A method and system for protecting a vehicle from a collision during a charging process

Abstract

The application discloses a safety protection method and system after collision in a vehicle charging process, and belongs to the technical field of automobile safety, comprising the steps of multi-source perception data acquisition, Kalman filter fusion processing, slip amount prediction, false trigger classification, joint decision, protection execution and safety state prompting. The application inputs vehicle motion state data into a slip amount prediction model based on a long short-term memory network, obtains a vehicle future trajectory point coordinate sequence, and performs geometric overlap determination on the sequence, the charging pile position coordinates and the adjacent vehicle position coordinates. The predictive protection instruction is triggered 0.5s before the collision occurs. The problems of passive response after the collision and lag of high-pressure system cut-off in the traditional collision protection are solved. The safety protection window is changed from millisecond-level post-response to second-level pre-judgment. The collision probability and the secondary disaster risk after the collision are reduced.

Description

Technical Field

[0001] This application belongs to the field of automotive safety technology, and specifically relates to a safety protection method and system for vehicles after a collision during charging. Background Technology

[0002] With the rapid development of the new energy vehicle industry and the continuous expansion of charging infrastructure, vehicle safety during the charging process has become a key focus of the industry.

[0003] Existing technologies face various challenges in addressing the risks of charging collisions. Traditional collision sensors are not linked to the charging status, so airbags may fail to deploy because the trigger threshold is not reached; high-voltage relays cannot cut off power within a few hundred milliseconds, resulting in the battery pack continuously outputting high voltage and posing a risk of electric shock; charging harnesses have insufficient impact resistance, and they are prone to detachment during lateral vehicle movement, causing arcing and burning of the charging port; after the harness is broken, there is a lack of graded power-off protection, and the exposed copper core may cause a short circuit and fire; existing systems can only respond passively after a collision and cannot predict the risk of overlap between the vehicle's sliding trajectory and the charging station or adjacent vehicles; after a collision, the vehicle's low-voltage battery is prone to failure, causing the domain controller and battery management system to lose power and become paralyzed, preventing critical protective actions from being performed. Summary of the Invention

[0004] To address the aforementioned issues, this application provides a safety protection method and system for vehicles after a collision during charging. This method solves the problems of traditional collision protection, which can only respond passively after a collision and has a delayed high-voltage system cutoff. It also solves the problem of electric arc burns caused by live plugging and unplugging due to the high-voltage arc not being completely extinguished before physical separation, and the problem of power loss and paralysis of the domain controller and battery management system after the main power supply circuit fails due to a vehicle collision.

[0005] In a first aspect, embodiments of this application provide a safety protection method after a collision during vehicle charging, comprising the following steps: Acquire vehicle perimeter radar point cloud data, charging port housing vibration data, vehicle visual image data, and vehicle motion state data to obtain a multi-source perception dataset. The vehicle perimeter radar point cloud data, charging port housing vibration data and vehicle-mounted visual image data are fused to obtain the object motion state estimation results. Input the vehicle motion state data into the preset slip prediction model to obtain the prediction result of the vehicle's future trajectory; The vehicle perimeter radar point cloud data, charging port housing vibration data and vehicle visual image data are input into a preset false trigger classification model to obtain the risk level and false trigger probability value. Based on the object motion state estimation results, vehicle future trajectory prediction results, risk level and false trigger probability value, and the preset joint judgment conditions, a protection decision instruction is obtained. Based on the protection decision instructions, the system performs active power disconnection of the high-voltage circuit, physical isolation of the charging port, and occupant protection actions, resulting in charging collision safety protection.

[0006] In one implementation, a multi-source sensing dataset is obtained, specifically including: Imaging millimeter-wave radar deployed around the vehicle collects radar point cloud data around the vehicle. Vibration data of the charging port housing is collected by a vibration sensor deployed inside the charging port housing. Acquire vehicle-mounted visual image data using vehicle-mounted visual acquisition equipment; Vehicle motion status data is acquired through wheel speed sensors and inertial measurement units; By aligning the vehicle's perimeter radar point cloud data, charging port housing vibration data, vehicle-mounted visual image data, and vehicle motion state data in a time sequence, a multi-source perception dataset is obtained.

[0007] In one embodiment, obtaining the object's motion state estimation result specifically includes: Based on the vehicle perimeter radar point cloud data, charging port housing vibration data and vehicle visual image data, state vectors and observation vectors are constructed, and the observation noise covariance matrix is ​​set. The weights of the observation noise covariance matrix are adjusted based on the real-time signal-to-noise ratio of each data point. Under abnormal lighting conditions, the weights of the onboard visual image data are reduced and the weights of the vehicle perimeter radar point cloud data are increased. When continuous micro-vibration is detected, the weights of the charging port housing vibration data are increased. When the deviation between a single data output and the fusion estimation result exceeds a preset threshold and continues for a preset period, the weight of that data is reduced and a degradation mode is triggered. When multiple data fail simultaneously, a safety mode is triggered. The state vector is updated based on the adjusted weights to obtain the object's motion state estimation result.

