Speed control method, device and equipment based on speed limit rule and medium

Abstract

This application provides a speed control method, apparatus, device, and medium based on speed limit rules, relating to the field of vehicle technology. The method includes: acquiring multi-source information characterizing the vehicle's driving environment, and determining the current speed limit scenario state of the vehicle based on the multi-source information; determining a pre-trained decision tree corresponding to the speed limit scenario state; inputting the multi-source information into the decision tree and outputting the vehicle's target speed; and controlling the vehicle's speed according to the target speed. The method of this application effectively solves the problems of insufficient flexibility, high resource consumption, and poor practicality in speed control under complex speed limit rules in related technologies.

Description

Technical Field

[0001] This application relates to the field of vehicle technology, and in particular to a speed control method, device, equipment and medium based on speed limit rules. Background Technology

[0002] In non-congested scenarios such as highways and urban expressways, vehicles need to respond in real time to speed limit and recommended speed information from multiple sources and types, including high-precision maps, visual traffic signs, electronic eyes, temporary construction zones, tunnels, severe weather warnings, V2X dynamic information, and special area regulations. These speed limit rules are intertwined, have different priorities, and change dynamically. The accuracy of speed decisions directly affects driving compliance, traffic efficiency, and the driving experience. Therefore, advanced driver assistance systems (ADAS) need to be able to intelligently integrate these multi-source speed limit rules and output a reasonable target speed to achieve safe and effective control of the vehicle's driving status.

[0003] In related technologies, when dealing with the aforementioned complex speed-limited scenarios, methods based on static priorities or fixed rules suffer from rigid decision-making, unnecessarily reducing traffic efficiency, and the rule system expands rapidly with the scenario, making it difficult to maintain. While deep learning-based methods offer greater flexibility, they suffer from large model parameters and high inference latency, making it difficult to meet real-time requirements on low-cost automotive-grade MCUs. Therefore, speed control under complex speed-limited rules generally suffers from insufficient flexibility, high resource consumption, and poor practicality. Summary of the Invention

[0004] This application provides a speed control method, device, equipment, and medium based on speed limit rules to solve the problems of insufficient flexibility, high resource consumption, and poor practicality in vehicle speed control under complex speed limit rules in related technologies.

[0005] Firstly, this application provides a speed control method based on speed limit rules, the method comprising the following steps:

[0006] Acquire multi-source information to characterize the vehicle's driving environment, and determine the current speed limit scenario state of the vehicle based on the multi-source information;

[0007] Determine the pre-trained decision tree corresponding to the speed-limited scenario state;

[0008] Input multi-source information into the decision tree and output the vehicle's target speed;

[0009] The vehicle speed is controlled according to the target speed.

[0010] In one embodiment of this disclosure, determining the current speed limit scenario state of a vehicle based on multi-source information includes: fusing the multi-source information to generate a standardized feature vector; inputting the feature vector into a pre-trained state machine to output the current speed limit scenario state of the vehicle; wherein the speed limit scenario state includes at least one or more of steady-state cruise, electronic eye avoidance, tunnel passage, dynamic speed limit zone, severe weather impact, and composite rule zone.

[0011] In one embodiment of this disclosure, the standardized feature vector is a fixed-dimensional numerical vector extracted from multi-source information after spatiotemporal alignment and validity verification. The feature vector includes at least one of the following: the instantaneous maximum constraint speed limit representing the lowest value among all speed limit sources; a speed limit scenario conflict flag representing whether there are two or more valid speed limit rules with different values; a driver style coefficient representing the driver's preference for efficiency and comfort; and a positional correlation feature representing the distance of the vehicle from the next speed limit change point.

[0012] In one embodiment of this disclosure, the decision tree is a state-specific decision tree that corresponds one-to-one with the speed limit scenario state; wherein, each state-specific decision tree is pre-configured as a lightweight tree model generated by independently training: using driving data collected under the same speed limit scenario state as training samples and using discrete speed values ​​formed by discretizing the actual driving speed under the scenario after being processed as an integer multiple of the set speed as training labels.

[0013] In one embodiment of this disclosure, the lightweight tree model corresponding to the state-specific decision tree is trained by applying depth pruning and feature pruning constraints during training; the state-specific decision tree is stored in the storage medium with an implicit index structure.

[0014] In one embodiment of this disclosure, the multi-source information includes the instantaneous maximum constraint speed limit at the current moment; inputting the multi-source information into a decision tree and outputting the vehicle's target speed includes: obtaining the discretized vehicle speed value stored in the leaf node of the decision tree as the preliminary target vehicle speed; comparing the preliminary target vehicle speed with the instantaneous maximum constraint speed limit at the current moment; in response to the preliminary target vehicle speed being greater than the instantaneous maximum constraint speed limit, correcting the target vehicle speed to a legal discrete vehicle speed value that is not greater than the instantaneous maximum constraint speed limit, and using the corrected vehicle speed as the final output target vehicle speed.

[0015] In one embodiment of this disclosure, the method further includes: receiving an incremental data packet sent via in-vehicle wireless communication, the incremental data packet containing a newly added speed limit scenario state, the corresponding state transition conditions, and a new state-specific decision tree corresponding to the newly added speed limit scenario state; and dynamically updating the state machine logic and decision tree model library without restarting the main program to support the identification of the newly added speed limit scenario state and vehicle speed decision.

[0016] Secondly, embodiments of this disclosure provide a speed control device based on speed limit rules, the speed control device based on speed limit rules comprising:

[0017] The determination module is used to acquire multi-source information to characterize the vehicle's driving environment and, based on the multi-source information, determine the current speed limit scenario state in which the vehicle is located.

[0018] The acquisition module is used to determine the pre-trained decision tree corresponding to the speed limit scenario state;

[0019] The analysis module is used to input multi-source information into the decision tree and output the vehicle's target speed.

[0020] The execution module is used to control the vehicle speed according to the target vehicle speed.

[0021] Optionally, the determination module is specifically used to perform fusion processing on multi-source information to generate standardized feature vectors; input the feature vectors into a pre-trained state machine, and output the speed limit scenario state to which the current vehicle belongs; wherein the speed limit scenario state includes at least one or more of steady-state cruise, electronic eye avoidance, tunnel passage, dynamic speed limit zone, severe weather impact, and composite rule zone.

[0022] Optionally, the determining module specifically includes a standardized feature vector, which is a fixed-dimensional numerical vector extracted from multi-source information after spatiotemporal alignment and validity verification. The feature vector includes at least one of the following: the instantaneous maximum constraint speed limit representing the lowest value among all speed limit sources; a speed limit scenario conflict flag representing whether there are two or more valid speed limit rules with different values; a driver style coefficient representing the driver's preference for efficiency and comfort; and a positional correlation feature representing the distance of the vehicle from the next speed limit change point.

