Torque adjusting method and vehicle

Abstract

The invention discloses a torque adjusting method and a vehicle, and belongs to the technical field of vehicle power systems. Comprising the steps of determining whether a target vehicle is in a slip working condition or not based on a difference between an original feature vector and a reconstructed feature vector corresponding to a multi-dimensional state parameter of the target vehicle at a current moment; under the condition that the target vehicle is in the slipping working condition, the current slipping type of the target vehicle is determined based on the difference between the original feature vector and the reconstructed feature vector and the parameter change information of the multi-dimensional state parameter at the multiple moments; and adjusting the output torque of the target vehicle based on the current slip type, the M target state parameters and the target torque of the whole vehicle at the current moment. Thus, the multi-dimensional state parameters of the vehicle are synthesized to determine whether the vehicle is in the slip working condition and determine the current slip type, accurate determination of the slip working condition and the slip type can be achieved, accurate adjustment of the output torque can be achieved subsequently, and the vehicle can break away from the slip working condition more quickly.

Description

Technical Field

[0001] This application relates to the field of vehicle powertrain technology, and in particular to a torque adjustment method and a vehicle. Background Technology

[0002] When vehicles travel on wet or slippery surfaces with low traction, wheel slippage is a common occurrence. Currently, in driving scenarios, accurately identifying wheel slippage and taking timely corrective action is crucial.

[0003] Generally, when a vehicle slips, there will be significant differences in the wheel speeds. Related technologies determine whether the vehicle is slipping by acquiring wheel speeds at multiple moments, calculating the changes in wheel speeds at these moments, and analyzing the differences between the four wheel speeds. If the vehicle is slipping, appropriate measures are then taken on the slipping wheels to help the vehicle recover from the slippage.

[0004] However, the above method determines whether a vehicle is slipping by using a single-dimensional wheel speed signal. In complex road scenarios, the road environment is complex and ever-changing. Relying solely on wheel speed may not be able to accurately determine the vehicle's slipping condition. Consequently, when the vehicle is slipping, it is impossible to achieve precise control of the vehicle to help it get out of the slipping condition as quickly as possible. Summary of the Invention

[0005] This application provides a torque adjustment method, device, vehicle, and storage medium. It can comprehensively determine whether a vehicle is in a slipping condition by integrating multi-dimensional state parameters. Furthermore, when the vehicle is in a slipping condition, it determines the current slip type by comprehensively considering the severity of the slip and the changes in the multi-dimensional state parameters. This allows for accurate determination of whether the vehicle is in a slipping condition and accurately identifies the current slip type. Subsequently, based on the current slip type, precise adjustment of the vehicle's output torque can be achieved, enabling the vehicle to recover from the slipping condition more quickly. The technical solution includes the following:

[0006] Firstly, a torque adjustment method is provided, the method comprising: Based on the difference between the original feature vector and the reconstructed feature vector corresponding to the multi-dimensional state parameters of the target vehicle at the current moment, it is determined whether the target vehicle is in a slipping condition. The reconstructed feature vector is obtained by reconstructing the features of the multi-dimensional state parameters. When the target vehicle is in a slipping condition, the current slipping type of the target vehicle is determined based on the difference between the original feature vector and the reconstructed feature vector, and the parameter change information of the multi-dimensional state parameters at multiple times. Based on the current slippage type, M target state parameters, and the target torque of the vehicle at the current moment, the output torque of the target vehicle is adjusted. The target state parameters are state parameters related to the slippage state.

[0007] In this application, feature reconstruction is performed on the multi-dimensional state parameters of the target vehicle at the current moment to obtain the reconstructed feature vectors corresponding to the multi-dimensional state parameters. Normally, there is a certain correlation between the various state parameters, and feature reconstruction is based on this correlation. Therefore, under normal circumstances, the reconstructed feature vector should be consistent with the original feature vector. However, this correlation is broken during slippage. Therefore, the reconstructed feature vector obtained after reconstructing the multi-dimensional state parameters based on this correlation will differ from the original feature vector. Thus, based on the difference between the original feature vector and the reconstructed feature vector of the multi-dimensional state parameters, it is possible to accurately determine whether the target vehicle is in a slippage condition. Then, when the target vehicle is in a slippage condition, based on the difference between the original feature vector and the reconstructed feature vector, and the parameter changes of the multi-dimensional state parameters at multiple times, the current slippage type of the target vehicle is determined. Since the difference between the original feature vector and the reconstructed feature vector can represent the severity of vehicle slippage, for example, the greater the difference, the more severe the slippage, the parameter change information of multi-dimensional state parameters at multiple times can represent the dynamic change trend of the target vehicle. Thus, by combining these two aspects of parameters, the current slippage type can be accurately determined. Subsequently, based on the current slippage type, M target state parameters, and the target torque of the whole vehicle at the current time, the torque of the target vehicle can be precisely adjusted, so that the target vehicle can get out of the slippage condition as soon as possible.

[0008] Optionally, determining whether the target vehicle is in a skidding condition based on the difference between the original feature vector and the reconstructed feature vector includes: Based on the i-th dimension vector of the original feature vector and the i-th dimension vector of the reconstructed feature vector, determine the vector difference of the i-th dimension; Based on the vector differences between the original feature vector and the reconstructed feature vector in multiple dimensions, the comprehensive difference between the original feature vector and the reconstructed feature vector is calculated; If the combined difference between the original feature vector and the reconstructed feature vector is less than or equal to a preset difference threshold, it is determined that the target vehicle is not in a slipping condition. If the combined difference between the original feature vector and the reconstructed feature vector is greater than the preset difference threshold, the target vehicle is determined to be in a slipping condition.

[0009] In the above method, by calculating the dimensional difference between the original feature vector and the reconstructed feature vector, the difference between the original feature vector and the reconstructed feature vector can be measured from different dimensions. This allows for a fine capture of the difference between the two, and a more accurate comprehensive difference can be obtained. Based on the accurate comprehensive difference, it can be determined whether the target vehicle is in a slipping condition.

[0010] Optionally, the method further includes: The multi-dimensional state parameters of the target vehicle at the current moment are compressed and encoded to obtain a compressed feature vector; The compressed feature vector is decoded and reconstructed to obtain the reconstructed feature vector; And, determining the current slip type of the target vehicle based on the difference between the original feature vector and the reconstructed feature vector, and the parameter change information of the multi-dimensional state parameters at multiple times, includes: Based on the compressed feature vector, the difference between the original feature vector and the reconstructed feature vector, and the parameter change information of the multi-dimensional state parameters at multiple times, the current slip type of the target vehicle is determined.

[0011] In the above method, since the compressed feature vector can represent the key information of the multi-dimensional state parameters at the current moment, determining the current slip type of the target vehicle based on the compressed feature vector, the difference between the original feature vector and the reconstructed feature vector, and the parameter change information of the multi-dimensional state parameters at multiple moments is equivalent to determining the current slip type based on the key state information at the current moment, the dynamic change information of the target vehicle at multiple moments, and the degree of abnormality of the current state of the target vehicle. By combining the three pieces of information, a more accurate current slip type can be determined.

[0012] Optionally, determining the current slippage type of the target vehicle based on the compressed feature vector, the difference between the original feature vector and the reconstructed feature vector, and the parameter change information of the multi-dimensional state parameters at multiple times includes: The compressed feature vector, the difference, and the parameter change information are input into the slip type recognition model, and the slip type recognition model outputs multiple predicted slip types and the confidence scores corresponding to the multiple predicted slip types. The predicted slip type with the highest confidence among the multiple predicted slip types is determined as the current slip type.

[0013] In the above method, by deploying the slippage type recognition model in the vehicle-side controller, the slippage type recognition model can identify the slippage type based on the compressed feature vector, the difference between the original feature vector and the reconstructed feature vector, and the parameter change information of the multi-dimensional state parameters at multiple times. This can achieve rapid identification of the current slippage type of the target vehicle, thereby reducing computational latency and improving the response speed of slippage determination.

[0014] Optionally, adjusting the output torque of the target vehicle based on the current slippage type, M target state parameters, and the target torque of the vehicle at the current moment includes: Based on the current slippage type and the M target state parameters, a target allocation ratio is determined, which represents the ideal torque distribution ratio of multiple wheels of the target vehicle. Based on the target torque of the vehicle and the target distribution ratio, the output torque of the target vehicle is adjusted.

[0015] In the above method, since the current slippage type is known, it's equivalent to knowing which wheels of the target vehicle are currently slipping. Therefore, based on the target vehicle's current actual state, the specific adjustments needed can be determined, such as which adjustment strategy to apply to each wheel (which wheel's torque needs to be increased, which wheel's torque needs to be decreased, etc.), which is subsequently expressed as a torque distribution ratio. Furthermore, the M target state parameters can represent the actual state characteristics of the target vehicle that indicate slippage, such as the adhesion between each wheel and the torque of each wheel. Based on this, combined with the currently slipping wheel, the direction of adjustment can be determined; for example, if the torque of the slipping wheel is too high, it needs to be reduced; if the torque of the non-slipping wheels is too low, it needs to be increased. In this case, based on the current slippage type and the M target state parameters, the torque distribution ratio that can remove the target vehicle from the slippage condition can be determined, thus enabling the target vehicle to escape the slippage condition as quickly as possible.

[0016] Optionally, determining the target allocation ratio based on the current slippage type and the M target state parameters includes: The current slip type and the M target state parameters are fuzzed to obtain multiple first membership degrees and M groups of second membership degrees. The multiple first membership degrees represent the probability that the current slip type belongs to multiple slip levels. The M groups of second membership degrees are the membership degrees corresponding to the M target state parameters. Each group of second membership degrees in the M groups includes multiple second membership degrees. Based on the multiple first membership degrees and the M groups of second membership degrees, N reference membership degrees and N groups of reference levels are determined from the fuzzy rule base. Each of the N groups of reference levels includes multiple allocation ratio levels, which are used to indicate the possible torque distribution ratio of multiple wheels. The N reference membership degrees are used to represent the confidence level of the N groups of reference levels. The target allocation ratio is determined based on the N reference membership degrees and the N sets of reference levels.

[0017] The above method is equivalent to performing fuzzy reasoning based on different first membership degrees and different second membership degrees to obtain a set of reference levels and reference membership degrees corresponding to the first membership degree and the second membership degree. In this way, by reasoning based on different vehicle states and different slipping wheels, the reference membership degrees and reference levels corresponding to slipping wheels at different times and in different vehicle states can be obtained. Therefore, by combining the reference membership degrees and reference levels corresponding to slipping wheels at different times and in different vehicle states, the target allocation ratio can be accurately determined.

[0018] Optionally, determining the target allocation ratio based on the N reference membership degrees and the N sets of reference levels includes: The maximum reference membership degree among the N reference membership degrees is determined as the target membership degree. The set of reference levels corresponding to the target membership degree is determined as the target level; Obtain multiple ratio ranges corresponding to multiple allocation ratio levels in the target level; The target allocation ratio is determined based on the multiple ratio ranges.

[0019] The above method is equivalent to defuzzifying by combining N reference membership degrees and N sets of reference levels, which can yield a clear torque distribution ratio. Then, torque adjustment is performed based on the target distribution ratio, thus achieving precise adjustment of the target vehicle's torque, which in turn allows the target vehicle to get out of the slippage condition as quickly as possible.

[0020] Optionally, adjusting the output torque of the target vehicle based on the target torque of the whole vehicle and the target distribution ratio includes: According to the target allocation ratio, the target torque of the whole vehicle is divided into multiple ideal torques, and the multiple ideal torques are used to represent the torque that should be allocated to multiple wheels; Based on the multiple ideal torques, determine the target torque for the front axle and the target torque for the rear axle; Based on the target torque of the front axle and the target torque of the rear axle, the torque of multiple wheels of the target vehicle is adjusted.

