Short-Term Power Load Forecasting Based On Local Characteristic-scale Decomposition-Sample Entropy And Elman Neural Network

Short-Term Power Load Forecasting is the basis of power system planning, and the accuracy of Load Forecasting directly affects the safety, economy and stable operation of the power system. However, the Power Load is influ enced by many factors such as the economy, temperature, electricity price and etc., and it's essentially a nonlinear, non-stationary time series. Local characteristic-scale decomposition(LCD) is a new time-frequency analysis method, which can be used to decompose the complex non-stationary signal adaptively into the sum of single component signals-Intrinsic scale component(ISC),and it is very suitable for analyzing Nonlinear and non-stationary power load series. So, this paper proposes a short-term load forecasting model based on local characteristic scale decomposition which is for the first t ime applied in Load Forecasting.First of all, aiming at the nonlinear, non-stationary characteristics of power load, it improves the accuracy of load forecasting, b y using the LCD method to decompose the original power load series into a series of components of which instantaneous frequency has the physical meaning. And it carefully analyzes the influencing factors of each component, in order to better grasp the inhe rent law of the change of the load and to improve the accuracy of load forecasting.Secondly, in order to reduce the computational load and to improve the efficiency of load forecasting, it analyzes the complexity of decomposed components by using sample e ntropy theory, and it rebuilds new subsequences according to the sample entropy value. Comprehensive considering how each sub sequence is influenced by factors such as temperature and the type of date, different Elman neural network is established to predi ct each sub series respectively.Finally, Because the power load forecasting is a dynamic model of time, it's difficult to gain good prediction effect using static BP neural network. Therefore, dynamic Elman neural network is used to forecast the power load in the paper. At the same time, in order to validate the model of LCD-Sample Entropy-Elman prediction, three kinds of sample entropy prediction models —that is Elman, BP, LCD-Sample Entropy-BP—are established for short term load forecasting.The simulation results show that Elman neural network model has higher prediction accuracy than the static model of BP neural network, and verifies the validity of the analysis of LCD- Sample Entropy.