A data-driven vehicle accident mitigation method, system, device and medium

Abstract

The present invention discloses a data-driven vehicle accident mitigation method, system, device and medium. The method includes: obtaining a multi-parameter combination, including static parameters of the vehicle itself, real-time state parameters, occupant information, restraint system parameters of the vehicle for the occupants, as well as state parameters and relative positions of surrounding vehicles; calculating in real time the risk of a vehicle collision accident based on the obtained parameters and judging the possibility of avoiding the collision accident; predicting the degree of accident damage based on the obtained parameters; aiming at minimizing the degree of accident damage and considering the feasibility of parameter transformation, searching for the optimal matching parameters in a multi-dimensional parameter space; wherein the multi-dimensional parameter space includes static parameters of the vehicle itself, real-time state parameters, occupant information, and restraint system parameters of the vehicle for the occupants; and performing adjustment operations on the state parameters and restraint system parameters of the vehicle according to the optimal matching parameters. The present invention can improve the safety performance of the vehicle in extreme cases where an accident cannot be avoided.

Description

Technical Field

[0001] The present invention belongs to the technical field of vehicle autonomous driving, and particularly relates to a data-driven vehicle accident mitigation method, system, device and medium. Background Technique

[0002] Autonomous driving technology was first proposed for the purpose of improving road traffic safety. The perception system is used to obtain information about other participants outside the vehicle and detect the risk of accident occurrence in real time. In the case where accidents cannot be avoided, current main research mostly focuses on the category of occupant injury prediction in accidents, exploring the prospects of technologies such as regression analysis and neural networks applied to this field, and focusing on the accuracy and real-time performance of injury prediction. Since occupant injury is related to many factors such as vehicle speed, collision angle and occupant posture, this results in a certain high-dimensional input matrix for the injury prediction model. Neural networks of various structures may be the optimal method for processing high-dimensional information to solve injury prediction. However, existing research does not pay much attention to the growth of the prediction model accuracy, the data utilization rate is not high, and there is less research on exploring accident injury mitigation methods based on injury prediction results. In addition, under the premise of minimizing occupant injury, the feasibility analysis of mitigation operations has not been carried out, increasing the probability of additional risks. Summary of the Invention

[0003] The present invention provides a data-driven vehicle accident mitigation method, system, device and medium to improve the safety performance of vehicles in extreme cases where accidents cannot be avoided.

[0004] To achieve the above technical objectives, the present invention adopts the following technical solutions:

[0005] A data-driven vehicle accident mitigation method includes:

[0006] Step S10, obtaining multi-parameter combinations, including static parameters of the vehicle itself, real-time state parameters, occupant information, restraint system parameters of the vehicle for the occupants, and state parameters and relative positions of surrounding vehicles;

[0007] Step S20, calculating the risk of the vehicle having a collision accident in real time according to the state parameters and relative positions of the vehicle itself and surrounding vehicles, and judging the possibility of avoiding the collision accident;

[0008] Step S30, in the case where the accident cannot be avoided, calculating pre-collision parameters between the vehicle and surrounding vehicles: collision point, collision speed, collision angle, collision coincidence degree, and predicting the accident injury degree according to the pre-collision parameters and the size parameters of the mutually colliding vehicles;

[0009] Step S40: Search for the optimal matching parameters in the multi-dimensional parameter space with the goal of minimizing the accident damage degree and considering the feasibility of parameter transformation. The multi-dimensional parameter space for the search includes the static parameters of the vehicle itself, real-time state parameters, occupant information, and the restraint system parameters of the vehicle for the occupants.

[0010] Step S50: Adjust the state parameters of the vehicle and the restraint system parameters of the vehicle for the occupants according to the optimal matching parameters.

[0011] Further, the state parameters include speed, acceleration, and direction, and the restraint system parameters of the vehicle for the occupants include seat position, seat inclination, seat belt usage, and airbag equipment.

[0012] Further, in step S30, a pre-trained accident damage degree prediction model is used to predict accident damage.

[0013] Further, after the adjustment operation in step S50, record the damage degree of the actual collision accident that occurred, and construct a new training sample with its optimal matching parameters to dynamically update and maintain the accident damage degree prediction model.

[0014] Further, the method for constructing the training sample of the accident damage degree prediction model is as follows:

[0015] Adopt the random sampling or distribution curve-based sampling method to search in the distribution space of each pre-collision parameter, and sample to obtain a multi-dimensional parameter matrix. Normalize each parameter in the multi-dimensional parameter matrix, and use the normalized multi-dimensional parameter matrix as the input set of the training sample of the accident damage degree prediction model.

