摘要
针对BP算法存在的缺陷 ,如训练速度慢 ,易收敛于局部极小点及全局搜索能力弱等 ,利用遗传算法能够进行全局最优化搜索这一特点 ,提出了一种新的用于BP网络训练的混合算法 ,即遗传算法与改进的BP算法相结合的混合训练方法 .将所提出的混合训练方法应用于神经网络式电力负荷预测中 ,结果表明 :所提出的算法与单一的BP算法相比 ,不仅可避免陷入局部极小点 。
To avoid the shortcomings in BP algorithm,such as slowness in training speed,convergence to the Local minimum,and weakness in global search,a new hybird algorithm for network is presented.Then the hybrid algorithm is applied to neural network based load forecasting.The training and testing results show that hybird algorithm can not only avoid convergence to the local minimum,but also improve the training speed for the neural network and accuracy for load forecasting,as compared with the simple BP algorithm.