Optimization method, system, storage medium and electronic device of vehicle restraint system

Abstract

This invention discloses an optimization method, system, storage medium, and electronic device for a vehicle constraint system. The method includes: using a constraint system model to be optimized, predicting personnel injury based on the current design parameters and training scenario parameters of the vehicle constraint system to obtain a first predicted injury value; using an adversarial scenario generation model, generating adversarial scenario parameters based on the current design parameters and training scenario parameters; using the constraint system model to be optimized, predicting personnel injury based on the current design parameters and adversarial scenario parameters to obtain a second predicted injury value; obtaining a comprehensive loss function based on the first and second predicted injury values; updating the current design parameters based on the comprehensive loss function, and returning to the step of using the constraint system model to be optimized to predict personnel injury based on the current design parameters and training scenario parameters of the vehicle constraint system, until a preset iteration condition is met to obtain the target design parameters.

Description

Technical Field

[0001] This invention relates to the field of automotive passive safety and intelligent design, and in particular to an optimization method, system, storage medium, and electronic device for a vehicle constraint system. Background Technology

[0002] With the diversification of intelligent driving technology and intelligent cockpit scenarios, occupant protection also needs to provide multi-scenario, multi-dimensional protection. As intelligent cockpit technology upgrades and L2+ level autonomous driving becomes more widespread, user demands for occupant restraint systems have shifted from "passive collision protection" to "active prediction, personalized adaptation, and full-scenario coverage."

[0003] The optimization goal of vehicle restraint systems is to minimize the risk of injury to occupants in a collision. The current mainstream optimization paradigm is an iterative cycle of "design-simulation verification-parameter adjustment," which is highly dependent on the engineer's experience and the preferred test scenarios, and also suffers from long iteration cycles and high costs. Summary of the Invention

[0004] This invention aims to at least partially solve one of the technical problems in related technologies. To this end, one objective of this invention is to propose an optimization method for vehicle constraint systems that can significantly shorten the development cycle, reduce later modification costs, and also solve the problem of "verification lag".

[0005] The second objective of this invention is to propose an optimization system for a vehicle constraint system.

[0006] A third objective of this invention is to provide a computer-readable storage medium.

[0007] The fourth objective of this invention is to provide an electronic device.

[0008] To achieve the above objectives, a first aspect of the present invention proposes an optimization method for a vehicle constraint system. The method includes: using a constraint system model to be optimized, predicting personal injury based on the current design parameters and training scenario parameters of the vehicle constraint system to obtain a first predicted injury value, wherein the training scenario parameters are scenario parameters for a preset collision involving the vehicle; using an adversarial scenario generation model, generating adversarial scenario parameters based on the current design parameters and the training scenario parameters, wherein the adversarial scenario parameters are used to increase the predicted injury value output by the constraint system model to be optimized; using the constraint system model to be optimized, predicting personal injury based on the current design parameters and the adversarial scenario parameters to obtain a second predicted injury value; obtaining a comprehensive loss function based on the first predicted injury value and the second predicted injury value; updating the current design parameters based on the comprehensive loss function, and returning to the step of predicting personal injury using the constraint system model to be optimized based on the current design parameters and training scenario parameters of the vehicle constraint system, until a preset iteration condition is met to obtain target design parameters.

[0009] The vehicle constraint system optimization method according to embodiments of the present invention adopts a dynamic game theory approach. Through the competition between the constraint system model to be optimized and the adversarial scenario generation model, the performance boundary is continuously expanded. The optimization result is more robust and can cope with a wider range of unknown scenarios. The optimization and adversarial testing are transformed from a serial to a parallel and real-time interactive relationship. The defects of the constraint system model to be optimized are exposed in the early design stage. The constraint system model to be optimized enhances its "immunity" in the iteration process. This greatly compresses the "design-defect exposure-redesign" cycle, significantly shortens the development cycle, reduces the cost of later modifications, and also solves the problem of "verification lag".

[0010] In addition, the optimization method for the vehicle constraint system proposed in the above embodiments of the present invention may also have the following additional technical features: According to an embodiment of the present invention, the adversarial scenario generation model includes a policy subnet, a planning subnet, and a perturbation subnet connected in sequence. The step of generating adversarial scenario parameters using the adversarial scenario generation model based on the current design parameters and the training scenario parameters includes: using the policy subnet to classify damage based on damage features to obtain multiple damage types, and determining a target damage type from the multiple damage types, wherein the damage features are determined by the current design parameters and the training scenario parameters; using the planning subnet to determine target collision scenario parameters associated with the target damage type; and using the perturbation subnet to apply perturbations to the target collision scenario parameters to generate the adversarial scenario parameters.

[0011] According to one embodiment of the present invention, the step of using the strategy subnet to classify damage based on damage features to obtain multiple damage types, and determining a target damage type from the multiple damage types, includes: inputting the damage features into the strategy subnet, having the strategy subnet classify damage based on the damage features to obtain multiple damage types, and then calculating the vulnerability score corresponding to each damage type; and recording the damage type corresponding to the maximum vulnerability score as the target damage type.

[0012] According to one embodiment of the present invention, applying perturbation to the target collision scene parameters using the perturbation subnet includes: using the perturbation subnet to identify the physical laws corresponding to the target collision scene parameters, and applying perturbation to the target collision scene parameters according to the physical laws corresponding to the target collision scene parameters.