[0021] In the above method, since the target allocation ratio is the torque allocation ratio of multiple wheels that can enable the target vehicle to get out of the slipping condition, the torque (multiple ideal torques) allocated according to the target allocation ratio is the torque that should be allocated to multiple wheels that can enable the target vehicle to get out of the slipping condition. In other words, by adjusting the torque of multiple wheels to these multiple ideal torques, the target vehicle can get out of the slipping condition.

[0022] Optionally, adjusting the torque of multiple wheels of the target vehicle based on the target torque of the front axle and the target torque of the rear axle includes: Subtract the actual torque of the front axle from the target torque of the front axle to obtain the torque difference of the front axle; Subtract the actual torque of the rear axle from the target torque of the rear axle to obtain the torque difference of the rear axle; Based on the front axle torque difference, the front axle motor is controlled to adjust the front axle output torque, and the front axle output torque is distributed to the front wheels according to the front wheel torque distribution ratio in the target distribution ratio. Based on the rear axle torque difference, the rear axle motor is controlled to adjust the rear axle output torque, and the rear axle output torque is distributed to the rear wheels according to the rear wheel torque distribution ratio in the target distribution ratio.

[0023] Secondly, a torque adjustment device is provided, the device comprising: The first determining module is used to determine whether the target vehicle is in a slipping condition based on the difference between the original feature vector and the reconstructed feature vector corresponding to the multi-dimensional state parameters of the target vehicle at the current moment. The reconstructed feature vector is obtained by reconstructing the features of the multi-dimensional state parameters. The second determining module is used to determine the current slippage type of the target vehicle based on the difference between the original feature vector and the reconstructed feature vector, and the parameter change information of the multi-dimensional state parameters at multiple times when the target vehicle is in a slippage condition. The torque adjustment module is used to adjust the output torque of the target vehicle based on the current slippage type, M target state parameters, and the target torque of the vehicle at the current moment. The target state parameters are state parameters related to the slippage state.

[0024] Optionally, the first determining module is used to: Based on the i-th dimension vector of the original feature vector and the i-th dimension vector of the reconstructed feature vector, determine the vector difference of the i-th dimension; Based on the vector differences between the original feature vector and the reconstructed feature vector in multiple dimensions, the comprehensive difference between the original feature vector and the reconstructed feature vector is calculated; If the combined difference between the original feature vector and the reconstructed feature vector is less than or equal to a preset difference threshold, it is determined that the target vehicle is not in a slipping condition. If the combined difference between the original feature vector and the reconstructed feature vector is greater than the preset difference threshold, the target vehicle is determined to be in a slipping condition.

[0025] Optionally, the device further includes: The compression encoding module is used to compress and encode the multi-dimensional state parameters of the target vehicle at the current moment to obtain a compressed feature vector. A decoding and reconstruction module is used to decode and reconstruct the compressed feature vector to obtain the reconstructed feature vector; And, the second determining module is used for: Based on the compressed feature vector, the difference between the original feature vector and the reconstructed feature vector, and the parameter change information of the multi-dimensional state parameters at multiple times, the current slip type of the target vehicle is determined.

[0026] Optionally, the second determining module is specifically used for: The compressed feature vector, the difference, and the parameter change information are input into the slip type recognition model, and the slip type recognition model outputs multiple predicted slip types and the confidence scores corresponding to the multiple predicted slip types. The predicted slip type with the highest confidence among the multiple predicted slip types is determined as the current slip type.

[0027] Optionally, the torque adjustment module is used for: Based on the current slippage type and the M target state parameters, a target allocation ratio is determined, which represents the ideal torque distribution ratio of multiple wheels of the target vehicle. Based on the target torque of the vehicle and the target distribution ratio, the output torque of the target vehicle is adjusted.

[0028] Optionally, the torque adjustment module is used for: The current slip type and the M target state parameters are fuzzed to obtain multiple first membership degrees and M groups of second membership degrees. The multiple first membership degrees represent the probability that the current slip type belongs to multiple slip levels. The M groups of second membership degrees are the membership degrees corresponding to the M target state parameters. Each group of second membership degrees in the M groups includes multiple second membership degrees. Based on the multiple first membership degrees and the M groups of second membership degrees, N reference membership degrees and N groups of reference levels are determined from the fuzzy rule base. Each of the N groups of reference levels includes multiple allocation ratio levels, which are used to indicate the possible torque distribution ratio of multiple wheels. The N reference membership degrees are used to represent the confidence level of the N groups of reference levels. The target allocation ratio is determined based on the N reference membership degrees and the N sets of reference levels.

[0029] Optionally, the torque adjustment module is used for: The maximum reference membership degree among the N reference membership degrees is determined as the target membership degree. The set of reference levels corresponding to the target membership degree is determined as the target level; Obtain multiple ratio ranges corresponding to multiple allocation ratio levels in the target level; The target allocation ratio is determined based on the multiple ratio ranges.

[0030] Optionally, the torque adjustment module is used for: According to the target allocation ratio, the target torque of the whole vehicle is divided into multiple ideal torques, and the multiple ideal torques are used to represent the torque that should be allocated to multiple wheels; Based on the multiple ideal torques, determine the target torque for the front axle and the target torque for the rear axle; Based on the target torque of the front axle and the target torque of the rear axle, the torque of multiple wheels of the target vehicle is adjusted.

[0031] Optionally, the torque adjustment module is used for: Subtract the actual torque of the front axle from the target torque of the front axle to obtain the torque difference of the front axle; Subtract the actual torque of the rear axle from the target torque of the rear axle to obtain the torque difference of the rear axle; Based on the front axle torque difference, the front axle motor is controlled to adjust the front axle output torque, and the front axle output torque is distributed to the front wheels according to the front wheel torque distribution ratio in the target distribution ratio. Based on the rear axle torque difference, the rear axle motor is controlled to adjust the rear axle output torque, and the rear axle output torque is distributed to the rear wheels according to the rear wheel torque distribution ratio in the target distribution ratio.

[0032] Thirdly, a vehicle is provided, the vehicle comprising: Memory, used to store executable program code; A processor is configured to call and run the executable program code from the memory, causing the vehicle to perform the torque adjustment method described above.

[0033] Fourthly, a computer-readable storage medium is provided, the computer-readable storage medium storing a computer program that, when executed by a processor, implements the torque adjustment method described above.

[0034] Fifthly, a computer program product containing instructions is provided, which, when run on a computer, causes the computer to perform the steps of the torque adjustment method described above.

[0035] It is understood that the beneficial effects of the second, third, fourth, and fifth aspects mentioned above can be found in the relevant descriptions in the first aspect above, and will not be repeated here. Attached Figure Description

[0036] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0037] Figure 1 This is a schematic diagram of a scenario for a torque adjustment method provided in an embodiment of this application; Figure 2 This is a schematic diagram illustrating the implementation environment of a torque adjustment method provided in an embodiment of this application; Figure 3 This is a flowchart of a torque adjustment method provided in an embodiment of this application; Figure 4 This is a flowchart of another torque adjustment method provided in the embodiments of this application; Figure 5 This is a schematic diagram of the structure of a torque adjustment device provided in an embodiment of this application; Figure 6 This is a schematic diagram of the structure of a vehicle provided in an embodiment of this application. Detailed Implementation

[0038] To make the objectives, technical solutions, and advantages of this application clearer, the embodiments of this application will be described in further detail below with reference to the accompanying drawings.

[0039] It should be understood that "multiple" as mentioned in this application refers to two or more. In the description of this application, unless otherwise stated, " / " indicates "or," for example, A / B can mean A or B; "and / or" in this document is merely a description of the relationship between related objects, indicating that three relationships can exist, for example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Furthermore, to facilitate a clear description of the technical solutions of this application, the terms "first," "second," etc., are used to distinguish identical or similar items with essentially the same function and effect. Those skilled in the art will understand that the terms "first," "second," etc., do not limit the quantity or execution order, and that "first," "second," etc., do not necessarily imply differences.

[0040] First, the terms used in the embodiments of this application will be explained.

[0041] 1. Feature Reconstruction: The essence of feature reconstruction is to compress and reconstruct the input data through an encoder-decoder structure, and finally generate reconstructed features that are highly similar to the original input.

[0042] The main workflow of feature reconstruction includes: first, compressing and encoding the original high-dimensional features to extract key information and outputting a low-dimensional latent feature vector. Then, mapping the low-dimensional latent feature vector back to the original space and reconstructing features with the same structure as the original high-dimensional features to obtain the reconstructed feature vector.

[0043] 2. Fuzzy Rules: Fuzzy rules can describe the relationship between input and output variables. Fuzzy rules are generally formulated in the form of "IF <condition> THEN <conclusion>". For example, IF high temperature AND moderate humidity THEN air conditioner set to cooling mode.

[0044] 3. Membership function: This is a mathematical tool used to characterize fuzzy sets. It describes the membership relationship between an element and a fuzzy set. Generally, the value of the membership function of an element can take the range [0,1]. Different values ​​can represent the degree to which an element truly belongs to a certain fuzzy set.

[0045] For example, fat people belong to a fuzzy set. The membership function value of a person weighing 80 kilograms belonging to the fat people can be 0.9, and the membership function value of a person weighing 70 kilograms belonging to the fat people can be 0.8.

[0046] 4. Fuzzy Reasoning: Fuzzy reasoning is used to process imprecise and uncertain information through a series of fuzzy rules and membership functions, and to draw reasonable conclusions.

[0047] Fuzzy reasoning can be divided into four steps: fuzzification, fuzzy rule formulation, reasoning methods, and defuzzification. Fuzzification is the process of obtaining the membership degree of an element to various fuzzy sets based on a specific input element using membership functions. Fuzzy rule formulation can be fuzzy rules pre-formulated by technicians based on experience. Reasoning methods are methods for obtaining fuzzy conclusions based on the formulated fuzzy rules and the membership degrees of the input element to relevant fuzzy sets. Defuzzification is the process of transforming the fuzzy conclusions into specific, precise outputs.

[0048] Before describing the torque adjustment method provided in the embodiments of this application, the application scenarios of the embodiments of this application will be explained first.

[0049] For example, Figure 1 This is a schematic diagram illustrating a scenario of a torque adjustment method provided in an embodiment of this application. See also... Figure 1 , Figure 1 This includes vehicle 101.

[0050] For example, when vehicle 101 is driving on a road, such as on a muddy road, the wheels of vehicle 101 are prone to slipping on such a low-traction surface.

[0051] In related technologies, the presence of wheel slippage in vehicle 101 is determined by acquiring the wheel speeds of vehicle 101 at multiple times, calculating the changes in wheel speeds at those multiple times, and the differences between wheel speeds. In some cases, the slipping wheel can also be determined based on the wheel speed, and corresponding measures can be taken for the corresponding wheel, such as braking the slipping wheel or increasing torque on the non-slipping wheel, to help the vehicle get out of the slipping condition as soon as possible.

[0052] However, in some complex road environments, such as when multiple types of road surfaces appear alternately, the wheel slippage conditions are also more complex. Based solely on the wheel speeds at the aforementioned multiple moments, it may not be possible to accurately determine whether vehicle 101 is in a slippage condition, nor can it accurately determine which wheel of vehicle 101 is slipping. Consequently, it is impossible to accurately control the vehicle subsequently, which will increase the slippage time of vehicle 101 and prevent vehicle 101 from quickly getting out of trouble.

