Technical Field

The present invention relates to a system for classification and optimization of timeseries data by using artificial intelligence algorithms.

Background of the Invention

Optimization is a standard transaction in automated time-series estimation systems being used today. However, time and resource are needed in an optimization transaction. Cost of performing optimization individually in a system running a plurality of time-series is high.

Considering the studies in the state of the art, it is understood that there is need for a system which classifies artificial intelligence algorithms of time-series patterns (unsupervised learning) and carries out optimization transaction by selecting one time-series from each class.

The United States patent document no. US2019317952, an application in the state of the art, discloses a system for clustering and improving data by analyzing time data generally. The said invention discloses carrying out the analysis of data hierarchically so as to generate more accurate predictions. The time data mentioned in the invention represent one or more time-series. A flow enabling to generate a machine learning model so as to make the prediction mentioned in the said invention accurately, is disclosed. The machine learning model is a mathematical artificial intelligence model that can learn from, categorize data and make predictions about data. The machine model learning model can classify input data among two or more classes by analyzing. It is enabled to predict a result based on input data and to identify patterns or trends in input data; identify a distribution of input data by identifying patterns and/or trends in input data. In embodiments of the current invention, a hierarchical analysis is generated automatically and it can be a recommended hierarchy. A recommended hierarchy enables to provide the best possible result with the attributes required to generate results that are optimized more accurately. Using a hierarchy optimized over a reference enables to obtain results providing high accuracy.

Summary of the Invention

An objective of the present invention is to realize a system which enables to make prediction more quickly, more cost-efficiently and more accurately without needing a separate optimization transaction for each time-series.

Detailed Description of the Invention

“Classification and Optimization System on Time-Series” realized to fulfil the objective of the present invention is shown in the figure attached, in which:

Figure l is a schematic view of the inventive system.

The components illustrated in the figure are individually numbered, where the numbers refer to the following:

1. System

2. Database

3. Classification server

4. Optimization server

5. Action server The inventive system (1) for classification and optimization of time-senes data by using artificial intelligence algorithms comprises: at least one database (2) which is configured to store the data about timeseries; at least one classification server (3) which is configured to receive the data about time-series from the database (2) and then to classify these data by artificial intelligence algorithms, and to transmit the classification information to the database (2); at least one optimization server (4) which is configured to receive classification information from the database (2) and then to determine a method used for optimizing the information; at least one action server (5) which is configured to receive the optimization method determined by the optimization server (4) and the parameters about the time-series from the database (2), and to apply the determined method to the time-series in the related class.

The database (2) included in the inventive system (1) is configured to store data about time-series.

The classification server (3) included in the inventive system (1) is configured to receive the time-series data kept in the database (2) and then to classify these data by pre-determined unsupervised artificial intelligence algorithms. The classification server (3) is configured to transmit the outputs about the classification realized by it to the database (2).

The optimization server (4) included in the inventive system (1) is configured to receive the information about the related time-series from the database (2) by selecting a time-series from each class obtained by the classification server (3). In a preferred embodiment of the invention, the optimization server (4) is configured to find the optimum methods and parameters of the value of a class number via Bayesian optimization by using the silhouette score of a classification algorithm on the received data and to transmit them to the database (2).

The action server (5) included in the inventive system (1) is configured to receive the classification data, the time-series data and the optimization methods and parameters of each class calculated, from the database (2). The action server (5) is configured to apply the optimum methods and parameters determined with respect to a class, to all time-series included in there related class. The action server (5) is configured to transmit the results of the optimization applied to the time-series, to the database (2).

In the inventive system (1), the database (2) stores the data about the time-series on it at first. The classification server (3) carries out a classification transaction by accessing these data and records these data in the database (2). The optimization server (4) receives the information about the related time-series by selecting at least one time-series from each class. It determines the methods and parameters related to the optimization with the received information. The action server (5) applies the optimum methods and parameters, that are determined to belong to a class, to all time-series in the related class and transmits them to the database (2). In different embodiments of the invention, each server (3, 4, 5) can store its outputs in different databases (2).

With the present invention, classification of artificial intelligence algorithms of time-series patterns (unsupervised learning) and optimization transactions can be carried out by selecting one time-series from each class.

Within these basic concepts; it is possible to develop various embodiments of the inventive system (1); the invention cannot be limited to examples disclosed herein and it is essentially according to claims.