[0016] Use the multi-dimensional parameter matrix as the input of the finite element simulation software, deploy the corresponding pre-collision scenario and the human body model simulating the occupant in the simulation software, and obtain the collision response information of the human body model by running the simulation software. Then, according to the knowledge of injury biomechanics, solve the injury indexes of each part of the occupant after the collision accident occurs, and use the comprehensive value of the multi-part injury indexes as the output label of the training sample of the accident damage degree prediction model.

[0017] Further, regarding the feasibility of considering parameter transformation in step S40, for the state parameters of the vehicle itself, specifically, the yaw rate lateral velocity v y acceleration a, and rotation angle θ are constrained as follows:

[0018]

[0019]

[0020] |θ| ≤ θ lim

[0021] |a| ≤ a lim

[0022] where α r represents the rear wheel slip angle, v x and v y respectively represent the longitudinal speed and lateral speed of the vehicle's center of mass, is the yaw rate, l r and l f respectively represent the distances from the vehicle's center of mass to the rear and front axles; α r,lim is the threshold value of the rear wheel slip angle, is the linear cornering stiffness of the rear wheel; θ lim represents the steering angle limit of the vehicle, a lim represents the vehicle acceleration limit.

[0023] Furthermore, the iterative termination conditions for searching for the optimal matching parameters in the multi-dimensional parameter space in step S40 include: the accident damage degree corresponding to the currently optimal matching parameter reaches an acceptable range, or the iterative time of the search reaches the preset maximum iterative time.

[0024] A data-driven vehicle accident mitigation system, comprising: a sensing module, a collision prediction module, a damage prediction module, a high-speed search module, and a control constraint module;

[0025] The sensing module is used to: obtain multi-parameter combinations, including the static parameters of the vehicle itself, real-time state parameters, occupant information, parameters of the vehicle's restraint system for the occupants, and the state parameters and relative positions of surrounding vehicles;

[0026] The collision prediction module is used to: calculate the risk of the vehicle having a collision accident in real time based on the state parameters and relative positions of the vehicle itself and surrounding vehicles, and judge the possibility of avoiding the collision accident;

[0027] The damage prediction module is used to: calculate the pre-collision parameters between the vehicle and surrounding vehicles in the case where the accident cannot be avoided: the collision point, collision speed, collision angle, and collision overlap degree, and predict the accident damage degree based on the pre-collision parameters and the size parameters of the mutually colliding vehicles;

[0028] The high-speed search module is used to: search for the optimal matching parameters in the multi-dimensional parameter space with the goal of minimizing the accident damage degree and considering the feasibility of parameter transformation; wherein, the multi-dimensional parameter space for the search includes the vehicle's own state parameters and the parameters of the vehicle's restraint system for the occupants;

[0029] The control constraint module is used to: adjust the state parameters of the vehicle and the parameters of the vehicle's restraint system for the occupants according to the optimal matching parameters.

[0030] An electronic device includes a memory and a processor. A computer program is stored in the memory. When the computer program is executed by the processor, the processor implements the data-driven vehicle accident mitigation method described in any one of the above.

[0031] A computer-readable storage medium stores a computer program. When the computer program is executed by a processor, the data-driven vehicle accident mitigation method described in any one of the above is implemented.

[0032] Beneficial effects

[0033] On the one hand, the present invention promotes the application extension of the existing damage prediction model to the damage mitigation strategy, improving the safety performance of autonomous vehicles; on the second hand, aiming at the fact that the existing damage prediction methods ignore the scalability of the model, a model training method with a data closed-loop is proposed to realize the dynamic update and maintenance of the model; on the third hand, in the search link of the optimal parameter matrix of the present application, a feasibility analysis model of parameter transformation operation is additionally introduced for multi-objective optimization search, which can avoid the occurrence of additional accidents caused by excessive pursuit of damage mitigation. Therefore, when the present invention is applied to intelligent vehicles, it can significantly improve the safety performance of vehicles in extreme situations where accidents are unavoidable, make up for the deficiencies of existing collision avoidance systems in this situation, and also provide new ideas for the development of future intelligent vehicle safety functions. Description of the drawings

[0034] Figure 1 It is a flowchart of a data-driven accident damage mitigation method.

[0035] Figure 2 It is a schematic diagram of accident collision avoidance.

