A charging anxiety behavior prediction system and prediction method

Abstract

The present invention relates to the technical field of charging behavior prediction, and more particularly to a charging anxiety behavior prediction system and method. The prediction system includes an acquisition unit, an image preprocessing unit, a vehicle remaining mileage monitoring unit, and a cloud storage computing platform. In the cloud storage computing platform, the driver's facial features and head deflection movements are input into a charging demand prediction model, which then outputs a prediction result of whether charging behavior occurs and records and stores it. The prediction result is compared with the remaining power data uploaded from the dashboard in real time to evaluate the accuracy of the prediction, continuously train and optimize the charging demand prediction model, and then use the trained charging demand prediction model to predict the driver's charging needs. The present invention solves the technical problem of predicting the charging demand of electric vehicles.

Description

Technical Field

[0001] The present invention relates to the technical field of charging behavior prediction, and in particular to a charging anxiety behavior prediction system and prediction method. Background Art

[0002] The widespread adoption of electric vehicles still faces numerous technical and market obstacles. First, due to the limited capacity of their onboard batteries, electric vehicles generally have a lower range than traditional fuel-powered vehicles. Second, due to limitations in battery technology, battery charging speeds are far slower than refueling fuel-powered vehicles. Finally, the development of supporting infrastructure for electric vehicles is slow. In current urban and intercity transportation networks, the number of electric vehicle charging stations is generally lower than that of fuel-powered vehicles. Limited range and low coverage of charging stations have led to range anxiety about pure electric vehicles, which has limited consumers' willingness to purchase them and owners' willingness to use them. Overall, range anxiety has become a major obstacle to the large-scale development of electric vehicles.

[0003] Range anxiety refers to the mental distress or anxiety that drivers experience when driving an electric vehicle due to concerns about insufficient range. Since range anxiety is prevalent among electric vehicle users, research on range anxiety is of great significance. To alleviate range anxiety, common methods include: 1) making the energy storage of vehicle battery packs more efficient; 2) expanding charging infrastructure and increasing output power to provide targeted charging solutions for homes, offices, convenience stores, and highway rest stops; 3) using existing facilities and technologies to provide extended range assistance for vehicles (such as commercially available PHEV models); 4) more efficient and lightweight vehicle platforms; 5) multi-modal auxiliary transportation systems and services, supplemented by private electric vehicles that can address urban and last-mile needs; and 6) battery replacement technology and the construction of battery swap stations. However, existing public technologies lack research on the prediction of drivers' charging anxiety behavior. Only by fully predicting drivers' charging anxiety behavior can the range anxiety problem be better solved. Summary of the Invention

[0004] In view of this, an object of the present invention is to provide a charging anxiety behavior prediction system and prediction method. The prediction method based on the prediction system can more accurately predict the charging anxiety behavior of the driver.

[0005] The present invention solves the above technical problems through the following technical means:

[0006] One aspect of the present invention is to provide a charging anxiety behavior prediction system, the charging anxiety behavior prediction system comprising:

[0007] An acquisition unit, which acquires the driver's facial image in real time through an on-board camera deployed in the electric vehicle;

[0008] An image preprocessing unit, communicatively connected to the acquisition unit, and configured to perform image preprocessing on the facial image;

[0009] The vehicle remaining mileage monitoring unit transmits data with the electric vehicle instrument panel to obtain the remaining power information on the instrument panel in real time;

[0010] The cloud storage computing platform is used to store the charging demand prediction model and receive the remaining power information of the dashboard in real time. It receives the pre-processed facial image and recognizes and judges the facial expression features and head deflection movements. The facial expression features and head deflection movements are used as input to the charging demand prediction model, and then the prediction result of whether there is charging behavior is output and recorded and stored. The prediction result is compared with the remaining power data of the dashboard uploaded from the vehicle in real time to evaluate the accuracy of the prediction, and the charging demand prediction model is continuously trained and optimized.

[0011] Furthermore, the image pre-processing unit is integrated into the in-vehicle processor and includes:

[0012] An image acquisition module, configured to receive the facial image transmitted by the acquisition unit, wherein the facial image is a facial image of the driver acquired at intervals of Δt1 by aiming a vehicle-mounted camera at the driver's upper body or face;

[0013] An image preprocessing module, in communication with the image acquisition module, for performing face recognition, data enhancement, normalization, and grayscale processing on the facial image;

[0014] The first communication module uses a CAN bus or SerDes to connect to the acquisition unit, and is used to transmit the facial image collected by the acquisition unit to the image acquisition module, and uses a 5G network to connect to the cloud storage computing platform to upload the preprocessed facial image to the cloud storage computing platform.

