiment is described of a system that advises on a desired moment of insemination. The fertility management related action decisions in these embodiments merely include deciding on whether or not to inseminate or to determining the best moment to do so. In later described embodiments, a similar system is described that advises on any action decisions mentioned above; thus also to perform pregnancy tests or decide on a group change of the animal. Below, a number of these fertility management action decisions and their relevancy is briefly explained: · Post-calving check: Several checks in the first few days (up to 28 days) after calving to confirm that the reproductive system of the cow recovered from calving. This means check that the placenta has been completely expelled, the ovaries and uterus are back in a non-pregnant status. This requires manual labor. · (Hormone) treatment: Sometimes it may be needed to treat cows with hormones because they show only very weak (or no) heat signs. There are costs associated with this. The model could assist in determining when this is the best option. · Insemination: A decision to inseminate has been explained hereinbefore, and may be desired in order to maintain a sufficient milk production or in order to breed a desired animal for meat production. · Do-not-breed decision: At a certain stage in lactation it is more profitable to stop inseminating a cow that is not pregnant yet and replace her with a new animal. This is a decision that in the current systems is usually based on number of days in lactation, but this is a decision that should in fact be based on expected future revenue of this cow, costs of insemination, availability of replacement animals and so on. · Culling a do-not-breed animal: Once a cow is marked as ‘Do Not Breed’ you have to decide when to cull her (sell her to the slaughterhouse). This depends on her future expected revenue. This is very different for a beef and a dairy cow. A beef cow can gain weight and that means a higher slaughter weight. A dairy cow still produces milk that you can sell. The optimal time of culling is a decision that can benefit from a support system. · Pregnancy check: When to do a pregnancy check. This always require manual labor and therefore can be optimized as well. · Abortion check: Actually this also is a pregnancy check, but can be seen more as a confirmation of whether the cow is still pregnant or not anymore. The action is the same as with the pregnancy check, but the use of its outcome may be different and therefore it is mentioned separately here. · Dry off (for dairy cows only): When to stop collecting the milk of this cow. Cows need some time to prepare for the next lactation. This depends on their expected calving date, current milk production, etc. · Move to close up pen: A pen were cows are moved to about 3 weeks before calving, usually with a special type of diet. This diet is usually more expensive. It is not desired that cows spend too much time in the close up pen, but if the time spent in this pen is too short, the animal can have difficulties in the next lactation. · Move to maternity pen: just before calving, cows are moved to maternity pens. They have more comfort in these pens (more space, straw bedding, extra human care, etc), but these pens are limited in capacity. Figure 2 schematically illustrates a training method for training a machine learning data processing model 40. The machine learning data processing model 40 comprises an input side 43 and an output side 48, and the core of the data processing model 40 is formed by decision model 45. Decision model 45 may for example be a mathematical algorithm, such as a mathematic regression model, but may as well be formed by for example a neural network or a random forest algorithm. Controller 50 controls the training of the decision model 45. On the output side 48, the network may have one possible output node per option (inseminate / do not inseminate). The output node may provide a score representing the expected future reward or penalty (i.e. negative reward) when choosing this option. Every time step, the actual rewards are accumulated and used to update the score of previous steps. In accordance with the present invention, using for example a system 3 as is illustrated in figure 1, a number of input parameters may be determined for use as training data for training the decision model 45. For example as first training data, a heat probability signal may be determined using the labels 21 of cows 1. Other input data that may be used as first training data may for example include decision trigger signals such as heat history data that may be obtained from database 88 of the central animal management server 85 milk yield history data of the at least one animal or the herd of a present milk yield obtainable from server 85, database 88 or directly from the milking system 28, age data of the at least one animal which also may be obtained from the server 85 the momentary number of days that the animal is in milk which may be obtained directly from the animal worn labels 21 or the server 85 or the milking system 28, or a pregnancy status of one or more animals in the herd which may be obtained from the server 85, the labels 21 or a different source. Second training data may also be obtained from the system 3 of figure 1 for use as input data 43 to the decision model 45. This second data relates to fertility management history data. The fertility management history data relates for example to data on earlier fertility management action decisions, and may for example include insemination data for the animal, or data on a performed pregnancy test, an earlier feed change. The fertility management history data may also include data on the birth of offspring, such as calving data for cows. Insemination data may be indicative of the decision to inseminate the animal or a moment of insemination. For example, once the moment of insemination has been planned in the system 85, this data may be obtained there from. Furthermore, once the insemination has taken place