Automobile control method and device based on combinatorial search
By acquiring user voice data for intent parsing and combined search, cross-domain control strategies are generated, solving the problem that existing in-vehicle intelligent systems cannot comprehensively consider user intent and vehicle status, thus achieving deeper and more accurate intelligent vehicle control.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- DONGFENG LIUZHOU MOTOR
- Filing Date
- 2026-01-21
- Publication Date
- 2026-05-05
AI Technical Summary
Existing in-vehicle intelligent systems cannot comprehensively consider the vehicle's real-time status, external environment, and underlying atomic action combinations when understanding users' complex and multi-layered interactive intentions, resulting in a lack of depth and accuracy in intelligent services.
By acquiring user voice data for intent parsing, a formal task description is generated. A heuristic graph search algorithm is used to perform a combined search in a pre-set atomic action library. A multi-objective evaluation algorithm is combined to generate and select cross-domain alternative control strategies, and the model is optimized based on user feedback.
It enhances the depth and accuracy of automotive intelligent control services, enabling the generation of multi-dimensional optimized control strategies that meet user needs and improve the system's intelligence level.
Smart Images
Figure CN121979045A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of vehicle intelligent control technology, and in particular to a vehicle control method and device based on combinatorial search. Background Technology
[0002] With the continuous improvement of vehicle intelligence and connectivity, in-vehicle intelligent operating systems have become a core component of modern automobiles. Existing in-vehicle intelligent systems, such as those developed based on Android Automotive, AliOS, and QNX platforms, primarily focus on infotainment, navigation, and voice control. However, these existing technologies have significant shortcomings in addressing complex and multi-layered user interaction intentions and providing intelligent, proactive, and comprehensive solutions.
[0003] Existing technologies can only provide a single, linear response after understanding user intent. For example, for a user intent to optimize energy consumption strategies, existing systems may only provide suggestions such as "switch to economy driving mode" or display the nearest charging station. They cannot comprehensively consider the vehicle's real-time status, the external environment, and the combination of underlying atomic actions that the vehicle can perform (such as adjusting the output power curve of the drive motor, coordinating the duty cycle of the air conditioning compressor, adjusting the intensity of regenerative braking, and pre-planning kinetic energy recovery strategies that include specific slope sections). This results in a lack of depth and accuracy in intelligent automotive services. Summary of the Invention
[0004] This application provides a vehicle control method and apparatus based on combinatorial search, which can generate feasible, cross-domain alternative control strategies in a preset library of atomic actions that can be executed by the vehicle and a multi-dimensional environmental state space.
[0005] In a first aspect, embodiments of this application provide a vehicle control method based on combinatorial search, comprising: Acquire user voice data and perform intent parsing to obtain optimization goals and constraints; Generate a formal task description based on the optimization objective and constraints; Acquire user profiles, vehicle environment data, and vehicle status data, and construct a state space vector; Using a heuristic graph search algorithm, a combined search is performed in a pre-set atomic action library based on a formal task description and state space vector to obtain multiple alternative control strategies; A multi-objective evaluation algorithm is used to calculate the comprehensive score of each alternative control strategy; The target control strategy is determined and executed based on the comprehensive score or user selection instructions.
[0006] Furthermore, the above-mentioned acquisition of user voice data and intent parsing yields optimization objectives and constraints, including: Acquire user voice data and perform text conversion to obtain the text requested by the user; By inputting the user's request text into a natural language understanding model for the automotive field, the optimization objective and constraints are obtained.
[0007] Furthermore, vehicle environmental data includes external temperature, GPS location, navigation route planning, and information on charging stations along the way; vehicle status data includes current vehicle speed, motor torque, battery state of charge, and status information of various electrical devices.
[0008] Furthermore, the preset atomic action library includes multiple atomic actions, each of which includes an action number, action description, state variables, adjustable parameter range of the state variables, action execution time, and dependency conditions.
[0009] Furthermore, the above-mentioned heuristic graph search algorithm performs a combined search within a pre-defined atomic action library based on the formal task description and state space vectors to obtain multiple candidate control strategies, including: Using the state space vector as the root node of the search tree, a search is performed in the pre-set atomic action library based on the constraints and dependencies of atomic actions in the formal task description to obtain compatible atomic actions. Use compatible atomic actions as state expansion nodes of the search tree until the search depth reaches the preset depth, the search time exceeds the preset time limit, or the cumulative cost of the search tree satisfies the optimization objective in the formal task description. By combining the various state expansion nodes in the search tree according to the search order, alternative control strategies are obtained.
