Forecasting Of Electric Vehicle Charging Demand Based On Data Driven

The transformation of automotive industry to electrification has become a significant trend of sustainable social development.Electric vehicles,with the advantages of environmental protection and energy saving,will become an important direction for the development of automobile industry in the future.Especially in recent years,China attaches great importance to the development of electric vehicles,and the number of electric vehicles in numerous cities has increased significantly.The random charging demand brought by massive electric vehicles is a huge challenge to the safety and stability of power system.It is of great significance to forecast the electric vehicle charging demand accurately and effectively.With the advent of the era of big data,electric power companies in domestic major cities have established electric vehicle monitoring platforms to realize real-time monitoring of massive electric vehicles’ state of charge.The massive monitoring data contains a large amount of useful system information which makes it possible to forecast electric vehicle charging demand using data-driven method.As a starting point,this thesis takes charging demand data of urban district as the research object based on the measured electric vehicle charging demand data of a city in southwestern China.Considering the complexity of city districts’ power demand,a charging demand forecasting method based on time series decomposition and ensemble method and a forecasting model based on chaos theory and data dimension reduction are proposed by combining non-linear theory with machine learning method and considering different scenarios of single district and multi-district.The corresponding problems have been optimized to achieve a more efficiently and accurately short-term forecasting of electric vehicle charging demand.As to the research scenario of single urban district,a charging demand forecasting model based on time series decomposition and ensemble method is proposed in this thesis to analyze the problem of the complexity of the charging demand time series.Using the idea of divide and conquer,the original charging demand time series is first decomposed into several simple time series components by empirical mode decomposition.Secondly,considering large quantities of components are more likely to lead to the accumulation of errors and the complexity of the calculation,the fuzzy entropy is used to calculate the complexity of each component,and the components are classified and combined into new subsequences.Supported vector regression(SVR)and long short-term memory(LSTM)neural networks were used as base learners for the subsequences of different frequency.The input data is composed of the forecasting results of the base learner,weather data and the original charging demand time lags data with strong correlation.Finally,a fully connected neural network is trained to obtain the final forecasting result.Simulation results of one-step and multi-step forecasting based on measured data shows that the proposed method has higher forecasting accuracy than the traditional prediction model.In order to solve the problem that single district data cannot reflect the spatial correlation of charging demand among districts and lack of complex system information,this thesis researches the multi-district scenarios.Firstly,phase space reconstruction of multi-district charging demand time series is realized based on chaos theory,and then chaotic characteristics of the time series of each district are analyzed,which proves that the electric vehicle charging demand of urban districts has chaotic characteristics.Secondly,in view of the high dimensionality of phase space reconstructed data of multivariate time series,Laplace mapping is used to achieve data dimensionality reduction to eliminate redundant and irrelevant features and reduce the complexity of data dimension.Finally,the dimension-reduced phase space reconstructed data and the original data of the target district are used to form the model’s input,and the SVR model optimized by the differential evolution algorithm is used to realize the one-step and multistep forecasting.The experiments of the measured data show that the multi-district data contains more complex system information which is conducive to multi-step forecasting after phase space reconstruction.Laplace mapping can extract features that are beneficial to multi-step forecasting.Meanwhile,differential evolution algorithm is effective for parameter optimization of SVR model.Compared with other methods,the proposed method has better forecasting effect and has great value for improving the multi-step forecasting accuracy of electric vehicle charging demand.