Research And Application Of Short-term Power Load Forecasting Based On Gated Recurrent Unit Neural Network

Electric energy plays an pivotal role in the development of country’s industry.Effective and accurate predict the electric load of factories can improve the stability,safety and reliability of their power systems.It has important practical significance for energy utilization and industrial development.It has important practical significance for energy utilization and industrial development.Improving forecasting accuracy is the core issue of industrial load forecasting.According to the peculiarities of industrial load,for the problems of single forecasting method and low precision at present,it is based on methods such as signal decomposition,deep learning and intelligent optimization.In this article,the short-term forecast of industrial consumer power is investigated.The main research contents of this paper are as follows:(1)The advantages and disadvantages of the Gated Recurrent Unit(GRU)and the Back-Propagation Network(BP)and the corresponding prediction models are compared and analyzed.The experimental results show that the GRU prediction results are more accurate and the prediction advantages are more obvious.Therefore,GRU is used as the basic prediction algorithm in this paper.Aiming at the problem that the prediction accuracy of the GRU model is affected by the randomness of the parameters set by human experience.The Sparrow Search Algorithm(SSA)is introduced to realize the automatic iterative search of the optimal parameters of the GRU model.The real power load data of an industrial user is used as experimental data,and the prediction results before and after optimization are compared.The results show that the prediction accuracy of the optimized model is 35.58% higher than the original model,and the prediction accuracy of GRU model is significantly improved.(2)According to industrial electricity data is non-linear and non-stationary,a Complementary Ensemble Empirical Mode Decomposition(CEEMD)is proposed.Comparing with Empirical Mode Decomposition(EMD),CEEMD is more suitable for short-term power load prediction.Then,a hybrid prediction model based on CEEMDSSA-GRU was constructed.The results showed that the prediction accuracy of the CEEMD-SSA-GRU model reached 99.36%,which was significantly higher than other models.And the three error evaluation indexes of MAPE,MAE and RMSE are the lowest,and the prediction effect is the best.(3)In order to better facilitate the management method of load data information for industrial production customers,an intelligent energy management system is designed and achieved in this paper.The CEEMD-SSA-GRU prediction model mentioned in this paper is used in the load prediction analysis control module of the system software.After the system software is used in the specific operation of industrial production customers,the prediction model in this paper has achieved remarkable practical results,which is very convenient for industrial users to arrange production planning and energy management.The CEEMD-SSA-GRU prediction model proposed in this paper has the characteristics of strong persistence,high accuracy and high prediction accuracy.Shortterm forecasts of industrial electrical loads and applications under consideration are of practical value..