[0015] Furthermore, the cloud storage computing platform includes:

[0016] a second communication module, configured to receive the preprocessed facial image sent by the image preprocessing unit, and receive the instrument panel remaining power information sent by the vehicle remaining mileage monitoring unit;

[0017] The human expression recognition module is in communication with the image preprocessing unit, extracts features from the preprocessed facial image, and uses an SVM classifier to identify the driver's expression features to determine whether the driver has anxious expression features;

[0018] A head deflection recognition module is in communication with the image preprocessing unit and performs head deviation detection and line of sight deviation detection based on the preprocessed facial image. It is used to identify the driver's head deflection movement and detect whether the driver is looking down at the remaining mileage meter on the car instrument panel.

[0019] a charging demand prediction module, configured to input the facial expression features and the head deflection movement into the charging demand prediction model, and then output a prediction result of whether there is a charging behavior and record and store it;

[0020] The comparison module is used to compare the prediction result with the remaining power data on the instrument panel in real time to evaluate the accuracy of the prediction.

[0021] Another aspect of the present invention is to provide a method for predicting charging anxiety behavior, comprising:

[0022] The driver's facial image is collected through an on-board camera deployed inside the electric vehicle;

[0023] The facial image is transmitted to an in-vehicle processor, the driver's face position is located by using a histogram of oriented gradients, and data preprocessing such as data enhancement, normalization, and grayscale conversion is performed, and the preprocessed facial image is transmitted to a cloud storage computing platform;

[0024] In the cloud storage computing platform, based on the pre-processed facial image, the driver's facial expression features and head deflection movements are identified and judged;

[0025] The driver's facial features and head deflection movements are used as input into the charging demand prediction model to determine whether the driver currently has a need to charge. The judgment result is compared with the current remaining power of the electric vehicle. The prediction model is continuously verified and improved until the prediction result better matches the remaining power and the prediction accuracy is high enough. Finally, the trained prediction model and each prediction result are stored in the cloud storage computing platform.

[0026] Furthermore, the steps of identifying and judging the facial expression features are as follows:

[0027] Based on the preprocessed facial image, given an image d∈R containing m pixels m×1 , d(x)∈R p×1 , used to index the p feature points of the image, x represents the p feature points, h is the nonlinear feature function at each feature point, h(d(x))∈R 128p×1 Represents the SIFT feature vector extracted from p features, and 128 SIFT features are extracted from each feature point. Assume that the correct feature point of a normal face is x * , Φ * Represents x *The eigenvalue is taken at , so the detection objective function of facial feature points is as follows:

[0028] (x+Δx)=h(d(x+Δx)-Φ * ) 2

[0029] The calculation of facial feature points is achieved by solving the optimal problem in Δx. * and Δx are known,

[0030]

[0031] During the testing phase, Φ * Unknown, take the derivative with respect to Δx, let f'(x+Δx)=0 to get

[0032]

[0033] in ΔΦ0=Φ0-Φ * , Through training, a series of descent directions {R k} and {b k}So as to update x, that is

[0034] x k =x k-1 +R k-1 Φ k-1 +b k-1

[0035] Through a series of iterations, x k It will eventually converge to the feature point x * ;

[0036] During the testing phase, a local region of 32×32 SIFT features is extracted for each feature point, and PCA is used for dimensionality reduction, preserving 98% of the energy for each image. During the testing process, the average shape of the training sample image is used as the initial shape. A series of descent directions and offsets are learned to update the initial shape of the test image. Multiple iterations are performed until convergence, ultimately achieving accurate positioning of facial feature points.

[0037] The located facial expression features are divided into four areas: eyebrows, forehead, eyes, and mouth. The SVM classification method is used to classify the expression and determine whether the driver has an anxious or nervous expression.

[0038] Furthermore, the recognition and judgment of the head deflection action includes head deviation detection and line of sight deviation detection, as follows:

[0039] Head deviation detection: determines the head deviation angle by changes in facial contour area and the position of facial features. It is set to be determined as head deviation when the head deflection angle exceeds 35 degrees;

[0040] Gaze deviation detection determines the gaze by analyzing the position of the iris center relative to the eye contour. When the gaze angle falls within the angle range of the human eye in the car dashboard area, it is determined to be looking at the odometer.

[0041] Furthermore, the head deviation detection is performed by pre-establishing a three-dimensional model, manually calibrating the facial feature points of the three-dimensional model, and then randomly rotating and translating these feature points. The two-dimensional points are projected through the camera imaging model to obtain a mapping of the two-dimensional facial feature points to the head posture. The head deflection angle can be obtained through linear regression based on the facial feature points detected by SDM.

