带温度补偿的神经网络用于电力系统短期负荷预测  

Short-term Load Forecasting by Neural Network with Temperature Compensating

作　　者：夏经德[1,2] 夏道止[2] 

机构地区：[1]淄博科汇电气有限公司 [2]西安交通大学,陕西西安710049

出　　处：《西北电力技术》2006年第3期1-4,共4页Northwest China Electric Power

摘　　要：带温度补偿的神经网络结构和短期负荷预测方法在前向神经网络的输入和输出节点上对负荷引入了相应的温度补偿,其中所涉及的临界温度和温度补偿系数将与前向神经网络本身的权系数一起通过训练而自动获得。对陕西电网2001年1月开始连续45个月的预测试验结果表明,高温日期的平均负荷预测精度比用常规前向神经网络高3％-6％。神经网络的训练采用求解无约束最优化问题的BFGS算法,不但保证了神经网络学习的收敛性,而且可以减少隐节点的数目,使神经网络的推广能力和预测精度显著提高。In order to improve the load forecasting accuracy in hot days, a novel artificial neural network with temperature compensation is proposed with the aid of operating experience in power systems that the electric load will be approximately increased a certain percentage when the temperature is raised one degree centigrade. The novel neural network is an extension of the conventional feed-forward neural network in which the load data in the input and output is changed to the corresponding factitious loads through compensating by temperature coefficients. The critical temperature and compensating coefficients are automatically and simultaneously obtained with the weighting coefficients through neural network training. The tests results of load forecasting for continuous 45 months of Shaanxi power system show that the load forecasting accuracy in the hot days can be increased 3% -6% by using the proposed neural network compared to that obtained by conventional feed-forward neural network. The neural network is trained by BFGS algorithm, so that the convergence of learning is guaranteed, and the number of neural in the hidden level can be greatly reduced which makes the generalization capability and forecasting accuracy significantly increased.

关 键 词：电力系统 短期负荷预测 神经网络 温度补偿BFGS优化算法 

分 类 号：TM714[电气工程—电力系统及自动化]