Research On The Method Of Stock Price Prediction Based On The Bp Neural Network And Intelligent Algorithm

With the deepening of economic reform, people pay more and more attention to stock market, and improve the ways of stocks prediction. This thesis optimize particle swarm optimization(PSO) and genetic algorithm(GA), present their hybrid algorithm, and using BP neural network to combine with PSO, GA and their hybrid algorithm to forecast stocks. This thesis’ main content as follows:This thesis firstly introduce the basic concept, flow, advantage and disadvantage of neural network, particle swarm optimization and genetic algorithm. Secondly, the writer put forward the improved inertia weight method PSO, the selected seed GA. In the improved inertia weight method PSO, when inertia weight come down to a value, and the difference of adaptive value between two steps come down to another value, the inertia weight will increase a third value. In the selected seed GA, it copies the population twice, one copy for crossover, one copy for variation. Then, combine three copies, take out the one third individual of higher adaptive value as the next generation of population. Furthermore, using 4 test functions to test these algorithms. The results are that the improved inertia weight method PSO and the selected seed GA better than normal method.Next, the writer came up PSO and GA in parallel algorithm. At the first of each step of this algorithm, there is an information exchange mechanism. Take out the highest 10 adaptive value individuals of PSO and GA, sort according to their adaptive value, and put the higher 10 adaptive value individuals into PSO and GA, replace the take out individuals. Then, proceed the PSO and GA iterations at the same time. This thesis use the former 4 test functions to test this mixed algorithm and other two PSO and GA serial algorithms. The results say the PSO and GA in parallel algorithm is the best of the three.Last, this thesis uses better improvement effects of the improved inertia weight method PSO, the selected seed GA, and the PSO and GA in parallel algorithm, combining with BP neural network, to forecast stocks, and comparing with existing methods. This thesis uses Matlab 2010 b to write programs, uses the Sheffield University’s genetic algorithm toolbox gatbx run the simulation experiment. The numerical results say that the selected seed GA combines with BP neural network method is best.