[0013] According to one embodiment of the present invention, the network architecture of the adversarial scenario generation model adopts one of conditional generative adversarial network, policy gradient reinforcement learning agent, and differentiable parameterized perturbation network.

[0014] According to one embodiment of the present invention, the method further includes: updating the current attack parameters of the adversarial scenario generation model using the second predicted damage value.

[0015] According to one embodiment of the present invention, updating the current attack parameters of the adversarial scenario generation model using the second predicted damage value includes: calculating the gradient of the second predicted damage value with respect to the current attack parameters; determining the update direction and update step size of the current attack parameters based on the gradient; and adjusting the update step size of the current attack parameters along the update direction to update the current attack parameters.

[0016] According to an embodiment of the present invention, obtaining a comprehensive loss function based on the first predicted damage value and the second predicted damage value includes: determining a first loss based on the first predicted damage value; determining a second loss based on the second predicted damage value; and obtaining the comprehensive loss function based on the first loss and the second loss.

[0017] To achieve the above objectives, a second aspect of the present invention proposes an optimization system for a vehicle constraint system. The system includes: a first prediction module, used to predict personal injury using a constraint system model to be optimized, based on the current design parameters and training scenario parameters of the vehicle constraint system, to obtain a first predicted injury value, wherein the training scenario parameters are scenario parameters for a preset collision involving the vehicle; an adversarial generation module, used to generate adversarial scenario parameters using an adversarial scenario generation model, based on the current design parameters and the training scenario parameters, wherein the adversarial scenario parameters are used to increase the predicted injury value output by the constraint system model to be optimized; a second prediction module, used to predict personal injury using the constraint system model to be optimized, based on the current design parameters and the adversarial scenario parameters, to obtain a second predicted injury value; a loss calculation module, used to obtain a comprehensive loss function based on the first predicted injury value and the second predicted injury value; and a parameter update module, used to update the current design parameters based on the comprehensive loss function and return to the step of predicting personal injury using the constraint system model to be optimized, based on the current design parameters and training scenario parameters of the vehicle constraint system, until a preset iteration condition is met to obtain target design parameters.

[0018] To achieve the above objectives, a third aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the optimization method for the vehicle restraint system as described above.

[0019] To achieve the above objectives, a fourth aspect of the present invention provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, characterized in that, when the computer program is executed by the processor, it implements the optimization method of the vehicle restraint system as described above.

[0020] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0021] Figure 1 This is a flowchart of an optimization method for a vehicle restraint system according to an embodiment of the present invention; Figure 2 This is a flowchart illustrating the generation of adversarial scenario parameters according to an embodiment of the present invention; Figure 3 This is a flowchart illustrating the determination of the target damage type according to an embodiment of the present invention; Figure 4 This is a schematic diagram of an optimization system for a vehicle restraint system according to an embodiment of the present invention; Figure 5 This is a structural block diagram of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0022] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.

[0023] The optimization techniques for related vehicle restraint systems employ traditional optimization methods based on parameter-sensitive design spaces. A typical process involves selecting a finite number of sample points within a pre-defined parameter space using experimental design methods (e.g., airbag deployment time 10-30 ms, seatbelt force limiting 2-4 kN). Next, high-fidelity CAE (Computer-Aided Engineering) simulations are performed on each sample point to obtain damage indicators. Finally, based on these discrete sample points, a medical response surface model or surrogate model is used to fit the relationship between damage and parameters, and an optimization algorithm is used to find the optimal solution. The fundamental flaw of this technique is that the entire optimization process is a passive, static sampling and fitting process, and its optimization effect is severely limited by whether the initial sample point selection adequately covers the true "performance boundary." For complex, nonlinear restraint systems, with a limited sample size, optimization is prone to getting trapped in local optima and cannot detect "design blind spots" existing in unknown parameter combinations or extreme scenarios.

[0024] Furthermore, optimization techniques for related vehicle constraint systems employ rule-based or predefined scenario libraries for testing. Some publicly available solutions attempt to introduce more test scenarios, such as establishing a scenario library containing multiple standard collision conditions (e.g., NCAP (New Car Assessment Programme) regulations) and some extended conditions. After optimization, the vehicle constraint system needs to be validated against this scenario library. A fundamental flaw in this testing method is its brute-force scenario enumeration and verification. Its scenario library is predefined and fixed, unable to dynamically generate new test cases targeting current design weaknesses as the design changes. Therefore, it can only validate known scenarios and cannot proactively uncover unknown risks.

[0025] To address the aforementioned problems, embodiments of the present invention provide an optimization method, system, storage medium, and electronic device for a vehicle restraint system. The optimization method, system, storage medium, and electronic device of the vehicle restraint system according to embodiments of the present invention will be described in detail below with reference to the accompanying drawings and specific implementation details.

[0026] Figure 1 This is a flowchart of an optimization method for a vehicle restraint system according to an embodiment of the present invention. Figure 1 As shown, the optimization method for the vehicle constraint system may include: S101, using the constraint system model to be optimized, perform personnel injury prediction based on the current design parameters of the vehicle constraint system and the training scenario parameters to obtain the first predicted injury value. The training scenario parameters are the scenario parameters when the vehicle experiences a preset collision.

