PSO-BP Neural Network Based Prediction Study for Nonlinear Problems

The Back Propagation (BP) neural network is commonly used for predicting nonlinear problems. However, this model is known to converge slowly and is prone to local optimisations. To address this issue, using a population intelligence optimisation algorithm to optimise the structure of the BP neural network is highly significant. This paper aims to explore the prediction of nonlinear problems by establishing the prediction model according to the three-layer architecture mode of BP neural network and using the characteristics of global optimisation of the particle swarm algorithm (particle swarm optimization, PSO) to optimise the initial weights and thresholds of BP neural network, so as to propose a kind of prediction model of nonlinear problems based on PSO-BP neural network. Based on the experimental results, the coefficient of determination R² of the PSO-BP neural network is 29.41% higher than that of the single BP neural network. Moreover, the RMSE is reduced by 64.08%. These findings demonstrate that the PSO algorithm-optimised BP neural network has superior prediction accuracy and stability for nonlinear problems than that of the single BP neural network. This provides constructive optimisation of the structure of the BP neural network.