Design Of Multi-Time Scale Electric Vehicle Charging Load Forecasting Model Based On Deep Learning

Power load forecasting technology has been a difficult and important research hotspot in energy power systems for a long time.The accurate power load forecasting model plays an important role in maintaining the safe,stable,and cost-effective operation of the power system.With the rapid development of technology,artificial intelligence and big data technologies are playing an important role in various industries such as communications,power,and manufacturing.The traditional power load types include personal residential power consumption,industrial and commercial power consumption,etc.The related load type prediction technology research has been approaching maturity,and good prediction accuracy has been achieved at the macro level.With the development of new energy vehicles,the traditional power grid faces the high power,strong sudden change,and high random power loads when large-scale new energy vehicles are charged,which puts great pressure on the grid system,especially the distribution network.The high-precision multi-time scale charge forecast of new energy vehicles is expected to alleviate the impact and impact of electric vehicles on the power grid and provide important support for the optimal planning of power systems and safe and economic operation.Electric vehicle charging load is different from traditional electric load.It has more random behavior.As the scale of electric vehicles increases and the amount of data increases,traditional models can no longer meet the current prediction accuracy requirements.In order to solve the challenge of electric vehicle charging load to the existing load forecasting technology,based on the time series depth neural network technology under the background of artificial intelligence,this paper proposes a charging load forecasting method for electric vehicles with different time scales.The main contents and innovations of this thesis are as follows:1.A long-term short-term memory network(LSTM)electric vehicle charging load prediction deep learning model is proposed.The charging load data of Shenzhen Yonglian Co.,Ltd.’s Liuyue charging station for one year is used for datapre-processing and feature analysis.2.Aiming at the problem of safety monitoring of charging piles in charging station monitoring system,an Encoder-Decoder LSTM model of minute-level ultra-short-term electric vehicle charging load is proposed,and the experimental results are compared with five other artificial intelligence time series prediction algorithms(artificial neural network,recurrent neural network,gated recurrent units,stacked auto-encoder,bilateral long short-term memory).The proposed E-LSTM model is in three the average prediction error in this time step scenario is only 2%,which is about 9% lower than ANN;3.Aiming at the problems of charging station hour-ahead scheduling and day-ahead scheduling,the influence of many factors on charging load model is analyzed deeply.Specifically,because of the large time span of hourly load data,the variation law is difficult to capture,this paper chooses the five characteristics of charging time,holiday/weekday type,the peek\flat\valley charging period,real-time electricity price,and real-time temperature are used as input features to predict the hourly level and daily charging load.The computational redundancy brought by,introduces the attention mechanism into the Encoder-Decoder LSTM model,proposes an EA-LSTM hour-ahead/day-ahead charging load prediction model,and with the other three advanced sequence prediction algorithms(artificial neural network,recurrent neural network,long short-term memory)to compare experimental results.The proposed model has an average prediction error of 1.17% in the hour-level charge load prediction task,which is about 3% lower than that of ANN.In the daily charge load prediction task,the average prediction error is 8.73% in the four time-step scenarios.Compared to ANN,it is reduced by 2.74%.The high-precision charging load prediction method for electric vehicles proposed in this article has a decisive effect on flexible access,real-time monitoring of large-scale electric vehicle charging at the grid level,and day-ahead and short-term energy storage scheduling of thermal power units,and it is safe and economical for power system operation.Is of great significance.