影响特大型电网短期负荷的因素十分复杂 ,负荷变化属非平稳的随机过程。用神经网络 BP算法并采用变步长、加动量因子以及不等权、半随机初始解等方法 ,使迭代迅速收敛 ,减小预测误差、提高学习精度 ;用目标误差作为结束网络迭代条件 ;用测点模型矩阵预测不同季节、不同时段的负荷 ,在短期负荷预测中取得较满意结果 The factors that affect the short-term load of very large power grids are very complicated. The load change is a non-stationary stochastic process. BP neural network algorithm and the use of variable step, plus momentum factor and the unequal weight, semi-random initial solution and other methods, so that the rapid convergence of the iteration to reduce the prediction error and improve learning accuracy; target error as the end of the network iteration conditions; Point model matrix predicts load in different seasons and periods, and obtains satisfactory results in short-term load forecasting