Short-term Power Load Forecasting On Combination Methods

With the rapid development of economy, there is an increasing demand for electricity. The power load forecasting plays a more and more important role for ensuring safe and stable operation of the power system network reasonable planning and scheduling. How to improve the accuracy of power load forecasting is an important problem what is urgently needed to solve. In recent years, many experts has found and developed a variety of prediction methods for this. But the short-term power load forecasting is influenced by many factors, change law is extremely complex and highly nonlinear, so far no one recognized as the most appropriate forecasting methods. But a single forecasting method cannot have satisfied the requirements for accuracy. In order to improve the forecasting accuracy of short-term power load, this article puts forward a kind of process neural networks model by the empirical mode decomposition method and particle swarm optimization method with the combination of prediction methods.The pretreatment of the raw data has a great influence for the accuracy of power load forecasting. Short-term power load forecasting influenced by many factors, complex, strong nonlinear, which adds great difficulty to data modeling. And, therefore, in this paper, using the empirical mode decomposition method preferred to deal with the raw data. The original sequence to carry on the empirical mode decomposition (EMD) into several independent subsequence (the IMF). Each subsequence of strong cyclical regularity is relatively stable, improves the degree of executable for the prediction of the late.In view of the traditional neural network which is difficult to solve the problem of large sample of learning and generalization and unreflecting input process for the cumulative effects of time is difficult to satisfy the requirement of real time in the process of engineering practice. This paper builds process neural networks, It makes the system can directly process the type of data, and uses the orthogonal basis function expansion method solved the problem of the functional approximation and generalization of large sample. Neuron algorithm is easy to fall into local minima for process defects such as slow convergence speed, the particle swarm optimization algorithm is introduced into the process neural network training. Particle swarm optimization algorithm is a global optimization algorithm, the algorithm has strong adaptability characteristic such as concise and easy to implement good robustness, and has been applied successfully in many research fields, and has a deep intelligence background. But the basic particle swarm optimization algorithm has the problems about Premature convergence, poor optimization ability, slow late convergence speed. In the end. this paper designs the accelerate disturbance particle swarm (ADPSO).In the process of learning of the particles, which gives up the best experience learning,Only depends on the optimal value of the group. At the same time increase the disturbance factor to keep the diversity of particles to avoid local minima. Test results show that the improved particle swarm and then use this method to optimize parameters of process neural networks.Finally, in a certain area of Zhengzhou historical load data proves the combination forecast model to forecast, The simulation results show that the proposed combination of the short-term power load forecasting method is more accurate than the traditional power load forecasting. Timeliness also has obviously improved. This will help to improve the efficiency of power system that makes the forecast of power system more feasible.