Research On Load Forecasting Based On Combination Method

With the rapid development of today’s economy, as well as the primary energy which are mainly fossil fuels gradually be exhausted and the continuous increase of the degree of environment pollution, the high quality of secondary energy-electrical energy plays a more and more important role in human life and production status. Power load forecasting is of great significance to ensure security and stable operation of power system and reasonable of power grid. Therefore, raising the precision of power load forecasting has great practical significance to ensure of reasonable and efficient use of existing power. Based on the in-depth study on the national industrial electricity consumption time sequence, the improved variable weight combination forecasting model is established in this paper based on Elman neural network and wavelet neural network.For the problem of electric power load data displays a non-linear feature and even chaos under the influence of multiple factors, Elman neural network method based on phase space reconstruction for industrial month electricity consumption in the whole society forecasting is adopted. By using small-data method to calculate the largest Lyapunov exponent, the chaos of load time series is judged, and the optimal delay time and the best embedding dimension are determined for phase space reconstruction. The topology structure of Elman neural network is determined. Finally, the measured data is brought into the chaos-Elman model for training. By forecasting and simulating the measured data, the results indicate that the model achieve better prediction effect, and validate the correctness and effectiveness of the combination of time series phase space reconstruction and Elman neural network.The wavelet neural network based on genetic algorithm industrial electricity consumption forecasting method is built after deep study on the nonlinear and uncertainty problems of electric power load data. The fitness function values of each individual in the method are calculated which based on the sum of the squared errors from wavelet neural network. Then the new population is generated through the selection, crossover and mutation operation. This process is repeated until the best fitness value corresponding to an individual is found out or a maximum iteration steps is attained. Finally, the optimized parameters are taken to the wavelet neural network to predict and simulate the national industrial month consumption data. Comparing the model with the wavelet neural network, the results showed that the method achieved good prediction efficiency and got higher prediction precision.Finally, an improved variable weight combination forecast model is proposed and built in this paper based on the advantages and disadvantages of each model. Verified by measured data, the improved combination prediction model plays the advantages of the above two kinds of forecasting models and improves the prediction accuracy.