Research On New Method For Load Forecasting

The power system load forecasting based on the characteristics of the power load, looking for the law of load changes and building forecasting models in mathematical methods, estimates and projections the future power load change. In the power system, short-term power load forecasting is an important work of scheduling department and management department. Compared to the traditional short-term load forecasting methods, neural network has a strong convergence, Information memory capacity learning ability and nonlinear mapping ability. The main work of the paper is to study the neural network model applied to short-term load forecasting.First of all, the paper describes the characteristics of the power load and selects the input samples and output samples according to its similarity of data and week using some data pre-processing technology. Secondly, the paper studies the theory, the learning algorithm and the improving program of three typical neural network theories. It includes back-propagation neural network, radial basis function neural networks and support vector machine. Finally, the paper proposed a forecasting model of radial basis function neural network improved with particle swarm optimization.In the paper, short-term load forecasting models are proposed to predict the96time points'power load of next day in one city of Jiangxi province. Experimental results show that neural networks applied to short-term power load forecasting is feasible and practical and the model of radial basis function neural network improved with particle swarm optimization has the advantages of high forecasting accuracy, output stability and fast convergence speed, etc. In addition, the average relative error of the improved model less than1.7%, the forecasting accuracy is improved.