[0010] Furthermore, the above-mentioned determination and execution of the target control strategy based on comprehensive scoring or user selection instructions includes: Arrange the alternative control strategies in descending order of their comprehensive scores; The top-ranked alternative control strategy is selected as the target control strategy.
[0011] Furthermore, the method also includes: Obtain the execution results and user feedback information of the target control strategy; Optimize the natural language understanding model and multi-objective evaluation algorithm based on user feedback; The heuristic graph search algorithm is optimized based on the comparison between the execution results and the optimization objective.
[0012] Secondly, embodiments of this application provide a vehicle control device based on combined search, comprising: The intent parsing module is used to acquire user voice data and perform intent parsing to obtain optimization goals and constraints. The task formalization module is used to generate formal task descriptions based on optimization objectives and constraints. The acquisition module is used to acquire user profiles, vehicle environment data, and vehicle status data, and construct a state space vector. The combinatorial search module is used to perform a combinatorial search in a preset atomic action library based on a heuristic graph search algorithm, according to a formal task description and a state space vector, to obtain multiple alternative control strategies. The scoring module is used to calculate the comprehensive score of each alternative control strategy using a multi-objective evaluation algorithm; The execution module is used to determine and execute the target control strategy based on the comprehensive score or user selection instructions.
[0013] Thirdly, embodiments of this application provide a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it performs the steps of a vehicle control method based on combinatorial search as described in any of the above embodiments.
[0014] Fourthly, embodiments of this application provide a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of a vehicle control method based on combinatorial search as described in any of the above embodiments.
[0015] In summary, compared with the prior art, the beneficial effects of the technical solution provided in this application include at least the following: This application provides a vehicle control method based on combinatorial search. After parsing user voice data to obtain a formal task description, a state space vector is constructed from user profile, vehicle environment data, and vehicle state data. A heuristic graph search algorithm is then used to perform a combinatorial search in a preset atomic action library based on the formal task description and the state space vector to obtain multiple alternative control strategies. This application, through an innovative combinatorial search mechanism, can generate feasible, cross-domain alternative control strategies in the preset atomic action library executable by the vehicle and the multi-dimensional environmental state space. Then, a multi-objective evaluation algorithm is used to determine the target control strategy and execute it, thereby improving the depth and accuracy of vehicle intelligent control services. Attached Figure Description
[0016] Figure 1 A flowchart illustrating a vehicle control method based on combinatorial search, provided as an exemplary embodiment of this application.
[0017] Figure 2 This is a structural diagram of an automobile control device based on combinatorial search, provided as an exemplary embodiment of this application. Detailed Implementation
[0018] Next, the technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments.
[0019] All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present application without creative efforts shall fall within the scope of protection of the present application.
[0020] Please refer to Figure 1 , the embodiments of the present application provide a vehicle control method based on combinatorial search, including: Step S1, obtain user voice data and perform intent parsing to obtain an optimization target and constraint conditions.
[0021] Specifically, the above-mentioned obtaining user voice data and performing intent parsing to obtain an optimization target and constraint conditions includes: Step S11, obtain user voice data and perform text conversion to obtain user requirement text.
[0022] Step S12, input the user requirement text into a natural language understanding model in the vehicle field to obtain an optimization target and constraint conditions.
[0023] Among them, the natural language understanding model in the vehicle field can be obtained by fine-tuning the BERT and GPT architectures after trimming and optimization in the field, and then training with a large amount of corpus related to vehicle scenarios, such as instructions, vehicle function descriptions, user queries, etc.
[0024] Step S2, generate a formal task description according to the optimization target and constraint conditions.
[0025] Specifically, from the structured semantics output by the natural language understanding model, extract the optimization target (such as the lowest energy consumption, the most comfortable, the fastest arrival, etc.) and constraint conditions (such as not exceeding XX battery power, ensuring the vehicle interior temperature is between 22-24°C, etc.), and then convert the abstract optimization target and constraint conditions into a formal expression that can be processed by a computer. For example, convert the optimization of energy consumption into the objective function Minimize(Energy_Consumption), and convert but don't want to consume too much power into the constraint condition Energy_Consumption<Threshold, so as to obtain a formal task description.
