Weight parameter determination method based on linear optimization, vehicle, device and medium

Abstract

The application provides a weight parameter determination method based on linear optimization, a vehicle, equipment and a medium, and belongs to the technical field of automatic driving. The method comprises the following steps: constructing a cost matrix according to the cost value of each candidate path of the vehicle under a preset driving condition; each row of the cost matrix corresponds to a candidate path, and each column corresponds to the cost value of an index type; obtaining an optimal motion trajectory under the preset driving condition, and determining an optimal path index from the multiple candidate paths according to the optimal motion trajectory; based on the cost matrix and the optimal path index, an optimization objective function is established, wherein the weight parameter is used as a decision variable, and the optimal path total cost value is the minimum; based on a linear optimization algorithm, the optimization objective function is solved to obtain an optimal weight parameter combination that meets a preset constraint condition, and the optimal weight parameter combination comprises a weight parameter corresponding to an index type. The application can automatically and efficiently determine the optimal weight parameter combination of the cost function in local path planning.

Description

Technical Field

[0001] This invention relates to the field of autonomous driving technology, specifically to a method for determining weight parameters based on linear optimization, a vehicle, equipment, and a medium. Background Technology

[0002] Autonomous driving is an automotive technology that enables vehicles to drive autonomously without direct human driver intervention. In the field of autonomous driving and intelligent vehicles, local path planning is a core component for achieving safe and efficient driving.

[0003] In the local path planning process of autonomous driving, it is typically necessary to generate multiple candidate trajectories (i.e., path clusters) in real time based on the vehicle's current state, road structure, and obstacle distribution. The comprehensive performance of each trajectory is then evaluated using a multi-objective cost function, ultimately selecting the optimal path for execution. Commonly used metrics include lateral deviation (distance from the reference path), curve length (total trajectory length), curve curvature (reflecting driving smoothness and safety), and relative distance to obstacles (collision risk). However, determining the weights of these metric types presents numerous challenges in practical applications. The significant differences in the magnitude of different metric types lead to a large range in weights, further complicating weight determination.

[0004] Currently, the weighting of these indicator types often relies on extensive experimentation and experience; however, this method has significant drawbacks: First, it is time-consuming and laborious, and it is difficult to quickly find the optimal weight combination that meets expectations.

[0005] Second, manually set weight combinations are often suboptimal solutions in local areas and cannot guarantee global optimality. In particular, in complex dynamic scenarios (such as dense obstacles and sharp bends), fixed weights may not be able to adapt to changing environmental constraints.

[0006] Third, slight adjustments to a single weight may lead to drastic changes in the planning results ("parameter sensitivity"), while weights optimized for a specific scenario may have significantly reduced performance under other conditions ("insufficient generalization").

[0007] Therefore, there is an urgent need for an efficient and automated method to determine scientific and reasonable weights so that the multi-objective cost function can accurately reflect the balance between driving safety, comfort and efficiency, which has become a key bottleneck in improving the performance of autonomous driving planning algorithms. Summary of the Invention

[0008] In view of this, it is necessary to provide a method, vehicle, equipment and medium for determining weight parameters based on linear optimization, in order to solve the technical problems of difficulty in determining weights in the prior art, low efficiency of relying on manual weight setting and difficulty in obtaining the optimal combination of weight parameters.

[0009] To address the aforementioned technical problems, in a first aspect, the present invention provides a method for determining weight parameters based on linear optimization, comprising: A cost matrix is ​​constructed based on the cost of each candidate path under preset driving conditions; each row of the cost matrix corresponds to one candidate path, and each column corresponds to the cost of an index type. Obtain the optimal motion trajectory under the preset driving conditions, and determine the optimal path index from multiple candidate paths based on the optimal motion trajectory; Based on the cost matrix and the optimal path index, an optimization objective function is established that uses weight parameters as decision variables and minimizes the total cost value of the optimal path. Based on a linear optimization algorithm, the optimization objective function is solved to obtain the optimal combination of weight parameters that satisfies the preset constraints. The optimal combination of weight parameters includes weight parameters corresponding to the index type.

[0010] In one possible implementation, constructing a cost matrix based on the cost of each candidate path under preset driving conditions includes: Set the path generation parameters and the preset driving conditions; A path cluster is generated based on the path generation parameters and the preset driving conditions; the path cluster includes n candidate paths, where n≥2 and n is a positive integer. Calculate the cost of all the indicator types for each of the candidate paths; The cost matrix is ​​constructed based on the cost value of the corresponding indicator type for each candidate path.

[0011] In one possible implementation, the index types include at least lateral deviation values, curve length values, curve curvature values, and obstacle distance values.

[0012] In one possible implementation, obtaining the optimal trajectory under the preset driving condition and determining the optimal path index from multiple candidate paths based on the optimal trajectory includes: Acquire real trajectory data generated by a driving expert driving the vehicle under the preset driving conditions, and use the real trajectory data as the optimal motion trajectory; The optimal trajectory is matched with all the candidate paths for similarity, and the index corresponding to the candidate path with the highest similarity is determined as the index of the optimal path.