[0027] Implementable, constrained system model to be optimized This could be a lightweight damage prediction surrogate model, such as a multilayer perceptron. The constrained system model to be optimized. The input is a concatenated vector. ,in, The current design parameters of the vehicle constraint system to be optimized (e.g.) ), These are the training scenario parameters (such as the speed and angle of the vehicle when a preset collision occurs), and their output is the predicted damage value (occupant injury index value (e.g., ...). It should be noted that the constraint system model to be optimized in the embodiments of the present invention is pre-trained using historical simulation data and remains unchanged during the optimization of the vehicle constraint system, serving as a fast evaluator of the current design performance of the vehicle constraint system.

[0028] Specifically, obtain the training scene parameter set. , training scene parameter set Training scene parameters Input the constraint system model to be optimized. The constraint system model is based on the current design parameters of the vehicle constraint system (such as seat belts, airbags, and seats). and training scene parameters Personnel injury prediction is performed to obtain the first predicted injury value.

[0029] It should be noted that the training scenario parameters in the training scenario parameter set of this embodiment of the invention can be scenario parameters when a vehicle actually collides, or scenario parameters when a simulated collision occurs. This embodiment of the invention does not limit the method of obtaining the training scenario parameters.

[0030] S102, using an adversarial scenario generation model, adversarial scenario parameters are generated based on the current design parameters and training scenario parameters. The adversarial scenario parameters are used to increase the predicted damage value output by the model of the constraint system to be optimized.

[0031] Adversarial Scene Generation Model in Embodiments of the Invention The goal is to generate a model of the constrained system to be optimized. Scenario parameters with higher damage values, i.e., finding a constrained system model to optimize. Adversarial testing scenario.

[0032] Specifically, the current design parameters and training scene parameters Input to adversarial scenario generation model Adversarial scenario generation model Based on current design parameters and training scene parameters Generates the current design parameters for the vehicle constraint system. The worst-case scenario parameters are used to obtain the current design parameters. Corresponding adversarial scenario parameters .

[0033] S103. Using the constraint system model to be optimized, personnel damage is predicted based on the current design parameters and the parameters of the adversarial scenario, and a second predicted damage value is obtained.

[0034] Specifically, the current design parameters and adversarial scenario parameters Input to the constrained system model to be optimized Constraint system model to be optimized Based on current design parameters and adversarial scenario parameters Personnel injury prediction is performed to obtain a second predicted injury value.

[0035] S104, the comprehensive loss function is obtained based on the first predicted damage value and the second predicted damage value.

[0036] Specifically, the first loss is determined based on the first predicted damage value and its corresponding loss function. The second loss is determined based on the second predicted damage value and its corresponding loss function. According to the first loss Second loss Determine the comprehensive loss function .

[0037] S105, update the current design parameters according to the comprehensive loss function, and return to the step of using the constraint system model to be optimized to predict personnel injury based on the current design parameters of the vehicle constraint system and the training scenario parameters, until the preset iteration conditions are met and the target design parameters are obtained.

[0038] Specifically, according to the comprehensive loss function Update the constraint system model to be optimized Current design parameters Repeat the above steps until the preset iteration conditions are met, such as minimizing the comprehensive loss function. To obtain the target design parameters .

[0039] The vehicle constraint system optimization method in this invention adopts a dynamic game theory approach. By competing with the constraint system model to be optimized and the adversarial scenario generation model, the performance boundary is continuously expanded, resulting in more robust optimization results that can cope with a wider range of unknown scenarios. The optimization and adversarial testing are transformed from a sequential process to a parallel and real-time interactive process, exposing the defects of the constraint system model to be optimized in the early design stage. The constraint system model to be optimized simultaneously enhances its "immunity" during iteration, greatly compressing the "design-defect exposure-redesign" cycle, significantly shortening the development cycle, reducing the cost of later modifications, and solving the problem of "verification lag".

[0040] The vehicle constraint system optimization method of this invention transforms the optimization process from a static mathematical optimization problem into a dynamic, adversarial game-based self-evolutionary process of the vehicle constraint system. Specifically, by constructing an autonomous closed loop of "performance weakness detection - adversarial scenario generation - design reinforcement," intelligent optimization is achieved, enabling the vehicle constraint system to self-expose defects and perform targeted enhancements. Compared with testing methods based on a fixed scenario library, the test scenarios used to test the vehicle constraint system in this invention are not preset, but are dynamically and in real-time generated by a model that engages in adversarial game with the current vehicle constraint system, aiming to continuously challenge the limits of the current vehicle constraint system.

[0041] In one embodiment of the present invention, such as Figure 2 As shown, the adversarial scenario generation model may include a policy subnet, a planning subnet, and a perturbation subnet connected in sequence. Using the adversarial scenario generation model, adversarial scenario parameters are generated based on the current design parameters and training scenario parameters, and may include: S201 utilizes a policy subnet to classify damage based on damage features, resulting in multiple damage types. The target damage type is then determined from these multiple damage types, where the damage features are determined by the current design parameters and training scenario parameters.

[0042] In practice, the damage characteristics are based on the current design parameters of the vehicle constraint system model to be optimized. and training scene parameters Damage features are obtained by feature extraction.