[0042] Furthermore, the line of sight deviation detection includes the following steps:

[0043] Eye contour detection, using the SDM detection algorithm to obtain the eye feature area in the facial expression features;

[0044] Iris center positioning, simplify the eyeball model into a circular object, obtain the center point by analyzing the gradient vector, let c represent the possible pupil position, where d i Expressed as a normalized displacement vector, g i Expressed as a gradient vector, when c is the true center of the circle, d i and g i have the same displacement vector, otherwise there will be an angle between the two. When calculating, a weight w is assigned to each possible center point c. c , find the actual center of the circle using the formula

[0045]

[0046] The sight angle is calculated using the relative position of the iris center and the eye contour, and sight deviation is classified using the SVM classification method. The sight deviation classification is divided into two categories: when the sight is in the instrument panel area and when the sight is not in the instrument panel area.

[0047] Furthermore, the training steps of the charging demand prediction model are as follows:

[0048] Assume that N groups of data are used for model training and the correlation coefficient of each influencing factor is calculated:

[0049] The remaining mileage data of the car is used as the reference sequence Y0, and the six factors that influence the judgment of anxiety behavior, namely the eyes, eyebrows, mouth, forehead features, the number of times the driver looks down at the dashboard and the number of times the head deflects in a certain period of time, are used as the comparison sequence X. i, these 7 sequences form a matrix, import the input data, and generate the reference sequence

[0050] Y0=(Y0)((t)),(t=1,2,...,N)

[0051] Comparing sequences

[0052] X i ={X i (t),(i=1,2,...,6)(t=1,2,...,N)}

[0053] Among them, X i (t) represents the value of the Nth group of data on index i, and the matrix [Y0,X1,X2,X3,X4,X5,X6] is obtained T First, the variable sequence is dimensionless to form the matrix [Y'0,X'1,X'2,X'3,X'4,X'5,X'6] T

[0054]

[0055] According to Δ 0i (t)=|Y'0(t)-X' i (t)|(i=1,2,...,6;t=1,2,...,10) Calculate the absolute difference between the reference sequence and the six comparison sequences, obtain the absolute difference matrix, and find the minimum and maximum numbers in the absolute difference matrix, which are expressed as Δ min and Δ max ;

[0056] The correlation coefficient is calculated through the correlation coefficient matrix,

[0057]

[0058] Where ρ is the resolution coefficient, which ranges from 0 to 1, and ε 0i (t) is the correlation coefficient between the i-th group of data and the i-th comparison sequence,

[0059] Calculate the value of each correlation γ 0i ,

[0060]

[0061] By sorting, we can get the influence of each influencing factor, take the correlation of each influencing factor as the weight, perform weighted calculation, and find the average credibility of N groups of data.

[0062] Furthermore, the judgment steps of the charging demand prediction model are as follows:

[0063] The six factors that influence the judgment of anxious behavior, namely, the features of eyes, eyebrows, mouth, forehead, the number of times the driver looks down at the dashboard and the number of times the head turns in a certain period of time, are input into the charging demand prediction model. The weights obtained in the training phase are used for weighted processing to obtain the credibility Ψ * ,when When , it is determined that the driver corresponding to this group of data has a high probability of generating charging demand behavior during this time period.

[0064] The present invention uses an on-board camera to capture facial images and identify the driver's facial expressions and head movements. These expressions and head movements are then input into a trained charging demand prediction model to predict the driver's charging needs, thereby better addressing range anxiety. The charging demand prediction model in the present invention extracts a large number of driver facial expressions and head movements, then compares and trains them with information about the remaining charge of the electric vehicle to produce a relatively accurate charging demand prediction model. Using this relatively accurate charging demand prediction model, it is possible to predict the probability that the corresponding driver will generate charging demand behavior within a certain time period. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] Figure 1 This is a system architecture diagram of a charging anxiety behavior prediction system according to the present invention;

[0066] Figure 2 This is a flow chart of a method for predicting charging anxiety behavior according to the present invention;

[0067] Figure 3 This is a model diagram of the relative position of the iris center and the eye contour. DETAILED DESCRIPTION

[0068] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0069] The terms "first" and "second" in the specification, claims and drawings of the present invention are used to distinguish different objects rather than to describe a specific order.

