【作者】 陈艳；

【导师】 程春田；

【作者基本信息】 大连理工大学 ， 水文学及水资源， 2006， 硕士

【摘要】 电力系统的短期负荷预测是指以周、天、小时为单位的负荷预测,它是现代电力系统运行研究中的重要课题之一。电力系统短期负荷预测的结果是研究电力系统规划问题、电力系统经济运行及其调度自动化的重要依据。随着电力市场化改革的不断深入,短期负荷预测在电力系统中更显得日益重要。实践证明,在电力系统发展日趋复杂的今天,各种传统的负荷预测技术已经越来越难以满足电力部门越来越高的负荷预测精度要求,所以应用智能算法进行电力系统的短期负荷预测,提高负荷预测的精度和稳定性,具有十分重要的意义。 本文首先概述了电力系统短期负荷的原理、特点、研究现状及发展趋势,对电力系统的短期负荷预测的各种传统方法及现代方法进行了综述,并重点研究了人工神经网络在短期电力负荷预测中的应用。针对神经网络中最常用的BP算法所存在的收敛速度慢,容易陷入局部极值的问题,介绍了几种改进的快速学习算法。仿真结果表明改进的BP快速训练算法有效改善了BP算法的缺点,提高了神经网络用于电力系统短期负荷预测的效率和精度。然后从BP神经网络的理论入手,考虑到在BP网络中最初的权值和网络的构造对预测结果的精度影响最大,采用遗传算法优化BP神经网络的初始权值和隐层节点数,从而避免了神经网络结构确定和初始权值选择的盲目性,使得负荷预测在更加合理的网络结构上进行。实例证明,用遗传神经网络进行短期负荷的预测可以提高预测的精度。最后,以云南省昆明地区为代表,研究该地区的负荷特性,在此基础上,将负荷预测按日期分为不同类型的预测模型,并将对负荷影响很大的天气因素归一化后输入神经网络,使得预测更加合理,更能适用于一般情况。更多还原

【Abstract】 The short-term load forecasting of electric power system, predicting electric load for a period of hours, days, or weeks, is an important research area of electric power system’s operation. It is the important foundation of the study on electric system planning problem, economical running and dispatcher automation. Furthermore, with the establishment of power market, short-term load forecasting will play a more important role in the future. With the power system becoming more and more complex, it’s demonstrated that those traditional load-forecasting technologies can’t satisfy the requirement of load forecasting accuracy, which becomes more and more strict. So using intelligent technologies to improve the forecasting accuracy and stability of the load forecasting of electric power system is a new character of the short-term load forecasting field of electric power system.Firstly, the principle, features, current status and development of the electric power system short-term load forecasting are generalized in this thesis. And then it makes a summary of many traditional and modern load-forecasting technologies, especially, introduces the application of ANN in electric power system short-term load forecasting. In order to improve BP algorithm’s efficiency, this paper gives several improved training algorithms used in BP neural network. Considering that the number of nerve cell in hidden layer, initial weight and unit’s bias value are the most important factors to the forecasting’s precision of ANN, genetic algorithm is used to choose a more reasonable frame of ANN. Genetic algorithm is good for deciding the proper fabric of net, and help the ANN to conquer it’s disfigurement. GA-BP algorithm makes use of the strongpoints of GA and BP algorithm, the results of the example show that GA-BP algorithm is better than BP algorithm only. At last, make weather as a input factor of BP neural network, the example of Kun Ming’s load forecasting indicates that this method has better definition and is more suitable for common condition.更多还原