[0053] Therefore, this application provides a torque adjustment method that can be applied to the driving scenarios of the aforementioned vehicles. This torque adjustment method can determine whether a target vehicle is in a slipping condition based on the difference between the original feature vector and the reconstructed feature vector of multi-dimensional state parameters. Then, if the target vehicle is in a slipping condition, the current slipping type of the target vehicle can be determined based on the difference between the original and reconstructed feature vectors and the parameter changes of the multi-dimensional state parameters at multiple times. Subsequently, corresponding measures can be taken based on the current slipping type to restore the vehicle to normal operating conditions.

[0054] Under normal circumstances, there are certain correlations between the various state parameters of a vehicle. Feature reconstruction is based on these correlations to reconstruct the multi-dimensional state parameters. Normally, the reconstructed feature vector should be consistent with the original feature vector. However, under abnormal circumstances, these correlations are broken. Therefore, the reconstructed feature vector obtained after reconstructing the multi-dimensional state parameters based on these correlations will differ from the original feature vector. Thus, based on the difference between the original and reconstructed feature vectors of the multi-dimensional state parameters, it is possible to accurately determine whether the target vehicle is in a slipping condition. Furthermore, since the difference between the original and reconstructed feature vectors can represent the degree of deviation of the vehicle's various state parameters (and thus reflect the degree of slippage), and the parameter changes of the multi-dimensional state parameters at multiple times can represent the dynamic change trend of the target vehicle, combining these two aspects of parameters can accurately determine the current slippage type. Subsequently, based on the current slippage type, M target state parameters, and the current target torque of the vehicle, precise adjustments can be made to the target vehicle's torque, allowing the target vehicle to quickly escape the slippage condition.

[0055] The implementation environment involved in the embodiments of this application will be described below.

[0056] For example, Figure 2 This is a schematic diagram illustrating the implementation environment of a torque adjustment method provided in this application embodiment. See also... Figure 2 , Figure 2 This includes vehicle 201 and cloud 202.

[0057] The vehicle 201 includes multiple signal sources, which provide state parameters of the vehicle 201 in multiple dimensions. These signal sources may include various sensors, functional modules, and controllers. In this embodiment, the multiple signal sources may include four wheel speed sensors, two motor controllers (front axle motor controller and rear axle motor controller), a brake pressure sensor, a pedal displacement sensor, a lateral acceleration sensor, a longitudinal acceleration sensor, a gyroscope, a battery voltage module, a battery current module, a vehicle height sensor, a GPS (Global Positioning System) module, a temperature sensor, a humidity sensor, an atmospheric pressure sensor, and a high-precision adhesion sensor, etc.

[0058] Vehicle 201 also includes a vehicle central gateway, a data acquisition device, and a domain controller. In this embodiment, the multiple signal sources can be connected to the vehicle central gateway, and the data acquisition device can be connected between the vehicle central gateway and the domain controller. It is used to acquire multi-dimensional status data in parallel through a high-speed CAN (Controller Area Network) bus, an Ethernet bus, and a LIN (Local Interconnect Network) bus, and send the multi-dimensional status parameters to the domain controller, so that the domain controller can determine the slip condition, the slip type, and generate a torque adjustment strategy based on the multi-dimensional status parameters.

[0059] In some embodiments, vehicle 201 can communicate with cloud 202 via a wireless connection. In this case, vehicle 201 can package and send the multi-dimensional state data collected in each detection cycle, the judgment of slippage condition, the judgment of slippage type, and the torque adjustment strategy to cloud 202, so that cloud 202 can store the slippage identification process of each detection cycle as feedback data.

[0060] The torque adjustment method provided in the embodiments of this application will be explained in detail below.

[0061] Figure 3 This is a flowchart illustrating a torque adjustment method provided in an embodiment of this application. This method can be applied to a vehicle's domain controller. See also... Figure 3 The method includes the following steps 301-303.

[0062] Step 301: Based on the difference between the original feature vector and the reconstructed feature vector corresponding to the multi-dimensional state parameters of the target vehicle at the current moment, determine whether the target vehicle is in a slipping condition. The reconstructed feature vector is obtained by reconstructing the multi-dimensional state parameters.

[0063] This multi-dimensional state parameter is used to represent the instantaneous state of the target vehicle in multiple dimensions, and it can represent the complete motion and control state of the target vehicle at a certain moment. This includes multi-dimensional state data such as the vehicle's longitudinal state, lateral state, electric drive system state, and the state of the environment in which the target vehicle is located. In this embodiment, the multi-dimensional state parameter can be derived from the above... Figure 1 The data collected from multiple signal sources described in the embodiments, such as the multi-dimensional state parameters, may include wheel speeds of multiple wheels of the target vehicle, GPS vehicle speed, lateral acceleration, longitudinal acceleration, angular velocity, brake pedal opening, accelerator pedal opening, target torque of the front and rear axle motors, actual output torque of the front and rear axle motors, motor speed and temperature, vehicle height, voltage and current data of the power system, attitude parameters of the target vehicle (such as pitch angle, roll angle, yaw angle), ambient temperature, humidity, atmospheric pressure, current road type, and ground adhesion coefficient, among other parameters. In some embodiments, vehicle parameters, such as vehicle identification, vehicle configuration, drive mode, vehicle load, and tire type, may also be mixed into the multi-dimensional state data.

[0064] In some embodiments, a detection period can be preset, for example, to 100ms. That is, throughout the entire process from the moment the target vehicle is powered on until it is powered off, multi-dimensional state parameters can be collected in each detection period. This allows for the determination of the target vehicle's slippage condition based on these multi-dimensional state parameters at regular intervals. Therefore, in the current detection period, multi-dimensional state parameters at the current moment can be collected to determine whether the target vehicle is currently in a slippage condition.

[0065] The original feature vector is the feature vector obtained by directly extracting features from the original multi-dimensional state parameters, and it is used to represent the features of the original multi-dimensional state parameters.

[0066] Reconstructing a feature vector refers to restoring and reconstructing the features of the reconstructed multidimensional state parameters based on the key information of the original multidimensional state parameters.

[0067] Under normal circumstances, there are certain correlations between the various state parameters of a target vehicle. Feature reconstruction is based on these correlations to restore the key information of the multi-dimensional state parameters. Therefore, under normal circumstances, the reconstructed feature vector should be close to the original feature vector. However, this correlation is broken under slippage conditions. For example, when the target vehicle slips, there are significant inconsistencies between the acceleration, wheel speed, motor response, and overall vehicle motion state of some wheels. This leads to an imbalance in the correlation between various signals. Consequently, the reconstructed feature vector obtained after reconstructing the key information of the multi-dimensional state parameters based on normal correlations will differ from the original feature vector (because there are no normal correlations between the original multi-dimensional state parameters). Thus, by using the original feature vector and the reconstructed feature vector corresponding to these multi-dimensional state parameters, it is possible to accurately determine whether the target vehicle is in a slippage condition.

[0068] It should be understood that feature reconstruction maps features to a feature space with the same dimensions as the input data. Therefore, the reconstructed feature vector has the same dimensions as the original feature vector.

[0069] In this case, step 301 can be performed as follows: based on the i-th dimension of the original feature vector and the i-th dimension of the reconstructed feature vector, determine the vector difference in the i-th dimension; based on the vector differences between the original feature vector and the reconstructed feature vector in multiple dimensions, calculate the comprehensive difference between the original feature vector and the reconstructed feature vector; if the comprehensive difference between the original feature vector and the reconstructed feature vector is less than or equal to a preset difference threshold, determine that the target vehicle is not in a slipping condition; if the comprehensive difference between the original feature vector and the reconstructed feature vector is greater than a preset difference threshold, determine that the target vehicle is in a slipping condition.

[0070] In the above steps, for each dimension of the original feature vector and the reconstructed feature vector, the difference between the original feature vector and the reconstructed feature vector is calculated for each dimension, which is equivalent to calculating the dimension-wise difference. Then, the dimension-wise difference of each dimension is combined to obtain the overall difference between the original feature vector and the reconstructed feature vector. Finally, the overall difference is measured. If the overall difference is less than or equal to the preset difference threshold, it means that the difference between the original feature vector and the reconstructed feature vector is small, that is, the original feature vector and the reconstructed feature vector are relatively close. This indicates that the correlation between the original multi-dimensional state parameters is consistent with the correlation between the reconstructed multi-dimensional state parameters, which means that the target vehicle is currently driving normally and has not slipped.

[0071] If the overall difference is greater than the preset difference threshold, it indicates that the difference between the original feature vector and the reconstructed feature vector is large. This means that the correlation between the original multi-dimensional state parameters is inconsistent with the correlation between the reconstructed multi-dimensional state parameters. In other words, the correlation between the multi-dimensional state parameters has been broken, indicating that the target vehicle is currently in a slipping condition.

[0072] In the above method, by calculating the dimensional difference between the original feature vector and the reconstructed feature vector, the difference between the original feature vector and the reconstructed feature vector can be measured from different dimensions. This allows for a fine capture of the difference between the two, and a more accurate comprehensive difference can be obtained. Based on the accurate comprehensive difference, it can be determined whether the target vehicle is in a slipping condition.

[0073] It is worth noting that, before step 301, the following steps can also be performed to obtain the reconstructed feature vector corresponding to the multidimensional state parameter.

[0074] One possible approach is to compress and encode the multi-dimensional state parameters of the target vehicle at the current moment to obtain a compressed feature vector; then decode and reconstruct the compressed feature vector to obtain the reconstructed feature vector.

[0075] In the above steps, by compressing and encoding the multi-dimensional state parameters of the target vehicle at the current moment, the resulting compressed feature vector can represent the key information of the multi-dimensional state parameters. Then, by decoding and restoring the key information of the multi-dimensional state parameters, the reconstructed feature vector corresponding to the multi-dimensional state parameters can be obtained.

[0076] One possible approach is to perform the above steps using a feature reconstruction model to obtain the reconstructed feature vector. This feature reconstruction model can be an autoencoder consisting of an encoder-decoder structure, and can include both an encoder and a decoder.

[0077] In this case, the multi-dimensional state parameters of the target vehicle at the current moment can be compressed and encoded by the encoder to obtain a compressed feature vector; the compressed feature vector can then be decoded and reconstructed by the decoder to obtain the reconstructed feature vector.

[0078] It is worth noting that the feature reconstruction model can be trained before performing the above operations.

[0079] One possible approach is for the server to obtain a first dataset, which includes multiple first training samples, and use these multiple first training samples to train the neural network model to obtain the feature reconstruction model.

[0080] These multiple first training samples can be pre-set. Each of these multiple first training samples can be multi-dimensional state data of the vehicle during normal driving.

[0081] This neural network model can include multiple network layers, including an input layer, multiple hidden layers, and an output layer. The input layer is responsible for receiving input data; the output layer is responsible for outputting the processed data, that is, the reconstructed feature vector of the input data; the multiple hidden layers are located between the input and output layers, responsible for processing the data, and are invisible to the outside world. For example, this neural network model can be a neural network composed of an encoder-decoder structure. The model training process can be implemented using unsupervised training methods.

[0082] Specifically, when the server trains the neural network model using multiple first training samples, for each of these first training samples, this first training sample can be used as input data into the neural network model to obtain output data (reconstructed feature vectors). A loss function is then used to determine the loss value between the output data and the original feature vectors corresponding to the input data. The parameters in the neural network model are then adjusted based on this loss value. After adjusting the parameters of the neural network model based on each of these first training samples, the neural network model with adjusted parameters is the feature reconstruction model.

