A method for improving the metrological accuracy of a gross vehicle weight detection device.

Abstract

This invention discloses a method for improving the metrological accuracy of a total mass detection device for heavy haulage vehicles, belonging to the field of calibration and standardization technology for detection devices. First, the axle load data of the heavy haulage vehicle under different test conditions is used as the dataset for training a deep fitting neural network model, and this dataset is cleaned, augmented, and partitioned. Then, a deep fitting neural network model is constructed, and the model is thoroughly trained on the sample data in the dataset, with training parameters adjusted in real time based on the training progress. Finally, by simply inputting the load data of each axle of the heavy haulage vehicle under test, collected in real time, into the trained deep fitting neural network model, the total mass of the heavy haulage vehicle under test can be predicted. Using this method, identification errors can be effectively reduced, dependence on the identification accuracy of the axle load detection device can be decreased, and the identification accuracy under different vehicle speeds and adverse road conditions can be improved.

Description

Technical Field

[0001] This invention belongs to the field of calibration and standardization technology of testing devices, specifically relating to a method for improving the metrological accuracy of a total mass testing device for heavy-duty transport vehicles based on a deep fitting neural network. Background Technology

[0002] In recent years, with the continuous advancement of power grid upgrades, rail transit construction, new energy equipment deployment, and chemical plant expansion, the scale of oversized and overweight cargo transportation has shown an exponential growth trend. Statistics show that in 2024, the number of oversized and overweight cargo transportation operations nationwide exceeded 1.5 million batches, creating an urgent need for technological upgrades and management optimization of the transportation supervision system. In the oversized and overweight cargo transportation management system, the accurate measurement of the total mass of vehicles and cargo is a core element of supervision. This parameter is not only the legal basis for transportation permit approval but also a key technical indicator for determining overloading violations. In current transportation supervision practice, traditional measurement methods suffer from large measurement errors, leading to prolonged approval cycles, reduced reliability, and consequently affecting the construction progress of related engineering projects. Therefore, it is urgent to build an intelligent measurement system to achieve synergistic optimization of rapid safety assessment and efficient approval. With the breakthrough development of intelligent detection technology, data-driven nonlinear regression models provide an innovative solution for accurate prediction of mass parameters. Summary of the Invention

[0003] To address the problems existing in the prior art, the purpose of this invention is to provide a method for improving the metrological accuracy of a total mass detection device for heavy-duty transport vehicles based on a deep fitting neural network. This invention focuses on the innovative integration of intelligent deep fitting algorithms and heavy-duty axle load detection devices. By establishing a multi-dimensional feature coupling dataset, a metrological accuracy improvement method with adaptive metrological error compensation function is constructed. This method not only solves the problem of low accuracy in traditional metrological methods, but also provides reliable technical support for building intelligent heavy-duty metrological equipment.

[0004] To achieve the above objectives, the present invention adopts the following technical solution:

[0005] A method for improving the metrological accuracy of a gross vehicle weight detection device includes the following steps:

[0006] Step 1: The axle load detection device measures all axle load data of the heavy-duty transport vehicle under different test conditions to form a source dataset with the same axle count information;

[0007] Step 2: After cleaning and data augmentation preprocessing of all sample data in the source dataset, a qualified dataset is formed;

[0008] Step 3: Obtain the total number of load axes in the qualified dataset and construct a deep fitting neural network model that matches the input dimension with the total number of load axes;

[0009] Step 4: Randomly select a portion of the qualified dataset as the training set and use it to train the weight parameters in the deep fitting neural network model;

[0010] Step 5: Adjust training parameters and activation function types in real time to avoid gradient vanishing, gradient exploding, getting stuck in local optima, overfitting, and underfitting problems during the training process of deeply fitted neural network models;

[0011] Step Six: Use a portion of the qualified dataset as the test set. Input the sample data from the test set into the trained deep fitting neural network model and evaluate the relative error between its output value and the label value. If it is less than 1.5%, the deep fitting neural network model training is complete; otherwise, repeat steps four and five.

[0012] Step 7: Extract the weight parameters and obtain their dimensionality information from the deep fitting neural network model trained in Step 6; then use the extracted weight parameters to update the weight parameters of the deep fitting neural network model with the same input dimension in the data processing module of the axle load detection device, to obtain the calibrated axle load detection device, wherein the calibrated axle load detection device is deployed with multiple deep fitting neural network models with different input dimensions.