[0070] The embodiment of the present invention provides a charging anxiety behavior prediction system. Figure 1As shown, the system includes an acquisition unit, an image preprocessing unit, a vehicle remaining mileage monitoring unit, and a cloud storage computing platform. The acquisition unit acquires the driver's facial image in real time via an onboard camera deployed in the electric vehicle. The image preprocessing unit is in communication with the acquisition unit and is used to perform image preprocessing on the facial image. The vehicle remaining mileage monitoring unit transmits data with the electric vehicle dashboard to acquire the dashboard remaining power information in real time. The cloud storage computing platform is used to store the charging demand prediction model and receive the dashboard remaining power information in real time. The platform receives the preprocessed facial image and recognizes and determines the facial expression features and head deflection movements. The facial expression features and head deflection movements are used as input to the charging demand prediction model. The platform then outputs a prediction result of whether charging behavior has occurred and records and stores the result. The prediction result is compared with the dashboard remaining power data uploaded from the vehicle in real time to evaluate the accuracy of the prediction and continuously train and optimize the charging demand prediction model.

[0071] In one embodiment of the present invention, an image preprocessing unit is integrated into an in-vehicle processor and includes an image acquisition module, an image preprocessing module, and a first communication module. The image acquisition module is configured to receive the facial image transmitted by the acquisition unit, wherein the facial image is an image of the driver's face captured at intervals of Δt1 by an onboard camera aimed at the driver's upper body or face; the image preprocessing module is communicatively connected to the image acquisition module and is configured to perform face recognition, data enhancement, normalization, and grayscale processing on the facial image; the first communication module is configured to connect to the acquisition unit via a CAN bus or SerDes and is configured to transmit the facial image captured by the acquisition unit to the image acquisition module, and to connect to a cloud storage computing platform via a 5G network and upload the preprocessed facial image to the cloud storage computing platform.

[0072] In one embodiment of the present invention, the cloud storage computing platform includes a second communication module, a human expression recognition module, a head deflection recognition module, a charging demand prediction module, and a comparison module.

[0073] a second communication module, configured to receive the preprocessed facial image sent by the image preprocessing unit, and receive the instrument panel remaining power information sent by the vehicle remaining mileage monitoring unit;

[0074] The human expression recognition module is in communication with the image preprocessing unit, extracts features from the preprocessed facial image, and uses an SVM classifier to identify the driver's expression features to determine whether the driver has anxious expression features;

[0075] A head deflection recognition module is in communication with the image preprocessing unit and performs head deviation detection and line of sight deviation detection based on the preprocessed facial image. It is used to identify the driver's head deflection movement and detect whether the driver is looking down at the remaining mileage meter on the car instrument panel.

[0076] a charging demand prediction module, configured to input the facial expression features and the head deflection movement into the charging demand prediction model, and then output a prediction result of whether there is a charging behavior and record and store it;

[0077] The comparison module is used to compare the prediction result with the remaining power data on the instrument panel in real time to evaluate the accuracy of the prediction.

[0078] Among them, the vehicle-mounted camera in the first acquisition unit can be set to 1 or 2. When 1 vehicle-mounted camera is set, the installation position of the vehicle-mounted camera has higher requirements, and the vehicle-mounted camera needs to be installed in a position where it can clearly capture the front image and facial image of the driver's head; when 2 vehicle-mounted cameras are set, one camera is used to aim at the driver's face, and the other camera is used to capture the front of the driver's head.

[0079] Among them, the remaining power information can be directly transmitted from the in-vehicle processor to the cloud storage computing platform. This method is more accurate and convenient, and is preferred for this embodiment. The dashboard can also be photographed by a camera for data processing, and the remaining power information can be extracted in the processor and then transmitted to the cloud storage computing platform.

[0080] The embodiment of the present invention also provides a charging anxiety behavior prediction method, such as Figure 2 Shown, including:

[0081] S10. Capturing a facial image of the driver using an onboard camera deployed inside the electric vehicle;

[0082] S20. The facial image is transmitted to the in-vehicle processor, the driver's face position is located by the oriented gradient histogram, and data preprocessing such as data enhancement, normalization, and grayscale is performed, and the preprocessed facial image is transmitted to the cloud storage computing platform;

[0083] S30. In the cloud storage computing platform, based on the pre-processed facial image, the driver's facial features and head deflection movements are identified;

[0084] S40. The driver's facial features and head deflection movements are used as input into the charging demand prediction model to determine whether the driver currently has a need for charging. The judgment result is compared with the current remaining power of the electric vehicle. The prediction model is continuously verified and improved until the prediction result better matches the remaining power and the prediction accuracy is high enough. Finally, the trained prediction model and each prediction result are stored in the cloud storage computing platform.