[0026] Step S3, obtain user portraits, vehicle environment data and vehicle status data and construct a state space vector.
[0027] Among them, the vehicle environment data includes external temperature, GPS location, navigation planned route and information of charging stations along the way; the vehicle status data includes current vehicle speed, motor torque, state of charge of the battery and status information of each electrical device.
[0028] Step S4: Using a heuristic graph search algorithm, a combined search is performed in a preset atomic action library based on the formal task description and state space vector to obtain multiple alternative control strategies.
[0029] The preset atomic action library includes multiple atomic actions, each of which includes an action number, action description, state variables, adjustable parameter range of the state variables, action execution time, and dependency conditions.
[0030] For example, a certain atomic action in the preset atomic action library can be represented as: Action_ID: AC_Compressor_Power, Desc: Adjust the power of the air conditioner compressor, Params: {power_level: [0,10]}, Effect: {cabin_temp_change_rate: +0.1*power_level°C / min, power_draw: +200*power_level W}.
[0031] Specifically, the above-mentioned heuristic graph search algorithm performs a combined search in a pre-set atomic action library based on the formal task description and state space vector to obtain multiple candidate control strategies, including: Step S41: Using the state space vector as the root node of the search tree, search in the preset atomic action library according to the constraints in the formal task description and the dependency conditions of atomic actions to obtain compatible atomic actions.
[0032] Step S42: Use compatible atomic actions as state expansion nodes of the search tree until the search depth reaches a preset depth, the search time exceeds a preset time limit, or the cumulative cost of the search tree satisfies the optimization objective in the formal task description.
[0033] Step S43: Combine the various state expansion nodes in the search tree according to the search order to obtain alternative control strategies.
[0034] The core of this invention is the combinatorial search algorithm, which employs a heuristic graph search algorithm (such as the A* algorithm, a variant of Monte Carlo Tree Search (MCTS), etc.) to explore a vast combinatorial space composed of atomic actions. The combinatorial search algorithm uses the state space vector constructed from the current state as the root node of the search tree. Each time a node is expanded, one or more compatible atomic actions (satisfying dependency and constraint conditions) are selected and applied to the current state, generating a new state node. The cumulative cost from the root node to the latest node is calculated. When the search reaches a preset depth, times out, or the cumulative cost satisfies the objective function, the algorithm backtracks the search path, sequentially combining the atomic actions corresponding to the visited nodes to form a candidate control strategy. Typically, the combinatorial search algorithm searches multiple times and retains several superior paths, thereby generating multiple candidate control strategies.
[0035] Step S5: Use a multi-objective evaluation algorithm to calculate the comprehensive score of each alternative control strategy.
[0036] The multi-objective evaluation algorithm calculates a comprehensive score for each alternative control strategy, based on multiple dimensions, such as: (1) Target alignment: The degree to which the optimization target is achieved after the strategy is implemented (such as the estimated energy saving).
[0037] (2) Feasibility / Safety: Does the strategy comply with the physical limits of the vehicle and traffic safety regulations?
[0038] (3) Comfort impact: The negative impact of the strategy on the comfort of drivers and passengers (such as rapid acceleration and frequent start-stop).
[0039] (4) Personalization matching degree: The degree of consistency between the strategy and the user's historical preferences (obtained from user profile).
[0040] The scores from the above multiple dimensions are weighted and fused according to the current task type and the weight preferences in the user profile (for example, a user values energy saving more than absolute speed) to obtain the comprehensive score of each alternative control strategy.
[0041] Step S6: Determine and execute the target control strategy based on the comprehensive score or user selection instructions.
[0042] Specifically, the above-mentioned determination and execution of the target control strategy based on comprehensive scoring or user selection instructions includes: Step S61: Sort the candidate control strategies in descending order according to their comprehensive scores.
[0043] Step S62: Select the candidate control strategy ranked first as the target control strategy.
[0044] Specifically, after sorting and displaying the various alternative control strategies, users can actively select them (by clicking on the touchscreen or inputting "the xth one" via voice) or directly select the alternative control strategy with the highest comprehensive score as the target control strategy.
[0045] After obtaining the target control strategy, the target control strategy is translated into specific, time-sequential vehicle control commands (such as CAN message sequences) and passed to the corresponding vehicle controller for execution.