[0013] Step 8: Match the corresponding dimension of the deep fitting neural network model in the calibrated axle load detection device with the total number of load axles of the heavy-duty transport vehicle to be tested. Then input the axle load data of the heavy-duty transport vehicle to be tested into the successfully matched deep fitting neural network model to obtain the final predicted value of the total mass of the heavy-duty transport vehicle.

[0014] The different test conditions include driving speed, ramp pattern, road surface smoothness, and changes in cargo position.

[0015] The axle load detection device includes a weighing platform, a ramp, and a data processing module equipped with a deep fitting neural network model.

[0016] To match each sample data with a deep fitting neural network model, first determine the total number of load axes in each qualified sample, and the input dimension of the constructed deep fitting neural network model must be the same as this total number.

[0017] The deep fitting neural network model includes one input layer, three hidden layers, and one output layer; wherein, the input layer receives the axle load data of each load axle of the heavy-duty transport vehicle, and the output layer directly outputs the total mass of the heavy-duty transport vehicle.

[0018] The method for randomly selecting a portion of the qualified dataset as the training set is as follows: a simple random sampling algorithm is used to select a portion of the qualified dataset as the training set, and the remainder is used as the test set; wherein, the number of training set samples accounts for 80%-92% of the total number of samples.

[0019] The specific methods for real-time adjustment of training parameters and activation function types are as follows: First, replace the activation function type and use gradient clipping strategies to avoid gradient vanishing or gradient exploding problems; use a dynamic learning rate parameter adjustment mechanism to avoid getting trapped in local optima; and avoid overfitting or underfitting problems by adapting the number of nodes in each hidden layer.

[0020] The activation function is Sigmoid, Tanh, or ReLU.

[0021] Compared with the prior art, the present invention has the following advantages:

[0022] 1. This invention improves the estimation accuracy of the total mass of heavy-duty transport vehicles using traditional axle load detection devices. Currently, the estimation error of the total mass of heavy-duty transport vehicles based on traditional axle load measurement technology is still at a certain level. Within the range. However, this invention collects actual axle load data from tests and inputs iterative training into a deep fitting neural network model, thereby controlling the total mass prediction error to within 2.5%.

[0023] 2. This invention eliminates the crucial steps of calibration and error correction in traditional axle load measurement systems, thus saving production costs to some extent. This is because the embedded deep fitting neural network model can continuously adjust the weight parameters during training, thereby achieving error compensation and calibration. Attached Figure Description

[0024] Figure 1 This is a flowchart illustrating the overall process of improving the measurement accuracy of the total mass detection device for oversized transport vehicles according to the present invention.

[0025] Figure 2 This is a picture of the actual axle load detection device.

[0026] Figure 3 The flowchart shows the algorithm for verifying the compliance of sample data.

[0027] Figure 4 This is a flowchart of a data augmentation algorithm. Detailed Implementation

[0028] like Figure 1 As shown, a method for improving the measurement accuracy of a gross vehicle weight detection device includes the following steps:

[0029] Step 1: Construct a dataset for training and testing, where each sample in the dataset is all axle load data of a large-item transport vehicle obtained from the axle load detection device under different test conditions;

[0030] Step 2: Perform preprocessing operations such as cleaning, data augmentation, and dataset partitioning on all sample data in the dataset to form a qualified dataset;

[0031] Step 3: Obtain the total number of load axes in the qualified dataset and construct a deep fitting neural network model that matches the input dimension with the total number of load axes;

[0032] Step 4: Randomly select a portion of the qualified dataset as the training set and use it to train the weight parameters in the deep fitting neural network model;

[0033] Step 5: Adjust training parameters and activation function types in real time to avoid problems such as gradient vanishing, gradient exploding, getting stuck in local optima, overfitting, and underfitting that occur during the training process of a deeply fitted neural network model.

[0034] Step Six: Use a portion of the qualified dataset as the test set. Input the sample data from the test set into the trained deep fitting neural network model and evaluate the relative error between its output value and the label value. If it is less than 1.5%, the deep fitting neural network model training is complete. Otherwise, repeat steps four and five.

[0035] Step 7: Extract relevant weight parameters and obtain their dimensionality information from the deep fitting neural network model trained in Step 6; then use the exported relevant weight parameters to obtain the weight parameters of the deep fitting neural network model with the same input dimensionality information in the data processing module of the axle load detection device to obtain the calibrated axle load detection device. The calibrated axle load detection device is equipped with multiple deep fitting neural network models with different input dimensions.

