A method for joint generalization simulation

Abstract

This invention belongs to the field of vehicle networking and specifically relates to a joint generalization simulation method. The method includes the following steps: Step 1, establishing a simulation element set; Step 2, selecting key elements to combine into a key element set; Step 3, building a complete vehicle model, exporting the vehicle model, and importing it into a data processing center; Step 4, building dynamic / static scene models and inputting them into the data processing center; Step 5, completing the generalization simulation through the data processing center. In this invention, based on the joint simulation of dynamics simulation software, scene simulation software, and a data processing center, the problem of unconvincing single simulations and the time-consuming modeling of multiple simulations is addressed. While ensuring model accuracy, the platform is further developed to achieve the goal of generalizing the key elements of the research scenario, which is very helpful in reducing design time costs and controlling model consistency in simulation.

Description

Technical Field

[0001] This invention belongs to the field of vehicle networking, and specifically relates to a joint generalized simulation field. Background Technology

[0002] Currently, the selection of simulation software for ADAS and C-V2X is limited. Simulation scenario construction often relies on manual setup of single scenarios, making batch testing impossible. Furthermore, generalized simulations are typically only applicable to specific simulation software, resulting in low model accuracy. In other words, simulation modeling is slow, making it difficult to implement high-dimensional test cases. Summary of the Invention

[0003] To address the aforementioned problems, the present invention aims to provide a method for joint generalized simulation, which offers fast simulation modeling and facilitates the testing of high-dimensional test cases.

[0004] To achieve the above objectives, the technical solution of the present invention is as follows.

[0005] A method for joint generalization simulation, characterized by comprising the following steps: Step 1: Construct a set of simulation elements from the simulation elements. Step 2: For a specific test scenario, filter the set of simulation elements and select key elements to combine into a key element set. Step 3: Based on the key elements in the key element set, build the complete vehicle model in the dynamics simulation software, export the complete vehicle model, and import it into the data processing center. Step 4: Based on the key elements in the key element set, build dynamic / static scene models in the scene simulation software and input the dynamic / static scene models into the data processing center; Step 5: The data processing center finds the key elements that need to be generalized in the dynamic / static scene model and the vehicle model, and generalizes these key elements to complete the generalization simulation through the data processing center.

[0006] This method utilizes a combined simulation based on dynamics simulation software, scene simulation software, and a data processing center. To address the issues of unconvincing single simulations and time-consuming modeling multiple simulations, the platform is further developed while ensuring model accuracy. This approach aims to generalize the key elements of the research scenario, which is very helpful in reducing design time costs and ensuring model consistency in simulation.

[0007] Furthermore, the simulation element set includes four categories of simulation elements: people, vehicles, roads, and environment. These four categories are broad categories, which can be further subdivided into more subcategories.

[0008] Furthermore, when selecting key elements, it is determined whether more than one category of the selected key elements is missing. If more than one category is missing, the selection fails. Here, "one category" refers to a broad category, such as people, vehicles, roads, and environment. If any two or more of these four categories are missing, the selection fails.

[0009] Furthermore, a key element screening matrix is ​​established. The horizontal axis includes the influence on self-class elements and the existence of semantic duplication within self-class elements; the vertical axis includes the main research parameters and the influence on other classes. The key element screening matrix includes δ, ε, and two θ values. One θ value has the horizontal axis representing the influence on self-class elements and the vertical axis representing the influence on other classes. The other θ value has the horizontal axis representing the existence of semantic duplication within self-class elements and the vertical axis representing the main research parameters. ε value has the horizontal axis representing the influence on self-class elements and the vertical axis representing the main research parameters. δ value has the horizontal axis representing the existence of semantic duplication within self-class elements and the vertical axis representing the influence on other classes. Key research parameters: Subjective judgment, determining whether the simulation element is a key element of interest in the research scenario; 1 for yes, 0 for no. Semantic duplication within a class: An objective judgment is made to determine whether the simulation element has semantic duplication with other elements of the same class. If yes, the value is 0; if no, the value is 1. Impact on other categories: Objectively judge whether the simulation element has an impact on other major categories of simulation elements; if yes, the value is 0.6, otherwise it is 0. Impact on self-class: Objectively judge whether the simulation element has an impact on the simulation element of this class; if yes, the value is 0.6, otherwise it is 0. In the key element screening matrix, the θ value depends on the maximum value of the horizontal and vertical axes, the ε value is the value of the main research parameter, and the δ value is the value that has an impact on other classes; after obtaining the screening matrix representing the element, the F-norm of the matrix is ​​calculated. When the F-norm = 0, it proves that the element has no value for the study of the scene and should be discarded. When the F-norm is greater than 1.5, it proves that the element has a profound impact on the scene and must be retained; When 0 < F-norm < 1.5, the value is directly proportional to the research value of the element for the scene.

