Research On Home Electricity Consumption Prediction Method Based On Deep Learning

With the change of modern life style,the importance of household electricity load forecasting is increasingly significant.The continuous development of smart grid technology enables the power system to collect power load data more efficiently,providing a data basis for the research of artificial intelligence technology in power load forecasting.Household power load forecasting is one of the important research topics of smart grid technology.Accurate load forecasting has important practical significance for maintaining the stable operation of the power grid system.However,household electricity data has a variety of complex time series characteristics,which makes it difficult to fully extract and use sequence knowledge.Based on the characteristics of household electricity data,this paper improves the prediction performance of the model from many aspects.In view of the problem of insufficient accuracy of household electricity load forecasting caused by the low utilization rate of covariate information in the existing in-depth learning model,by focusing on the linkage relationship between different features and main variables,the expression ability of important features is enhanced,the disturbance of unimportant features on model training is weakened,and the ability of the model to grasp effective information is improved.In view of the low accuracy of household electricity load forecasting caused by the insufficient extraction of cycle features in the existing models,the attention of the model to the prediction location in the historical period of similar periods is strengthened,and the mining ability of cycle information is improved.The main research contents of this paper are summarized as follows:(1)In order to solve the problem of insufficient accuracy of household electricity consumption prediction caused by insufficient utilization of covariate information in existing models,this paper proposes a Feature-Attention Household Electricity Forecast Network(FANet)based on feature attention mechanism.The FANet model improves the residual connection of temporal convolutional network(TCN)and adds location code to TCN,Then the improved TCN is used for feature coding.Then,according to the principle of self-attention mechanism,a feature attention mechanism for the covariate matrix is proposed to strengthen the feature expression ability of important covariates by analyzing the linkage between covariates and main variables.Finally,the improved TCN is used to decode the intermediate representation to obtain the final output of the model,which solves the problem of low utilization of covariate information.(2)In response to the problem of insufficient accuracy in predicting household electricity load caused by insufficient mining of sequence period information in existing models,this article focuses on the obvious periodicity of household electricity load data.Based on the existing FANet models,the model has strengthened its ability to extract period information,A Jump Cycle Focus Household Electricity Forecast Network(JCFNet)based on skip cycle attention is proposed.The JCFNet model first uses the feature attention mechanism of the FANet model to enhance the expression ability of covariate information,and then inputs the enhanced higher-order features into the trend modeling component and the cycle modeling component for training,Using GRU as a trend modeling component to extract the long-term and short-term trend features of the sequence,CNN and Bi GRU as cycle modeling components,and transmitting sequence information across cycles,paying attention to the contextual information of the predicted position in the corresponding position of the historical cycle.Finally,using a fully connected layer to fuse trend and periodic features,and adding residual connections to avoid network degradation,the final output of the model is obtained.The JCFNet model solves the problem of difficult to accurately model periodic features and improves the prediction accuracy of the model by focusing on the similarity of historical periods across cycles.(3)Based on the Python framework,this paper designs and implements the FANet model and JCFNet model.In order to verify the validity and rationality of the model,this paper carried out a comparative experiment and a ablation experiment on the European household electricity data set provided by ENTSO-E.Through the comparative experiment,the validity of the FANet model and the JCFNet model on the household electricity data set was verified.Through ablation experiments,the rationality of FANet model and JCFNet model architecture design is proved.