[0023] Optionally, the acquisition module specifically includes a decision tree that corresponds one-to-one with the speed limit scenario state; wherein each state-specific decision tree is pre-configured as a lightweight tree model generated through independent training, using driving data collected under the same speed limit scenario state as training samples and discrete speed values ​​formed by discretizing the actual driving speed under the scenario into an integer multiple of the set speed.

[0024] Optionally, the acquisition module specifically includes: the lightweight tree model corresponding to the state-specific decision tree is trained by applying depth pruning and feature pruning constraints during training; the state-specific decision tree is stored in the storage medium with an implicit index structure.

[0025] Optionally, the analysis module is specifically used to: if the multi-source information includes the instantaneous maximum constraint speed limit at the current moment; obtain the discretized vehicle speed value stored in the leaf node of the decision tree as the initial target vehicle speed; compare the initial target vehicle speed with the instantaneous maximum constraint speed limit at the current moment; and in response to the initial target vehicle speed being greater than the instantaneous maximum constraint speed limit, correct the target vehicle speed to a legal discrete vehicle speed value that is not greater than the instantaneous maximum constraint speed limit, and use the corrected vehicle speed as the final output target vehicle speed.

[0026] Optionally, the determining module is also used to receive incremental data packets sent via vehicle wireless communication, the incremental data packets containing the newly added speed limit scenario state, the corresponding state transition conditions, and the new state-specific decision tree corresponding to the newly added speed limit scenario state; and to dynamically update the state machine logic and decision tree model library without restarting the main program, so as to support the identification of the newly added speed limit scenario state and the vehicle speed decision.

[0027] Thirdly, embodiments of this application provide a control device, including: a memory and a processor;

[0028] The memory stores instructions that the computer executes;

[0029] The processor executes computer execution instructions stored in memory, causing the processor to perform a speed control method based on a speed limit rule, as described in the first aspect of this disclosure.

[0030] Fourthly, embodiments of this disclosure also provide a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, are used to implement the speed control method based on speed limit rules as described in the first aspect of this disclosure.

[0031] Fifthly, embodiments of this disclosure also provide a computer program product comprising computer execution instructions, which, when executed by a processor, are used to implement the speed control method based on speed limit rules as described in the first aspect of this disclosure.

[0032] The speed control method, apparatus, device, and medium based on speed limit rules provided in this disclosure acquire multi-source information and determine the current speed limit scenario state of the vehicle. Then, it calls a pre-trained decision tree corresponding to the scenario state, inputs the multi-source information into the decision tree, and infers and outputs the target vehicle speed. Finally, it controls the vehicle speed based on the target speed. Thus, the infinitely complex problem of fusing multi-source speed limit rules is transformed into the problem of identifying and independently modeling finite discrete scenarios. This solves the logic bloat and maintenance dilemmas caused by rule cross-conflict in traditional rule-based methods at the architectural level. It also avoids the high computational cost and low interpretability defects of learning methods due to model complexity. It achieves the technical effects of scene adaptability flexibility, human-like decision-making, and lightweight vehicle platform while strictly adhering to the maximum speed limit regulations, significantly improving the engineering practicality and system scalability of speed control in complex speed limit scenarios. Attached Figure Description

[0033] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0034] Figure 1 This diagram illustrates an application scenario of the speed control method, apparatus, device, and medium based on speed limit rules provided in this disclosure.

[0035] Figure 2 A flowchart illustrating a speed control method based on speed limit rules provided in one embodiment of this disclosure;

[0036] Figure 3 A flowchart of a speed control method based on speed limit rules provided in yet another embodiment of this disclosure;

[0037] Figure 4 A schematic diagram of a speed control device based on speed limit rules provided in yet another embodiment of this disclosure;

[0038] Figure 5 This is a schematic diagram of the structure of a control device provided in yet another embodiment of this disclosure.

[0039] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0040] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0041] Speed ​​control based on speed limit rules is a fundamental safety function of intelligent driving systems. Its core challenge lies in the complex characteristics of speed limit information in the vehicle's driving environment, which is multi-source, heterogeneous, dynamic, and potentially conflicting. Static speed limits from high-precision maps, temporary signs perceived visually, dynamic warnings from vehicle-to-infrastructure (V2I) systems, and regulations for specific areas often coexist with overlapping priorities. The system must complete information fusion, rule arbitration, and speed decision-making within milliseconds. The essence of this problem is to provide real-time, compliant, and human-like responses to an infinite number of combined speed limit scenarios under the constraints of limited onboard resources. Its complexity increases exponentially with the number of scenarios, making it one of the main technical bottlenecks for the mass production and deployment of advanced driver assistance systems (ADAS).

[0042] In related technologies, speed control schemes based on static priorities simplify multi-source speed limiting to a single numerical comparison, resulting in rigid decision-making and an inability to handle rule overlap. While state machine-based schemes possess scene classification capabilities, states and rules require manual pre-setting, creating a fundamental contradiction between scene granularity and system maintainability. Deep learning-based methods offer improved flexibility, but their large model parameter count and high inference latency make them unsuitable for meeting real-time requirements in low-cost automotive-grade controllers, and their black-box nature prevents functional safety certification. All these solutions seek improvement within the same technical paradigm, consistently failing to simultaneously address the triple constraints of decision-making flexibility, system lightweighting, and safety compliance.

[0043] The speed control method, device, equipment, and medium based on speed limit rules provided in this application classify and aggregate complex speed-limited environments through a macro-level scene state machine, and then independently configure a micro-level decision tree for each scene to realize vehicle speed reasoning. Furthermore, by recognizing scenes, the infinite combination of rules is reduced to a finite discrete state, fundamentally curbing logical expansion; through a dedicated decision tree driven by anthropomorphic driving data, flexible vehicle speeds are output while strictly adhering to speed limits; through the layered collaboration of these two approaches, the system combines the determinism of rule-based methods with the adaptability of data-based methods, and simultaneously overcomes the maintenance, resource, and compliance bottlenecks of traditional solutions with an extremely lightweight architecture.

[0044] Figure 1 The application scenario diagrams of the speed control method, device, equipment, and medium based on speed limit rules provided in this application are as follows: Figure 1As shown, during vehicle operation, the vehicle control system 100 automatically collects vehicle status data and environmental data 110, and when it is determined that the vehicle is about to enter the ramp environment, it controls the vehicle speed in real time according to the corresponding control strategy 120.

[0045] It should be noted that, Figure 1 The scenario shown includes an on-board control system, vehicle status data and environmental data. The control strategy is only illustrated by one or a specific number of examples, but this disclosure is not limited to this. That is to say, the number of on-board control systems, vehicle status data and environmental data, and control strategies can be arbitrary.

[0046] The technical solution of this application and how the technical solution of this application solves the above-mentioned technical problems are described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of this application will now be described with reference to the accompanying drawings.

