【作者】 郗亚辉；

【导师】 王熙照；

【作者基本信息】 河北大学 ， 计算机应用技术， 2006， 硕士

【摘要】 负荷预测是实现电力系统安全、经济运行的基础，对一个电力系统而言，提高电网运行的安全性和经济性，改善电能质量，都依赖于准确的负荷预测。中长期负荷预测的意义在于帮助决定新发电机组的安装与电网的规划、增容和改建，是电力规划部门的重要工作之一。 本文基于对人工神经网络基础理论的研究，利用多层前向网络的逼近能力，论述了利用BP神经网络进行中长期负荷预测的原理和方法。针对BP算法由于初始权值和阈值选取不当而陷入局部极小点、网络收敛速度慢的缺点，给出了将遗传算法引入网络初始权值和阈值优化的具体方法。在分析了电力负荷结构及影响负荷预测因素的基础上，提出了基于人工神经网络的负荷预测模型。模型以人口、主要国民经济行业总产值等社会经济指标为自变量，以电量为因变量，较全面地考虑了各社会经济指标对电量需求的影响，取得了较好的预测精度。更多还原

【Abstract】 Load forecasting is the basic issue to ensure the stable and economical operation of power system. For a power system, both the improvement of electricity quality and the increase of stability and economy depend on the precise load forecasting. The significance of medium and long-term load forecasting lies in the help it offers with the installation of generators and the planning, capacity increase and rebuilding of the electric network, which are among the most important work of the power planning utilities.Based on the basic theory research of ANN, the dissertation expounds the theory and methods of medium and long-term load forecasting with BP neural network, employing the approach ability of Multi-layer Feedforward Neural Network. However, BP algorithm can enter into local optimal point and slow down the convergence rate because of the improper selection of initial weight and threshold. To overcome these weaknesses, the dissertation elaborates how to introduce the genetic algorithm into the optimization of ANN initial weight and threshold. Besides, the dissertation advances the load forecasting model on the basis of artificial neural network, after it analyzes the structure of electricity load and the factors that influence the load forecasting. The model makes a fuller consideration of the effect of every social economic index on the demand of electricity quantity and achieves a higher precision of forecasting, with social economic indexes such as population, gross output value of main national economy trades, etc. as the independent variables and quantity of electricity as the variable.更多还原