The Data Mining Of Charging Piles And The Research Of Charging Capacity Prediction Algorithm

With good environmental protection,energy saving and other characteristics,electric vehicles(EV)have now become one of the development trends and hot spots.As a source of electric vehicle power supply,the electric vehicle charging station is an important supporting infrastructure for the development of electric vehicles.However,due to the lack of reasonable technical analysis guidance and operational deficiencies of electric pile enterprises,the development of the electric vehicle industry is restricted.Data mining technology technologies have been widely used in the field of intelligent travel and achieve expressive performance.This article aims to use the technologies to analyze the large amount of charging data accumulated during the operation of electric pile enterprises and provide guidance to management platform.There are three main research problems in this paper,and first is the intelligent prediction of the failure of the charging pile.By digging the fault records of the charging piles to establish a prediction model that then accurately predict the fault state of the charging piles at a certain period in the future,which can help maintenance personnel to more accurately locate the charging piles fault module and carry out accurate maintenance and repair.Second is user behavior analysis of electric vehicles.Using data mining technology to analyze the charging behavior of EV users can provide a reference for the charging pile operation management platform,which is conducive to improving the charging service of the charging management platform and the promoting of EV.Third is the intelligent prediction of the electricity quantity of the charging station.Predicting the user's power consumption is conducive to the charging pile management platform to do a good job of power consumption scheduling in advance,and it is also conducive to the protection of the power grid system.In this paper,the charging pile fault prediction adopts the method of machine learning.By training multiple machine learning,tuning model parameters and comparing the prediction results,the best prediction model are obtained.The charging behavior of electric vehicle users is analyzed through two data mining algorithms: Apriori and K-means.The K-means algorithm is applied to user clustering,and then the article analyzes the charging behavior characteristics of all categories in detail.Another algorithm is Apriori,and it is used to study the impact of weather and holidays on user charging.In addition,through statistical methods,the article analyzes the charging characteristics of the three typical areas and the relationship between temperature and charging,and then gives corresponding market suggestions.The individual pile group and the overall charging capacity have different influencing factors.For example,the individual pile group is more susceptible to the influence of local weather factors and seasonal changes and holidays,however the overall charging capacity curve has richer hidden features.In order to improve the accuracy of the prediction,two different methods are used to predict the single charging station and the overall charging capacity.For the prediction of the charging capacity of a single pile group,this paper uses an Encode-decoder model based on the Attention mechanism with a fully connected network to learn external feature,the experimental results show that the model has a good performance.In this paper,a three-layer network model is proposed for the prediction of the overall charge.The three-layer network is used to learn the three hidden features of the history,trend and periodicity of the charge curve,and the experimental results show that the model can achieve better prediction accuracy.