Non-invasive Detection-based Charging Load Probability Prediction System and Method

Abstract

The present invention relates to the field of charging load prediction technology, and in particular to a charging load probability prediction system and method based on non-invasive detection, which is used to solve the problem of charging load prediction and planning. The charging load probability prediction system based on non-invasive detection includes an image acquisition unit, a temperature detection unit, a positioning unit, a vehicle networking service system, and a cloud storage computing platform. In the cloud storage computing platform, vehicle images and real-time temperature data collected by traffic cameras can be received in real time, and the data obtained by vehicle image recognition can be classified according to the temperature data range, and input into the corresponding charging load prediction model to obtain the probability of when and where the target will be charged and the charging load. Through time accumulation, the probability of when and where to charge and the charging load can be used for the planning and construction of charging facilities in certain areas. The present invention can predict the charging demand of electric vehicles in the prediction area and reasonably select the layout planning of charging facilities.

Description

Technical Field

[0001] The present invention relates to the technical field of charging load prediction, and particularly to a charging load probability prediction system and method based on non-intrusive detection. Background Art

[0002] The charging load of electric vehicles has strong randomness in both time and space. With the popularization of electric vehicles in the future, their charging load will have an increasingly greater impact on the operation of urban distribution networks, especially for small and medium-sized urban power grids. After a large-scale access of electric vehicle charging loads to the power grid, on the one hand, it affects the power quality of the distribution network, bringing voltage deviation, three-phase imbalance and harmonic pollution to the distribution network, further widening the peak-valley difference and directly affecting the reliability of the power grid; on the other hand, the large-scale access causes changes in the power loss of the distribution network and the service life of transformers, affecting the economic operation of the power grid.

[0003] The prediction of electric vehicle charging load is the basis for analyzing the impact of electric vehicle access on the power grid, planning and controlling the operation of the distribution network, realizing the two-way interaction between electric vehicles and the power grid, and coordinating the research of electric vehicles with other energy and transportation systems. Due to the randomness of the charging behavior of electric vehicles in time and space, the prediction of charging load involves very complex influencing factors, and different consideration angles will form different load prediction models and results. In order to ensure the normal and reliable operation of the urban power grid, it is necessary to accurately predict the future charging load of electric vehicles. It is urgent to investigate and analyze the impact of large-scale access of electric vehicles on aspects such as the power grid structure, power quality, load curve, and dispatching control, and form an adaptive plan for the coordinated development of the power grid and electric vehicles, so as to more effectively promote the popularization and application of new energy electric vehicles. Currently, the methods for predicting the load of electric vehicles are mainly based on the influencing factors of electric vehicle load, and are divided into short-term load prediction methods for power systems, Monte Carlo simulation methods, and other new electric vehicle load prediction methods. However, from the current research methods, the influence of temperature control load is not considered, that is, in colder or hotter seasons, the turning on of the in-vehicle air conditioner will increase the power consumption of electric vehicles, exacerbate the "range anxiety" of drivers, and make the charging demand of electric vehicles more frequent.

[0004] The frequent charging behavior of electric vehicle users caused by environmental temperature exacerbates the grid load in typical seasons, and temperature has become one of the influencing factors that cannot be ignored in the prediction of electric vehicle charging load. Currently, there is little research on taking environmental temperature as one of the influencing factors for charging demand prediction. Summary of the Invention

[0005] In view of this, the purpose of the present invention is to provide a charging load probability prediction system and method based on non-intrusive detection, which predicts the charging demand of electric vehicles by adding the influence factor of environmental temperature, so as to reasonably select the layout plan of charging facilities.

[0006] The present invention solves the above technical problems through the following technical means: One aspect of the present invention is to provide a charging load probability prediction system based on non-intrusive detection, including:

[0007] An image acquisition unit that captures pictures of passing vehicles through traffic cameras deployed on various roads and processes the picture data;

[0008] A temperature detection unit that obtains the ambient temperature through a temperature sensor installed beside the traffic camera;

[0009] A positioning unit installed on the traffic camera for obtaining the position of the traffic camera, and thus obtaining the position of the electric vehicle at the shooting moment;

[0010] A vehicle networking service system that obtains information about surrounding charging stations at this time based on the geographical location of the traffic camera;

[0011] A cloud storage and computing platform that is communicatively connected to the image acquisition unit, the temperature detection unit, the vehicle networking service system, and the positioning unit, and is used for processing and storing the historical vehicle pictures collected by the image acquisition unit, and for training a charging load prediction model by identifying license plate information, the deflection action of the vehicle owner's head, and the facial expression features of the vehicle owner from the historical vehicle picture data, in combination with the historical air conditioner opening state data and historical charging records of electric vehicles, and for receiving in real time the vehicle pictures collected and processed by the traffic camera and the real-time temperature data of the temperature sensor, classifying the data obtained by identifying the vehicle pictures according to the temperature data range, and inputting them into the corresponding charging load prediction models to obtain the probability of when and where the target will charge and the charging load, and through time accumulation, it is used for the planning and construction of charging facilities in certain areas.