[0083] In some embodiments, the preset difference threshold in the above-mentioned slippage condition judgment process can be set in advance by a technician. In other embodiments, the preset difference threshold can be set based on the difference distribution between the output data learned during the feature reconstruction model training process and the original feature vectors corresponding to the input data.

[0084] It's worth noting that the goal of model training is to minimize the difference between the original feature vectors corresponding to the output and input data. This allows the model to output reconstructed feature vectors that are close to the feature vectors corresponding to the input data, enabling the model to fully learn the correlations between various signals during normal vehicle operation. After training, this feature reconstruction model can reconstruct multi-dimensional state parameters during normal driving, and the reconstructed feature vectors are quite close to the original feature vectors. However, for multi-dimensional state parameters under abnormal conditions, the correlations between these parameters are inconsistent with the correlations between various signals during normal driving. Therefore, the reconstructed feature vectors obtained by this feature reconstruction model from the state data under abnormal conditions differ significantly from the original feature vectors, which can then be used to identify whether the vehicle is skidding.

[0085] It should be understood that when the target vehicle is not in a slipping condition, there is no need to proceed with the subsequent slipping category determination process. Instead, the subsequent slipping category determination and torque adjustment process will only continue when the target vehicle is determined to be in a slipping condition. This avoids the need to determine the slipping category based on data during normal driving, thereby reducing the computational load and improving the overall discrimination efficiency, which in turn forms an effective resource scheduling mechanism.

[0086] When it is determined that the target vehicle is in a slipping condition, the slipping type of the target vehicle can be determined, and corresponding measures can be taken to get the target vehicle out of the slipping condition as soon as possible, so that steps 302 and 303 can be executed.

[0087] Step 302: When the target vehicle is in a slipping condition, the current slipping type of the target vehicle is determined based on the difference between the original feature vector and the reconstructed feature vector, and the parameter change information of the multi-dimensional state parameters at multiple times.

[0088] It should be understood that the difference between the original feature vector and the reconstructed feature vector can indicate the degree of anomalousness of the target vehicle's current state.

[0089] The parameter changes of these multi-dimensional state parameters at multiple moments can represent the dynamic changes of the target vehicle over time, revealing specific slippage information. In some embodiments, multi-dimensional state parameters from multiple moments prior to the current moment can be obtained. These parameters are then compared with the current moment to obtain the parameter changes of each state parameter at multiple moments. Furthermore, the rate of change of these multi-dimensional state parameters at multiple moments can be calculated. For example, for vehicle speed, the rate of change of vehicle speed at multiple moments can be calculated; for motor temperature, the rate of increase / decrease of motor temperature at multiple moments or the amount of change in motor temperature at multiple moments can be calculated. Subsequently, by analyzing the parameter change information at multiple moments, it can be determined which wheel is slipping.

[0090] The slip type indicates which wheel of the target vehicle is currently slipping, and may further indicate the severity of the slippage. In some embodiments, the slip type may include no slippage, left front wheel slippage, right front wheel slippage, left rear wheel slippage, right rear wheel slippage, multiple wheel slippage simultaneously, etc. When indicating the severity of slippage, the slip type may also be categorized as no slippage, slight left front wheel slippage, severe left front wheel slippage, slight right front wheel slippage, severe right front wheel slippage, slight left rear wheel slippage, severe left rear wheel slippage, slight right rear wheel slippage, severe right rear wheel slippage, severe multiple wheel slippage, etc.

[0091] In the above method, the current slip type is determined by the difference between the original feature vector and the reconstructed feature vector and the parameter change information of the multi-dimensional state parameters at multiple times. This is equivalent to determining the slip type based on the degree of abnormality and dynamic change information of the current state of the target vehicle, thereby determining a more accurate and detailed slip type.

[0092] In one possible approach, after compressing and encoding the multi-dimensional state parameters to obtain a compressed feature vector, step 302 can be performed as follows: based on the compressed feature vector, the difference between the original feature vector and the reconstructed feature vector, and the parameter change information of the multi-dimensional state parameters at multiple times, determine the current slippage type of the target vehicle.

[0093] Since compressed feature vectors can represent key information about multi-dimensional state parameters at the current moment, the above method, which determines the current slip type of the target vehicle based on the compressed feature vector, the difference between the original feature vector and the reconstructed feature vector, and the parameter change information of the multi-dimensional state parameters at multiple moments, is equivalent to determining the current slip type based on the key state information at the current moment, the dynamic change information of the target vehicle at multiple moments, and the degree of abnormality of the current state of the target vehicle. By combining the three pieces of information, a more accurate current slip type can be determined.

[0094] One possible approach to determine the current slip type of a target vehicle based on the compressed feature vector, the difference between the original feature vector and the reconstructed feature vector, and the parameter change information of the multi-dimensional state parameters at multiple times can be as follows: input the compressed feature vector, the difference between the original feature vector and the reconstructed feature vector, and the parameter change information of the multi-dimensional state parameters at multiple times into a slip type recognition model; output multiple predicted slip types and their corresponding confidence scores through the slip type recognition model; and determine the slip type with the highest confidence score among the multiple predicted slip types as the current slip type.

[0095] It should be understood that after inputting the compressed feature vector, the difference between the original feature vector and the reconstructed feature vector, and the parameter change information of the multi-dimensional state parameters at multiple times into the slippage type recognition model, the compressed feature vector, the difference between the original feature vector and the reconstructed feature vector, and the parameter change information of the multi-dimensional state parameters at multiple times are normalized to unify the scale of the input data and avoid model bias caused by different scales.

[0096] These multiple predicted slip types are consistent with the set slip types. For example, if 6 slip types are set, then these multiple predicted slip types are these 6 slip types. The confidence level is used to indicate the probability that the current slip type belongs to a certain predicted slip type. The higher the confidence level, the greater the probability that the current slip type belongs to a certain predicted slip type. Therefore, in the above method, it is equivalent to first predicting the probability that the current slip type belongs to each of the predicted slip types through the slip type identification model, and then determining the predicted slip type with the highest probability as the current slip type.

[0097] The slip type recognition model is used for slip classification. In this embodiment, the slip type recognition model can be a support vector machine model.

[0098] In the above method, by deploying the slippage type recognition model in the vehicle-side controller, the slippage type recognition model can identify the slippage type based on the compressed feature vector, the difference between the original feature vector and the reconstructed feature vector, and the parameter change information of the multi-dimensional state parameters at multiple times. This can achieve rapid identification of the current slippage type of the target vehicle, thereby reducing computational latency and improving the response speed of slippage determination.

[0099] Step 303: Adjust the output torque of the target vehicle based on the current slippage type, M target state parameters, and the target torque of the vehicle at the current moment.

[0100] Where M is an integer greater than or equal to 1.

[0101] Target state parameters refer to parameters related to vehicle slippage among multi-dimensional state parameters. Specifically, they can be parameters that directly reflect the slippage state and can be calculated based on the multi-dimensional state parameters at the current moment. In this embodiment, some of the M target state parameters are derived from the multi-dimensional state parameters, while others are calculated based on them (e.g., wheel speed differences between wheels, adhesion differences between wheels). As an example, the M target state parameters may include the current torque value of each wheel, the current vehicle speed, and the adhesion differences between wheels (calculated based on the adhesion of each wheel). In some embodiments, they may also include the motor load rate of the front and rear axle motors, the current temperature, the remaining power of the power battery, and the driver's current acceleration intention level (quantified by the accelerator pedal opening).

[0102] In the above method, by using the current slippage type, M target state parameters and the current target torque of the vehicle, it is equivalent to knowing the slipping wheels of the target vehicle. Then, based on the current actual state of the target vehicle, the corresponding torque adjustment strategy is determined. Finally, based on the target torque of the vehicle, the corresponding torque adjustment strategy is executed, which can make the target vehicle gradually get out of the slippage condition.

[0103] One possible approach is that the operation of step 303 may include the following steps (1)-(2).

[0104] (1) Determine the target allocation ratio based on the current slippage type and M target state parameters.

[0105] The target distribution ratio is used to represent the ideal torque distribution ratio of multiple wheels of a target vehicle. In the embodiments of this application, the target distribution ratio refers to the torque distribution ratio of multiple wheels that enables the target vehicle to get out of the slipping condition.

[0106] Since the current slip type is known, which is equivalent to knowing which wheels of the target vehicle are currently slipping, we can determine how to make specific adjustments based on the current actual state of the target vehicle. For example, we can determine what adjustment strategies to implement for each wheel (which wheel's torque needs to be increased, which wheel's torque needs to be decreased, etc.), which will then be expressed in the form of torque distribution ratios.

[0107] In the above method, the M target state parameters can represent the actual state characteristics of the target vehicle that can exhibit a slippage state, such as the adhesion between each wheel and the torque of each wheel. Based on this, combined with the current slipping wheel, the direction of adjustment can be determined. For example, if the torque of the slipping wheel is too high, the torque needs to be reduced; if the torque of the non-slipping wheel is too low, the torque needs to be increased. In this case, based on the current slippage type and the M target state parameters, the torque distribution ratio that can enable the target vehicle to escape the slippage condition can be determined, thus allowing the target vehicle to escape the slippage condition as quickly as possible.

[0108] One possible approach is to determine the target allocation ratio based on the current slippage type and M target state parameters using fuzzy reasoning. Specifically, the operation of determining the target allocation ratio based on the current slippage type and M target state parameters can include the following steps A-C.

[0109] A. Fuzzyen the current slippage type and M target state parameters respectively to obtain multiple first membership degrees and M groups of second membership degrees.

[0110] It should be understood that in this embodiment, multiple slip levels can be set for slip types. For example, multiple slip levels may include three levels: "low," "medium," and "high." Multiple first membership degrees represent the probability that the current slip type belongs to each of the multiple slip levels. That is, multiple first membership degrees include the probability that the current slip type belongs to the "low" slip level, the probability that the current slip type belongs to the "medium" slip level, and the probability that the current slip type belongs to the "high" slip level. Furthermore, this embodiment also sets multiple membership functions corresponding to the slip levels. These membership functions are the calculation rules for the probability that the current slip type belongs to a certain slip level. Therefore, when calculating the probability that the current slip type belongs to a certain slip level, it can be calculated using the membership function corresponding to that slip level.

[0111] The M groups of second membership degrees represent the membership degrees corresponding to the M target state parameters, meaning one group of second membership degrees corresponds to the membership degree of one target state parameter. In this embodiment, each group of M second membership degrees includes multiple second membership degrees. It should be understood that in this embodiment, multiple state levels can also be set for each of the M target state parameters. For example, the M target state parameters may include the current torque value of each wheel, the current vehicle speed, and the adhesion between each wheel.

[0112] For the target state parameter of the current torque value of each wheel, multiple state levels can be set as multiple torque levels (such as "high", "medium", and "low"). The multiple second membership degrees included in the set of second membership degrees corresponding to the target state parameter of the current torque value of each wheel are used to represent: the probability that the current torque value of each wheel belongs to the "high" torque level, the probability that the current torque value of each wheel belongs to the "medium" torque level, and the probability that the current torque value of each wheel belongs to the "low" torque level. Similarly, multiple membership functions corresponding to multiple torque levels can be set. When calculating the probability that the current torque value belongs to a certain torque level, it can be calculated using the membership function corresponding to that torque level.

[0113] For the target state parameter of current vehicle speed, multiple state levels can be set as multiple speed levels (such as "high", "medium", and "low"). Then, the multiple second membership degrees included in the set of second membership degrees corresponding to the target state parameter of current vehicle speed are used to represent: the probability that the current vehicle speed belongs to the "high" speed level, the probability that the current vehicle speed belongs to the "medium" speed level, and the probability that the current vehicle speed belongs to the "low" speed level, respectively. Similarly, multiple membership functions can be set for different speed levels. When calculating the probability that the current vehicle speed belongs to a certain speed level, it can be calculated using the membership function corresponding to that speed level.

