CN-121479210-B - Reconfigurable operator combination determination method and device for driving assistance

CN121479210BCN 121479210 BCN121479210 BCN 121479210BCN-121479210-B

Abstract

The application discloses a reconfigurable operator combination determining method and device for driving assistance, and belongs to the technical field of electric digital data processing. The method comprises the steps of obtaining key feature information of preset road condition data of a vehicle and hardware performance information of a tensor processing unit of the vehicle, selecting candidate operators based on the key feature information and the hardware performance information, determining disturbance indexes of the candidate operators, pairing the candidate operators in pairs to obtain basic operator pairs, determining matching scores of the basic operator pairs according to the disturbance indexes of the candidate operators in the basic operator pairs, selecting a preset number of basic operator pairs according to the matching scores from high to low as the candidate operator pairs, and determining reconfigurable operator combinations to be applied to the vehicle according to the candidate operator pairs. According to the technical scheme, reasonable screening of the reconfigurable operator combination is achieved by accurately matching road condition data characteristics, TPU hardware characteristics and the synergistic disturbance influence of the quantization operator, so that the instantaneity of auxiliary driving control is improved.

Inventors

WANG LIYU
LIU MAN
DONG WENQIANG

Assignees

广州万协通信息技术有限公司

Dates

Publication Date: 20260508
Application Date: 20260107

Claims (7)