[0043] To realize the model of the constrained system to be optimized The output interpretability analysis (e.g., through gradient saliency maps) utilizes the policy subnet to identify the current design parameters of the vehicle constraint system at a macroscopic level. The most vulnerable damage type is denoted as the target damage type. (e.g., “chest compression”).

[0044] Specifically, the model of the constraint system to be optimized is used based on the current design parameters of the vehicle constraint system. and training scene parameters Feature extraction is performed to obtain damage features. These damage features are then input into a policy subnet, which classifies the damage features and determines the target damage type from multiple damage types. .

[0045] S202, using the planning subnet, determines the target collision scene parameters associated with the target damage type.

[0046] In an example implementation, a planned subnet is used to determine the target damage type from the meso-level. Identify at least one set of key physical parameters to be adjusted, denoted as the target collision scenario parameters. (For example, if the chest is attacked, then) This may include "collision offset rate" and "occupant displacement distance".

[0047] S203 uses a perturbation subnet to apply perturbations to the target collision scene parameters, generating adversarial scene parameters.

[0048] In practice, the perturbation subnet affects the target collision scenario parameters at the microscopic level. Apply perturbations to generate adversarial scenario parameters .

[0049] In this embodiment of the invention, the adversarial scenario generation model adopts a hierarchical adversarial scenario generator. This generator is a cascaded decision and generation network consisting of three layers: a strategy subnet, a planning subnet, and a perturbation subnet. Through a three-layer strategy of "macro-meta-micro", it dynamically generates adversarial test scenarios targeting the weaknesses of the current design.

[0050] In one specific embodiment of the present invention, such as Figure 3 As shown, using a policy subnet, damage is classified according to damage characteristics to obtain multiple damage types, and the target damage type is determined from these multiple damage types. This may include: S301, input the damage features into the strategy subnet, and the strategy subnet classifies the damage based on the damage features. After obtaining multiple damage types, it calculates the vulnerability score corresponding to each damage type. S302, the damage type corresponding to the maximum vulnerability score is denoted as the target damage type.

[0051] In practice, the policy subnet can be a pre-trained small classification network. The input to the policy subnet is the model of the constraint system to be optimized. Based on the current design parameters of the vehicle constraint system and training scene parameters Damage features are extracted. The policy subnet classifies the damage features into multiple damage types and calculates the vulnerability score for each type. The vulnerability scores for each damage type are compared, and the damage type with the highest vulnerability score is designated as the target damage type. .

[0052] In one embodiment of the present invention, using a planning subnet to determine target collision scene parameters associated with the target damage type may include: Input the target damage type into the planning subnet to obtain the target collision scene parameters associated with the target damage type.

[0053] In one specific embodiment, the planning subnet can employ a database encapsulating pre-defined domain knowledge rules (IF (condition) - THEN (conclusion) rules). Specifically, the target damage type... Input is fed into the database mentioned above, which is based on the target damage type. Perform a query to obtain at least one target collision scene parameter associated with the target damage type. .

[0054] In another specific embodiment, the planning subnetwork may employ a pre-trained prediction subnetwork, which can achieve... The mapping.

[0055] In one embodiment of the present invention, a perturbation subnet is used to apply perturbation to the target collision scene parameters, including: The perturbation subnet is used to identify the physical laws corresponding to the target collision scene parameters, and perturbation is applied to the target collision scene parameters according to the physical laws corresponding to the target collision scene parameters.

[0056] Specifically, the perturbation subnet identifies each target collision scenario parameter in the target design parameter set. The corresponding physical laws, for the parameters of the corresponding target collision scene Applying a smooth perturbation that conforms to its physical laws ensures the parameters of the generated adversarial scenario are... It is physically feasible.

[0057] For example, such as through a conditional variational autoencoder, learning in a given Under the condition of reasonable scenario perturbation The probability distribution is determined and sampled from it. For example, Bézier curves or superpositions of sine functions can be used to generate smooth time series perturbations.

[0058] In one embodiment of the present invention, the network architecture of the adversarial scenario generation model may be one of conditional generative adversarial networks, policy gradient reinforcement learning agents, or differentiable parameterized perturbation networks.

[0059] Adversarial Scene Generation Model in Embodiments of the Invention Any of the above architectures can be used, and other architecture types can be selected as needed. This embodiment of the invention does not limit the network architecture used in the adversarial scenario generation model.

[0060] For example, when the adversarial scenario generation model adopts a conditional generative adversarial network architecture, it can utilize adversarial generators. With random noise and current design parameters Using the above three-layer strategy as input, adversarial scenario parameters are generated. and output the parameters of the adversarial scenario. Adversarial generator Discriminator in The aim is to distinguish between real-world high-risk scenarios and generated scenarios. This is achieved through adversarial training, specifically by developing an adversarial generator. Learn to generate parameters for the current design Realistic high-risk scenarios.

[0061] For example, when the adversarial scenario generation model employs a policy gradient reinforcement learning agent, the state is defined as the current design parameters. Action is defined as the adjustment amount of scene parameters. The reward is defined as The increase in the output damage value. The agent employs the above three-layer strategy to learn the attack strategy by maximizing cumulative rewards.

[0062] For example, when the adversarial scene generation model employs a differentiable parameterized perturbation network, a neural network is constructed, taking into account the scene parameters when a vehicle experiences a preset collision. The above three-layer strategy is used to resist scene parameters. The generation of output perturbation ,make Training is performed by directly calculating the gradient of the loss function with respect to the perturbed network parameters and then backpropagating.

