Research And Implementation Of The Intelligent Securities Analysis System Base On BP Neural Network

The application of the artificial intelligence in financial area has always been a hottopic in the field of both computer and finance. Mainly on the basis of artificial neuralnetworks, this essay will predict and analyze the future price changes of securities,aiming to assist the investors to make correct investment decisions and improve theirperformance, as well as lower their risks. Currently, there are some different predictiveanalysis methods in the securities market; However, few of them could provide avaluable and efficient predicative analysis on the future movements of the securities.The movement of the securities market by its own nature is kind of extremelycomplicated non-linear problem, and the artificial neural networks are able to deal withsuch issues by imitating the creatures’ reactions in the natural world. Thus, it isreasonable that the artificial neural networks can make a predicative analysis about thefuture change of the securities market more efficiently.This essay introduces briefly the working principle of artificial neural networks,together with the description of feed forward neural networks, feedback neural networksas well as self-organization neural networks. It conducts study on the basis of feedforward neural networks. At first, error-back propagation (BP) neural networks isanalyzed and improved with the use of levenberg-marquardt algorithm, additionalmomentum and adaptive learning rate algorithm. The comparison results show that theimproved BP neural networks with the application of levenberg-marquardt algorithm isbetter than two other improved methods in terms of convergence speed, while, inferiorin both the local and global search abilities. It’s therefore unlikely to apply such BPneural networks in the field of prediction and analysis. In order to address thoseproblems, this article presents the solution of optimizing the improved BP neuralnetworks through the utilization of simulated annealing genetic algorithm (SAGA). Byintegrating elitist model simulated annealing algorithm and genetic algorithm, SAGA isable to fully play the features of both strong local and global search abilities, avoidingfalling into local minima in simulated annealing method, and making up for the shortagein local search of genetic algorithm. Next, a new BP neural networks is employed topredict and analyze variations of future prices and tendencies of securities. In the caseof Shanghai stock index, different neural networks are applied to predict stock pricefluctuation and make analysis. It’s seen that the new neural network after optimizationmakes great improvement in stability, convergence speed and prediction precision.Finally, on the basis of the proposed BP neural networks, a system is developed for theprediction and analysis of the change of securities prices or trends in the future. Inconsideration of too much factors influencing stock market, the function of adjustmentover neural networks input information is added, improving data completeness and also reducing the noise interference. Security market is always turbulent and has strongrandomness. In order to improve flexibility of the system, the feature of parameteradjustment is supplemented to it. Through the adjustment of parameters, networkperformance can be tuned in response to market situation.