Theory of stock market prediction based on BP neural network is discussed and the prediction model of stock market is established using the three-layer feed-forward neural network. The problems including the structure of network, the number of hidden nodes, choice and pretreatment of sample data and the determination of preliminary parameters are discussed. By analyzing BP neural network's model, structure and training rule, a time series prediction model based on BP neural network are constructed. Issues like network size, network generalization are also studied. A BP network model with single hidden layer and single output is used to forecast the trend of the index of the coming day. Levenberg-Marquardt algorithm is used to train the network, and the number of nodes of hidden layer is refined. The BP network simulation program is written in MATLAB. And fuzzy logic is also introduced to revise the result using fuzzy rules. The simulation result shows that the purposed method is effective. |