[0047] Figure 2 This is a flowchart illustrating the speed control method based on speed limit rules provided in an embodiment of this application. The following is a summary of the process. Figure 2 The main process of the speed control method based on speed limit rules is explained below:

[0048] S201. Obtain multi-source information to characterize the vehicle's driving environment, and determine the current speed limit scenario state of the vehicle based on the multi-source information.

[0049] Specifically, in this embodiment of the speed control method based on speed limit rules, the executing entity is a system, an onboard processor, or a cloud server capable of acquiring and analyzing vehicle driving status and road information. For ease of explanation, it will be referred to as the system below.

[0050] The system continuously collects multi-source information characterizing the vehicle's driving environment through the onboard sensing and communication unit.

[0051] Multi-source information includes at least the following categories: First, static road attribute data provided by high-precision maps and navigation systems, including road grade, design speed, number of lanes, geographical location and associated speed limit values ​​of special road sections such as tunnels and bridges; Second, traffic sign information captured in real time by visual perception systems through forward-facing cameras, covering the location and speed limit values ​​of fixed speed limit signs, variable message signs, and electronic eye monitoring equipment.

[0052] Thirdly, the vehicle-road cooperative communication unit receives dynamic event information from roadside units or cloud service platforms, such as speed limits in temporary construction areas, recommended speeds for severe weather warnings, and speeds for traffic congestion relief; fourthly, the vehicle's precise latitude and longitude, heading angle, current instantaneous speed, and turn signal status output by the positioning and vehicle status unit; and fifthly, the driver's mode selection or driving preference tendencies learned by the system through the human-machine interface.

[0053] After acquiring the above multi-source heterogeneous information, the system first performs spatiotemporal alignment and validity verification on it.

[0054] Spatiotemporal alignment here refers to unifying information with different sampling frequencies and transmission delays into the same time reference coordinate system, and spatially matching the location of visually recognized traffic signs and the scope of V2X events with geofences in high-precision maps to determine whether the current vehicle is within the effective area of ​​the speed limit rule.

[0055] The validity verification is based on the confidence level of the information source, signal strength, historical packet loss rate, and consistency with other information sources. Each rate limit information is assigned a reliability weight, and interference data that is obviously abnormal or has too low confidence is filtered out.

[0056] Based on the preprocessed multi-source information, the system can make a qualitative judgment on the macro speed limit scenario in which the vehicle is currently located.

[0057] The classification of scene types is based on typical road conditions that drivers can clearly perceive and adjust their speed accordingly during actual driving.

[0058] For example, when a speed camera is detected within a specified distance ahead of the vehicle and the current vehicle speed is higher than the speed limit at that location, the system determines that it is entering a speed camera avoidance scenario; when high-precision map matching shows that the vehicle is less than the set distance from the tunnel entrance and the light sensor value continues to decrease, the system determines that it is entering a tunnel passage scenario; when the V2X unit receives a temporary construction or event warning broadcast by the roadside equipment and the vehicle is within the warning area, the system determines that it is entering a dynamic speed limit zone scenario; when the vehicle is in a basic road speed limit environment and no of the above special constraints are activated, the system determines that it is entering a steady-state cruise scenario.

[0059] Multiple scenarios may be triggered simultaneously. The system selects the scenario with the highest priority as the current speed limit scenario state based on the preset conflict arbitration logic. For example, the priority of dynamic construction information is usually higher than the speed limit of the tunnel foundation.

[0060] This scene recognition process does not rely on complex reasoning calculations, but is based on explicit threshold conditions and lookup table logic to ensure that the state determination is completed in a fixed time within each control cycle, providing a deterministic scene context for subsequent vehicle speed decisions.

[0061] S202. Determine the pre-trained decision tree corresponding to the speed limit scenario state.

[0062] Specifically, the system maintains a mapping table between scene states and decision tree models. This table is pre-built offline and stored in non-volatile memory.

[0063] Each rate-limited scenario state corresponds to an independently stored decision tree model, and there is a strict one-to-one correspondence between the scenario state and the decision tree.

[0064] Once the current speed-limited scenario state is determined and a scenario identifier is output, the system can use this identifier as an index to locate and load the corresponding decision tree model into the runtime memory area via a direct table lookup. The entire process requires no traversal search or conditional judgment, and its time complexity is constant, which is strictly compatible with high-frequency control loops.

[0065] The pre-training process of the decision tree is completed during the offline development phase of the system, without consuming real-time computing resources on the vehicle. The server pre-collects driving data of the vehicle in real road environments for each speed limit scenario.

[0066] Taking tunnel traffic as an example, the system collects traffic data of vehicles in multiple tunnels of different types and lengths, and simultaneously records the speed change curves of vehicles before entering the tunnel, while driving in the tunnel, and after exiting the tunnel.

[0067] In some embodiments, during the data collection process, data corresponding to non-congested periods with smooth traffic flow can be deliberately selected to minimize interference from vehicles ahead and ensure that the collected vehicle speeds primarily reflect the driver's subjective response strategy to speed limit rules.

[0068] After the data collection is completed, the raw vehicle speed data can be cleaned to ensure compliance, removing any records that exceed the maximum speed limit on the road. The vehicle speed can then be discretized and quantized at set intervals to form a training dataset with clear labels.

[0069] In some embodiments, during the driving data compliance cleaning phase, in addition to removing data frames that exceed the immediate maximum speed limit, the system further removes vehicle speed samples from atypical steady-state cruise conditions such as excessive longitudinal acceleration, excessively close following distance, and lane changing / overtaking. This scenario filtering ensures that the training dataset retains the driver's pure response strategy to the speed limit rule itself under undisturbed conditions, thus improving the scenario generalization ability of the decision tree.

[0070] During training, decision tree models can employ splitting criteria based on information gain or the Gini coefficient to recursively construct a binary tree structure from top to bottom. Each internal node contains a feature comparison condition, while the leaf nodes directly store a discretized target vehicle speed value.

[0071] Unlike traditional decision trees that aim to maximize prediction accuracy, this approach actively imposes depth and leaf node number constraints during training, forcing the tree model to grow into a shallow, compact structure. For example, the maximum tree depth is set to no more than five levels, and the maximum number of leaf nodes is sixteen, thus ensuring that the storage space and inference path length of a single model are kept at extremely low levels.

[0072] After training converges, the model is subjected to structured compression, redundant comparison nodes are removed, and the tree nodes are stored linearly as a continuous array.

[0073] Each state-specific decision tree generated is a lightweight tree model, whose size and execution efficiency can meet the stringent constraints of automotive-grade microcontrollers on code storage space and worst-case execution time.

[0074] S203. Input multi-source information into the decision tree and output the target vehicle speed.

[0075] Specifically, within each control cycle, the system uses the preprocessed multi-source information of the current moment as an input vector and inputs it into the loaded scenario-specific decision tree.