[0085] In one embodiment of the present invention, the steps of identifying and judging the facial expression features in step S30 are as follows:

[0086] Based on the preprocessed facial image, given an image d∈R containing m pixels m×1 , d(x)∈R p×1 , used to index the p feature points of the image, x represents the p feature points, h is the nonlinear feature function at each feature point, h(d(x))∈R 128p×1 Represents the SIFT feature vector extracted from p features, and 128 SIFT features are extracted from each feature point. Assume that the correct feature point of a normal face is x * , Φ * Represents x * The eigenvalue is taken at , so the detection objective function of facial feature points is as follows:

[0087] (x+Δx)=h(d(x+Δx)-Φ * ) 2

[0088] The calculation of facial feature points is achieved by solving the optimal problem in Δx. * and Δx are known,

[0089]

[0090] During the testing phase, Φ * Unknown, take the derivative with respect to Δx, let f'(x+Δx)=0 to get

[0091]

[0092] in ΔΦ0=Φ0-Φ * , Through training, a series of descent directions {R k} and {b k}So as to update x, that is

[0093] x k =x k-1 +R k-1 Φ k-1 +b k-1

[0094] Through a series of iterations, x k It will eventually converge to the feature point x * ;

[0095] During the testing phase, a local region of 32×32 SIFT features is extracted for each feature point, and PCA is used for dimensionality reduction, preserving 98% of the energy for each image. During the testing process, the average shape of the training sample image is used as the initial shape. A series of descent directions and offsets are learned to update the initial shape of the test image. Multiple iterations are performed until convergence, ultimately achieving accurate positioning of facial feature points.

[0096] The located facial expression features are divided into four areas: eyebrows, forehead, eyes, and mouth. The SVM classification method is used to classify the expression and determine whether the driver has an anxious or nervous expression.

[0097] In one embodiment of the present invention, the recognition and determination of the head deflection action in step S30 includes head deviation detection and line of sight deviation detection as follows:

[0098] Head deviation detection: determines the head deviation angle by changes in facial contour area and the position of facial features. It is set to be determined as head deviation when the head deflection angle exceeds 35 degrees;

[0099] Gaze deviation detection determines the gaze by analyzing the position of the iris center relative to the eye contour. When the gaze angle falls within the angle range of the human eye in the car dashboard area, it is determined to be looking at the odometer.

[0100] In one embodiment of the present invention, head deviation detection is performed by pre-establishing a three-dimensional model, manually calibrating the facial feature points of the three-dimensional model, and then randomly rotating and translating these feature points. The two-dimensional points are projected through the camera imaging model to obtain a mapping of the two-dimensional facial feature points to the head posture. The head deflection angle can be obtained through linear regression based on the facial feature points detected by SDM.

[0101] In one embodiment of the present invention, the line of sight deviation detection comprises the following steps:

[0102] Eye contour detection, using the SDM detection algorithm to obtain the eye feature area in the facial expression features;

[0103] Iris center positioning, simplify the eyeball model into a circular object, obtain the center point by analyzing the gradient vector, let c represent the possible pupil position, where d i Expressed as a normalized displacement vector, g i Expressed as a gradient vector, when c is the true center of the circle, d i and g i have the same displacement vector, otherwise there will be an angle between the two. When calculating, a weight w is assigned to each possible center point c. c , find the actual center of the circle using the formula

[0104]

[0105] The sight angle is calculated by using the relative position of the iris center and the eye contour. The relative position model of the iris center and the eye contour is as follows: Figure 3 As shown by Figure 3 The relevant feature parameters are extracted as follows:

[0106]

[0107] The above three feature parameters are used to form the feature vector X (FeatureX, FeatureY, FeatureXY) of gaze detection, and the classifier trained by SVM is used to classify gaze deviation. Among them, the feature vector passed into the SVM is X, and the output category is divided into two categories: when the gaze is in the dashboard area and when the gaze is not in the dashboard area.

