Electricity Load Forecasting Based On GM(1,1)Power Model Optimized By Particle Swarm Algorithm

Electric power is the pillar of the national economy, along with the rapid development ofelectrification in economy and people’s daily life, the demand for electricity consumptioncontinued to increase. Therefore, the power load forecasting technology which impactsthe decision-making becomes more and more important. The level of forecastingaccuracy directly impacts on whether the power sector could formulate reasonableeconomic power deployment plan, etc. Therefore, exploring relative effective ways toimprove the accuracy of power load forecasting has great practical significance.This paper based on the domestic and abroad study in load forecasting, grey predictiontheory and the particle swarm optimization. Depth studied the power load forecastingtechniques from experienced, classic and new technology three aspects. Then the paperstudied the grey prediction model, researched on the development and basic principles ofthe grey system theory. Detailed introduced the GM（1,1）model and grey Verhulstmodel, and analyzed the theoretical defects respectively. In order to solve the nonlinearoriginal series grey prediction problem, the concept of GM（1,1）power model isproposed, and depth studied its principles, procedures, parameters and the reasons toerror. Summarized through improving smoothness of the original data sequence, andoptimizing power model building background value, initial conditions, and exponentialto improve the prediction accuracy. In response to these problems, this paper processedoriginal data sequence with Cosine function, which could weaken the influence ofoutliers, improve the smoothness of the sequence and reduce reduction error. The particleswarm optimization algorithm was employed to optimize the parameters in GM（1,1）power model with the purpose of searching for the optimal parameters of the model,which could make up for the insufficient caused by given the parameters on experience.Finally, discussed the implementation process of the optimized model. The history loaddata of Beijing grid was employed to inspect the forecasting effect of the optimizedmodel. The numerical results and error analysis illustrated the model has a favorableprediction effect.