【作者】 刘双；

【导师】 杨丽徙；

【作者基本信息】 郑州大学 ， 电力系统及其自动化， 2003， 硕士

【摘要】 负荷预测是电力系统规划和运行研究的重要内容，是保证电力系统可靠供电和经济运行的前提，是电力系统规划建设的依据。负荷预测的准确程度将直接影响到投资、网络布局和运行的合理性。 由于配电网中长期负荷预测，会受到很多不确定因素的影响。因此，到目前为止，还没有哪一种方法能保证在任何情况下都能获得满意的预测结果。故具体预测时，宜结合本地区的实际情况，选用多种预测方法，不同方法的预测结果互相校核，最终确定预测值。 本文首先对负荷总量和分布预测的方法进行了综述，接着对负荷总量预测中的灰色预测方法和组合预测方法进行了深入的研究，然后将负荷分布预测中的分类分区法应用到工程实践中，同时对仿真法的相关内容也进行了一定的探讨。最后，将各种预测方法用计算机加以实现，开发了一套实用的配电网负荷预测软件。 负荷总量预测属于战略预测，是将整个规划地区的电量或负荷作为预测对象，它的结果决定了未来城市对电力的需求量和未来城市电网的供电容量。负荷总量预测的结果对城市供电电源点的确定和发电规划具有重要的指导意义，是城网规划的重要依据。 通过对灰色理论预测方法建模机理的研究，找出了灰色建模的局限性并提出了改进的方法。通过对负荷原始数据序列的预处理及优化，增强了灰色预测对波动负荷数据序列的处理能力，大大提高了灰色预测方法的适用范围和预测精度。利用等维新息递推GM(1，1)模型进行预测，保证了预测能够较为充分地利用新信息，既克服了简单灰色预测法中数学模型固定不变的弊病，又利用了灰色预测法短期预测精度高的优点，能够满足中长期负荷预测的要求。经过改进之后，扩展了普通GM(1，1)模型的适应范围，提高了预测精度。利用实例将改进模型与普通GM(1，1)模型进行比较，证明改进模型具有比普通GM(1，1)模型误差小、精度高的优点。实例证明改进后的GM(1，1)模型是一种较好的预测方法。 针对常规组合预测模型采用对单个预测模型进行加权处理的方法存在的不足，即要求参加组合预测的误差能保持稳定，但电力负荷预测结果的误差往往是非均匀性的。提出了基于人工神经网络的组合预测模型，利用人工神经网络对复杂非线性系统的拟合能力，通过网络训练自适应地调整各种预测模型的权重。同时，为了避免用常规语言建立人工神经网络负荷预测模型存在的模型结构复杂，郑州大学工学硕士论文训练时间长等缺点，利用MAJLAB神经网络工具箱建立组合预测模型，该模型不仅编程简单、而且收敛速度快。实际算例表明，该方法可以大大提高负荷预测的精度，在电力系统规划中具有广泛的应用前景。 空间负荷预测是对规划区域内负荷的地理位置和数值大小进行的预测，它提供未来负荷的空间分布信息。只有确定了配电网供电区域内未来负荷的空间分布，才能对变电站的位置和容量，主干线的型号和路径，开关设备的装设以及它们的投入时间等决策变量进行规划。 由于空间负荷预测涉及大量的空间信息，地理信息系统(GcograPhichiformation System，GIS)可以为空间负荷预测的数据收集、处理和预测结果的表示提供一个良好的平台。将GIS引入空间负荷预测，可以极大地减少数据收集量，是空间负荷预测方法实用化的必要步骤。 针对国内土地使用的实际情况，在空间负荷预测中采用了分类分区法，该方法是在分类负荷总量预测的基础上，根据城市规划用地图，计算分类负荷平均密度;再由小区面积构成、小区负荷同时率及修正系数求得小区最终负荷。该方法具有基础数据易于获得、能适应城市规划方案的变化、灵活性强等优点。同时，就分类分区法在预测过程中存在的一些问题进行了恰当的处理和改进。 针对己有负荷预测软件在数据收集、统计，模型、方法选用，结果处理等方面存在的问题，在将传统、实用的常规预测方法用计算机加以实现的同时，结合目前负荷预测领域的一些新发展、新成果，在地理信息系统平台的基础上开发了一套实用的配电网负荷预测软件。本软件采用OLE自动化技术用vC对MaPInfo进行集成二次开发。整个软件主要由地理信息系统、数据库系统、计算程序和系统界面四部分组成。本软件可为城网规划提供分类负荷、分类电量、总负荷、总电量和负荷分布预测。对电量和负荷的预测都提供了回归模型、趋势外推模型、灰色模型和组合模型等。使用者可以依据预测模型提供的图形分析和误差分析从多个预测模型中选择预测精度高的模型预测结果作为最后的预测结果。负荷分布预测采用了分类分区法，具有基础数据易于获得、能适应城市规划方案的变化、灵活性强等优点。负荷分布预测在地理信息系统的基础上完成了计算和预测结果的显示。更多还原

