【作者】 任海军；

【导师】 孙才新；

【作者基本信息】 重庆大学 ， 电气工程， 2010， 博士

【摘要】 配电网短期负荷预测是电力运营和调度部门的一项重要日常工作,负荷预测精度高低直接影响到电力系统运行的经济性、安全性和供电质量;中长期负荷预测是配电网规划的前提和基础,其准确性直接关系到规划方案的水平和质量,对配电网规划具有重要的指导意义,也是实现电力系统管理现代化的重要内容之一。本文综合运用多种智能计算方法,主要从预测工作的各个环节为线索,对短期负荷预测神经网络日最大负荷预测输入参数的选择、短期负荷多变量正则化回归混沌局部预测模型及方法,中长期组合预测模型和空间负荷预测模型的建立各方面都作了深入的研究,丰富了配电网负荷预测方法。针对短期负荷预测输入参数的选择问题,引入了基于概念格的属性约简算法,提出了基于概念格的神经网络日最大负荷输入参数选择算法,约简属性并选择与待预测负荷量相关性大的属性参数作为神经网络预测模型的输入,充分保证了预测模型输入参数的合理性,为提高神经网络模型的预测精度提供了保证,并且能有效地减少神经网络模型的计算工作量。配电网系统是一个非线性系统,并表现出一定的混沌行为,为了能够重构出更为准确的相空间,提高短期负荷预测的精确度,选取对负荷影响程度最大的实感温度因素,构建了多变量时间序列。首先采用互信息法和最小预测误差法确定时间延迟和嵌入维数,并依据重构参数进行短期负荷多变量时间序列的相空间重构,针对局域预测法中邻近点个数少,不能满足最小二乘估计条件的问题,提出了基于正则化回归的多变量时间序列混沌局部预测模型,并利用该模型对实际的电力短期负荷进行了预测,提高了预测的精度。在分析了各种配电网中长期负荷总量预测方法基础上,提出了基于层次结构的中长期负荷变权组合预测方法。将组合预测中的最重要的权重确定问题按层次进行分解,构成一个递阶层次结构,从而将模型的评价指标进行综合考虑。从拟合误差指标选取了方差—协方差优选组合预测方法、发展相关性指标选取了灰色关联分析方法,二者分别来确定方案层到准则层的相对权重,再采用熵值法确定拟合误差和发展相关性这两个指标的相对权重,进而得到组合预测模型的最终组合权重,并将等维信息的概念引入到组合预测中,实现了变权组合预测。在分析了各种配电网空间负荷预测方法与的基础上,提出了基于概念格和元胞自动机模型的配电网空间负荷预测方法。在小区土地使用类型动态演变的过程中,采用元胞自动机理论对预测区域的各小区土地使用类型的进行动态模拟,并根据预测区域土地利用实际发展情况,确定元胞自动机规则的转换时间和迭代时间,使预测结果更合理,更加符合区域发展规划和实际的空间负荷分布发展变化规律。收集到各个小区土地使用决策规则后,引入概念格理论,采用概念格理论中的属性约简算法对影响规划地区土地使用类型的条件属性进行约简,得出约简后的小区土地使用类型的决策规则。在空间负荷预测的整个过程中,采用用地仿真法“自上而下”负荷分配的思想,将预测区域的负荷总量按照预测的空间分布分配到各个小区,从而实现了配电网空间负荷预测。更多还原

【Abstract】 Short-term load forecasting of distribution network is an important daily work for the dispatching and operating department of power system. The level of forecast precision directly impacts on the safety, economy and quality of power supply of power system operation. Mid-long term load forecasting is the prerequisite and foundation of distribution network planning, which accuracy directly counts for the level and quality of layout scheme, and it has instructive significance to reasonably layout distribution network. It is also one of the important contents of accomplishing power system management modernization.In this paper, many kinds of Intelligent Computing methods were synthetically adopted. As various aspects of the predicting work being the clues, the input parameters of daily maximum load forecasting to short-term load forecasting neural network were selected, and the models and methods of short-term load multi-variable regularized regression chaos partial forecasting, mid-long combination forecasting and spatial load forecasting are established, which enriched the methods of load forecasting of distribution network. The primary research was as follows:(1) As for the input parameter selection of short-term load forecasting, a property reduction arithmetic based on concept lattice was introduced, and input parameter selecting arithmetic of neural network was also put forward. The property was reduced, and we chosed property parameter that has good relativity to forecasting load as input of forecasting model of neural network. It completely ensured the rationality of input parameters of the forecasting model, and provides guarantee on the forecasting Accuracy of the neural network model, moreover, it could effectively reduce the amount of calculation for the neural network model.(2)Distribution system is a nonlinear system, showing some certain chaotic behaviors. The multivariate time series were constructed, in order to reconstruct the more accurate phase space, and to enhance the precision of short-term load forecasting. We chosed the effective temperature factors with the greatest impact on the load. Firstly, time delay and embedding dimension were confirmed with the methods of mutual information and the minimum predicting error. Secondly, according to reconstruction parameters, the phase space of short-term load multivariate time series was reconstructed. Thirdly, aiming at few neighboring points in the partial predicting method that can not satisfy least-square estimate condition, multivariate time series chaos partial forecasting model based on the regularized regression was presented. Moreover, such model was carried on in practical power load forecast, and the forecasting accuracy was enhanced.(3)Based on analyzing all kinds of forecasting methods for mid-long term load amounts of distribution network, the changeable combination forecasting method based on hierarchy structure was presented. Firstly, the weight determining is the most important issue. We decomposed it as hierarchy structure, and a recursion order hierarchy structure was formed. The evaluation indicator of models was synthesized considered. Secondly, the variance-covariance optimum combined forecasting method was selected from fitting error indexes and the grey association analysis method was selected from developing relative indexes, which separately determined the relative weight from the solution level to the principal level. Thirdly, adopting the entropy method, the relative weights of two indexes as fitting error and developing relative were confirmed, and then the final combination weight of the combination forecasting method was obtained. The conception of dimensional information was brought in the combination forecasting, and the changeable combination forecasting was achieved.(4)Based on analyzing all kinds of forecasting methods for spatial load of distribution network, the spatial load forecasting method for distribution network based on concept lattice and cellular automaton was presented. In the dynamic process of residential land-using types, the theory of cellular automaton was brought in to achieve the dynamic simulation of each residential land-using type. According to the actual state of development in planning areas, the conversion and iterative time of cellular automaton was confirmed, which made predicting outcomes be more reasonably and be in accordance with the actual developing discipline of spatial load in planning areas. In the process of obtaining the decision rules of residential land-using, the theory of concept lattice was introduced. Adopting the attribute reduction algorithm in the theory of concept lattice, all kinds of condition attributes affecting land-using types of planning areas were reduced. The decision rules of future residential land-using types were obtained.In the whole process of spatial load forecasting, the idea of the Land Usage Method for“from top to bottom”load distribution was adopted. With the load amounts forecasting in planning areas, its outcomes were assigned to each area, thus the spatial load forecasting in planning areas was accomplished.更多还原