[0010] That is, in cases where there is semantic repetition among multiple key elements, the key element with the larger norm value is directly selected and retained, while the one with the smaller norm value is discarded.

[0011] Furthermore, when selecting key elements, it is determined whether the selected key elements can be reproduced in the dynamic / static scene model or the vehicle model, and key elements that cannot be reproduced are deleted.

[0012] Furthermore, the construction of the complete vehicle model specifically includes: overall vehicle dimensions, aerodynamic parameters, transmission system parameters, steering system, braking system, and suspension system. These parameters correspond to key elements within the vehicle and environmental categories.

[0013] Furthermore, the construction of dynamic / static scene models specifically includes: road information, lane information, and traffic participant information. These parameters correspond to key elements under the categories of people and roads.

[0014] Furthermore, if sensor information is added when building dynamic / static scene models, the design of dynamic / static scene models can also include sensor information; if sensor information is added when building dynamic / static scene models, the frequency of ADAS model algorithms in the data processing center can be set to be consistent with the simulation frequency in the data processing center.

[0015] The beneficial effects of this invention are that, in this method, based on the joint simulation of dynamic simulation software, scene simulation software, and data processing center, the problem of single simulation lacking persuasiveness and multiple simulations taking too long to model is addressed. On the basis of ensuring model accuracy, the platform is further developed to achieve the purpose of generalizing the key elements of the research scene, which is very helpful in reducing design time costs and controlling the consistency of model simulation. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the framework of the present invention.

[0017] Figure 2 It is a key element screening matrix. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0019] This embodiment provides a method for joint generalization simulation; its specific solution is a scenario generalization method based on a simulation software platform, which mainly consists of three steps.

[0020] (Step 1) First, to obtain the simulation element set, which consists of key and secondary elements, prior knowledge is used to screen the key elements. Specific test cases are designed to extract the key elements from the experimental scenario, and target parameters to be extracted are designed for the test. An element model is then built in the simulation software by classifying the elements. The process from Step 1 to Step 2 and then to Step 3 is a unidirectional and irreversible generalization scenario step. Step 1 provides the scenario elements simulated in Step 2 and provides the generalized elements for Step 3.

[0021] The specific details are as follows: First, assuming the principle that all elements of the real physical world cannot be exhaustively represented in the simulation environment, the set of elements obtained for a specific test scenario is: Based on the principles of scenario testing, a subjective judgment is made on the set of elements, and key elements are filtered out. The set of key elements after filtering is a subset, denoted as... The element set can be considered as the set of general elements contained in the research scenario, usually without considering the limitations of simulation and the overlap of semantic construction. The key element set is selected from the element set to achieve the goal of constructing the target test scenario with the shortest description, thereby shortening the test cycle and making the test more efficient.

[0022] The scene element testing and judgment rules are as follows: The set of elements should ideally include four categories: people, vehicles, roads, and environment. For each research scenario, elements are derived by analyzing each category. For example, when screening the environment in an urban expressway, this includes major categories such as road type, lane attributes, road surface alignment, and traffic signs. These major categories are then broken down into smaller, specific elements. It's important to note that not all scenarios include all four categories; in certain scenarios, one element may be missing (e.g., pedestrians are not considered in highway research). If more than one element is missing, the test scenario judgment rule becomes invalid.

[0023] The rules for determining key elements of a scene are as follows: ① Elements belonging to the same major category but with different wording but similar semantics need to be judged based on their specific circumstances. However, the selection process will adhere to the principle of including all elements as much as possible while minimizing semantic redundancy. Specific selection rules are as follows: A key element screening matrix exists: Impact on other humans θ δ elements Influence on self-type elements There is repetition in the semantics of the self-class. Key research parameters: subjective judgment, whether it is a key element of the research scenario (1 for yes, 0 for no); Semantic duplication within a self-category: Objectively judge whether there is semantic duplication with other elements of the self-category item, such as weather conditions and lighting conditions; if yes, the value is 0, and if no, the value is 1. Impact on other categories: Objectively judge whether it affects scene elements of other categories, such as weather conditions of the environmental category affecting the road surface adhesion coefficient of the road category. If yes, the value is 0.6; otherwise, it is 0.