[0046] Taking the user's voice input as "Help me optimize energy consumption strategy, but don't let the car get too cold," the processing of the domain NLU model after converting the voice into text is as follows: 1. Identify domain keywords: "optimize energy consumption" (core action), "strategy" (indicating the need for a systematic solution), and "don't let the car get too cold" (constraint); 2. Perform semantic role labeling: action (optimize) - object (energy consumption strategy), condition (maintain) - attribute (in-car temperature) - relation (above) - value (lower limit of comfort threshold); 3. Link to the vehicle knowledge graph: associate "energy consumption" with the power system, air conditioning system, etc., and "in-car temperature" with air conditioning, seat heating, etc.
[0047] The above analysis results are then converted into a formal task description: Objective function: Minimize(Total_Energy_Consumption_over_Route) Hard constraint: Cabin_Temperature >= T_comfort_low (e.g., 20°C) Soft constraints / preferences: Prefer(Smooth_Driving) (read from user profile) Optimization scope: Route: Current_Location to Destination Vehicle status data is read from the vehicle's CAN bus: current speed, motor torque, battery SOC, air conditioning settings and power consumption, and the status of all electrical appliances. Environmental data is obtained from the environmental perception system: outside temperature, GPS location, planned navigation route (including gradient information), remaining mileage, and charging station information along the route. Data is read from the user profile: the user's temperature sensitivity range and historical driving style rating (preferring a stable driving style). Supplementary data is obtained from the cloud: weather forecast for the next route and real-time traffic flow prediction. A comprehensive, multi-dimensional state space vector S0 is output, for example: S0 = {soc: 65%, speed: 80km / h, cabin_temp: 22°C, ext_temp: -5°C, route_elevation_gain: +300m,...}.
[0048] Using S0 as the starting point, we choose the evaluation function f(n) = g(n) + h(n), where n represents a state node, and g(n) is the actual cost from the starting point to node n. Here, g(n) is defined as the estimated total energy consumed from S0 to state n. h(n) is the heuristic cost from node n to the optimization objective, defined here as "the minimum theoretical energy consumption of the remaining distance, which can be estimated using the remaining distance, the vehicle's basic rolling resistance, and the drag coefficient, etc.." We then select an atomic action A from the atomic action library that is executable and compatible with the current state n (the current state is S0 during the first selection). Atomic action A may include: Adjust_Cruise_Speed(speed: 75km / h) Set_AC_Mode(mode: eco, target_temp: 20°C) Preheat_Battery_On_Approach(station_id: XYZ) Suggest_Pitstop_At_Station(station_id: XYZ, charge_to: 80%) Apply vehicle dynamics and energy consumption models to simulate and calculate the new state n' generated after executing A, and update g(n). Add the new state n' to the open list, record its parent node and the action A that caused the state transition, and repeat the expansion process until a state node is found whose corresponding "remaining mileage" is 0 (reaching the destination), or the search reaches the maximum depth or time limit.
[0049] Starting from the termination node and tracing back to the starting node S0, record all actions along the path. This action chain represents an alternative control strategy, such as: reducing speed to 75 km / h, switching the air conditioning to ECO mode and setting the temperature to 20°C, and suggesting entering charging station A to recharge to 80% after 30 km. Retain the K paths with the lowest actual cost to generate K alternative control strategies.
[0050] A multi-objective evaluation algorithm is used to simulate and deduce each alternative control strategy, and multi-dimensional scores are calculated: Score_energy(Strategy_i) = Estimated percentage of energy saved Score_comfort(Strategy_i) = Estimated comfort decline index Score_feasibility(Strategy_i) = Whether the strategy is 100% feasible (Boolean value converted to score) Score_personal(Strategy_i) = Cosine similarity(strategy feature vector, user preference vector) The overall score is calculated as follows: Final_Score_i = w1*Score_energy + w2*Score_comfort + w3*Score_feasibility + w4*Score_personal, where the weights {w1, w2, w3, w4} can be set by default.
[0051] Finally, the first three strategies are presented on the vehicle's infotainment screen in the form of cards, such as: "Strategy A (Optimal): Gentle driving + energy-saving air conditioning, expected to save 12% of electricity, and keep the interior temperature at 20°C", "Strategy B: Maintain the current speed, only slightly adjust the air conditioning, expected to save 5% of electricity", and "Strategy C: Charge for 15 minutes at the next service area, and you can use the air conditioning more freely afterwards".