[0013] In one possible implementation, establishing an optimization objective function based on the cost matrix and the optimal path index, which uses weight parameters as decision variables and minimizes the total cost value of the optimal path, includes: Extract the vector in the cost matrix that corresponds to the optimal path index as the optimal cost vector; Create a weight parameter vector corresponding to the weight coefficients of the indicator type according to the order of the cost values ​​in the cost matrix; The optimization objective function is calculated based on the optimal cost vector and the weight parameter vector.

[0014] in, The optimization objective function is... The row sequence of the cost matrix, Let be the optimal cost vector. This is the weight parameter vector.

[0015] In one possible implementation, the step of solving the objective function based on a linear optimization algorithm to obtain the optimal combination of weight parameters that satisfies preset constraints includes: The optimization objective function and the preset constraints are transformed into standard linear programming equations, and the candidate weight parameter combinations are obtained by solving them. Calculate the total generation value corresponding to all the candidate paths based on the candidate weight parameters, and verify whether the total generation value of the optimal path is the minimum value. If so, the candidate weight parameter combination is determined to be the optimal weight parameter combination.

[0016] In one possible implementation, the preset constraints include at least one of the following: all weight parameters are non-negative; the sum of all weight parameters is 1; and the total value of the candidate paths other than the optimal path is greater than or equal to the total value of the optimal path.

[0017] Secondly, the present invention also provides an intelligent vehicle, comprising: The construction module is used to construct a cost matrix based on the cost of each candidate path under preset driving conditions; each row of the cost matrix corresponds to one candidate path, and each column corresponds to the cost of an index type. The acquisition module is used to acquire the optimal motion trajectory under the preset driving conditions, and determine the optimal path index from multiple candidate paths based on the optimal motion trajectory; A module is established to establish an optimization objective function that minimizes the total cost value of the optimal path, based on the cost matrix and the optimal path index; The processing module is used to solve the objective function based on a linear optimization algorithm to obtain the optimal combination of weight parameters that satisfies preset constraints. The optimal combination of weight parameters includes weight parameters corresponding to the index type.

[0018] Thirdly, the present invention also provides an electronic device, including a memory and a processor, wherein, The memory is used to store programs; The processor, coupled to the memory, is used to execute the program stored in the memory to implement the steps in the linear optimization-based weight parameter determination method described in any of the above implementations.

[0019] Fourthly, the present invention also provides a computer-readable storage medium for storing a computer-readable program or instructions, which, when executed by a processor, can implement the steps in the linear optimization-based weight parameter determination method described in any of the above implementations.

[0020] The beneficial effects of this invention are as follows: The linear optimization-based weight parameter determination method provided by this invention can generate multiple candidate paths under preset driving conditions in local path planning for autonomous driving. Each candidate path has multiple indicator types (such as lateral deviation, curve length, curve curvature, distance to obstacles, etc.), and the cost values ​​corresponding to these indicator types are organized into a cost matrix. Under preset driving conditions, the optimal trajectory is obtained through real-vehicle manual driving data or other methods. An optimization objective function is established based on the cost matrix and the optimal path index, which quantifies the optimization objective as minimizing the total cost value of the optimal path, making the optimization process more explicit and operable. By adjusting the weight parameters, the optimal combination of weight parameters can be found, making the evaluation cost function more reasonably reflect the importance of different indicators. Furthermore, by solving the above-mentioned objective function using a linear optimization algorithm, pre-defined constraints need to be considered during the solution process. In this way, the objective function and pre-defined constraints are transformed into the standard form of linear programming, which enables efficient solution of the optimization problem and quick finding of the optimal combination of weight parameters, greatly saving time and effort. Moreover, the optimal combination of weight parameters obtained based on the linear optimization algorithm can better reflect the importance of different index types, making path planning more scientific and reasonable, and improving the accuracy and efficiency of path planning. Attached Figure Description To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 A schematic flowchart of an embodiment of the weight parameter determination method based on linear optimization provided by the present invention; Figure 2 For the present invention Figure 1 A schematic diagram of an embodiment of S100; Figure 3 For the present invention Figure 2 A schematic diagram of an embodiment of S200; Figure 4 For the present invention Figure 3 A schematic diagram of an embodiment of S300; Figure 5 For the present invention Figure 3 A schematic diagram of an embodiment of the S400; Figure 6 A schematic diagram of an embodiment of the intelligent vehicle provided by the present invention; Figure 7 A schematic diagram of an embodiment of the electronic device provided by the present invention. Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0023] In the description of the embodiments of the present invention, unless otherwise stated, "multiple" means two or more. "And / or" describes the relationship between related objects, indicating that there can be three relationships. For example, A and / or B can represent three situations: A exists alone, A and B exist simultaneously, and B exists alone.

[0024] The terms "first," "second," etc., used in the embodiments of this invention are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, a technical feature defined with "first" or "second" may explicitly or implicitly include at least one of that feature.

[0025] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the invention. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0026] Before demonstrating the embodiments, the following terms will be explained.

[0027] A linear programming solver is a tool or software used to solve linear programming problems. Linear programming is a mathematical optimization method used to maximize or minimize a linear objective function under a set of linear constraints. The objective function is a linear function that includes multiple decision variables. The constraints are linear inequalities or equality conditions that limit the range of values ​​for the decision variables. The main function of a linear programming solver is to find the values ​​of the decision variables that satisfy all constraints, thus maximizing the objective function. The solver achieves this through algorithms, including the Simplex Method, the Interior Point Method, and the Branch and Bound method.

