Method and device for predicting charging demand of electric vehicle

Abstract

The invention discloses an electric vehicle charging demand prediction method and device, and relates to the field of energy management, and the method comprises the steps: obtaining the historical driving data of a vehicle and the historical behavior data of a driver; performing feature extraction on the historical driving data and the historical behavior data, and constructing a charging behavior portrait of the driver; determining the charging or discharging probability of the vehicle based on the charging behavior portrait of the driver and the current driving data of the vehicle; and predicting the charging demand of the power grid based on the charging or discharging probabilities of all vehicles served by the current power grid. Different drivers have different sensitivities to the electricity price and the residual electricity quantity, so that the charging behavior portraits of the drivers directly influence whether the vehicle selects charging or discharging or not, and also influence the charging or discharging time. Therefore, the probability prediction of charging or discharging is more accurate in combination with the driving data of the vehicle and the charging behavior portrait of the driver, and the dispatching of the power grid is more accurate.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of energy management, in particular to a method and device for predicting charging demand of an electric vehicle. BACKGROUND

[0002] With the continuous increase of the number of electric vehicles, the large-scale concentrated charging of electric vehicles in peak periods brings significant load fluctuation and scheduling pressure to the power system. As an important regulation and control means of the power system, demand response can guide users to adjust their electricity consumption behavior through price signals or incentive mechanisms, thereby achieving peak load shifting and balancing power supply and demand, and playing an important role in the interaction between electric vehicles and the power grid in the future. However, existing research and technical solutions usually only predict the demand response potential of users based on the state of charge, charging habits and simple price response of electric vehicles. Although this method can reflect the charging demand of the vehicle to some extent, it cannot fully depict the multi-dimensional characteristics of individual drivers. SUMMARY

[0003] The purpose of the present application is to provide a method and device for predicting the charging demand of an electric vehicle. Different drivers have different sensitivities to electricity prices and remaining power, so the charging behavior portrait of the driver directly affects whether the vehicle chooses to charge or discharge, and also affects the time of charging or discharging. Therefore, the combination of the driving data of the vehicle and the charging behavior portrait of the driver is more accurate in predicting the probability of charging or discharging, and more precise in scheduling the power grid.

[0004] To solve the above technical problems, the present application provides a method for predicting the charging demand of an electric vehicle, comprising:

[0005] obtaining historical driving data of the vehicle and historical behavior data of the driver, wherein the historical driving data includes driving mileage and state of charge data, and the historical behavior data includes facial image data and voice data;

[0006] extracting features from the historical driving data and the historical behavior data to construct a charging behavior portrait of the driver;

[0007] determining the charging or discharging probability of the vehicle based on the charging behavior portrait of the driver and the current driving data of the vehicle;

[0008] predicting the charging demand of the power grid based on the charging or discharging probability of all vehicles served by the current power grid.

[0009] On the other hand, the features of the historical driving data and the historical behavior data are extracted to construct a charging behavior portrait of the driver, comprising:

[0010] The historical behavior data is feature extracted to obtain basic attributes, behavior characteristics and psychological characteristics of the driver;

[0011] The basic attributes include a vehicle type, an operation time and an operation power of the vehicle in which the driver is located;

[0012] The behavior characteristics include a state of charge, a charging time and an electricity price of the vehicle;

[0013] The psychological characteristics include a range anxiety degree and an electricity price sensitivity, the range anxiety degree representing a relationship between the charging time and the state of charge, and the electricity price sensitivity representing a relationship between the charging time and the electricity price;

[0014] A charging behavior portrait of the driver is constructed based on the basic attributes, the behavior characteristics and the psychological characteristics.

[0015] On the other hand, the extraction process of the psychological characteristics includes:

[0016] Facial feature points in the facial image data are feature extracted, the facial feature points including eyebrows, eyes and a mouth;

[0017] The expression of the range anxiety degree is:

[0018] ;

[0019] wherein, is a range anxiety degree of an i-th driver, is a first weight coefficient of the range anxiety degree, is a second weight coefficient of the range anxiety degree, is a feature of the eyebrows, is a third weight coefficient of the range anxiety degree, is a feature of the eyes, is a fourth weight coefficient of the range anxiety degree, is a feature of the mouth, and is a fusion function;

[0020] The extraction process of the electricity price sensitivity includes:

[0021] The electricity price sensitivity and a charging or discharging probability are determined according to the state of charge, the charging time and the electricity price of the vehicle;

[0022] The expression of the electricity price sensitivity is:

[0023] ;

[0024] The expression of the charging or discharging probability is:

[0025] ;

[0026] wherein, is the probability of the driver choosing to charge or discharge under the current conditions, is the bias term, is the price coefficient, is the price change value, is the state of charge coefficient, is the state of charge of the vehicle at time t, is the price sensitivity of the i-th driver, is the individual price coefficient of the j-th driver, is the value of ∣ ∣max among all drivers, is the fusion function.

[0027] In another aspect, the psychological characteristics further include speech emotion, and the extraction process of the speech emotion includes:

[0028] extracting a mel-frequency cepstral coefficient from the collected speech data, the mel-frequency cepstral coefficient representing pitch and rhythm;

[0029] inputting the mel-frequency cepstral coefficient into a pre-trained convolutional neural network for emotion classification and quantification;

[0030] obtaining a speech emotion score output by the convolutional neural network.