[0063] In one embodiment of the present invention, a comprehensive loss function is obtained based on a first predicted damage value and a second predicted damage value, including: The first loss is determined based on the first predicted damage value; The second loss is determined based on the second predicted damage value; The combined loss function is obtained based on the first loss and the second loss.

[0064] Specifically, the performance loss of the training scenario is calculated based on the first predicted damage value and its corresponding loss function, thus obtaining the first loss. Based on the second predicted damage value and its corresponding loss function, the performance loss function for the adversarial scenario is calculated, and the second loss is obtained. Based on the first loss Second loss Calculate the constrained system model to be optimized Comprehensive loss function .

[0065] In one embodiment, the comprehensive loss function :

[0066] in, For the comprehensive loss function, The first loss, This is the second loss. This is the weight for the second loss.

[0067] Second loss in the embodiments of the present invention The comprehensive loss function of the constrained system model to be optimized The penalty items.

[0068] In one embodiment of the present invention, the optimization method for the vehicle constraint system may further include: The current attack parameters of the adversarial scenario generation model are updated using the second predicted damage value.

[0069] This invention's embodiments optimize and update current design parameters. At the same time, it can also generate models for adversarial scenarios. Current attack parameters Update the adversarial scenario generation model to generate scenarios that can cause higher damage.

[0070] In one embodiment of the present invention, updating the current attack parameters of the adversarial scenario generation model using a second predicted damage value may include: Calculate the gradient of the second predicted damage value with respect to the current attack parameters; The update direction and update step size of the current attack parameters are determined based on the gradient; Adjust the update step size along the update direction to update the current attack parameters.

[0071] Specifically, the gradient of the second predicted damage value with respect to the current attack parameters is calculated based on the second predicted damage value of the current period and the previous period, thus obtaining the gradient of the current attack parameters. According to this gradient Determine the update direction and update step size of the current attack parameters, and adjust the update step size along the update direction to update the current attack parameters.

[0072] gradient in the embodiments of the present invention As a model for generating adversarial scenarios The updated gradient. This updated gradient... Instructions were given on how to fine-tune the adversarial scenario generation model. Current attack parameters This is to generate scenarios that can cause greater damage. Adversarial scenario generation model. Current attack parameters Update along this gradient direction: ,in, Its learning rate.

[0073] In one embodiment of the invention, a second predicted damage value can also be calculated relative to the current design parameters. The gradient is used to obtain the current design parameters. gradient Current design parameters in this embodiment of the invention gradient Directly reveals the current design parameters In adversarial scenario parameters The vulnerability direction is as follows. Embodiments of this invention do not directly rely on the first damage gradient. Update current design parameters This can serve as a punitive signal.

[0074] It should be noted that the mathematical essence of game theory is adversarial optimization based on gradients. This invention employs a dual-model evaluation and gradient feedback mechanism for adversarial optimization.

[0075] This invention constructs an optimization environment in which a constraint system model to be optimized and an adversarial scenario generation model coexist in a game.

[0076] The optimization method in this invention addresses two deeply interrelated problems in traditional optimization methods for constrained systems: 1) The "optimization myopia" problem: because optimization can only be performed within a limited, pre-selected parameter and scenario space, the system cannot recognize and cope with unknown risks beyond its performance boundaries, and the optimization results may perform poorly in a wider range of potential scenarios. 2) The "verification lag" problem: the exposure of design defects heavily relies on time-consuming later physical verification or accidental discovery, lacking a means to proactively, efficiently, and automatically expose the weakest link of the system during the digital design phase, resulting in long design iteration cycles and high costs.

[0077] The optimization method of the vehicle constraint system according to an embodiment of the present invention will be described in detail below: Initialization and Game Theory Start-up: Loading the Constraint System Model to be Optimized and initialize the adversarial scenario generation model. .

[0078] Using the constrained system model to be optimized Based on the current design parameters of the vehicle restraint system and training scene parameters Feature extraction is performed, and personnel injury prediction is based on the extracted injury features to obtain the first predicted injury value. Adversarial scenario generation model. From a macroscopic level, based on the model of the constraint system to be optimized Extracted damage features, select the most promising target damage type. (e.g., "concentrate on increasing chest compression"); then from the meso-level, based on the target lesion type The specific direction for adjusting scene parameters is planned (such as increasing the collision angle or changing the occupant's seating position). Target collision scene parameters associated with the target damage type are selected. Finally, at the micro level, continuous perturbations conforming to physical laws are applied to the selected target collision scene parameters to generate the final adversarial scene parameters. Parameters for adversarial scenarios Input into the current constraint system model to be optimized The second predicted damage value is obtained.

[0079] Two gradients can be calculated using a game evaluator: 1) Adversarial model update gradient: calculate the second predicted damage value with respect to the current attack parameters. gradient Used to update the anti-scene generation model To increase the resilience of scene generation models 1) Attack power; 2) Design model penalty gradient, calculate the second predicted damage value against the current design parameters. gradient It should be noted that, This demonstrates the vulnerability of the current design in such scenarios.

