【作者】 卢峥；

【导师】 张帆；

【作者基本信息】 湖南大学 ， 电力电子与电力传动， 2007， 硕士

【摘要】 短期负荷预测是电力系统安全经济运行的前提,随着电力系统的市场化,高质量的短期负荷预测越来越显得重要和迫切。本文对电力负荷预测的各种方法进行了深入的研究,在总结现有研究成果的基础上,将卡尔曼滤波理论、人工神经网络和模糊控制等方法综合运用于短期电力负荷预测。短期电力负荷的预测要依靠大量的历史负荷数据,这些数据的有效性直接影响着预测结果的准确程度。因此在进行负荷预测之前,必须对选择的历史负荷数据进行补缺和去伪的预处理。本文把负荷值作为状态变量建立电力负荷的随机状态空间模型,采用卡尔曼滤波滤除历史负荷数据中的异常值,并在标准卡尔曼滤波的基础上将残差变化率引入到异常值的判别标准中,既能准确的判别出异常值又能避免误判,从而提高对负荷数据的滤波精度,为负荷预测提供更加有效的数据样本集。在电力负荷分析中,一般把负荷分为基本负荷分量和变动性负荷分量两部分。本文采用善于对非线性关系进行逼近的BP神经网络对基本负荷分量进行预测。针对BP算法存在的缺点及网络权值的修改在算法中的重要性,将卡尔曼预测的寻优原理与BP算法相结合,把神经网络中各神经元之间连接权值的学习作为扩展卡尔曼滤波中的状态向量进行最优估计,使其迭代收敛速度加快,缩短了神经网络训练的时间,同时也克服了容易陷入局部最小值的缺点。对于由天气条件、工作日类型、临时发生的重大事件等不确定因素引起的变动性负荷分量,由模糊逻辑构造各因素特有的隶属函数,建立模糊规则库,对神经网络预测得到的负荷基本分量进行修正。所以综合卡尔曼滤波、神经网络和模糊逻辑各自的优缺点,本文采用建立组合预测模型的方法来进行短期负荷预测,利用神经网络的学习能力,完成基本负荷分量的预测,并利用模糊逻辑对基本负荷分量进行修正。基于这种组合预测模型,以长沙市的历史负荷数据为例,进行了未来24小时短时负荷预测,仿真结果显示了这种方法的有效性和准确性。更多还原

【Abstract】 Short term load forecasting (STLF) is the precondition of economic and secure operation of power system,STLF with high quality is getting more and more important and exigent along with the marketable of power system. Base on the in-depth study of load forecasting methods and the sum-up of latest research productions, the theories such as Kalman filter, artificial neural network (ANN) and fuzzy logic are integrated into STLF in the thesis.Since a mass of historical load data is needed in STLF and the validity of it have a straight relation on the veracity of forecasting result, pretreatment with filling vacancy and getting rid of outliers is indispensable before forecasting. Load data are regarded as state variable and the stochastic state space model of power load is established in the thesis. The Kalman filter is used to eliminate the outliers, moreover the variety of residual difference is introduced to the distinguish standard of outliers. It made the determination more exactly and can avoid the mistaken judgement. From this way it can advanced the filtering precision and provide the most effective data swatch muster for STLF.Power load data are usually classified into elementary heft and mutative heft,To the former method of ANN is considered effective , the classical method for training a multilayer feed-forward ANN called back-propagation(BP) algorithm is used to forecast the elementary load in the thesis,which is good at approaching the nonlinear relations. The optimal estimation theory of Kalman filtering is integrated into BP algorithm, which estimate the network weights as the state vectors of the extended Kalman filter. As result the iterative convergent rate is quicken and the shortcoming of local convergence is improved. To the mutative load, fuzzy logic deals with the factors such as air temperature and holidays, etc,It modify the result forecasted by ANN. So base on the characteristics of Kalman filter, ANN and fuzzy logic, a combined load forecasting model is presented.The combined load forecasting model is used in application of the short-term load forecasting of Changsha power system. The case proves that the algorithm proposed this thesis is effective and veracity.更多还原