PURPOSE: To execute an equal flow rate control absorbing a mutual interference operation by using various control parameters as samples, evaluating a difference between the parameter and the sample when the output of a neural network is set to be the manipulated variable of a flow rate control valve and adjusting the connection degree of the network.

CONSTITUTION: Respective charge valves 4-1 to 4-4 are operated by using the manipulated variables MV1-MV4 which a neural network 10 outputs. When disturbance occurs in one of return systems, the respective manipulated variables M1-MV4 are simultaneously adjusted in a direction where the disturbance is converged and disturbance is immediately converged. A learning function part 14 learning the network 10 prepares respective control parameters in a specified control pattern, analyzes it and generates an experience rule. For learning, an initial value such as the weight coefficient of the network 10 is set by simulation or the like. Furthermore, the parameter when the output of the network 10 is used as MV4 is self-evaluated by using the experience rule.