[0012] Further, the image acquisition unit includes:

[0013] Traffic cameras deployed on various traffic roads for capturing pictures of passing vehicles;

[0014] An image preprocessing module communicatively connected to the traffic camera for performing data enhancement, normalization, and grayscale preprocessing on the passing vehicle pictures;

[0015] A first communication module that uses a 5G network to connect to the cloud storage and computing platform and uploads the preprocessed vehicle images to the cloud storage and computing platform;

[0016] The temperature detection unit includes:

[0017] A temperature sensor installed beside the traffic camera for monitoring the ambient temperature;

[0018] The third communication module uses a 5G network to connect to the cloud storage and computing platform and uploads the ambient temperature to the cloud storage and computing platform;

[0019] The positioning unit includes:

[0020] The positioning module is used to obtain the position of the traffic camera, and then obtain the position of the electric vehicle at the shooting moment;

[0021] The fourth communication module uses a 5G network to connect to the cloud storage and computing platform and uploads the position information data to the cloud storage and computing platform;

[0022] The vehicle networking service system includes:

[0023] The database module is used to store the charging station information of various places;

[0024] The charging record module is used to store the historical charging data of various electric vehicles;

[0025] The air conditioner usage status record module is used to store the historical air conditioner usage status data of various electric vehicles;

[0026] The fifth communication module uses a 5G network to connect to the cloud storage and computing platform and uploads the charging station information of various places, the historical charging data of electric vehicles, and the historical air conditioner usage status data to the cloud storage and computing platform.

[0027] Furthermore, the cloud storage and computing platform includes:

[0028] The historical image database module is used to store the pre-processed historical vehicle pictures to form a historical database;

[0029] The image recognition module is used to perform license plate information, the head deflection action of the vehicle owner, and the facial expression features of the vehicle owner on the electric vehicle in the processed vehicle pictures;

[0030] The model training module identifies license plate information, the head deflection action of the vehicle owner, and the facial expression features from the historical vehicle picture data, and combines the historical air conditioner on status data and historical charging records to train a charging load prediction model;

[0031] The model prediction module is used to receive in real time the vehicle pictures collected and processed by the traffic camera and the real-time temperature data of the temperature sensor, classify the data obtained by identifying the vehicle pictures according to the temperature data range, input them into the corresponding charging load prediction model, obtain the probability of when and where the target will charge and the charging load, and through time accumulation, obtain the charging facility planning and construction in certain areas;

[0032] The second communication module uses a 5G network to connect the image acquisition unit, the temperature detection unit, the vehicle networking service system, and the positioning unit.

[0033] Another aspect of the present invention is to provide a method for predicting the probability of charging load based on non-invasive detection, including the following steps:

[0034] Collect historical vehicle pictures passing by through traffic cameras deployed on the road, and establish a picture database;

[0035] Perform data processing on the historical vehicle pictures to obtain license plate information, the deflection action of the vehicle owner's head, and the facial expression features of the vehicle owner;

[0036] Based on the license plate information, obtain the charging records of the electric vehicle through the vehicle networking system, including the charging location and charging time;

[0037] Taking the deflection action of the vehicle owner's head and the facial expression features of the vehicle owner as independent variables, and the charging time and charging location as dependent variables, establish a prediction model for the charging load demand of each vehicle owner respectively;

[0038] Obtain the historical air-conditioning usage status of the electric vehicle through the vehicle networking system, and train and optimize the charging load demand prediction model through the air-conditioning turning-on information to obtain the first load demand prediction model when the air-conditioning is on and the second load demand prediction model when the air-conditioning is off;

[0039] Detect the ambient temperature through the temperature sensor beside the traffic camera, clarify the corresponding temperature of the photo taken at a certain moment, and judge whether the in-vehicle air-conditioning is turned on. If the in-vehicle air-conditioning is turned on, input it into the first load demand prediction model for charging load prediction. If the air-conditioning is not turned on, input it into the second load demand prediction model for charging load prediction;

[0040] Combined with the positioning unit of the traffic camera, the probability of when and where an electric vehicle will charge at a certain moment and the charging load prediction can be obtained.

[0041] Furthermore, the picture database is composed of historical picture data of m cameras, t moments, and n targets, with a total of mnt basic pictures.

