Study On The Support Vector Machine Model And Integrated Model Of Monthly Load Forecasting

Load forecasting is a traditional topic in power system, with the fast development of modern electric power systems, the operation of power market requires high precision of load forecasting. With the development of new theory, there are more studies in theory and complemented methods of load forecasting. However, compared with short-term load forecasting, there's relatively lack of research on medium-term and long-term load forecasting which correspondingly has larger difficulty. For monthly load, the monthly load forecasting belongs to medium-term load forecasting, it possesses dual property of growth and volatility simultaneously, so it makes the load variations to possess complex non-linear combined characteristics. Therefore, it needs to establish appropriate forecasting model for monthly load own characteristics.Support vector machine (SVM) is a new machine algorithm based on statistical learning theory, it has some remarkable characteristics such as good generalization performance, the absence of local minima and fast computing speed. Thus it is successful to be applied to the load forecasting area. But there is no general theory and method on a proper setting of its parameters yet, which constrains the application in a certain extent.Aiming at the problem in parameters selection of SVM, an algorithm based on the chaotic particle swarm optimization was presented through the improvement of particle swarm optimization algorithm. Also, a golden section is designed to calculate the above model and divides the particle swarm into a good group and a poor group. What is more, the adjustable mechanism of the adaptive particle velocity is introduced, it can make proposed algorithm easily jump out of local optimum with effective dynamic adaptability and be more effective in searching the global optimal solution.Based on the analysis of monthly load characteristics, this paper identifies the type of input sample. Based on the mathematical statistics and three-point flat principle, an algorithm is presented to identify and correct anomalous data, it has the advantage of little man-made interference, simplicity, practicality and high efficiency. These two points above provide a preparation for accurate forecasting, finally a support vector machine model based on adaptive chaotic particle swarm optimization (ACPSO-SVM) for monthly load forecasting is established.Integrated forecasting model can make full use of the provided information in single forecasting model and give more appropriate or complete description of the natural law of load development, it will improve the prediction accuracy. The improved gray model which can reflect the growth characteristics well is used to model the annual development series in this paper. Combined with the support vector machine model which has good ability to describe the complex non-linear function, a integrated grey support vector machine model is established and its weight is optimized by the use of the improved particle swarm optimization. This integrated model is based on the relative correlative degree, thus it broadens the application of this degree.The above models are applied to the monthly load forecasting in power system and compared with conventional methods by the simulation examples. These results show that two models presented in this paper can improve the prediction accuracy to some extent, and they have better performance and significance in monthly load forecasting.