[0008] In one implementation, obtaining the vehicle's future trajectory prediction result specifically includes: Vehicle wheel speed data, vehicle acceleration data, and road friction coefficient are extracted from vehicle motion state data to construct a time-series input sequence; Input the time-series input sequence into the preset slip prediction model, and output the future trajectory point coordinate sequence of the vehicle; The coordinate sequence of the vehicle's future trajectory points is geometrically overlapped with the coordinates of the charging pile and the coordinates of adjacent vehicles. If trajectory overlap is found, a trajectory overlap warning signal is generated. The trajectory overlap warning signal is combined with the coordinate sequence of the vehicle's future trajectory points to form the prediction result of the vehicle's future trajectory.

[0009] In one implementation, obtaining the risk level and the probability of false triggering includes: Input the vehicle's perimeter radar point cloud data, charging port housing vibration data, and vehicle-mounted visual image data into a preset false trigger classification model, and output the initial risk level and false trigger probability value. When the probability of false triggering exceeds the preset filtering threshold, the initial risk level output is suppressed and the current state is maintained; When the false trigger probability value does not exceed the preset filtering threshold, the initial risk level is selectively increased based on whether the vehicle's future trajectory prediction results contain trajectory overlap warning signals, thus obtaining the risk level.

[0010] In one implementation, obtaining a protection decision instruction specifically includes: When the risk level reaches the preset high-voltage trigger level and the object's speed exceeds the preset speed threshold, a high-voltage cutoff command is generated. When the risk level reaches the preset airbag trigger level and the presence of an occupant is detected, an airbag deployment command is generated. When the risk level reaches the preset airbag trigger level and no occupant is detected, a high-pressure cutoff command is generated. The command to disconnect high voltage, the command to deploy airbags, or the command to disconnect high voltage only shall be determined as the protection decision command.

[0011] In one embodiment, based on the protection decision command, active power disconnection of the high-voltage circuit, physical isolation of the charging port, and occupant protection actions are executed, specifically including: The high-voltage busbar short circuit is extinguished by using a pyrotechnic short-circuit switch; After the high-voltage bus short circuit is extinguished, a release command is sent to the charging port mechanical pin mechanism after a preset safety delay time. The mechanical pin mechanism of the charging port performs a pop-out action according to the release command, completing the physical isolation of the charging port; Depending on whether the protection decision command includes an airbag deployment command, the airbag deployment action is selectively executed.

[0012] In one embodiment, the process of performing active power disconnection of the high-voltage circuit, physical isolation of the charging port, and occupant protection actions further includes: It adopts a dual-redundant power supply architecture, one of which is the power supply circuit for the vehicle's low-voltage battery, and the other is an independent low-voltage circuit for the power battery; By connecting the supercapacitor module in parallel with the vehicle's low-voltage battery, a buffer power supply is provided during power switching. When the on-board low-voltage battery voltage is detected to be lower than the preset switching threshold or a collision risk signal is received, the system automatically switches to the independent low-voltage circuit of the power battery to ensure that the control unit is continuously powered.

[0013] In one embodiment, after obtaining the charging collision safety protection result, the method further includes: Generate corresponding safety status indication signals based on risk levels; The safety status indication signal is output to the vehicle human-machine interface and the light-emitting indicator around the charging port to provide multimodal safety status prompts.

[0014] Secondly, embodiments of this application provide a safety protection system for vehicles after a collision during charging, including: The multi-source acquisition module is used to acquire point cloud data of the vehicle's perimeter radar, vibration data of the charging port housing, vehicle-mounted visual image data, and vehicle motion state data to obtain a multi-source perception dataset. The data fusion module is used to fuse the point cloud data of the vehicle's perimeter radar, the vibration data of the charging port housing, and the vehicle-mounted visual image data to obtain the estimation result of the object's motion state. The trajectory prediction module is used to input vehicle motion state data into a preset slip prediction model to obtain the predicted trajectory of the vehicle in the future. The risk identification module is used to input the vehicle's surrounding radar point cloud data, charging port housing vibration data and vehicle visual image data into a preset false trigger classification model to obtain the risk level and false trigger probability value. The joint assessment module is used to compare the object motion state estimation results, vehicle future trajectory prediction results, risk level and false trigger probability value with preset joint assessment conditions to obtain protection decision instructions. The protection execution module is used to execute active power disconnection of the high-voltage circuit, physical isolation of the charging port and occupant protection actions according to the protection decision instructions, so as to obtain the charging collision safety protection results.

[0015] Compared with the prior art, this application has the following advantages: (1) This invention obtains the future trajectory point coordinate sequence of the vehicle by inputting the vehicle motion state data into the slip amount prediction model based on the long short-term memory network, and performs geometric overlap determination with the charging pile location coordinate and the adjacent vehicle location coordinate. The predictive protection command is triggered 0.5s before the collision occurs, which solves the problem that traditional collision protection can only passively respond after the collision and the high voltage system cut-off is delayed. It realizes the leap from millisecond-level post-event response to second-level pre-event prediction of safety protection window, reducing the probability of collision and the risk of secondary disasters after the collision. (2) The present invention completes the short circuit extinguishing of the high voltage bus within 2ms by using a pyrotechnic short circuit switch, and sends a release command to the mechanical pin mechanism of the charging port after a 3ms safety delay, so that the mechanical pin completes the pop-out action within 5ms. This solves the problem of electric arc burns caused by the high voltage arc not being completely extinguished before physical separation, and realizes the timing coordination of high voltage electrical cut-off and mechanical physical isolation, eliminating the risk of short circuit fire and electric shock caused by the live separation of the charging port; (3) This invention constructs a dual-circuit redundant power supply architecture consisting of an on-board low-voltage battery power supply circuit and an independent low-voltage power battery circuit, and uses a supercapacitor module to provide buffer power during power supply switching. This solves the problem of the domain controller and battery management system losing power and paralyzing after the main power supply circuit fails due to a vehicle collision. It achieves seamless redundancy protection of the control power supply and ensures the reliable execution of key protective actions such as high-voltage power failure and airbag deployment under extreme conditions.