[0036] Figure 3 It is a training flowchart of a neural network damage prediction model.

[0037] Figure 4 It is a flowchart of a multi-objective optimal parameter search method.

[0038] Figure 5 It is a two-degree-of-freedom vehicle dynamics model.

[0039] Figure 6 It is an architecture diagram of a data-driven accident damage mitigation system. Detailed implementation manners

[0040] In order to make the above-mentioned objects, features, and advantages of the application more obvious and understandable, the following will describe the specific implementation manners of the present application in detail with reference to the accompanying drawings. Many specific details are set forth in the following description to facilitate a full understanding of the present application. However, the present application can be implemented in many other ways different from this description, and those skilled in the art can make similar improvements without departing from the connotation of the present application. Therefore, the present application is not limited by the specific implementations disclosed below.

[0041] Embodiment 1

[0042] This embodiment provides a data-driven vehicle accident mitigation method. The following takes the example of this method being used for a single vehicle terminal. At the same time, this method is also applicable to the server side, which has higher storage capacity and faster computing speed.

[0043] The method of this embodiment refers to Figure 1 As shown, it includes the following steps:

[0044] Step S10: Obtain a multi-parameter combination, including the static parameters of the vehicle itself, real-time state parameters, occupant information, restraint system parameters of the vehicle for the occupants, and the state parameters and relative positions of surrounding vehicles.

[0045] The static parameters of the vehicle include the type, size information, chassis height, etc. of the vehicle. The state parameters include speed, acceleration, direction, etc. The occupant information includes the position, gender, height, age, and sitting posture of the occupants. The restraint system parameters of the vehicle for the occupants include the seat position, seat inclination, seat belt usage, airbag equipment, etc.

[0046] In this embodiment, a perception system (such as millimeter-wave radar and other sensors) is used to obtain information such as the speed, acceleration, distance, and relative position of surrounding vehicles in real time. At the same time, through the in-vehicle communication bus, vehicle information such as the vehicle speed and acceleration of the host vehicle, as well as information such as the sitting posture of the occupants detected by in-vehicle sensors and the usage status of the restraint system, are obtained in real time.

[0047] Step S20: According to the state parameters and relative positions of the vehicle itself and surrounding vehicles, calculate the risk of a vehicle collision accident in real time, and judge the possibility of avoiding a collision accident.

[0048] The on-vehicle computing module can calculate whether there is a collision risk for the current vehicle based on the multi-parameter combination obtained in step S10 (usually judged by the time to collision (TTC)). Generally, a threshold TTC is set. When the time t for the vehicle to reach the collision point from the current position TTCWhen TTC, it is regarded as having a collision risk). If so, the corresponding key parameters such as the collision speed, collision point, collision angle, and collision overlap at the pre-collision moment can be solved according to the existing kinematic knowledge for use in the subsequent damage mitigation strategy calculation system.

[0049] Judging whether an accident can be avoided under a collision risk can be achieved by using existing technologies. For example, refer to Figure 2 As shown: The initial position of the host vehicle (x 0 , y 0 ), the extreme collision avoidance position (y e , y e ), the position of the target vehicle (x m , y m ), the width B1 of the host vehicle, and the width B2 of the target vehicle.

[0050] (1) Longitudinal operation cannot avoid: That is, when the vehicle decelerates to 0 at the maximum deceleration, the two vehicles continue to collide

[0051] V 1y = a 1max t

[0052] D y ≤(V 1y t - 0.5a 1max t 2 )+(V 2y t - 0.5a 2y t 2 ) (1)

[0053] (2) Lateral operation cannot avoid: The position of the target vehicle (x m , y m )

[0054] Taking the local collision avoidance path of a univariate fifth-degree polynomial as an example:

[0055] y = c 0 + c 1 x + c 2 x 2 + c 3 x 3 + c 4 x 4 + c 5 x 5

[0056] y′ = c 1 + 2c 2 x 1 + 3c 3 x 2 + 4c 4 x 3 + 5c 5 x 4

[0057] The initial position of the host vehicle (x 0 , y 0 ), and the end position (x e , y e ) satisfy:

[0058] y 0 = c 0 = 0

[0059] y 0 ' = c 0 + c 1 = 0

[0060] y e = c 0 + c 1 x e + c 2 x e 2 + c 3 x e 3 + c 4 x e 4 + c 5 x e 5

[0061] y e ' = c 1 + 2c 2 x e + 3c 3 x e 2 + 4c 4 x e 3 + 5c 5 x e 4 = 0

[0062] It can be obtained that: c 0 = c 1 = c 2 = 0

[0063]

[0064]

[0065]

[0066] The fitting collision avoidance path formula of the unary fifth - degree polynomial is obtained as:

[0067]

[0068] The extreme case of collision avoidance failure is y e -y m is greater than or equal to half of the sum of the widths:

[0069]

[0070] In summary, when both formula (1) and (2) are satisfied, it is determined that the collision cannot be avoided. In a more optimal embodiment, an alarm warning can further be carried out after a collision risk is detected.