[0024] Impact on its own class: Objectively judge whether it affects the scene elements of this class. For example, if the weather affects the rainfall, the value is 0.6 if it does, and 0 if it does not.

[0025] Based on the analysis of the elements, four values ​​are obtained and then filled into the matrix above. In the matrix, the θ value depends on the maximum value of the horizontal and vertical coordinates, the ε value is the main research parameter value, and δ is the value that affects other classes. After obtaining the screening matrix representing the element, the F-norm of the matrix is ​​calculated. The magnitude of the norm value represents the influence of the element on the overall scene.

[0026] When the F-norm = 0, it proves that the element has no value for the study of the scene and should be discarded. When the F-norm is greater than 1.5, it proves that the element has a profound impact on the scene and must be retained; When 0 < F-norm < 1.5, the value is directly proportional to the research value of the element for the scene.

[0027] ② The element must be reproducible in the simulation scene; otherwise, it will be deleted. When an element selected by rule ① conflicts with this rule, this rule will be used as the judgment condition.

[0028] (Step Two) Next, based on the scenarios generated by the test cases, a model is built in the simulation software according to the element attributes. The specific settings are as follows: ① Dynamic model building: Design a complete vehicle model. Specifically, this includes: 1) Overall vehicle dimensions (length, width, height, wheelbase, track width, minimum ground clearance, front and rear overhangs, approach angle, departure angle, etc.); 2) Aerodynamic parameters (drag coefficient, lateral force coefficient, lift coefficient, roll coefficient, pitch coefficient, point of aerodynamic application, frontal area, air density, etc.); 3) Transmission system parameters (drive method, natural frequency and damping of the transmission system). It should be noted that the transmission model should be designed according to the selected drive method.

[0029] 4) Steering system (pinion torque, controls front wheel steering angle) 5) Braking system (pipeline transmission delay time, ABS settings, brake pressure distribution, actuator inertial hysteresis time) 6) Suspension system (suspension center of gravity Y-coordinate, unsprung mass, left and right wheel moments of inertia, etc.) ② Construction of some dynamic / static scene models By building dynamic / static scene models in simulation software, as detailed below: 1) Road information (road starting point, length, width, curve radius, road alignment) 2) Lane information (lane width, number of lanes, lane markings) 3) Traffic participant information (participant type, participant location, movement status, direction of movement, and movement path) 4) Sensor information (For vehicles equipped with ADAS sensors, it is also necessary to set the sensor type, sensor installation location, maximum number of recognitions, recognition frequency, recognition range, signal transmission method, etc. For vehicles equipped with V2X OBU, it is necessary to set the interface type, transmission protocol, transmission range, etc.).

[0030] ③ Algorithm Model Construction If sensor information is added in step ②, the algorithm frequency of the ADAS model being measured usually needs to match the simulation frequency set in the simulation software. This model is generally used in the simulation as the research object or as an invariant for studying other factors.

[0031] (Step 3) The first two steps are only applicable to building a single simulation scene. After the scene model is built, find the parameter module that needs to be generalized and rename it. The specific steps are as follows: Create a new script file in the "Data Processing Center" and type the following code: SetPointValues=[A1, A2, A3, A4...An]; (Sets the parameter group that the key element A needs to traverse in the simulation) spv_length = length(SetPointValues); SetPointValues1 = [B1, B2, B3, B4...Bm]; (Sets the parameter group that the key element B needs to traverse in the simulation) spv_length1 = length(SetPointValues1); SetPointValues2 = [C1, C2, C3, C4...Cp]; (Sets the parameter group that the key element C needs to traverse in the simulation) spv_length2 = length(SetPointValues2); ...(It is necessary to generalize the h key element parameters, so set the statements above the h group) for j = spv_length1:-1:1 (the name of the "length" corresponding to the key element A) fprintf('n: = %d\n',j) ("n" is the number of times loop A is executed) in = Simulink.SimulationInput('Project_address'); (Physical location of the project file) in = in.setBlockParameter('addressA','Value',num2str(SetPointValues1(j))); (addressA is the location of module A in Simulink) fprintf('AA = %d\n',SetPointValues1(j)) for k = spv_length2:-1:1 (the name of the "length" corresponding to the key element B) fprintf('m:=%d\n',k) ("m" is the number of times the B loop is executed) in = Simulink.SimulationInput('Project_address'); (Physical location of the project file) in = in.setBlockParameter('addressB','Value',num2str(SetPointValues2(k))); (addressB is the location of module B in Simulink) fprintf('CC = %d\n',SetPointValues2(k)) ...(The number of nested for loops corresponds to the number of key elements that are generalized.) simout=sim(in); end End End (the number of `end` iterations is the same as the number of `For` iterations) The above method enables generalized simulation; once the code is entered, the platform can automatically perform the generalized simulation. The text in parentheses following the code above is a description of that line of code.