[0042] Furthermore, the recognition and acquisition of the license plate information include the following steps:

[0043] Preprocess the license plate image, convert the vehicle picture from the RGB channel to the HSV channel, convert the HSV channel image into a grayscale image, and then perform binarization and morphological processing on the grayscale image;

[0044] License plate positioning, use the function cv2.findContours() to perform rectangular detection on the grayscale image after morphological processing, locate the license plate area, and segment out the license plate area;

[0045] License plate character segmentation, and perform license plate horizontal correction, removal of license plate borders and rivets, and character segmentation operations on the segmented license plate area in sequence. The license plate horizontal correction includes tilt angle detection and tilt correction;

[0046] License plate character recognition, converting the characters on the image into feature vectors through HOG feature extraction, and classifying and discriminating through the SVM classification algorithm;

[0047] Obtain the result, and identify whether it is a new energy vehicle and the license plate number through classification and discrimination.

[0048] Furthermore, the acquisition of the head deflection action of the vehicle owner is as follows:

[0049] Perform data processing on the pictures of passing vehicles, obtain vehicle type information and the driving position information of the vehicle owner, and judge the corresponding direction between the head deflection direction of the vehicle owner and the dashboard through the vehicle type information and the driving position information of the vehicle owner;

[0050] According to the law of human head movement, under the condition of assuming that the human head rotates along the head center axis without swinging left and right in the vertical direction, the line segment connecting the two eyes when the driver's head faces directly forward is represented as line segment The spatial connection line segment between the two eyes after the head rotates through an angle θ is represented as line segment Numerically, line segment and are equal. The projection of the distance between the two eyes after the head rotates on the forward plane is represented as line segment Through geometric analysis, it can be obtained that the acute angle between line segment and line segment is equal to the angle through which the head rotates. Then the head rotation angle calculation formula is represented as

[0051]

[0052] If the calculated θ value is negative, it is judged that the head rotates to the left. If the θ value is positive, it is judged that the head rotates to the right.

[0053] Furthermore, the acquisition of the facial expression features of the vehicle owner is as follows:

[0054] In the feature sampling layer, perform image data preprocessing, face detection, face localization, convolutional feature learning, and feature sampling on the vehicle pictures. Input the original vehicle image sequence x = {x 1 ,x 2 ,…,x T}, preprocess the image to eliminate the influence of illumination; at time t = 1, use the Fast-CNN method to quickly locate the face image and perform segmentation; at subsequent times, quickly track and segment the face image; then, the face image enters the convolutional neural network for learning, and through cross-convolution and pooling, generate image abstract features; finally, input to the K-means sampling layer, which averages and samples the x T-K+1 , x T-K+2 , …, x T A total of K consecutive image features are averaged and sampled, and used as input features to enter the recurrent network for learning;

[0055] The main loop of the RNN and the information memory of the LSTM take the convolutional sampling feature vector obtained from the feature sampling layer as input; then, according to the time series, enter the recurrent network, and the recurrent unit LSTM extracts information to generate state information; finally, output the feature vector for classification;

[0056] Classify the output sequence of the feature vectors learned by the recurrent network layer. Let T be the sequence length and L be the length of the label. Then, when the sequence T appears, there are L to the power of T possible label paths, and each of them is called a "Path". Its conditional probability formula

[0057]

[0058] Among them, π (t) represents the output path π at time t, and y t represents the network output of the RNN at time t,

[0059] The predicted label conditional probability is expressed as the sum of the conditional probabilities of the corresponding Paths,

[0060]

[0061] Among them, V represents the operator that converts the output path π to the target L,

[0062] When the input sequence is trained, the label with the maximum conditional probability of the input image expression at the current time is:

[0063]

[0064] Identify various expressions of the car owner, including anger, disgust, fear, happiness, sadness, surprise, and neutral.

[0065] Furthermore, the charging load demand prediction model is established as follows:

[0066] Assume that N groups of data are used for model training, and perform random forest algorithm analysis:

[0067] a. Use the charging behavior of electric vehicles as the prediction variable X o, taking the eight factors of anger, disgust, fear, happiness, sadness, surprise, neutrality, and head deflection shown by the driver within one hour before charging as the influencing variable X i ;

[0068] b. Let the number of sampling times b = 1, 2, …, B s , repeat steps c and d;

[0069] c. Randomly select a sub - sample set from X o and X i as the training set through the bootstrap resampling technique;

[0070] d. Obtain a regression tree model r fb ;

[0071] After the training is completed, for a new sample x, the random forest model gives the predicted value of this sample by averaging the predicted values of all regression trees as:

[0072]

[0073] Further, in the step of determining whether to turn on the in - vehicle air conditioner, it is set that when the ambient temperature detected by the temperature sensor is 14°C to 28°C, it is the condition of not turning on the in - vehicle air conditioner, otherwise it is the condition of turning on the in - vehicle air conditioner.