[0016] Other features and advantages of this application will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the application. The objectives and other advantages of this application may be realized and obtained by means of the structures pointed out in the description, claims and drawings. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 A schematic diagram of the process provided by the present invention is shown; Figure 2 The electronic signal control logic matrix provided by this invention is shown. Figure 3 A schematic diagram of the structure provided by the present invention is shown. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0020] See Figure 1 This invention provides a safety protection method for vehicles after a collision during charging. This method achieves end-to-end charging collision safety protection through steps such as multi-source sensing data acquisition, Kalman filter fusion processing, slip amount prediction, false trigger classification, joint decision-making, protection execution, and safety status indication. The specific implementation methods of each step are described in detail below with reference to embodiments.

[0021] Step 1: Acquire multi-source sensing data The system acquires radar point cloud data around the vehicle, vibration data of the charging port housing, vehicle-mounted visual image data, and vehicle motion state data to obtain a multi-source perception dataset.

[0022] Imaging millimeter-wave radars deployed around the vehicle collect radar point cloud data around the vehicle. The radars are deployed on both sides of the front bumper, both sides of the rear bumper, and the left and right sides of the vehicle corresponding to the charging port, totaling 6 radar nodes, covering 360° circumferential detection. The radars are connected to the vehicle domain controller via a flexible data rate bus on the controller area network, with a data transmission rate of 500kbps and a data frame format conforming to the International Organization for Standardization ISO 11898-1 standard. The radar ranging range is 0.1m to 200m with a ranging accuracy of ±0.1m, the speed range is -150km / h to 150km / h with a speed accuracy of ±0.5km / h, and the angular resolution is ±1° horizontally and ±2.5° vertically.

[0023] Vibration data of the charging port housing is collected by a vibration sensor deployed inside the charging port housing. The vibration sensor is mechanically isolated from the charging port electromagnetic lock. The vibration sensor has a sensitivity of 0.01g, a sampling frequency of 1kHz, and a range of 0.01g to 1g. When the range is exceeded, a saturation value is output. The vibration sensor output is a digital signal, which is directly connected to the vehicle domain controller through the integrated circuit bus protocol.

[0024] The vehicle-mounted visual image data is acquired through an in-vehicle visual acquisition device. The device has a field of view of 120° horizontally and 90° vertically, ensuring that there are no blind spots within a 5m radius around the charging port. The in-vehicle visual acquisition device is connected to the vehicle domain controller through a second-generation serial camera interface bus with a mobile industry processor interface, and the frame rate is no less than 15fps.

[0025] Vehicle motion data is acquired through wheel speed sensors and an inertial measurement unit. The wheel speed sensors output four-wheel speed pulse signals, and the inertial measurement unit outputs three-axis acceleration and three-axis angular velocity signals.

[0026] The above four types of data are time-series aligned. Using the system clock of the vehicle domain controller as a unified time reference, timestamps are marked on each data. Based on the nearest neighbor principle of timestamps, the vehicle perimeter radar point cloud data, charging port housing vibration data, vehicle visual image data, and vehicle motion state data are aligned to the same time to obtain a multi-source perception dataset.

[0027] Step 2: Kalman filter fusion processing Kalman filtering is used to fuse the vehicle's perimeter radar point cloud data, charging port housing vibration data, and vehicle-mounted visual image data. The fusion weights are dynamically adjusted based on the real-time signal-to-noise ratio of each data point to obtain the object's motion state estimation result.

[0028] State vectors and observation vectors are constructed based on vehicle perimeter radar point cloud data, onboard visual image data, and charging port housing vibration data. The state vector includes the object's position, velocity, acceleration, and category identifier in three-dimensional space, specifically represented as follows: ; in, These are the position coordinates of the object in three-dimensional space. Let be the velocity components of the object in three-dimensional space. The acceleration components of an object in three-dimensional space. Superscript is used to identify object categories. This represents the transpose of a vector.

[0029] The observation vector is constructed from vehicle perimeter radar point cloud data, onboard visual image data, and charging port housing vibration data after feature extraction, and is specifically represented as follows: ; in, These are the object position and velocity observations obtained by clustering the radar point cloud data around the vehicle. These are the object position and category observations obtained after target detection from vehicle-mounted visual image data. The impact strength observation value is obtained by frequency domain analysis of the vibration data of the charging port housing.

[0030] Set the state transition matrix With observation matrix State transition matrix Constructed based on a uniformly accelerated motion model; observation matrix Constructed based on the observable dimensions of each data source; define the process noise covariance matrix. Covariance matrix of observation noise Process noise covariance matrix Based on the sensor calibration error and the randomness of target motion, the observation noise covariance matrix is ​​set. It is a three-dimensional diagonal matrix, whose diagonal elements correspond to the observation noise variance of the vehicle perimeter radar point cloud data, vehicle-mounted visual image data, and charging port housing vibration data, respectively.

