An estimation and modeling method of electric vehicle charging load

Abstract

The application relates to an electric vehicle charging load estimation and modeling method, and is designed in view of the technical problems that existing similar methods rarely use CNN and LSTM to quickly acquire charging space-time characteristics of different time periods and different scenes, leading to low precision of model evaluation, and rarely use BP neural networks to fit the charging preferences of users and establish a smooth charging station charging route channel. The gist is that the method acquires the travel trajectory of the electric vehicle, excavates the travel mode and charging mode of the electric vehicle user, uses fusion CNN and LSTM to respectively extract space-time characteristics of the travel mode data and the charging mode data of different scenes and different time periods, and uses BP neural networks based on the above space-time characteristics to comprehensively realize the traffic network state to estimate the charging load in different time periods. Meanwhile, the method uses BP neural networks to realize charging load prediction of different scenes and different time periods in combination with the traffic state, and fits the charging preferences of users.

Description

TECHNICAL FIELD

[0001] The application relates to an electric vehicle charging load system and is an electric vehicle charging load estimation and modeling method. BACKGROUND

[0002] At present, an accurate electric vehicle charging load estimation model can provide reliable real-time charging prediction information for optimal operation and control of a distribution network and provide fine charging load space-time characteristic input for distribution network planning. Existing electric vehicle charging load technologies, such as the application number 201710952465.0 disclosed in the Chinese patent document, the application publication date is February 13, 2018, and the invention name is "a charging load prediction method for electric vehicles considering space-time distribution"; and the application number 202110978765.2 disclosed in the Chinese patent document, the application publication date is December 31, 2021, and the invention name is "a charging load prediction method for electric vehicles considering data correlation", and the application number 202011446332.4 disclosed in the Chinese patent document, the application publication date is March 26, 2021, and the invention name is "a fault diagnosis method based on CNN and LSTM deep feature fusion". However, in addition to the above methods, most of the current electric vehicle charging load estimation model research generally uses a single model, such as a Monte Carlo model to predict user travel behavior. This model is difficult to well reflect the diversity of different user charging demand characteristics, resulting in simulation errors. Therefore, it is necessary to conduct in-depth research on the travel mode of users and predict the charging demand of users in combination with the travel mode, rather than only focusing on the change rule of the electric quantity of electric vehicles. At the same time, the existing similar model is less likely to consider the randomness and dispersion of charging behavior in the existing electric vehicle penetration rate high scene. A CNN and LSTM electric vehicle charging load estimation model is proposed to realize the estimation of charging load in different time periods. SUMMARY

[0003] To overcome the above shortcomings, the purpose of the present application is to provide an electric vehicle charging load estimation and modeling method to the field, which mainly solves the technical problems that the existing similar method is less likely to use CNN and LSTM to quickly obtain the charging space-time characteristics of different time periods and different scenes, resulting in low precision of model evaluation, and less likely to use BP neural network to fit the charging preference of users and establish a smooth charging station charging route channel. The purpose is realized through the following technical solutions.

[0004] The application discloses an estimation and modeling method of electric vehicle charging load, which comprises the following steps: obtaining a travel trajectory of an electric vehicle, mining a travel mode and a charging mode of an electric vehicle user, adopting a CNN and an LSTM to respectively extract space-time features of travel mode data and charging mode data in different scenes and different time periods, and estimating charging load in different time periods based on the space-time features and a BP neural network and a comprehensive traffic network state.

[0005] Then, the charging access time is arranged in sequence with the travel mode data and the charging mode data, the travel mode data and the charging mode data comprise a travel mileage quantity at a moment before charging, a charging duration, a residual power at a moment before charging, a charging start time and a charging end time, and time features of a certain time period and a certain scene are extracted through the LSTM.

[0006] After the LSTM processing, the number of hidden layer units of the LSTM is n, and finally, an n*m-dimensional time feature vector is output from the last hidden layer:

[0007] After obtaining the original expression of the CNN and the LSTM, attention features are introduced to fuse the feature expression, and a fused space-time feature expression f map is formed.

[0008] Based on the formula 7, the following is obtained:

[0009]

[0010]

[0011] f map is the current expression of the space-time features of the travel mode and the charging mode.

[0012] The BP neural network takes the current traffic state and the space-time features of the travel mode and the charging mode as inputs to estimate the charging load of the electric vehicle in a certain time period and a certain scene.

[0013] ​​​​​The travel trajectory of the electric vehicle is obtained, and the travel mode and charging mode of the electric vehicle user are mined. The travel time, travel location, travel duration, residence time and residence location of the electric vehicle are obtained through the GPS point of the electric vehicle. The coincidence degree of the parking position and the charging station is combined to determine whether the user has charging behavior and the state of charge (SOC) of the electric vehicle battery when charging. The charging mode of the user is obtained, including the travel characteristics, charging method, travel behavior and charging behavior.