[0108] In one embodiment of the present invention, the training steps of the charging demand prediction model in step S40 are as follows:

[0109] Assume that N groups of data are used for model training and the correlation coefficient of each influencing factor is calculated:

[0110] The remaining mileage data of the car is used as the reference sequence Y0, and the six factors that influence the judgment of anxiety behavior, namely the eyes, eyebrows, mouth, forehead features, the number of times the driver looks down at the dashboard and the number of times the head deflects in a certain period of time, are used as the comparison sequence X. i , these 7 sequences form a matrix, import the input data, and generate the reference sequence

[0111] Y0=(Y0)((t)),(t=1,2,...,N)

[0112] Comparing sequences

[0113] X i ={X i (t),(i=1,2,...,6)(t=1,2,...,N)}

[0114] Among them, X i (t) represents the value of the Nth group of data on index i, and the matrix [Y0,X1,X2,X3,X4,X5,X6] is obtained T First, the variable sequence is dimensionless to form the matrix [Y'0,X'1,X'2,X'3,X'4,X'5,X'6] T

[0115]

[0116] According to Δ 0i (t)=|Y'0(t)-X'i (t)|(i=1,2,...,6;t=1,2,...,10) Calculate the absolute difference between the reference sequence and the six comparison sequences, obtain the absolute difference matrix, and find the minimum and maximum numbers in the absolute difference matrix, which are expressed as Δ min and Δ max ;

[0117] The correlation coefficient is calculated through the correlation coefficient matrix,

[0118]

[0119] Where ρ is the resolution coefficient, which ranges from 0 to 1, and ε 0i (t) is the correlation coefficient between the i-th group of data and the i-th comparison sequence,

[0120] Calculate the value of each correlation γ 0i ,

[0121]

[0122] By sorting, we can get the influence of each influencing factor, take the correlation of each influencing factor as the weight, perform weighted calculation, and find the average credibility of N groups of data.

[0123] In one embodiment of the present invention, the determination steps of the charging demand prediction model in step S40 are as follows:

[0124] The six factors that influence the judgment of anxious behavior, namely, the features of eyes, eyebrows, mouth, forehead, the number of times the driver looks down at the dashboard and the number of times the head turns in a certain period of time, are input into the charging demand prediction model. The weights obtained in the training phase are used for weighted processing to obtain the credibility Ψ * ,when When , it is determined that the driver corresponding to this group of data has a high probability of generating charging demand behavior during this time period.

[0125] The above embodiments are intended only to illustrate the technical solutions of the present invention and are not intended to limit the scope of the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art will appreciate that the technical solutions of the present invention may be modified or replaced with equivalents without departing from the spirit and scope of the technical solutions of the present invention, and such modifications or equivalents shall be encompassed by the claims of the present invention. Any techniques, shapes, and structures not described in detail herein are well known.

Claims

1. A charging anxiety behavior prediction system, characterized in that: The charging anxiety behavior prediction system includes: An acquisition unit, which acquires the driver's facial image in real time through an on-board camera deployed in the electric vehicle; An image preprocessing unit, communicatively connected to the acquisition unit, and configured to perform image preprocessing on the facial image; The vehicle remaining mileage monitoring unit transmits data with the electric vehicle instrument panel to obtain the remaining power information on the instrument panel in real time; A cloud storage computing platform is used to store the charging demand prediction model and receive real-time remaining power information from the dashboard. The platform receives pre-processed facial images and identifies and determines facial features and head deflection movements. The facial features and head deflection movements are input into the charging demand prediction model, and the platform outputs and records a prediction result of whether charging behavior is required. The platform compares the prediction result with the remaining power data uploaded from the dashboard in real time to evaluate the accuracy of the prediction and continuously trains and optimizes the charging demand prediction model. The training method of the charging demand prediction model is as follows: Assume that N groups of data are used for model training and the correlation coefficient of each influencing factor is calculated: The remaining mileage data of the car is used as the reference sequence Y0, and the six factors that influence the judgment of anxiety behavior, namely the eyes, eyebrows, mouth, forehead features, the number of times the driver looks down at the dashboard and the number of times the head deflects in a certain period of time, are used as the comparison sequence X. i , these 7 sequences form a matrix, import the input data, and generate the reference sequence Y0=(Y0)((t)),(t=1,2,...,N) Comparing sequences X i ={X i (t),(i=1,2,...,6)(t=1,2,...,N)} Among them, X i (t) represents the value of the Nth group of data on index i, and the matrix [Y0,X1,X2,X3,X4,X5,X6] is obtained T , First, the variable sequence is dimensionless to form the matrix [Y0',X'1,X'2,X'3,X'4,X'5,X'6] T According to Δ 0i (t)=|Y0'(t)-X' i (t)|(i=1,2,...,6;t=1,2,...,10) Calculate the absolute difference between the reference sequence and the six comparison sequences, obtain the absolute difference matrix, and find the minimum and maximum numbers in the absolute difference matrix, which are expressed as Δ min and Δ max ; The correlation coefficient is calculated through the correlation coefficient matrix, Where ρ is the resolution coefficient, which ranges from 0 to 1, and ε 0i (t) is the correlation coefficient between the i-th group of data and the i-th comparison sequence, Calculate the value of each correlation γ 0i , By sorting, we can get the influence of each influencing factor, take the correlation of each influencing factor as the weight, perform weighted calculation, and find the average credibility of N groups of data. The judgment steps of the charging demand prediction model are as follows: The six factors that influence the judgment of anxious behavior, namely, the features of eyes, eyebrows, mouth, forehead, the number of times the driver looks down at the dashboard and the number of times the head turns in a certain period of time, are input into the charging demand prediction model. The weights obtained in the training phase are used for weighted processing to obtain the credibility Ψ * ,when When , it is determined that the driver corresponding to this group of data has a high probability of generating charging demand behavior during this time period.

