Artificial Neural Network Co-optimization Algorithm Based on Differential Evolution

BP network training algorithm is based on the error gradient descent to modify weights, which leads to the inevitable problem of a local minimum point. Some researchers have presented some amending ways and made some remarkable achievements. But combining others algorithm for adjusting the weights of BP network is few. At present, a new evolution algorithm called as differential evolution is used wildly. The differential evolutionary algorithm as a global search algorithm has many advantages, especially its optimizing speed. However, the differential evolutionary algorithm has its lack such as the capacity of local search is not as good as the BP algorithm. This paper combines the BP algorithm and the differential evolutionary algorithm´s superiority to complete neural network weights and threshold value adjustments.