[0074] The charging load probability prediction method based on non - intrusive detection of the present invention uses the method of non - intrusive detection of charging probability and charging load. After obtaining the load demand prediction model, everything does not need to be associated with the vehicle owner, avoiding affecting the vehicle owner's mood and causing a large error in the prediction result. The charging load probability prediction method based on non - intrusive detection of the present invention receives the image data collected and processed by the traffic camera in real - time, as well as the real - time temperature data of the temperature sensor, classifies the image data according to the temperature result, and inputs it into the corresponding load demand prediction model, so as to predict the probability and charging load of when and where the target electric vehicle will charge. The present invention can predict the charging demand of electric vehicles in the prediction area, so as to reasonably select the layout plan of charging facilities. Brief Description of the Drawings

[0075] Figure 1 is the system architecture schematic diagram of the charging load probability prediction system based on non - intrusive detection of the present invention;

[0076] Figure 2 is the step - by - step flow schematic diagram of the charging load probability prediction method based on non - intrusive detection of the present invention;

[0077] Figure 3 is the simplified diagram of the driver's head rotation model. Detailed implementation mode

[0078] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative work shall fall within the protection scope of the present invention.

[0079] The terms "first", "second", etc. in the description and claims of the present invention and the above-mentioned accompanying drawings are used to distinguish different objects, rather than to describe a specific order.

[0080] An electric vehicle charging load probability prediction system based on non-invasive detection is provided in an embodiment of the present invention, as Figure 1 shown, including an image acquisition unit, a temperature detection unit, a positioning unit, a vehicle networking service system, and a cloud storage and computing platform.

[0081] The image acquisition unit captures pictures of passing vehicles through traffic cameras deployed on various roads and processes the picture data; the temperature detection unit obtains the ambient temperature through a temperature sensor installed beside the traffic camera; the positioning unit is installed on the traffic camera and is used to obtain the position of the traffic camera, so as to obtain the position of the electric vehicle at the shooting moment; the vehicle networking service system obtains information about surrounding charging stations based on the geographical location of the traffic camera.

[0082] The cloud storage and computing platform is communicatively connected to the image acquisition unit, the temperature detection unit, the vehicle networking service system, and the positioning unit, and is used for processing and storing the historical vehicle pictures collected by the image acquisition unit, and identifying license plate information, the deflection action of the vehicle owner's head, and the facial expression features of the vehicle owner from the historical vehicle picture data, combining the historical air conditioner opening state data and historical charging records of electric vehicles to train a charging load prediction model, and for receiving in real time the vehicle pictures collected and processed by the traffic camera and the real-time temperature data of the temperature sensor, classifying the data obtained by identifying the vehicle pictures according to the temperature data range, and inputting them into the corresponding charging load prediction model to obtain the probability of when and where the target will charge and the charging load. Through time accumulation, the probability of when and where to charge and the charging load can be used for the planning and construction of charging facilities in certain areas.

[0083] In an embodiment of the present invention, the image acquisition unit includes a traffic camera, an image preprocessing module, and a first communication module. The traffic camera is deployed on each traffic road to capture pictures of passing vehicles; the image preprocessing module is communicatively connected to the traffic camera and is used for preprocessing the pictures of passing vehicles by performing data enhancement, normalization, and grayscale processing; the first communication module is connected to the cloud storage computing platform using a 5G network and uploads the preprocessed vehicle images to the cloud storage computing platform;

[0084] The temperature detection unit includes a temperature sensor and a third communication module. The temperature sensor is installed beside the traffic camera and is used for monitoring the ambient temperature; the third communication module is connected to the cloud storage computing platform using a 5G network and uploads the ambient temperature to the cloud storage computing platform.

[0085] The positioning unit includes a positioning module and a fourth communication module. The positioning module is used for obtaining the location of the traffic camera and further obtaining the location of the electric vehicle at the shooting moment; the fourth communication module is connected to the cloud storage computing platform using a 5G network and uploads the location information data to the cloud storage computing platform.

[0086] The vehicle networking service system includes a database module and a charging record module. The database module is used for storing charging station information in various places; the charging record module is used for storing the historical charging data of various electric vehicles; the air conditioner usage status record module is used for storing the historical air conditioner usage status data of various electric vehicles; the fifth communication module is connected to the cloud storage computing platform using a 5G network and uploads the charging station information in various places, the historical charging data of electric vehicles, and the historical air conditioner usage status data to the cloud storage computing platform.

