摘要
针对配电网复杂多变的特点 ,提出了用神经网络和最优化算法相结合进行配电网超短期负荷预测的研究方法。采用分时段的负荷预测方法 ,大大缩小了网络规模。在神经网络的训练中 ,采用变步长的 BP算法 ,并实行远小近大加权均方的误差计算原则。运用了遗传算法和模拟退火两种最优化算法分别与神经网络算法相结合 ,并进行了比较。在遗传算法中首次引入了聚合度的概念。当两种算法结果相差不大时 ,用它们的平均值作为最后结果 ,进一步提高了预测精度 ,尤其是提高了重大节假日这一预测难点的精度。运用本算法编制了实用性软件 ,并对潍坊地区的真实负荷进行了预测 ,结果较好地满足了现场的要求。
In allusion to the characteristics of agility and levity of distribution network, this article puts forward a method for load forecasting of the distribution network, combining ANN method and optimization methods. In this article, the load is forecasted one hour by one hour, so the size of the ANN is greatly reduced. BP algorithm using variable step length is used for the training of the ANN. The error of every day is multiplied by a weight. The later the date is, the bigger the value of the weight is. Then the RMSE of these errors is used to train the network. GA and SA are combined to ANN respectively and the results are compared. The convergence degree is introduced into GA for the first time. When the two results are close, the average of them is regarded as the final result, and the forecasting precision is improved further, especially the precision of the holiday, which is hard to forecast. The real load of Weifang area is forecasted by the practicable software using this algorithm and the result can meet the demand.
出处
《电力系统自动化》
EI
CSCD
北大核心
2001年第22期45-48,共4页
Automation of Electric Power Systems