Short-time Load Forecasting Based On Phase Space Reconstruction And Neural Network

Short-term load forecasting is one of power load forecast, and it plays a very important role on power system economic and stable operation. Besides, it is also an important basis of load management and grid dispatching. Because the power short-term load data is affected by the weather, holidays, seasonal change, and temperature and so on, the discrete dates collected by power department appear complex nonlinear feature, which increases the difficulty of short-term load forecasting.Chaotic time series is a young subject developed in recent years, because of the advantage of discarding each factors affecting load forecasting, it becomes a hot area of studying forecasting. Neural network is an intelligent technology simulating human brain, and it has incomparable nonlinear processing power. Combining the two and using them to load forecasting causes more and more attention of researchers, this is also the subject of this paper.First of all, this article introduces various short-term load forecasting methods, the basic theory of chaos, chaos time series, and phase space reconstruction theory simply. According to the judgment methods of chaotic time series, this paper proves the chaos characteristics of short-term load, and calculates the load data’s delay time and embedding dimension used in later example.Secondly, this paper uses part method of chaotic time series prediction to construct weighted and one order approach step prediction model, and introduces RBF neural network to correct error for the predicted results. The example analysis shows that the model can reach a certain prediction precision, and decreases MRE for nearly20%after introducing error correction, which improves the performance of the model prediction.And then, this paper combines phase space reconstruction and RBF neural network to construct the RBF neural network forecast model. The example analysis shows that the method can achieve satisfactory accuracy. Finally, this article combines phase space reconstruction of chaos theory, Chebyshev polynomial with neural network combination to design a kind of single input Chebyshev orthogonal basis neural network prediction model which can approximate any nonlinear functions. And then, a multiple input Chebyshev orthogonal basis neural network dynamic prediction model, which can meet the demands of practical prediction, is constructed based on the previous model. Example analysis proves the model’s feasibility and its good prediction performance.