Font Size: a A A

Applications And Researches Of Data Mining Method In Short-term Load Forecasting

Posted on:2008-01-15Degree:DoctorType:Dissertation
Country:ChinaCandidate:Z Y WangFull Text:PDF
GTID:1102360242964328Subject:Power system and its automation
Abstract/Summary:PDF Full Text Request
Short-term load forecasting (STLF) is aimed at predicting system load over an interval of one day or one week. It plays an important role in power system planning and operation. Basic operating functions such as unit commitment, economic dispatch, fuel scheduling and unit maintenance can be performed efficiently with an accurate forecast. STLF becomes increasingly important since many countries have privatized and deregulated their power systems, and electricity has been turned intoa commodity to be sold and bought at market prices.With the development of the electrical information technology, lots of load related data are stored in each power company's databases. We are data rich, but information poor. How can we make use of these data and obtain some valuable knowledge to give us some instruction on how to increase the precision of STLF are an important task. Data mining is a process that uses a variety of data analysis tools to extract implicit, previously unknown, and potentially useful information from data. In this dissertation, some data mining methods are used to analyze the load data of Hangzhou, Zhejiang Province, China. Some useful conclusions are drawn, which help us to improve the work of STLF.The dissertation begins by introducing the concept and development of short-term load forecasting and the definition of data mining, summarizing popularly used data mining techniques and its applications in power system load forecasting. The main research works are as follows.The second chapter is devoted to the data cleaning of the historical electrical load data. Based on the statistical methods, we are capable of identifying the missing data and abnormal data. And then a grey interpolation approach based on forward and back grey prediction model is proposed to correct the missing part or abnormal part of the historical load data. This approach deduces the missing value, which can make the best of all information in time zone of missing point.Artificial neural networks (ANN) are more commonly used for load forecasting. However, there still exist some difficulties in choosing the input variables and selecting an appropriate architecture of the networks. A novel feature selection based ANN for STLF is presented in this paper. The fuzzy-rough sets theory further extends the rough set concept through the use of fuzzy equivalence classes and is presented as a tool to extract principal case attributes and determine the initial weights of ANN. In the sequel, the ANN module is trained using historical daily load and weather data selected to perform the final forecast. Domain knowledge is needed to decide the fuzzy membership function in the process of fuzzy-rough set based feature selection. While the knowledge is not available in some times, mutual information based feature selection method is proposed in this situation. To demonstrate the effectiveness of the approach, short-term load forecasting was performed on the Hang Zhou Electric Power Company (HZEPC) in China, and the testing results show that the proposed model is feasible and promising for load forecasting.An improved Case-based Reasoning (CBR) system is presented to solve STLF problem with the aid of Self-organizing Maps (SOM) and mutual information method. CBR is composed of the steps of case representation, indexing, retrieval, and adaptation, and the key idea in CBR involves the use of already existing knowledge about objects or situations to predict aspects of similar objects. SOM are trained as a cluster tool in order to organize the old cases with the purpose of speeding up the CBR process. Mutual information method is employed to measure the input attributes and determine the initial weights of the case. This method not only uses case-specific knowledge of past problems, but also uses additional knowledge derived from the clusters of cases and it provides a new way for selecting proper features and feature weights.The fifth chapter presents a procedure for determining typical load profiles (TLP) based on clustering methods and assigning a customer to a particular TLP based on classification rules, using actual measurements of industrial and commercial consumers' load profiles. Various approaches can be used for grouping customers that exhibit similar electrical behavior into customer classes. The preprocessed measured load profiles are clustered with different clustering algorithms and we compare the results by means of three indicators, k -means method and RIPPER method are selected for determining the representative clusters and the classification rules respectively. The results demonstrate the efficiency of the proposed procedure.A Multi-agent based substation load forecasting model is introduced in the sixth chapter. The load data and customers' data are first obtained and clustered into some classes. Each class has its own load profiles, and we make the estimation of the load of each class based on different approaches. Aggregating the prediction load value of each class, we get the overall consumed load of the substation. A multi-agent model is proposed, due to different class has different profiles and the ratio of the classes in each substation is different either.
Keywords/Search Tags:Data mining, short-term load forecasting, grey theory, fuzzy-rough set, mutual information, case-based reasoning, artificial neural network, multi-agent
PDF Full Text Request
Related items