1. A reconfigurable operator combination determination method for driving assistance, the method comprising: Acquiring key characteristic information of preset road condition data of a vehicle and hardware performance information of a tensor processing unit of the vehicle, wherein the preset road condition data comprises image data and at least one radar data; Selecting a candidate operator from a preset operator pool based on the key characteristic information and the hardware performance information, and determining a disturbance index of the candidate operator, wherein the disturbance index comprises a cache disturbance index, a bandwidth disturbance index and a synchronous disturbance index; The method comprises the steps of selecting candidate operators in a preset operator pool based on key feature information and hardware performance information, determining compatibility information of each operator in the preset operator pool on preset road condition data based on the key feature information, wherein the key feature information comprises data type information, real-time demand information and resource demand information, determining adaptation degree reference information of each operator in the preset operator pool on a tensor processing unit based on the hardware performance information, wherein the adaptation degree reference information comprises parallel computing core utilization rate, data processing efficiency, memory bandwidth occupancy rate and calculation force loss; The method comprises the steps of establishing a flow resistance topological graph according to hardware performance information, generating calculation fluid data according to data type information and attribute information of each operator in a preset operator pool, and determining predicted flow velocity of the calculation fluid data as data processing efficiency according to the flow resistance topological graph; The method comprises the steps of determining the predicted flow rate of calculated fluid data according to the flow resistance topological graph as data processing efficiency, wherein the method comprises the steps of mapping the fluid attribute of the calculated fluid data into injection pressure values of input nodes of the flow resistance topological graph, distributing reference initial pressure values for other nodes in the flow resistance topological graph, calculating the instantaneous flow of the calculated fluid data flowing through each side according to the pressure distribution of the flow resistance topological graph, and updating the pressure values of all nodes in the flow resistance topological graph according to the instantaneous flow; Pairing the candidate operators pairwise to obtain a basic operator pair, and determining the matching score of the basic operator pair according to the disturbance indexes of two candidate operators in the basic operator pair; And selecting a preset number of basic operator pairs as candidate operator pairs according to the sequence from high to low of the matching score, and determining a reconfigurable operator combination to be applied to the vehicle according to the candidate operator pairs for assisting driving when the vehicle runs.
2. The reconfigurable operator combination determination method for driving assistance according to claim 1, wherein the determining a reconfigurable operator combination to be applied of the vehicle from the candidate operator pair includes: Counting the occurrence times of each operator in the candidate operator pairs, and determining the initial score of each operator according to the occurrence times and the matching score; The operators are arbitrarily combined to obtain candidate reconfigurable operator combinations, and estimated cooperative efficiency of the candidate reconfigurable operator combinations on the tensor processing unit is determined based on the hardware performance information; Determining a total score of the candidate reconfigurable operator combination according to the initial score and the estimated cooperative efficiency; and determining a reconfigurable operator combination to be applied of the vehicle in the candidate reconfigurable operator combinations according to the total score.
3. The reconfigurable operator combination determination method for driving assistance according to claim 2, wherein the hardware performance information includes a calculation unit peak computing power, a memory bandwidth, and a buffer capacity; accordingly, the determining, based on the hardware performance information, the estimated synergistic efficiency of the candidate reconfigurable operator combination on the tensor processing unit includes: Determining a computing unit utilization weight based on the computing unit peak computing power, and determining a memory access overhead weight based on the memory bandwidth and the cache capacity; Constructing a collaborative efficiency computing function according to the computing unit utilization rate weight and the memory access overhead weight; Substituting the calculated amount and the memory access amount of each operator in the candidate reconfigurable operator combination into the collaborative efficiency calculation function to obtain the estimated collaborative efficiency of the candidate reconfigurable operator combination on the tensor processing unit.
4. The reconfigurable operator combination determination method for driving assistance according to claim 1, wherein the determining the matching score of the base operator pair from the disturbance indexes of the two candidate operators of the base operator pair includes: Respectively adding and calculating the buffer disturbance indexes, the bandwidth disturbance indexes and the synchronous disturbance indexes of the two candidate operators in the basic operator pair to obtain a sub-term disturbance cost, and carrying out weighted summation calculation on the sub-term disturbance cost to obtain a total disturbance cost; Respectively carrying out absolute difference calculation on the buffer disturbance index, the bandwidth disturbance index and the synchronous disturbance index value of the two candidate operators in the basic operator pair to obtain a sub-term complementary gain value, and carrying out summation calculation on the sub-term complementary gain value exceeding a preset complementary gain threshold value to obtain a total complementary gain value; Subtracting the total disturbance cost from a preset basic score to obtain a first calculation result, multiplying the total complementary gain value by a preset complementary gain coefficient to obtain a second calculation result, and adding the first calculation result and the second calculation result to obtain a matching score of the basic operator pair.
5. A reconfigurable operator combination determination apparatus for assisting driving, the apparatus comprising: the information acquisition module is used for acquiring key characteristic information of preset road condition data of the vehicle and hardware performance information of a tensor processing unit of the vehicle, wherein the preset road condition data comprises image data and at least one radar data; the candidate determining module is used for selecting a candidate operator from a preset operator pool based on the key characteristic information and the hardware performance information and determining a disturbance index of the candidate operator, wherein the disturbance index comprises a cache disturbance index, a bandwidth disturbance index and a synchronous disturbance index; The candidate determining module is specifically configured to determine compatibility information of each operator in the preset operator pool on the preset road condition data based on the key feature information, wherein the key feature information comprises data type information, real-time demand information and resource demand information, determine adaptation degree reference information of each operator in the preset operator pool on the tensor processing unit based on the hardware performance information, wherein the adaptation degree reference information comprises parallel computing core utilization rate, data processing efficiency, memory bandwidth occupancy rate and calculation force loss; The candidate determining module is specifically used for constructing a flow resistance topological graph according to the hardware performance information, generating calculation fluid data according to the data type information and attribute information of each operator in a preset operator pool, and determining the predicted flow velocity of the calculation fluid data as data processing efficiency according to the flow resistance topological graph; The candidate determining module is specifically configured to map the fluid attribute of the calculated fluid data to an injection pressure value of an input node of the flow resistance topological graph, and allocate a reference initial pressure value to other nodes in the flow resistance topological graph; calculating the instantaneous flow of the calculated fluid data flowing through each side according to the pressure distribution of the flow resistance topological graph, and updating the pressure values of all nodes in the flow resistance topological graph according to the instantaneous flow; repeatedly executing the steps until the variation norm of the pressure distribution of the flow resistance topological graph is smaller than a preset convergence threshold value, determining the instantaneous flow of the output side of the flow resistance topological graph as a predicted flow velocity, and obtaining data processing efficiency; The score determining module is used for pairwise pairing the candidate operators to obtain a basic operator pair, and determining the matching score of the basic operator pair according to the disturbance indexes of two candidate operators in the basic operator pair; And the combination determining module is used for selecting a preset number of basic operator pairs as candidate operator pairs according to the sequence from high to low of the matching score, and determining a reconfigurable operator combination to be applied to the vehicle according to the candidate operator pairs for assisting driving when the vehicle runs.
6. An electronic device comprising a processor, a memory and a program or instruction stored on the memory and executable on the processor, which when executed by the processor implements the reconfigurable operator combination determination method for driving assistance of any one of claims 1-4.
7. A readable storage medium, characterized in that a program or instructions is stored on the readable storage medium, which when executed by a processor implements the reconfigurable operator combination determination method for assisting driving according to any one of claims 1-4.

