Vehicle weight calibration method and cloud service system based on artificial neural network

Through the vehicle load-load cloud service platform based on artificial neural network, combined with the microservice architecture, the problems of heavy load-load measurement and calibration methods and low data transmission efficiency of traditional vehicle load-load measurement and calibration methods are solved, high-precision calibration and real-time data transmission are realized, and the system response is improved Speed ​​and concurrency capabilities.

CN113984175BActive Publication Date: 2024-11-15NORTHEASTERN UNIV AT QINHUANGDAO
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202111245044.7
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2021-10-26
Publication Date
2024-11-15
Estimated Expiration
2041-10-26

AI Technical Summary

Technical Problem

The calibration method of traditional vehicle load capacity measurement sensors is heavy and not accurate enough, and the vehicle cloud service platform has problems such as low efficiency, long response time, and inability to meet high concurrency requirements in terms of data transmission and interactivity.

Method used

Using the vehicle load-load cloud service platform calibration method based on artificial neural network, the calibration model is constructed by obtaining the actual weight and sensor data during the loading process, using the cloud platform to realize real-time data transmission and interaction, and using the microservice architecture to improve the system Flexibility and scalability.

Benefits of technology

It realizes high-precision calibration and real-time data transmission of vehicle load capacity, reduces operation and maintenance costs, improves the system's response speed and concurrency capabilities, and supports multiple users to be online at the same time.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN113984175B_ABST
    Figure CN113984175B_ABST
Patent Text Reader

Abstract

The present invention belongs to the field of vehicle load measurement methods, and in particular, relates to a vehicle load cloud service platform calibration method based on artificial neural network and its architecture. The method collects vehicle data, encodes it, and then sends it to the cloud service platform. The forward proxy program and the back-end data processing program are run in the server of the cloud service platform, and the data is saved in the memory cache database Redis database; the data decoding in the Redis database is read; the network structure of the artificial neural network is constructed, divided into a training set and a test set, and the training data is input into the artificial neural network for training; the test set is input into the trained artificial neural network structure for verification, the accuracy of the load detection is calculated, and the result model of the artificial neural network is optimized and saved in the relational database MySQL database of the corresponding user. When the user sends a request to the cloud service platform, real-time and accurate vehicle load information can be obtained.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the technical field of vehicle load measurement methods, and in particular relates to a vehicle load cloud service platform calibration method based on an artificial neural network and its architecture. Background Art

[0002] The sensor required for measuring vehicle load in the prior art is a vehicle self-load sensor. Because it is greatly affected by the surrounding environment and the detection accuracy is not enough, the sensor needs to be calibrated frequently. The traditional sensor calibration method is to calibrate the sensor offline. First, use instruments and equipment with a certain accuracy to detect the weight of the goods, as an input signal, and input it into the sensor to be calibrated; the output of the sensor is measured by the instrument. Then, the output of the sensor is compared with the input to obtain a series of curves, and then the relationship between the input and output of the sensor is established to describe the relationship curve between the input and output, that is, the calibration curve. Although this traditional calibration method is simple, the workload is very large. Every time the calibration operation is performed, the calibration result needs to be written back to the sensor to be calibrated, and the workload is heavy. Therefore, an artificial neural network model can be used to construct a system for calibrating the vehicle load to improve the accuracy of load detection.

[0003] The artificial neural network model is composed of a large number of nodes (or neurons) connected to each other. Each node represents a specific output function, called an excitation function. The connection between each two nodes represents a weighted value for the signal passing through the connection, called a weight, which is equivalent to the memory of the artificial neural network. With the development of Internet business, the present invention proposes a method for calibrating the vehicle load weight through a cloud service platform to form a system for collecting vehicle load information, and realize interaction with users and real-time transmission of data.

[0004] The main task of the current vehicle cloud service platform for vehicle services is mostly vehicle dispatching. The system for collecting vehicle load information needs to be developed and improved. There are obstacles in the interaction between it and the user. The data cannot be transmitted in real time and the interactivity is poor. As a result, the data transmission speed is not fast enough, the back-end takes a long time to process data, and high data concurrency cannot be achieved. The existing traditional development model, such as web development, is a monolithic development. All functions are packaged in a WAR package, with basically no external dependencies, and deployed in a JEE container (Tomcat, JBoss, WebLogic), including all logic such as DO / DAO, Service, UI, etc. However, the disadvantages are: low efficiency: developers change the code in the same project, waiting for each other, and conflicts continue; difficult maintenance, the code functions are coupled together, and if a fault occurs, you don’t know where to start; inflexible, long build time, a small problem may cause the entire application to paralyze; any small modification requires the reconstruction of the entire project, which is time-consuming and labor-intensive; poor stability and insufficient scalability: unable to meet business needs under high concurrency.

