Research On Electricity Prediction And Electricity Strategy Recommendation Based On Edge Computing

The extensive development of electrical energy demand,the reform of electricity system as well as the opening up of electricity market put higher demands on electricity system.As the premise of proving the high-quality and steady working of the electricity system,and the basis of formulating various electricity sale packages,accurate electricity prediction and reasonable electricity consumption strategy can improve the efficiency and operation safety of the power grid.In today’s rapid evolution environment of electric energy system,fresh problems fall on how to forecast electricity accurately.The development of edge computing and the rise of smart electricity meters provide conditions for accurate electricity prediction and reasonable power consumption strategy recommendation of power metering system.The improvement of real-time performance,data integrity and security can further energize the whole power grid cycle.In this paper,based on the edge computing power metering system,combined with the current most advanced machine learning and deep learning methods,the electricity prediction method suitable for edge computing and the power package recommendation that can be combined with the electricity prediction are mainly studied.Mainly completed the following work:(1)By combing and analyzing the related theories of electricity prediction and edge computing,aiming at the limited computing resources and strong real-time characteristics of edge computing environment,a linear regression online prediction model based on SGD optimization is proposed.According to the characteristics of linear regression algorithm and edge computing real-time data flow,the continuous feature discretization method is improved under edge computing.Then,in order to further improve the nonlinear expression ability of the online prediction model,the FTRLFM online prediction model based on edge computing is proposed by using FM instead of linear regression.The experimental results show that the proposed two online prediction models can adapt to the environment of limited computing resources under the edge computing,and at the same time,compared with the traditional linear regression method,they have a greater improvement in each index of measuring the prediction error of electricity.(2)In order to improve the ability of offline electricity prediction,using ensemble learning method,combined with XGBoost,LightGBM and Stacking stack framework,a hybrid Boosting offline prediction model based on Stacking is proposed.According to the characteristics of XGBoost algorithm,word2 vec is used to vectorize the discrete features.The experimental results show that the MSE index of the offline prediction model is 10.2%higher than that of the online prediction model.(3)In order to deal with abnormal data,combining with the characteristics of edge computing real-time data stream,a sparse anomaly sensing framework is proposed.In order to combine the strong real-time performance of the online prediction model with the strong ability of the offline prediction model,a time-varying combined prediction model based on sparse anomaly perception was proposed.The effectiveness of the sparse anomaly perception framework in the environment of abnormal data acquisition is verified through experiments.The results show that the MSE index is further improved by about 9.6%,and the time-varying combined prediction model successfully combines the advantages of online prediction model and offline prediction model.(4)In order to further capture the trend or change of power users’ electricity consumption habits and preferences in the power package recommendation,a package recommendation algorithm based on electricity prediction and AFM algorithm is proposed by combining the electricity prediction results and the AFM algorithm.The experimental results show that the recommended Top K precision can be improved under the appropriate K value.In summary,taking the electric metering system under edge computing as the background,this paper primarily studies electricity prediction,electric strategy recommendation,model selection,feature improvement and simulation experiments.The results point out that the model presented in this paper can decrease the prediction error under the environment of edge computing,and increase the recommendation precision,which has a certain reference significance.