Self-learning type commercial vehicle weight identification and confidence evaluation method

Abstract

The invention discloses a self-learning type commercial vehicle re-identification and confidence evaluation method, which comprises the following steps of: firstly, extracting vehicle operation data required by a model from a multi-source data set by utilizing a big data extraction technology; then, abnormal frames in the vehicle operation data are identified and preprocessed according to a data cleaning strategy, and expansion operation of vehicle operation features and road structure features is completed; and constructing a load reference generation model based on vehicle longitudinal dynamics by adopting a recursive least square algorithm with a forgetting factor, and constructing a load identification output and confidence evaluation model by adopting an optimized Informer neural network so as to complete a scheme of vehicle load identification and confidence evaluation. According to the device, real-time identification of the load of the whole vehicle is realized by using the data of the Internet of Vehicles, the in-transit whole-course supervision of the standardized heavy freight vehicle is facilitated, and a reliable basis is provided for the optimization of the vehicle travel utilization rate.

Description

Technical Field

[0001] The present invention belongs to the field of vehicle parameter identification, and in particular relates to a self-learning commercial vehicle load identification and confidence assessment method. Background Art

[0002] With the development of technologies such as big data science, 5G communications and artificial intelligence, the electrification, networking and intelligence of automobiles have gradually become the development direction of the automotive industry. Intelligent networked vehicles refer to vehicles equipped with advanced on-board sensors, controllers, actuators and other equipment, which enable vehicles to have complex environmental perception, intelligent decision-making, collaborative control and other functions, greatly facilitating the acquisition of various parameters under vehicle operating conditions; combined with modern communication and network technologies, the storage volume and accuracy of Internet of Vehicles data have been significantly improved; under this development trend, vehicle parameter identification and data mining based on massive Internet of Vehicles data have gradually become research hotspots.

[0003] In recent years, with the rapid development of the transportation industry, my country's road freight transport volume accounts for 73.8% of the total freight volume. At present, there are two main measures to achieve full-time on-the-way supervision of freight vehicles. One is to install additional gravity sensors, and the other is to have drivers manually input data and upload it. However, both of the above methods have certain problems: installing gravity sensors will increase vehicle costs, and it is difficult to promote general vehicles and small-scale fleets; and the authenticity and accuracy of data uploaded manually by drivers will be affected by personal factors of the drivers.

[0004] As the Internet of Vehicles technology becomes increasingly mature, the identification and estimation of vehicle weight based on the vehicle operating condition measurement data of intelligent connected vehicles is one of the important application solutions of Internet of Vehicles data. Through the research and development of vehicle load estimation algorithms, the vehicle's cargo transportation status can be obtained in real time, and the vehicle's on-the-road driving status can be monitored throughout the process, thereby achieving effective supervision of vehicle overloading and avoiding the occurrence of transportation accidents to a certain extent; at the same time, by obtaining the real-time vehicle load, while avoiding abnormal loading and unloading behavior of trucks, it can also reflect the travel utilization rate of vehicles in the corresponding area, thereby helping freight companies to reasonably allocate trucks, avoid the waste of vehicle transportation resources, and improve transportation efficiency. Summary of the invention

[0005] Purpose of the invention: The present invention provides a self-learning commercial vehicle load identification and confidence assessment method, which can realize the full-process supervision of freight vehicles on the way.

[0006] Technical solution: The self-learning commercial vehicle load identification and confidence assessment method described in the present invention comprises the following steps:

[0007] (1) Vehicle operation initialization: including the initialization of the GPS module, vehicle management system, vehicle communication module and vehicle static parameters. The vehicle static parameters include the following fields: vehicle identification number, vehicle model, engine type, vehicle width, vehicle height, rear axle speed ratio, gear ratio of each gear, wheel rolling radius, and engine maximum torque;

[0008] (2) Multi-source data acquisition: Data acquired through the GPS module include timestamp, elevation, and vehicle speed; data acquired through the CAN bus include engine speed, engine torque, and gear information;

[0009] (3) Operation data detection and cleaning: Implement corresponding cleaning strategies for erroneous frame data such as missing and noise in the operation data, and then take pre-processing measures such as interpolation and filtering for the operation data that does not meet the continuity conditions;

[0010] (4) Vehicle operation characteristics expansion: Based on the vehicle system transmission characteristics and road structure characteristics, the pre-processed vehicle operation data is used to obtain the road slope characteristics, vehicle acceleration and jerk characteristics of the entire road section at all times;

