Short-term Power Load Forecasting Based On Weighted Similarity And Weighted Support Vector Machine

The power load forecasting is related to the stability of the electricity market and therational allocation of power resources. It is of great significance for the entire power industry.In the short-term load forecasting, the dimension of load data is high, various influentialfactors are different, each sample exists difference and etc. In this paper, based on thetraditional similarity and supports vector machine, a weighted similarity and weighted supportvector machine algorithm is proposed to address the short-term power load forecastingproblem, and the main work is outlined as follows:(1) Adopt the weighted similarity method to select the similar days. The factorsincluding maximum temperature, minimum temperature, weather conditions, season, weekand date interval are considered in this study. Firstly, analyze the impact of various factors toquantify each factor, and then obtain the comprehensive factor of load data using principalcomponent analysis; Secondly, calculate the relative degree between all factors and the abovecomprehensive factor, and get the weight of various factors based on the relative degree, thenthe similarity is calculated by using the formula of weighted similarity; Finally, select thesimilar days in term of similarity.(2) Use the weighted support vector machine to predict the96-point load. In theShort-term power load data, there exist some differences among the sample data. Thesimilarity calculated by the weighted similarity not only reflects the relationship betweensample data and prediction day data, but also reflects the degree of importance of the sampledata. Therefore, add a weighting coefficient to the support vector machine model and selectthe similarity as the weight coefficients in order to resolve the difference of sample data.Compared with the traditional similarity and support vector machine method, the method hashigher accuracy rate. Finally, this paper studies holiday load forecasting, including statutoryholiday load forecasting and non-statutory holiday load forecasting, and obtain a holiday loadforecasting method. By case study, this method has good accuracy and effect.