[0028] This invention provides a method for determining weight parameters based on linear optimization, as well as a vehicle, equipment, and medium, which are described below.

[0029] Figure 1 The following is a schematic flowchart of an embodiment of the weight parameter determination method based on linear optimization provided by the present invention, as shown below. Figure 1 As shown, the method for determining weight parameters based on linear optimization includes: S100. Construct a cost matrix based on the cost of each candidate path under preset driving conditions; each row of the cost matrix corresponds to one candidate path, and each column corresponds to the cost of an index type.

[0030] It should be noted that: Pre-set driving conditions can be established, and local path planning can be performed based on these conditions to obtain n candidate paths. Then, the cost value of each candidate path for m indicator types is calculated, where m and n are positive integers, and m and n ≥ 2. Thus, an n×m cost matrix is ​​constructed based on the cost values ​​of all candidate paths under the pre-set driving conditions. Each row of the cost matrix corresponds to a candidate path, and each column corresponds to the cost value of an indicator type. For example, for n candidate paths and m indicator types, the size of the cost matrix C is n×m, where Cij represents the cost value of the i-th candidate path for the j-th indicator type. This matrix approach organizes complex path evaluation data, making the data more structured and facilitating subsequent processing and analysis. The matrix format allows for a direct comparison of the advantages and disadvantages of different candidate paths across various indicator types, providing a clear data foundation for subsequent optimization.

[0031] S200: Obtain the optimal motion trajectory under the preset driving conditions, and determine the optimal path index from multiple candidate paths based on the optimal motion trajectory.

[0032] It should be noted that under preset driving conditions, the optimal trajectory is obtained through real-vehicle manual driving data or other methods. The optimal trajectory is the most ideal driving path under the specified conditions, as determined by the operator. In high-fidelity simulation environments such as CARLA and Prescan, experienced engineers ("experts") can manually control a virtual vehicle to complete what is considered the safest, most comfortable, and most efficient lane-changing maneuver under set lane-changing and overtaking conditions. The vehicle's time-state sequence (such as position, speed, acceleration, and heading angle) during this operation is recorded as the optimal trajectory. Alternatively, in a closed test track or on a specific road section, a professional test driver can drive a real vehicle equipped with high-precision positioning (such as RTK-GPS) and an inertial measurement unit (IMU) to complete the lane-changing and overtaking maneuver. The collected data, after cleaning and preprocessing, is used as the optimal trajectory. Then, the obtained optimal trajectory is matched with multiple candidate paths, and the candidate path closest to the optimal trajectory is selected to obtain the optimal path. The optimal path index is obtained based on the index of this optimal path in the cost matrix. Determining the optimal path index based on real vehicle data or other reliable data makes the optimization process more closely aligned with actual application needs, thereby improving the reliability and practicality of the optimization results.

[0033] S300. Based on the cost matrix and the optimal path index, establish an optimization objective function that uses weight parameters as decision variables and minimizes the total cost value of the optimal path.

[0034] It should be noted that: Let the cost matrix be C, the weight parameter vector be w, and the optimal path index be k. Then the total cost of the optimal path can be expressed as Ck·w, where Ck is the vector of the optimal path index in the k-th row of the cost matrix. Based on the cost matrix and the optimal path index, an optimization objective function can be established. This objective function quantifies the optimization objective as minimizing the total cost of the optimal path, making the optimization process more explicit and operable. Subsequently, the optimal combination of weight parameters can be found by adjusting the weight parameters.

[0035] S400. Based on the linear optimization algorithm, the optimization objective function is solved to obtain the optimal weight parameter combination that satisfies the preset constraints. The optimal weight parameter combination includes weight parameters corresponding to the index type.

[0036] It should be noted that linear optimization algorithms (such as the simplex method or interior point method) can be used to solve the above-mentioned objective function. Pre-defined constraints need to be considered during the solution process to obtain the optimal combination of weight parameters. This allows for the rapid and accurate determination of the weight parameters of the evaluation cost function in local path planning, avoiding the tedious process of manually and repeatedly determining parameters in traditional methods. This significantly saves time and effort, improves the efficiency and quality of path planning, and provides stronger technical support for local path planning of autonomous vehicles.

[0037] In summary, in local path planning for autonomous driving, multiple candidate paths can be generated under preset driving conditions. Each candidate path has multiple indicator types (such as lateral deviation, curve length, curve curvature, and distance to obstacles). The costs corresponding to these indicator types are organized into a cost matrix. Under preset driving conditions, the optimal trajectory is obtained through real-vehicle manual driving data or other methods. An optimization objective function is established based on the cost matrix and the optimal path index. The optimization objective can be quantified as minimizing the total cost of the optimal path, making the optimization process more explicit and operable. By adjusting the weight parameters, the optimal combination of weight parameters can be found, making the evaluation cost function more reasonably reflect the importance of different indicators. Furthermore, the above optimization objective function is solved using a linear optimization algorithm. During the solution process, preset constraints need to be considered. Thus, the optimization objective function and preset constraints are transformed into the standard form of linear programming, enabling efficient solution of the optimization problem and rapid finding of the optimal weight parameter combination, greatly saving time and effort. Moreover, the optimal weight parameter combination obtained based on the linear optimization algorithm can better reflect the importance of different indicator types, making path planning more scientific and reasonable, and improving the accuracy and efficiency of path planning. In some embodiments of the present invention, such as Figure 2 As shown, step S100 includes: S110, Set the path generation parameters and the preset driving conditions.