[0036] Step 8: Match the corresponding dimension of the deep fitting neural network model in the calibrated axle load detection device with the total number of load axles of the heavy-duty transport vehicle to be tested. Then input the axle load data of the heavy-duty transport vehicle to be tested into the successfully matched deep fitting neural network model to obtain the final predicted value of the total mass of the heavy-duty transport vehicle.

[0037] Example

[0038] 1. Construct datasets for training and testing.

[0039] To ensure sufficient generalization ability of the trained deep fitting neural network model, the diversity of the dataset should be guaranteed during its construction. Therefore, this invention focuses on acquiring relevant sample data from different test environments, such as driving speed, ramp pattern, road surface smoothness, and cargo position. After the raw data collection is completed, data cleaning operations must be performed to remove missing values, duplicate data, and outliers.

[0040] Figure 2 The axle load detection device used for data acquisition in this invention mainly consists of three parts: a ramp, a weighing platform, and a data processing module. The ramp reduces the elevation difference between the weighing platform and the road surface, preventing severe bumps or vibrations when the load axles of the heavy-duty transport vehicle pass over the platform. The weighing sensor in the weighing platform is the U10F dynamic force sensor manufactured by HBM (Germany), with a range of up to 15 tons. To meet the testing requirements of common heavy-duty transport vehicles, 10 U10F dynamic force sensors are installed in parallel inside the weighing platform, resulting in a total range of 150 tons. The data processing module deploys deep fitting neural network models corresponding to the input dimensions for heavy-duty transport vehicles with different numbers of load axles.

[0041] The axle load data of each load axle can be directly obtained from the weighing platform data output interface provided by the axle load detection device, based on the data information of a sample in the source dataset. After all the load axles of the heavy-duty transport vehicle to be tested have passed through the weighing platform, the data information of these axles is packaged into a single sample data. Then, by changing the test conditions, multiple sample data can be obtained.

[0042] For example, keeping the cargo placement position unchanged, and under the same road conditions and a fixed vehicle speed, five samples from the source dataset were obtained from five test environments: no approach ramp, short approach ramp, medium approach ramp, long approach ramp, and ultra-long approach ramp, as shown in Table 1:

[0043] Table 1 (Unit: kg)

[0044]

[0045] The i-th sample uses uppercase letters. Represented. Then, the j-th axle load data in the i-th sample is represented by... To obtain the label value for each sample group, a more precise large-item weighing device is used to directly weigh the total mass of the vehicle to be tested. Since the mass of the loaded cargo remains unchanged, the label values ​​for all samples in Table 1 are as follows. All are the same, denoted as =45000kg.

[0046] This invention continues by keeping the other three environmental conditions constant, and collecting seven samples from the source dataset when the test vehicle randomly passed the weighing platform at a speed of 3-7 kilometers per hour, as shown in Table 2:

[0047] Table 2 (Unit: kg)

[0048]

[0049] Furthermore, different test conditions can be set through permutations and combinations to obtain numerous test sample data under different label data. For example, taking a heavy transport vehicle with a total mass label value of 78,400 kg as the test object, by changing two of the four variables—driving speed, ramp style, road surface smoothness, and cargo position—the following data in Table 3 can be obtained:

[0050] Table 3 (Unit: kg)

[0051]

[0052] Next, this invention further uses a heavy transport vehicle with a total mass label value of 138,000 kg as the test object. At a fixed vehicle speed of 5.1 km / h, different parameters in three aspects—ramp style, road surface smoothness, and cargo position—are combined. After cleaning and data augmentation processing in step 2, the following five sets of training sample data are obtained:

[0053] Table 4 (Unit: kg)

[0054]

[0055] 2. Data cleaning and data augmentation preprocessing

[0056] After obtaining the source dataset, the authenticity of individual axle weight data and the final accumulated total mass data is verified. The data verification algorithm flow is as follows: Figure 3 As shown:

[0057] 1. If the sum of the axle weight data for each sample has a relative error compared to the label value... Data exceeding 13% is considered outlier and its usability requires further verification.

[0058] 2. From the qualified sample data where the label value is still 45,000 kg, randomly select a set of data and calculate the mean of its axle load data. ;

[0059] 3. Place the sample to be verified Each axle load data with the mean Compare and calculate the relative error. ;

[0060] 4. Statistical analysis of the samples to be verified middle The total number greater than 35% ;

[0061] 5. If this total If the value is less than or equal to 2, the nearest-neighbor average replacement algorithm is used to replace the outlier, and the updated sample can be used as a qualified sample; otherwise, the sample to be verified is discarded. .

