Font Size: a A A

Applications And Researches Of Data Mining And Artificial Intelligence Theory In Short-term Electrical Load Forecasting

Posted on:2006-02-18Degree:DoctorType:Dissertation
Country:ChinaCandidate:L FengFull Text:PDF
GTID:1102360152990827Subject:Electrical engineering, power system and its automation
Abstract/Summary:PDF Full Text Request
Short-term load forecasting (STLF) is an important task of power utilities, which is wildely used in the dispatching and operation planning of power systems, and the accurate load forecasting is helpful to the security and stability of power systems as well as to reducing the generation casts. With the establishment and development of the power matket, STLF will play more and more important role in power systems.In order to improve the accuracy and correlative of the historical data, to reduce the redundance of the input vectors of a neural network, and to optimize the structure of a neural network, the composing and characters of electric load are discussed, the influences of the correlative factors for STLF are analized, and the actuality and the existing problems of STLF are studied in this paper. The main research works are as follows.The load of power systems is an unsteady stochastic process. Among those observed values there may exist some "unhealthy data" due to the effect of various factors. These unhealthy data, participating the training of neural networks intermingled with normal data, badly affect the accuracy of load forecasting. Based on the traditional clustering algorithm - CURE, an improved outlier mining algorithm applied with entropy for electric load is proposed in this paper. It can identify and modify the unhealthy data effectively, and can provide useful information for control centers.Analyzing the regularity of historical load data, and the relationship between the electric loads and the outside enviorment factors, we have found that good load patterns can provide comprehensive and accurate historical samples. So a fuzzy classify system is proposed based on the multi-objective genetic algorithm in this paper, it solved the problem of dimensional disaster coming form the multidimentional input vectors, and paid attention to the accuracy and interpretative of the classify rules. Furthermore, association rule mining is used in the algorithm to reduce the spaces that the genetic algorithm is seeking for, which can improve the searching ability of the genetic algorithm, so better classification results are gained. The proposed algorithm can provid comprehensive and accurate training data for load forecasting, which can improve the accuracy of STLF, especially for the ananemous days with less historical data.There are so many factors that influenced STLF. How to justify and select the correlative factors is the key to improve the performance of load forecasting. A novel model based on the rough sets theory is proposed in the paper. It eliminates the redundant attributes by the attribute reduction algorithm in the rough sets theory, and uses the decision rules to deside the structure and initial weights of the neural network. For the crude domain knowledge applied in the neural network, the load forecasting model becomes clearer and more transpatent, and better performance of load forecasting is gained then.
Keywords/Search Tags:Short-term Load Forecasting, Data Mining, Artificial Intelligence, Information Theory, Clustering Analysis, Classify, Association Rule Mining, Rough Sets Theory, Genetic Optimum Algorithm, Fuzzy Classifying System, Fuzzy Nueral Network
PDF Full Text Request
Related items