[0031] The formula for calculating the real-time signal-to-noise ratio is: ; in, For the first Real-time signal-to-noise ratio of each data source For the first The signal power of each data source For the first Noise power of each data source, These correspond to the vehicle's perimeter radar point cloud data, in-vehicle visual image data, and charging port housing vibration data, respectively.

[0032] The adjustment formula for the weights of the observation noise covariance matrix is ​​as follows: ; in, This is the adjusted observation noise covariance matrix. This represents the operation of constructing a diagonal matrix. These are the signal-to-noise ratio weighting coefficients for vehicle perimeter radar point cloud data, in-vehicle visual image data, and charging port housing vibration data, respectively. These are the real-time signal-to-noise ratios of the three types of data mentioned above.

[0033] Under abnormal lighting conditions, the real-time signal-to-noise ratio (SNR) of vehicle-mounted visual image data decreases, and the diagonal elements in its observation noise covariance matrix increase, correspondingly reducing the fusion weight. Specifically, when the image entropy value of the vehicle-mounted visual image data is below 5.0, the visual weight drops to 0.3. When continuous micro-vibration is detected, the real-time SNR of the charging port housing vibration data increases, and its fusion weight increases accordingly. Specifically, when the signal dominant frequency energy of the charging port housing vibration data accounts for more than 80%, the vibration weight increases to 0.5.

[0034] Perform standard Kalman filter recursive operations. The state prediction equation is: ; The covariance prediction equation is: ; The Kalman gain calculation equation is as follows: ; The state update equation is: ; The covariance update equation is: ; in, For the first Prior state estimation at time 10:00 For the first Posterior state estimation at time 10:00. To estimate the covariance matrix a priori, To estimate the covariance matrix posteriorly, For the first The Kalman gain matrix at time t. For the first The observation vector at time t, For the first Posterior state estimation at time 10:00. To estimate the covariance matrix posteriorly, For identity matrix, superscript Indicates matrix transpose, superscript This represents finding the inverse of a matrix.

[0035] When the deviation between the output of a single data source and the fusion estimation result exceeds a preset threshold and persists for a preset period, the weight of that data source is reduced and a degradation mode is triggered. Specifically, if the location deviation exceeds 0.5m, the speed deviation exceeds 2km / h, or the category confidence deviation exceeds 0.3, and these deviations exceed the limits for three consecutive sampling periods, the data source is marked as a suspected fault, its weight is reduced to 0.1, and a degradation mode is triggered. When two or more data sources fail simultaneously, a safety mode is triggered, and the highest level of protection status is directly output.

[0036] The state is updated based on the adjusted weights, and the motion state estimation result of the object is obtained. This result includes the optimal estimates of the object's position, velocity, acceleration and category label in three-dimensional space.

[0037] Step 3: Slip Prediction The vehicle motion state data is input into a preset slip prediction model to obtain the prediction result of the vehicle's future trajectory.

[0038] Vehicle wheel speed data, vehicle acceleration data, and road friction coefficient are extracted from vehicle motion state data to construct a time-series input sequence. The road friction coefficient is estimated through wheel speed fluctuations, and its calculation formula is as follows: ; in, The coefficient of friction of the road surface. The standard deviation of the current wheel speed fluctuation. This is a baseline value for the standard deviation of wheel speed fluctuation under high-friction road surfaces.

[0039] Timing input sequence Represented as: ; in, For time-series input sequences, For the first The input feature vector at time t, , For the first Vehicle wheel speed data at any given time. For the first Vehicle acceleration data at any given time. For the first The coefficient of road surface friction at time t. For time step.

[0040] The preset slip prediction model is constructed using a Long Short-Term Memory (LSTM) network. This network comprises an input layer, a first LSTM layer, a second LSTM layer, a fully connected layer, and an output layer. The input layer receives the temporal input sequence; the first LSTM layer contains 128 hidden units to extract short-term dynamic features from the temporal data; the second LSTM layer contains 128 hidden units to extract long-term dependency features from the temporal data; the fully connected layer maps the output of the second LSTM layer to the trajectory prediction space; and the output layer outputs the coordinate sequence of the vehicle's future trajectory points. The temporal input sequence is constructed by extracting vehicle wheel speed data, vehicle acceleration data, and road friction coefficient from the past 2 seconds, with a time step of 20 and a sampling interval of 0.1 seconds. The output layer outputs the planar coordinate sequence of 5 trajectory points within the next 0.5 seconds, with a sampling interval of 0.1 seconds.

[0041] The formula for updating the cell state of the Long Short-Term Memory layer is: ; ; ; ; ; ; in, Output for the forget gate. For input gate output, The candidate cell state is... For the unit state, For output gate output, Output in hidden state It is the Sigmoid activation function. The hyperbolic tangent activation function is used. These are the weight matrices for the forget gate, input gate, candidate cell state, and output gate, respectively. These are the corresponding bias vectors. This is an element-wise multiplication operation. For the first The input feature vector at time t, For the first Output the hidden state at any given moment.