[0014] The GPS point of the electric vehicle obtains the travel time, travel location, travel duration, residence time and residence location of the electric vehicle based on the GPS data of the electric vehicle trajectory of the Internet of Vehicles platform, that is, the Internet of Vehicles platform records the electric vehicle trajectory data in the form of GPS. First, the electric vehicle trajectory data needs to be extracted, including the time and geographic information of the GPS point;

[0015] The GPS data is preprocessed. Since the GPS data is affected by random factors, the vertical distance between the GPS point and the road is certainly not the same. Then the GPS point needs to be preprocessed and reflected on the center line of the road. Combined with time and geographic factors, the LCSS algorithm is used to obtain the charging mode of the user, including the charging start time, charging end time and charging location;

[0016] (Formula 1);

[0017] Where, △T is the time accuracy, set to half an hour, δ(Li(u), Lj(v)) is a coincidence formula, when the charging stations of two users (user u and user v) coincide, the value is 1, otherwise 0; If COL is greater than 1 / 3, it means that the electric vehicle is charging in the same charging station. Assuming that there are n users charging within a time accuracy of △T;

[0018] Electric vehicle battery remaining capacity calculation. In the charging mode, the remaining capacity of the electric vehicle and the daily travel distance;

[0019] (Formula 2);

[0020] Where, Initial battery capacity, The remaining capacity at the previous time before charging. Then, through the method of big data, the average remaining capacity of all electric vehicles at the previous time before charging at a certain charging station is calculated every week, with half an hour as a statistical period.

[0021] (Formula 3);

[0022] The charging load statistics of a charging station at a certain time, the average duration of user charging from the start of charging at a certain delta T time accuracy to the end of the moment:

[0023] (Formula 4);

[0024] Wherein C represents the average capacity of n user batteries at the end of delta T time accuracy, when delta T is 30 minutes, i is considered to be 1 at the beginning of each delta T, and 30 minutes is considered to be the end of each delta T, and the average capacity of n user batteries at the end of each delta T is obtained. The charging efficiency of the charging station, Pc is the average power of the charging station.

[0025] The method is aimed at the randomness and dispersion of the charging behavior in the existing electric vehicle penetration rate high scene, proposes a fusion CNN and LSTM electric vehicle charging load estimation model, which effectively extracts the travel mode data, charging mode data of different scenes and different time periods, and extracts the space-time characteristics of different time periods, different scenes, electric vehicle driving characteristics, driving behavior (driving characteristics and driving behavior are collectively referred to as travel mode), charging method and charging behavior (charging method and charging behavior are collectively referred to as charging mode). Based on the above space-time characteristics, the BP neural network is used to comprehensively consider the traffic network state to estimate the charging load in different time periods.

[0026] The modeling method of the present application is scientific, the model precision is high, the charging preference of the user is met, the congested road section can be avoided well, the travel time is saved, and the charging load of the electric vehicle is reduced; It is suitable as an electric vehicle charging load estimation and modeling method, and a technical improvement of similar models and methods. Embodiment

[0027] The present application will be further described in detail by the specific implementation steps.

[0028] (1) Obtain the travel trajectory of the electric vehicle, and mine the travel mode and charging mode of the electric vehicle user.

[0029] The travel time, travel location, travel duration, residence time and residence location of the electric vehicle are obtained through the GPS points of the electric vehicle, the coincidence degree of the residence position and the charging station is combined, the charging behavior of the user and the state of charge (SOC) of the electric vehicle battery during charging are determined, the charging mode of the user is obtained, and the charging mode includes travel characteristics, charging method, travel behavior and charging behavior. At the same time, due to the complex change rule of the battery of the electric vehicle and the high nonlinearity of the data, it is difficult to establish an accurate mathematical model to describe it. The method describes the characteristics of the travel characteristics, charging method, travel behavior and charging behavior of the electric vehicle by fusing CNN and LSTM methods, and obtains the spatio-temporal distribution characteristics of the travel characteristics, charging method, travel behavior and charging behavior of the electric vehicle in different time periods, which provides data support for short-term prediction of charging load.

[0030] 1. GPS data extraction of electric vehicle trajectory based on vehicle networking platform, that is, the vehicle networking platform records the electric vehicle trajectory data in the form of GPS. First, the electric vehicle trajectory data needs to be extracted, including the time and geographic information of the GPS points.

[0031] 2. Preprocessing of GPS data. Since GPS data is affected by random factors, the perpendicular distance between GPS points and roads is definitely not the same. Then, the GPS points need to be preprocessed and reflected on the center line of the road.

[0032] The perpendicular processing of the GPS trajectory points of the user and the road center line can be regarded as the position of the user; at the same time, since the GPS points of the user will have a large number of points at the traffic light intersection or when the car is stopped, the foot of the road center line needs to be merged. If the distance between the points is less than 100 meters, the above points will be merged. The position of the merged point is at the center position of the two footprints, and the time of the merged point is the average value of the time of the two footprints.