[0011] (5) Design of load-carrying joint identification system: Establish a load-carrying benchmark generation model based on the recursive least squares method, and iterate the vehicle operation data to obtain the distribution benchmark of the estimated load; establish a load-carrying regression identification model based on the improved Informer encoder, and output the load-carrying identification results through GPS source data and extended operation characteristics;

[0012] (6) Confidence assessment and model fine-tuning: Based on the dynamic load distribution benchmark and the output of the neural network load regression model, complete the confidence assessment of the load identification result; store the vehicle operation data and total weight benchmark data at all times of the entire journey, perform offline data training and fine-tune the load regression identification model after the current journey ends, and redeploy the model after the training is completed;

[0013] Furthermore, the implementation process of step (3) is as follows:

[0014] Considering the differences in the operating characteristics of non-homologous data, artificial features are extracted and marked for the error frames in the GPS source data and CAN bus source data respectively, and the error frame data samples are separated based on the continuity criterion. The error frame features and cleaning strategies of multi-source data are formulated as follows.

[0015] GPS source data: Error frames mainly come from data interruptions. Combined with the data feedback timestamp, they can be divided into short-term data loss and long-term signal interruption. The time threshold δ is set to 30s. For short-term data loss less than this threshold, linear interpolation is performed using the nearest non-error data value before and after the error frame. For long-term signal interruptions exceeding this threshold, the data during the period is set to zero and discarded.

[0016] CAN bus source data: Error frame data is mainly characterized by short-term data loss, high-frequency measurement noise and discontinuous mutations.

[0017] Short-term data missing is mainly aimed at vehicle gear data detection. Based on the principle of continuous change of vehicle gear, short-term single-point missing data is filled with non-abnormal data at the previous moment.

[0018] High-frequency measurement noise is mainly used for engine speed and engine torque data detection, and a Savitzky-Golay filter is designed to process the noise data.

[0019] Discontinuous mutations are mainly aimed at engine torque data detection, and the characteristic detection standards are set as follows:

[0020]

[0021] Among them, R is the engine torque data at a certain moment, T is the timestamp at the corresponding moment, t is the calculation moment, z is the time step, and δ is the set detection threshold; for the data points obtained according to the above detection standards, the linear interpolation method is used to cover the data values ​​at the corresponding moment.

[0022] Furthermore, the implementation process of step (4) is as follows:

[0023] After the vehicle operation data is cleaned, the vehicle operation speed characteristics are obtained using the pre-processed vehicle operation data based on the vehicle parameter attributes and operation characteristics:

[0024]

[0025] Then the vehicle running acceleration and jerk characteristics are obtained:

[0026]

[0027] Where n is the engine speed, r is the wheel rolling radius, i g is the transmission ratio, i0 is the rear axle speed ratio, Δt is the time step, V t is the vehicle speed, a t is the acceleration value, j is the jerk.

[0028] Based on the construction standards and changing characteristics of expressways, the road slope is obtained by using the pre-processed speed data and elevation data:

[0029]

[0030] Where l is the time step of backward calculation; h t is the elevation value at time t, h t+nis the elevation value at time t+n, in meters; v i is the vehicle speed value, in units of

[0031] The point slope calculation is performed according to the distance span of 100m, and the elevation data is guaranteed to change monotonically within a single calculation step, and the time step k satisfies:

[0032]

[0033] (h i -h i-1 )·(h i+1 -h i )>0,(i=t+1,t+2,…,t+n-1)

[0034] Based on the above calculation strategy and formula, all running time period data are traversed to obtain the road slope characteristics, vehicle acceleration and jerk characteristics of the entire journey.

[0035] Furthermore, the load joint identification system of step (5) is designed as follows:

[0036] Based on the vehicle dynamics model, the reasonable distribution range of the estimated load is obtained, and the vehicle longitudinal dynamics model under driving conditions is adopted:

[0037]

[0038] Among them, g is the acceleration of gravity, M is the mass of the car, f is the friction coefficient, α is the road slope, C D is the air resistance coefficient, A is the frontal area of ​​the vehicle, ρ is the air density, σ is the conversion coefficient of the vehicle's moment of inertia, v is the vehicle's speed, a is the vehicle's acceleration, and T tq is the engine torque, i g is the transmission ratio, i0 is the rear axle speed ratio, η T is the total efficiency of the transmission system, r is the rolling radius of the wheel;

[0039] The recursive least squares algorithm with forgetting factor is used to obtain the distribution benchmark of the estimated load based on the above vehicle dynamics model filtering, and the loss function of the filtering algorithm is constructed:

[0040]

[0041] Among them, λ represents the forgetting factor, z(i) represents the actual output of the system, φ(k) represents the system input, Represents the parameter vector to be estimated at time k.

