Research On Short-term Load Forecasting Method Based On Chaos And Wavelet Neural Network

The short-term load forecasting is the precondition and foundation for the electiric department to effectively develop the deployment plan, reasonably arrange the generator sets’ working, stopping and maintenance, and effectively ensure the power system’s steaty operation. So the study of the short-term load forecasting has great significance.Firstly, the actual load time series is reconstructed based on the chaos theory and phase space reconstruction theory. The C-C method is used to calculate the time delay τ and embedding dimension m, which are the important parameter of the phase space reconstruction. Then based on the load time series phase space reconstruction, we use the method with a small dataset to calculate the chaotic featrue of Lyapunov exponent, which can be used to distinguish the chaotic characteristic of the time series. Through the simulation, the chaotic characterstic of the load time series is proved, and the predictability is analysised.Secondly, the load time series phase space reconstruction and wavelet neural network (WNN) is combined to estabilish the short-term load forecasting model which based on the WNN and phase space. Then the design of the wavelet neural network is done. For comparative analysis, the short-term load forecasting model based on the Back Propagation neural network (BPNN) and phase space is established too.The simulation result shows the WNN model based on phase space has better prediction performance.Finally, after analyzing the shortcomings of the basic particle swarm optimization, three improved strategies to the basic particle swarm optomizization (PSO) are integrated and the comprehensive particle swarm optomizization (CPSO) is got. Then the CPSO’s performance is tested by some performance testing functions and the test results show that integrating three improved strategies is effective. At last, after analyzing the shortcomings of the traditional wavelet neural network based on the gradient decreased learning algorithm, the basic particle swarm optomizization (PSO) and comprehensive particle swarm optomizization (CPSO) are used to be the learning algorithm for the WNN model based on phase space. Then the PSO-WNN model and CPSO-WNN model are established and they are used in the short-term load forecastiong area. Through the simulation and comparative analysis, the wavelet neural network model based on particle swarm optimization can obtain higher prediction accuracy.