[0080] The optimization algorithm simultaneously obtains the model of the constrained system to be optimized. Performance loss in training scenarios and penalties from adversarial test scenarios This constitutes a comprehensive loss. The optimization algorithm minimizes To update the model of the constraint system to be optimized Design parameters This step allows optimization to simultaneously consider both general performance and robustness against adversarial attacks. Updated design parameters. and attack parameters The effects are applied to the two models respectively, initiating a new round of game theory. This process is repeated until the constrained system model to be optimized is reached. Generative models in adversarial scenarios The generated adversarial scenarios are stable and the overall performance meets the requirements.

[0081] It should be noted that the game theory approach can also employ a multi-agent reinforcement learning framework, in which the constraint system model to be optimized and the adversarial scenario generation model are regarded as two agents, with their action spaces being the design parameter adjustment and the scenario parameter adjustment, respectively, and their reward functions being opposites (one agent's reward damage decreases, while the other agent's reward damage increases), achieving equilibrium through competitive comparative learning.

[0082] The vehicle constraint system optimization method of this invention constructs a "co-progressive system of adversarial stress testing and reinforcement".

[0083] Specifically, optimizing the current design parameters of the vehicle constraint system In this study, a weakness detection mechanism based on a dual-model game theory approach is employed. The optimization system uses two parallel models: one is the constrained system model to be optimized, and the other is the adversarial scenario generation model corresponding to that constrained system model. The goal of the adversarial scenario generation model is no longer to fit the real data distribution, but rather to maximize the predicted damage of the constrained system model to be optimized. The constrained system model to be optimized and the adversarial scenario generation model constitute a game relationship: the adversarial scenario generation model strives to find scenario parameters that can "defeat" the current constrained system model to be optimized, while the constrained system model to be optimized defends against these attacks through optimization. This mechanism transforms the goal of "finding weaknesses" into a clear and optimizable mathematical problem (maximizing damage), thereby automatically and continuously uncovering high-risk scenario combinations that are difficult to reach with traditional sampling methods, fundamentally solving the problem of "optimization myopia."

[0084] The adversarial scenario generation model in this embodiment of the invention adopts a hierarchical and interpretable parameterized generation architecture. The adversarial scenario generation model abandons the method of indiscriminate random perturbation of the entire parameter space, and designs a hierarchical adversarial generator with an attack strategy divided into three layers: 1) Macro-strategy layer: determining the attack type (e.g., whether the attack point is chest protection or neck protection); 2) Meso-parameter layer: adjusting key physical parameters (e.g., collision offset rate, occupant displacement degree) under the selected strategy; 3) Micro-perturbation layer: applying continuous small perturbations to the selected parameters in accordance with physical laws. The decision at each layer is based on the interpretability analysis of the response of the current constraint system model to be optimized. Specifically, this architecture makes adversarial attacks no longer "black box" noise, but a targeted test case generation process with clear objectives, traceable processes, and clear physical meaning. This not only identifies weaknesses but also explains their causes (e.g., "in the case of occupant displacement and a large collision velocity gradient, the existing airbag deployment logic leads to premature head contact"), providing direct guidance for subsequent optimization of the constraint system model to be optimized.

[0085] The vehicle constraint system optimization method of this invention adopts a real-time closed-loop evolution process coupled with online optimization. Adversarial testing is deeply embedded in the optimization loop of the constraint system model to be optimized. In each iteration of the optimization algorithm, not only is the performance of the parameters of the current constraint system model under normal scenarios evaluated, but an adversarial scenario generation model is also simultaneously invoked to obtain the latest "worst-case scenario" for the current constraint system model and the resulting damage value. This damage value is added as a key penalty term to the comprehensive loss function of the constraint system model to be optimized, ensuring that the optimization process is always carried out under "adversarial pressure," driving the constraint system model to evolve towards robustness in adversarial scenarios.

[0086] In this invention, the adversarial scenarios are dynamically generated from the game process, are adaptive and infinite, and are specifically designed to address the weaknesses of the current constrained system model to be optimized. This allows for the discovery of blind spots and risks that exceed human prior experience, resulting in more thorough verification. The results are interpretable, outputting a "worst-case scenario" and its causal analysis, providing a clear performance boundary and vulnerability diagnostic report.

[0087] The vehicle constraint system optimization method of this invention involves a constraint system model to be optimized and an adversarial scenario generation model, and optimizes the current design parameters of the vehicle constraint system. In this process, an adversarial scenario generation model is used to dynamically generate adversarial test scenarios targeting the current design weaknesses of the vehicle constraint system. Simultaneously, the performance of the constraint system model to be optimized is evaluated in both the training and adversarial test scenarios. High damage values ​​in the adversarial test scenarios are used as a penalty term and incorporated into the comprehensive loss function of the constraint system model to be optimized. The optimization algorithm updates both the constraint system model to be optimized and the adversarial scenario generation model by minimizing this comprehensive loss function. During this optimization process, the constraint system model to be optimized and the adversarial scenario generation model engage in a game, continuous adversarial process, and evolution until the performance of the optimized system converges, forming an integrated intelligent design closed loop of "testing-optimization".

[0088] This invention provides an optimization system for a vehicle constraint system.

[0089] Figure 4 This is a schematic diagram of an optimization system for a vehicle restraint system according to an embodiment of the present invention. Figure 4 As shown, the optimization system 100 of the vehicle constraint system may include a first prediction module 10, an adversarial generation module 20, a second prediction module 30, a loss calculation module 40, and a parameter update module 50.