[0042] The load benchmark generation model based on recursive least squares is designed as follows:

[0043]

[0044] The load regression identification model is built using the Informer-based neural network architecture. The model input parameters are selected as: vehicle speed, acceleration, jerk, road elevation, road slope, and the model output parameter is the load regression identification result.

[0045] Furthermore, the implementation process of step (6) is as follows:

[0046] The load distribution output by the recursive least squares algorithm is used to construct a 3σ distribution interval, and the interval average is used as the confidence evaluation benchmark. The confidence of the load regression identification model output of the Informer encoder is evaluated, and MAE is used as the evaluation indicator:

[0047]

[0048] in, is the output result of the load regression identification model, is the average value of the load reference interval;

[0049] The confidence assessment criteria are:

[0050]

[0051] The load regression identification results and confidence evaluation results are iterated with the time series and continuously output along with the vehicle operation until the vehicle working trip ends; the load regression identification results and confidence evaluation results in each time dimension are recorded in the vehicle storage unit;

[0052] After the vehicle's current working trip is completed, the confidence of the load identification results within the entire trip is counted, and the load identification data with a confidence level lower than 10% is marked. The average value of the load identification benchmark at the corresponding moment is used as the fine-tuning label to generate a fine-tuning dataset for the Informer encoder for offline training. After the model fine-tuning is completed, it is redeployed.

[0053] Through the above model calculation and identification, the load identification result and confidence assessment value can be output, and the data set generation, model self-learning and model redeployment can be completed after the vehicle journey ends.

[0054] Beneficial effects: Compared with the prior art, the beneficial effects of the present invention are as follows: 1. The present invention takes into account the technical advantages and technical level of existing intelligent networked freight vehicles, combines the content and characteristics of vehicle networking data, and adopts mature data preprocessing methods for typical vehicle networking data types. Based on the acquired vehicle and road status information, the weight of the whole vehicle is accurately identified; 2. The present invention adopts the recursive least squares algorithm with forgetting factor and the optimized Informer neural network model, which can realize the real-time identification of the whole vehicle load and the high-precision output of the result confidence; 3. The present invention can realize the automatic learning and fine-tuning correction of the model. With the accumulation of application scope and application time, the accuracy and versatility of the identification regression model will continue to improve; 4. The present invention can be applied to many fields related to commercial vehicles. In the vehicle design stage, the operating load conditions of vehicles in the segmented market are mastered to make the vehicle development target design more accurate. In the vehicle use stage, cargo and transportation capacity are accurately matched, cargo loading rate is optimized, vehicle energy consumption evaluation is improved, intelligent maintenance strategy calculation, calculation of used car residual value, and even participate in the calculation of vehicle intelligent predictive driving strategy. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 It is a flow chart of the present invention. DETAILED DESCRIPTION

[0056] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0057] A self-learning commercial vehicle load identification and confidence assessment method of the present invention comprises the following steps:

[0058] (1) Vehicle operation initialization: including the initialization of the GPS module, vehicle management system, vehicle communication module and vehicle static parameters. The vehicle static parameters include the following fields: vehicle identification number, vehicle model, engine type, vehicle width, vehicle height, rear axle speed ratio, gear ratio of each gear, wheel rolling radius, and engine maximum torque;

[0059] (2) Multi-source data acquisition: Data acquired through the GPS module include timestamp, elevation, and vehicle speed; data acquired through the CAN bus include engine speed, engine torque, and gear information;

[0060] (3) Operation data detection and cleaning: Implement corresponding cleaning strategies for erroneous frame data such as missing and noise in the operation data, and then take pre-processing measures such as interpolation and filtering for the operation data that does not meet the continuity conditions;

[0061] (4) Vehicle operation characteristics expansion: Based on the vehicle system transmission characteristics and road structure characteristics, the pre-processed vehicle operation data is used to obtain the road slope characteristics, vehicle acceleration and jerk characteristics of the entire road section at all times;

[0062] (5) Design of load-carrying joint identification system: Establish a load-carrying benchmark generation model based on the recursive least squares method, and iterate the vehicle operation data to obtain the distribution benchmark of the estimated load; establish a load-carrying regression identification model based on the improved Informer encoder, and output the load-carrying identification results through GPS source data and extended operation characteristics;