[0076] The reasoning process of a decision tree follows a deterministic binary tree traversal protocol: starting from the root node, it reads the feature index and comparison threshold stored in that node, extracts the value of the corresponding feature dimension from the input vector, and compares it with the threshold as an integer or floating-point number; if the feature value is less than or equal to the threshold, it jumps to the left child node; otherwise, it jumps to the right child node. The above process is repeated until the leaf node is reached.

[0077] Because each decision tree is forced to be pruned into a shallow structure, the path length from the root node to any leaf node is limited to a small number of comparison operations (e.g., 3 to 7). The entire reasoning process does not involve loops, backtracking, or recursive calls, and its execution path and execution time are completely determined.

[0078] Meanwhile, the leaf nodes do not store continuous probability distributions or regression values, but rather discretely encoded integer vehicle speed values. This value is directly used as the initial target vehicle speed output determined by the system in this cycle.

[0079] For example, in a tunnel passage scenario, a leaf node of the decision tree may store the value 80, indicating that the system determines that, under the current feature combination, setting the target speed of 80 kilometers per hour is most consistent with the decision-making habits of an excellent driver in this scenario.

[0080] The discretization granularity of the output value can be set to an integer multiple of five kilometers per hour or ten kilometers per hour according to engineering calibration requirements. This design makes the setpoint received by the downstream controller inherently stable and will not produce continuous jitter due to small fluctuations in the input characteristics.

[0081] In some embodiments, to improve the scene adaptability of the decision tree, the dedicated tree for different scenes only uses a subset of features that are strongly related to that scene for reasoning.

[0082] For example, tunnel traffic decision trees primarily rely on features such as distance from the tunnel entrance, distance from the tunnel exit, tunnel lighting conditions, and driver style coefficients, while ignoring information with low relevance to the scenario, such as weather conditions and road curvature. This feature pruning strategy reduces the complexity of the tree model and prevents irrelevant features from introducing unnecessary noise into the decision logic.

[0083] The entire reasoning process can be implemented based on pure integer operations, without the need for a floating-point coprocessor. On mainstream automotive-grade control chips, the time consumed for a single reasoning operation is extremely short, fully meeting the real-time requirements of advanced driver assistance systems for the perception-decision-control link.

[0084] S204. Control the vehicle speed according to the target speed.

[0085] Specifically, the system uses the output discretized target vehicle speed value as the desired setpoint and broadcasts it to the downstream actuator responsible for longitudinal motion control at fixed intervals via the vehicle controller LAN bus or a higher bandwidth vehicle Ethernet link.

[0086] After receiving the target vehicle speed, the downstream controller compares it with the vehicle's current actual speed, calculates the speed deviation, and generates an accelerator pedal opening request or a braking pressure request through proportional-integral-derivative control or its engineered variants, driving the actuator to make the vehicle's actual speed follow the target vehicle speed.

[0087] For example, taking a highway cruise scenario, when the system determines that it is currently in a steady-state cruise state and the driver's style coefficient is in the economic mode range, the decision tree may output 110 kilometers per hour as the target speed for several consecutive control cycles; when the vehicle enters the tunnel entrance area, the scenario state switches to tunnel passage, and the target speed smoothly transitions to 80 kilometers per hour within several cycles.

[0088] This discrete, stepped target vehicle speed sequence provides a clear control reference point for downstream controllers, effectively avoiding frequent alternating throttle and brake actions caused by continuous small fluctuations in the target value, thereby reducing actuator wear and vehicle energy consumption, and improving ride smoothness.

[0089] While control commands are issued, the system continuously monitors the vehicle's tracking response to the target speed. If the deviation between the actual speed and the target speed continuously exceeds the preset tolerance range due to factors such as slope, load, or changes in road surface adhesion coefficient, the system can adjust the target speed output as needed or request torque compensation from a higher-level decision-making module.

[0090] Thus, the entire process forms a closed-loop feedback chain of perception, decision-making, and execution, ensuring that the vehicle can stably and reliably execute the speed commands decided by this solution under different driving conditions.

[0091] In addition, the system will present key decision information for each control cycle, including the current speed limit scenario status, target vehicle speed, and real-time maximum constraint speed limit, in a visual or signal-based form on the human-machine interface, so that the driver can clearly understand the system's behavioral intent and enhance their trust in using the assisted driving functions.

[0092] The speed control method based on speed limit rules provided in this application acquires multi-source information and determines the current speed limit scenario state of the vehicle. Then, it calls a pre-trained decision tree corresponding to this scenario state, inputs the multi-source information into the decision tree, and infers and outputs the target vehicle speed. Finally, it controls the vehicle speed based on the target speed. This transforms the infinitely complex problem of fusing multi-source speed limit rules into the problem of identifying and independently modeling finite discrete scenarios. It solves the logic bloat and maintenance dilemmas caused by rule cross-conflict in traditional rule-based methods from an architectural perspective. Simultaneously, it avoids the high computational cost and low interpretability defects of learning methods due to model complexity. It achieves the technical effects of scene adaptability flexibility, human-like decision-making, and lightweight vehicle platform while strictly adhering to the maximum speed limit regulations, significantly improving the engineering practicality and system scalability of speed control in complex speed limit scenarios.

[0093] Figure 3 Another embodiment of the speed control method based on speed limit rules provided in this disclosure, in Figure 2 Based on the illustrated embodiment, the following is combined with Figure 3 The implementation process of the speed control method based on speed limit rules is explained in detail, and it specifically includes the following steps:

[0094] S301. Obtain multi-source information to characterize the vehicle's driving environment.

[0095] Specifically, this step is related to Figure 2 The corresponding steps in the illustrated embodiments are the same and will not be repeated here.

[0096] In some embodiments, the multi-source information also includes geofence information for special areas. High-precision maps pre-mark the geographical boundaries of special road sections such as school areas, hospital areas, accident-prone sections, and noise-sensitive areas, along with their suggested or legal speed limits. When a vehicle's location enters such a geofence, the system includes it as one of the sources for calculating the immediate maximum speed limit. Some special area speed limits have time-limited attributes; for example, school areas are only effective during weekday school hours. The system needs to combine local clock and calendar information to comprehensively determine the activation status of the speed limit rule.

[0097] S302. Perform fusion processing on multi-source information to generate standardized feature vectors.

[0098] Specifically, the system first performs spatiotemporal alignment and validity verification on the collected multi-source heterogeneous information. For details on the implementation, please refer to [link / reference needed]. Figure 2 The corresponding descriptions in the illustrated embodiments will not be repeated here.

[0099] Based on cleaned and aligned multi-source information, the system extracts and generates a set of fixed-dimensional numerical feature vectors.

[0100] Furthermore, the standardized feature vector is a fixed-dimensional numerical vector extracted from multi-source information after spatiotemporal alignment and validity verification. It includes at least one of the following: the instantaneous maximum constraint speed limit representing the lowest value among all speed limit sources; a speed limit scenario conflict flag representing whether there are two or more valid speed limit rules with different values; a driver style coefficient representing the driver's preference for efficiency and comfort; and a positional correlation feature representing the distance of the vehicle from the next speed limit change point.

