Research On Neural Network Training Based On Intelligent Algorithm

With the advent of the era of big data,the application scenarios of neural network models represented by artificial intelligence will continue to increase,and the training data will also continue to be enriched.However,the high-dimensional training data also brings difficulties to the training of the network.In the traditional back propagation algorithm and its variants,the time of core gradient operation increases geometrically with the increase of the dimension of the data set.This paper presents a new model training method of neural network based on swarm intelligence algorithm which is widely used in recent years.It is of great significance to reduce the time complexity of neural network training algorithm and improve the training efficiency.In this paper,we propose a scheme to train the artificial neural network model by using swarm intelligence algorithm,and further improve the swarm intelligence algorithm on the trained artificial neural network model.The multi-layer perceptron and long and short memory network model are trained by swarm intelligence algorithms based on genetic algorithm,gray Wolf algorithm and whale algorithm,and the feasibility is verified in the experiment.The intelligent optimization algorithm was improved in the training artificial neural network model to verify the effectiveness of the improved strategy for users.The specific work is as follows:(1)A multi-layer perceptron training method is constructed,which takes the vector composed of multi-layer perceptron node weight and bias as the individual information of swarm intelligence algorithm,takes the mean square error as the objective function,and uses swarm intelligence algorithm to optimize the multi-layer perceptron training method in the solution space.(2)Improve the gray Wolf algorithm used in training.When the gray Wolf algorithm is training the multi-layer perceptron,it can solve the unbalanced problem of its solution space discovery and exploration stage,and improve its local retrieval ability at the same time.The effectiveness of the improved strategy is verified by experiments.(3)Swarm intelligence algorithm is applied to training short and long memory network of time-series data set.Firstly,the parameter matrix of short and long memory network is mapped into vectors,and these vectors are combined into individual information of Swarm intelligence algorithm.Then,the training model is optimized in solution space with absolute mean error as the objective function.On this basis,the improved swarm intelligence algorithm of training short and long memory network is improved to verify the effectiveness of the improved strategy.The experimental results show that the two neural network training strategies proposed in this paper have certain advantages in the training neural network model,and can improve the overall performance of the trained network.