Research On Electric Vehicle Charging Load Prediction Based On Data Driven

With the transformation of clean energy and the gradual maturity of battery technology,the large-scale popularization of electric vehicles has become an inevitable trend driven by both national policies and market effects.However,its large-scale disordered charging behavior can lead to numerous hidden dangers of power quality such as line overload,harmonic distortion,voltage imbalance,and so on;At the same time,with the rapid development of 5G communication technology and intelligent cloud terminals,multi-dimensional,high-quality large data sets based on EV have emerged as the times require.Therefore,using data driven methods to scientifically and accurately predict EV charging loads has important implications and significance for the safe,economic,and stable operation of China’s power system.Aiming at the difficulty of EV charging load association data mining and the difficulty of defining data specifications,firstly,complete EV charging load association data mining from three dimensions:vehicle inherent attributes,natural social attributes,and owner behavior preference attributes;Secondly,standardize the associated data from three dimensions:field name,field type,and field constraint;Finally,the basic theory of associative data preprocessing is elaborated from two dimensions:data quality processing and data specification processing,laying a theoretical and data foundation for subsequent EV charging load prediction models.In the residential area sequential charging load forecasting scenario,ignoring the spatial information of associated data and aiming at the difficulty in selecting similar day data for the forecast day,a similar day clustering model based on the CLARANS algorithm is proposed;Based on the clustering results,a bidirectional evolutionary GRU network model for collaborative learning of historical and future information is constructed;The analysis of numerical examples shows that the time series charging load prediction model for residential areas that combines the CLARANS algorithm and the bidirectional evolutionary GRU network model has strong applicability and advantages over the other five conventional network models in terms of prediction accuracy,fitting degree,interpretability,and training time.In the multi functional area spatiotemporal charging load forecasting scenario,using the spatiotemporal information of the associated data,an improved Floyd algorithm model that replaces the total travel distance with the total travel cost is proposed for the path planning problem between traffic nodes in the predicted area to complete the optimal travel path planning between any traffic nodes;Based on the optimal path planning,a user charging behavior decision-making strategy model integrating switching decision,mode decision,and load decision is constructed to simulate the charging behavior decision of users at each optimal path sub node,and ultimately traverse all EV driving paths to complete EV space-time charging load prediction.The analysis of a numerical example shows that the spatiotemporal distribution trend of EV charging load in different scenarios of the multi-functional regional spatiotemporal charging load prediction model that integrates the improved Floyd algorithm and user decision-making strategies are mutually verified with objective travel laws,demonstrating the feasibility and rationality of the proposed model.In summary,this article focuses on the prediction of EV timing charging loads in residential areas and EV spatiotemporal charging loads in multi-functional areas.Using multi-attribute association data information to drive corresponding models has achieved ideal prediction results,which will have a positive impact on the orderly scheduling and safe operation of the power grid.