2. A charging anxiety behavior prediction system according to claim 1, characterized in that: The image pre-processing unit is integrated into the in-vehicle processor and includes: An image acquisition module, configured to receive the facial image transmitted by the acquisition unit, wherein the facial image is a facial image of the driver acquired at intervals of Δt1 by aiming a vehicle-mounted camera at the driver's upper body or face; An image preprocessing module, in communication with the image acquisition module, for performing face recognition, data enhancement, normalization, and grayscale processing on the facial image; The first communication module uses a CAN bus or SerDes to connect to the acquisition unit, and is used to transmit the facial image collected by the acquisition unit to the image acquisition module, and uses a 5G network to connect to the cloud storage computing platform to upload the preprocessed facial image to the cloud storage computing platform.

3. A charging anxiety behavior prediction system according to claim 2, characterized in that: The cloud storage computing platform includes: a second communication module, configured to receive the preprocessed facial image sent by the image preprocessing unit, and receive the instrument panel remaining power information sent by the vehicle remaining mileage monitoring unit; The human expression recognition module is in communication with the image preprocessing unit, extracts features from the preprocessed facial image, and uses an SVM classifier to identify the driver's expression features to determine whether the driver has anxious expression features; A head deflection recognition module is in communication with the image preprocessing unit and performs head deviation detection and line of sight deviation detection based on the preprocessed facial image. It is used to identify the driver's head deflection movement and detect whether the driver is looking down at the remaining mileage meter on the car instrument panel. a charging demand prediction module, configured to input the facial expression features and the head deflection movement into the charging demand prediction model, and then output a prediction result of whether there is a charging behavior and record and store it; The comparison module is used to compare the prediction result with the remaining power data on the instrument panel in real time to evaluate the accuracy of the prediction.

4. A method for predicting charging anxiety behavior, characterized in that: Applied to the system according to any one of claims 1 to 3, the method comprising: The driver's facial image is collected through an on-board camera deployed inside the electric vehicle; The facial image is transmitted to an in-vehicle processor, the driver's face position is located by using a histogram of oriented gradients, and data preprocessing such as data enhancement, normalization, and grayscale conversion is performed, and the preprocessed facial image is transmitted to a cloud storage computing platform; In the cloud storage computing platform, based on the pre-processed facial image, the driver's facial expression features and head deflection movements are identified and judged; The driver's facial features and head deflection movements are used as input into the charging demand prediction model to determine whether the driver currently has a need to charge. The judgment result is compared with the current remaining power of the electric vehicle. The prediction model is continuously verified and improved until the prediction result better matches the remaining power. Finally, the trained prediction model and each prediction result are stored in the cloud storage computing platform.

5. A charging anxiety behavior prediction method according to claim 4, characterized in that: The steps of identifying and judging the facial expression features are as follows: Based on the preprocessed facial image, given an image d∈R containing m pixels m×1 , d(x)∈R p×1 , used to index the p feature points of the image, x represents the p feature points, h is the nonlinear feature function at each feature point, h(d(x))∈R 128p×1 Represents the SIFT feature vector extracted from p features, assuming that the correct feature point of the normal face is x * , Φ * Represents x * The eigenvalue is taken at , so the detection objective function of facial feature points is as follows: (x+Δx)=h(d(x+Δx)-Φ * ) 2 The calculation of facial feature points is achieved by solving the optimal problem in Δx. * and Δx are known, During the testing phase, Φ * Unknown, take the derivative with respect to Δx, let f'(x+Δx)=0 to get in Φ0=h(d(x0)), ΔΦ0=Φ0-Φ * , Through training, a series of descent directions {R k } and {b k }So as to update x, that is x k =x k-1 +R k-1 F k-1 +b k-1 Through a series of iterations, x k It will eventually converge to the feature point x * ; During the testing phase, a local region of 32×32 SIFT features is extracted for each feature point, and PCA is used for dimensionality reduction, preserving 98% of the energy for each image. During the testing process, the average shape of the training sample image is used as the initial shape. A series of descent directions and offsets are learned to update the initial shape of the test image. Multiple iterations are performed until convergence, ultimately achieving accurate positioning of facial feature points. The located facial expression features are divided into four areas: eyebrows, forehead, eyes, and mouth. The SVM classification method is used to classify the expression and determine whether the driver has an anxious or nervous expression.