this may likewise be notified to the server 85 or be automatically detected for example by registration of a proximity of a cow 1 wearing a label 21 to a location where insemination takes place. Any such data may be used as second training data to be provided as input data 43 to the decision model 45. After the fertility management history data has been obtained as second training data, third training data may be obtained from the system 3 which is indicative of a momentary status of the herd. The momentary status of the herd may for example indicate the herd size, the milk yield, a number of pregnant animals in the herd, resources data indicative of available resources from management of the herd or data such as health data of the herd. In a first phase of training method, the above first training data, second training data and third training data may be used in order to train a neural network, such as illustrated in figure 3 to be discussed later, in taking fertility management action decisions, for example on insemination, based these input parameters. For example, turning to figure 2, the first training data and third training data may be used as input data elements 43-1 through 43-m. The second training data may be proposed as output 48, and based on this, the neurons 66 and 67 illustrated in figure 3 can be trained in order to propose the decision 48 as output on the input parameters 43-1 through 43-m. Furthermore, the controller 50 may calculate, based on the first training data and the third training data, and in particular the animal management data, a status score for the herd status. The status score indicates whether the decision 48 in the output of the decision model 45 corresponds with a favorable output on the herd status in view of the requirements set. For example, suppose the system would take an insemination decision and the insemination decision taken thereafter led to a decrease in milk yield of the herd, while sufficient resources are available for management, the result of the insemination decision is suboptimal and therefore the future reward is lowered. However, if the insemination has led to an increase in milk yield, the future reward is increased. For calculating the future reward of the individual options, the controller 50 received the input parameters 43-1 through 43-m (as indicated by arrow 52), and calculates the output values 48 (as indicated by arrow 51). The option with the highest estimated future reward is taken as the insemination decision. The expected future reward calculated by controller 50 can be made more precise by adding more information, represented by neurons 66 and 67, to the decision model 45. During this first training phase, the machine learning data processing model 40 will follow the farmers decisions on insemination, and the policy being determined by the decision model 45 will be improved based on the outcomes and the calculated status score by the controller 50, by accordingly adapting the specific parameters of the neurons 66 and 67. This may be referred to as off-policy learning during the method of training. In a second phase of the training method, the policy that is advised by the machine learning data processing model 40 may then be used to provide recommendations to a farmer on fertility management action decisions. These could include action decisions such as deciding on a moment of insemination, performing a pregnancy check, deciding on a group change for the at least one animal, deciding on a feed change, a decision to dry off, a decision to do not breed, or a decision to take no action with respect to the. Technically, the training method in the first phase and in the second phase is largely similar, in the sense that also in the second phase the actual decision is taken by the farmer. The second training data being the fertility management history data for the at least one animal is also during the second training phase provided as input by the farmer to the machine learning data processing model 40. However, by suggesting the fertility management action decision to the farmer, the farmer may start to follow the decision and thereby the herd status and training of the data processing model 40 will adapt towards the situation where the fertility management action decision will be taken automatically. Referring to figure 3, a neural network having three layers is schematically illustrated. In figure 3, the input positions 43-1 to 43-n are illustrated as input neurons 43 in an input layer 55 of the neural network. The first hidden layer 57 consists of a plurality of neurons 66, the number of neurons 66 in the first hidden layer 57 may be chosen as desired or deemed necessary. Each neuron 66 in principle receives each of the input values x1 through xn from the input neurons 43-1 to 43-n. At the input of each neuron 66, the received values from neurons 43-1 to 43-n may be multiplied with a weighing value which is in figure 3 schematically indicated as W01,i,j. Herein, the value i refers to the neurons 43-i (e.g. 43-2 where i=2) in the input layer 55, and the value j refers to the respective neurons 66 in the first hidden layer 57. Each of the neurons 66 in the first layer 57 may likewise apply an additional multiplier uj. Furthermore, optionally each of the neurons 66 may be assigned an activation function that may implement the non-linear character (where applicable) of the data processing model in some embodiments. To each of the neurons 66 in the first hidden layer 57, a bias value 61 may further be applied, which is a constant value that is added to the output of the neuron 66. The output value of each neuron 66 of the first hidden layer 57 is thereafter provided to each of the neurons 67 in the second hidden layer 58. Again, these output values are for each pair of sending neurons 66 from layer 57 and receiving neurons 67 from layer 58, multiplied with a weighing value w12,j,k. The value j herein again refers to the respective sending neuron 66 in the first hidden layer 57, and the value k herein refers to the receiving neuron 67 in the second