【作者】 郑金；

【导师】 罗勇；

【作者基本信息】 郑州大学 ， 控制理论与控制工程， 2012， 硕士

【摘要】 电力系统的正常运行是关系到国计民生的头等问题,如何保证电力系统正常运行时现在及将来科研工作的主要课题。而负荷预测为电力系统正常运行提供了可靠的依据,尤其短期电力负荷预测具有重要的意义。负荷预测精度的好坏影响整个国家经济、社会、生活的正常进行。本文叙述了电力系统负荷预测的背景和研究意义,以及负荷预测的概念、特点等,并叙述了几种常见的预测评价指标。阐述了电力负荷的种类和规律。在此基础上,分析了指数平滑法、灰色预测模型和人工神经网络模型三种单一短期电力负荷预测方法。文中接着论述了组合预测方法的概念、特点及数学模型。叙述了多种固定权组合预测方法：等权平均组合预测、基于误差平法和最小的思想来确定权重系数以及它们的计算公式及步骤。重点分析了变权系数组合预测模型中的模糊变权系数组合模型方法及神经网络组合预测方法。为了克服了BP神经网络缺点,本文依次采用了基于小波神经网络和粒子群优化的小波神经网络两种组合预测方法,并引入模糊聚类分析的方法筛选小波神经网络的训练样本,使小波神经网络的训练更具有针对性。通过仿真验证了这两种方法预测精度比BP神经网络都有了改善,并且粒子群优化算法改善了小波神经网络预测精度。叙述了一种新的递归神经网络即回声状态网络的结构特点、学习过程及预测方法。并在已有对电力负荷混沌性分析的基础上,通过电力负荷时间序列相空间重构,提出了一种基于回声状态网络的短期电力负荷预测方法,通过实验仿真验证了这种方法在电力负荷预测中的有效性。并建立了一种短期负荷的回声状态网络组合预测模型,仿真验证了该方法的有效性。更多还原

【Abstract】 The normal operation of the power system is related to the national economy, the main topic of how to ensure the normal operation of the power system current and future research work.And power system load forecasting in order to ensure the normal operation of the power system, provides a reliable basis, especially short-term power load forecasting.The prediction accuracy of electric power system of good or bad influences electric power system and its economic value and operation safety of power supply quality. At present, there are many load forecasting model, both single model and combination prediction model.This paper describes the research meaning and background of short-term power system load forecasting and the concepts and characteristics of the load forecasting, analyzing the composition, characteristics and regularity of the power load, and citing several forecast evaluation index. On this basis, the method of time series and artificial neural network model of short-term load forecasting method is analyzed. Time series is divided into stationary time series and non-stationary time series, accordingly, the different specific time series forecasting method are proposed, and the steps of time series models is pointed out. The paper discusses the concept, features and mathematical model of the combination forecasting method. Several common fixed combination of prediction methods are pointed out, such as equally weighted average combination of forecasts, determining the weight coefficients based on the error level and the formula and steps are described. The combined model of fuzzy variable weight coefficient method and neural network combination forecasting method of variable weight coefficient prediction model are analyzed. In order to overcome the shortcomings of the BP neural network, two prediction methods are proposed on the base of and PSO(particle swarm optimization)-WNN(wavelet neural network).And fuzzy clustering analysis method is introduced to filer the training samples of wavelet neural network. The prediction accuracy of those two methods has been improved, and particle swarm optimization is very important. The structural characteristics network learning process, and prediction methods of the echo state networks in a new recurrent neural are described. In the light of chaos analysis of the power load,and on the basis of the power load phase space reconstruction, a short-term load forecasting method based on echo state networks is proposed. The effectiveness of this method in power load forecasting is showed through experimental simulation. And a short-term load combinations prediction model based on echo state network is proposed. Simulation results shows that this model is very good.更多还原