Research On Intelligent Optimization Based Power Load Forecasting Methods

Nowadays, in China, the total energy consumption is growing and the energy production and consumption pose a great challenge to environment. The central government and the government at all levels have recognized that electrical energy, a new clean energy, is the pillar of economic development. In recent years, the complicated interrelationship between power supply and economic growth makes the government, enterprises and a lot of experts begin to focus on the interrelationship between power consumption and demand prediction. Therefore, the prediction of electrical load has been an active topic in the field of management, electrical and computer science.In this thesis, by constructing mathematical model and application prototype system that is based on intelligent optimization algorithms, the trend of future electrical load in the long and mid-term is predicted and the situation of demand and supply of electrical industry is analyzed. This thesis also provides precautions for electrical load problems that might occurred within government and enterprises. The analyses can benefit the steady, fast and sustainable economic growth. The content and innovation of this thesis can be summarized as follows:1.Given the strong dynamic and nonlinearity of the short-term electrical load, an improved non-dominated Sorting Genetic Algorithm(NSGAⅡ) and the NSGAⅡ-CV-SVR optimized by cross validation is provided. First of all, the thesis takes the capability of NSGA Ⅱ-CV-SVR for fitting and forecasting nonlinear data, targets on the difficulties of identifying parameters on NSGA Ⅱ-CV-SVR, and adopts the improved non-dominated Sorting Genetic Algorithm Ⅱ, NSGA Ⅱ to optimize the parameters setting in SVR model. Therefore, the SVR model will be more accurate and can better forecast the short term electrical load.2.Based on the typical load pattern extraction and load pattern similar matching, a model and method for predicting the short-term load curve is proposed. Through multi-type clustering algorithm,the similar power users’ load curves can be divided into different groups and load pattern from different groups can also be extracted. By matching the different load model that is selected from different groups of models, the short-term load curve of users can be predicted. First, this thesis showed the model of classifying the load curve and the clustering algorithm that is often used. It also fixed the cluster validity indicator that is often used. Second, it proposed the predicting model and steps of the load curve which is based on matching similar load model. This part can be divided into four modules in terms of data collecting and dealing module, load clustering parameter setting module, load curve clustering and character extracting module and pattern matching and load predicting module. Third, this thesis conducted extensive experiments on 106 real load curves by K-means algorithm, fuzzy c-means algorithm and hierarchical cluster. The above-mentioned effectiveness of the patterns and method is also tested.3.By using the combining prediction of Gray model and BP-neural network, the medium-and-long term electrical load in varied environment can be predicted. On one hand, as the electrical load is a set of data with strong nonlinear and volatility, the predicting model of neural network can well represent its strong nonlinear calculating ability and provide a better fitting and predicting accuracy. On the other hand, as the electrical load is affected by varied factors, the prediction would lack of relative data. But the grey model can provide better prediction with little information and sample and uncertain environment. Based on the traditional grey model, this paper optimized the initial value, background value and the solution of the parameter. This thesis also improved the traditional BP-neural system through introducing the power factor, adaptive learning rate and upgraded genetic algorithm.4.The source software, Liferay, is utilized as the developing platform to predict electrical load through the prototype system. First, this thesis designed the whole system from several aspects, such as data access, structure and function. Second, this thesis carefully designed the function module and data of the system. Third, this thesis established a system through developing language, and explained the system from four modules, in terms of data preparation, data anomalies warning, data analyzing and presentation and electrical load forecasting.