Analysis On End Effects Of Local Wave And Its Application To Power System Short-term Load Forecasting

Power system is the key part of the healthy development of national economy and society, as well as being the infrastructure of people’s life. High-precision power load forecast data can provide effective information for electric power system scheduling, the installation of generating units, power grid planning etc, so as to guarantee the stable operation of the power system.However, in the influence of weather changes, social events, holidays and other factors, the power load data exists strong nonlinear and non-stationary, which will affect the accuracy of prediction results if modeling directly. In view of this, this paper combined with the local wave decomposition, particle swarm optimization algorithm and least squares support vector machine, proposing a hybrid forecasting method, to build the DEMD-LSSVM model based on the four-midpoint estimation method for short term load forecasting. The main idea is to use local wave method suppressing the end effect to decompose the original load, effectively smoothing load data, then to build LSSVM modeling prediction for each component of different frequency characteristics with the parameter optimization by PSO, finally to reconstruct components to obtain the forecast value.In essence, the local wave method can decompose nonlinear signals into a series of intrinsic mode function to reduce non-stationary of signals, which have different characteristic length as they are decomposed progressively according to these IMF’s respective fluctuation scale or trend. But the end effect of local wave method will reduce the quality of decomposition, and even produce false components, thus affecting the accuracy of the prediction model. This paper designs the simulation of some traditional methods on restraining end effect, such as extreme extending method, envelope linearity extension method, semi-periodical—semi-symmetrical extending method, waveform matching method and four-midpoint estimation method, comparing five evaluating indexes of each component and the reconstruction of main mode of components, to analyze the applicability and inhibitory effect of the five methods.Finally, take the load data of Xi Chang power grid station for example, which is comprehensive load under low-voltage, including hundreds of small hydroelectric power plants. It is more difficult to predict power load under this condition of weak power generation plan, the paper establishes the DEMD-PSVM and DEMD-LSSVM prediction model based on four-midpoint estimation method, and through the analysis and comparison of EMD-PSVM model and DEMD-PSVM model to verify the conclusion that four point estimate method inhibiting end effect can improve the accuracy of powerload forecasting to a certain extent. The DEMD-LSSVM model has higher prediction accuracy, and has certain advantages in the ability to capture the historical data model, in other words, the improved model’s generalization ability is excellent and practical. Then,aiming at the holiday power load prediction, the paper analyzes the DEMD-SVM,DEMD-PSVM and DEMD-LSSVM models, showing the model prediction error decreases one by one, and uses mean absolute percentage error and error distribution diagram to validate the DEMD-LSSVM model has strong adaptability and better accuracy in load forecasting, having certain advantages compared with other methods.Therefore, local wave decomposition and least square support vector forecasting model based on fourth point estimate method has certain research value and social significance,which provides a new research idea for short-term load forecasting of electric power system.