[0038] It should be noted that the path generation parameters include path geometry parameters, vehicle dynamics constraints, and obstacle safety parameters. Path geometry parameters include lateral offset direction (controlling the left and right boundaries of the path within the lane), longitudinal prediction distance (the length the path extends forward), lateral sampling interval (determining path density; smaller values ​​indicate more paths), and number of path nodes (determining the number of control points, affecting path smoothness). Vehicle dynamics constraints include maximum permissible curvature (ensuring the path curvature does not exceed the vehicle's minimum turning radius), maximum lateral acceleration (ensuring ride comfort), and maximum rate of change of heading angle (limiting steering wheel speed). Obstacle safety parameters include obstacle expansion layer radius, obstacle cost attenuation coefficient, and minimum safe distance threshold. Preset driving conditions include straight driving, curves, obstacle avoidance, and lane changing, categorized into urban road scenarios, highway scenarios, rural road scenarios, severe weather scenarios, special scenarios (construction areas, accident scenes, emergency avoidance scenarios, etc.), nighttime driving scenarios, and complex terrain scenarios.

[0039] S120. Generate a path cluster based on the path generation parameters and the preset driving conditions; the path cluster includes n candidate paths, where n≥2 and n is a positive integer.

[0040] It should be noted that when the path generation parameters and preset driving conditions are input into the path generator, the path generator can use the B-spline generation algorithm to perform local path planning on the input information and generate a path cluster containing n candidate paths.

[0041] S130. Calculate the cost of all the indicator types for each candidate path.

[0042] S140. Construct the cost matrix based on the cost value of the index type corresponding to each candidate path.

[0043] In some embodiments of the present invention, the index types include at least lateral deviation value, curve length value, curve curvature value, and obstacle distance value.

[0044] It should be noted that the following are assumed to be the indicator types: lateral deviation, curve length, curve curvature, and obstacle distance. For each candidate path, multiple discrete points of equal arc length are taken, with sampling intervals typically between 0.1 and 0.5 meters. The nearest neighbor to the current discrete point on the global reference line (e.g., lane centerline) is found. The Euclidean distance between the current discrete point and its corresponding neighbor is calculated. The Euclidean distances of all discrete points are summed, and the result is divided by the total number of discrete points on the candidate path to obtain the lateral deviation. For each candidate path, its cost for each indicator type is calculated. Specifically, the candidate path is discretized into an equidistant point column with a fixed step size. This column includes multiple discrete points with equal intervals. The distances between any two adjacent discrete points are accumulated to obtain the curve length of the candidate path. The curvature of each discrete point on the candidate path is calculated, and the curvatures of all discrete points on the candidate path are summed and the arithmetic mean is taken to obtain the curve curvature of the candidate path. Calculate the distance from each discrete point on the candidate path to the nearest obstacle. Obstacle distances are then calculated based on the distances from all discrete points on the candidate path to the nearest obstacle. An obstacle distance can be the minimum of all distances from the discrete points on the candidate path to the nearest obstacle, or it can be the arithmetic mean of all distances. After calculating the cost value for each candidate path for each indicator type, the cost matrix is ​​constructed based on the cost value for each indicator type corresponding to each candidate path.

[0045] This application ensures a repeatable and consistent testing environment by pre-setting driving conditions and path generation parameters, while providing flexible testing scenarios and a solid foundation for evaluating path planning algorithms. The path clusters generated by the pre-set driving conditions and path generation parameters offer a variety of possible driving paths, increasing the diversity of path planning and adapting to different driving scenarios, such as curves, obstacle avoidance, and lane changes. Furthermore, by calculating the cost value of each candidate path, the merits of the paths are quantified into specific numerical values, making the comparison of candidate paths more objective and scientific. Data-driven evaluation also improves the reliability and practicality of the algorithm. Furthermore, the cost value is structured into a cost matrix, facilitating subsequent processing and analysis, and providing input for linear optimization algorithms, making the optimization process more efficient and accurate. Thus, a comprehensive, scientific, and efficient method is provided for optimizing autonomous driving path planning algorithms, ensuring the reliability and safety of the algorithm in various complex driving scenarios.

[0046] In some embodiments of the present invention, such as Figure 3 As shown, step S200 includes: S210. Obtain the real trajectory data generated by the driving expert driving the vehicle under the preset driving conditions, and use the real trajectory data as the optimal motion trajectory.

[0047] It should be noted that: under preset driving conditions (such as curves, obstacle avoidance, lane changes, etc.), real trajectory data of actual vehicle driving is collected. This real trajectory data includes vehicle status, environmental information, and timestamps. Vehicle status includes position coordinates (x, y), heading angle (θ), speed (v), acceleration (a), etc. Environmental information includes lane lines, obstacle positions, etc. Using this real trajectory data from actual vehicle driving, the optimal trajectory at each sampling time t under the preset driving conditions is determined.