[0052] In the active mode authorized by the user, the alternative control strategy with the highest Final_Score will be directly selected, and a sequence of control commands will be sent to each controller in the vehicle through the actuator interface for automatic execution.
[0053] The above embodiments provide a vehicle control method based on combinatorial search. After parsing the user's voice data to obtain a formal task description, the method constructs a state space vector from the user profile, vehicle environment data, and vehicle state data. Then, using a heuristic graph search algorithm, it performs a combinatorial search in a preset atomic action library based on the formal task description and the state space vector to obtain multiple alternative control strategies. This application, through an innovative combinatorial search mechanism, can generate feasible, cross-domain alternative control strategies in the preset atomic action library executable by the vehicle and the multi-dimensional environmental state space. Then, a multi-objective evaluation algorithm is used to determine the target control strategy and execute it, thereby improving the depth and accuracy of vehicle intelligent control services.
[0054] In some embodiments, the method further includes: Step S7: Obtain the execution results and user feedback information of the target control strategy.
[0055] Step S8: Optimize the natural language understanding model and multi-objective evaluation algorithm based on user feedback.
[0056] Step S9: Optimize the heuristic graph search algorithm based on the comparison results of the execution results and the optimization objective.
[0057] Specifically, collect user selections (such as clicking on strategy B), user adjustments (such as changing the suggested temperature of strategy A from 20°C to 21°C), and actual vehicle data after strategy execution (actual trip energy consumption report).
[0058] The formal task description of this interaction, the state space vector S0, the list of alternative control strategies, user feedback information, and execution results are stored in the database as a training sample.
[0059] If a user frequently chooses a strategy that scores high on Score_comfort but slightly lower on Score_energy, the system uses reinforcement learning or online learning methods to automatically increase w2 (comfort weight) and decrease w1 (energy weight).
[0060] If the actual energy consumption is consistently found to be significantly higher than the estimated energy consumption during the search (h(n) is too optimistic), the energy consumption model parameters are corrected using actual data to make h(n) more accurate. For fuzzy constraints such as "don't let the car get too cold," when users frequently manually change the system-recommended temperature from 20°C to 21°C, this application reinforces the knowledge that "the user's individual comfort temperature limit is 21°C," and directly uses T_comfort_low = 21°C in subsequent formal transformations.
[0061] The above embodiments construct a continuous learning optimization framework. By learning from explicit user feedback (such as choices and ratings) and implicit feedback (such as the effect of whether the strategy is ultimately adopted and long-term behavioral data), the strategy generation algorithm is continuously optimized to better match users' personalized preferences, thereby improving the system's intelligence level and user satisfaction.
[0062] Please see Figure 2 Another embodiment of this application provides a vehicle control device based on combinatorial search, comprising: The intent parsing module 101 is used to acquire user voice data and perform intent parsing to obtain optimization goals and constraints.
[0063] Task formalization module 102 is used to generate a formal task description based on the optimization objective and constraints.
[0064] The acquisition module 103 is used to acquire user profiles, vehicle environment data and vehicle status data and construct a state space vector.
[0065] The combinatorial search module 104 is used to perform a combinatorial search in a preset atomic action library based on a heuristic graph search algorithm, according to the formal task description and state space vector, to obtain multiple alternative control strategies.
[0066] The scoring module 105 is used to calculate the comprehensive score of each alternative control strategy using a multi-objective evaluation algorithm.
[0067] The execution module 106 is used to determine and execute the target control strategy based on the comprehensive score or user selection instructions.
[0068] The specific limitations of the vehicle control device based on combinatorial search provided in this embodiment can be found in the embodiment of the vehicle control method based on combinatorial search described above, and will not be repeated here. Each module in the above-described vehicle control device based on combinatorial search can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each module.
[0069] This application provides a computer device that may include a processor, memory, network interface, and database connected via a system bus. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage medium. The network interface communicates with external terminals via a network connection. When the computer program is executed by the processor, it causes the processor to perform the steps of a vehicle control method based on combinatorial search as described in any of the above embodiments.
[0070] The working process, working details, and technical effects of the computer device provided in this embodiment can be found in the above embodiment of a vehicle control method based on combinatorial search, and will not be repeated here.
