对前馈型神经网络股价预测的一点改进

An Improvement of Prediction of Stock Prices Based on Feed-forward Neural Network

为减小预测误差,目前的方法是:选择对股价有显著影响的输入变量;调整网络结构及选择合适的参数;优化学习算法.这些方法都是以实际股价作为网络的预测.本文提出的方法是以相邻两天的股价差价作为网络的预测.本文在相同的数据上分别建立股价预测模型和股价的差价预测模型,并对它们的结果进行对比分析,结果表明差价预测模型的预测误差小,预测效果好.

In order to reduce the prediction error, the present methods are to choose the significant parameters that form the inputs of the neural network, adjust the architecture of the neural network, choose appropriate parameters, optimize the Algorithm. These methods all have been taken the actual stock price for the network prediction. The proposed method has taken the difference between the stock price of the next day and that of the present day for the network prediction. The article has been to establish the stock price model and the price difference model on the same data, and make a comparative analysis of the results. The experimental study indicated that the most accurate prediction results are produced by neural networks trained to predict the difference between the stock price of the next day and that of the present day.