Interval feedback neural network-based uncertain system modeling method

Abstract

The invention provides an interval feedback neural network-based uncertain system modeling method, and relates to the technical field of system modeling in the industrial process. The method comprisesthe steps of collecting actual input output data pairs of a system; collecting system data pairs under a UBB condition; normalizing actual point value inputs and actual interval output data corresponding to the actual point value inputs; performing offline training on an interval feedback neural network by taking the normalized data as training data to obtain a trained interval feedback neural network; and testing the trained interval feedback neural network by utilizing test samples and finishing output value prediction. According to the interval feedback neural network-based uncertain system modeling method, network weight learning is performed by utilizing a nonlinear approximation capability of the interval feedback neural network and adopting an error back propagation-based gradientdescend algorithm, so that the input of a large amount of neurons and the demand on a system mechanism model are avoided; and the method is widely suitable for the modeling process of the high-order dynamic system with UBB errors.

Description

technical field [0001] The invention relates to the technical field of system modeling in industrial processes, in particular to an uncertain system modeling method based on an interval feedback neural network. Background technique [0002] In the actual system, due to the incompleteness of measurement and estimation, the disturbance and uncertainty in the industrial process, the obtained data are often inaccurate, and the equivalent model of the system determined by the input and output data is not accurate enough. Therefore, it is more reasonable to use an uncertain system model to describe an actual process. Expressing uncertainty as interval numbers is a simple and effective way to deal with it, and a system with interval parameters is called an interval system. If a deterministic system model is used to describe, there will inevitably be a certain deviation between the actual system and this model. [0003] The system description based on the unknown but bounded error...

AI Technical Summary

Problems solved by technology

Interval feedback neural network can solve the problem of high-order dynamic system modeling by feedforward neural network due to its own structure of memory and adaptability to time-varying characteristics, so it can be used as an effective means of modeling uncertain systems

Images