[0101] Specifically, the real-time maximum speed limit is determined by the system iterating through all active and valid speed limit information at the current moment and extracting the corresponding speed limit values. The system then selects the minimum value among these as the absolute speed limit that the vehicle must not exceed. This feature represents a rigid constraint at the regulatory level and serves as the baseline parameter to ensure driving compliance.

[0102] The speed limit scenario conflict flag indicates that the system will detect whether there are two or more active speed limit rules that have passed validity checks and have different speed limit values. If such a rule exists, the flag will be set to true; otherwise, it will be set to false. This feature is a key identifier for identifying complex speed limit scenarios and triggering complex arbitration logic.

[0103] The driver style coefficient is a quantitative coefficient generated by the system through the human-machine interface by reading the driver's preset driving mode selection, or by online learning based on long-term driving behavior since vehicle startup. This coefficient is typically normalized to a set range; a smaller value indicates a greater preference for an economical and conservative driving style, while a larger value indicates a greater preference for a sporty and efficient driving style. This feature allows speed decisions to reflect the individualized habits of different drivers.

[0104] In some embodiments, the driver style coefficient is generated using two modes: static configuration and dynamic learning. In static configuration mode, the driver selects preset modes such as Economy, Comfort, and Sport via the central control screen, and the system maps these modes to fixed coefficient values. In dynamic learning mode, the system uses a long driving cycle after vehicle startup as a time window to statistically analyze features such as the percentage deviation of the driver's free-driving speed relative to the road speed limit under conditions without interference from the vehicle ahead, common acceleration ranges, and following distance. The system updates the style coefficient in real time using clustering or regression algorithms and visualizes the learning results for the driver to confirm whether to adopt.

[0105] The distance to the next speed limit change point is calculated by the system using high-precision map road topology information and the vehicle's current location to search forward for locations where road attributes change (such as tunnel entrances, construction zone start points, speed camera locations, and boundaries between road sections with different speed limits), and then calculating the longitudinal path distance from the vehicle's current location to that location. This feature provides a distance basis for a phased, smooth deceleration strategy.

[0106] The standardized feature vector with fixed dimensions can be composed of the above features, or it can include a part of them, or it can also contain other information besides the above dimensions. The values ​​of each dimension are normalized or discretized to adapt to the uniform requirements of the input format of the subsequent state machine and decision tree.

[0107] In some embodiments, the information corresponding to the aforementioned feature vectors includes traffic signs identified by the visual perception system, such as variable message signs. The content displayed on the variable message signs changes in real time according to traffic control strategies. The system dynamically updates the currently effective speed limit information by verifying the consistency of recognition results across multiple consecutive frames of images and combining this with a confidence score. In cases where the speed limit displayed on the variable message sign is inconsistent with the base speed limit on the high-precision map, the system prioritizes using the variable message sign values ​​in the calculation of subsequent real-time maximum constraint speed limits.

[0108] S303. Input the feature vector into the pre-trained state machine and output the current speed limit scene state to which the vehicle belongs.

[0109] Among them, the speed limit scenario states include at least one or more of the following: steady-state cruise, electronic eye avoidance, tunnel passage, dynamic speed limit zone, severe weather impact, and complex rule zone.

[0110] Specifically, the system maintains a pre-built finite state machine internally.

[0111] The state space of this state machine covers various types of typical speed-limited scenarios in non-congested road sections, namely the speed-limited scenario states mentioned above. Each state represents a macroscopic road condition characteristic that drivers can clearly perceive and adjust their vehicle speed accordingly.

[0112] The state machine is constructed using an offline rule table-driven approach. During development, engineers can define clear entry and exit conditions for each state based on driving experience and traffic regulations. For example, the entry condition for the electronic eye avoidance state is: an electronic eye device is detected within a specified distance ahead and the current vehicle speed is greater than the speed limit at that location minus a set tolerance margin; the exit condition is: the vehicle passes the location or the vehicle speed has dropped below the speed limit.

[0113] These conditions are expressed as thresholds for one or more dimensions of the feature vector and organized into a lookup table structure.

[0114] In each control cycle, the system inputs the feature vector of the current moment into the state machine. The state machine starts from the current state and evaluates in turn whether the transition conditions of each state are satisfied.

[0115] Since the transition conditions are all implemented using direct comparison operations, and the number of states is usually strictly limited to a finite number (e.g., no more than 10), the time complexity of the entire state determination process is constant, and can be completed in microseconds. The determination result outputs a unique scene state identifier, which will serve as the query key for indexing the decision tree in subsequent steps.

[0116] The composite rule zone state is a special state specifically designed for scenarios involving conflicts between multiple speed limit rules. When the speed limit conflict flag is true, the system does not simply force a selection from the existing states, but actively switches to the composite rule zone state.

[0117] This state does not pre-determine any fixed rules or arbitration logic. Instead, it treats the conflict itself as a scenario and entrusts it to a dedicated decision tree for data-driven, anthropomorphic processing.

[0118] This fundamentally avoids the complex problem of manually arranging priorities when rules intersect, a problem inherent in traditional solutions.

[0119] S304. Determine the pre-trained decision tree corresponding to the speed limit scenario state.

[0120] Specifically, the system maintains a static mapping table between scene states and decision tree models. This table is stored in non-volatile memory before the system leaves the factory. Each rate-limited scene state corresponds to an independently stored and trained decision tree model, and the two have a strict one-to-one correspondence.

[0121] Once the current scene state identifier is obtained, the system uses this identifier as an index to locate the starting address of the corresponding decision tree in the storage medium through direct array addressing, and loads it into the runtime memory area.

[0122] The entire process does not involve traversal search or conditional branching, and the execution time is completely deterministic.

[0123] Furthermore, the decision tree is a state-specific decision tree that corresponds one-to-one with the state of the speed-limited scenario;

[0124] Each state-specific decision tree is pre-configured as follows: using driving data collected under the same speed limit scenario as training samples, and using discrete speed values ​​formed by discretizing the actual driving speed under that scenario into integer multiples of the set speed as training labels, and generating a lightweight tree model through independent training.

[0125] Specifically, each state-specific decision tree is generated through an independent training process based on real human driving data collected in that scenario.

[0126] After data collection, the system performs compliance cleaning on the raw vehicle speed data, removing any data frames that exceed the road's maximum speed limit or clearly violate safe following distance regulations. The cleaned continuous speed values ​​are then discretized and quantized, rounding the actual speed to the nearest integer multiple according to a set gear interval. The gear interval is typically set to five or ten kilometers per hour, and the rounding rule uses either rounding to the nearest whole number or rounding down, determined by engineering calibration. The discretized speed values ​​are the labels used in the decision tree training phase.