[0033] 3. Combined with time and geographical factors, the LCSS algorithm is used to obtain the charging mode of the user, including charging start time, charging end time and charging location;

[0034] (Formula 1);

[0035] Wherein, △T is the time accuracy, which is set to half an hour, and δ(Li(u), Lj(v)) is a coincidence formula, which is 1 when the charging stations of two users coincide, otherwise it is 0; if COL is greater than 1 / 3, it means that the electric vehicle is charging in the same charging station. Assuming that there are n users charging within a time accuracy of △T.

[0036] 4. The remaining battery power of an electric vehicle is calculated. In the charging mode, the remaining battery power of an electric vehicle and the daily driving distance;

[0037] (Formula 2);

[0038] wherein, the initial battery power, the remaining battery power at the previous time before charging; then, by means of big data, the average remaining battery power of all electric vehicles at the previous time before charging at a certain charging station is calculated every week in a half-hour statistical period;

[0039] (Formula 3);

[0040] 5. The charging load of a charging station at a certain time is calculated. The average charging time of users from the start of charging to the end of charging at a certain △T time accuracy is obtained:

[0041] (Formula 4);

[0042] wherein, C represents the average capacity of n user batteries at the end of △T time accuracy, when △T is 30 minutes, i is considered to be 1 at the start of each △T, and 30 minutes at the end of each △T to obtain the average capacity of n user batteries at the end of charging; the charging efficiency of the charging station, and Pc is the average power of the charging station.

[0043] At this point, the driving distance, charging time, remaining battery power at the previous time before charging, charging start time, charging end time, and charging location of users in the entire city area before charging are calculated and obtained.

[0044] (2) The CNN and LSTM are fused to extract the space-time features of the travel mode data and the charging mode data in different scenarios and different time periods.

[0045] 1. The CNN is used to obtain the spatial features of a certain time period and a certain scenario. The original input is the data of the driving distance, charging time, remaining battery power at the previous time before charging, charging start time, and charging end time of all connected electric vehicles at a T time length and n samples (for example: n electric vehicle users of all large shopping center charging stations) before charging. After a series of CNN processing, a 1×n-dimensional spatial feature vector is obtained by n neurons: (Formula 5);

[0046] 2, the time sequence of charging access is arranged into travel mode data and charging mode data, the travel mode data and the charging mode data include the driving mileage number at the moment before charging, the charging time length, the residual power at the moment before charging, the charging start time and the charging end time, and the time characteristics of a certain time period and a certain scene are extracted through LSTM.

[0047] After the LSTM processing, the number of hidden layer units of the LSTM is n, and the last hidden layer outputs an n x m-dimensional time characteristic vector: (Formula 6);

[0048] After obtaining the original expression of the CNN and the LSTM, the attention feature is introduced to fuse the feature expression, and the fused spatio-temporal feature expression f map is formed. (Formula 7);

[0049] Based on formula 7, the following is obtained:

[0050] (Formula 8);

[0051] (Formula 9);

[0052] f map is the current expression of the spatio-temporal characteristics of the travel mode and the charging mode.

[0053] (3) A BP neural network is used to estimate the charging load of the electric vehicle in a certain time period and a certain scene, with the current traffic state and the spatio-temporal characteristics of the travel mode and the charging mode as inputs.

[0054] At this point, the charging load of each scene in the entire city field at each time (△T time accuracy) can be predicted according to the actual spatio-temporal characteristics and the traffic state. The invention uses the fusion of CNN and LSTM to extract the spatio-temporal characteristics of the travel mode and the charging mode data of a certain time period and a certain scene; since the charging data has randomness and periodicity, the use of CNN and LSTM can quickly obtain the charging spatio-temporal characteristics of different time periods and different scenes, and improve the accuracy of model evaluation. At the same time, this method uses a BP neural network to realize the prediction of charging load in different scenes and different time periods in combination with the traffic state, which is divided into four types: smooth, light congestion, congestion and severe congestion. The charging load prediction of the charging user and the power distribution network in a certain time period provides data support. This method combined with the traffic state can well fit the charging preferences of users, and in the case of congested traffic roads, users can avoid congested sections and charge at charging stations with relatively smooth sections, realizing the estimation of charging load in different time periods.

[0055] The above-mentioned convolutional neural network (CNN) is a kind of feedforward neural network containing convolution calculation and having a deep structure, is one of the representative algorithms of deep learning, has a representation learning ability, can perform translation invariant classification on input information according to its hierarchical structure, and is also called "translation invariant artificial neural network". The convolutional neural network is constructed by imitating the visual perception mechanism of biology, can perform supervised learning and unsupervised learning, the convolution kernel parameter sharing in the hidden layer and the sparsity of the interlayer connection enable the convolutional neural network to learn the grid features such as pixels and audio with a small amount of calculation, has a stable effect and has no additional feature engineering requirement for data.