hidden layer 58. In the same way as above, to each of the neurons 67 in the second hidden layer 58 an activation function may be applied, and the neurons 67 may be biased with a value 62. In the example of figure 3, the output values of the neurons 67 in the second hidden layer 58 are passed on to the summation layer 59. This layer on one hand contains a numerator neuron 69 and on the other hand a denominator neuron 70. The summation layer 59 in fact forms a third hidden layer of the neural network. The weights w ^23,k,l indicatively indicate which weight the output value of which neuron 67 in the second hidden layer 58 receives for passing on this output value to the summation layer 59. Again, herein the value k indicates the respective neuron 67 in the second hidden layer 58, and the value l in the example shown in figure 3 may be l=1 or l=2 for indicating the respective receiving neuron 69 or 70 in layer 59. By dividing the value from the numerator neuron 69 by the denominator neuron 70, the output value 48 at the output 60 is obtained. The output values 48 indicate the expected future reward of each option of the fertility management action decision. The option with the highest feature reward is chosen. The above examples are not limitative and other possibilities may likewise be considered in this respect. Furthermore, the decision model 45 may be a different type of machine learning data processing model that is suitable for being applied in a reinforced learning method. For example, without being exclusive or limiting, the machine learning data processing model may be designed in accordance with any of the following algorithms: linear or non-linear regression, logistic regression, linear or non-linear discriminant analysis, decision trees, naive Bayes, K-nearest Neighbors, learning vector quantization, support vector machine, random forest, or deep neural network. In figure 4 a third training phase or operational phase of the method of the present invention is schematically illustrated. In this phase, the machine learning data processing model 40 receives inputs 43 in order to calculate an output 48 using the decision model 45. The output 48 is indicative of action 18, which may relate to a decision in favor of insemination or a decision not in favor of insemination of the at least animal. Alternatively, the output 48 is indicative of action 18 may relate to a decision to perform a pregnancy test, for example after an animal that was considered to be pregnant shows signs of being in heat again. The action 18 may also be indicative of any of the other action decisions mentioned herein before. In the third phase of the method of the present invention, the machine learning data processing model 40 autonomously performs the fertility management action decisions, and therefore the result 48 triggers the action 18 to be performed on the animal or the herd: for example an insemination decision may be taken. The environment is schematically illustrated by reference numeral 10, which may be an environment as illustrated in figure 1 consisting of a herd cows 1 and some facilities such as a barn 5, milking arrangement 28, feed distributor, etc. When the action is followed, the effect thereon on the environment 10 will be monitored and will provide the new input parameters for machine learning data processing model 40. This is indicated by the arrows 12, 13 and 14 which respectively relates to the first training data 12 comprising the decision trigger signals, the second training data 13 comprising the insemination data to provide the machine learning data processing model 40 with information on those animals that have been inseminated or will be inseminated, and the fourth training data 14 indicative of the herd status. The herd status may indicate for example the milk yield or the forecasted milk yield from the herd, the current or forecasted herd size, the availability of resources, the surplus or shortage of resources, optionally including the availability of resources over time such as the history or forecast thereof, the number of projected parturitions and the forecast thereof overtime e.g. to identify a cumulation of simultaneous parturitions that may give rise to a shortage in resources, etc. The first, second and third training data 12, 13 and 14 are to be provided at the input of the decision model 45. The decision model 45 will calculate a next action 18 to be performed and indicate this at its output 48. However, based on the present herd status 14, the controller 50 will also calculate a status score 15 in order to implement the reinforced learning component as feedback on the previous action taken. The status score for the herd status is an evaluation score that indicates how well or badly the action 18 contributes to the objective. For example, the score may be rewarded for any developments in the herd status that correspond with the objectives within the requirements said. In other words, the status score is rewarded for an increase in milk yield (if this is the objective) or an increase in herd size (if this is the objective), as long as this new result (i.e. the increased milk yield or herd size) is manageable with the present and forecasted resources for managing the herd. For example, if the milk yield increases so much that the present milk storage tank of the farm becomes too small, then the penalty for overshooting the storage capacity will be larger or more important than the reward for increasing the milk yield, so that the net score will be a penalty for this result. Thus penalties will be provided for any undesired result in view of the objective, for example a decrease in milk yield or a decrease in herd size that is counter to the objective set, or a projected shortage of resources. Rewards will be provided for meeting the objectives. In the above example, if the milk yield is increased such as optimally fill the milk storage container (e.g. for 98%) without overshooting before being emptied by a milk truck, and within the regular planned normal availability of staff on the farm (without requiring extra personnel), the objective may be well met and thereby may yield a high score. The magnitude of the reward or penalty may thus be based on how good or bad the developments in the herd status match the objectives and or may be dependent on the seriousness of any mismatch with the requirements. The status score 15 will be used to adapt the values and algorithms in the neurons 66 and 67 of the hidden layers 57 and 58. Figure 5 schematically illustrates a method 30 in accordance with an embodiment of the present invention. The method 30 starts by obtaining, from the environment 10, as first training data 12, one or more design triggers which at least include a heat probability signal that indicates whether or not one or more of the animals 1 of the herd are in heat. The heat probability signal may be taken from one or more animal worn labels 21 that are worn by the animals 1. Furthermore, other information may be used as decision trigger signals, such as whether or not a cow 1 has recently calved, or whether or not there are any specific reasons why such an animal 1 must be treated differently. For example, based on the wireless signals 80 from labels 21 and 21’ of a cow 1 and its calf 2 in a calving pen of a barn 5, which are received via base station 81 in the barn 5, the animal management server 85 may determine for this particular cow 1 that any signals that are indicative of heat are to be ignored for this cow because the cow 1 needs to recover from calving. Other information that may be taken from animal management server 85 or other systems of the farm, such as the milking system 28, may for example include heat history data or one or more animals indicative of when an animal has been in heat, milk yield history data of one or more animals or of the complete herd, which is indicative (for an animal) how its milk production has evolved overtime, age data for the animal, or a momentary number of days that the animal is in milk (produces sufficient milk), or a pregnancy status of the animal. In step 32 as second training data, insemination data is obtained from the system 85 or the sensors in the environment 10. For example, a projected insemination planning for one or more animals may be obtained which indicates for which animals it is already be decided when they will be inseminated. Furthermore, in step 34, third training data may be determined. The third training data may be indicative of a herd status of a momentary status of the herd. The herd status may for example indicate the present or projected milk yield, or present of projected herd size, as discussed herein before. Other animal management data may likewise be obtained, such as resource data for the farm, indicative of a number of calving pens, a number of present labor resources, the availability of milking systems 28, the size of the barn 5 and its capacity in terms of a number of animals that it may hold, or the evolution or forecast of any of these parameters. In particular, the second training data may include any fertility management history data, i.e. data that relates for example to earlier fertility management action decisions, and may for example include the abovementioned insemination data for the animal, data on a performed pregnancy test, or an earlier feed change. Such fertility management history data may also include data on the birth of offspring, such as calving data for cows. In step 36, as fourth training data, a status score for the herd status may be determined which is an important parameter because it implements the reinforced learning aspect of the training method. The status score represents a reward or penalty for a future expected herd status and is calculated as a criterium before, by rewarding or analyzing the score dependent on whether or not the animal management data is indicative of and achievement of one or more of the objectives or indicates a development counter thereto. The status score will be used by the autonomous learning data processing model to adapt its internal parameter in order to provide a better matching decision the output. In step 38, the first, second, third and fourth data will be used by the autonomous learning data processing model 40 in order to determine the desired moment of insemination of the at least one animal. This will enable the autonomous learning data processing model to perform a step of determining the desired moment of insemination based on future input data (i.e. input data to be received at a later moment) in order to achieve a desired status score. This future input data may comprise one or more future decision trigger signals and training is performed based on the first, second, third and fourth training data. Thus, by training the autonomous learning data processing model in this manner, the model will be enabled to process input data that will be received in the future in order to provide a fertility management action decision (i.e. any one of determining a desired moment of insemination, deciding to perform a pregnancy check, deciding on a group change for the at least one animal, deciding on a feed change (e.g. to feed concentrates or to feed standard feed), a decision to dry off, a decision to do not breed, or a decision to take no action with respect to the animal), such as to achieve a desired goals or objective (e.g. maximize milk yield, optimize costs versus yield, preventing resource shortages or capacity issues). In another embodiment of the invention, the system 3 is enabled to advise on any of the abovementioned fertility management action decisions. To this end, the system 3 not only may receive insemination data, but any fertility management history data. For example, such fertility management history data may include data on any earlier action decisions that were taken, e.g. whether or not a pregnancy test was perform and the result thereof, whether the animal was placed in a different group of the herd, whether any dietary changes were made such as the adding of concentrates to the animal’s feed, when or whether the animal was