[0114] By performing the above fuzzification on each current target state parameter, we can obtain the probability that each target state parameter belongs to multiple state levels of the corresponding parameter, and thus obtain the M groups of second membership degrees.

[0115] B. Based on multiple first membership degrees and M groups of second membership degrees, determine N reference membership degrees and N groups of reference levels from the fuzzy rule base.

[0116] Where N is an integer greater than or equal to 1.

[0117] Each of the N reference levels includes multiple distribution ratio levels, which are used to indicate the possible torque distribution ratios of multiple wheels. The N reference membership degrees are used to represent the confidence levels of the N reference levels.

[0118] The fuzzy rule base includes multiple fuzzy rules, which can be used by technicians to infer the relationship between slip types, M target state parameters, and torque distribution ratios of multiple wheels. Specifically, it infers the distribution ratio level based on different slip types belonging to different slip levels and each of the M target state parameters corresponding to different state levels. Each distribution ratio level corresponds to a corresponding torque distribution ratio range.

[0119] As an example, the M target state parameters include the current torque value of each wheel, the current vehicle speed, and the adhesion between each wheel. The fuzzy rule base can then include multiple fuzzy rules as follows.

[0120] First, when the slip type is "low" slip level for the left front wheel, the torque values ​​of each wheel are "high", "low", "high", and "low" torque levels respectively, the vehicle speed is "low" speed level, and the adhesion between each wheel is "low", "high", "low", and "high" adhesion levels respectively, then the corresponding distribution ratio levels are "low" for the left front wheel, "high" for the right front wheel, "low" for the left rear wheel, and "high" for the right rear wheel.

[0121] Second, when the slippage type is "medium" slippage level for the left rear wheel, the torque values ​​of each wheel are "small", "small", "large", and "small" torque levels respectively, the vehicle speed is "low" speed level, and the adhesion between each wheel is "high", "high", "low", and "high" adhesion levels respectively, then the corresponding distribution ratio levels are "high" for the left front wheel, "high" for the right front wheel, "low" for the left rear wheel, and "high" for the right rear wheel.

[0122] Third, when the slippage type is a multi-wheel "high" slippage level, the torque values ​​of each wheel are respectively "large", "small", "large", "large" torque levels, the vehicle speed is a "low" speed level, and the adhesion between each wheel is respectively "low", "high", "low", "low" adhesion levels, then the corresponding distribution ratio levels are: left front wheel "low", right front wheel "high", left rear wheel "low", and right rear wheel "low".

[0123] The above fuzzy rules are merely illustrative examples. Specific fuzzy rules are formulated by technicians based on historical experience. The above examples do not constitute a limitation on the embodiments of this application.

[0124] Since the current slippage level is fuzzified into multiple slippage levels, and each of the M target state parameters is also fuzzified into multiple state levels corresponding to the parameter, different slippage levels and different state levels of each target state parameter will correspond to different allocation ratio levels for multiple wheels. Therefore, based on multiple first membership degrees and M groups of second membership degrees, N reference membership degrees and N groups of reference levels can be determined from the fuzzy rule base. The multiple allocation ratio levels included in each of these N groups of reference levels are also the possible allocation ratio levels for multiple wheels inferred under different slippage levels and different state levels of each target state parameter.

[0125] In the above method, the fuzzy rules are formulated by technicians based on experience gained from a large number of road tests and actual control feedback. They take into account multiple factors, as well as the uncertainty and intersection of these factors. Therefore, the fuzzy rules can determine a relatively accurate level of the torque distribution ratio of multiple wheels after fuzzification, thereby allowing for the subsequent determination of a relatively accurate ideal torque distribution ratio of multiple wheels.

[0126] C. Determine the target allocation ratio based on N reference membership degrees and N sets of reference levels.

[0127] Step B above yields the possible wheel allocation ratios for different target state parameters under different slip levels, and the probability (N reference membership degrees) that these possible wheel allocation ratios belong to the ideal allocation ratio.

[0128] In this case, multiple torque distribution ratios can be selected from N reference levels based on N reference membership degrees as the target distribution ratio.

[0129] One possible approach is to determine the target allocation ratio based on N reference membership degrees and N sets of reference levels. This can be done by: obtaining the largest reference membership degree from the N reference membership degrees and determining it as the target membership degree; determining the set of reference levels corresponding to the target membership degree as the target level; obtaining multiple ratio intervals corresponding to multiple allocation ratio levels in the target level; and determining the target allocation ratio based on these multiple ratio intervals.

[0130] It should be understood that in the embodiments of this application, each allocation ratio level corresponds to a corresponding ratio range, and different allocation ratio levels correspond to different ratio ranges. For example, an allocation ratio level of "high" can correspond to a ratio range of [50%, 60%]. An allocation ratio level of "low" can correspond to a ratio range of [10%, 20%], and an allocation ratio level of "medium" can correspond to a ratio range of [21%, 49%].

[0131] Multiple distribution ratio levels are distribution ratio levels corresponding to multiple wheels, and multiple ratio ranges are the possible ranges in which the torque distribution ratios corresponding to multiple wheels may fall.

[0132] The highest reference membership degree, which is the probability that the distribution ratio of multiple wheels in a set of reference levels belongs to the ideal distribution ratio level, is the highest. Therefore, by determining the highest reference membership degree as the target membership degree and the set of reference levels corresponding to the target membership degree as the target level, multiple ratio intervals corresponding to multiple distribution ratio levels in the target level are obtained. Finally, the target distribution ratio determined based on these multiple ratio intervals will be the most ideal torque distribution ratio under the current slippage condition.

[0133] The above method is equivalent to defuzzifying by combining N reference membership degrees and N sets of reference levels, which can yield a clear torque distribution ratio. Then, torque adjustment is performed based on the target distribution ratio, thus achieving precise adjustment of the target vehicle's torque, which in turn allows the target vehicle to get out of the slippage condition as quickly as possible.

[0134] The operation of determining the target distribution ratio based on these multiple ratio ranges can be as follows: for any one of the multiple ratio ranges, the median value in this ratio range can be determined as the torque distribution ratio of the corresponding wheel.

[0135] Furthermore, after determining the torque distribution ratios of multiple wheels, if the sum of the torque distribution ratios of multiple wheels is 1, then the torque distribution ratio of these multiple wheels can be determined as the target distribution ratio. If the sum of the torque distribution ratios of multiple wheels is not 1, it indicates that there is a problem with the torque distribution of these multiple wheels. In this case, the sum of the torque distribution ratios of the multiple wheels needs to be adjusted to 1, and finally, the adjusted torque distribution ratio is determined as the target distribution ratio.

[0136] The above method is equivalent to performing fuzzy reasoning based on different first membership degrees and different second membership degrees to obtain a set of reference levels and reference membership degrees corresponding to the first membership degree and the second membership degree. In this way, by reasoning based on different vehicle states and different slipping wheels, the reference membership degrees and reference levels corresponding to slipping wheels at different times and in different vehicle states can be obtained. Therefore, by combining the reference membership degrees and reference levels corresponding to slipping wheels at different times and in different vehicle states, the target allocation ratio can be accurately determined.

[0137] One possible approach is to implement steps A through C using a fuzzy inference model. Specifically, the fuzzy inference model is used to fuzzify the current slippage type and the M target state parameters, resulting in multiple first membership degrees and M groups of second membership degrees. Based on these first membership degrees and M groups of second membership degrees, N reference membership degrees and N groups of reference levels are inferred. Finally, based on these N reference membership degrees and N groups of reference levels, the target allocation ratio is determined.

[0138] This fuzzy inference model is a neural network model that embeds fuzzy inference processes into a neural network. By setting a rule layer in the neural network model, the rule layer can pre-learn the inference relationship of torque distribution ratios corresponding to different slip types and vehicle states. The output layer can have four channels, each corresponding to the torque distribution ratio of one of the four wheels. That is, this fuzzy inference model receives the current slip type and M target state parameters, performs the aforementioned inference operation in the rule layer, and finally outputs the ideal distribution ratio (target distribution ratio) for each of the four wheels through the four-channel output layer.

[0139] In the above method, by embedding the fuzzy inference process into the neural network, the neural network can learn the fuzzy inference process, thereby learning the inference relationship between different slip types, different vehicle states and torque distribution ratios, and obtaining a fuzzy inference model. When it is necessary to determine the torque adjustment strategy, the fuzzy inference model can receive the slip type and target state parameters, and quickly infer the torque adjustment strategy, thereby enabling rapid control of the target vehicle after slippage, so as to further complete the torque adjustment of individual wheels. The whole process has low response delay and high control accuracy.

[0140] It is worth noting that the torque distribution ratio (target distribution ratio) of multiple wheels of the target vehicle can be re-determined through the above step (1) based on fuzzy reasoning. After the target distribution ratio is determined, the torque of the target vehicle can be adjusted based on the target distribution ratio so that the target vehicle can get out of the slipping condition as much as possible by adjusting the torque distribution of multiple wheels.

[0141] (2) Adjust the output torque of the target vehicle based on the target torque of the whole vehicle and the target distribution ratio.

[0142] One possible approach is to perform step (2) as follows: distribute the target torque of the whole vehicle to multiple ideal torques according to the target distribution ratio; determine the target torque of the front axle and the target torque of the rear axle based on the multiple ideal torques; and adjust the torque of multiple wheels of the target vehicle based on the target torque of the front axle and the target torque of the rear axle.

[0143] These multiple ideal torques are used to represent the torque that should be distributed to multiple wheels, that is, the torque that should be distributed to multiple wheels that can enable the target vehicle to get out of the slipping condition.

[0144] The front axle target torque refers to the torque that the front axle motor should output, while the rear axle target torque refers to the torque that the rear axle motor should output.

[0145] In the above steps, the target torque of the whole vehicle is first distributed to multiple wheels according to the target distribution ratio, and the torque that each wheel may be allocated is calculated. Then, the target torque of the front axle (by adding the torques allocated to the two front wheels) and the target torque of the rear axle (by adding the torques allocated to the two rear wheels) are calculated. Subsequently, the torque of multiple wheels is adjusted based on the calculated target torques of the front axle and the rear axle.

[0146] In the above method, since the target allocation ratio is the torque allocation ratio of multiple wheels that can enable the target vehicle to get out of the slipping condition, the torque (multiple ideal torques) allocated according to the target allocation ratio is the torque that should be allocated to multiple wheels that can enable the target vehicle to get out of the slipping condition. In other words, by adjusting the torque of multiple wheels to these multiple ideal torques, the target vehicle can get out of the slipping condition.

[0147] Furthermore, in this embodiment, the target vehicle includes a front axle motor and a rear axle motor. The front axle motor is used to control the output of the front axle torque, and the rear axle motor is used to control the output of the rear axle torque. Thus, the front axle motor can control the torque of the two front wheels, and the rear axle motor can control the torque of the two rear wheels. Therefore, by adjusting the target torque of the front axle motor (target torque of the front axle) and the target torque of the rear axle motor (target torque of the rear axle), the torque of the four wheels can be adjusted.