[0071] This application provides a computer-readable storage medium storing a computer program thereon. When executed by a processor, the computer program implements the steps of a vehicle control method based on combinatorial search as described in any of the above embodiments. The computer-readable storage medium refers to a data storage medium, which may include, but is not limited to, floppy disks, optical disks, hard disks, flash memory, USB flash drives, and / or Memory Sticks. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The working process, details, and technical effects of the computer-readable storage medium provided in this embodiment can be found in the embodiments of a vehicle control method based on combinatorial search described above, and will not be repeated here.
[0072] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and RAMbus dynamic RAM (RDRAM).
[0073] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.
[0074] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.
Claims
1. A vehicle control method based on combinatorial search, characterized in that, include: Acquire user voice data and perform intent parsing to obtain optimization goals and constraints; Generate a formal task description based on the optimization objective and the constraints; Acquire user profiles, vehicle environment data, and vehicle status data, and construct a state space vector; Using a heuristic graph search algorithm, a combined search is performed in a preset atomic action library based on the formal task description and the state space vector to obtain multiple alternative control strategies; A multi-objective evaluation algorithm is used to calculate the comprehensive score of each of the candidate control strategies; The target control strategy is determined and executed based on the comprehensive score or user selection instructions.
2. The vehicle control method based on combinatorial search according to claim 1, characterized in that, The process of acquiring user voice data and performing intent parsing to obtain optimization objectives and constraints includes: The user's voice data is acquired and converted into text to obtain the text requested by the user; The user request text is input into a natural language understanding model for the vehicle domain to obtain the optimization objective and constraints.
3. The vehicle control method based on combinatorial search according to claim 1, characterized in that, The vehicle environmental data includes external temperature, GPS location, navigation route planning, and information on charging stations along the way; the vehicle status data includes current vehicle speed, motor torque, battery state of charge, and status information of various electrical devices.
4. The vehicle control method based on combinatorial search according to claim 2, characterized in that, The preset atomic action library includes multiple atomic actions, each of which includes an action number, action description, state variable, adjustable parameter range of the state variable, action execution time, and dependency conditions.
5. The vehicle control method based on combinatorial search according to claim 4, characterized in that, The heuristic graph search algorithm is used to perform a combined search in a preset atomic action library based on the formal task description and the state space vector to obtain multiple candidate control strategies, including: Using the state space vector as the root node of the search tree, a search is performed in the preset atomic action library according to the constraints in the formal task description and the dependency conditions of the atomic actions to obtain compatible atomic actions. The compatible atomic actions are used as state expansion nodes of the search tree until the search depth reaches a preset depth, the search time exceeds a preset time limit, or the cumulative cost of the search tree satisfies the optimization objective in the formal task description. The alternative control strategies are obtained by combining the various state extension nodes in the search tree according to the search order.
6. The vehicle control method based on combinatorial search according to claim 1, characterized in that, The step of determining and executing the target control strategy based on the comprehensive score or user selection instruction includes: Arrange the candidate control strategies in descending order according to the comprehensive score; The candidate control strategy ranked first is selected as the target control strategy.
7. The vehicle control method based on combinatorial search according to claim 5, characterized in that, Also includes: Obtain the execution results and user feedback information of the target control strategy; Optimize the natural language understanding model and multi-objective evaluation algorithm based on the user feedback information; The heuristic graph search algorithm is optimized based on the comparison between the execution results and the optimization objective.
8. A vehicle control device based on combinatorial search, characterized in that, include: The intent parsing module is used to acquire user voice data and perform intent parsing to obtain optimization goals and constraints. The task formalization module is used to generate a formal task description based on the optimization objective and the constraints. The acquisition module is used to acquire user profiles, vehicle environment data, and vehicle status data, and construct a state space vector. The combined search module is used to perform a combined search in a preset atomic action library based on the formal task description and the state space vector using a heuristic graph search algorithm to obtain multiple alternative control strategies. The scoring module is used to calculate the comprehensive score of each of the candidate control strategies using a multi-objective evaluation algorithm; The execution module is used to determine and execute the target control strategy based on the comprehensive score or user selection instruction.
9. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the vehicle control method based on combinatorial search as described in any one of claims 1 to 7.
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 steps of the vehicle control method based on combinatorial search as described in any one of claims 1 to 7.