[0062] It should be noted that for certain goods with uneven centroids (such as wind turbine blades) or high density goods, the load on some axle bearings of heavy-duty transport vehicles may be overly concentrated, thus a larger deviation in axle load data is likely reasonable. Therefore, it is recommended that the error limit for axle load data be set within 25% to 35%. Furthermore, the relative error of conventional axle weighing for measuring the total weight of heavy-duty transport vehicles can currently be controlled within 15%. Therefore, the error limit for total mass can be set between 10% and 20%. After the deep fitting neural network model has been trained on a qualified dataset, more anomalous data with an error limit between 15% and 20% can be used to further train this deep fitting neural network model, which can further improve its generalization and ability to handle unknown disturbances.

[0063] Using the original sample data in Table 2, the calculation process of the outlier detection algorithm is explained in detail. This involves examining the samples to be verified in Table 2. The sum of their axle loads for

[0064] (1)

[0065] Further calculate the sum of axle loads With tag value The relative error R1:

[0066] (2)

[0067] because The value is less than 13% of the specified maximum error parameter; therefore, further verification is needed to determine if each axle load data in this sample is an outlier. It should be noted that, to ensure the reliability of the original test data, this invention manually verifies a set of qualified sample data for different labels. Subsequent sample data with the same label value are then processed as follows: Figure 3The verification algorithm shown automatically verifies the compliance of samples, thereby improving the efficiency and accuracy of dataset collection. Here, we can assume that samples 2-4 in Table 1 are qualified data after manual or automatic verification. Then, if sample 2 in Table 1 is selected as the standard data using a normal distribution random sampling algorithm, its average axle weight can be calculated. Then, calculate the relative deviation between the load data of the first load shaft in Sample 1 and the standard value:

[0068] (3)

[0069] Similarly, the relative errors of other load axes can be calculated as follows: , , , , , , , , , It can be observed that... to The values ​​are all less than or equal to 35%, therefore the data for sample 1 is acceptable.

[0070] Next, continue to verify sample 5 and calculate. With tag value Relative error:

[0071] (4)

[0072] The total mass error can be detected. It is qualified. Continuing to use sample 2 as standard data, the axle load data were verified, and the relative error data for axles 2, 3, 7, and 8 were found to be as follows: , , , All of them significantly exceeded the error limit of 35%. Among them, abnormal data may have been obtained due to sensor malfunction, therefore, sample 5 needs to be discarded.

[0073] As shown in Table 2, due to sensor malfunction or unstable transmission line network, the load data corresponding to axle 5 in sample 2 was lost. In this case, the data dimension of sample 2 changed from 11 to 10. Such data with inconsistent dimensions cannot be directly used to train the neural network. Therefore, to address this situation, the present invention uses an arithmetic mean algorithm to fill in the gaps. That is, summing up the remaining axle load data in the sample and then dividing by the total number of axles yields the following result:

[0074] (5)

[0075] Therefore, the load data for axle 5 in sample 2 will be filled with 3958 kg. Furthermore, numerical verification revealed that the axle load data for each axle in samples 6 and 7 are identical. Therefore, sample 7 is discarded, resulting in 6 samples suitable for training.

[0076] To further expand the number of relevant samples in the dataset, this invention employs data augmentation techniques. Specifically, the randn function in MATLAB software, which generates normally distributed load data for each sample, is used to randomly generate the load data for the relevant axes. The specific calculation process of the data augmentation algorithm is as follows: Figure 4 As shown. First, the j-th axle weight data in sample i is... As the mean parameter of the normal distribution, a larger variance value can be set for sample i, provided that the generated data can pass the automatic verification algorithm. This can further enhance the diversity of training sample data. The variance value recommended in this invention... The allowable range is 5-150. Next, the j-th axle load data in the new sample i. The following formula can be used to calculate:

[0077] (6)

[0078] randn will randomly generate a specific value between 0 and 1. In this way, random numbers randn will be generated again and the next axle load data in the standard sample will be selected as the mean, so that the specific data of the corresponding load axle can be calculated again using formula (6).