[0148] The operation of adjusting the torque of multiple wheels of the target vehicle based on the target torque of the front axle and the target torque of the rear axle can be as follows: subtract the actual torque of the front axle from the target torque of the front axle to obtain the torque difference of the front axle; subtract the actual torque of the rear axle from the target torque of the rear axle to obtain the torque difference of the rear axle; based on the torque difference of the front axle, control the front axle motor to adjust the output torque of the front axle, and distribute the output torque of the front axle to the front wheels according to the front wheel torque distribution ratio in the target distribution ratio; based on the torque difference of the rear axle, control the rear axle motor to adjust the output torque of the rear axle, and distribute the output torque of the rear axle to the rear wheels according to the rear wheel torque distribution ratio in the target distribution ratio.

[0149] The actual torque of the front axle refers to the actual torque output by the front axle motor, and the actual torque of the rear axle refers to the actual torque output by the rear axle motor.

[0150] In the above steps, by subtracting the actual torque of the front axle from the target torque of the front axle, and subtracting the actual torque of the rear axle from the target torque of the rear axle, the difference between the current output torque of the front axle motor and the torque it should output, and the difference between the current output torque of the rear axle motor and the torque it should output, can be obtained. Subsequently, the output torque of the front and rear axle motors is adjusted based on these differences, ensuring that the output torque of the front axle motor reaches the target torque of the front axle, and the output torque of the rear axle motor reaches the target torque of the rear axle. Then, the output torque of the front axle motor is transmitted to the front wheels according to the torque distribution ratio of the two front wheels, ensuring that the two front wheels receive the torque they should receive. Similarly, the output torque of the rear axle motor is transmitted to the rear wheels according to the torque distribution ratio of the two rear wheels, ensuring that the two rear wheels receive the torque they should receive. In other words, differential torque control can be used to adjust the torque of all four wheels, allowing each wheel to quickly recover from slippage after torque adjustment.

[0151] In the above implementation, after calculating the torque difference between the front axle and the rear axle, the domain controller can generate a motor control command. This command instructs the torque adjustment based on the torque difference. The motor control command is then sent in real-time to the front and rear axle motor controllers via the high-voltage control interface. Based on the instructions in the motor control command, the front and rear axle motor controllers control the front and rear axle motors to perform the actual torque output control actions.

[0152] In the above steps (1)-(2), by first determining the target distribution ratio based on the current slip type and M target state parameters, it is equivalent to first determining the torque distribution ratio of multiple wheels that can enable the target vehicle to get out of the slip condition. Then, the output torque is adjusted based on the overall vehicle target torque and this target torque distribution ratio, so that the target vehicle can get out of the slip condition as soon as possible.

[0153] Steps 301-303 above describe the identification and control process of slippage conditions in three stages: identification of slippage conditions of the target vehicle, identification of slippage type, and torque adjustment. Thus, the embodiments of this application can realize the whole process of slippage conditions and improve the processing efficiency of slippage conditions.

[0154] It should be noted that the models or neural networks involved in feature reconstruction, slippage type recognition, and fuzzy inference described in the embodiments of this application can be trained before actual inference.

[0155] During the training process, the first step is to obtain the dataset used for training.

[0156] It should be understood that the sample data included in the dataset should cover influencing factors such as different driving conditions, different road environments, and different driving styles. In the embodiments of this application, the collected data covers no less than one thousand hours of actual road driving, including driving conditions such as congested urban roads, highway cruising roads, mountain roads with curves, icy and slippery roads, and gravel and unpaved roads, and also covers factors such as different weather, temperature, humidity, tire wear, and driving style. The collected data includes parameter signals (multi-dimensional state parameters) provided by multiple signal sources of the vehicle. In some embodiments, it may also include vehicle configuration information (such as vehicle identification, drive mode, tire model, vehicle load, etc.).

[0157] Since the relationships between various state parameters differ among vehicles with different configurations, the model can be trained by adding vehicle configuration information. This allows the model to learn the specific relationships between parameters under slippage conditions for vehicles with different configurations, enabling it to process vehicles with different configurations accordingly and thus improve the accuracy of slippage condition recognition for vehicles with different configurations.

[0158] It should be understood that during the aforementioned data acquisition process, the vehicle's slippage state can be recorded simultaneously. For example, in each detection cycle, the aforementioned vehicle data (parameter signals provided by different signal sources) can be detected. This vehicle data can serve as sample data, and the vehicle's slippage state in the current detection cycle (including whether slippage occurs and the type of slippage) can be recorded as a sample marker for the current frame's sample data. The vehicle's slippage state can be determined in three ways: first, the vehicle automatically triggers slippage recognition (for example, if the wheel speed of a certain wheel increases instantaneously, but the vehicle speed does not change significantly, this can be judged as slippage); second, during data acquisition, the testing personnel judge the slippage state based on subjective driving experience combined with parameter signals from various signal sources; and third, feedback from a high-precision adhesion sensor. A high-precision adhesion sensor can detect the friction between the tire and the road surface in real time, thereby identifying the slippage state based on changes in the friction coefficient between the tire and the road surface.

[0159] It is worth noting that after the data is collected, it needs to be preprocessed first, and then the preprocessed data is processed to form the corresponding training dataset.

[0160] One possible approach is to first perform a signal integrity check after data acquisition, discarding data collected during periods with significant frame drops or sensor malfunctions to ensure the integrity and accuracy of the data acquisition. Next, noise filtering and signal smoothing are applied. During this process, moving averages and wavelet filtering can be used based on empirical rules to eliminate short-term spikes and high-frequency interference. Finally, outlier correction is performed. This involves identifying "single-point jumps" and "extreme points"—data that suddenly deviates from the normal range but whose preceding and following time-series data are normal, and data significantly above or below the normal range that are not interference signals. For "single-point jumps" and "extreme points," reasonable values ​​can be used for backfilling or interpolation to make the data more reasonable, thus ensuring the integrity and rationality of the entire time-series data chain.

[0161] After preprocessing the collected data, the format of all data can be standardized, and the collected data can be divided into multiple training samples according to the temporal relationship. Each training sample includes sample data and sample label. The sample data can be the parameter signals (multi-dimensional state parameters) provided by multiple signal sources in the data collected within a detection period, and the sample label is the slippage state recorded in the same detection period.

[0162] It is worth noting that the slip state includes whether or not slipping occurs. In this case, multiple training samples can be divided into a first dataset and a second dataset based on the sample labels. The first dataset includes training samples with no slipping labels, which are used to train the feature reconstruction model in this application embodiment. The second dataset includes training samples with slipping labels and slipping types, which are used to verify the feature reconstruction model.

[0163] The training process of the feature reconstruction model has been explained in step 301 above, and will not be repeated here. It should be noted that after the feature reconstruction model is trained, it can be validated based on the second dataset.

[0164] The second dataset includes multiple second training samples. Each training sample includes sample data and sample labels. The sample data consists of parameter signals (multi-dimensional state parameters) provided by multiple signal sources within a detection period. The sample labels indicate whether slippage exists in the current detection period and the corresponding slippage type.

[0165] One possible approach is to input any one of the training samples from multiple second training samples into the feature reconstruction model, causing the model to output a reconstructed feature vector. Then, based on the difference between the reconstructed feature vector and the original feature vector of the sample data in the training sample, it can be determined whether the current detection cycle is slipping.

[0166] It should be understood that the training objective of this feature reconstruction model is to minimize the difference between the output data and the original feature vectors of the sample data. Therefore, for sample data corresponding to samples with labels indicating no slippage, this feature reconstruction model can reconstruct a reconstructed feature vector that is close to the original feature vector of that sample data. However, for sample data corresponding to samples with labels indicating slippage, this feature reconstruction model cannot reconstruct a reconstructed feature vector that is close to the original feature vector of that sample data; that is, it can detect vehicle slippage.

[0167] In this case, the feature reconstruction model can output the compressed feature vector obtained by compressing and encoding the sample data, as well as the difference between the reconstructed feature vector and the original feature vector of the sample data, as training data for the slip type recognition model, so as to further train the slip type recognition model's ability to distinguish slip types.

[0168] Furthermore, after the feature reconstruction model is trained, it can be deployed in the domain controller of the target vehicle and run in an independent thread in the domain controller without affecting the logic execution cycle of the main processor. The entire response chain has low latency, thus meeting the real-time operation requirements of the whole vehicle.

[0169] The training process of the slippage type recognition model and the fuzzy inference model is explained below.

[0170] (1) Training the slippage type recognition model The training process may include: training the original support vector machine model based on the third training samples in the third dataset to obtain the slippage type recognition model.

[0171] In this embodiment, the third dataset includes multiple third training samples. Each third training sample includes sample data and sample labels. The sample data may include a compressed feature vector obtained after processing the second training samples, and the difference between the reconstructed feature vector corresponding to the sample data of the second training sample and the original feature vector. Furthermore, when the feature reconstruction model outputs this data (compressed feature vector and the difference), sample data from a time period prior to the time sequence of the sample data in this second training sample can be obtained to calculate the parameter change information of the sample data at multiple times. In this case, the sample data of the third training sample may include a compressed feature vector obtained after processing the second training samples, the difference between the reconstructed feature vector corresponding to the sample data of the second training sample and the original feature vector, and the parameter change information of the sample data of the second training sample at multiple times. The sample label is the slippage type corresponding to the sample data.

[0172] In one possible implementation, for any one of the multiple third training samples, this third training sample is input into the original support vector machine model, and the original support vector machine outputs a predicted slip type. Based on the difference between the predicted slip type and the sample label, the original support vector machine model is trained to obtain the slip type recognition model.

[0173] During training, a third training sample is input into the original support vector machine model. Based on the difference between the predicted slippage type and the sample label, training parameters such as learning rate, batch size, and training epochs can be continuously adjusted to make the predicted slippage type output by the model closer to the sample label. In this embodiment, the model parameters can be optimized using a grid search method.

[0174] In some embodiments, since multi-wheel slippage conditions are relatively rare in reality, the sample data corresponding to multi-wheel slippage conditions that can be collected are also relatively small. In the case of a small amount of sample data corresponding to multi-wheel slippage conditions, data augmentation can be used to generate more multi-wheel slippage samples, thereby further improving the model's recognition accuracy for multi-wheel slippage conditions.

[0175] After training, the skid type recognition model can be evaluated. As an example, cross-validation can be used to assess the model's accuracy and generalization ability. Furthermore, once the model achieves a certain level of accuracy, the model parameters can be fixed as a set of boundary support vectors and kernel function parameters. Subsequently, by directly loading this set of boundary support vectors and kernel function parameters, the skid type recognition model can be deployed on vehicles.

[0176] Furthermore, this slippage type recognition model is deployed after the vehicle's domain controller, and can be invoked immediately whenever a target vehicle is detected to be in a slippage condition. The model has a low latency per invocation and can achieve a slippage type recognition response in milliseconds.

[0177] (2) Training the fuzzy inference model The training process can be as follows: obtain a fourth training set, which includes multiple fourth training samples, and use these multiple fourth training samples to train the neural network model to obtain a fuzzy inference model.

[0178] Each of the multiple fourth training samples includes sample data and sample labels. The sample data may include a slip type and M state parameters related to the slip state. The sample label is the optimal torque distribution ratio corresponding to the sample data, which can be determined based on the torque adjustment process of the slip sample during the slip recovery process.

[0179] The neural network model may include multiple network layers, including an input layer, multiple intermediate hidden layers, and an output layer. The input layer receives input data (including sample data and labels of the fourth training sample); the output layer outputs processed data (torque distribution ratios of multiple wheels); the multiple hidden layers are located between the input and output layers and are responsible for processing the data; these hidden layers are not visible to the outside world. In this embodiment, the intermediate hidden layers are network structures embedding fuzzy inference rules, including rule layers and activation mapping layers.