[0071] Step S30, in the case where an accident cannot be avoided, calculate the pre-collision parameters between the vehicle and surrounding vehicles: the collision point, the collision speed, the collision angle, and the collision coincidence degree, and predict the accident damage degree according to the pre-collision parameters and the size parameters of the mutually colliding vehicles.

[0072] In this embodiment, a pre-trained accident damage degree prediction model is specifically used for accident damage prediction.

[0073] In implementation, the parameters input into the accident damage degree prediction model are data after some necessary preprocessing. The preprocessing includes: performing mean normalization processing on continuous variables such as speed and normalizing them to a dimensionless state with values ranging from 0 to 1; for discrete variables such as the sitting postures of occupants, generally using the one-hot encoding method to perform non-discriminatory value conversion on them. After the data processing is completed, it can be input into the damage prediction model, and finally, the information on the damage degree of the occupants that may be brought about without taking any measures in the current state can be obtained.

[0074] Among them, referring to Figure 3 as shown, the method for constructing the training samples of the accident damage degree prediction model is:

[0075] (1) Adopt the random sampling or the sampling method based on the distribution curve to search in the parameter distribution space of the multi-parameter combination described in step S10, and sample to obtain a multi-dimensional parameter matrix; perform normalization processing on each parameter in the multi-dimensional parameter matrix, and use the normalized multi-dimensional parameter matrix as the input set of the training samples of the accident damage degree prediction model.

[0076] In implementation, based on the existing research foundation, the distribution curves of variables such as the speed, collision point, and collision coincidence degree of the vehicle when an accident occurs are statistically obtained. For these information variables with statistics, sampling can be performed based on their distribution laws in subsequent searches. For some key information such as the occupant posture, it may be temporarily impossible to obtain their distribution laws in accidents, and then a simple random sampling method can be used to extract them according to their different values. In each loop sampling process, after all variables are sampled, they can be spliced into an information matrix as the input data for subsequent simulation tests.

[0077] (2) Use the multi-dimensional parameter matrix as the input of the finite element simulation software, deploy the corresponding pre-crash scenario and the human body model simulating the occupant in the simulation software, and obtain the collision response information of the human body model by running the simulation software; then, according to the knowledge of injury biomechanics, solve the injury indexes of each part of the occupant after the collision accident, such as the head injury evaluation index HIC, the neck injury evaluation index N ij etc., and use the comprehensive value of the multi-part injury indexes as the output label of the training sample of the accident injury degree prediction model.

[0078] The finite element simulation software that may be involved includes but is not limited to software such as HyperMesh, Ansys, Abaqus, etc. At the same time, the establishment of some vehicle and human finite element models may be involved.

[0079] In this embodiment, the neural network adopted by the prediction model can be a mainstream deep neural network, a convolutional neural network, and various other evolved networks, which are used to explore a certain mapping relationship between the input information and the occupant injury. In this embodiment, the activation function of the input layer adopts the Relu function, which can achieve a faster convergence speed; the activation function of the output layer selects the sigmoid function to map the prediction result to the interval [0,1]; the loss function adopts the mean square error MSE between the predicted value and the actual injury result.

[0080] After obtaining the training sample set according to the above method, reasonably divide the existing data set into a training set, a validation set, and a test set according to a certain proportion, and then repeatedly train the neural network model for many times. At the same time, continuously adjust the hyperparameters such as the number of network layers, the number of neurons, and the learning rate, and select the model with the highest prediction accuracy as the finally trained and optimal accident injury degree prediction model.

[0081] When constructing the training sample by fusing the advanced sampling technology as described above, the distribution laws of each variable in the actual accident statistics are considered, which can ensure that the training data set conforms to the real characteristics, can maximize the parameter coverage rate, so as to achieve a better training effect and improve the prediction accuracy of the accident injury prediction model.