[0005] Therefore, in order to make the system architecture more agile, respond to business needs in a timely manner, improve user experience and reduce costs, the microservice architecture can meet the above requirements. The microservice architecture aims to decouple the solution by decomposing functions into discrete services. Its role is to decompose functions into discrete services, thereby reducing the coupling of the system and providing more flexible service support. It can be understood as: splitting a large single application and service into several or even dozens of supporting microservices, which can expand a single component instead of the entire application stack to meet the service level agreement. During the creation process, applications are created around business domain components, which can be developed, managed and iterated independently. Using cloud architecture and platform-based deployment, management and services in decentralized components makes product delivery simpler. Its essence is to use some services with relatively clear functions and relatively refined business to solve larger and more practical problems. Its characteristics are: dividing the organization according to business rather than technology, forming a system composed of distributed services, automated operation and maintenance, high fault tolerance, rapid evolution and iteration. Summary of the invention

[0006] The purpose of the present invention is to provide a vehicle load cloud service platform calibration method and architecture based on artificial neural network, so as to solve the technical problems that the calibration curve needs to be re-established each time during the traditional calibration operation, which makes the workload heavy, and the transmission speed is not fast enough when using network information, the back-end takes a long time to process data, and it is not possible to transmit transparently in real time, and the interactivity is poor.

[0007] In order to solve the above technical problems, the present invention discloses a vehicle weight calibration method based on artificial neural network, the method comprising:

[0008] Obtain actual weight data of the goods loaded at each stage of the loading process;

[0009] Collect sensor detection data at various stages of the vehicle's loading process;

[0010] Input the actual weight data and the sensor detection data corresponding to each stage of the loading process into the final calibration model to output sensor calibration relationship data;

[0011] The sensor calibration relationship data is analyzed to form a calibration curve which is sent to a terminal and presented to a user.

[0012] Preferably, the final calibration model is trained in the following manner:

[0013] Acquire multiple sets of training data, the training data including actual weight data for training, sensor detection data for training, and corresponding calibration relationship data;

[0014] Construct an artificial neural network structure and input training data for training to obtain the final calibration model.

[0015] Preferably, the acquiring of multiple sets of training data, the training data including actual weight data for training, sensor detection data for training and corresponding calibration relationship data, includes:

[0016] Obtain training actual weight data for each extreme of the multi-pack loading process and training sensor detection data corresponding to each stage;

[0017] Obtain multiple sets of calibration relationship data for training by calculating actual weight data for training and sensor detection data for training;

[0018] The training data includes actual weight data for training, sensor detection data for training and corresponding calibration relationship data.

[0019] Preferably, the input training data is trained to obtain a final calibration model, including:

[0020] Dividing the training data into training set data and test set data;

[0021] Construct an artificial neural network structure and input training set data for training to form an intermediate model;

[0022] The test set is input into the intermediate model for verification, the prediction accuracy is calculated, and the parameters of the artificial neural network are optimized and adjusted through the test set. The final adjusted network model is used as the final calibration model.

[0023] Preferably, the actual weight data and the sensor detection data are uploaded to a cloud platform for storage, and the analysis of the sensor calibration relationship data to form a calibration curve is completed on the gimbal and sent to the user.

[0024] Preferably, the artificial neural network structure can regularly scan the input features, perform matrix element multiplication and summation on the input features to obtain the deviation: where (i,j)∈{0,1,...L l+1}, The summation part in the formula is equivalent to solving a cross correlation; b is the deviation, Z l and Z l+1 represents the artificial input and output of the l+1th layer, also known as the feature map, L l+1 Z l+1 The size of the eigenvalue is assumed to be the same length and width; Z(i,j) corresponds to the length of the eigenvalue, and K is the number of channels of the eigenvalue.

[0025] Preferably, the cloud platform is provided with a forward proxy program Netty program and a backend processing program developed by Spring-boot.

[0026] Preferably, the artificial neural network structure adopts a local perception area, and / or adopts a shared weight method so that repeated units use the same parameter settings, and / or perform downsampling in the spatial domain.

