The Data Mining Research Of Charging Control Platform For New Energy Vehicles

The research in this paper is mainly based on the charging stations data provided by a charging equipment company.The data of charging stations include charging process data and geographic information data of charging stations.The purpose of this paper is to predict the charging amount of charge stations,which includes the charging amount of single charging station and the charging amount of system.There are three purposes for research and prediction of charging amount.1)Let the charging equipment company adjust the price dynamically according to the amount of charging.For example,in order to encourage consumers to charge more and avoid a waste of resources,the company can lower the electricity price when the charge is low;They can increase the price of electricity when the charge is high which can earn more.2)When we find that the peak charging amount is expected to occur in the future,the company could purchase electricity from relevant departments in advance.3)For some "zombie stations " which have very little charging amount,new stations can be kept away from these stations.The prediction of charging amount in this paper is different from the general time series prediction.Firstly,there may be wrong date from the charging equipment company provide which should be removed in the early date processing.Secondly,charging amount of charging stations will fluctuate randomly,so it is difficult to predict the curve accurately.Finally,the curves of charging amount change regularly.We need to analyze the law of these curves and classify these curves to improve the accuracy of the prediction.The main work of this paper includes the following aspects.First of all,the charging equipment company provides us with data.We are responsible for receiving and simply processing data using big data technology.The main technology we use is Kafka +Spark Streaming + HBase.Then,this paper analyses the characteristics of some data,mainly about charging amount.The specific work includes: Using the spectrum diagram to analyze the charging curves and judge which one has periodicity;The correlation coefficient between temperature and charging amount is calculated,which shows that season has little effect on charging amount;Analyzing the influence of holidays on charging amount and the trend of charging amount curves.Next,we build a model to predict the charge amount.The prediction of charging amount is divided into the prediction of single charging station and the prediction of total charging amount.On the basis of the prediction of the single charging station,We classify single charging station before predicting.Non-periodic single station is directly used Encoder-decoder + LSTM model.For periodic single station,periodic prediction network is added on the basis of proximity model.On the basis of the prediction of the total charge(the whole country and several cities),we also incorporate the trend characteristics and the influence of holidays and the dynamic electricity price.Finally,we analyze and compare the effect of adding each module model to improve the prediction accuracy.And we also compare the prediction accuracy of our model and some other models,including HA(historical average),ARIMA and Xgboost.The main criteria are RMSE and MAPE.Experiments show that our model can effectively predict the charge amount and can improve the accuracy of the prediction.