|
Artificial Neural Network (Artificial Neural Network, ANN) nonlinear systems using computer simulations of biological neural tissue . It has a strong self-organization , adaptive learning , parallel processing and high fault tolerance . So far , a number of scholars have proposed dozens of neural network model , and successfully applied in many fields . BP network as the most widely used neural network has the advantages of simple structure , mature technology . However , BP algorithm based on gradient descent methods , there is slow network convergence , easy to fall into local minimum point . Firstly, the development process of the artificial neural network , the basic principles of learning , classification and application made ??a more detailed analysis, the principle of BP algorithm , the learning process and the presence of defects , at home and abroad and introduced improved BP algorithm Research . Based on in-depth analysis of BP algorithm , a comprehensive improvement algorithm , is about to adaptively adjust the learning rate dynamically adjust the activation function combining . In each of the algorithm in the learning process , according to the error changes dynamically adjust the learning rate and activation function , thereby enhancing the convergence performance of the algorithm . The experiments show that the comprehensive improvement BP algorithm can effectively improve the convergence performance of the network . Finally, this article will be improved BP neural network is used in the prediction of merchandise exports . Established based on improved BP neural network prediction model of commodity exports , merchandise exports were forecast , and a comparative study with other commonly used forecasting methods . The results show that the higher the accuracy of the results than other prediction method based on improved BP neural network forecasting model .
|