Study On BP Neural Network Based On Improved Glowworm Swarm Optimization Algorithm

BP neural network is one of the most important neural networks at present, which has the advantages of simple structure, stable working state, being easily implemented in hardware, and is widely used in pattern recognition, classification and prediction, system simulation and image processing and many other fields. However, BP neural network has been found that there are many defects in the process of wide application, such as being sensitive to initial value, being easy to fall into local minimum, the hidden layer structure is difficult to determine, and so on. To solve these problems, a method of using improved glowworm swarm optimization algorithm to train the weights and thresholds of BP neural network is put forward.Glowworm swarm optimization algorithm is an optimization algorithm based on swarm intelligence, and can quickly find the global optimum. In order to avoid the glowworm swarm optimization algorithm emerging the phenomena of premature convergence, slow convergence speed in the later period and low convergence precision, a glowworm swarm optimization algorithm is put forward based on hybrid mutation. The algorithm using chaotic mutation and boundary mutation hybrid mutation method to increase the diversity of glowworm swarm, makes the glowworms fall into the local extremum can promptly jump out, and does not appear the phenomenon of too many glowworms gather near the boundary. Experimental results with six standard test functions show that the improved glowworm swarm optimization algorithm has higher convergence speed, convergence accuracy and convergence rate than the basic glowworm swarm optimization algorithm.The initial weights and thresholds of BP neural network are generated randomly, and the overall distribution of the initial value has great influence on whether the network will fall into local minimum, and the performance and fitting effect of network. Using the improved glowworm swarm optimization algorithm combined with BP algorithm to obtain better initial weights and thresholds of network, and to start the network learning. Through simulation results show that, the hybrid optimization algorithm has higher testing accuracy and better fitting ability on the problem of classification and prediction, avoids the problems of being sensitive to initial value and being easy to fall into local minimum in BP neural network, and improves the generalization ability, convergence speed and learning capability of BP neural network, to verify the feasibility and effectiveness of the optimized algorithm.Determine the influence factors and forecasting evaluation functions of GDP, and use the BP neural network which is optimized by glowworm swarm algorithm with hybrid mutation to establish GDP forecasting model of Nanning city. Combining the collected sample data for training and testing, the analysis of the prediction results show that, the optimized algorithm has higher prediction accuracy than other algorithms.Finally, the work is summarized and the future research direction is put forward.