[0027] In addition, the present disclosure also discloses a vehicle-mounted weight calibration cloud service system based on artificial neural network.

[0028] The cloud service system adopts a microservice architecture, including mobile devices and cloud service platforms;

[0029] The mobile device enables relevant personnel to perform corresponding calibration operations through communication services, and receives and stores data on the input-output relationship of the sensor;

[0030] The cloud service platform is used to implement the calibration method described in any one of claims 1 to 8. Preferably, the cloud service platform includes: a user service module, a calibration module and a vehicle information query module to realize the cloud calibration service of the self-loading sensor; the user service module, the calibration module and the vehicle information query module are implemented through the application service Spring-boot framework according to the MVC design method and the modular method, and complete the functions of user service, calibration service and query service, so that users can understand the vehicle load status in real time. .

[0031] The vehicle-mounted cloud service platform calibration method based on artificial neural network and its architecture have the following advantages:

[0032] 1. The calibration method of vehicle load measurement using artificial neural network is used to calibrate the vehicle load through the cloud service platform, forming a system for collecting vehicle load information, realizing interaction with users and real-time transmission of data.

[0033] 2. The Netty program is used in the forward proxy program to achieve high data concurrency; the back-end processing program processes data through the back-end program developed by Spring-boot, which improves cloud throughput and reduces response time.

[0034] 3. Use local sensing areas in the artificial neural network structure to improve the accuracy of training loads; use a shared weight method so that repeated units use the same parameter settings; downsample in the spatial domain so that a complete calibration curve can be obtained with less sampled data.

[0035] 4. Establish a microservice architecture, split the entire system into multiple microservices according to business needs, break it down into parts and implement them one by one, and finally coordinate to complete the functions of the entire system. The modular design of the microservice architecture enables each part of the system to have its own independent functions, making the system easy to maintain, expandable, and highly reusable. Through the mobile client of the Android system, users can view the vehicle's self-load information in real time, solving the problem of poor real-time transmission of information between the server and the user. Compared with traditional monitoring systems, it can cope with higher concurrency capabilities and is more conducive to realizing multiple users online at the same time. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 A flow chart of the cloud service platform of the present invention detecting the load weight of a truck;

[0037] Figure 2 It is an artificial neural network model diagram of the present invention;

[0038] Figure 3 is a sensor calibration curve diagram of the present invention;

[0039] Figure 4 This is a microservice architecture diagram of the present invention;

[0040] Figure 5 This is a throughput test chart of the cloud service platform of the present invention. DETAILED DESCRIPTION

[0041] In order to better understand the purpose, structure and function of the present invention, the vehicle-mounted heavy cloud service platform calibration method based on artificial neural network and its architecture are further described in detail below in conjunction with the accompanying drawings.

[0042] The vehicle-load cloud service platform calibration method based on artificial neural network provided by the present invention is: collect the load data detected by sensors of multiple trucks on site at the same time, encode the data, send the data to the cloud service platform, run the forward proxy program and the back-end data processing program in the server of the cloud service platform, save the data in the memory cache database Redis database, read the data in the Redis database through the back-end processing program, and decode the data. Construct the network structure of the artificial neural network, divide the data into a training set and a test set, use the training data as the input data of the artificial neural network, and train the artificial neural network; input the test set data into the trained artificial neural network structure for verification, calculate the accuracy of load detection, and adjust and optimize the parameters of the result model of the artificial neural network through the test set. Then save the trained model in the relational database MYSQL database of the corresponding user, and when the user sends a request to the cloud service platform to obtain data, the trained result is presented on the front end. The present invention can display the load data of large trucks in real time and accurately.

[0043] like Figure 1-2 As shown, the vehicle-mounted cloud service platform calibration method based on artificial neural network provided by the present invention comprises the following specific steps:

[0044] S1. Start sensor detection, collect sensor detection data during the entire loading process of the vehicle, upload it as the underlying data, and encode the detection data in real time and send it to the cloud service platform. For example: when the vehicle is loading or unloading, the sensor collects data of 0x34 and 0x56, where 0x34 is the high eight bits and 0x56 is the low eight bits, and sends it to the cloud service platform. Use the JT-808 protocol to encode the transmitted data, such as: 7e090000080130818577970007345600000001cf09. Among them: 7e0900 is the start bit, 0008 is the message body attribute; 013081857797 is the mobile phone number; 0007 is the sensor number, 3456 is the monitoring data of the sensor; 000001 is the interval time of 1 second; cf is the data check; the check method is XOR check; 09 is the end bit. In the process of data collection, the JT / T808 protocol is used to formulate the message body of each monitoring data and design the communication protocol.