[0042] The training process of the pre-defined slip prediction model is as follows: A training dataset is constructed, collecting vehicle motion state data and corresponding real future trajectory data under different road conditions and external force scenarios. Vehicle wheel speed data, vehicle acceleration data, and road friction coefficient are used as input samples, and the coordinates of vehicle trajectory points within the next 0.5 seconds are used as output labels. The sampling interval is 0.1 seconds, and the number of output trajectory points is 5. The mean squared error loss function is used, and its formula is: ; in, This is the mean squared error loss value. The total number of training samples, The number of trajectory points for each sample and It equals 5. For the j-th sample, the first... The true coordinates of each trajectory point For the j-th sample, the first... The predicted coordinates of each trajectory point were used. An adaptive moment estimation optimizer was employed for iterative updates of the model parameters, with an initial learning rate of 0.001. Iteration continued until the loss function converged or a preset number of iterations (500) were reached, resulting in a trained prediction model for the preset slip amount. Validation results showed that the average trajectory prediction error on the test set was less than 0.15m, and the maximum error was less than 0.3m.

[0043] The time-series input sequence is input into the trained preset slip prediction model, which outputs a sequence of coordinates of the vehicle's future trajectory points. ; in, This is a sequence of coordinates for the vehicle's future trajectory points. For the first The coordinates of the predicted trajectory point at time [time]. , To predict the number of trajectory points.

[0044] The geometric overlap between the vehicle's future trajectory point coordinate sequence and the charging pile location coordinates and the location coordinates of adjacent vehicles is determined. Using the vehicle's circumscribed rectangle as a reference, the Euclidean distance between each trajectory point in the vehicle's future trajectory point coordinate sequence and the charging pile location coordinates and the location coordinates of adjacent vehicles is calculated using the following formula: ; in, For the first The Euclidean distance between the predicted trajectory point and the target obstacle. For the first The planar coordinates of the predicted trajectory points The coordinates of the target obstacle are given. The target obstacle includes the charging station and adjacent vehicles.

[0045] If the Euclidean distance between any trajectory point and the target obstacle is less than a preset safe distance threshold, trajectory overlap is determined, and a trajectory overlap warning signal is generated. The trajectory overlap warning signal is combined with the vehicle's future trajectory point coordinate sequence to form the vehicle's future trajectory prediction result.

[0046] Step 4: Accidental Trigger Classification The vehicle's perimeter radar point cloud data, charging port housing vibration data, and vehicle-mounted visual image data are input into a preset false trigger classification model to obtain the risk level and false trigger probability value.

[0047] The pre-defined false trigger classification model is constructed using a Long Short-Term Memory (LSTM) network, comprising an input layer, an LSM layer, a fully connected layer, a risk level output branch, and a false trigger probability output branch. The input layer receives multi-source feature vectors formed by feature extraction from vehicle perimeter radar point cloud data, charging port housing vibration data, and in-vehicle visual image data. The LSM layer contains 128 hidden units and models the temporal dependencies of the multi-source feature vectors. The fully connected layer maps the output of the LSM layer to the classification space. The risk level output branch outputs the initial risk level using the Softmax function. The false trigger probability output branch outputs the false trigger probability value using the Sigmoid function. The risk levels are divided into four levels: Level 0 (Safe), Level 1 (Warning), Level 2 (Danger), and Level 3 (Emergency). The criteria for Level 1 (Warning) are: vibration data of the charging port housing detects a signal greater than 0.01g for 50ms, and the vehicle's perimeter radar point cloud data confirms the object's distance is less than 5m; or, vehicle-mounted visual image data and vehicle perimeter radar point cloud data detect an object's speed greater than 10km / h and a distance less than 3m, with a false trigger probability not exceeding 0.85. The criteria for Level 2 (Danger) are: a preset slip prediction model determines that the trajectory overlaps with the charging pile or adjacent vehicle within 0.5s, with a false trigger probability not exceeding 0.85. The criteria for Level 3 (Emergency) are: vibration data of the charging port housing detects an impact signal greater than 0.5g, or vehicle perimeter radar point cloud data confirms an object's speed greater than 30km / h and a distance less than 1m.

[0048] The formula for calculating the Softmax function is: ; in, The output probability is the k-th risk level. This is the logical value corresponding to the k-th risk level in the output of the fully connected layer, where K is the total number of risk levels.

[0049] The formula for calculating the Sigmoid function is: ; in, This is the probability value for false triggering. This is the logical value corresponding to the erroneously triggered branch in the output of the fully connected layer.

[0050] The training process for the pre-defined false trigger classification model is as follows: A training dataset is constructed by collecting historical vehicle perimeter radar point cloud data, charging port housing vibration data, and vehicle-mounted visual image data under charging scenarios. The corresponding true risk levels and false trigger labels are manually labeled. Risk levels are divided into four levels: Level 0 (Safe), Level 1 (Warning), Level 2 (Danger), and Level 3 (Emergency). The risk level output branch is trained using the cross-entropy loss function, with the following formula: ; in, This represents the cross-entropy loss value. The total number of training samples, The total number of risk levels and It equals 4. For the first The sample at the th A true label for each risk level. For the first The sample at the th The predicted probabilities for each risk level are used. The false trigger probability output branch is trained using a binary cross-entropy loss function, with the following formula: ; in, This represents the binary cross-entropy loss value. For the first The false triggering of the true label in a sample, For the first The false trigger prediction probability of each sample. The total loss function is the weighted sum of the two branch loss functions, and its formula is: ; in, This is the total loss value. These are the loss weight coefficients for the risk level branch and the false trigger probability branch, respectively. An adaptive moment estimation optimizer is used for end-to-end training, iterating until the total loss function converges, thus obtaining the pre-set false trigger classification model after training.

