CNN-ARX model-based linear primary inverted pendulum system modeling method and CNN-ARX model-based linear primary inverted pendulum system model

Abstract

The invention discloses a CNN-ARX model-based linear primary inverted pendulum system modeling method and a CNN-ARX model-based linear primary inverted pendulum system modeling model. An inverted pendulum is a relatively complex system with strong nonlinearity and instability, an accurate mechanism mathematical model of the system is difficult to obtain, and dynamic characteristics of the system can be described by adopting the CNN-ARX model. According to the invention, a convolutional neural network (CNN) technology, a local linearization method and a state-dependent ARX model are used to construct a CNN-ARX model structure; parameters of the convolutional neural network are optimized through an RMSprop algorithm, and the output of the convolutional neural network hidden layer is randomlyreturned to zero by adopting a random inactivation technology (dropout), so that the interdependence between nodes is reduced, and the risk of model over-fitting is reduced. According to the method,the prediction precision and robustness of the inverted pendulum system identification model can be improved, and the method has relatively high practical value and application prospect.

Description

technical field [0001] The invention relates to the field of engineering design and optimization, which can utilize input and output discrete time series data of a linear one-stage inverted pendulum system, in particular a modeling method for a linear one-stage inverted pendulum system based on a CNN-ARX model. Background technique [0002] The linear one-stage inverted pendulum is a kind of strongly nonlinear, unstable, and relatively complex system. It can be modeled in a data-driven manner, and the relationship model between the input and output variables of the system can be constructed to describe the dynamic characteristics of the inverted pendulum system. As a nonlinear modeling method, the state-dependent ARX model has the advantages of the ability to describe the nonlinear dynamic characteristics of the coefficients of the state-dependent function and the advantages that the autoregressive structure is easy to apply to control. Linear time series modeling is widely ...

AI Technical Summary

Problems solved by technology

As a nonlinear modeling method, the state-dependent ARX model has the advantages of the ability to describe the nonlinear dynamic characteristics of the coefficients of the state-dependent function and the advantages that the autoregressive structure is easy to apply to control. Linear time series modeling is widely used in the field of linear time series modeling. One of the core problems of using the state-dependent ARX model to model the inverted pendulum system is to find the appropriate structure of the state-dependent function coefficients. Using the neural network to approximate the coefficients of the state-dependent ARX model can be Obtain higher-performance identification models, such as RBF-ARX, DBN-ARX, etc. In theory, the neural network can fit any complex nonlinear relationship when the network level is deep enough. In practical applications, the neural network may There are problems such as underfitting, overfitting and gradient disappearance

Images