【Abstract】 Load forecasting is an important research content of power system planning and running, and is a premise for reliable supplying and economic running, and also is the basis of power system planning and construction. Exact degree of load forecasting shall directly affect rationality of investment, network layout and running.Because the medium and long term load forecasting in distribution network is affected by many uncertain factors, up to now, no methods can obtain the satisfying forecasting results at all instances. So at the time of practical forecasting, choosing many forecasting methods according to the actual instance, and checking forecasting results of different method, finally forecasting results are confirmed.This thesis firstly gives a summarization for load gross and distribution forecasting. Then a thorough research into grey forecasting method and combined forecasting method in load gross forecasting is carried through. And the classified-divisional load density method in load distribution forecasting is applied to project, at the same time a discussion on correlative content of emulation method is gone along. Finally each method is realized with computer program, and practical software for load forecasting in distribution network is developed.Load gross forecasting belongs to stratagem forecasting. It makes the electric power or load of the whole planning region as forecasting object. Its results decide the urban demand for electric power and the supplying capacity of urban distribution network in the future. The results of load gross forecasting have important guidance significance for ascertaining the location of power supply and generating planning. It is the important basis of distribution network planning.Through the research into modeling mechanism of grey forecasting method, the shortages of grey mechanism are found and some improved measures are put forward. Through the pretreatment and optimization to historical load data, the ability of grey forecasting dealing with fluctuant load data is strengthened, and the application range and forecasting precision are also enhanced. By using equally dimensional, new information’s grey model for forecasting, new information is used in the forecasting, which not only overcomes the shortcoming that the math model is changeless in simplegrey forecasting method; but also makes use of the advantage of the high precision in-m-short term grey forecasting. So it satisfies the request for the medium and long term load forecasting. By improving, the applicable range of the common grey model is extended, and the precision of the common grey model is enhanced. The improved model is compared with the common grey model by calculation example, which shows that the improved model has the advantages of small error and high precision. The calculation example proves that the improved grey model is a good forecasting method.Aiming at the shortage of usually adopting linear combination based upon proportion for single forecasting model in combined forecasting model, namely requiring forecasting models errors that take part in combined forecasting to keep stabilization, but the forecasting results errors are not usually well-proportioned in power load forecasting. The combined forecasting model based upon artificial neural network is presented, which makes use of the shaping of artificial neural network for complicated non-linear system, automatically adjusts the proportion for forecasting models by network training. At the same time, to escape the disadvantages that the combined forecasting model based upon artificial neural network is established using conventional language, such as complicated in structure, longer in training. Establishing combined forecasting model using artificial neural network on the basis of MATLAB toolbox. This model is not only simpler in programming, but also quicker in convergence. The calculation example shows that this method can greatly improve forecasting precision, has extensive application perspective in power更多还原