Description

Reconfigurable operator combination determination method and device for driving assistance Technical Field The application belongs to the technical field of electric digital data processing, and particularly relates to a reconfigurable operator combination determining method and device for driving assistance. Background In the auxiliary driving technology, road condition data can be obtained through a vehicle-mounted sensor, then the road condition data is inferred through a model, and corresponding control instructions are output, so that accurate regulation and control (such as self-adaptive cruising, emergency obstacle avoidance and the like) on the running state of the vehicle are realized, and whether the process is smooth or not depends on efficient operation of model reasoning. The core influencing factors of the model reasoning efficiency are rationality of the reconfigurable operator combination, wherein the operators are basic model calculation units, the combination mode of the operators influences core elements such as calculation power distribution and data transmission efficiency, the reasonable reconfigurable operator combination can improve the reasoning speed and stability, and the unreasonable reconfigurable operator combination can lower the efficiency and finally influence the auxiliary driving landing effect. Currently, with the development of multi-sensor fusion application, road condition data presents multi-source heterogeneous characteristics, and vehicle-mounted Tensor Processing Units (TPU) carried by different vehicle types have significant differences in hardware architecture. However, the existing operator selection scheme is limited to single performance index type selection, so that the multi-source heterogeneous nature and the differentiated processing requirement of the current road condition data cannot be considered, the suitability of the operator and the TPU hardware characteristics is ignored, and the fact that the reconfigurable operator combination cannot accurately meet the targeted calculation requirement and the real-time processing standard in the current auxiliary driving scene is easily caused. Disclosure of Invention The application provides a reconfigurable operator combination determining method and device for driving assistance, and aims to realize reasonable screening of the reconfigurable operator combination by accurately matching road condition data characteristics, TPU hardware characteristics and quantization operator cooperative disturbance influence, so that the real-time performance of driving assistance control is improved. In a first aspect, the present application provides a reconfigurable operator combination determination method for assisting driving, the method comprising: Acquiring key characteristic information of preset road condition data of a vehicle and hardware performance information of a tensor processing unit of the vehicle, wherein the preset road condition data comprises image data and at least one radar data; Selecting a candidate operator from a preset operator pool based on the key characteristic information and the hardware performance information, and determining a disturbance index of the candidate operator, wherein the disturbance index comprises a cache disturbance index, a bandwidth disturbance index and a synchronous disturbance index; Pairing the candidate operators pairwise to obtain a basic operator pair, and determining the matching score of the basic operator pair according to the disturbance indexes of two candidate operators in the basic operator pair; And selecting a preset number of basic operator pairs as candidate operator pairs according to the sequence from high to low of the matching score, and determining a reconfigurable operator combination to be applied to the vehicle according to the candidate operator pairs for assisting driving when the vehicle runs. Optionally, the determining the reconfigurable operator combination to be applied of the vehicle according to the candidate operator pair includes: Counting the occurrence times of each operator in the candidate operator pairs, and determining the initial score of each operator according to the occurrence times and the matching score; The operators are arbitrarily combined to obtain candidate reconfigurable operator combinations, and estimated cooperative efficiency of the candidate reconfigurable operator combinations on the tensor processing unit is determined based on the hardware performance information; Determining a total score of the candidate reconfigurable operator combination according to the initial score and the estimated cooperative efficiency; and determining a reconfigurable operator combination to be applied of the vehicle in the candidate reconfigurable operator combinations according to the total score. Optionally, the hardware performance information includes a calculation unit peak computing power, a memory bandwidth and a cache capac