[0087] In an embodiment of the present invention, the cloud storage computing platform includes a historical image database module, an image recognition module, a model training module, a model prediction module, and a second communication module. The historical image database module is communicatively connected to the image preprocessing module and is used to store preprocessed historical vehicle pictures to form a historical database. The image recognition module is communicatively connected to the image preprocessing module and is used to identify license plate information, the head deflection action of the vehicle owner, and the facial expression features of the vehicle owner in the processed vehicle pictures. The model training module identifies license plate information, the head deflection action of the vehicle owner, and the facial expression features from the historical vehicle picture data, and trains a charging load prediction model in combination with the historical air conditioner opening state data and historical charging records of electric vehicles. The model prediction module is used to receive in real time the vehicle pictures collected and processed by traffic cameras and the real-time temperature data of temperature sensors, classify the data identified from the vehicle pictures according to the temperature data range, input them into the corresponding charging load prediction models, obtain the probability of when and where the target will charge and the charging load, and through time accumulation, obtain the charging facility planning and construction in certain areas. The second communication module uses a 5G network to connect the image acquisition unit, the temperature detection unit, the vehicle networking service system, and the positioning unit.

[0088] Another embodiment of the present invention provides a charging load probability prediction method based on non-intrusive detection, as Figure 2 shown, including the following steps:

[0089] S10. Shoot and collect historical vehicle pictures passing by through traffic cameras deployed on the road to establish a picture database;

[0090] S20. Process the historical vehicle pictures to obtain license plate information, the head deflection action of the vehicle owner, and the facial expression features;

[0091] S30. Based on the license plate information, obtain the charging records of the electric vehicle through the vehicle networking system, including the charging location and charging time;

[0092] S40. Use the head deflection action and facial expression features of the vehicle owner as independent variables, and the charging time and charging location as dependent variables to establish a charging load demand prediction model for each vehicle owner respectively;

[0093] S50. Obtain the historical air conditioner usage status of the electric vehicle through the vehicle networking system, and train and optimize the charging load demand prediction model through the air conditioner opening information to obtain a first load demand prediction model when the air conditioner is on and a second load demand prediction model when the air conditioner is off;

[0094] S60. Detect the ambient temperature through the temperature sensor beside the traffic camera, clarify the corresponding temperature of the photo taken at a certain moment, determine whether the in-vehicle air conditioner is turned on, and set that when the ambient temperature detected by the temperature sensor is 14°C to 28°C, it is the condition for not turning on the in-vehicle air conditioner, otherwise it is the condition for turning on the in-vehicle air conditioner. If the in-vehicle air conditioner is turned on, input it into the first load demand prediction model for charging load prediction; if the air conditioner is not turned on, input it into the second load demand prediction model for charging load prediction;

[0095] S70. Then, combined with the positioning unit of the traffic camera, it is possible to obtain the probability and charging load prediction of when and where an electric vehicle will charge at a certain place and time.

[0096] In an embodiment of the present invention, the picture database is composed of historical picture data of m cameras, t moments, and n targets, with a total of mnt basic pictures.

[0097] In an embodiment of the present invention, the identification and acquisition of license plate information include the following steps:

[0098] B10. Preprocess the license plate image, convert the vehicle picture from the RGB channel to the HSV channel, convert the HSV channel image into a grayscale image, and then perform binarization and morphological processing on the grayscale image;

[0099] B20. License plate positioning, use the function cv2.findContours() to perform rectangular detection on the grayscale image after morphological processing, locate the license plate area, and segment out the license plate area;

[0100] B30. License plate character segmentation, perform license plate horizontal correction, removal of license plate borders and rivets, and character segmentation operations on the segmented license plate area in sequence. License plate horizontal correction includes tilt angle detection and tilt correction;

[0101] B40. License plate character recognition, convert the characters on the image into feature vectors through HOG feature extraction, and classify and discriminate through the SVM classification algorithm;

[0102] B50. Obtain the result, and identify whether it is a new energy vehicle and the license plate number through classification and discrimination.

[0103] In an embodiment of the present invention, the acquisition of the owner's head deflection action is as follows:

[0104] Perform data processing on the pictures of passing vehicles, obtain vehicle type information and the owner's driving position information, and judge the corresponding direction of the owner's head deflection direction and the dashboard through the vehicle type information and the owner's driving position information;

[0105] According to the law of human head movement, under the condition that it is assumed that the human head rotates along the central axis of the head without swinging left and right in the vertical direction, the head rotation model of the driver can be simplified as shown in Figure 3 the following model. The line segment connecting the two eyes when the driver's head is facing directly forward is represented as line segment After the head rotates through an angle of θ, the spatial connection line segment between the two eyes is represented as line segment Since the distance between a person's two eyes does not change, therefore, numerically, line segment and are equal. The projection of the distance between the two eyes after head rotation on the positive plane is represented as line segment That is, the distance between the two eyes shown in the two-dimensional image after the head rotates. Through geometric analysis, it can be obtained that the acute angle between line segment and line segment is equal to the angle through which the head rotates. Then, the head rotation angle calculation formula is expressed as

[0106]

[0107] If the calculated θ value is negative, it is determined that the head rotates to the left; if the θ value is positive, it is determined that the head rotates to the right. If the head rotation angle to the right is greater than 30 degrees, it is determined that the driver turns his head to look at the dashboard.

