BP Neural Network Forecasting Model Based On PSO

The stock market as an economic "barometer" plays a very important role in Chinese economy, an effective stock forecasting occupies an important position in the area of financial investment. However, the stock market is affected by the policy, the economy, the psychology of investors, and other complicated factors. It is a very complex nonlinear dynamic system with typical features of the complex uncertainty.With the capability of approximating any nonlinear system and speciality of self-learning and self-adapting, Neural network can extract from the historical data the knowledge of economic activities automaticly, Thus neural network is adaptive for solving problems of economical prediction. The experiments have shown that the method of modeling stock market with neural network leads to a satisfying result in near-period stock prediction.But the traditional neural network has problems of studying slow, easy to fall into the local minimum value and disadvantages of low prediction accuracy. To address these shortcomings, the PSO (Particle Swarm Optimization) algorithm is applied to the optimization of BP Neural networks in this paper, and the PSO algorithm is optimized, the appropriate BP neural network optimization algorithm is selected based on a large number of experiments. The main jobs done in the paper are:Firstly, the predictability of the stock system and the factors of influencing it are stated in the paper. The methods of estimating the influence of the intervention on the stock system home and aboard are expounded.Secondly, proper stock index prediction model based on BP neural network is set up. Define the prediction model structure base on experiments; combinig the stock market law, select the input and output variables, and learn new law from the historical data to predict the future trends.Finally, based on studying shortcomings of BP algorithm, the PSO algorithm is applied on the optimization of the BP neural networks. the weights and valve values of the neural network are trained with PSO in order to fast the leaming speed; a proper PSO algorithm is chosen from the different PSO optimization algorithms based on a fair amount of experiments and a BP neural network forecasting model based on PSO is built.It is proofed that the decreasing strategy with control of nonlinear factors for PSO algorithm in optimization of BP neural networks have achieved better results.