[0079] This invention utilizes a data augmentation algorithm to generate five sets of grid sample data as shown in Table 5. Next, using the augmented sample data related to sample 1 in Table 5, the calculation process of the data augmentation algorithm will be described in detail. First, the first axle weight data corresponding to sample 1 in Table 4 is... The mean parameter of the normal distribution is used; then, a random number of 0.8 is generated using the randn function; next, the variance value is... Set to 100; finally, the relevant data of axis 1 corresponding to sample 1 in Table 5 can be calculated using formula (6):

[0080] (7)

[0081] Table 5 (Unit: kg)

[0082]

[0083] 3. Construct a deep fitting neural network model

[0084] This invention designs a deep fitting neural network with three hidden layers. The sample data provided in this embodiment reveals that each sample contains load data for 11 axes. Therefore, the input layer of the deep fitting neural network model is set with 11 nodes, each used to receive the load data for each axis. Furthermore, the node parameters of each hidden layer need to be reasonably determined according to the complexity of the fitting problem. Too many node parameters can easily lead to overfitting, while too few node parameters can easily lead to underfitting. Therefore, the node parameters of each hidden layer need to be continuously adjusted during training. Finally, the number of nodes in the first hidden layer is set to 8, the number of nodes in the second hidden layer is set to 16, and the number of nodes in the third hidden layer is set to 32. It should be noted that the number of nodes in the hidden layers is recommended to be an integer multiple of 2, which helps improve the computational efficiency of the neural network. Since the output layer directly outputs the predicted total mass of the heavy-duty transport vehicle, it can be set to 1.

[0085] 4. Dataset Splitting

[0086] To ensure the representativeness of the extracted training set and reduce sampling error, this invention employs a stratified sampling method based on different total mass label values. This embodiment collected a total of 30 gridded dataset samples, including augmented sample data. Specifically, there were 10 samples with a total mass label value of 45,000 kg, 10 samples with a total mass label value of 78,400 kg, and 10 samples with a total mass label value of 138,000 kg. First, the sampling ratio of the test set to the training set was set to 1:9. Then, one data point was sequentially extracted from each of the sample sets with label values ​​of 45,000 kg, 78,400 kg, and 138,000 kg to form the test set as shown in Table 6. The remaining samples were used as the training set for training the neural network. If the prediction performance of the trained neural network does not meet the given performance metrics, the dataset can be re-splitted and retrained.

[0087] Table 6 (Unit: kg)

[0088]

[0089] 5. Adjust parameters based on training results.

[0090] First, the Levenberg-Marquardt optimization algorithm is used to correct the weight parameters during backpropagation. It offers advantages such as fast convergence and strong real-time performance in handling nonlinear fitting problems. Next, the maximum number of training epochs is set, which is the maximum number of iterations the network can perform during training. A larger epochs setting results in longer training time. Therefore, a smaller value, such as 20, is initially set in the training phase. However, significant errors still exist when testing the trained neural network using a test set, so the epochs are set to 100. Next, the learning rate is set, which mainly controls the step size of network weight adjustments. A larger learning rate can accelerate convergence but may lead to oscillations or instability; a smaller learning rate results in more stable training but slower convergence. This invention sets the learning rate to 0.1. Finally, learning rate decay controls the speed of parameter updates by dynamically adjusting the learning rate. In the early stages of training, a larger learning rate allows for rapid convergence to the vicinity of the optimal solution; as training progresses, gradually decreasing the learning rate allows the network to perform a more refined search near the optimal solution, avoiding missing the optimal solution due to excessively large step sizes. Therefore, in this embodiment, the learning decay rate is set to 0.8. Experiments have shown that training the deep fitting neural network model using the above training parameters yields good training results.

[0091] 6. Predictive performance evaluation

[0092] The predictive performance of the deep fitting neural network model was evaluated using the test set data shown in Table 6. Inputting the axle weight data for each sample in Table 6 into the deep fitting neural network model yielded the following total masses: , =78428.95708kg and =137370.2722kg, as shown in the third row of Table 7. The traditional method for calculating total mass is the sum of all axle load data. Based on the axle load data for each sample in Table 6, the total mass can be calculated. , =72660kg and =136910kg, as shown in the second row of Table 7.

[0093] Table 7 (Unit: kg)

[0094]

[0095] The relative error generated by the total mass prediction algorithm provided by this invention is calculated using the following formula:

[0096] (8)

[0097] in This represents the label value corresponding to sample i. The calculation results are as follows: =0.16%, =0.04%, =0.46%, as shown in the third row of Table 8. It can be seen that each relative error is less than 1.5%, therefore, the approximation fitting capability of this deep fitting neural network model has met the specified performance requirements, and it can be deployed in the data processing module.

