基于改进遗传算法的BP神经网络自适应优化设计

Self-adaptation Optimize BP Neural Network Design Based on the Genetic Algorithms

BP(Back Propagation)神经网络在网络训练中存在着局部最优问题,其算法收敛过慢、局部收敛不理想,影响其工作性能.针对以上不足以及传统神经网络设计规模庞大等问题,提出了一种由EGA(改进的遗传算法)确定网络拓扑结构和训练网络的方法,该方法通过实数编码、自适应多点变异等操作有效地优化了网络拓扑结构和网络参数,从而有效缩小了网络规模和提高了BP网络训练的速度以及收敛的有效性.最后结合了番茄常见病害诊断的实例说明了此方法的可行性.

BP （Back Propagation） Neural networks is in the presence of the local optimization in the Neural networks training. The algorithm have slow convergence and the local convergence problem which impact the neural networks work performance. In order to cover these shortcomings and solves the size＇s hugeness and the low efficiency of the net problem in the traditional NN designing, the action principles of BP-Neural network＇s structure are analyzed, and a new method is formed which is confirmed from the Enhance genetic algorithms （EGA）. The method can identify network configuration and network training methods. By adopting the number coding, self-adaptable multi-point variations operation, this method can effectively reduce the network size and the network convergence time, increase the network training speed. Tomatoes disease diagnosis examples illustrate the feasibility of this approach.