Implementation of Diagonal Recurrent Neural Network Controller in Multiple Platforms

Abstract

The present invention relates to a method for implementing a diagonal recursive neural network controller in multiple platforms, which is implemented as follows: establish a DRNN neural network controller, which includes a DRNC neural network and a DRNI neural network; correspond to the DRNC neural network and the DRNI neural network Select the learning and training method; establish the topology structure of the DRNN neural network controller through the configuration of the computing unit module; establish the mapping relationship between the computing unit module and the general middleware; establish the mapping relationship between the general middleware and the target platform; Deploy to the target platform; analyze the mapping relationship between the general middleware and the target platform on the target platform; run the DRNN neural network controller on the target platform. The method proposed by the present invention can be mutually transplanted and shared among various different types of computing platforms, supports online configuration and debugging, and its learning, training and testing applications can all be realized in the lower computer, which can meet the long-term stable work of industrial sites needs.

Description

technical field [0001] The invention relates to the technical field of industrial automation software control, in particular to a method for realizing a diagonal recursive neural network controller in multiple platforms. Background technique [0002] In complex process industrial systems, the controlled objects are mostly multi-input and multi-output dynamic time-varying parameter systems, and it is difficult to establish an accurate mathematical model; at the same time, the conventional PID control method is also difficult in the precision control and decoupling control of nonlinear systems. It is difficult to achieve the ideal control effect. As an intelligent control method, neural network control technology can fully approximate the dynamic behavior of unknown nonlinear objects, and can make up for the limitations of conventional PID methods. It is often used to solve some control problems of nonlinear systems that are difficult to model. However, the commonly used mult...

AI Technical Summary

Problems solved by technology

[0013] The purpose of the present invention is to provide a method for implementing a diagonal recursive neural network controller in multiple platforms, which is used to solve technical problems such as difficult transplantation and sharing of DRNN neural network algorithm programs in the prior art on different types of computing devices, and to Overcome the deficiencies of existing technologies

Images