[0048] S220. The optimal motion trajectory is matched with all the candidate paths for similarity, and the index corresponding to the candidate path with the highest similarity is determined as the index of the optimal path.

[0049] It should be noted that: the optimal path that is closest to the optimal trajectory among the candidate paths is identified and its index is marked as the optimal path index. Common trajectory similarity metrics used in similarity matching include endpoint location matching, overall trajectory similarity, and keypoint matching. Endpoint location matching compares the Euclidean distance between the endpoints of candidate paths and the optimal trajectory. Overall trajectory similarity calculates the similarity between the optimal trajectory and each candidate path (e.g., DTW dynamic time warping, Hausdorff distance, Fréchet distance, etc.). Keypoint matching involves taking multiple keypoints on the optimal trajectory (e.g., one point every 5 meters) and calculating the deviation between these keypoints and the candidate paths.

[0050] Assume the optimal trajectory is sampled over a time interval to obtain a set of keypoints. Each candidate path also samples a set of keypoints, with the same number of points as the keypoints on the optimal trajectory, aligned temporally or spatially. Based on the keypoint sets and the sampled point sets of a candidate path, the root mean square error (RMSE) between the optimal trajectory and the candidate path is calculated. The candidate path with the smallest RMSE is selected as the optimal path. For more accurate matching, multiple indicators (such as position, heading angle, curvature, etc.) can be considered simultaneously. Since the optimal trajectory is a continuous time series, while the candidate path is a spatial path, the optimal trajectory and candidate path are time-space aligned before similarity matching. This can be achieved by sampling the candidate path at the same longitudinal distance or over time to align it with the optimal trajectory.

[0051] This application provides real-world trajectory data that reflects the behavioral patterns of driving experts in actual driving. This data, based on real driving experience, reflects the optimal path selection in real-world driving. Using real-world trajectory data from driving experts as the optimal trajectory ensures that the optimization process is based on reliable and realistic data, improving the credibility of the optimization results. Furthermore, the preset driving conditions cover various driving scenarios, such as curves, obstacle avoidance, and lane changes. By obtaining real-world trajectory data from driving experts under these preset driving conditions as the optimal trajectory, the optimal trajectory can be adjusted according to different conditions. This allows the algorithm to dynamically adapt to environmental changes and different driving scenarios, enhancing its adaptability. Furthermore, by matching the optimal trajectory with all candidate paths based on similarity, the most similar candidate path can be found. The merits of the paths are quantified into specific numerical values, making the path comparison more objective and scientific. Automatically matching the path with the highest similarity reduces human intervention and improves the automation level of path planning. In addition, using real-world trajectory data as the optimal trajectory ensures that the path planning algorithm can generate paths similar to the expert's trajectory, improving the accuracy and reliability of path planning.

[0052] In some embodiments of the present invention, such as Figure 4 As shown, step S300 includes: S310. Extract the vector in the cost matrix that corresponds to the optimal path index as the optimal cost vector.

[0053] It should be noted that: after obtaining the optimal path index through S220, the cost matrix is ​​then adjusted according to the optimal path index. k The row vector corresponding to the optimal path index is obtained, and this row vector is transposed into a column vector to obtain the optimal cost vector.

[0054] S320. Create a weight parameter vector corresponding to the weight coefficients of the index type according to the arrangement order of the cost values ​​in the cost matrix.

[0055] It should be noted that the weight parameter vector is a row vector. The order of the indicator types is fixed when constructing the cost matrix M. Therefore, the created weight parameter vector must strictly follow this order to arrange the weight parameters corresponding to the indicator types. In other words, the order of the weight parameters corresponding to the indicator types in the row direction of the weight parameter vector is aligned with the column order of the indicators. For example, when the indicator types in multiple columns in the row direction of the cost matrix are lateral deviation value, curve length value, curve curvature value, and obstacle distance value, the generated weight parameter vector = […]. w 1, w 2, w 3, w 4], w1, w 2, w 3, w 4 represents the weighting coefficients corresponding to the lateral deviation value, curve length value, curve curvature value, and obstacle distance value, respectively.

[0056] S330. The optimization objective function is calculated based on the optimal cost vector and the weight parameter vector:

[0057] in, The optimization objective function is... The row sequence of the cost matrix, Let be the optimal cost vector. This is the weight parameter vector.

[0058] It should be noted that: extracting an optimal cost value vector from the cost matrix, retaining only the row of data most relevant to the optimal trajectory as the optimal cost value vector, reduces the dimensionality of the n×m two-dimensional table to a 1×m one-dimensional vector, eliminates redundant information, and avoids irrelevant path noise interfering with subsequent optimization.

[0059] The subjective fact of the optimal trajectory is transformed into a mathematical objective: finding an optimal combination of weight parameters such that the total cost of the target path most similar to the optimal trajectory is less than the total cost of any other path. This eliminates the need for repeated manual weight adjustments; instead, by solving a mathematical problem, an objective combination of weight parameters consistent with expert judgment is derived. This significantly reduces reliance on human experience and improves the system's automation and consistency. The optimized objective function is calculated based on the optimal cost vector and the weight parameter vector. While ensuring real-time engineering performance, expert experience is efficiently transformed into interpretable, reproducible, and scalable optimal weights. The weight adjustment problem, which relies on subjective human experience, is successfully transformed into a linear programming problem with convexity, efficient solvability, interpretable solutions, and guaranteed decision advantage. This provides a powerful, reliable, and practical framework for autonomous driving systems to automatically learn expected decision preferences in different scenarios.

