A Nonlinear Inverse Model Control Method Based on Improved Extreme Learning Machine

Abstract

The invention discloses an improved extreme learning machine-based nonlinear inverse model control method. The improved extreme learning machine-based nonlinear inverse model control method comprises the following steps that: population initialization is performed on MAPSO, and relevant operation parameters of the MAPSO are set; MAPSO optimization is performed on ELM parameters including input weights and hidden layer threshold values, so that appropriate ELM parameters can be determined; MAPSO-ELM is utilized to perform direct modeling on inverse of a nonlinear discrete controlled object; and a trained MAPSO-ELM inverse model is directly combined with an original system, so that inverse model control of a nonlinear system can be realized, and the methods terminates. With the improved extreme learning machine-based nonlinear inverse model control method adopted, prediction accuracy and generalization ability can be effectively improved. According to the improved extreme learning machine-based nonlinear inverse model control method, the improved extreme learning machine (MAPSO-ELM) is directly applied to the inverse model control of the nonlinear system, and therefore, problems such as difficult modeling of an inverse model of the nonlinear system as well as over-learning, low training speed and high possibility of falling into local optimal values of a traditional inverse model control method can be solved.

Description

technical field [0001] The invention relates to a nonlinear system inverse model control method, in particular to a nonlinear system inverse model control method based on multi-agent particle swarm (MAPSO) optimization of extreme learning machine parameters. Background technique [0002] The inverse system method is a feedback linearization decoupling method, which finds its inverse model under the premise of the existence of the inverse of the system, and forms a pseudo-linear system in series with the original system, so as to complete the feedback linearization. At present, the commonly used methods used in inverse model control of nonlinear systems include neural network, support vector regression (SVR) algorithm and other methods. Although these algorithms have achieved good results in the application of inverse model control in nonlinear systems, due to the limitations of these algorithms (the selection of neural network structure types is based on experience selection...

AI Technical Summary

Problems solved by technology

Although these algorithms have achieved good results in the application of inverse model control in nonlinear systems, due to the limitations of these algorithms (the selection of neural network structure types is based on experience selection, there are problems such as local optimum, SVR algorithm There are problems such as robustness, sparsity, and large-scale operations), making problems such as over-learning, slow training speed, and local minima unavoidable

Images