[0090] The system comprises the following modules: a first prediction module 10, which uses the constraint system model to be optimized to predict personal injury based on the current design parameters and training scenario parameters of the vehicle constraint system, and obtains a first predicted injury value; an adversarial generation module 20, which uses an adversarial scenario generation model to generate adversarial scenario parameters based on the current design parameters and training scenario parameters; a second prediction module 30, which uses the constraint system model to be optimized to predict personal injury based on the current design parameters and adversarial scenario parameters, and obtains a second predicted injury value; a loss calculation module 40, which obtains a comprehensive loss function based on the first and second predicted injury values; and a parameter update module 50, which updates the current design parameters based on the comprehensive loss function and returns to the step of using the constraint system model to be optimized to predict personal injury based on the current design parameters and training scenario parameters of the vehicle constraint system, until a preset iteration condition is met to obtain the target design parameters.

[0091] It should be noted that other specific implementations of the vehicle restraint system optimization system provided in the embodiments of the present invention can be found in other specific implementations of the vehicle restraint system optimization method of the above embodiments of the present invention.

[0092] The vehicle constraint system optimization system in this embodiment of the invention involves a constraint system model to be optimized and an adversarial scenario generation model, and optimizes the current design parameters of the vehicle constraint system. In this process, an adversarial scenario generation model is used to dynamically generate adversarial test scenarios targeting the current design weaknesses of the vehicle constraint system. Simultaneously, the performance of the constraint system model to be optimized is evaluated in both the training and adversarial test scenarios. High damage values ​​in the adversarial test scenarios are used as a penalty term and incorporated into the comprehensive loss function of the constraint system model to be optimized. The optimization algorithm updates both the constraint system model to be optimized and the adversarial scenario generation model by minimizing this comprehensive loss function. During this optimization process, the constraint system model to be optimized and the adversarial scenario generation model engage in a game, continuous adversarial process, and evolution until the performance of the optimized system converges, forming an integrated intelligent design closed loop of "testing-optimization".

[0093] This invention provides a computer-readable storage medium.

[0094] In this embodiment, a computer program is stored on a computer-readable storage medium, and when the computer program is executed by a processor, it implements the vehicle restraint system optimization method as described above.

[0095] This invention provides an electronic device.

[0096] In this embodiment, the electronic device may include a memory and a processor. The memory stores a computer program, and when the computer program is executed by the processor, it implements the optimization method of the vehicle restraint system as described above.

[0097] Figure 5 This is a structural block diagram of an electronic device according to an embodiment of the present invention.

[0098] like Figure 5 As shown, the electronic device 500 includes a processor 501 and a memory 503. The processor 501 and the memory 503 are connected, for example, via a bus 502. Optionally, the electronic device 500 may also include a transceiver 504. It should be noted that in practical applications, the transceiver 504 is not limited to one type, and the structure of this electronic device 500 does not constitute a limitation on the embodiments of the present invention.

[0099] Processor 501 may be a CPU (Central Processing Unit), a general-purpose processor, a DSP (Digital Signal Processor), an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. It can implement or execute the various exemplary logic blocks, modules, and circuits described in conjunction with the disclosure of this invention. Processor 501 may also be a combination that implements computational functions, such as including one or more microprocessor combinations, a combination of a DSP and a microprocessor, etc.

[0100] Bus 502 may include a pathway for transmitting information between the aforementioned components. Bus 502 may be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus, etc. Bus 502 can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 5 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0101] The memory 503 stores a computer program corresponding to the optimization method of the vehicle restraint system in the above embodiments of the present invention. This computer program is controlled and executed by the processor 501. The processor 501 executes the computer program stored in the memory 503 to implement the content shown in the foregoing method embodiments.

[0102] Among them, electronic devices 500 include, but are not limited to: mobile terminals such as mobile phones, laptops, digital radio receivers, PDAs (personal digital assistants), PADs (tablet computers), PMPs (portable multimedia players), and in-vehicle terminals (such as in-vehicle navigation terminals), as well as fixed terminals such as digital TVs and desktop computers. Figure 5 The electronic device 500 shown is merely an example and should not be construed as limiting the functionality and scope of use of the embodiments of the present invention.

[0103] It should be noted that the logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be specifically implemented in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of computer-readable media include: an electrical connection having one or more wires (electronic device), a portable computer disk drive (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Alternatively, the computer-readable medium may be paper or other suitable media on which the program can be printed, since the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in a computer memory.

[0104] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0105] The computer-readable storage medium and electronic device in the embodiments of the present invention optimize the model of the constraint system to be optimized based on the above-mentioned vehicle constraint system optimization method, determine the optimal design parameters of the vehicle constraint system, significantly shorten the development cycle, reduce the cost of later modification, and also solve the problem of "verification lag".

[0106] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0107] In the description of this invention, it should be understood that the terms "center," "longitudinal," "lateral," "length," "width," "thickness," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," "outer," "clockwise," "counterclockwise," "axial," "radial," and "circumferential" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing this invention and simplifying the description, and are not intended to indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0108] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this invention, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0109] In this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," "linking," and "fixing," etc., should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components, unless otherwise explicitly limited. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0110] In this invention, unless otherwise explicitly specified and limited, "above" or "below" the second feature can mean that the first feature is in direct contact with the second feature, or that the first feature is in indirect contact with the second feature through an intermediate medium. Furthermore, "above," "over," and "on top" of the second feature can mean that the first feature is directly above or diagonally above the second feature, or simply that the first feature is at a higher horizontal level than the second feature. "Below," "below," and "under" the second feature can mean that the first feature is directly below or diagonally below the second feature, or simply that the first feature is at a lower horizontal level than the second feature.