[0056] The above-mentioned long short-term memory network (LSTM) is an extension of the recurrent neural network, is a kind of RNN, is more advanced than the ordinary RNN, and generally uses the LSTM when using the RNN, and now few people use the most basic version of the RNN because the LSTM has better effect. The long short-term memory network is suitable for learning important experiences with a long time lag in the middle, the unit of the LSTM is used as a building block of an RNN layer, the RNN layer is usually called an LSTM network, the LSTM enables the RNN to remember their input for a long time, the LSTM contains their information in the memory, which is very similar to the memory of a computer, because the LSTM can read, write and delete information from the memory.

Claims

1. An estimation and modeling method of electric vehicle charging load, the method comprises the following steps: obtaining the travel trajectory of the electric vehicle, mining the travel mode and charging mode of the electric vehicle user, extracting the space-time features of the travel mode data and charging mode data in different scenes and different time periods by fusing CNN and LSTM, and estimating the charging load in different time periods based on the space-time features and the traffic network state by using BP neural network; characterized in that: the travel mode data and charging mode data are arranged in chronological order according to the charging time, and the travel mode data and charging mode data include the driving mileage number at the moment before charging, the charging time, the remaining power at the moment before charging, the charging start time and the charging end time, and the time features of a certain time period and a certain scene are extracted by using LSTM; The CNN obtains spatial features of a certain time period and a certain scene. The original input is data of the driving mileage number of all the connected electric vehicles before charging, the charging time, the remaining power before charging, the charging start time, and the charging end time of n samples with a time length of T. After a series of processing of the CNN, a spatial feature vector of 1xn dimensions is obtained by n neurons: (Formula 5) based on formula 7, we get: After the LSTM processing, the last hidden layer output is a time feature vector of n x m dimensions when the number of hidden layer units of the LSTM is n: (Equation 6); After obtaining the original expression of CNN and LSTM, the attention feature is introduced to fuse the feature expression, and the fused spatio-temporal feature expression f is formed map The representation method of feature fusion is: (Equation 7); the BP neural network takes the current traffic state and the space-time features of the travel mode and charging mode as input to estimate the charging load of the electric vehicle in a certain time period and a certain scene. (Formula 8); (Formula 9); f map is the current expression of the space-time characteristics of travel mode and charging mode.

2. The method of estimating and modeling electric vehicle charging load according to claim 1, wherein the travel trajectory of the electric vehicle, the travel mode and the charging mode of the electric vehicle user are obtained by obtaining the travel time, travel location, driving time, residence time and residence location of the electric vehicle through the GPS points of the electric vehicle, combining the coincidence degree of the parking position and the charging station, determining whether the user has charging behavior and the state of charge (SOC) of the electric vehicle battery when charging, obtaining the charging mode of the user, and the charging mode includes driving characteristics, charging method, driving behavior and charging behavior.

3. The method of estimating and modeling electric vehicle charging load according to claim 1, wherein the GPS data of the electric vehicle trajectory based on the vehicle networking platform is extracted, that is, the vehicle networking platform records the electric vehicle trajectory data by using GPS, first, the electric vehicle trajectory data needs to be extracted, including the time and geographic information of the GPS points; 4. The method of estimation and modeling of electric vehicle charging load according to claim 3, characterized in that the GPS data is preprocessed, since the GPS data is affected by random factors, the vertical distance between the GPS points and the road is definitely different, then the GPS points need to be preprocessed and reflected on the center line of the road; combined with time and geographical factors, the charging mode of the user is obtained by using the LCSS algorithm, including the charging start time, charging end time and charging location; wherein, △T is the time accuracy, set to half an hour, and δ(Li(u), Lj(v)) is a coincidence formula, when the charging stations of two users coincide, the value is 1, otherwise 0; if COL is greater than 1 / 3, it means that the electric vehicle is charging in the same charging station, assuming that there are n users charging in a certain △T time accuracy; (Formula 1); electric vehicle battery remaining power calculation, in the charging mode, the remaining power of the electric vehicle and the daily driving distance; the charging load statistics of a charging station at a certain time, the average duration of user charging from the start of charging to the end of the time at the site: (Formula 2); wherein, initial battery power, the average residual power of all electric vehicles at a certain charging station at a time before charging; (Formula 3); ​ (Formula 4); Wherein, C represents the average capacity of n user batteries at the end of the △T time accuracy, when △T takes 30 minutes, each △T starting time considers i is 1, and each △T ending time considers 30 minutes to obtain the average capacity of n user batteries at the end; The charging efficiency of the charging station, Pc is the average power of the charging station.

Images