[0063] (6) Confidence assessment and model fine-tuning: Based on the dynamic load distribution benchmark and the output of the neural network load regression model, complete the confidence assessment of the load identification result; store the vehicle operation data and total weight benchmark data at all times of the entire journey, perform offline data training and fine-tune the load regression identification model after the current journey ends, and redeploy the model after the training is completed;

[0064] Furthermore, the implementation process of step (3) is as follows:

[0065] Considering the differences in the operating characteristics of non-homologous data, artificial features are extracted and marked for the error frames in the GPS source data and CAN bus source data respectively, and the error frame data samples are separated based on the continuity criterion. The error frame features and cleaning strategies of multi-source data are formulated as follows.

[0066] GPS source data: Error frames mainly come from data interruptions. Combined with the data feedback timestamp, they can be divided into short-term data loss and long-term signal interruption. The time threshold δ is set to 30s. For short-term data loss less than this threshold, linear interpolation is performed using the nearest non-error data value before and after the error frame. For long-term signal interruptions exceeding this threshold, the data during the period is set to zero and discarded.

[0067] CAN bus source data: Error frame data is mainly characterized by short-term data loss, high-frequency measurement noise and discontinuous mutations.

[0068] Short-term data missing is mainly aimed at vehicle gear data detection. Based on the principle of continuous change of vehicle gear, short-term single-point missing data is filled with non-abnormal data at the previous moment.

[0069] The high-frequency measurement noise is mainly used for engine speed and engine torque data detection. A Savitzky-Golay filter is designed to process the noise data. The filter is constructed using a quadratic polynomial, and the filter window length is set to 19.

[0070] Discontinuous mutations are mainly aimed at engine torque data detection, and the characteristic detection standards are set as follows:

[0071]

[0072] Among them, R is the engine torque data at a certain moment, T is the timestamp at the corresponding moment, t is the calculation moment, z is the time step, and δ is the set detection threshold; for the data points obtained according to the above detection standards, the linear interpolation method is used to cover the data values ​​at the corresponding moment.

[0073] Furthermore, the implementation process of step (4) is as follows:

[0074] After the vehicle operation data is cleaned, the vehicle operation speed characteristics are obtained using the pre-processed vehicle operation data based on the vehicle parameter attributes and operation characteristics:

[0075]

[0076] Then the vehicle running acceleration and jerk characteristics are obtained:

[0077]

[0078] Where n is the engine speed, r is the wheel rolling radius, i g is the transmission ratio, i0 is the rear axle speed ratio, Δt is the time step, V t is the vehicle speed, a t is the acceleration value and j is the jerk.

[0079] Based on the construction standards and changing characteristics of expressways, the road slope is obtained by using the pre-processed speed data and elevation data:

[0080]

[0081] Where l is the time step of backward calculation; h t is the elevation value at time t, h t+n is the elevation value at time t+n, in meters; v i is the vehicle speed value, in units of

[0082] The point slope calculation is performed according to the distance span of 100m, and the elevation data is guaranteed to change monotonically within a single calculation step, and the time step k satisfies:

[0083]

[0084] (h i -h i-1 )·(h i+1 -h i )>0,(i=t+1,t+2,…,t+n-1)

[0085] Based on the above calculation strategy and formula, all running time period data are traversed to obtain the road slope characteristics, vehicle acceleration and jerk characteristics of the entire journey.

[0086] Furthermore, the load joint identification system of step (5) is designed as follows:

[0087] Based on the vehicle dynamics model, the reasonable distribution range of the estimated load is obtained, and the vehicle longitudinal dynamics model under driving conditions is adopted:

[0088]

[0089] Among them, g is the acceleration of gravity, M is the mass of the car, f is the friction coefficient, α is the road slope, C D is the air resistance coefficient, A is the frontal area of ​​the vehicle, ρ is the air density, σ is the conversion coefficient of the vehicle's moment of inertia, v is the vehicle's speed, a is the vehicle's acceleration, and T tq is the engine torque, i g is the transmission ratio, i0 is the rear axle speed ratio, η T is the total efficiency of the transmission system, r is the rolling radius of the wheel;

[0090] The recursive least squares algorithm with forgetting factor is used to obtain the distribution benchmark of the estimated load based on the above vehicle dynamics model filtering, and the loss function of the filtering algorithm is constructed:

[0091]

[0092] Among them, λ represents the forgetting factor, z(i) represents the actual output of the system, φ(k) represents the system input, Represents the parameter vector to be estimated at time k.