[0108] In an embodiment of the present invention, the acquisition of the facial expression characteristics of the vehicle owner is as follows:

[0109] In the feature sampling layer, image data preprocessing, face detection, face localization, convolutional feature learning, and feature sampling are performed on the vehicle pictures. The original vehicle image sequence x = {x 1 , x 2 , …, x T} is input. The image is preprocessed to eliminate the influence of illumination. At time t = 1, the Fast-CNN method is used to quickly locate the face image and segment it. At subsequent times, the face image is quickly tracked and segmented. Then, the face image enters the convolutional neural network for learning, and through cross-convolution and pooling, image abstract features are generated. Finally, it is input into the K-means sampling layer, and this layer performs average sampling on a total of K consecutive image features of x T-K+1 , x T-K+2 , …, x T as the input feature to enter the recurrent network for learning;

[0110] The RNN main loop and LSTM information memory take the convolutional sampling feature vector obtained from the feature sampling layer as the input; then, according to the time series, it enters the recurrent network, and the recurrent unit LSTM extracts information to generate state information; finally, the output feature vector is classified;

[0111] Classify the output sequence of the feature vectors learned by the recurrent network layer. Let T be the sequence length and L be the length of the label. Then, when the sequence T appears, there are L^T possible label paths, and each of these possibilities is called a "Path". Its conditional probability formula

[0112]

[0113] where, π (t) represents the output path π at time t, and y t represents the network output of the RNN at time t.

[0114] The predicted label conditional probability is expressed as the sum of the conditional probabilities of the corresponding Paths.

[0115]

[0116] where, V represents the operator that converts the output path π into the target L.

[0117] When the input sequence is being trained, the label with the maximum conditional probability for the input image expression at the current moment is:

[0118]

[0119] Identify various expressions of the car owner, such as anger, disgust, fear, happiness, sadness, surprise, and neutral.

[0120] In an embodiment of the present invention, the charging load demand prediction model is established as follows:

[0121] Assume that N sets of data are used for model training, and grey relational analysis is performed:

[0122] Assume that N sets of data are used for model training, and random forest algorithm analysis is performed:

[0123] a. Take the charging behavior of the electric vehicle as the prediction variable X o , and take the eight factors of anger, disgust, fear, happiness, sadness, surprise, neutral, and head deflection shown by the driver's face within one hour before charging as the influencing variable X i ;

[0124] b. Let the number of sampling times b = 1, 2,..., B s , and repeat steps c and d;

[0125] c. Randomly select a sub-sample set from X o , X i as the training set through the self-repeated sampling technique;

[0126] d. Obtain a regression tree model r through the training set fb;

[0127] After the training is completed, for a new sample x, the random forest model gives the predicted value of the sample by averaging the predicted values of all regression trees. It is:

[0128]

[0129] The above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the purpose and scope of the technical solutions of the present invention, and they should all be covered within the scope of the claims of the present invention. The technologies, shapes, and structures not detailedly described in the present invention are all well-known technologies.

Claims

1. A charging load probability prediction system based on non-invasive detection, characterized in that, it includes: An image acquisition unit that captures pictures of passing vehicles through traffic cameras deployed on various roads and processes the picture data; A temperature detection unit that obtains the ambient temperature through a temperature sensor installed beside the traffic camera; A positioning unit installed on the traffic camera to obtain the position of the traffic camera, and thus obtain the position of the electric vehicle at the shooting moment; A vehicle networking service system that obtains information about surrounding charging stations based on the geographical location of the traffic camera; A cloud storage and computing platform, which is communicatively connected to the image acquisition unit, the temperature detection unit, the vehicle networking service system, and the positioning unit, and is used for processing and storing the historical vehicle pictures collected by the image acquisition unit, and for identifying license plate information, the deflection action of the vehicle owner's head, and the facial expression features of the vehicle owner from the historical vehicle picture data, combining the historical air conditioner on / off state data and historical charging records of electric vehicles to train a charging load prediction model, and for receiving in real time the vehicle pictures collected and processed by the traffic camera and the real-time temperature data of the temperature sensor, classifying the data obtained by identifying the vehicle pictures according to the temperature data range, inputting them into the corresponding charging load prediction models, obtaining the probability of when and where the target will charge and the charging load, and through time accumulation, being used for the planning and construction of charging facilities in certain areas.