[0031] In another aspect, after feature extraction is performed on the historical driving data and the historical behavior data and a charging behavior portrait of the driver is constructed, the method further includes:

[0032] nonlinearly mapping the charging behavior portrait by a neural network to obtain a low-dimensional feature space and an embedding vector;

[0033] calculating a similarity based on the embedding vector and a trained clustering center, and generating a soft assignment probability matrix;

[0034] generating a label as a clustering group label of the driver by taking a clustering index corresponding to a maximum probability in each row of the soft assignment probability matrix;

[0035] determining the charging or discharging probability of the vehicle based on the charging behavior portrait of the driver and the current driving data of the vehicle, including:

[0036] determining the charging or discharging probability of the vehicle based on the clustering group label of the driver and the current driving data of the vehicle.

[0037] In another aspect, determining the charging or discharging probability of the vehicle based on the clustering group label of the driver and the current driving data of the vehicle includes:

[0038] Based on the clustering labels and the current driving data of the vehicle, the potential probability distribution of transferable power is obtained by using conditional variational autoencoder modeling.

[0039] The expression for the conditional variational autoencoder is:

[0040] ;

[0041] in, Let R be the potential probability distribution of the transferable power, x be the driver's transferable power, c be the current driving data of the vehicle, and z be the latent variable. The decoder operation results are based on the driver's transferable power, the vehicle's current driving data, the clustering labels, and the latent variables. The decoder operation results are based on the driver's transferable power, the vehicle's current driving data, the clustering labels, and the latent variables. For the weights of the decoder, For encoder bias;

[0042] The vehicle's driving needs are predicted based on its current driving data and historical driving data.

[0043] The upper limit of the vehicle's battery capacity is determined based on the aforementioned driving requirements;

[0044] The probability of the driver's behavior is determined based on the driver's charging behavior profile;

[0045] The probability of charging or discharging the vehicle is determined based on the potential probability distribution of the transferable power, the upper limit of the vehicle's battery, and the probability of the driver's behavior.

[0046] The expression for the probability of charging or discharging the vehicle is:

[0047] ;

[0048] in, Let be the probability of charging or discharging the i-th vehicle at time t. The probability of charging or discharging for the i-th driver at time t. To obtain the minimum value, Let be the upper limit of the battery for the i-th vehicle at time t. for The expected value, R is the driver-transferable power. The driving data is at time t. Let be the clustering label for the i-th driver.

[0049] In another aspect, determining the battery upper limit of the vehicle based on the driving demand comprises:

[0050] predicting the predicted driving range and the minimum state of charge in h hours in the future according to the charging behavior portrait of the driver and the historical driving data;

[0051] The expression of the predicted driving range is:

[0052]

[0053] The expression of the minimum state of charge is:

[0054]

[0055] wherein, is the predicted driving range in h hours in the future, is the average value of driving range per unit time, is the minimum state of charge required for driving in h hours in the future, is a safety redundancy coefficient, is the average energy consumption of the vehicle, and C is the battery capacity;

[0056] determining the battery upper limit of the vehicle based on the predicted driving range and the minimum state of charge in h hours in the future;

[0057] The expression of the battery upper limit is:

[0058]

[0059] wherein, is the battery upper limit of the ith vehicle at time t, is the maximum limit of the charging facility type or the battery charging power, is the state of charge of the ith vehicle at time t, is the battery efficiency, is the scheduling time scale.

[0060] In another aspect, determining the behavior probability of the driver based on the charging behavior portrait of the driver comprises:

[0061] determining the individual charging or discharging probability of the driver, and the expression of the individual charging or discharging probability is:

[0062]

[0063] wherein, is the individual charging or discharging probability of the ith driver at time t, is a bias term, is a price interaction coefficient, ​​​​is the electricity price sensitivity of the ith driver, is the electricity price change at time t, is the anxiety coefficient, is the mileage anxiety of the ith driver, is the speech emotion feature coefficient, is the speech emotion score of the ith driver, is the state of charge coefficient, is the state of charge of the ith vehicle at time t, is the proximity coefficient, is the fast charging station proximity indicator, which quantifies the influence of fast charging station accessibility on driver response behavior, is the fusion function;

[0064] The expression of the fast charging station proximity indicator is:

[0065] ;

[0066] wherein, is the travel time of the ith driver to the nearest available fast charging station, is a preset time threshold;

[0067] The expression of the travel time of the ith driver to the nearest available fast charging station is:

[0068] ;

[0069] wherein, is the road edge length, is the travel speed, is the feasible path of the ith driver to the s-th fast charging station, is the nearest s-th fast charging station.

[0070] On the other hand, the charging demand of the power grid is predicted based on the charging or discharging probability of all vehicles served by the current power grid, including:

[0071] The charging or discharging probability of all vehicles served by the current power grid at the current time is adjusted based on the charging or discharging probability of all vehicles served by the current power grid at the previous time. The expression of the adjusted charging or discharging probability of the vehicles is:

[0072] ;

[0073] wherein, is the charging or discharging probability of the t-th vehicle at t+1 time, is the charging or discharging probability of the t-th vehicle at t time, is a gain coefficient, is the normalized excitation strength at time t, characterization of substitution;

[0074] The charging or discharging probability of the i-th vehicle at time t is adjusted according to the number of vehicles served by the power grid, and the expression of the adjusted charging or discharging probability of the i-th vehicle at time t is:

[0075]

[0076] wherein, is the adjusted charging or discharging probability of the i-th vehicle at time t, is the charging or discharging probability of the i-th vehicle at time t, is a group effect coefficient, is the total number of drivers at time t simultaneously requiring parameter response, and N is the total number of drivers;

[0077] The sum of the charging or discharging probabilities of all adjusted vehicles at time t is obtained to obtain the charging demand of the power grid at time t, and the expression of the charging demand of the power grid at time t is:

[0078]

[0079] wherein, is the charging demand of the power grid at time t.

