Smith pre-estimation control method based on improved neural network

Abstract

The invention relates to a Smith pre-estimation control method based on an improved neural network. The method comprises the following steps of acquiring input and output data of a large-lag controlled object as a preliminary sample and performing time lag elimination; determining number of input layer nodes and output layer nodes of a BP neural network, and furthermore determining the number of hidden layers and the number of hidden layer nodes; determining the length of a genetic algorithm individual; obtaining an optimal individual by means of the genetic algorithm, decoding the optimal individual for obtaining an optimal initial weight and threshold of the BP neural network; training the BP network, and identifying a controlled object and a non-lag part of the controlled object by means of the BP network; and transmitting input and output information after BP neural network identification to a PIC controller which is set by the RBF neural network, and realizing pre-estimation control by means of the PID controller after setting. The Smith pre-estimation control method can realize better controlling on the large-lag controlled object.

Description

technical field [0001] The invention relates to the technical field of automatic control, in particular to a Smith predictive control method based on an improved neural network. Background technique [0002] In the actual industrial production process, the controlled object not only has a volume delay, but also widely exists different degrees of pure hysteresis. Lag is extremely detrimental to the performance of the control system. Within the lag time range, the controller cannot make corresponding adjustments because it cannot obtain the actual change of the controlled quantity at the current time from the closed-loop loop. The output of the control signal and the controlled quantity The action of the deviation is inconsistent, which will increase the adjustment time and overshoot of the system, and even cause the system to lose control due to instability in severe cases. It is generally believed that if the ratio of the pure lag time to the time constant of the process is...

AI Technical Summary

Problems solved by technology

At present, some scholars have proposed to use the BP neural network model to construct an online identifiable neural network Smith predictor by learning the relationship between the input and output of the controlled process. Inherent flaws in extremely small points that reduce their usefulness and reliability

In addition, the generalization ability of BP neural network is limited, so the method of online weight adjustment is generally used for identification. During the dynamic adjustment process, the identification output may have a large oscillation, so simply use BP neural network identification Large lag objects and then general control may still lead to system instability

Images