2. The charging load probability prediction system based on non-invasive detection according to claim 1, characterized in that, the image acquisition unit includes: Traffic cameras deployed on various traffic roads for capturing pictures of passing vehicles; An image preprocessing module communicatively connected to the traffic camera for performing data enhancement, normalization, and grayscale preprocessing on the passing vehicle pictures; A first communication module that uses a 5G network to connect to the cloud storage and computing platform and uploads the preprocessed vehicle images to the cloud storage and computing platform; The temperature detection unit includes: A temperature sensor installed beside the traffic camera for monitoring the ambient temperature; A third communication module that uses a 5G network to connect to the cloud storage and computing platform and uploads the ambient temperature to the cloud storage and computing platform; The positioning unit includes: A positioning module for obtaining the position of the traffic camera, and thus obtaining the position of the electric vehicle at the shooting moment; A fourth communication module that uses a 5G network to connect to the cloud storage and computing platform and uploads the position information data to the cloud storage and computing platform; The vehicle networking service system includes: A database module for storing information about charging stations in various places; A charging record module for storing historical charging data of various electric vehicles; An air conditioner usage status record module for storing historical air conditioner usage status data of various electric vehicles; A fifth communication module that uses a 5G network to connect to the cloud storage and computing platform and uploads the information about charging stations in various places, the historical charging data of electric vehicles, and the historical air conditioner usage status data to the cloud storage and computing platform.

3. The charging load probability prediction system based on non-invasive detection according to claim 2, characterized in that, The cloud storage computing platform includes: A historical image database module for storing pre - processed historical vehicle pictures to form a historical database; An image recognition module for recognizing license plate information, the deflection action of the vehicle owner's head, and the facial expression features of the vehicle owner in the processed vehicle pictures; A model training module for training a charging load prediction model by identifying license plate information, the deflection action of the vehicle owner's head, and the facial expression features from historical vehicle picture data, and combining the historical air - conditioner opening state data and historical charging records of electric vehicles; A model prediction module for receiving in real - time the vehicle pictures collected and processed by the traffic camera and the real - time temperature data of the temperature sensor, classifying the data obtained by recognizing the vehicle pictures according to the temperature data range, inputting them into the corresponding charging load prediction model, obtaining the probability of when and where the target will charge and the charging load, and through time accumulation, obtaining the charging facility planning and construction in certain areas; A second communication module using a 5G network to connect the image acquisition unit, the temperature detection unit, the vehicle - to - everything service system, and the positioning unit.

4. A method for predicting the probability of charging load based on non - invasive detection, Characterized in that, It includes the following steps: Taking pictures of passing historical vehicle pictures through traffic cameras deployed on the road to establish a picture database; Processing the historical vehicle pictures to obtain license plate information, the deflection action of the vehicle owner's head, and the facial expression features; Based on the license plate information, obtaining the charging record of the electric vehicle through the vehicle - to - everything system, including the charging location and charging time; Taking the deflection action of the vehicle owner's head and the facial expression features as independent variables, and the charging time and charging location as dependent variables, respectively establishing a charging load demand prediction model for each vehicle owner; Obtaining the historical air - conditioner usage status of the electric vehicle through the vehicle - to - everything system, and training and optimizing the charging load demand prediction model through the air - conditioner opening information to obtain a first load demand prediction model when the air - conditioner is on and a second load demand prediction model when the air - conditioner is off; Detecting the ambient temperature through a temperature sensor beside the traffic camera, clarifying the corresponding temperature of the picture taken at a certain moment, judging whether the in - vehicle air - conditioner is turned on. If the in - vehicle air - conditioner is turned on, inputting it into the first load demand prediction model for charging load prediction; if the air - conditioner is not turned on, inputting it into the second load demand prediction model for charging load prediction; Among them, it is set that when the ambient temperature detected by the temperature sensor is 14℃ - 28℃, it is the condition of not turning on the in - vehicle air - conditioner, otherwise it is the condition of turning on the in - vehicle air - conditioner; Combined with the positioning unit of the traffic camera, the probability of when and where an electric vehicle will charge at a certain moment and the charging load prediction can be obtained.

5. The method for predicting the probability of charging load based on non - invasive detection according to claim 4, Characterized in that, The picture database is composed of historical picture data of m cameras, t moments, and n targets, with a total of mnt basic pictures.