[0080] To solve the above technical problems, the application further provides an electric vehicle charging demand prediction device, comprising:

[0081] a memory for storing a computer program;

[0082] a processor for executing the computer program to realize the steps of the electric vehicle charging demand prediction method.

[0083] The application provides an electric vehicle charging demand prediction method and device, relates to the field of energy management, and comprises the following steps: obtaining historical driving data of vehicles and historical behavior data of drivers; performing feature extraction on the historical driving data and the historical behavior data to construct a charging behavior portrait of the drivers; determining a charging or discharging probability of the vehicles based on the charging behavior portrait of the drivers and current driving data of the vehicles; and predicting a charging demand of a power grid based on the charging or discharging probabilities of all vehicles served by the current power grid. Different drivers have different sensitivities to electricity prices and residual electricity, so the charging behavior portrait of the drivers directly affects whether the vehicles choose to charge or discharge and also affects the time of charging or discharging, so the application is more accurate in predicting the charging or discharging probability in combination with the driving data of the vehicles and the charging behavior portrait of the drivers and is more accurate in scheduling the power grid. BRIEF DESCRIPTION OF DRAWINGS

[0084] ​​In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the prior art and embodiments. Obviously, the drawings described below are only some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor on the basis of these drawings.

[0085] Figure 1 A flowchart of a method for predicting electric vehicle charging demand provided by the present application;

[0086] Figure 2 A structural schematic diagram of a device for predicting electric vehicle charging demand provided by the present application. DETAILED DESCRIPTION

[0087] The core of the present application is to provide a method and device for predicting electric vehicle charging demand. Different drivers have different sensitivities to electricity prices and remaining power, so the driver's charging behavior portrait directly affects whether the vehicle chooses to charge or discharge, and also affects the charging or discharging time. Therefore, the present application combines the driving data of the vehicle and the charging behavior portrait of the driver to more accurately predict the probability of charging or discharging, and more accurately schedule the power grid.

[0088] In order to make the objects, technical solutions and advantages of the embodiments of the present application clearer, the following will combine the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0089] Figure 1 A flowchart of a method for predicting electric vehicle charging demand provided by the present application, the method for predicting electric vehicle charging demand comprising:

[0090] S11: obtaining historical driving data of the vehicle and historical behavior data of the driver, the historical driving data including driving mileage and state of charge data, and the historical behavior data including facial image data and voice data;

[0091] The driver multi-modal data includes vehicle driving data, historical charging behavior data and driver physiological behavior data. The driving mileage, speed and real-time state of charge data of the vehicle are collected by a vehicle networking collection module. The historical charging selection records including charging start / end time, charging power, charging pile type data and corresponding electricity price are collected by a charging pile interaction module. The facial image data and voice data are collected by an in-vehicle sensor.

[0092] S12: Feature extraction is performed on historical driving data and historical behavior data to construct a charging behavior portrait of the driver;

[0093] The feature extraction obtains basic attributes, behavior characteristics and psychological characteristics. The basic attributes cover the static background information of the driver. The behavior characteristics are derived from vehicle driving data and historical charging behavior, such as real-time state of charge, charging time and electricity price record. The psychological characteristics include the range anxiety degree calculated through facial image and FACS coding, the electricity price sensitivity based on historical charging selection and binary Logit model fitting, and the speech emotion score obtained by extracting the Mel frequency cepstrum coefficient from the speech data and identifying it through the convolutional neural network, thereby comprehensively depicting the decision motivation and internal state of the driver and laying a foundation for subsequent accurate assessment of demand response potential.

[0094] S13: Determine the charging or discharging probability of the vehicle based on the charging behavior portrait of the driver and the current driving data of the vehicle;

[0095] It should be noted that the driver driving the vehicle can charge at the charging pile, at which time the driver needs to pay, or the driver can discharge to the power grid. If it is determined whether the vehicle needs to be charged or discharged based only on the current electricity of the vehicle, it is not accurate enough.

[0096] Specifically, the charging behavior portrait of different drivers directly affects the charging time of the vehicle. If the driver is sensitive to the mileage, even if the vehicle continues to drive, it does not affect the normal operation, and the driver will choose to charge. At this time, only relying on the electricity of the vehicle to make a judgment, the mileage-sensitive driver is not accurate enough to predict in this way. If the driver is sensitive to the electricity price, he or she may choose to charge when the electricity price is low, and may not choose to charge or even discharge to the power grid when the electricity price is high. At this time, only relying on the electricity of the vehicle to make a judgment, the electricity price-sensitive driver is not accurate enough to predict in this way.

[0097] S14: Predict the charging demand of the power grid based on the charging or discharging probability of all vehicles served by the current power grid.

[0098] The total demand response potential of the group refers to the total value of the net potential of the power grid that can be safely dispatched after aggregating the individual demand response potential of all electric vehicle users and correcting the dynamic mechanism feedback and group interaction effect.

[0099] The application provides a method for predicting the charging demand of an electric vehicle, and relates to the field of energy management, which comprises obtaining historical driving data of the vehicle and historical behavior data of the driver; extracting features from the historical driving data and the historical behavior data to construct a charging behavior portrait of the driver; determining the charging or discharging probability of the vehicle based on the charging behavior portrait of the driver and the current driving data of the vehicle; and predicting the charging demand of the power grid based on the charging or discharging probability of all vehicles served by the power grid. Different drivers have different sensitivities to electricity prices and remaining power, so the charging behavior portrait of the driver directly affects whether the vehicle chooses to charge or discharge, and also affects the time of charging or discharging, so the prediction of the charging or discharging probability based on the driving data of the vehicle and the charging behavior portrait of the driver is more accurate, and the scheduling of the power grid is more accurate.