[0180] When the server trains the neural network model using multiple fourth training samples, for each of these fourth training samples, the input data from that fourth training sample is fed into the neural network model to obtain output data. A loss function is then used to determine the loss value between the output data and the sample labels in that fourth training sample. The parameters of the neural network model are then adjusted based on this loss value. After adjusting the parameters of the neural network model based on each of these fourth training samples, the adjusted neural network model is the fuzzy inference model.

[0181] After training the fuzzy inference model, it can be deployed in the domain controller of the target vehicle. This allows the fuzzy inference model to receive the slippage type and related state parameters in real time, enabling it to quickly respond and control after detecting vehicle slippage, and complete the torque adjustment action for each wheel.

[0182] It is worth noting that after training the feature reconstruction model, slippage type recognition model, and fuzzy inference model separately, the individual models can be jointly optimized to determine the optimal state of information transmission, response order, and parameter coupling among the feature reconstruction model, slippage type recognition model, and fuzzy inference model, thereby ensuring the overall control effect and actual vehicle response capability after the three models are connected in series.

[0183] It should be understood that the joint optimization of various models mainly focuses on output stability and response speed. During the optimization process, confusion matrix analysis can be used to analyze the misclassification rate and missed disc rate of slippage types, paying particular attention to front and rear wheel confusion, multi-wheel misidentification, and recognition delay. The kernel function parameters and label distribution of the slippage type recognition model can also be adjusted to improve the ability to distinguish between minority class samples. Furthermore, for fuzzy inference models, the evaluation process can adjust parameters such as membership function, number of rules, and model weight distribution according to different vehicle models, making the output results more closely match the real-world vehicle control requirements.

[0184] In addition, all models are jointly evaluated using a unified evaluation index system. The evaluation dimensions include multiple data such as torque response time, slippage recovery time, wheel stability, driver's subjective control feeling, and energy efficiency, so as to evaluate the degree of joint optimization of the models from multiple dimensions.

[0185] It is worth noting that information can be exchanged between the various models through a unified data structure, and all features maintain a fixed dimension and format. This avoids communication interruptions or anomalies between models due to inconsistent data formats. Furthermore, each model operates using an asynchronous triggering mechanism. The feature reconstruction model runs continuously, while the slippage type recognition module and fuzzy inference model are triggered only when a slippage condition is detected. This reduces the overall system load and improves processing efficiency.

[0186] It is worth noting that after each model is deployed to the vehicle and the entire vehicle is online, real vehicle data can be collected and uploaded remotely via OTA (Over-The-Air) technology. The cloud can also fine-tune each model based on the OTA real vehicle data to improve slip recognition accuracy and control response efficiency.

[0187] For example, all data related to slippage triggered in the target vehicle, including multi-dimensional state parameters collected in each detection cycle, compressed feature vectors of multi-dimensional state parameters, reconstructed feature vectors, differences between reconstructed and compressed feature vectors, slippage type, torque distribution ratio of multiple wheels output by the fuzzy inference model, final output torque of the front and rear axle motors, and slippage generation time, etc., that is, each slippage processing process is completely recorded to form a data chain. All data on this data chain will be encapsulated into structured data packets in a unified format and uploaded to the cloud periodically through the target vehicle's TBOX or 5G communication module. The data upload cycle can be pre-configured.

[0188] After receiving the data uploaded by the vehicle, the cloud can classify and store all slippage times according to different dimensions (such as vehicle type, road type, weather conditions, tire model, etc.).

[0189] Furthermore, a retraining dataset can be constructed based on the uploaded data, and the retraining dataset can be sent to an offline training platform for a new round of model training. The slippage type recognition model can regenerate the classification boundary support vector machine and kernel function parameters based on the new samples, thereby optimizing the recognition accuracy of the original model in various scenarios.

[0190] After each model completes a new round of training, a new version parameter configuration file can be generated. This file is then uniformly numbered and pushed to the vehicle's terminal by the cloud-based version management system. Upon receiving the file, the vehicle performs an upgrade according to the preset upgrade strategy. After the upgrade is complete, the new model version can be automatically activated, while the old model version is backed up. Manual rollback or multi-version switching and debugging is supported.

[0191] In this embodiment, the entire data feedback mechanism possesses the capabilities of automatic data collection, analysis, training, and distribution. It can complete periodic model optimization without manual intervention, improving system adaptability and maintainability. Simultaneously, it provides a transfer learning foundation for future vehicle models and supports rapid deployment across different platforms. All data generated during the mechanism's operation is encrypted, adheres to vehicle manufacturer data security standards, contains no user privacy information, and is used solely for vehicle dynamic behavior recognition and control strategy optimization. It is not publicly accessible, does not enter user-end systems, and is not displayed to the public.

[0192] To facilitate understanding, we will now combine... Figure 4 The torque adjustment method provided in the embodiments of this application will be described by way of example, for example, Figure 4 This is a flowchart of another torque adjustment method provided in the embodiments of this application.

[0193] like Figure 4As shown, the target vehicle collects multi-dimensional state parameters during the current detection cycle. Then, a feature reconstruction model compresses and reconstructs these parameters, yielding compressed and reconstructed feature vectors. The difference between the reconstructed and original feature vectors is then calculated, and this difference is used to determine whether the target vehicle is in a slipping condition.

[0194] If the target vehicle is not in a slipping condition, no further processing is performed. Instead, the multi-dimensional state parameter collection step is directly executed to continue collecting multi-dimensional state parameters in the next detection cycle and determine whether the target vehicle is in a slipping condition.

[0195] When the target vehicle is in a slipping condition, the slipping type identification model is used to identify the slipping type and obtain the current slipping type of the target vehicle.

[0196] The current slippage type is then input into the fuzzy inference model. Based on the current slippage type and M target state parameters, the fuzzy inference model performs fuzzy inference to obtain the ideal torque distribution ratio for multiple wheels. Subsequently, the torque of the front axle motor and the rear axle motor is adjusted according to the inferred ideal torque distribution ratio for multiple wheels, thereby adjusting the torque of each wheel and enabling the target vehicle to get out of the slippage condition as quickly as possible.

[0197] Furthermore, after the target vehicle recovers from slippage, OTA data feedback can be performed. This allows data generated at each stage—including multi-dimensional state parameter collection, slippage condition identification, slippage type identification, torque distribution ratio inference, and motor torque adjustment—to be packaged and uploaded to the cloud. The cloud can store each slippage data point to accumulate training datasets for subsequent model iterations and enhancement training. Following this, based on the real-vehicle slippage data obtained through OTA feedback, each model can be fine-tuned to improve its processing accuracy and better reflect real-world driving conditions.

[0198] In this embodiment, the domain controller reconstructs the multi-dimensional state parameters of the target vehicle at the current moment to obtain the reconstructed feature vectors corresponding to the multi-dimensional state parameters. Normally, there is a certain correlation between the various state parameters, and feature reconstruction is based on this correlation. Therefore, under normal circumstances, the reconstructed feature vector should be consistent with the original feature vector. However, this correlation is broken during slippage. Therefore, the reconstructed feature vector obtained after reconstructing the multi-dimensional state parameters based on this correlation will differ from the original feature vector. Thus, based on the difference between the original feature vector and the reconstructed feature vector of the multi-dimensional state parameters, it is possible to accurately determine whether the target vehicle is in a slippage condition. Then, when the target vehicle is in a slippage condition, based on the difference between the original feature vector and the reconstructed feature vector, and the parameter changes of the multi-dimensional state parameters at multiple times, the current slippage type of the target vehicle is determined. Since the difference between the original feature vector and the reconstructed feature vector can represent the severity of vehicle slippage, for example, the greater the difference, the more severe the slippage, the parameter change information of multi-dimensional state parameters at multiple times can represent the dynamic change trend of the target vehicle. Thus, by combining these two aspects of parameters, the current slippage type can be accurately determined. Subsequently, based on the current slippage type, M target state parameters, and the target torque of the whole vehicle at the current time, the torque of the target vehicle can be precisely adjusted, so that the target vehicle can get out of the slippage condition as soon as possible.

[0199] Figure 5 This is a schematic diagram of a torque adjustment device provided in an embodiment of this application. The torque adjustment device can be implemented by software, hardware, or a combination of both, and can be part or all of a vehicle, which can be described below. Figure 6 The vehicle shown. See also Figure 5 The device includes: a first determining module 501, a second determining module 502, and a torque adjustment module 503.

[0200] The first determining module 501 is used to determine whether the target vehicle is in a slipping condition based on the difference between the original feature vector and the reconstructed feature vector corresponding to the multi-dimensional state parameters of the target vehicle at the current moment. The reconstructed feature vector is obtained by reconstructing the features of the multi-dimensional state parameters. The second determining module 502 is used to determine the current slip type of the target vehicle based on the difference between the original feature vector and the reconstructed feature vector, and the parameter change information of the multi-dimensional state parameters at multiple times when the target vehicle is in a slip condition. The torque adjustment module 503 is used to adjust the output torque of the target vehicle based on the current slip type, M target state parameters and the target torque of the vehicle at the current moment. The target state parameters are state parameters related to the slip state.

[0201] Optionally, the first determining module 501 is used for: Based on the i-th dimension vector of the original feature vector and the i-th dimension vector of the reconstructed feature vector, determine the vector difference of the i-th dimension; Based on the vector differences between the original feature vector and the reconstructed feature vector in multiple dimensions, the comprehensive difference between the original feature vector and the reconstructed feature vector is calculated. If the overall difference between the original feature vector and the reconstructed feature vector is less than or equal to a preset difference threshold, it is determined that the target vehicle is not in a slipping condition. If the combined difference between the original feature vector and the reconstructed feature vector is greater than a preset difference threshold, the target vehicle is determined to be in a slipping condition.

[0202] Optionally, the device further includes: The compression encoding module is used to compress and encode the multi-dimensional state parameters of the target vehicle at the current moment to obtain a compressed feature vector. The decoding and reconstruction module is used to decode and reconstruct the compressed feature vector to obtain the reconstructed feature vector; And, the second determining module 502 is used for: Based on the differences between the compressed feature vector, the original feature vector and the reconstructed feature vector, and the parameter changes of multi-dimensional state parameters at multiple times, the current slip type of the target vehicle is determined.

[0203] Optionally, the second determining module 502 is specifically used for: The compressed feature vector, difference, and parameter change information are input into the slip type recognition model, which then outputs multiple predicted slip types and their corresponding confidence levels. The slip type with the highest confidence among multiple predicted slip types is determined as the current slip type.

[0204] Optionally, the torque adjustment module 503 is used for: Based on the current slippage type and M target state parameters, the target allocation ratio is determined. The target allocation ratio is used to represent the ideal torque distribution ratio of multiple wheels of the target vehicle. The output torque of the target vehicle is adjusted based on the target torque of the whole vehicle and the target distribution ratio.

[0205] Optionally, the torque adjustment module 503 is used for: The current slip type and M target state parameters are fuzzed to obtain multiple first membership degrees and M groups of second membership degrees. The multiple first membership degrees represent the probability that the current slip type belongs to multiple slip levels. The M groups of second membership degrees are the membership degrees corresponding to the M target state parameters. Each group of second membership degrees in the M groups includes multiple second membership degrees. Based on multiple first membership degrees and M groups of second membership degrees, N reference membership degrees and N groups of reference levels are determined from the fuzzy rule base. Each of the N groups of reference levels includes multiple allocation ratio levels, which are used to indicate the possible torque distribution ratio of multiple wheels. The N reference membership degrees are used to represent the confidence level of the N groups of reference levels. The target allocation ratio is determined based on N reference membership degrees and N sets of reference levels.