[0082] Step S40, with the goal of minimizing the accident injury degree and considering the feasibility of parameter transformation, search for the optimal matching parameters in the multi-dimensional parameter space.

[0083] Among them, the multi-dimensional parameter space to be searched includes: the static parameters of the vehicle itself and the real-time state parameters, the occupant information, and the restraint system parameters of the vehicle for the occupant. The specifically optimizable parameters refer to the parameters that can be actively adjusted among them, including the vehicle's own state parameters and the restraint system parameters of the vehicle for the occupant, excluding the parameters that cannot be actively adjusted, such as the static parameters of the vehicle itself and the occupant information.

[0084] In this embodiment, the specific steps for searching for the optimal matching parameters in the multi-dimensional parameter space are as follows with reference to Figure 4 shown, including:

[0085] (1) Using sampling techniques such as simple random sampling or sampling based on distribution curves, search in the multi-dimensional parameter space of each pre-collision parameter to obtain a target parameter matrix, which is used as the input of the result evaluation model;

[0086] In this embodiment, based on the neighborhood search process, a new parameter matrix can be obtained by continuously changing one or several parameters. It should be noted that the initialization of the parameter matrix is a random search based on the neighborhood, while the update of the parameter matrix during the optimization process is achieved through certain heuristic rules to accelerate the search effect.

[0087] (2) The result evaluation model evaluates the input target parameter matrix from two dimensions: the accident damage degree and the feasibility of parameter transformation. Among them, calculate the pre-collision parameters according to the target parameter matrix in step S30, and then use the accident damage degree prediction model to output the accident damage degree. The evaluation of the feasibility of parameter transformation mainly considers the stability performance of the vehicle to prevent additional accidents. Its input variables mainly include vehicle motion parameters (speed, direction, etc.), and the in-vehicle parameters (seat angle, occupant sitting posture, etc.) are not considered. Since the hazard of rear axle side slip is greater when the vehicle experiences yaw instability, the rear wheel side slip angle is restricted. Combining with the two-degree-of-freedom vehicle dynamics model (such as Figure 5 shown), the specific calculation formula is as follows:

[0088]

[0089] Convert the constraint into a constraint on the vehicle's lateral velocity v y , and we can get:

[0090]

[0091] In the formula, α r represents the rear wheel side slip angle, α r,lim is the rear wheel side slip angle threshold, v x and v y respectively represent the longitudinal velocity and lateral velocity of the vehicle's center of mass, is the yaw angular velocity, l r and l f respectively represent the distances from the vehicle's center of mass to the rear and front axles, is the linear side slip stiffness of the rear wheel. When the vehicle state satisfies equations (3) and (4), it proves that the currently searched parameters can ensure the yaw stability of the vehicle.

[0092] In addition, it is also necessary to ensure that the vehicle acceleration a and the steering angle θ are within the limit range, as shown in the following formula:

[0093]

[0094] (3) Cache the optimal solution in the target parameter matrix according to the accident damage degree output by the prediction model. On the one hand, it is used for the subsequent selection of the optimal matching parameters, and on the other hand, it is used for the repeated check during the search of the target parameter matrix.

[0095] (4) On the premise of considering the feasibility of parameter transformation, select the multi-parameter combination with the lowest accident damage degree in the optimal solution as the optimal matching parameter.

[0096] In implementation, on the basis that the accident damage degree corresponding to the currently input target parameter matrix is alleviated compared with that of the target parameter matrix in the previous round of iteration, then conduct the evaluation of the feasibility of parameter transformation. When both of the two optimization objectives are satisfied (that is, the damage is alleviated and the evaluation of the feasibility of parameter transformation), update the current target parameter matrix, and at the same time continue to search for the parameter combination according to the current search direction (that is, repeat the search heuristic rule in step (1)). If only the condition of damage mitigation is satisfied but the operation is not feasible, then continue to randomly search and iterate within its neighborhood for the input parameters of the feasibility analysis, and other parameters remain unchanged during this process. If neither of the two optimization objectives is satisfied, store the current parameter matrix in the infeasible solution space to avoid repeated search, and then update the existing parameter matrix with reference to the method in step (1) of iteration.

[0097] In implementation, classify and store the search results in a timely manner. According to the evaluation results of the parameter matrix, put them into the infeasible solution buffer and the feasible solution buffer respectively, so as to avoid repeated search and reduce the cumulative calculation cost.