[0100] On the basis of the above embodiments:

[0101] In some embodiments, the historical driving data and the historical behavior data are extracted to construct a charging behavior portrait of the driver, comprising:

[0102] The historical behavior data is extracted to obtain basic attributes, behavior characteristics and psychological characteristics of the driver;

[0103] The basic attributes include the vehicle type, running time and running power of the vehicle driven by the driver;

[0104] The behavior characteristics include the state of charge, charging time and electricity price of the vehicle;

[0105] The psychological characteristics include the range anxiety and the electricity price sensitivity, the range anxiety representing the relationship between the charging time and the state of charge, and the electricity price sensitivity representing the relationship between the charging time and the electricity price;

[0106] The charging behavior portrait of the driver is constructed based on the basic attributes, behavior characteristics and psychological characteristics.

[0107] The multi-modal data of the driver includes vehicle driving data, historical charging behavior data and driver physiological behavior data; the vehicle driving data includes real-time state of charge data; the historical charging behavior data includes historical charging selection records and corresponding charging time and electricity price data; and the driver physiological behavior data includes facial image data and voice data.

[0108] The higher the range anxiety of the driver is, the more the remaining power of the driver when choosing to start charging, and the higher the electricity price sensitivity of the driver, the higher the influence of the electricity price on the charging or discharging.

[0109] In some embodiments, the extraction process of the psychological characteristics comprises:

[0110] Feature points in the facial image data are extracted, and the feature points include eyebrow, eye and mouth;

[0111] The expression of the mileage anxiety is:

[0112] ;

[0113] wherein, is the mileage anxiety of the ith driver, is the first weight coefficient of the mileage anxiety, is the second weight coefficient of the mileage anxiety, is the feature of the eyebrow, is the third weight coefficient of the mileage anxiety, is the feature of the eye, is the fourth weight coefficient of the mileage anxiety, is the feature of the mouth, is a fusion function;

[0114] The extraction process of the electricity price sensitivity includes:

[0115] The electricity price sensitivity and the charging or discharging probability are determined according to the state of charge, charging time and electricity price of the vehicle;

[0116] The expression of the electricity price sensitivity is:

[0117] ;

[0118] The expression of the charging or discharging probability is:

[0119] ;

[0120] wherein, is the probability of the driver choosing to charge or discharge under the current condition, is a bias term, is a price coefficient, is a price change value, is a state of charge coefficient, is the state of charge of the vehicle at time t, is the electricity price sensitivity of the ith driver, is the personal price coefficient of the jth driver, is the value of finding the maximum value of | in all drivers, is a fusion function.

[0121] ​The range anxiety is obtained by using the Dlib library to detect the facial feature points of the face image data, using FACS coding to extract the facial features, and inputting the facial features into a range anxiety function.

[0122] In some embodiments, the psychological features further include voice emotion, and the extraction process of the voice emotion includes:

[0123] Mel-frequency cepstral coefficients are extracted from the collected voice data, and the Mel-frequency cepstral coefficients represent pitch and rhythm;

[0124] The Mel-frequency cepstral coefficients are input into a pre-trained convolutional neural network for emotion classification and quantification;

[0125] The voice emotion score output by the convolutional neural network is obtained.

[0126] The psychological features further include the voice emotion score, which is obtained by extracting Mel-frequency cepstral coefficients from voice data and inputting the Mel-frequency cepstral coefficients into a convolutional neural network.

[0127] The "voice emotion" here specifically refers to identifying the emotional state of the driver by analyzing the acoustic features of the driver's voice. The system first extracts Mel-frequency cepstral coefficients from the voice data to capture key features such as pitch and rhythm, then inputs the Mel-frequency cepstral coefficients into a convolutional neural network for emotion classification and quantification, and finally outputs a score representing the current emotional state of the driver (such as anxiety or calmness), which will be used as a key psychological feature parameter to more accurately predict the driver's decision-making behavior in demand response.

[0128] In specific implementation, a voice collection module is also configured, and Mel-frequency cepstral coefficients are extracted from the collected voice data and input into a convolutional neural network to output a voice emotion score Vi∈[0, 1], which is used as a supplement to the psychological features.

[0129] The embodiment collects multi-modal data of the driver including vehicle driving data, historical charging behavior data, and physiological behavior data of the driver, breaking through the limitations of traditional methods that only rely on vehicle state (such as SOC) and simple charging records. Specifically, the physiological behavior data of the driver's facial expression and voice emotion are introduced, so that the system can directly or indirectly quantify the psychological features of the driver (such as range anxiety, price sensitivity, etc.), thereby constructing a multi-modal user portrait that truly reflects the decision-making motivation of the driver, and improving the accuracy of potential assessment.

[0130] In some embodiments, after feature extraction is performed on the historical driving data and historical behavior data to construct a charging behavior portrait of the driver, the following steps are further included:

[0131] The charging behavior portrait is nonlinearly mapped by a neural network to convert a low-dimensional feature space to obtain an embedding vector;

[0132] The similarity is calculated based on the embedding vector and the trained cluster center, and a soft assignment probability matrix is generated;

[0133] The cluster index corresponding to the maximum probability of each row of the soft assignment probability matrix is taken to generate a label as the cluster grouping label of the driver.