[0093] The load benchmark generation model based on recursive least squares is designed as follows:

[0094]

[0095] in:

[0096]

[0097] φ(k)=gα(k)+gf+σ(k)u

[0098] θ=M

[0099] The load regression identification model is built using the Informer-based neural network architecture. The model input parameters are selected as: vehicle speed, acceleration, jerk, road elevation, road slope, and the model output parameter is the load regression identification result.

[0100] The encoder structure of Informer is designed as follows: encoding feature dimension (120, 7), expanded feature dimension (120, 128), encoder layer number 4, fully connected layer number 256, sampling factor number 10, multi-head attention number 8, output feature dimension (1, 1).

[0101] Furthermore, the implementation process of step (6) is as follows:

[0102] The load distribution output by the recursive least squares algorithm is used to construct a 3σ distribution interval, and the interval average is used as the confidence evaluation benchmark. The confidence of the load regression identification model output of the Informer encoder is evaluated, and MAE is used as the evaluation indicator:

[0103]

[0104] in, is the output result of the load regression identification model, is the average value of the load reference interval;

[0105] The confidence assessment criteria are:

[0106]

[0107] The load regression identification results and confidence evaluation results are iterated with the time series and continuously output along with the vehicle operation until the vehicle working trip ends; the load regression identification results and confidence evaluation results in each time dimension are recorded in the vehicle storage unit;

[0108] After the vehicle's current working trip is completed, the confidence of the load identification results within the entire trip is counted, and the load identification data with a confidence level lower than 10% is marked. The average value of the load identification benchmark at the corresponding moment is used as the fine-tuning label to generate a fine-tuning dataset for the Informer encoder for offline training. After the model fine-tuning is completed, it is redeployed.

[0109] Through the above-mentioned model operation and identification, the load identification result and the confidence assessment value can be output, and the data set generation, model self-learning and model redeployment can be completed after the vehicle journey. The above is only a preferred specific implementation method of the present invention, but the protection scope of the present invention is not limited to this. Any technician familiar with the technical field within the technical scope disclosed by the present invention, according to the technical solution and its inventive concept of the present invention, make equivalent substitutions or changes, which should be covered by the protection scope of the present invention.

Claims

1. A self-learning commercial vehicle load identification and confidence assessment method, characterized in that: The following steps are involved: (1) Vehicle operation initialization: including the vehicle's GPS module, vehicle management system, vehicle communication module, and vehicle static parameter initialization; (2) Multi-source data acquisition: Data acquired through the GPS module include timestamp, elevation, and vehicle speed; data acquired through the CAN bus include engine speed, engine torque, and gear information; (3) Operation data detection and cleaning: Implement corresponding cleaning strategies for erroneous frame data such as missing and noise in the operation data, and then take pre-processing measures such as interpolation and filtering for the operation data that does not meet the continuity conditions; (4) Vehicle operation characteristics expansion: Based on the vehicle system transmission characteristics and road structure characteristics, the pre-processed vehicle operation data is used to obtain the road slope characteristics, vehicle acceleration and jerk characteristics of the entire road section at all times; (5) Design of load-carrying joint identification system: Establish a load-carrying benchmark generation model based on the recursive least squares method, and iterate the vehicle operation data to obtain the distribution benchmark of the estimated load; establish a load-carrying regression identification model based on the improved Informer encoder, and output the load-carrying identification results through GPS source data and extended operation characteristics; (6) Confidence assessment and model fine-tuning: Based on the dynamic load distribution benchmark and the output of the neural network load regression model, the confidence assessment of the load identification results is completed; The vehicle operation data and total weight benchmark data at each moment of the entire journey are stored, and the load regression identification model is trained and fine-tuned offline after the current journey ends. After the training is completed, the model is redeployed.