[0098] The parameter update condition in the optimal parameter buffer is to satisfy both the damage evaluation model and the feasibility evaluation model at the same time. The iterative termination conditions for the search include: the accident damage degree corresponding to the currently optimal matching parameter reaches the acceptable range, or the iterative time of the search reaches the preset maximum iterative time. The specific expressions are as follows:

[0099]

[0100] where, Injury index is the comprehensive index of occupant injury, which is closely related to indicators such as the head injury evaluation index HIC and the neck injury evaluation index N ij and other indicators; ε is the maximum acceptable damage level; t and t max are the iterative time and the maximum iterative time respectively.

[0101] Step S50, adjust the state parameters of the vehicle and the restraint system parameters of the vehicle for the occupant according to the optimal matching parameters.

[0102] In implementation, according to the existing known conversion relationships (such as the correspondence between acceleration and throttle), the optimal matching parameters obtained in step S40 are successively converted into information that can be directly read by the vehicle operating system, including: information such as throttle opening, steering wheel angle, and seat back motor angle. Through a series of operations, the vehicle is set to the target ideal state, thereby achieving the purpose of reducing injuries. Among them, the above-mentioned throttle is not limited to fuel vehicles only, but can also correspond to electric vehicles, and generally refers to the vehicle speed control component.

[0103] In a more optimal embodiment, after the adjustment operation in step S50, there is also a step S60: record the degree of injury in an actual collision accident and construct a new training sample with the optimal matching parameters for dynamic update and maintenance of the prediction model for accident injury degree.

[0104] In implementation, after the vehicle state adjustment operation according to step S50, the optimal matching parameters obtained in step S40 are spliced with the actual injury information to form a new training sample data and supplement it to the existing training sample set, and then the existing injury prediction model is trained again to improve the prediction accuracy and stability of the model and achieve the closed-loop utilization of data. The above-mentioned injury information is mainly obtained through medical examinations after an accident, which ensures the authenticity value of this data. The retraining of the above-mentioned prediction model mainly refers to continuing to train on the basis of the existing prediction model with the accumulated new data and fine-tuning the model, rather than starting a new model training with all the data in the database. In addition, the dynamic update period of the model can be based on a time interval or on a certain amount of new data reserve.

[0105] In the accident injury mitigation method based on data-driven provided in this embodiment, considering the premise that a collision accident is inevitable, a method for mitigating occupant injuries is given. By using the perception system and in-vehicle system to obtain vehicle information and in-vehicle occupant and restraint system parameters in real time, and adopting kinematic methods, the vehicle information can be further converted into collision moment information, and finally a pre-collision parameter matrix is obtained and input into a pre-trained injury prediction model to obtain the possible occupant injury situation without taking any measures in the current state. Through a designed optimal parameter search method, this method aims to minimize occupant injury and the feasibility of transformation operations, and searches for optimal parameters in the parameter space. Then, according to the finally obtained optimal matching parameters, the vehicle state (such as throttle opening, steering wheel angle, seat inclination, etc.) is correspondingly switched to achieve the purpose of mitigating injuries. Finally, the obtained optimal matching parameters and the real injury information of the occupant after the accident are supplemented to the database for dynamic update and maintenance of the prediction model.

[0106] In summary, the method of this embodiment can achieve the effect of reducing the injuries of accident occupants, making up for the deficiencies of existing research. In addition, by considering the feasibility of mitigation measures, it avoids additional accidents (such as vehicle rollover) caused by parameter mismatches. At the same time, by returning the final effect obtained from the actual mitigation operation to the sample database, it realizes the closed-loop utilization of data and can gradually improve the accuracy and stability of the prediction model.

[0107] Embodiment 2

[0108] This embodiment provides a data-driven vehicle accident mitigation system, referring to Figure 6 As shown, it includes: a perception module, a collision prediction module, an injury prediction module, a high-speed search module, and a control constraint module;

[0109] The perception module is used to: obtain a multi-parameter combination, including the static parameters of the vehicle itself, real-time state parameters, occupant information, the restraint system parameters of the vehicle on the occupants, as well as the state parameters and relative positions of surrounding vehicles;

[0110] The collision prediction module is used to: calculate the risk of the vehicle having a collision accident in real time according to the state parameters and relative positions of the vehicle itself and surrounding vehicles, and judge the possibility of avoiding the collision accident;

[0111] The injury prediction module is used to: in the case where the accident cannot be avoided, calculate the pre-collision parameters between the vehicle and surrounding vehicles: collision point, collision speed, collision angle, collision overlap, and predict the accident injury degree according to the pre-collision parameters and the size parameters of the vehicles colliding with each other;

[0112] The high-speed search module is used to: aim at minimizing the accident injury degree and consider the feasibility of parameter transformation to search for the optimal matching parameters in the multi-dimensional parameter space; among them, the multi-dimensional parameter space searched includes the state parameters of the vehicle itself and the restraint system parameters of the vehicle on the occupants;

[0113] The control constraint module is used to: adjust the state parameters of the vehicle and the restraint system parameters of the vehicle on the occupants according to the optimal matching parameters.

