Dynamic model identification method and system based on recurrent neural network

Abstract

The invention discloses a dynamic model identification method and system based on a recurrent neural network. The system includes a dynamic model identification module, a module for converting a recurrent neural network model into a transfer function model, and a nonlinear module for converting a recurrent neural network into a state space model and the recurrent neural network. The dynamic modelidentification module includes a forward calculation module, an error calculation module, and a back-propagation calculation module. The method includes first establishing the recurrent neural networkmodel; converting the recurrent neural network model into the transfer function model or the state space model as needed; or non-linearizing the recurrent neural network model. The dynamic model identification method and system based on the recurrent neural network are simple to calculate and convenient to implement, the training time is quick, and mutual conversion can be performed with the transfer function model and the state space model. The dynamic model identification method and system can be well popularized to non-linear situations, a good fitting effect can be achieved, and the dynamic model identification method and system can be applied to industrial control model identification very conveniently.

Description

technical field [0001] The invention relates to thermal power engineering and automatic control methods, in particular to a dynamic model identification method and system based on a cyclic neural network. Background technique [0002] The thermal control process generally adopts the step response experiment to obtain the non-parametric representation method of the step response curve, and processes the non-parametric model into the expression form of the approximate transfer function model. The parameters obtained by the graphing method mainly include the following types: [0003] 1. Use the tangent method to obtain the delay time on the curve, the response speed and the self-balancing coefficient to obtain the parameters such as the time constant, amplification factor and order of the inertial link. [0004] 2. Since the tangent method only uses the data of the inflection point on the step response curve, there may be large differences due to different plotting results. The...

AI Technical Summary

Problems solved by technology

However, this method has a large computational workload, so the chance of errors in numerical calculations is greatly increased, and it is not intuitive, and it is difficult to find unreasonable results and find out the reasons

[0006] Therefore, the method of drawing is not able to obtain the transfer function model more accurately

Images