[0060] In some embodiments of the present invention, such as Figure 5 As shown, step S400 includes: S410. The optimization objective function and the preset constraints are transformed into the standard equation of linear programming, and the candidate weight parameter combination is obtained by solving the equation.

[0061] In some embodiments of the present invention, the preset constraints include at least one of the following: all weight parameters are non-negative; the sum of all weight parameters is 1; and the total value of the candidate paths other than the optimal path is greater than or equal to the total value of the optimal path.

[0062] It should be noted that the above optimization objective function and preset constraints are transformed into the standard linear programming form: min c T × w st A × w ≤ b , Aeq × w = beq Where min means to minimize. c It is the coefficient vector (i.e. the optimal cost vector) of the objective function, and its dimension must be consistent with the weight parameter vector. w It is the vector of decision variables to be optimized (i.e., the vector of weight parameters). T This indicates transpose. Here, `st` means "subject to" (satisfied with). A It is the coefficient matrix of the inequality constraints (with known constants). b st is the constant vector (a known constant) on the right-hand side of the inequality constraint. A × w ≤ b express w All inequality constraints must be satisfied (e.g., X × w < 800000). Aeq It is the coefficient matrix of the equality constraints (with known constants). beq It is the constant vector on the right side of the equality constraint (a known constant). Aeq × w = beq express w All equality constraints must be satisfied.

[0063] For example, each element stores a cluster of curve data at a sampling time, i.e., the index type includes: Colision_Risk_Value_All_B_splines1: Collision risk values ​​(n×1 vectors) for n curves, i.e. obstacle distance values.

[0064] Lat_All_B_splines1: Lateral deviation values ​​(n×1 vector).

[0065] Curv_all_B_splines1: Curve curvature values ​​(n×1 vector).

[0066] L_all_B_splines1: Curve length values ​​(n×1 vector).

[0067] optimalIndex: The index of the best curve labeled at the current time (either manually labeled or based on the true value, i.e., the index of the best path).

[0068] Set the initial weight values ​​to [1, 1, 1, 1]. Let the optimal cost vector X = [Colision_Risk_Value_All_B_splines1; Lat_All_B_splines1; Curv_all_B_splines1; L_all_B_splines1], and let the label values ​​be: X1=Colision_Risk_Value_All_B_splines1[optimalIndex], X2=Lat_All_B_splines1[optimalIndex], X3=Curv_all_B_splines1[optimalIndex], X4=L_all_B_splines1[optimalIndex].

[0069] The created optimization objective function is: Cost = X1×w1 + X2×w2 + X3×w3 + X4×w4. The optimization objective is to adjust the weight parameters w1, w2, w3, and w4 at each sampling time to minimize the total cost of the labeled optimal path index. The predefined constraints include: The weights are defined as non-negative: lb = 0, ub = 1, where lb is short for lower bound and ub is short for upper bound, indicating that the weight parameters cannot be lower than 0 and cannot exceed 1. st A×w≤800000, and w1+w2+w3+w4=1. The candidate weight parameters are calculated using an efficient LP solver, such as the simplex method or the interior-point method.

[0070] Among these, different cost metrics (such as collision risk vs. curve length) can vary by orders of magnitude (e.g., collision risk ranges from 0 to 100,000, and curve length ranges from 0 to 100). If the weights are not properly combined (e.g., w1 is extremely large), the total cost of some paths may reach unreasonably high values ​​(e.g., exceeding 10). 6This constraint, X×w < 800000, prevents the cost function value from exploding. The constraint forces the weight w to be relatively small (especially for cost metrics with large orders of magnitude), avoiding extreme weight combinations and limiting the feasible domain of the weight parameters. Furthermore, to ensure the rationality of path selection, the total cost of all candidate paths is limited to a certain range (e.g., X×w < 800000), ensuring that the optimization result does not select paths with extremely high individual costs (such as paths with extremely high collision risk).

[0071] S420. Calculate the total generation value corresponding to all the candidate paths according to the candidate weight parameters, and verify whether the total generation value of the optimal path is the minimum value. S430. If so, determine the candidate weight parameter combination as the optimal weight parameter combination.

[0072] It should be noted that for each sampling time t, the total cost of the optimal trajectory should be minimized (ideally less than the cost of all other trajectories). Substituting the candidate weight parameters and the cost value of each candidate path for different cost types into the above formula, the total cost value of each candidate path is calculated. The total cost value of all candidate paths is compared with the total cost value of the optimal path. If the total cost value of the optimal path is the minimum, then the candidate weight parameter combination is determined to be the optimal weight parameter combination. If the total cost value of the optimal path is not the minimum, the calculation of the cost matrix can be checked for correctness, or the labeling of the optimal path index k can be checked for reasonableness, or the quality of the generated candidate paths may be low, resulting in no path matching the optimal motion trajectory well. Adjustments can then be made based on the detection results.