6. The method for predicting charging anxiety behavior according to claim 5, characterized in that: The recognition and judgment of the head deflection action includes head deviation detection and line of sight deviation detection, as follows: Head deviation detection: determines the head deviation angle by changes in facial contour area and the position of facial features. It is set to be determined as head deviation when the head deflection angle exceeds 35 degrees; Gaze deviation detection determines the gaze by analyzing the position of the iris center relative to the eye contour. When the gaze angle falls within the angle range of the human eye in the car dashboard area, it is determined to be looking at the odometer.

7. The method for predicting charging anxiety behavior according to claim 6, characterized in that: The head deviation detection is achieved by pre-establishing a three-dimensional model, manually calibrating the facial feature points of the three-dimensional model, and then randomly rotating and translating these feature points. The two-dimensional points are projected through the camera imaging model to obtain a mapping from the two-dimensional facial feature points to the head posture. The head deflection angle can be obtained through linear regression based on the facial feature points detected by SDM.

8. The method for predicting charging anxiety behavior according to claim 7, characterized in that: The line of sight deviation detection comprises the following steps: Eye contour detection, using the SDM detection algorithm to obtain the eye feature area in the facial expression features; Iris center positioning, simplify the eyeball model into a circular object, obtain the center point by analyzing the gradient vector, let c represent the possible pupil position, where d i Expressed as a normalized displacement vector, g i Expressed as a gradient vector, when c is the true center of the circle, d i and g i have the same displacement vector, otherwise there will be an angle between the two. When calculating, a weight w is assigned to each possible center point c. c , find the actual center of the circle using the formula The sight angle is calculated using the relative position of the iris center and the eye contour, and sight deviation is classified using the SVM classification method. The sight deviation classification is divided into two categories: when the sight is in the instrument panel area and when the sight is not in the instrument panel area.

9. The method for predicting charging anxiety behavior according to claim 8, characterized in that: The training steps of the charging demand prediction model are as follows: Assume that N groups of data are used for model training and the correlation coefficient of each influencing factor is calculated: The remaining mileage data of the car is used as the reference sequence Y0, and the six factors that influence the judgment of anxiety behavior, namely the eyes, eyebrows, mouth, forehead features, the number of times the driver looks down at the dashboard and the number of times the head deflects in a certain period of time, are used as the comparison sequence X. i , these 7 sequences form a matrix, import the input data, and generate the reference sequence Y0=(Y0)((t)),(t=1,2,...,N) Comparing sequences X i ={X i (t),(i=1,2,...,6)(t=1,2,...,N)} Among them, X i (t) represents the value of the Nth group of data on index i, and the matrix [Y0,X1,X2,X3,X4,X5,X6] is obtained T , First, the variable sequence is dimensionless to form the matrix [Y0',X'1,X'2,X'3,X'4,X'5,X'6] T According to Δ 0i (t)=|Y0'(t)-X' i (t)|(i=1,2,...,6;t=1,2,...,10) Calculate the absolute difference between the reference sequence and the six comparison sequences, obtain the absolute difference matrix, and find the minimum and maximum numbers in the absolute difference matrix, which are expressed as Δ min and Δ max ; The correlation coefficient is calculated through the correlation coefficient matrix, Where ρ is the resolution coefficient, which ranges from 0 to 1, and ε 0i (t) is the correlation coefficient between the i-th group of data and the i-th comparison sequence, Calculate the value of each correlation γ 0i , By sorting, we can get the influence of each influencing factor, take the correlation of each influencing factor as the weight, perform weighted calculation, and find the average credibility of N groups of data.

10. The method for predicting charging anxiety behavior according to claim 9, characterized in that: The judgment steps of the charging demand prediction model are as follows: The six factors that influence the judgment of anxious behavior, namely, the features of eyes, eyebrows, mouth, forehead, the number of times the driver looks down at the dashboard and the number of times the head turns in a certain period of time, are input into the charging demand prediction model. The weights obtained in the training phase are used for weighted processing to obtain the credibility Ψ * ,when When , it is determined that the driver corresponding to this group of data has a high probability of generating charging demand behavior during this time period.

Images