Exchange Rate Forecasting Method Based On Particle Swarm Optimization And Probabilistic Neural Network Model

Foreign exchange market is a complex market, with a high degree of volatility characteristics. Exchange rate formation mechanism and the factors affecting exchange rate volatility is also very complex, is a nonlinear system, it is difficult to accurately forecast, That also makes the foreign exchange rate forecasting become the research focus in data mining area. The neural network is a nonlinear dynamic system with highly nonlinear global role, a good fault tolerance and robust adaptive auto-learning ability。In theory, neural networks can approximate nonlinear systems to arbitrary accuracy in solving foreign exchange rate prediction of such highly nonlinear problem that has unique advantages.Probabilistic neural network is the neural network making use of the Bayes theorem and the Bayesian minimum risk-based decision rules to classify the new samples. The neural network model based on the statistical theory, with advantages of short training time and difficult to converge to the local maximum point, but its network structure limits its output only as a category classification. In this context, the probabilistic neural network will be applied to the new field, exchange rate forecasting. Aim at this feature of Probabilistic neural network, data preprocessing the exchange rate, exchange rate volatility on classification, to achieve a qualitative and quantitative changes in order to probabilistic neural network to predict the trend of the exchange rate. Exchange rate in the basic probabilistic neural network prediction model, by changing the input vector dimension of the experiment, the final optimum embedding dimension input vector. In this model, based on particle swarm optimization algorithm is probabilistic neural network to optimize the smoothing factors, and applied to exchange rate forecasting, Obtained good results. Finally, particle swarm algorithm is easy to fall into local optimum conditions, the inertia weight particle swarm parameters for the dynamic adjustment, proven to improve the prediction accuracy, there are some practical applications.