[0127] In some embodiments, when discretizing driving data during the offline training phase of the decision tree, the system uses a rounding rule to round continuous actual vehicle speeds to the nearest integer multiple of the set gear. For vehicle speed values ​​that are exactly between two gears, the system can be configured to round up or down. This rule is calibrated based on vehicle type and target user group preferences and is fixed before training to ensure label determinism.

[0128] In some embodiments, after generating discretized labels for the driving data, the system also performs comfort filtering. This filtering is applied to the discretized vehicle speed label sequence, using a sliding window or median filtering algorithm to filter out abnormal sample points that frequently jump between adjacent discrete gears due to subtle driver pedal operations or sensor noise. The filtered label sequence exhibits a step-like stable characteristic, making the output of the trained decision tree more stable and preventing the downstream controller from generating unnecessary acceleration or deceleration actions due to frequent gear changes at the target vehicle speed.

[0129] Furthermore, the lightweight tree model corresponding to the state-specific decision tree is trained by applying depth pruning and feature pruning constraints during training; the state-specific decision tree is stored in the storage medium with an implicit index structure.

[0130] Specifically, during training, the system employs an information gain-based node splitting criterion to recursively construct a binary tree from top to bottom. Each internal node contains a feature index and a comparison threshold, while the leaf nodes directly store a discretized target vehicle speed value.

[0131] Unlike traditional decision trees that pursue ultimate accuracy, this approach actively applies several lightweight constraints during the training phase. One of these is feature pruning, which means retaining only feature dimensions that are strongly relevant to the scenario for splitting. For example, the tunnel traffic tree mainly uses features such as distance from the tunnel entrance, distance from the tunnel exit, and driver style coefficients, while ignoring irrelevant features such as road curvature and weather conditions.

[0132] The second is deep pruning, which forcibly limits the maximum number of tree growth layers, for example, setting it to a fixed 3 to 5 layers, to ensure that the path length from the root node to any leaf node does not exceed 7 comparisons.

[0133] Thirdly, there is a constraint on the number of leaf nodes, which limits the maximum number of leaf nodes in a single tree to control the model size.

[0134] After training convergence, the system optimizes the storage structure of the decision tree. Traditional decision trees typically link parent and child nodes using pointers. On resource-constrained microcontrollers, pointers occupy a large amount of space and have low addressing efficiency.

[0135] This scheme uses an implicit index structure for storage: all nodes of the binary tree are stored in a contiguous array in breadth-first order. For a node with index i in the array, the index of its left child node is twice i plus one, the index of its right child node is twice i plus two, and the index of its parent node is i minus one, divided by two and rounded down.

[0136] This storage method completely eliminates explicit pointers, and the model size of a single decision tree can be compressed to less than 256 bytes, which can be directly stored in the microcontroller's static random access memory or cache, achieving zero-wait loading.

[0137] In some embodiments, the decision tree model and state machine can also be dynamically updated, and the method includes the following steps:

[0138] Step A1: Receive incremental data packets sent via in-vehicle wireless communication.

[0139] The incremental data packet contains the newly added speed limit scenario state, the corresponding state transition conditions, and the new state-specific decision tree corresponding to the newly added speed limit scenario state.

[0140] Specifically, the system receives incremental data packets from the cloud service platform via the vehicle-mounted wireless communication module. These data packets use a differential encoding format and contain only the changes relative to the baseline version.

[0141] The incremental data packet contains at least three parts of information: first, the definition of the newly added rate-limited scenario state and its unique identifier; second, the set of state transition conditions associated with the newly added state, which describes the feature dimensions, comparison operators and thresholds in key-value pairs; and third, the complete model parameters of the new state-specific decision tree corresponding to the newly added state. This model has usually been pre-processed according to the aforementioned lightweight training and compression process and encapsulated in byte stream form.

[0142] Step A2: Without restarting the main program, dynamically update the state machine logic and decision tree model library to support the recognition of new speed limit scenario states and vehicle speed decisions.

[0143] Specifically, after receiving and completing the integrity verification, the system triggers a dynamic update process. This process does not require restarting the main program or interrupting the currently executing vehicle control task.

[0144] The system first appends the new state identifier and its transition condition table to the end of the state list of the state machine module; then it allocates storage space for the new tree in the decision tree model library and parses the tree parameter byte stream in the incremental package into an executable implicit index array structure; finally, it updates the mapping table between the scene state and the decision tree so that the new state identifier can be correctly indexed to the new tree.

[0145] The entire update process employs a double-buffer mechanism. After the data structure is reorganized in the background, the active pointer is switched through atomic operations to ensure that the system's decision logic is a complete and consistent version within any control cycle.

[0146] In some embodiments, the dynamic update process of the state machine and decision tree employs a version rollback protection mechanism. The system fully backs up the current state machine logic table and decision tree model library before installing incremental data packages. If the new version triggers a degradation alarm frequency exceeding a threshold within a preset observation period, or receives an explicit rollback command from the cloud, the system automatically switches back to the previous stable version to ensure high availability of vehicle control functions.

[0147] S305. Obtain the discretized vehicle speed values ​​stored in the leaf nodes of the decision tree as the initial target vehicle speed.

[0148] In this embodiment, the multi-source information includes the current instantaneous maximum constraint speed limit.

[0149] Specifically, after the system completes the loading of the decision tree and the input of the feature vector, it initiates the deterministic reasoning process.

[0150] The reasoning process starts from the root node and repeats the following operations until it reaches the leaf node: read the feature index stored in the current node and extract the corresponding dimension value from the feature vector; read the comparison threshold stored in the current node; compare the feature value with the threshold value. If the feature value is less than or equal to the threshold value, jump to the left child node; otherwise, jump to the right child node.

[0151] Because each decision tree undergoes deep pruning, the length of the inference path is strictly limited to 3 to 7 comparison operations.

[0152] When the process reaches a leaf node, the system directly reads the numerical field from the node's storage area. This field is not a continuous probability distribution or regression weight, but a discretized integer vehicle speed value in kilometers per hour, and the value is a multiple of five or ten. This value is the initial target vehicle speed determined by the system within the current control cycle.

[0153] Taking a tunnel passage scenario as an example, when the feature vector shows that the vehicle is 500 meters away from the tunnel entrance, the driver style coefficient is 1.08, and the current vehicle speed is 100 kilometers per hour, the corresponding leaf node of the decision tree may store the value 85, indicating that the system imitates the habits of a sporty driver and only slightly decelerates to 85 kilometers per hour at this distance; when the vehicle moves to 150 meters away from the entrance, another leaf node of the same decision tree may be activated, outputting the value 80, and completing the final deceleration.

[0154] In this process, the multi-source information must include the current instantaneous maximum constraint speed limit, which has been used as a one-dimensional input to the decision tree as a feature vector.

