Research And Application On The Intelligence Algorithm

In recent years, with the power system market, the load forecasting of electricity has become the essential part of an efficient power system planning and operation. Power system load forecasting refers to itself form the changes in load, the meteorological and the economy factors, through analysis and research of historical data to explore the internal change and contact of the power load, then development of the future electricity demand pre-estimates and projections. It is the power system security and economic operation of the base and is important to decisions in power systems. Because of the powers load forecasting is affected by the complexity of the historical load data and other random factors, thus making difficult to be accurately predicted. Comparison and analysis the advantages and disadvantages of traditional and artificial intelligence-based forecasting techniques, support vector machine is applied to short-term load forecasting in the thesis.Support Vector Machine is a novel machine learning method based on statistical learning theory. It has many merits such as simple structure, the global optimum and strong generalization ability. Also, it shows unique advantages in solving the small sample, nonlinear and other problems. The performance of SVM is determined by its parameters and kernel function, the bad setting will make the algorithm is poor. No uniform method of choose them, so usually use of artificially selecting and adjusting. Lack of theoretical guidance, the efficiency is poor. Differential Evolution (DE) algorithm is a novel intelligent optimization algorithm with many merits such as good performance of global optimization, fast convergence speed and good robust property, as a new algorithm, is not perfect in many aspects. A self-adapted DE is proposed to improve the performances, and used to search the parameters of SVM. In addition, SVM based on mixture kernel, which combined the advantage of global and local kernels, is proposed to solve the limitation of SVM based on single kernel.The applications of the support vector machine in the field of short-term load forecasting are comprehensively summarized in this thesis. Starting from the principle of support vector machine, the superiorities of it in short-term load forecasting are elaborated. In addition, some application problems, such as data pre-processing, kernel selection and parameters optimization, are analyzed in the thesis. Then a prediction model, based on mixture kernel function and the differential evolution algorithm, is forecasted for an area. Comparing with the artificial neural networks and the general support vector machine, it shows that the new model can improve the prediction accuracy, has an important practical value.