[0206] Optionally, the torque adjustment module 503 is used for: The target membership degree is determined by selecting the largest reference membership degree from N reference membership degrees. The set of reference levels corresponding to the membership degree of the target is determined as the target level; Obtain multiple ratio ranges corresponding to multiple allocation ratio levels in the target level; The target allocation ratio is determined based on multiple ratio ranges.

[0207] Optionally, the torque adjustment module 503 is used for: According to the target allocation ratio, the target torque of the whole vehicle is divided into multiple ideal torques, which are used to represent the torque that should be allocated to multiple wheels; Based on multiple ideal torques, determine the target torque for the front axle and the target torque for the rear axle; Based on the target torque of the front axle and the target torque of the rear axle, the torque of multiple wheels of the target vehicle is adjusted.

[0208] Optionally, the torque adjustment module 503 is used for: Subtract the actual torque of the front axle from the target torque of the front axle to obtain the torque difference of the front axle; Subtract the actual torque of the rear axle from the target torque of the rear axle to obtain the torque difference of the rear axle; Based on the front axle torque difference, control the front axle motor to adjust the front axle output torque, and distribute the front axle output torque to the front wheels according to the front wheel torque distribution ratio in the target distribution ratio; Based on the rear axle torque difference, the rear axle motor is controlled to adjust the rear axle output torque, and the rear axle output torque is distributed to the rear wheels according to the rear wheel torque distribution ratio in the target distribution ratio.

[0209] In this embodiment, by reconstructing the multi-dimensional state parameters of the target vehicle at the current moment, a reconstructed feature vector corresponding to the multi-dimensional state parameters is first obtained. Since there is a certain correlation between the various state parameters under normal circumstances, feature reconstruction is based on this correlation to reconstruct the multi-dimensional state parameters. Therefore, under normal circumstances, the reconstructed feature vector should be consistent with the original feature vector. However, this correlation is broken under slippage conditions. Therefore, the reconstructed feature vector obtained after reconstructing the multi-dimensional state parameters based on this correlation will have a certain difference from the original feature vector. Therefore, based on the difference between the original feature vector and the reconstructed feature vector of the multi-dimensional state parameters, it is possible to accurately determine whether the target vehicle is in a slippage condition. Then, when the target vehicle is in a slippage condition, based on the difference between the original feature vector and the reconstructed feature vector, and the parameter change information of the multi-dimensional state parameters at multiple moments, the current slippage type of the target vehicle is determined. Since the difference between the original feature vector and the reconstructed feature vector can represent the severity of vehicle slippage, for example, the greater the difference, the more severe the slippage, the parameter change information of multi-dimensional state parameters at multiple times can represent the dynamic change trend of the target vehicle. Thus, by combining these two aspects of parameters, the current slippage type can be accurately determined. Subsequently, based on the current slippage type, M target state parameters, and the target torque of the whole vehicle at the current time, the torque of the target vehicle can be precisely adjusted, so that the target vehicle can get out of the slippage condition as soon as possible.

[0210] It should be noted that the torque adjustment device provided in the above embodiments is only illustrated by the division of the above functional modules when the vehicle slips. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above.

[0211] The functional units and modules in the above embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit. Furthermore, the specific names of the functional units and modules are only for easy differentiation and are not intended to limit the scope of protection of the embodiments of this application.

[0212] The torque adjustment device and torque adjustment method provided in the above embodiments belong to the same concept. The specific working process and technical effects of the units and modules in the above embodiments can be found in the method embodiments section, and will not be repeated here.

[0213] Figure 6 This is a schematic diagram of the structure of a vehicle provided in an embodiment of this application.

[0214] For example, such as Figure 6 As shown, the vehicle 600 includes a memory 61 and a processor 60, wherein the memory 61 stores executable program code 62, and the processor 60 is used to call and execute the executable program code 62 to perform the torque adjustment method described above.

[0215] This embodiment can divide the vehicle into functional modules according to the above method example. For example, each function can be assigned to a separate module, or two or more functions can be integrated into one processing module. The integrated module can be implemented in hardware. It should be noted that the module division in this embodiment is illustrative and only represents one logical functional division. In actual implementation, there may be other division methods.

[0216] When each functional module is divided according to its corresponding function, the vehicle may include: a first determining module, a second determining module, and a torque adjustment module. It should be noted that all relevant content regarding the steps involved in the above method embodiments can be referenced from the functional descriptions of the corresponding functional modules, and will not be repeated here.

[0217] The vehicle provided in this embodiment is used to execute the torque adjustment method described above, and therefore can achieve the same effect as the above implementation method.

[0218] When using integrated units, the vehicle may include a processing module and a storage module. The processing module is used to control and manage the vehicle's actions. The storage module is used to support the vehicle in executing corresponding program code and data.

[0219] The processing module may be a processor or a controller, which can implement or execute various exemplary logic blocks, modules, and circuits shown in conjunction with the disclosure of this application. The processor may also be a combination of functions that implement computing capabilities, such as a combination of one or more microprocessors, a combination of digital signal processing (DSP) and a microprocessor, etc., and the storage module may be a memory.

[0220] This embodiment also provides a computer-readable storage medium storing computer program code. When the computer program code is run on a computer, the computer executes the aforementioned method steps to implement the torque adjustment method described in the above embodiment.

[0221] This embodiment also provides a computer program product that, when run on a computer, causes the computer to perform the aforementioned related steps to implement the torque adjustment method described in the above embodiment.

[0222] In this embodiment, the vehicle, computer-readable storage medium, computer program product, or chip are all used to execute the method described above. Therefore, the beneficial effects that can be achieved can be referred to the beneficial effects of the method described above, and will not be repeated here.

[0223] Through the above description of the embodiments, those skilled in the art will understand that, for the sake of convenience and brevity, only the division of the above functional modules is used as an example. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above.

[0224] In the embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are illustrative; for instance, the division of modules or units is merely a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another device, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0225] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A torque adjustment method, characterized in that, The method includes: Based on the difference between the original feature vector and the reconstructed feature vector corresponding to the multi-dimensional state parameters of the target vehicle at the current moment, it is determined whether the target vehicle is in a slipping condition. The reconstructed feature vector is obtained by reconstructing the features of the multi-dimensional state parameters. When the target vehicle is in a slipping condition, the current slipping type of the target vehicle is determined based on the difference between the original feature vector and the reconstructed feature vector, and the parameter change information of the multi-dimensional state parameters at multiple times. Based on the current slippage type, M target state parameters, and the target torque of the vehicle at the current moment, the output torque of the target vehicle is adjusted. The target state parameters are state parameters related to the slippage state.

2. The method as described in claim 1, characterized in that, Determining whether the target vehicle is in a slipping condition based on the difference between the original feature vector and the reconstructed feature vector includes: Based on the i-th dimension vector of the original feature vector and the i-th dimension vector of the reconstructed feature vector, determine the vector difference of the i-th dimension; Based on the vector differences between the original feature vector and the reconstructed feature vector in multiple dimensions, the comprehensive difference between the original feature vector and the reconstructed feature vector is calculated; If the combined difference between the original feature vector and the reconstructed feature vector is less than or equal to a preset difference threshold, it is determined that the target vehicle is not in a slipping condition. If the combined difference between the original feature vector and the reconstructed feature vector is greater than the preset difference threshold, the target vehicle is determined to be in a slipping condition.

3. The method as described in claim 1, characterized in that, The method further includes: The multi-dimensional state parameters of the target vehicle at the current moment are compressed and encoded to obtain a compressed feature vector; The compressed feature vector is decoded and reconstructed to obtain the reconstructed feature vector; And, determining the current slip type of the target vehicle based on the difference between the original feature vector and the reconstructed feature vector, and the parameter change information of the multi-dimensional state parameters at multiple times, includes: Based on the compressed feature vector, the difference between the original feature vector and the reconstructed feature vector, and the parameter change information of the multi-dimensional state parameters at multiple times, the current slip type of the target vehicle is determined.

4. The method as described in claim 3, characterized in that, The step of determining the current slip type of the target vehicle based on the compressed feature vector, the difference between the original feature vector and the reconstructed feature vector, and the parameter change information of the multi-dimensional state parameters at multiple times includes: The compressed feature vector, the difference, and the parameter change information are input into the slip type recognition model, and the slip type recognition model outputs multiple predicted slip types and the confidence scores corresponding to the multiple predicted slip types. The predicted slip type with the highest confidence among the multiple predicted slip types is determined as the current slip type.

5. The method as described in claim 1, characterized in that, The adjustment of the output torque of the target vehicle based on the current slippage type, M target state parameters, and the target torque of the vehicle at the current moment includes: Based on the current slippage type and the M target state parameters, a target allocation ratio is determined, which represents the ideal torque distribution ratio of multiple wheels of the target vehicle. Based on the target torque of the vehicle and the target distribution ratio, the output torque of the target vehicle is adjusted.

6. The method as described in claim 5, characterized in that, The step of determining the target allocation ratio based on the current slippage type and the M target state parameters includes: The current slip type and the M target state parameters are fuzzed to obtain multiple first membership degrees and M groups of second membership degrees. The multiple first membership degrees represent the probability that the current slip type belongs to multiple slip levels. The M groups of second membership degrees are the membership degrees corresponding to the M target state parameters. Each group of second membership degrees in the M groups includes multiple second membership degrees. Based on the multiple first membership degrees and the M groups of second membership degrees, N reference membership degrees and N groups of reference levels are determined from the fuzzy rule base. Each of the N groups of reference levels includes multiple allocation ratio levels, which are used to indicate the possible torque distribution ratio of multiple wheels. The N reference membership degrees are used to represent the confidence level of the N groups of reference levels. The target allocation ratio is determined based on the N reference membership degrees and the N sets of reference levels.

7. The method as described in claim 6, characterized in that, The determination of the target allocation ratio based on the N reference membership degrees and the N sets of reference levels includes: The maximum reference membership degree among the N reference membership degrees is determined as the target membership degree. The set of reference levels corresponding to the target membership degree is determined as the target level; Obtain multiple ratio ranges corresponding to multiple allocation ratio levels in the target level; The target allocation ratio is determined based on the multiple ratio ranges.

8. The method as described in claim 5, characterized in that, The adjustment of the output torque of the target vehicle based on the target torque of the whole vehicle and the target distribution ratio includes: According to the target allocation ratio, the target torque of the whole vehicle is divided into multiple ideal torques, and the multiple ideal torques are used to represent the torque that should be allocated to multiple wheels; Based on the multiple ideal torques, determine the target torque for the front axle and the target torque for the rear axle; Based on the target torque of the front axle and the target torque of the rear axle, the torque of multiple wheels of the target vehicle is adjusted.

9. The method as described in claim 8, characterized in that, The adjustment of the torque of multiple wheels of the target vehicle based on the target torque of the front axle and the target torque of the rear axle includes: Subtract the actual torque of the front axle from the target torque of the front axle to obtain the torque difference of the front axle; Subtract the actual torque of the rear axle from the target torque of the rear axle to obtain the torque difference of the rear axle; Based on the front axle torque difference, the front axle motor is controlled to adjust the front axle output torque, and the front axle output torque is distributed to the front wheels according to the front wheel torque distribution ratio in the target distribution ratio. Based on the rear axle torque difference, the rear axle motor is controlled to adjust the rear axle output torque, and the rear axle output torque is distributed to the rear wheels according to the rear wheel torque distribution ratio in the target distribution ratio.

10. A vehicle, characterized in that, The vehicles include: Memory, used to store executable program code; A processor for calling and running the executable program code from the memory, causing the vehicle to perform the method as described in any one of claims 1 to 9.

Images