[0073] This application automates, optimizes, and optimizes the solution process by transforming the objective function and preset constraints into standard linear programming equations and solving for candidate weight parameter combinations, ensuring the quality of the solution. Through independent verification loops (S430-S440), the results are endowed with extremely high reliability and robustness, meeting the stringent safety and reliability requirements of autonomous driving systems. The overall process design is rigorous and modular, possessing good repeatability, debuggability, and scalability, representing a complete parameter optimization methodology suitable for industrial-grade deployment.

[0074] To better implement the weight parameter determination method based on linear optimization in the embodiments of the present invention, based on the weight parameter determination method based on linear optimization, the corresponding method is as follows: Figure 6 As shown, this embodiment of the invention also provides an intelligent vehicle, the intelligent vehicle 600 comprising: The construction module 601 is used to construct a cost matrix based on the cost of each candidate path under preset driving conditions; each row of the cost matrix corresponds to one candidate path, and each column corresponds to the cost of an index type. The acquisition module 602 is used to acquire the optimal motion trajectory under the preset driving conditions, and determine the optimal path index from multiple candidate paths based on the optimal motion trajectory; Module 603 is used to establish an optimization objective function based on the cost matrix and the optimal path index, with weight parameters as decision variables and minimizing the total cost value of the optimal path. The processing module 604 is used to solve the optimization objective function based on a linear optimization algorithm to obtain the optimal weight parameter combination that satisfies the preset constraints. The optimal weight parameter combination includes weight parameters corresponding to the index type.

[0075] The intelligent vehicle 600 provided in the above embodiments can implement the technical solutions described in the above embodiments of the method for determining weight parameters based on linear optimization. The specific implementation principles of each module or unit can be found in the corresponding content in the above embodiments of the method for determining weight parameters based on linear optimization, and will not be repeated here.

[0076] like Figure 7 As shown, the present invention also provides an electronic device 700. The electronic device 700 includes a processor 701, a memory 702, and a display 703. Figure 7 Only some components of the electronic device 700 are shown, but it should be understood that it is not required to implement all the components shown, and more or fewer components may be implemented instead.

[0077] In some embodiments, processor 701 may be a central processing unit (CPU), microprocessor, or other data processing chip, used to run program code stored in memory 702 or process data, such as the linear optimization-based weight parameter determination method of the present invention.

[0078] In some embodiments, processor 701 may be a single server or a group of servers. The server group may be centralized or distributed. In some embodiments, processor 701 may be local or remote. In some embodiments, processor 701 may be implemented on a cloud platform. In one embodiment, the cloud platform may include a private cloud, public cloud, hybrid cloud, community cloud, distributed cloud, intranet, multi-cloud, etc., or any combination thereof.

[0079] In some embodiments, memory 702 may be an internal storage unit of electronic device 700, such as a hard disk or memory of electronic device 700. In other embodiments, memory 702 may also be an external storage device of electronic device 700, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc. equipped on electronic device 700.

[0080] Furthermore, the memory 702 may include both internal storage units of the electronic device 700 and external storage devices. The memory 702 is used to store application software and various types of data installed on the electronic device 700.

[0081] In some embodiments, display 703 may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, or an OLED (Organic Light-Emitting Diode) touchscreen. Display 703 is used to display information from electronic device 700 and to display a visual user interface. Components 701-703 of electronic device 700 communicate with each other via a system bus.

[0082] In one embodiment, when processor 701 executes a linear optimization-based weight parameter determination program in memory 702, the following steps can be implemented: Obtain the first linear optimization-based weight parameters to determine the model based on the first nuclide; Based on the deep transfer learning algorithm, a second model based on linear optimization weight parameters is constructed according to the first model determined by linear optimization weight parameters. A magnetic resonance image to be optimized based on a second nuclide is acquired, and a model is determined to optimize the magnetic resonance image based on a second linear optimization weight parameter. Wherein, the concentration of the first nuclide in the organism is greater than the concentration of the second nuclide in the organism, and / or, the transverse relaxation time of the first nuclide is greater than the transverse relaxation time of the second nuclide, and / or, the gyromagnetic ratio of the first nuclide is greater than the gyromagnetic ratio of the second nuclide.

[0083] It should be understood that when the processor 701 executes the linear optimization-based weight parameter determination program in the memory 702, in addition to the functions mentioned above, it can also perform other functions, as detailed in the description of the corresponding method embodiments above.

[0084] Furthermore, this embodiment of the invention does not specifically limit the type of electronic device 700 mentioned. Electronic device 700 can be a mobile phone, tablet computer, personal digital assistant (PDA), wearable device, laptop computer, or other portable electronic device. Exemplary embodiments of portable electronic devices include, but are not limited to, portable electronic devices running iOS, Android, Microsoft, or other operating systems. The aforementioned portable electronic device can also be other portable electronic devices, such as a laptop computer with a touch-sensitive surface (e.g., a touch panel). It should also be understood that in some other embodiments of the invention, electronic device 700 may not be a portable electronic device, but rather a desktop computer with a touch-sensitive surface (e.g., a touch panel).

[0085] Accordingly, this application also provides a computer-readable storage medium for storing a computer-readable program or instruction. When the program or instruction is executed by a processor, it can implement the steps or functions of the linear optimization-based weight parameter determination method provided in the above-described method embodiments.