[0155] However, it is worth noting that although decision trees can indirectly learn the relationship between the instantaneous maximum constraint speed limit and the output vehicle speed during the training process, the values ​​stored in their leaf nodes are not guaranteed to be strictly no greater than the instantaneous maximum constraint speed limit under any combination of inputs.

[0156] For example, in cases of model generalization boundary or rare feature combinations, the decision tree may output a speed value slightly higher than the red line.

[0157] S306. Compare the initial target speed with the current instantaneous maximum speed limit.

[0158] Specifically, the system extracts the instantaneous maximum constraint speed limit dimension value from the feature vector and compares it with the preliminary target vehicle speed output in the previous step.

[0159] The immediate maximum speed limit, as a rigid baseline derived from the fusion of multi-source speed limits, may not be a multiple of 5 or 10. For example, when the high-precision map limits speed to 80 km / h, V2X dynamic construction limits speed to 70 km / h, and visual recognition of temporary signs limits speed to 60 km / h, the immediate maximum speed limit is 60 km / h. When only the high-precision map limits speed to 120 km / h and no other speed limit sources are activated, the immediate maximum speed limit is 120 km / h. The unit of measurement for this value is consistent with the decision tree output, which is kilometers per hour.

[0160] S307. In response to the initial target vehicle speed being greater than the immediate maximum constraint speed limit, the target vehicle speed is corrected to a legal discrete vehicle speed value that is not greater than the immediate maximum constraint speed limit, and the corrected vehicle speed is used as the final output target vehicle speed.

[0161] Specifically, when the comparison results show that the initial target vehicle speed is greater than the immediate maximum constraint speed limit, the system activates the safety clamp correction logic.

[0162] The correction operation does not simply assign the target vehicle speed to the original value of the instantaneous maximum constraint speed limit, but rather maps it to a valid discrete gear according to the requirements of the discretized output format.

[0163] Specifically, the system first obtains the currently configured discrete gear interval, which is typically 5 kilometers per hour. Then, within a range that is no greater than the immediate maximum speed limit, the system searches for the largest integer multiple of that gear interval.

[0164] For example, assuming the immediate maximum speed limit is 82 km / h, the initial target speed output by the decision tree is 85 km / h. The system uses 5 km / h intervals for each gear, and the legal discrete gears not greater than 82 include 80, 75, 70, etc., with the maximum value being 80.

[0165] Therefore, the system will adjust the final output target speed to 80 km / h. For example, if the immediate maximum speed limit is 78 km / h and the initial target speed is 80 km / h, then the maximum value not exceeding 78 in the legal discrete gears is 75, and the system will output 75 km / h.

[0166] If the initial target vehicle speed is less than or equal to the immediate maximum constraint speed limit, the safety clamp correction logic will not take effect, and the system will directly use the discretized value stored in the leaf node of the decision tree as the final output target vehicle speed.

[0167] Therefore, the output target vehicle speed has two guarantees: first, the numerical form is strictly an integer multiple of the set gear, providing a stable set point for the downstream controller; second, the numerical value is strictly no greater than the highest constraint red line after the fusion of multi-source speed limits, ensuring that the vehicle does not exceed the speed limit in any scenario and meets regulatory compliance requirements.

[0168] S308. Speed ​​control of the vehicle is performed based on the target speed.

[0169] Specifically, this step is related to Figure 2 The corresponding steps in the illustrated embodiments are the same and will not be repeated here.

[0170] In some embodiments, the system incorporates a built-in system health monitoring module. This module continuously monitors operational metrics such as the confidence level of multi-source input information, state machine switching frequency, and decision tree inference time. When it detects that the visual recognition confidence level is below a threshold for multiple consecutive control cycles, the vehicle-road cooperative communication link is interrupted, the state machine switches at high frequency within a very short time, or the decision tree inference time exceeds a preset warning line, the system determines that it is currently in an abnormal operating state. At this time, the system automatically degrades to a fail-safe mode, abandons the decision tree output, and directly uses the vehicle speed value of the immediate maximum constraint speed limit rounded down to the nearest discrete gear as the target vehicle speed. Simultaneously, it reports a fault code through the on-board bus and prompts the driver on the human-machine interface.

[0171] The speed control method based on speed limit rules provided in this disclosure generates a standardized feature vector through multi-source information fusion. A state machine accurately identifies the speed limit scenario and calls the corresponding dedicated decision tree to output a discretized target vehicle speed. Simultaneously, a safety clamp is introduced to ensure the vehicle speed strictly does not exceed the red line, and the state machine and decision tree are dynamically updated via OTA incremental packages. Thus, the infinitely complex rule fusion problem is reduced to independent modeling of finite scenarios, fundamentally solving the problems of logical expansion and high maintenance costs of traditional rule-based methods, as well as the high resource consumption and poor scalability of learning methods. This enables lightweight deployment of the system on low-computing-power platforms, and the decision output combines human-like flexibility with regulatory compliance, while also possessing online iteration capabilities. This significantly improves the engineering practicality and lifecycle maintainability of the speed control system in complex speed-limited environments.

[0172] Figure 4 This is a schematic diagram of a speed control device based on speed limit rules provided in one embodiment of this disclosure. Figure 4 As shown, the speed control device 400 based on speed limit rules includes:

[0173] The determination module 410 is used to acquire multi-source information to characterize the vehicle's driving environment, and to determine the current speed limit scenario state of the vehicle based on the multi-source information.

[0174] The acquisition module 420 is used to determine the pre-trained decision tree corresponding to the speed limit scenario state;

[0175] Analysis module 430 is used to input multi-source information into the decision tree and output the target vehicle speed;

[0176] The execution module 440 is used to control the speed of the vehicle according to the target vehicle speed.

[0177] Optionally, the determining module 410 is specifically used to perform fusion processing on multi-source information to generate standardized feature vectors; input the feature vectors into a pre-trained state machine, and output the speed limit scene state to which the current vehicle belongs; wherein the speed limit scene state includes at least one or more of steady-state cruise, electronic eye avoidance, tunnel passage, dynamic speed limit zone, severe weather impact, and composite rule zone.

[0178] Optionally, the determining module 410 specifically includes a standardized feature vector that is a fixed-dimensional numerical vector extracted from multi-source information after spatiotemporal alignment and validity verification. The feature vector includes at least one of the following: the instantaneous maximum constraint speed limit representing the lowest value among all speed limit sources; a speed limit scenario conflict flag representing whether there are two or more valid speed limit rules with different values; a driver style coefficient representing the driver's preference for efficiency and comfort; and a positional correlation feature representing the distance of the vehicle from the next speed limit change point.

[0179] Optionally, the acquisition module 420 specifically includes a decision tree that corresponds one-to-one with the speed limit scenario state; wherein each state-specific decision tree is pre-configured as a lightweight tree model generated by independently training: using driving data collected under the same speed limit scenario state as training samples and using discrete speed values ​​formed by discretizing the actual driving speed under the scenario after being processed as an integer multiple of the set speed as training labels.