[0134] The charging or discharging probability of the vehicle is determined based on the charging behavior portrait of the driver and the current driving data of the vehicle, including:

[0135] The charging or discharging probability of the vehicle is determined based on the cluster grouping label of the driver and the current driving data of the vehicle.

[0136] The cluster grouping label generated by the system through the deep embedding clustering method is essentially an anonymous group number automatically assigned by the model. These labels take the form of numerical indexes such as "Group 0", "Group 1", "Group 2", etc., with each number representing a group of drivers with similar multi-modal feature patterns. The generation of these labels does not rely on artificial pre-set categories, but rather on the automatic division of features based on similarity after nonlinear dimensionality reduction of the user portrait by a neural network. The specific group characteristics corresponding to each number can only be interpreted by analyzing the original data of the users in that group. The core value of these cluster labels lies in serving as a conditional variable for subsequent modeling. In step S4, the system combines this label with real-time scenario features to enable the conditional variational autoencoder to generate differentiated potential probability distributions for different group characteristics, thereby achieving more accurate personalized evaluation.

[0137] The obtained multi-modal user portrait is input into the trained deep embedding clustering model, and a neural network is used for nonlinear mapping to convert a low-dimensional feature space to obtain an embedding vector. The similarity is calculated based on the embedding vector and the trained cluster center (such as 10 center points corresponding to 10 pre-set clusters) to generate a soft assignment probability matrix (each element in the matrix represents the probability that sample i belongs to cluster j). The cluster index corresponding to the maximum probability of each row of the soft assignment probability matrix is taken to generate a hard label as the cluster grouping label of the driver.

[0138] When training the deep embedding clustering model, the optimization target distribution and the soft clustering distribution are used as the objective function, and the objective function is:

[0139] ;

[0140] In the formula, is the optimization objective function; is the target distribution; is the soft assignment probability matrix. is a more explicit and extreme classification target generated by the model after mathematically processing (squaring and amplifying the difference, readjusting the scale) the initial classification result of the model itself, which is used to correct the parameters of the model itself in the next round of training, so as to make a more determined classification. is a more explicit and extreme classification target generated by the model after mathematically processing (squaring and amplifying the difference, readjusting the scale) the initial classification result of the model itself, which is used to correct the parameters of the model itself in the next round of training, so as to make a more determined classification.

[0141] The embodiment adopts the DEC method to process the user portrait, and overcomes the problems of dimension processing difficulty and insufficient expression ability of the traditional clustering algorithm (such as K-means) in processing high-dimensional and nonlinear characteristics. The DEC method first performs nonlinear mapping and dimension reduction on high-dimensional portrait data through a neural network, retains and strengthens the information in a low-dimensional deep feature space, and then performs clustering, which improves the expression ability of nonlinear complex characteristics, so as to realize more accurate user grouping.

[0142] A dynamic response potential distribution model is constructed, and the potential probability distribution of the transferable power of the driver under a given situation is obtained by using a conditional variational autoencoder under the condition of the current situation characteristic set and the clustering group label, and the individual demand response potential is calculated by combining the physical constraints of the battery and the individual response probability.

[0143] The individual demand response potential refers to the maximum schedulable power that a single electric vehicle driver is willing and able to adjust the charging behavior when the power grid issues a dispatch signal (such as a price incentive).

[0144] In some embodiments, the charging or discharging probability of the vehicle is determined based on the clustering group label of the driver and the current driving data of the vehicle, including:

[0145] The potential probability distribution of the transferable power is obtained by using a conditional variational autoencoder based on the clustering group label and the current driving data of the vehicle.

[0146] The expression of the conditional variational autoencoder is:

[0147] ;

[0148] wherein, is the potential probability distribution of the transferable power, R is the transferable power of the driver, x is the current driving data of the vehicle, c is the clustering group label, and z is the latent variable, is the decoder operation result of the transferable power of the driver, the current driving data of the vehicle, the clustering group label, and the latent variable, is the decoder operation result of the transferable power of the driver, the current driving data of the vehicle, the clustering group label, and the latent variable, is the weight of the decoder, is the bias of the encoder;

[0149] predicting a driving demand of the vehicle based on current driving data and historical driving data of the vehicle;

[0150] determining a battery upper limit of the vehicle based on the driving demand;

[0151] determining a behavior probability of the driver based on a charging behavior portrait of the driver;

[0152] determining a charging or discharging probability of the vehicle based on a potential probability distribution of transferable power, the battery upper limit of the vehicle, and the behavior probability of the driver;

[0153] An expression of the charging or discharging probability of the vehicle is:

[0154] ;

[0155] wherein, is the charging or discharging probability of the ith vehicle at time t, is the individual charging or discharging probability of the ith driver at time t, is the minimum value, is the battery upper limit of the ith vehicle at time t, is the expectation of is the transferable power of the driver, is the driving data at time t, is the clustering label of the ith driver.

[0156] The purpose of this step is to use the conditional variational autoencoder to predict the possible charging power adjustment range of the driver as a condition of the group to which the driver belongs and the current scene, and output a probability distribution instead of a single determined value. This essentially upgrades the answer to “how much power can he transfer” from a determined value such as “2 kilowatts” to a probabilistic description of “80% possibility between 1.5 and 2.5 kilowatts”, thereby quantifying the uncertainty in the prediction and providing more reliable and informative decision-making basis for grid scheduling. The transferable power refers to the size of the charging power that the driver can flexibly adjust (transfer) when receiving a request from the grid, temporarily reducing power consumption, pausing charging, or feeding the electricity in the vehicle back to the grid. In combination with the potential probability distribution, the physical upper limit of the battery, and the individual response probability, the final individual demand response potential is calculated.