2. A self-learning commercial vehicle load identification and confidence assessment method according to claim 1, characterized in that: The implementation process of step (3) is as follows: Considering the differences in the operating characteristics of non-homologous data, artificial feature extraction and marking are performed on the error frames in the GPS source data and CAN bus source data respectively, and the error frame data samples are separated based on the continuity criterion. The error frame features and cleaning strategies of multi-source data are formulated as follows; GPS source data: Error frames mainly come from data interruption. Combined with the data feedback timestamp, they can be divided into short-term data loss and long-term signal interruption. The time threshold is set to δ = 30s; For short-term data loss less than the threshold, linear interpolation is performed using the nearest non-erroneous data value before and after the error frame; For long signal interruptions exceeding this threshold, the data during the period is set to zero and discarded; CAN bus source data: Error frame data is mainly characterized by short-term data loss, high-frequency measurement noise and discontinuous mutation; Short-term data missing is mainly for vehicle gear data detection. Based on the principle of continuous change of vehicle gear, short-term single-point missing data is filled with non-abnormal data at the previous moment; High-frequency measurement noise is mainly used for engine speed and engine torque data detection. Savitzky-Golay filter is designed to process the noise data. Discontinuous mutations are mainly aimed at engine torque data detection, and the characteristic detection standards are set as follows: Among them, R is the engine torque data at a certain moment, T is the timestamp at the corresponding moment, t is the calculation moment, z is the time step, and δ is the set detection threshold; for the data points obtained according to the above detection standards, the linear interpolation method is used to cover the data values ​​at the corresponding moment.

3. The self-learning commercial vehicle load identification and confidence assessment method according to claim 1, characterized in that: The implementation process of step (4) is as follows: After the vehicle operation data is cleaned, the vehicle operation speed characteristics are obtained using the pre-processed vehicle operation data based on the vehicle parameter attributes and operation characteristics: Then the vehicle running acceleration and jerk characteristics are obtained: Where n is the engine speed, r is the wheel rolling radius, i g is the transmission ratio, i0 is the rear axle speed ratio, Δt is the time step, V t is the vehicle speed, a t is the acceleration value, j is the jerk; Based on the construction standards and changing characteristics of expressways, the road slope is obtained by using the pre-processed speed data and elevation data: Where l is the time step of backward calculation; h t is the elevation value at time t, h t+n is the elevation value at time t+n, in meters; v i is the vehicle speed value, in units of The point slope calculation is performed according to the distance span of 100m, and the elevation data is guaranteed to change monotonically within a single calculation step, and the time step k satisfies: (h i -h i-1 )·(h i+1 -h i )>0,(i=t+1,t+2,...,t+n-1) Based on the above calculation strategy and formula, all running time period data are traversed to obtain the road slope characteristics, vehicle acceleration and jerk characteristics of the entire journey.

4. The self-learning commercial vehicle load identification and confidence assessment method according to claim 1, characterized in that: The load combined identification system of step (5) is designed as follows: Based on the vehicle dynamics model, the reasonable distribution range of the estimated load is obtained, and the vehicle longitudinal dynamics model under driving conditions is adopted: Among them, g is the acceleration of gravity, M is the mass of the car, f is the friction coefficient, α is the road slope, C D is the air resistance coefficient, A is the frontal area of ​​the vehicle, ρ is the air density, σ is the conversion coefficient of the vehicle's moment of inertia, v is the vehicle's speed, a is the vehicle's acceleration, and T tq is the engine torque, i g is the transmission ratio, i0 is the rear axle speed ratio, η T is the total efficiency of the transmission system, r is the rolling radius of the wheel; The recursive least squares algorithm with forgetting factor is used to obtain the distribution benchmark of the estimated load based on the above vehicle dynamics model filtering, and the loss function of the filtering algorithm is constructed: Among them, λ represents the forgetting factor, z(i) represents the actual output of the system, φ(k) represents the system input, represents the parameter vector to be estimated at time k; The load benchmark generation model based on recursive least squares is designed as follows: A load regression identification model is built using an Informer-based neural network architecture; the model input parameters are selected as: vehicle speed, acceleration, jerk, road elevation, and road slope, and the model output parameter is the load regression identification result.

5. The self-learning commercial vehicle load identification and confidence assessment method according to claim 1, characterized in that: The implementation process of step (6) is as follows: The load distribution output by the recursive least squares algorithm is used to construct a 3σ distribution interval, and the interval average is used as the confidence evaluation benchmark. The confidence of the load regression identification model output of the Informer encoder is evaluated, and MAE is used as the evaluation indicator: in, is the output result of the load regression identification model, is the average value of the load reference interval; The confidence assessment criteria are: The load regression identification results and confidence evaluation results are iterated with the time series and continuously output along with the vehicle operation until the vehicle working trip ends; the load regression identification results and confidence evaluation results in each time dimension are recorded in the vehicle storage unit; After the vehicle's current working trip is completed, the confidence of the load identification results within the entire trip is counted, and the load identification data with a confidence level lower than 10% is marked. The average value of the load identification benchmark at the corresponding moment is used as the fine-tuning label to generate a fine-tuning dataset for the Informer encoder for offline training. After the model fine-tuning is completed, it is redeployed.

Images