[0051] The vehicle's surrounding radar point cloud data, charging port housing vibration data, and vehicle-mounted visual image data are input into the pre-trained false trigger classification model, which outputs the initial risk level and false trigger probability value.

[0052] When the false trigger probability exceeds the preset filtering threshold of 0.85, the initial risk level output is suppressed and the current state is maintained. When the false trigger probability does not exceed 0.85, the initial risk level is selectively increased based on whether the vehicle's future trajectory prediction result contains a trajectory overlap warning signal. Specifically, if the vehicle's future trajectory prediction result contains a trajectory overlap warning signal, the initial risk level is increased by one level; if it does not contain a trajectory overlap warning signal, the initial risk level remains unchanged. The final risk level is then obtained.

[0053] Step 5: Joint Decision-Making The protection decision instruction is obtained by comparing the object motion state estimation result, the vehicle future trajectory prediction result, the risk level and the false trigger probability value with the preset joint judgment conditions.

[0054] The preset joint judgment conditions include high-pressure triggering conditions and airbag triggering conditions. The high-pressure triggering condition is that the risk level reaches the preset high-pressure triggering level, the object speed exceeds the preset speed threshold, and the false trigger probability value does not exceed the preset filtering threshold; the airbag triggering condition is that the risk level reaches the preset airbag triggering level, the presence of an occupant is detected, and the false trigger probability value does not exceed the preset filtering threshold.

[0055] The object's velocity is extracted from the object's motion state estimation results, specifically the magnitude of the velocity vector in the motion state estimation results. Its calculation formula is as follows: ; in, For the object's velocity, This refers to the three-dimensional velocity components in the estimation results of the object's motion state.

[0056] When the risk level reaches level 2 or above, the object speed exceeds 5 km / h, and the probability of false triggering does not exceed 0.85, a high-voltage disconnection command is generated; when the risk level reaches level 3, the presence of occupants is detected, and the probability of false triggering does not exceed 0.85, an airbag deployment command is generated; when the risk level reaches level 3, no occupants are detected, and the probability of false triggering does not exceed 0.85, a high-voltage disconnection command is generated only.

[0057] The command to disconnect high voltage, the command to deploy airbags, or the command to disconnect high voltage only shall be determined as the protection decision command.

[0058] Step 6: Implementation of Protection Measures See Figure 2 Based on the protection decision instructions, the system performs active power disconnection of the high-voltage circuit, physical isolation of the charging port, and occupant protection actions, resulting in charging collision safety protection.

[0059] The high-voltage bus short circuit is extinguished by a pyrotechnic short-circuit switch; the pyrotechnic short-circuit switch has a resistance of 2Ω, an action time of 2ms when the trigger current is 1.2A, or an action time of 0.5ms when the trigger current is 1.75A; the pyrotechnic short-circuit switch is directly connected to the vehicle domain controller pin, and the trigger current is output by the internal capacitor of the vehicle domain controller when triggered.

[0060] After the high-voltage bus short circuit is extinguished, a release command is sent to the charging port mechanical pin mechanism after a 3ms safety delay. This 3ms delay ensures that the high-voltage arc is extinguished before physical separation. The charging port mechanical pin mechanism performs a pop-out action according to the release command, completing the physical isolation of the charging port. The mechanical pin mechanism adopts an electromagnet-driven release spring triggering method, with a pull-in voltage of 12V and a pull-in current of 2A. The spring stiffness is determined through bench testing, the pop-out time does not exceed 5ms, and the pop-out impact force is less than 50N. The mechanical pin mechanism is made of 304 stainless steel with a nickel-plated surface.

[0061] Timing verification results show that the high-voltage bus voltage drops to a safe value 14ms after the initial time, the charging gun physically separates 20ms after the initial time, and no electric arc is generated.

[0062] Depending on whether the protection decision command includes an airbag deployment command, the airbag deployment action is selectively executed. If the protection decision command includes an airbag deployment command, the seat occupant detection sensor is triggered, and the airbag is deployed after confirming the presence of an occupant; if the protection decision command does not include an airbag deployment command, the airbag deployment action is skipped.

[0063] During the execution of active power disconnection of the high-voltage circuit, physical isolation of the charging port, and occupant protection actions, a dual-redundant power supply architecture is adopted to ensure the power supply of the control unit. One path is the on-board low-voltage battery power supply circuit, and the other is the independent low-voltage circuit for the power battery. The independent low-voltage circuit for the power battery is led out from the power battery pack, and outputs 12V voltage after being protected by a fuse, forming parallel redundancy with the on-board low-voltage battery.

[0064] A supercapacitor module is connected in parallel with the vehicle's low-voltage battery to provide buffer power during power switching. The supercapacitor module has a capacity of 100F, a withstand voltage of 16V, and an equivalent series resistance of less than 10mΩ. It is connected in parallel with the 12V vehicle low-voltage battery through a bidirectional DC-DC converter.