[0114] The specific implementation manners of the above-mentioned modules included in the vehicle accident mitigation system of this embodiment are the same as those described in the method of Embodiment 1, and will not be repeated here.

[0115] In a further embodiment, the memory included in the vehicle accident mitigation system is mainly used to record the optimal search results in each accident. After the actual occupant injury detection results are announced, they will be stored together in the data set memory for the dynamic update and maintenance of the subsequent injury prediction model. The buffer is mainly used to temporarily store the data that needs to be temporarily stored, such as the feasible solution set and the infeasible solution set during the search iteration process, and will be automatically cleared after the accident occurs. The above-mentioned modules communicate through the vehicle-mounted system to exchange the required information data. It should be noted that some modules, such as the high-speed search module, are not limited to vehicle-mounted modules, but also applicable to cloud systems with faster computing and stronger storage capabilities.

[0116] Embodiment 3

[0117] This embodiment provides an electronic device, including a memory and a processor. A computer program is stored in the memory. When the computer program is executed by the processor, the processor implements the data-driven vehicle accident mitigation method described in Embodiment 1.

[0118] Embodiment 4

[0119] A computer-readable storage medium stores a computer program. When the computer program is executed by a processor, the data-driven vehicle accident mitigation method described in Embodiment 1 is implemented.

[0120] The above embodiments are the preferred embodiments of the present application. Those of ordinary skill in the art can also make various transformations or improvements based on this. Without departing from the general concept of the present application, these transformations or improvements should all fall within the scope required to be protected by the present application.

Claims

1. A data-driven vehicle accident mitigation method, characterized in that, it includes: Step S10, obtaining a multi-parameter combination, including static parameters of the vehicle itself, real-time state parameters, occupant information, restraint system parameters of the vehicle for the occupants, and state parameters and relative positions of surrounding vehicles; Step S20, calculating the risk of a collision accident occurring to the vehicle in real time according to the state parameters and relative positions of the vehicle itself and surrounding vehicles, and judging the possibility of avoiding the collision accident; Step S30, in the case where the accident cannot be avoided, calculating pre-collision parameters between the vehicle and surrounding vehicles: collision point, collision speed, collision angle, collision coincidence degree, and predicting the accident damage degree according to the pre-collision parameters and the size parameters of the mutually colliding vehicles; Step S30 uses a pre-trained accident damage degree prediction model to predict accident damage; the training sample construction method of the accident damage degree prediction model is: Adopting a random sampling or distribution curve-based sampling method, searching in the distribution space of each pre-collision parameter, and sampling to obtain a multi-dimensional parameter matrix; Normalizing each parameter in the multi-dimensional parameter matrix respectively, and using the normalized multi-dimensional parameter matrix as the training sample input set of the accident damage degree prediction model; Taking the multi-dimensional parameter matrix as the input of the finite element simulation software, deploying the corresponding pre-collision scenario and the human body model simulating the occupant in the simulation software, obtaining the collision response information of the human body model by running the simulation software; and then, according to the knowledge of injury biomechanics, solving the injury indexes of each part of the occupant after the collision accident occurs, and taking the comprehensive value of the multi-part injury indexes as the training sample output label of the accident damage degree prediction model; Step S40, aiming at minimizing the accident damage degree and considering the feasibility of parameter transformation, searching for the optimal matching parameters in the multi-dimensional parameter space; among them, the searched multi-dimensional parameter space includes static parameters of the vehicle itself, real-time state parameters, occupant information, and restraint system parameters of the vehicle for the occupants; Considering the feasibility of parameter transformation in step S40, for the state parameters of the vehicle itself, specifically for the yaw rate lateral velocity v y , acceleration a, and steering angle θ are constrained as follows: |θ| ≤ θ lim |a| ≤ a lim where α r represents the rear wheel slip angle, v x and v y represent the longitudinal speed and lateral speed of the vehicle center of mass respectively, is the yaw rate, l r and l f are the distances from the vehicle center of mass to the rear and front axles respectively; α r,lim is the rear wheel slip angle threshold, is the linear cornering stiffness of the rear wheel; θ lim represents the steering angle limit of the vehicle, a lim represents the vehicle acceleration limit; Step S50, performing adjustment operations on the state parameters of the vehicle and the restraint system parameters of the vehicle for the occupants according to the optimal matching parameters.