[0157] In some embodiments, determining the battery upper limit of the vehicle based on the driving demand comprises:

[0158] predicting the predicted driving mileage and the minimum state of charge in the future h hours according to the charging behavior portrait of the driver and the historical driving data;

[0159] An expression of the predicted driving mileage is:

[0160] ;

[0161] The expression of the minimum state of charge is:

[0162] ;

[0163] wherein, is the predicted driving range in the future h hours, is the average value of driving range per unit time, is the minimum state of charge required for driving in the future h hours, is a safety redundancy coefficient, is the average energy consumption of the vehicle, and C is the battery capacity;

[0164] The upper limit of the battery of the vehicle is determined based on the predicted driving range in the future h hours and the minimum state of charge;

[0165] The expression of the upper limit of the battery is:

[0166] ;

[0167] wherein, is the upper limit of the battery of the i-th vehicle at time t, is the maximum limit of the charging facility type or the battery charging power, is the state of charge of the i-th vehicle at time t, is the battery efficiency, is the scheduling time scale.

[0168] The predicted driving range in the future h hours is predicted according to the user portrait and historical charging behavior data, and the physical upper limit of the battery is determined. This item does not consider the actions of the driver, but only considers the running condition of the vehicle.

[0169] In some embodiments, the behavior probability of the driver is determined based on the charging behavior portrait of the driver, including:

[0170] The individual charging or discharging probability of the driver is determined, and the expression of the individual charging or discharging probability is:

[0171] ;

[0172] wherein, is the individual charging or discharging probability of the i-th driver at time t, is a bias term, is a price interaction coefficient, is the electricity price sensitivity of the i-th driver, is the change of the electricity price at time t, is an anxiety coefficient, is the driving range anxiety of the i-th driver, is a speech emotion feature coefficient, is a speech emotion score of the ith driver, is a state of charge coefficient, is a state of charge of the ith vehicle at time t, is a proximity coefficient, is a fast charging station proximity indicator, used to quantify the influence of fast charging station accessibility on driver response behavior, is a fusion function;

[0173] The expression of the fast charging station proximity indicator is:

[0174] ;

[0175] wherein, is the travel time of the ith driver to the nearest available fast charging station, is a preset time threshold;

[0176] The expression of the travel time of the ith driver to the nearest available fast charging station is:

[0177] ;

[0178] wherein, is a road edge length, is a travel speed, is a feasible path of the ith driver to the s-th fast charging station, is the nearest s-th fast charging station.

[0179] The fast charging station proximity indicator is used to quantify the influence of fast charging station accessibility on driver response behavior. Real-time driver location information, traffic network data, and fast charging station location and state information are obtained, and the shortest path algorithm is used to calculate the travel time of the driver to the nearest available fast charging station. The travel time to the nearest available fast charging station is compared with the preset time threshold to obtain a binary fast charging station proximity indicator.

[0180] In some embodiments, the charging demand of the power grid is predicted based on the charging or discharging probability of all vehicles currently served by the power grid, including:

[0181] The charging or discharging probability of all vehicles currently served by the power grid at the current time is adjusted based on the charging or discharging probability of all vehicles currently served by the power grid at the previous time. The expression of the adjusted charging or discharging probability of the vehicle is:

[0182] ;

[0183] wherein, is the charging or discharging probability of the tth vehicle at t+1, is the charging or discharging probability of the tth vehicle at t, is the gain coefficient, is the normalized incentive intensity at t, characterizes the substitution,

[0184] The charging or discharging probability of the ith vehicle at t is adjusted according to the number of vehicles served by the power grid, and the expression of the adjusted charging or discharging probability of the ith vehicle at t is:

[0185] ;

[0186] wherein, is the adjusted charging or discharging probability of the ith vehicle at t, is the charging or discharging probability of the ith vehicle at t, is the group effect coefficient, is the total number of drivers at t who simultaneously demand response, and N is the total number of drivers;

[0187] The charging or discharging probabilities of all the vehicles at t are summed to obtain the charging demand of the power grid at t, and the expression of the charging demand of the power grid at t is:

[0188] ;

[0189] wherein, is the charging demand of the power grid at t.

[0190] The response willingness of the driver is dynamically updated according to the historical incentive effect. It changes the individual response probability from a fixed parameter to a dynamic variable that evolves with system feedback: if the previous price incentive successfully triggers the active response of the user, the system will appropriately increase its future response probability through this mechanism, so as to more accurately reflect the sustained impact of the incentive policy and the dynamic changes of the user behavior, so that the model has the ability of self-adaptive optimization. On the basis of the individual demand response potential, the behavior dynamics and system stability are considered, the user willingness is adjusted through the dynamic incentive feedback mechanism, and the system risk is avoided through the group interaction effect model, and finally the group total response potential more conducive to the safe dispatching of the power grid is aggregated. The normalized incentive intensity at T (when the parameter is applied, it can come from the electricity price / integral / preference, etc.)

[0191] For the dynamic mechanism feedback, it is specifically: according to the change of the normalized incentive intensity and the response probability obtained by the individual, the individual response probability at the next moment is adjusted.

[0192] For the group interaction effect model, it is specifically: in practical application, when a large number of electric vehicle users respond to the power grid scheduling instruction at the same time to adjust charging, a new load growth may be generated, and a'second impact' on the power grid is caused. In order to avoid this risk, the group interaction effect model is introduced, which dynamically adjusts the response potential of each individual according to the number of users in the group responding at the same time, to avoid excessive response causing the burden of the power grid.