[0065] When the on-board low-voltage battery voltage is detected to be below 10.5V or a collision risk signal is received, the system automatically switches to the independent low-voltage circuit for the power battery to ensure continuous power supply to the vehicle domain controller and battery management system. Switching verification results show that, simulating a voltage drop scenario after a collision with the on-board low-voltage battery, with an initial voltage of 12.6V and a 10A load discharge, the switching is triggered when the voltage drops to 10.5V. The measured switching time is 45ms. The supercapacitor maintains the vehicle domain controller's power supply voltage above 11.5V, and the control unit does not experience a reset or communication interruption.

[0066] The formula for calculating the buffer power supply duration of the supercapacitor module is as follows: ; in, To buffer the power supply duration, This refers to the capacity of the supercapacitor module. This refers to the allowable voltage drop range for the supercapacitor module. This is the load current of the control unit.

[0067] Step 7: Security Status Indicator After obtaining the charging collision safety protection results, a corresponding safety status indication signal is generated based on the risk level.

[0068] Establish a mapping relationship between risk levels and safety status indication signals. When the risk level is 0, a green indication signal is generated; when the risk level is 1, a yellow indication signal is generated; and when the risk level is 2 or 3, a red indication signal is generated.

[0069] The safety status indication signals are output to the in-vehicle human-machine interface and the luminous indicator around the charging port. The in-vehicle human-machine interface displays the corresponding color safety status icon on the head-up display, instrument panel, or central control screen. The luminous indicator around the charging port achieves color change through pulse width modulation (PWM) dimming, where level 0 green corresponds to a 100% duty cycle, level 1 yellow corresponds to a 50% duty cycle (a mixture of green and red), and levels 2 and 3 red correspond to a 100% duty cycle. The luminous indicator uses high-brightness LEDs with a luminous flux exceeding 500 lm and a nighttime visibility distance exceeding 50 m, conforming to the International Organization for Standardization (ISO) 15008 standard.

[0070] See Figure 3 Based on the same inventive concept, embodiments of this application provide a safety protection system for vehicles after a collision during charging, including: The multi-source acquisition module is used to acquire point cloud data of the vehicle's perimeter radar, vibration data of the charging port housing, vehicle-mounted visual image data, and vehicle motion state data to obtain a multi-source perception dataset. The data fusion module is used to perform Kalman filter fusion processing on the vehicle perimeter radar point cloud data, charging port housing vibration data and vehicle visual image data. The fusion weight is dynamically adjusted according to the real-time signal-to-noise ratio of each data to obtain the object motion state estimation result. The trajectory prediction module is used to input vehicle motion state data into a preset slip prediction model to obtain the predicted trajectory of the vehicle in the future. The risk identification module is used to input the vehicle's surrounding radar point cloud data, charging port housing vibration data and vehicle visual image data into a preset false trigger classification model to obtain the risk level and false trigger probability value. The joint assessment module is used to compare the object motion state estimation results, vehicle future trajectory prediction results, risk level and false trigger probability value with preset joint assessment conditions to obtain protection decision instructions. The protection execution module is used to execute active power disconnection of the high-voltage circuit, physical isolation of the charging port and occupant protection actions according to the protection decision instructions, so as to obtain the charging collision safety protection results.

[0071] Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A safety protection method after a collision during vehicle charging, characterized in that, Includes the following steps: Acquire vehicle perimeter radar point cloud data, charging port housing vibration data, vehicle visual image data, and vehicle motion state data to obtain a multi-source perception dataset. The vehicle perimeter radar point cloud data, charging port housing vibration data and vehicle-mounted visual image data are fused to obtain the object motion state estimation results. Input the vehicle motion state data into the preset slip prediction model to obtain the prediction result of the vehicle's future trajectory; The vehicle perimeter radar point cloud data, charging port housing vibration data and vehicle visual image data are input into a preset false trigger classification model to obtain the risk level and false trigger probability value. Based on the object motion state estimation results, vehicle future trajectory prediction results, risk level and false trigger probability value, and the preset joint judgment conditions, a protection decision instruction is obtained. Based on the protection decision instructions, the system performs active power disconnection of the high-voltage circuit, physical isolation of the charging port, and occupant protection actions, resulting in charging collision safety protection.

2. The safety protection method after a collision during vehicle charging according to claim 1, characterized in that, The multi-source sensing dataset was obtained, specifically including: Imaging millimeter-wave radar deployed around the vehicle collects radar point cloud data around the vehicle. Vibration data of the charging port housing is collected by a vibration sensor deployed inside the charging port housing. Acquire vehicle-mounted visual image data using vehicle-mounted visual acquisition equipment; Vehicle motion status data is acquired through wheel speed sensors and inertial measurement units; By aligning the vehicle's perimeter radar point cloud data, charging port housing vibration data, vehicle-mounted visual image data, and vehicle motion state data in a time sequence, a multi-source perception dataset is obtained.