[0086] Those skilled in the art will understand that all or part of the processes of the methods described in the above embodiments can be implemented by a computer program instructing related hardware (such as a processor, controller, etc.), and the computer program can be stored in a computer-readable storage medium. The computer-readable storage medium may be a disk, optical disk, read-only memory, or random access memory, etc.

[0087] The foregoing has provided a detailed description of the linear optimization-based weight parameter determination method, vehicle, equipment, and medium provided by this invention. Specific examples have been used to illustrate the principles and implementation methods of this invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of this invention. At the same time, those skilled in the art will recognize that, based on the ideas of this invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this invention.

Claims

1. A method for determining weight parameters based on linear optimization, characterized in that, include: Construct a cost matrix based on the cost of each candidate path under preset driving conditions; Each row of the cost matrix corresponds to a candidate path, and each column corresponds to the cost of an indicator type. Obtain the optimal motion trajectory under the preset driving conditions, and determine the optimal path index from multiple candidate paths based on the optimal motion trajectory; Based on the cost matrix and the optimal path index, an optimization objective function is established that uses weight parameters as decision variables and minimizes the total cost value of the optimal path. Based on a linear optimization algorithm, the optimization objective function is solved to obtain the optimal combination of weight parameters that satisfies the preset constraints. The optimal combination of weight parameters includes weight parameters corresponding to the index type.

2. The method for determining weight parameters based on linear optimization according to claim 1, characterized in that, The construction of the cost matrix based on the cost of each candidate path under preset driving conditions includes: Set the path generation parameters and the preset driving conditions; A path cluster is generated based on the path generation parameters and the preset driving conditions; the path cluster includes n candidate paths, where n≥2 and n is a positive integer. Calculate the cost of all the indicator types for each of the candidate paths; The cost matrix is ​​constructed based on the cost value of the corresponding indicator type for each candidate path.

3. The method for determining weight parameters based on linear optimization according to claim 2, characterized in that, The index type includes at least one of the following: lateral deviation value, curve length value, curve curvature value, and obstacle distance value.

4. The method for determining weight parameters based on linear optimization according to any one of claims 1 to 3, characterized in that, The step of obtaining the optimal trajectory under the preset driving conditions and determining the optimal path index from multiple candidate paths based on the optimal trajectory includes: Acquire real trajectory data generated by a driving expert driving the vehicle under the preset driving conditions, and use the real trajectory data as the optimal motion trajectory; The optimal trajectory is matched with all the candidate paths for similarity, and the index corresponding to the candidate path with the highest similarity is determined as the index of the optimal path.

5. The method for determining weight parameters based on linear optimization according to claim 4, characterized in that, The step of establishing an optimization objective function based on the cost matrix and the optimal path index, with weight parameters as decision variables and minimizing the total cost value of the optimal path, includes: Extract the vector in the cost matrix that corresponds to the optimal path index as the optimal cost vector; Create a weight parameter vector corresponding to the weight coefficients of the indicator type according to the order of the cost values ​​in the cost matrix; The optimization objective function is calculated based on the optimal cost vector and the weight parameter vector. in, The optimization objective function is... The row sequence of the cost matrix, Let be the optimal cost vector. This is the weight parameter vector.

6. The method for determining weight parameters based on linear optimization according to claim 5, characterized in that, The step of solving the objective function based on a linear optimization algorithm to obtain the optimal combination of weight parameters that satisfies preset constraints includes: The optimization objective function and the preset constraints are transformed into standard linear programming equations, and the candidate weight parameter combinations are obtained by solving them. Calculate the total generation value corresponding to all the candidate paths based on the candidate weight parameters, and verify whether the total generation value of the optimal path is the minimum value. If so, the candidate weight parameter combination is determined to be the optimal weight parameter combination.

7. The method for determining weight parameters based on linear optimization according to claim 1, characterized in that, The preset constraints include at least one of the following: all weight parameters are non-negative; the sum of all weight parameters is 1; and the total value of the candidate paths other than the optimal path is greater than or equal to the total value of the optimal path.

8. An intelligent vehicle, characterized in that, include: The construction module is used to build a cost matrix based on the cost of each candidate path under preset driving conditions. Each row of the cost matrix corresponds to a candidate path, and each column corresponds to the cost of an indicator type. The acquisition module is used to acquire the optimal motion trajectory under the preset driving conditions, and determine the optimal path index from multiple candidate paths based on the optimal motion trajectory; A module is established to establish an optimization objective function that minimizes the total cost value of the optimal path, based on the cost matrix and the optimal path index; The processing module is used to solve the objective function based on a linear optimization algorithm to obtain the optimal combination of weight parameters that satisfies preset constraints. The optimal combination of weight parameters includes weight parameters corresponding to the index type.

9. An electronic device, characterized in that, Including memory and processor, among which, The memory is used to store programs; The processor, coupled to the memory, is used to execute the program stored in the memory to implement the steps in the linear optimization-based weight parameter determination method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, Used to store computer-readable programs or instructions, which, when executed by a processor, can implement the steps in the linear optimization-based weight parameter determination method described in any one of claims 1 to 7.

Images