Research And Application Of Short Term Load Forecasting Based On CEEMDAN-LSTM Model

Under the promotion of energy conservation and emission reduction in electricity demand and a completely free electricity market operation model,load forecasting has become an effective means of balancing the power system and the electricity market.Accurate short-term load forecasting plays an important guiding and regulatory role in the operation of power companies.Considering that the volatility,continuity,and randomness of power loads increase the complexity and difficulty of load forecasting,how to effectively use nonlinear and non-stationary historical load data for power load forecasting has become a key issue.This article combines signal processing,neural networks,and related optimization algorithms to propose a long-term and short-term neural network combination prediction method based on the improved whale optimization algorithm.The combination prediction model is used to conduct relevant research and application design for power load prediction.The main research content of this article is as follows:(1)The Control variates is used to select parameters to build a prediction model based on the long short-term memory(LSTM)network.Combined with the data in this paper,the prediction results are obtained through prediction analysis and compared with the prediction results of the multi-layer feedforward BP neural network.The experimental results show that the short-term power load prediction accuracy of the LSTM model is better than that of BP,verifying the feasibility and advantages of selecting the LSTM neural network to build a prediction model.(2)The Whale Optimization Algorithm(WOA)is introduced to improve the randomness of model parameters,and the LSTM neural network is optimized through the Whale Algorithm.To solve the problem of easily falling into local optima during the optimization process of the whale algorithm,a roulette wheel method was used to change the optimization method of individual whale populations to facilitate jumping out of local optima,resulting in the improved whale optimization algorithm IWOA.The performance of WOA and IWOA was tested using standard equations to verify that the improved whale optimization algorithm IWOA has better convergence performance than the WOA algorithm.(3)The LSTM short-term power load forecasting model based on the improved whale algorithm is proposed.The IWOA algorithm is used to optimize the number of neurons in the hidden layer,learning rate,time step,and iteration times of the LSTM algorithm.The parameters that minimize the fitness value of the model are obtained to establish the LSTM forecasting model.By comparing the prediction results of the LSTM model and the IWOA LSTM model through experiments,it was verified that the performance of the IWOA optimized model has been improved.To solve the problem of model training efficiency caused by the nonlinear and non-stationary input data of the model,this paper proposes to combine the adaptive noise complete set empirical mode decomposition method CEEMDAN with the IWOA-LSTM model.Using CEEMDAN to decompose the input data,multiple modal components with different features were obtained.Each modal component was separately used as the input for the trained IWOA-LSTM model,and a CEEMDAN-IWOA-LSTM combined prediction model was constructed.The experimental comparison of the performance of EMD and CEEMDAN,as well as the prediction results of EMDIWOA-LSTM and CEEMDAN-IWOA-LSTM,showed that the prediction accuracy of the combined prediction model reached 99.05%,and various prediction evaluation indicators were superior to other prediction models,with the best performance and effectiveness.(4)The combination prediction model proposed in this article has been applied to real prediction scenarios,and a smart energy management system has been designed and implemented,achieving very good results in practical applications in factory environments.The combined prediction model based on CEEMDAN-IWOA-LSTM proposed in this article has the advantages of high prediction accuracy,excellent prediction speed,and strong generalization ability for industrial energy consumption.It has excellent practical application value for energy management of industrial users.