[0180] Optionally, the acquisition module 420 specifically includes: the lightweight tree model corresponding to the state-specific decision tree is trained by applying depth pruning and feature pruning constraints during training; the state-specific decision tree is stored in the storage medium with an implicit index structure.

[0181] Optionally, the analysis module 430 is specifically used to: if the multi-source information includes the instantaneous maximum constraint speed limit at the current moment; obtain the discretized vehicle speed value stored in the leaf node of the decision tree as the initial target vehicle speed; compare the initial target vehicle speed with the instantaneous maximum constraint speed limit at the current moment; in response to the initial target vehicle speed being greater than the instantaneous maximum constraint speed limit, correct the target vehicle speed to a legal discrete vehicle speed value that is not greater than the instantaneous maximum constraint speed limit, and use the corrected vehicle speed as the final output target vehicle speed.

[0182] Optionally, the determining module 410 is further configured to receive incremental data packets sent via vehicle wireless communication, the incremental data packets containing the newly added speed limit scenario state, the corresponding state transition conditions, and the new state-specific decision tree corresponding to the newly added speed limit scenario state; and dynamically update the state machine logic and decision tree model library without restarting the main program to support the identification of the newly added speed limit scenario state and vehicle speed decision.

[0183] In this embodiment, the speed control device based on speed limit rules solves the problems of insufficient flexibility, high resource consumption, and poor practicality in vehicle speed control under complex speed limit rules in related technologies by combining various modules.

[0184] Figure 5 This is a schematic diagram of the structure of a control device provided in one embodiment of the present disclosure, as shown below. Figure 5 As shown, the control device 500 includes a memory 510 and a processor 520.

[0185] The memory 510 stores a computer program that can be executed by at least one processor 520. This computer program is executed by at least one processor 520 to enable the control device to implement the speed control method based on speed limit rules as provided in any of the above embodiments.

[0186] The memory 510 and the processor 520 can be connected via a bus 530.

[0187] The relevant explanations can be understood by referring to the corresponding descriptions and effects in the method embodiments, and will not be repeated here.

[0188] One embodiment of this disclosure provides a computer-readable storage medium having a computer program stored thereon, the computer program being executed by a processor to implement the speed control method based on speed limit rules as provided in any of the above embodiments.

[0189] The computer-readable storage medium may be ROM, random access memory (RAM), CD-ROM, magnetic tape, floppy disk, and optical data storage device, etc.

[0190] One embodiment of this disclosure provides a computer program product comprising computer-executable instructions that, when executed by a processor, are used to implement the speed control method based on speed limit rules provided in any of the above embodiments.

[0191] 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 program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the methods described above. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0192] In the above embodiments, the descriptions of each embodiment have their own emphasis. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments. The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as the combination of these technical features does not contradict each other, it should be considered within the scope of this specification.

[0193] Other embodiments of this application will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this application are indicated by the following claims.

[0194] It should be understood that this application is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this application is limited only by the appended claims.

Claims

1. A speed control method based on speed limit rules, characterized in that, Includes the following steps: Acquire multi-source information to characterize the vehicle's driving environment, and determine the current speed-limited scenario state of the vehicle based on the multi-source information; Determine the pre-trained decision tree corresponding to the speed-limited scenario state; The multi-source information is input into the decision tree, and the target speed of the vehicle is output. Based on the target vehicle speed, the vehicle speed is controlled.

2. The method according to claim 1, characterized in that, Determining the current speed limit scenario state of the vehicle based on the multi-source information includes: The multi-source information is fused to generate a standardized feature vector; The feature vector is input into a pre-trained state machine, which outputs the current speed limit scene state to which the vehicle belongs. The speed limit scenario states include at least one or more of the following: steady-state cruise, electronic eye avoidance, tunnel passage, dynamic speed limit zone, severe weather impact, and composite rule zone.

3. The method according to claim 2, characterized in that, The standardized feature vector is a fixed-dimensional numerical vector extracted from the multi-source information after spatiotemporal alignment and validity verification. The feature vector includes at least one of the following: The instantaneous maximum constraint speed limit, representing the lowest value among all speed limit sources; A flag indicating whether there are two or more valid speed limit rules with different values ​​that conflict in a speed limit scenario; Driver style coefficients characterizing a driver's preference for efficiency and comfort; The positional correlation features characterizing the distance of a vehicle from the next speed limit change point.

4. The method according to claim 3, characterized in that, The decision tree is a state-specific decision tree that corresponds one-to-one with the speed limit scenario state. Each of the state-specific decision trees is pre-configured as follows: a lightweight tree model generated through independent training, using driving data collected under the same speed limit scenario as training samples and discrete speed values ​​formed by discretizing the actual driving speed under that scenario into an integer multiple of the set speed.

5. The method according to claim 4, characterized in that, The lightweight tree model corresponding to the state-specific decision tree is obtained during training by applying depth pruning and feature clipping constraints. The state-specific decision tree is stored in the storage medium using an implicit index structure.

6. The method according to claim 1, characterized in that, The multi-source information includes the current instantaneous maximum constraint speed limit; The step of inputting the multi-source information into the decision tree and outputting the target vehicle speed includes: Obtain the discretized vehicle speed values ​​stored in the leaf nodes of the decision tree as the initial target vehicle speed; The initial target vehicle speed is compared with the current instantaneous maximum constraint speed limit; In response to the initial target vehicle speed being greater than the instantaneous maximum constraint speed limit, the target vehicle speed is corrected to a legal discrete vehicle speed value that is not greater than the instantaneous maximum constraint speed limit, and the corrected vehicle speed is used as the final output target vehicle speed.

7. The method according to any one of claims 1 to 6, characterized in that, The method further includes: Receive incremental data packets sent via vehicle-mounted wireless communication. The incremental data packets contain a new speed limit scenario state, the corresponding state transition conditions, and a new state-specific decision tree corresponding to the new speed limit scenario state. Without restarting the main program, the state machine logic and decision tree model library are dynamically updated to support the recognition of the newly added speed limit scenario state and vehicle speed decision.

8. A speed control device based on speed limit rules, characterized in that, include: The determination module is used to acquire multi-source information characterizing the vehicle's driving environment and, based on the multi-source information, determine the current speed limit scenario state of the vehicle. The acquisition module is used to determine the pre-trained decision tree corresponding to the speed limit scenario state; The analysis module is used to input the multi-source information into the decision tree and output the target speed of the vehicle. The execution module is used to control the vehicle speed according to the target vehicle speed.

9. A control device, characterized in that, include: Memory, processor; The memory stores computer-executed instructions; The processor executes computer execution instructions stored in the memory, causing the processor to perform the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, are used to implement the method as described in any one of claims 1 to 7.

Images