[0193] Based on the summation of all individual demand response potentials corrected by the group interaction effect, the total group response potential that can be safely scheduled at the current time is obtained, and it is directly used as the basis for the load optimization decision of the power grid scheduling center.

[0194] Figure 2 A structure diagram of a tram charging demand prediction device is provided, and the tram charging demand prediction device comprises:

[0195] The memory 21 is used for storing a computer program.

[0196] The processor 22 is used for executing the computer program to realize the steps of the tram charging demand prediction method.

[0197] The tram charging demand prediction device provided in the application is introduced above, and will not be described here.

[0198] It should be further noted that in the present specification, the relationship terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply that there is any such actual relationship or order between the entities or operations. Moreover, the terms 'include', 'contain' or any other variants thereof are intended to cover non-exclusive inclusion, so that the process, method, article or equipment including a series of elements not only includes those elements, but also includes other elements not explicitly listed or inherent to such process, method, article or equipment. Without more limitations, the element defined by the statement 'including a' does not exclude the presence of another identical element in the process, method, article or equipment including the element.

[0199] Those skilled in the art will further realize that the mechanisms of the various examples described herein are capable of being implemented using any number of combinations of the described features. Accordingly, these examples are not limited to the mechanisms described herein, but rather, the intent is to cover all modifications and alternatives equivalent thereto. The preceding description of the examples is illustrative, and not restrictive. Many other examples will be apparent to those of skill in the art upon reviewing the above description. The scope of the examples should, therefore, be determined not with reference to the above description, but instead should be given to the appended claims, along with their full scope of equivalents.

[0200] The above description of disclosed examples is intended to be illustrative, and not restrictive. Many other examples will be apparent to those of skill in the art upon reviewing the above description. The scope of the examples should, therefore, be determined not with reference to the above description, but instead should be given to the appended claims, along with their full scope of equivalents.

Claims

1. A method for predicting electric vehicle charging demand, characterized in that, include: The vehicle's historical driving data and the driver's historical behavior data are acquired. The historical driving data includes mileage and state of charge data, and the historical behavior data includes facial image data and voice data. Feature extraction is performed on the historical driving data and the historical behavior data to construct a driver's charging behavior profile; The probability of charging or discharging the vehicle is determined based on the driver's charging behavior profile and the vehicle's current driving data. The charging demand of the power grid is predicted based on the charging or discharging probabilities of all vehicles currently served by the power grid.

2. The method for predicting electric vehicle charging demand as described in claim 1, characterized in that, Feature extraction is performed on the historical driving data and the historical behavior data to construct a driver's charging behavior profile, including: Feature extraction is performed on the historical behavior data to obtain the driver's basic attributes, behavioral characteristics, and psychological characteristics; The basic attributes include the vehicle model, running time, and operating power of the vehicle in which the driver is located; The behavioral characteristics include the vehicle's state of charge, charging time, and electricity price; The psychological characteristics include range anxiety and electricity price sensitivity. Range anxiety represents the relationship between charging time and state of charge, and electricity price sensitivity represents the relationship between charging time and electricity price. A driver's charging behavior profile is constructed based on the aforementioned basic attributes, behavioral characteristics, and psychological characteristics.

3. The method for predicting electric vehicle charging demand as described in claim 2, characterized in that, The process of extracting the psychological features includes: Feature extraction is performed on facial feature points in the facial image data, including eyebrows, eyes, and mouth; The expression for the range anxiety level is: ； in, Let i be the range anxiety level of the i-th driver. As the primary weighting factor for range anxiety, As the second weighted coefficient for range anxiety, Features of the eyebrows As the third weighted coefficient for range anxiety, Features of the eyes As the fourth weighted coefficient for range anxiety, Features of the mouth This is the fusion function; The process of extracting the electricity price sensitivity includes: The electricity price sensitivity and the probability of charging or discharging are determined based on the vehicle's state of charge, charging time, and electricity price. The expression for the electricity price sensitivity is: ； The expression for the probability of charging or discharging is: ； in, Given the probability that the driver will choose to charge or discharge under current conditions, For bias terms, For price coefficients, This represents the change in electricity price. The state of charge coefficient, Let t be the state of charge of the vehicle at time t. Let i be the electricity price sensitivity of the i-th driver. Let j be the personal price coefficient for the j-th driver. To find | among all drivers | The largest value, This is the fusion function.

4. The method for predicting electric vehicle charging demand as described in claim 2, characterized in that, The psychological features also include vocal emotion, and the process of extracting vocal emotion includes: Mel frequency cepstral coefficients are extracted from the collected speech data, and the Mel frequency cepstral coefficients represent pitch and rhythm; The Mel frequency cepstral coefficients are input into a pre-trained convolutional neural network for emotion classification and quantification. The speech emotion score output by the convolutional neural network is obtained.

5. The method for predicting electric vehicle charging demand as described in any one of claims 1 to 4, characterized in that, After extracting features from the historical driving data and historical behavior data to construct a driver's charging behavior profile, the method further includes: The charging behavior profile is nonlinearly mapped using a neural network to obtain a low-dimensional feature space and thus an embedding vector. The similarity between the embedded vectors and the trained cluster centers is calculated, and a soft assignment probability matrix is ​​generated. The clustering index corresponding to the highest probability is taken row by row of the soft assignment probability matrix to generate a label as the clustering label of the driver; Determining the probability of charging or discharging the vehicle based on the driver's charging behavior profile and the vehicle's current driving data includes: The probability of charging or discharging the vehicle is determined based on the driver's clustering label and the vehicle's current driving data.

