Research On Neural Network Model Based On IWO Algorithm And Its Application

Stock market is known as "barometer" and "alarm" of the national economy. As it has the characteristic that high-yield coexists with high-risk, so the research on stock market analysis and forecast is not only valued by the government, but also is paid attention by the majority of investors. So the research on stock market analysis and forecast has important theoretical significance and practical application value. In this paper, based on “Research and development of quantitative combination forecasting system based on non-linear mapping” of Talent Project in Liaoning Province, stock price of domestic enterprise is object for forecasting, the improved learning algorithm for feed-forward neural network and their corresponding prediction model are carried on depth inquiry.Neural network has characteristics of strong ability of nonlinear approximation, self learning and adaptation, which makes the neural network show great advantage in terms of stock price forecasting. But the traditional neural network based on BP algorithm has disadvantages of slow convergence speed, easily falling into local minimum value and low prediction accuracy. In order to overcome the defects of BP algorithm, a novel swarm intelligence optimization algorithm is introduced into, which is invasive weed optimization(IWO) algorithm, and learning algorithm of neural network based on IWO algorithm and its improved algorithm are proposed. On this basis, prediction model of neural network based on these algorithms and their application are carried on in-depth research. In this paper, the main research content includes the following aspects:On the basis of review of factors that affect stock price changes, related evaluation for stock price predicting and the forecasting methods for stock price at home and abroad, according to the prediction of neural network, the learning algorithm of feed-forward neural network and the application in feed-forward neural network of a variety of intelligent optimization algorithms are summarized, which provides the basic data for determining research direction which introduces IWO algorithm into neural network and establishs predicting model.On the basis of in-depth study of feed-forward neural network learning algorithm and IWO algorithm, in order to overcome the defects of BP algorithm, firstly, neural network learning algorithm based on IWO algorithm and the corresponding prediction model are proposed. In order to further improve the optimization accuracy and speed up the optimization of IWO, and to solve its disadvantage of easily falling into local optimum, on the basis of without changing the normal space radiation pattern of IWO, in this paper, the traditional complex method is used to improve the IWO, a CIWO algorithm is proposed, CIWO is used to train the neural network, and the corresponding forecast model is established. Aim at the defects of easily falling into local optimum and the low optimization accuracy in the later of IWO algorithm, leading to the problem of optimization accuracy being not high of IWO algorithm training the neural network. Combining species diversity of invasive weed algorithm with heuristic global searching characteristics of differential evolution, differential evolutionary invasion weed(DEIWO) algorithm is proposed for training the feed-forward neural network, and a forecasting model of neural network based on DEIWO is obtained.In this paper, taking Hunan Sany shares(600031) as the research object, using the above several kinds of prediction model for instance simulation, the experimental results show that the neural network model based on IWO has higher prediction accuracy than that of BP neural network. The learning algorithm of neural network based on CIWO algorithm and the learning algorithm based on DEIWO algorithm improve that of IWO algorithm from different angles, their corresponding prediction model of neural network established improve the performance of neural network based on basic IWO algorithm from different degree, their prediction model both have high prediction accuracy in simulation examples. It further demonstrates the feasibility and effectiveness of these improved models by comparative analysis.