6. The method for predicting the probability of charging load based on non - invasive detection according to claim 5, Characterized in that, The recognition and acquisition of the license plate information includes the following steps: Preprocessing of license plate images, converting vehicle pictures from the RGB channel to the HSV channel, converting the HSV channel images into grayscale images, and then performing binarization and morphological processing on the grayscale images; License plate location, using the function cv2.findContours() to perform rectangular detection on the grayscale image after morphological processing, locate the license plate area, and segment out the license plate area; License plate character segmentation, successively performing license plate horizontal correction, removal of license plate borders and rivets, and character segmentation operations on the segmented license plate area. The license plate horizontal correction includes tilt angle detection and tilt correction; License plate character recognition, converting the characters on the image into feature vectors through HOG feature extraction, and classifying and discriminating through the SVM classification algorithm; Obtain the results, and identify whether it is a new energy vehicle and the license plate number through classification and discrimination.

7. The method for predicting the charging load probability based on non-intrusive detection according to claim 6, characterized in that, the acquisition of the head deflection action of the vehicle owner is as follows: Perform data processing on pictures of passing vehicles, obtain vehicle type information and the driving position information of the vehicle owner, and judge the corresponding direction between the head deflection direction of the vehicle owner and the dashboard through the vehicle type information and the driving position information of the vehicle owner; According to the law of human head movement, on the condition that it is assumed that the human head rotates along the central axis of the head without swinging left and right in the vertical direction, the line segment formed by connecting the two eyes when the driver's head is facing directly forward is represented as line segment The spatial connection line segment between the two eyes after the head rotates through an angle θ is represented as line segment Numerically, line segment and are equal. The projection of the distance between the two eyes on the positive plane after the head rotates is represented as line segment Through geometric analysis, it can be obtained that the acute angle between line segment and line segment is equal to the angle through which the head rotates. Then the head rotation angle calculation formula is expressed as If the calculated θ value is negative, it is judged as a left head rotation, and if the θ value is positive, it is judged as a right head rotation.

8. The method for predicting the charging load probability based on non-intrusive detection according to claim 7, characterized in that, the acquisition of the facial expression features of the vehicle owner is as follows: In the feature sampling layer, image data preprocessing, face detection, face localization, convolutional feature learning, and feature sampling are performed on vehicle pictures. The original vehicle image sequence x = {x 1 , x 2 , …, x T} is input. Image preprocessing is carried out to eliminate the influence of illumination. At time t = 1, the Fast-CNN method is used to quickly locate the face image and perform segmentation. At subsequent times, the face image is quickly tracked and segmented. Then, the face image enters the convolutional neural network for learning, and through cross-convolution and pooling, abstract image features are generated. Finally, it is input into the K-means sampling layer, which performs average sampling on a total of K consecutive image features of x T-K+1 , x T-K+2 , …, x T and uses the result as the input feature to enter the recurrent network for learning; RNN main loop and LSTM information memory, obtaining the convolutional sampling feature vector from the feature sampling layer as the input; Then, enter the recurrent network according to the time series, and the recurrent unit LSTM extracts information to generate state information; finally, output the feature vector for classification; Classify the output sequence of the feature vectors learned by the recurrent network layer. Let T be the sequence length and L be the length of the label. Then when the sequence T appears, there are L to the power of T possible label paths, and each of them is used as a "Path", and its conditional probability formula Among them, π (t) represents the output path π at time t, and y t represents the network output of the RNN at time t. The predicted label conditional probability is expressed as the sum of the conditional probabilities of the corresponding Paths, where V represents the operator that converts the output path π into the target L, When the input sequence is trained, the input image expression at the current moment is the label with the maximum conditional probability: Identify various expressions of the vehicle owner, such as anger, disgust, fear, happiness, sadness, surprise, and neutral.

9. The method for predicting the charging load probability based on non-intrusive detection according to claim 8, characterized in that, the establishment of the charging load demand prediction model is as follows: Assume that N groups of data are used for model training and perform random forest algorithm analysis: a. Use the charging behavior of the electric vehicle as the predictive variable X o , and take the eight factors of anger, disgust, fear, happiness, sadness, surprise, neutrality, and head deflection shown on the driver's face within one hour before charging as the influencing variable X i ; b. Let the number of sampling times \(b = 1, 2, \ldots, B\) s , and repeat steps c and d; c. Randomly select a subset from X o , X i as the training set through the independent repeated sampling technique; d. Obtain a regression tree model r from the training set fb ; After training, for a new sample x, the random forest model gives the predicted value of this sample by averaging the predicted values of all regression trees as follows:

10. The method for predicting the charging load probability based on non-intrusive detection according to claim 9, characterized in that, in the step of judging whether the in-vehicle air conditioner is turned on, it is set that when the ambient temperature detected by the temperature sensor is 14°C to 28°C, it is the condition for not turning on the in-vehicle air conditioner, otherwise it is the condition for turning on the in-vehicle air conditioner.

Images