The Research And Application Of Neural Network Model Based On Particle Swarm Optimization In Prediction Of Stock Price

On basis of the deep analysis and research on artificial neural network that is one of hot spots on the international stock forecast, it exists some shortcomings that it is slow at the convergence rate, absent of ability for global search and easy to trap in local minimum. So it causes forecast precision of the stock price low finally. The neural network model based on Particle Swarm Optimization (PSO) and its algorithm is proposed. This model has utilized the principal components analysis (PCA) method to simplify system's input for improving efficiency by the choice of input variables. The main work is as follows:(1)BP algorithm: through the deep research on BP algorithm, it is used to forecast the realistic stock price by MATLAB. The experiment indicates that its forecast precision is not high so that it cannot meet the actual request.(2)PSO algorithm: based on the discussion and the comparison to PSO algorithm's improvement, it is determined that the PSO algorithm with inertia factor is the foundation of neural network algorithm based on PSO.(3) PCA method reduces dimensionality for the multi-dimensional variables: the method is introduced to reduce dimensionality for the original input variables.On the one hand, PCA reduces the input dimension and eliminates the correlation of all the variables. On the other hand, it enhances the convergence and stability of the network and simplifies the structure of the network. Experiments show: compared to the multi-variables without treatment, the number of its input variables is changed from 15 to 3, which greatly reduces the input dimension and run-time. Forecast accuracy has been improved.(4)Neural network model based on PSO and its algorithm: the algorithm is the center of this research. Through deep research on the BP algorithm, the PSO algorithm and the PCA method, the model and its algorithm is proposed. The experimental results show that this approach is more efficient than BP algorithm in prediction of the stock price.