[0032] Specifically, the choice of vehicle dynamics simulation software includes, but is not limited to, Carsim, Carmaker, and Carlasimulink; The choice of dynamic and static scene simulation software includes, but is not limited to, VTD, Prescan, and Prosivic (PTVvissim). The choice of data processing center includes, but is not limited to, Matlab, Python, and Visual Studio.

[0033] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for joint generalization simulation, characterized in that, The method includes the following steps: Step 1: Construct a set of simulation elements from the simulation elements. Step 2: For a specific test scenario, filter the set of simulation elements and select key elements to combine into a key element set. Step 2 specifically includes: establishing a key element screening matrix, where the horizontal axis includes the influence on self-class elements and the existence of semantic duplication within self-class elements; the vertical axis includes the main research parameters and the influence on other classes; the key element screening matrix is ​​set with δ, ε, and two θ; one θ has the horizontal axis representing the influence on self-class elements and the vertical axis representing the influence on other classes; the other θ has the horizontal axis representing the existence of semantic duplication within self-class elements and the vertical axis representing the main research parameters; the horizontal axis of ε has the influence on self-class elements and the vertical axis representing the main research parameters; the horizontal axis of δ has the existence of semantic duplication within self-class elements and the vertical axis representing the influence on other classes. Key research parameters: Determine whether the simulation element is a key element of interest in the research scenario; 1 for yes, 0 for no. Semantic duplication within a class: Determines whether the simulation element has semantic duplication with other elements of the same class; 0 indicates yes, 1 indicates no. Impact on other categories: Determine whether the simulation element affects simulation elements of other major categories; 0.6 if yes, 0 if no. Impact on self-class: Determine whether the simulation element has an impact on the simulation elements of this class; 0.6 if yes, 0 if no. In the key element screening matrix, the θ value depends on the maximum value of the horizontal and vertical axes, the ε value is the value of the main research parameter, and the δ value is the value that has an impact on other classes; after obtaining the screening matrix representing the element, the F-norm of the matrix is ​​calculated. When the F-norm = 0, it proves that the element has no value for the study of the scene and should be discarded. When the F-norm is greater than 1.5, it proves that the element has a profound impact on the scene and must be retained; When 0 < F-norm < 1.5, the value is directly proportional to the research value of the element for the scene; Step 3: Based on the key elements in the key element set, build the complete vehicle model in the dynamics simulation software, export the complete vehicle model, and import it into the data processing center. Step 4: Based on the key elements in the key element set, build dynamic / static scene models in the scene simulation software and input the dynamic / static scene models into the data processing center; Step 5: The data processing center finds the key elements that need to be generalized in the dynamic / static scene model and the vehicle model, and generalizes these key elements to complete the generalization simulation through the data processing center.

2. The method for joint generalization simulation according to claim 1, characterized in that, The simulation element set includes four categories of simulation elements: people, vehicles, roads, and environment.

3. The method for joint generalization simulation according to claim 2, characterized in that, When filtering out key elements, it is determined whether the number of missing categories of the filtered key elements is greater than one. If it is greater than one, the filtering is invalid.

4. The method for joint generalization simulation according to claim 2, characterized in that, When selecting key elements, determine whether the selected key elements can be reproduced in the dynamic / static scene model or the vehicle model, and delete key elements that cannot be reproduced.

5. The method for joint generalization simulation according to claim 2, characterized in that, The specific steps involved in building a complete vehicle model include: overall vehicle dimensions, aerodynamic parameters, transmission system parameters, steering system, braking system, and suspension system.

6. The method for joint generalization simulation according to claim 2, characterized in that, The specific elements included when building dynamic / static scene models are: road information, lane information, and traffic participant information.

7. The method for joint generalization simulation according to claim 6, characterized in that, When building dynamic / static scene models, sensor information is also included. If sensor information is added when building dynamic / static scene models, the frequency of the ADAS model algorithm in the data processing center is set to be consistent with the simulation frequency in the data processing center.

Images