Short-term Load Forecasting Based On Competitive ISPO And Twin Support Vector Regression

With the introduction of electricity price competition mechanism, and the deepening of the reform of electricity market, power sector had put forward higher requirements for short-term load forecasting accuracy and speed. Twin support vector regression(TSVR), which is developed on the basis of support vector regression(SVR), has advantages of simple structure, strong generalization ability, global optimum and faster training speed. However, the selection of twin support vector regression parameters are blindly, time-consuming, laborious and unable to realize parameters for automatic selection, which seriously influence the learning and generalization ability of the algorithm. In view of these shortcomings, this paper puts forward a new algorithm of Competitive Intelligent Single Particle Optimizer(CISPO) to optimize parameters of twin support vector regression. CISPO overcomes the shortcomings of intelligent single particle optimizer(ISPO), which has bad directionality, chaotic search scale and a large number of uncertain parameters in the process of searching the optimal solution. The algorithm by introducing competition factors, making it has better global searching ability. And making the parameters adjust with each iteration automatically, to avoid the problem of parameter selection of intelligent single particle optimizer. The CISPO-TSVR model was set up, and CISPO is used to optimize the parameters of TSVR, which automate the parameters optimization, develop the generalization ability of TSVR and improve the global searching ability of intelligent single particle optimizer. Applying CISPO-TSVR model to the short-term load forecasting of power system can effectively improve the accuracy and speed of the short-term load forecasting. Main contents include:Firstly, this paper systematically expound the research significance, overseas and domestic research status quo of power system short-term load forecasting, and analyzed commonly used methods for short-term load forecasting as well as their advantages and disadvantages. Then it go into details of machine learning and statistical learning theory, understanding the specific content of support vector regression theory and twin support vector regression theory. Whereafter, short-term load forecasting based on twin support vector regression was studied. The main factors influencing the load in a region of Guangdong province was determined by analyzing the load characteristics of the region. Influence of abnormal historical data to predict as well as methods to eliminate them were analyzed. Steps of short-term load forecasting based on twin support vector regression were introduced and the influences of parameters selection for twin support vector regression result were analyzed. Next the intelligent single particle optimizer was in-depth studied. By means of overcoming the existing shortcomings of chaotic search scale and a large number of uncertain parameters, competitive intelligent single particle optimizer was put forward. And the specific methods and steps of short-term load forecasting based on twin support vector regression and competitive intelligence single particle optimizer were studied in detail. Lastly we used the proposed prediction model to forecast the short-term load of one region in Guangdong province. Studies have shown that comparing with back-propagation neural network, SVR, TSVR and the TSVR optimized by particle swarm optimization algorithm, CISPO-TSVR method can effectively improve the velocity and precision of load forecasting, and it is more suitable for development of short-term load forecasting.