inseminated or when the animal has given birth to offspring. For example, figure 7 shows a graph indicating the milk yield (upper part) of a particular cow and the occurrence of heat detection triggers (lower part of the graph). The vertical axis of the upper graph shows the daily yield, which for example may be the milk output or the economic yield (e.g. daily profit expressed in relevant currency). The horizontal axis shows the expiring of time, expressed in days. The lower graph shows the presence of a decision trigger signal, i.e. a heat detection signal for the respective animal. The letters I, P, G, and c are indicative of insemination, pregnancy test, group change and calving. In the example provided, the cow involved has been inseminated successfully, but during pregnancy the cow again is in heat. Upon detecting heat during pregnancy, the system 3 may take any of the following actions. As a first option, the system may perform a pregnancy test to ensure that the cow is still pregnant (it may have lost the calve) and dependent on the outcome the cow may again be inseminated. Alternatively, the system may immediately decide to inseminate, without a pregnancy test. Another possible decision is to take no action at all at that moment. Figure 7 shows the first alternative (pregnancy test followed by insemination). A specific embodiment of the invention relates to a method of providing decision support data for determining a moment of insemination of at least one animal from a herd, wherein the herd comprises a plurality of animals. The method in accordance with this embodiment uses an autonomous learning data processing model that is trained as described herein before. The method comprises obtaining, from an animal management system, one or more decision trigger signals, wherein the decision trigger signals at least include a heat probability signal indicative of an above average probability that the at least one animal is in heat. For example, the trigger may be a signal from a heat detection system integrated in an animal tag, or alternatively a heat probability data obtained from an animal management server. The method further includes obtaining a herd status indicative of a momentary status of the herd, wherein the herd status is determined based on animal management data obtained from the animal management system. The method further includes determining, by the autonomous learning data processing model, a desired moment of insemination of the at least one animal based on the one or more decision trigger signals. After the desired moment of insemination and when the insemination has been carried out, the method includes the monitoring of the herd status in order to determine, based on the one or more animal management data, a momentary status score for the herd status. The status score represents a reward or a penalty dependent on whether or not the animal management data is indicative of a future expected achievement of one or more objectives. The above description of some embodiments of the invention is not to be interpreted as limiting on the invention. In particular, the skilled person will appreciate that the calculated score is just an example of how to reinforce the system’s learning abilities. Scores may be calculated in many different manners, that may all lead to reinforcement of a desired result or discouraging of undesired results. For example, another criterium to include in score calculation may be to make the reward or penalty dependent on how much the result is improved or deteriorated compared to a preceding result or over a number of steps. This may lead to the steepest improvement at any time towards the optimal result. The present invention has been described in terms of some specific embodiments thereof. It will be appreciated that the embodiments shown in the drawings and described herein are intended for illustrated purposes only and are not by any manner or means intended to be restrictive on the invention. It will be appreciated that the embodiments shown in the drawings and described herein are intended for illustrated purposes only and are not by any manner or means intended to be restrictive on the invention. It is believed that the operation and construction of the present invention will be apparent from the foregoing description and drawings appended thereto. The context of the invention discussed here is merely restricted by the essence and scope of the appended claims. It will be clear to the skilled person that the invention is not limited to any embodiment herein described and that modifications are possible which should be considered within the scope of the appended claims. Also kinematic inversions are considered inherently disclosed and to be within the scope of the invention. In the claims, any reference signs shall not be construed as limiting the claim. The term 'comprising' and ‘including’ when used in this description or the appended claims should not be construed in an exclusive or exhaustive sense but rather in an inclusive sense. Thus the expression ‘comprising’ as used herein does not exclude the presence of other elements or steps in addition to those listed in any claim. Furthermore, the words ‘a’ and ‘an’ shall not be construed as limited to ‘only one’, but instead are used to mean ‘at least one’, and do not exclude a plurality. Features that are not specifically or explicitly described or claimed may be additionally included in the structure of the invention within its scope. Expressions such as: "means for ...” should be read as: "component configured for ..." or "member constructed to ..." and should be construed to include equivalents for the structures disclosed. The use of expressions like: "critical", "preferred", "especially preferred" etc. is not intended to limit the invention. Additions, deletions, and modifications within the purview of the skilled person may generally be made without departing from the spirit and scope of the invention, as is determined by the claims. The invention may be practiced otherwise then as specifically described herein.