3. The safety protection method after a collision during vehicle charging according to claim 1, characterized in that, The results of the object's motion state estimation are obtained, specifically including: Based on the vehicle perimeter radar point cloud data, charging port housing vibration data and vehicle visual image data, state vectors and observation vectors are constructed, and the observation noise covariance matrix is ​​set. The weights of the observation noise covariance matrix are adjusted based on the real-time signal-to-noise ratio of each data point. Under abnormal lighting conditions, the weights of the onboard visual image data are reduced and the weights of the vehicle perimeter radar point cloud data are increased. When continuous micro-vibration is detected, the weights of the charging port housing vibration data are increased. When the deviation between a single data output and the fusion estimation result exceeds a preset threshold and continues for a preset period, the weight of that data is reduced and a degradation mode is triggered. When multiple data fail simultaneously, a safety mode is triggered. The state vector is updated based on the adjusted weights to obtain the object's motion state estimation result.

4. The safety protection method after a collision during vehicle charging according to claim 1, characterized in that, The predicted trajectory of the vehicle is obtained, specifically including: Vehicle wheel speed data, vehicle acceleration data, and road friction coefficient are extracted from vehicle motion state data to construct a time-series input sequence; Input the time-series input sequence into the preset slip prediction model, and output the future trajectory point coordinate sequence of the vehicle; The coordinate sequence of the vehicle's future trajectory points is geometrically overlapped with the coordinates of the charging pile and the coordinates of adjacent vehicles. If trajectory overlap is found, a trajectory overlap warning signal is generated. The trajectory overlap warning signal is combined with the coordinate sequence of the vehicle's future trajectory points to form the prediction result of the vehicle's future trajectory.

5. The safety protection method after a collision during vehicle charging according to claim 4, characterized in that, The risk level and false trigger probability value are obtained, including: Input the vehicle's perimeter radar point cloud data, charging port housing vibration data, and vehicle-mounted visual image data into a preset false trigger classification model, and output the initial risk level and false trigger probability value. When the probability of false triggering exceeds the preset filtering threshold, the initial risk level output is suppressed and the current state is maintained; When the false trigger probability value does not exceed the preset filtering threshold, the initial risk level is selectively increased based on whether the vehicle's future trajectory prediction results contain trajectory overlap warning signals, thus obtaining the risk level.

6. The safety protection method after a collision during vehicle charging according to claim 1, characterized in that, Receive protection decision instructions, specifically including: When the risk level reaches the preset high-voltage trigger level and the object's speed exceeds the preset speed threshold, a high-voltage cutoff command is generated. When the risk level reaches the preset airbag trigger level and the presence of an occupant is detected, an airbag deployment command is generated. When the risk level reaches the preset airbag trigger level and no occupant is detected, a high-pressure cutoff command is generated. The command to disconnect high voltage, the command to deploy airbags, or the command to disconnect high voltage only shall be determined as the protection decision command.

7. The safety protection method after a collision during vehicle charging according to claim 1, characterized in that, Based on the protection decision instructions, the system will execute active power disconnection of the high-voltage circuit, physical isolation of the charging port, and occupant protection actions, specifically including: The high-voltage busbar short circuit is extinguished by using a pyrotechnic short-circuit switch; After the high-voltage bus short circuit is extinguished, a release command is sent to the charging port mechanical pin mechanism after a preset safety delay time. The mechanical pin mechanism of the charging port performs a pop-out action according to the release command, completing the physical isolation of the charging port; Depending on whether the protection decision command includes an airbag deployment command, the airbag deployment action is selectively executed.

8. The safety protection method after a collision during vehicle charging according to claim 7, characterized in that, The process of performing active power disconnection of the high-voltage circuit, physical isolation of the charging port, and occupant protection actions also includes: It adopts a dual-redundant power supply architecture, one of which is the power supply circuit for the vehicle's low-voltage battery, and the other is an independent low-voltage circuit for the power battery; By connecting the supercapacitor module in parallel with the vehicle's low-voltage battery, a buffer power supply is provided during power switching. When the on-board low-voltage battery voltage is detected to be lower than the preset switching threshold or a collision risk signal is received, the system automatically switches to the independent low-voltage circuit of the power battery for power supply.

9. The safety protection method after a collision during vehicle charging according to claim 8, characterized in that, After obtaining the charging collision safety protection results, it also includes: Generate corresponding safety status indication signals based on risk levels; The safety status indication signal is output to the vehicle human-machine interface and the light-emitting indicator around the charging port to provide multimodal safety status prompts.

10. A safety protection system for vehicles after a collision during charging, characterized in that, include: The multi-source acquisition module is used to acquire point cloud data of the vehicle's perimeter radar, vibration data of the charging port housing, vehicle-mounted visual image data, and vehicle motion state data to obtain a multi-source perception dataset. The data fusion module is used to fuse the point cloud data of the vehicle's perimeter radar, the vibration data of the charging port housing, and the vehicle-mounted visual image data to obtain the estimation result of the object's motion state. The trajectory prediction module is used to input vehicle motion state data into a preset slip prediction model to obtain the predicted trajectory of the vehicle in the future. The risk identification module is used to input the vehicle's surrounding radar point cloud data, charging port housing vibration data and vehicle visual image data into a preset false trigger classification model to obtain the risk level and false trigger probability value. The joint assessment module is used to compare the object motion state estimation results, vehicle future trajectory prediction results, risk level and false trigger probability value with preset joint assessment conditions to obtain protection decision instructions. The protection execution module is used to execute active power disconnection of the high-voltage circuit, physical isolation of the charging port and occupant protection actions according to the protection decision instructions, so as to obtain the charging collision safety protection results.

Images