[0098] Next, the relative error in calculating the total mass using the traditional algorithm can be calculated using the following formula:

[0099] (9)

[0100] The calculation results are respectively =5.56%, =7.32%, =0.79%, as shown in the second row of Table 8.

[0101] Table 8

[0102]

[0103] For a given test set, an average cumulative error metric can be defined to describe the estimation accuracy of the overall quality estimation algorithm. Taking the test set provided in Table 6 as an example, the cumulative error of the traditional quality estimation algorithm can be calculated as follows:

[0104] (10)

[0105] The cumulative error introduced by the quality estimation algorithm provided by this invention is

[0106] (11)

[0107] Therefore, the accuracy of estimating the total mass of heavy-duty transport vehicles using a deep fitting neural network, as described in this invention, is significantly improved compared to traditional algorithms. =95.18%.

Claims

1. A method for improving the metrological accuracy of a gross vehicle weight detection device, characterized in that: Includes the following steps: Step 1: The axle load detection device measures all axle load data of the heavy-duty transport vehicle under different test conditions to form a source dataset with the same axle count information; Step 2: After cleaning and data augmentation preprocessing of all sample data in the source dataset, a qualified dataset is formed; Step 3: Obtain the total number of load axes in the qualified dataset and construct a deep fitting neural network model that matches the input dimension with the total number of load axes; Step 4: Randomly select a portion of the qualified dataset as the training set and use it to train the weight parameters in the deep fitting neural network model; Step 5: Adjust training parameters and activation function types in real time to avoid gradient vanishing, gradient exploding, getting stuck in local optima, overfitting, and underfitting problems during the training process of deeply fitted neural network models; Step 6: Use a portion of the qualified dataset as the test set, input the sample data from the test set into the trained deep fitting neural network model, and evaluate the relative error between its output value and the label value. If it is less than 1.5%, the deep fitting neural network model training is complete. Otherwise, repeat steps four and five; Step 7: Derive the weight parameters and obtain their dimensionality information from the deep fitting neural network model trained in Step 6; The exported weight parameters are then used to update the weight parameters of the deep fitting neural network model with the same input dimension in the data processing module of the axle load detection device, to obtain a calibrated axle load detection device, wherein the calibrated axle load detection device is deployed with multiple deep fitting neural network models with different input dimensions. Step 8: Match the corresponding dimension of the deep fitting neural network model in the calibrated axle load detection device with the total number of load axles of the heavy-duty transport vehicle to be tested. Then input the axle load data of the heavy-duty transport vehicle to be tested into the successfully matched deep fitting neural network model to obtain the final predicted value of the total mass of the heavy-duty transport vehicle. The different test conditions include driving speed, ramp pattern, road surface smoothness, and changes in cargo position; The deep fitting neural network model includes one input layer, three hidden layers, and one output layer; wherein, the input layer receives the axle load data of each load axle of the heavy-duty transport vehicle, and the output layer directly outputs the total mass of the heavy-duty transport vehicle; The specific methods for real-time adjustment of training parameters and activation function types are as follows: First, replace the activation function type and use gradient clipping strategies to avoid gradient vanishing or gradient exploding problems; use a dynamic learning rate parameter adjustment mechanism to avoid getting trapped in local optima; and avoid overfitting or underfitting problems by adapting the number of nodes in each hidden layer. The activation function is Sigmoid, Tanh, or ReLU.

2. The method for improving the metrological accuracy of the total mass detection device for oversized transport vehicles as described in claim 1, characterized in that: The axle load detection device includes a weighing platform, a ramp, and a data processing module equipped with a deep fitting neural network model.

3. The method for improving the metrological accuracy of the total mass detection device for oversized transport vehicles as described in claim 1, characterized in that: To match each sample data with a deep fitting neural network model, first determine the total number of load axes in each qualified sample, and the input dimension of the constructed deep fitting neural network model must be the same as this total number.

4. The method for improving the metrological accuracy of the total mass detection device for oversized transport vehicles as described in claim 1, characterized in that: The method for randomly selecting a portion of the qualified dataset as the training set is as follows: a simple random sampling algorithm is used to select a portion of the qualified dataset as the training set, and the remainder is used as the test set; wherein, the number of training set samples accounts for 80%-92% of the total number of samples.

Images