6. The method for predicting electric vehicle charging demand as described in claim 5, characterized in that, Determining the probability of charging or discharging the vehicle based on the driver's clustering label and the vehicle's current driving data includes: Based on the clustering labels and the current driving data of the vehicle, the potential probability distribution of transferable power is obtained by using conditional variational autoencoder modeling. The expression for the conditional variational autoencoder is: ； in, Let R be the potential probability distribution of the transferable power, x be the driver's transferable power, c be the current driving data of the vehicle, and z be the latent variable. The decoder operation results are based on the driver's transferable power, the vehicle's current driving data, the clustering labels, and the latent variables. The decoder operation results are based on the driver's transferable power, the vehicle's current driving data, the clustering labels, and the latent variables. For the weights of the decoder, For encoder bias; The vehicle's driving needs are predicted based on its current driving data and historical driving data. The upper limit of the vehicle's battery capacity is determined based on the aforementioned driving requirements; The probability of the driver's behavior is determined based on the driver's charging behavior profile; The probability of charging or discharging the vehicle is determined based on the potential probability distribution of the transferable power, the upper limit of the vehicle's battery, and the probability of the driver's behavior. The expression for the probability of charging or discharging the vehicle is: ； in, Let be the probability of charging or discharging the i-th vehicle at time t. The probability of charging or discharging for the i-th driver at time t. To obtain the minimum value, Let be the upper limit of the battery capacity for the i-th vehicle at time t. for The expected value, R is the driver-transferable power. The driving data is at time t. Let be the clustering label for the i-th driver.

7. The method for predicting electric vehicle charging demand as described in claim 6, characterized in that, Determining the upper limit of the vehicle's battery capacity based on the aforementioned driving requirements includes: Based on the driver's charging behavior profile and the historical driving data, predict the estimated driving mileage and minimum state of charge for the next h hours; The expression for the estimated driving mileage is: ； The expression for the minimum state of charge is: ； in, This is the estimated driving distance for the next h hours. This represents the average distance traveled per unit time. The minimum state of charge required for driving for the next h hours. For safety redundancy coefficient, Where C is the average energy consumption of the vehicle and C is the battery capacity. The upper limit of the vehicle's battery is determined based on the estimated driving range and minimum state of charge for the next h hours; The expression for the upper limit of the battery is: ； in, Let be the upper limit of the battery capacity for the i-th vehicle at time t. Maximum limits for charging facility type or battery charging power. Let t represent the state of charge of the i-th vehicle at time t. For battery efficiency, The scheduling time scale.

8. The method for predicting electric vehicle charging demand as described in claim 6, characterized in that, Determining the driver's behavior probability based on the driver's charging behavior profile includes: The individual charging or discharging probability of the driver is determined, and the expression for the individual charging or discharging probability is: ； in, The probability of charging or discharging for the i-th driver at time t. For bias terms, This is the price interaction coefficient. Let i be the electricity price sensitivity of the i-th driver. Let be the change in electricity price at time t. Anxiety level coefficient Let i be the range anxiety level of the i-th driver. The emotional feature coefficient of speech. Let i be the voice emotion score of the i-th driver. The state of charge coefficient, Let t represent the state of charge of the i-th vehicle at time t. The proximity coefficient, This is an indicator variable for the proximity of fast charging stations, used to quantify the impact of fast charging station accessibility on driver response behavior. This is the fusion function; The expression for the fast charging station proximity indicator variable is: ； in, Let be the travel time for the i-th driver to reach the nearest available fast charging station. The preset time threshold; The expression for the travel time for the i-th driver to reach the nearest available fast charging station is: ； in, The length of the roadside. For traffic speed, Let be the feasible path from the i-th driver to the s-th fast charging station. This is the sth fast charging station recently.

9. The method for predicting electric vehicle charging demand as described in claim 6, characterized in that, Based on the charging or discharging probabilities of all vehicles currently served by the power grid, the charging demand of the power grid is predicted, including: The charging or discharging probabilities of all vehicles currently served by the power grid are adjusted based on the charging or discharging probabilities of all vehicles served by the power grid at the previous time step. The expression for the adjusted charging or discharging probabilities of the vehicles is as follows: ； in, Let be the probability of charging or discharging the t-th vehicle at time t+1. Let be the probability of charging or discharging the t-th vehicle at time t. This is the gain coefficient. Let be the normalized excitation intensity at time t. Representational substitution; The charging or discharging probability of the i-th vehicle at time t is adjusted based on the number of vehicles served by the power grid. The expression for the adjusted charging or discharging probability of the i-th vehicle at time t is as follows: ； in, Let be the adjusted charging or discharging probability of the i-th vehicle at time t. Let be the probability of charging or discharging the i-th vehicle at time t. This is the group effect coefficient. Let N be the total number of drivers simultaneously responding to parameter demands at time t, where N is the total number of drivers. The charging or discharging probabilities of all vehicles at time t are summed to obtain the charging demand of the power grid at time t. The expression for the charging demand of the power grid is as follows: ； in, Let t be the charging demand of the power grid at time t.

10. A device for predicting the charging demand of electric vehicles, characterized in that, include: Memory, used to store computer programs; A processor, configured to execute the computer program to implement the steps of the method for predicting electric vehicle charging demand as described in any one of claims 1 to 9.

Images