2. The vehicle accident mitigation method according to claim 1, characterized in that, The state parameters include speed, acceleration and direction, and the restraint system parameters of the vehicle for the occupants include seat position, seat inclination angle, seat belt usage, and airbag equipment.

3. The vehicle accident mitigation method according to claim 1, characterized in that, After the adjustment operation in step S50, record the damage degree of the actual collision accident that occurs, and construct a new training sample with its optimal matching parameters to dynamically update and maintain the prediction model of the accident damage degree.

4. The vehicle accident mitigation method according to claim 1, characterized in that, The iterative termination conditions for searching for the optimal matching parameters in step S40 in the multi-dimensional parameter space include: the accident damage degree corresponding to the current optimal matching parameter reaches an acceptable range, or the iterative time of the search reaches the preset maximum iterative time.

5. A data-driven vehicle accident mitigation system, characterized in that, it includes: Perception module, collision prediction module, damage prediction module, high-speed search module, control and constraint module; The perception module is used to: obtain multi-parameter combinations, including static parameters of the vehicle itself, real-time state parameters, occupant information, parameters of the vehicle's restraint system for the occupant, and state parameters and relative positions of surrounding vehicles; The collision prediction module is used to: calculate the risk of a vehicle collision accident in real time based on the state parameters and relative positions of the vehicle itself and surrounding vehicles, and judge the possibility of avoiding the collision accident; The damage prediction module is used to: calculate the pre-collision parameters between the vehicle and surrounding vehicles in the case where the accident cannot be avoided: collision point, collision speed, collision angle, collision coincidence degree, and predict the accident damage degree based on the pre-collision parameters and the size parameters of the vehicles involved in the collision; Step S30 uses a pre-trained accident damage degree prediction model to perform accident damage prediction; the training sample construction method of the accident damage degree prediction model is: Adopt random sampling or sampling method based on distribution curve to search in the distribution space of each pre-collision parameter, and sample to obtain a multi-dimensional parameter matrix; Normalize each parameter in the multi-dimensional parameter matrix respectively, and use the normalized multi-dimensional parameter matrix as the training sample input set of the accident damage degree prediction model; Use the multi-dimensional parameter matrix as the input of the finite element simulation software, deploy the corresponding pre-collision scenario and the human model simulating the occupant in the simulation software, and obtain the collision response information of the human model by running the simulation software; furthermore, according to the knowledge of injury biomechanics, solve the injury indicators of each part of the occupant after the collision accident occurs, and use the comprehensive value of the multi-part injury indicators as the training sample output label of the accident damage degree prediction model; The high-speed search module is used to: aim at minimizing the accident damage degree and consider the feasibility of parameter transformation, and search for the optimal matching parameters in the multi-dimensional parameter space; among them, the multi-dimensional parameter space searched includes the vehicle's own state parameters and the vehicle's restraint system parameters for the occupant; Considering the feasibility of parameter transformation described in step S40, for the state parameters of the vehicle itself, specifically for the yaw rate lateral velocity v y , acceleration a, and steering angle θ are constrained as follows: |θ| ≤ θ lim |a| ≤ a lim where α r represents the rear wheel slip angle, v x and v y respectively represent the longitudinal speed and the lateral speed of the vehicle center of mass, is the yaw rate, l r and l f are the distances from the vehicle center of mass to the rear and front axles respectively; α r,lim is the rear wheel slip angle threshold, is the linear slip stiffness of the rear wheel; θ lim represents the steering angle limit of the vehicle, a lim represents the vehicle acceleration limit; The control and constraint module is used to: adjust the state parameters of the vehicle and the parameters of the vehicle's restraint system for the occupant according to the optimal matching parameters.

6. An electronic device, including a memory and a processor, and a computer program is stored in the memory, characterized in that, when the computer program is executed by the processor, the processor realizes the method according to any one of claims 1 to 4.

7. A computer-readable storage medium, on which a computer program is stored, characterized in that, when the computer program is executed by the processor, it realizes the method according to any one of claims 1 to 4.

Images