[0111] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. An optimization method for a vehicle constraint system, characterized in that, The method includes: Using the constraint system model to be optimized, personal injury prediction is performed based on the current design parameters and training scenario parameters of the vehicle constraint system to obtain a first predicted injury value. The training scenario parameters are the scenario parameters when the vehicle experiences a preset collision. Using an adversarial scenario generation model, adversarial scenario parameters are generated based on the current design parameters and the training scenario parameters, wherein the adversarial scenario parameters are used to increase the predicted damage value output by the constraint system model to be optimized; Using the constrained system model to be optimized, personnel injury prediction is performed based on the current design parameters and the adversarial scenario parameters to obtain a second predicted injury value; A comprehensive loss function is obtained based on the first predicted damage value and the second predicted damage value; The current design parameters are updated according to the comprehensive loss function, and the process returns to the step of using the constraint system model to be optimized to predict personnel injury based on the current design parameters of the vehicle constraint system and the training scenario parameters, until the preset iteration conditions are met and the target design parameters are obtained.

2. The optimization method for the vehicle constraint system according to claim 1, characterized in that, The adversarial scenario generation model includes a policy subnet, a planning subnet, and a perturbation subnet connected in sequence. The step of generating adversarial scenario parameters using the adversarial scenario generation model based on the current design parameters and the training scenario parameters includes: Using the policy subnet, damage is classified according to damage features to obtain multiple damage types, and a target damage type is determined from the multiple damage types, wherein the damage features are determined by the current design parameters and the training scenario parameters; Using the planning subnet, target collision scene parameters associated with the target damage type are determined; The perturbation subnet is used to perturb the target collision scene parameters to generate the adversarial scene parameters.

3. The optimization method for the vehicle constraint system according to claim 2, characterized in that, The process of using the strategy subnet to classify damage based on damage characteristics, obtaining multiple damage types, and determining the target damage type from these multiple damage types includes: The damage features are input into the strategy subnet, which classifies the damage based on the damage features, obtains multiple damage types, and then calculates the vulnerability score corresponding to each damage type. The damage type corresponding to the maximum vulnerability score is denoted as the target damage type.

4. The optimization method for the vehicle constraint system according to claim 2, characterized in that, The step of applying perturbation to the target collision scene parameters using the perturbation subnet includes: The perturbation subnet is used to identify the physical laws corresponding to the target collision scene parameters, and perturbation is applied to the target collision scene parameters according to the physical laws corresponding to the target collision scene parameters.

5. The optimization method for the vehicle constraint system according to claim 1, characterized in that, The network architecture of the adversarial scenario generation model adopts one of the following: conditional generative adversarial network, policy gradient reinforcement learning agent, or differentiable parameterized perturbation network.

6. The optimization method for the vehicle constraint system according to claim 1, characterized in that, The method further includes: The second predicted damage value is used to update the current attack parameters of the adversarial scenario generation model.

7. The optimization method for the vehicle constraint system according to claim 6, characterized in that, The step of updating the current attack parameters of the adversarial scenario generation model using the second predicted damage value includes: Calculate the gradient of the second predicted damage value with respect to the current attack parameters; The update direction and update step size of the current attack parameters are determined based on the gradient. The update step size is adjusted along the update direction to update the current attack parameters.

8. The optimization method for the vehicle constraint system according to claim 1, characterized in that, The step of obtaining a comprehensive loss function based on the first predicted damage value and the second predicted damage value includes: The first loss is determined based on the first predicted damage value; The second loss is determined based on the second predicted damage value; The comprehensive loss function is obtained based on the first loss and the second loss.

9. An optimization system for a vehicle constraint system, characterized in that, The system includes: The first prediction module is used to predict personal injury based on the current design parameters and training scenario parameters of the vehicle constraint system using the constraint system model to be optimized, and to obtain a first predicted injury value. The training scenario parameters are the scenario parameters when the vehicle causes a preset collision. The adversarial generation module is used to generate adversarial scenario parameters based on the current design parameters and the training scenario parameters using an adversarial scenario generation model. The adversarial scenario parameters are used to increase the predicted damage value output by the constraint system model to be optimized. The second prediction module is used to use the constraint system model to be optimized to predict personnel damage based on the current design parameters and the adversarial scenario parameters, and obtain a second predicted damage value. The loss calculation module is used to obtain a comprehensive loss function based on the first predicted damage value and the second predicted damage value; The parameter update module is used to update the current design parameters according to the comprehensive loss function and return to the step of using the constraint system model to be optimized to predict personnel injury based on the current design parameters of the vehicle constraint system and the training scenario parameters, until the preset iteration conditions are met and the target design parameters are obtained.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the optimization method for the vehicle restraint system as described in any one of claims 1-8.

11. An electronic device, comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the computer program is executed by the processor, it implements the optimization method for the vehicle restraint system as described in any one of claims 1-8.

Images