Control algorithm based on neural network deep learning

Abstract

The invention discloses a control algorithm based on neural network deep learning comprising the following steps: establishing a neural network, training a to-be-optimized neural network model, carrying out parameter quantification, training, calculating the precision before and after calculation, and if the precision of the optimized neural network model is judged to be reduced by more than 2% compared with the precision of the original to-be-optimized neural network model, cutting the to-be-optimized neural network model again until the precision of the optimized neural network model is reduced by less than 2% compared with the precision of the original neural network model to be optimized; then establishing a controller model, calculating input and output of each layer of the network, then calculating output of the controller, and performing real-time online adjustment on parameters of the fuzzy controller to obtain optimal control parameters. According to the invention, the control time is greatly shortened. The control precision can be effectively improved, and the control effect is idealized.

Description

technical field [0001] The invention relates to the field of artificial intelligence algorithms, in particular to a control algorithm based on neural network deep learning. Background technique [0002] The three cornerstones of artificial intelligence—algorithms, data, and computing power. Algorithms, as one of them, are very important and the most widely used. Algorithms are widely used in pattern recognition. [0003] Artificial Neural Networks (ANNs for short), also referred to as Neural Networks (NNs) or Connection Model, is a kind of algorithmic mathematics that imitates the behavior characteristics of animal neural networks and performs distributed parallel information processing. Model. This kind of network depends on the complexity of the system, and achieves the purpose of processing information by adjusting the interconnection relationship between a large number of internal nodes. It abstracts the human brain neuron network from the perspective of information pr...

AI Technical Summary

Problems solved by technology

[0005] (1) The complexity of the algorithm parameters is high: running the neural network algorithm model needs to consume very high hardware server resources;

[0006] (2) Neural network training is very time-consuming: Nowadays, no matter what type of neural network algorithm, it takes a long time to train in the training process, and finally a suitable model can be obtained, which will lead to the final required Time cost rises

[0007] (3) Low execution efficiency in an environment without GPU resources: the existing neural network training uses GPU graphics card resources for calculation, but the execution efficiency is greatly reduced on mobile platforms without GPU resources or GPU resources are not powerful enough

Images