[0045] S2. After the cloud service platform receives the data, it performs a forward proxy program to verify the check digit of the data to determine whether the data is sent correctly. If it is correct, it is saved in the Redis database. If it is incorrect, it is discarded. Specifically: After the cloud service platform receives the data, it uses the check digit to verify the data. If the check data is 0xcf, the data is saved in the Redis database. If the check is 0xaa, the data is deleted. There are two reasons for incorrect data. One is that the underlying sensor detects an error when it starts measuring data; the second is that the data is received incorrectly during the retransmission process.

[0046] S3. Process the data in the backend, read the data stored in the Redis database, decode the data, store it in the MySQL database, and divide the stored data into a training set and a test set. For example: according to the user's number 013081857797, convert the data 0x34 and 0x56 into decimal data 64.126 and the current load capacity 3.4 tons, and store them in the MySQL database. Repeat the loading or unloading process N times, and the database stores N times of detection data. The N times of data of the current user are randomly divided into a training set and a test set in the form of 7:3.

[0047] S4. Construct an artificial neural network structure and input the training set data for training. During the training process, the input features will be scanned regularly, and the matrix elements will be multiplied and the deviation will be added. where (i,j)∈{0,1,...L l+1}, The summation part in the formula is equivalent to solving a cross correlation. b is the deviation, Z l and Z l+1 represents the artificial input and output of the l+1th layer, also known as the feature map, L l+1 Z l+1 Here, we assume that the eigenvalues ​​have the same length and width. Z(i,j) corresponds to the length of the eigenvalue, and K is the number of channels of the eigenvalue.

[0048] The artificial neural network includes two processes: forward conduction and reverse transmission. S2 of the present invention uses the Netty program as a forward proxy program to achieve high data concurrency; S3 uses the back-end program developed by Spring-boot as a back-end processing program to process data, which improves cloud throughput and reduces response time, so that the load monitoring can be done in real time.

[0049] The back-end processing algorithm adopted by the present invention is a back-propagation algorithm, which is an algorithm applied to hierarchical feedforward ANN. The algorithm adopts the adam optimization algorithm to update the network weights and deviations according to the gradient of the loss function. The artificial neurons of the algorithm are organized and divided according to different levels, and the signals are sent in a forward manner, and then the error rate is propagated to the backward neurons in a feedback manner. The network collects input signals through neurons located in the input layer, and the output value of the network is given by neurons located in the output layer. There may be one or more layers of intermediate hidden layers. The back-propagation algorithm uses supervised learning, that is, providing input values ​​and output values ​​for it to calculate, and calculating the error (the error between the true value and the calculated value). Make its supervised learning as small as possible after learning the training samples. The training starts with the weight value as an arbitrary value, and the purpose is to constantly adjust the weight value to minimize the error.

[0050] The output function represented by each node, i.e. the excitation function, usually adopts a linear rectification function. Sigmod function and hyperbolic tangent function can also be used to complete nonlinear transformation of data in artificial neural network and solve the problem of insufficient expression and classification ability of linear model. The specific parameters of artificial neural network can be designed as:

[0051] The number of input neurons is 100, the number of output neurons is 100, the number of hidden layer neurons is 100, and the activation function is a linear rectifier function. The learning algorithm is Adam, the learning rate is 0.05, the minimum error is 0.03, the training step size is 100, the batch size is 25 groups, and the number of iterations is 100.

[0052] The artificial neural network of the present invention is similar to a biological neural network, which reduces the complexity of the network model and thus reduces the number of weights. The artificial neural network is trained by the back propagation algorithm, and the number of parameter estimates is relatively small. Compared with the original neural network algorithm, it is easier to train and has higher accuracy; local perception area, shared weights and downsampling in the spatial domain are adopted. The use of local perception can make adjacent areas in the load have greater correlation, and the correlation is also smaller when the distance is far, thereby improving the accuracy of training the load. Shared weights are mainly applicable to the process of training load data. After some repeated parameters are input into the artificial neural network, these repeated units can use the same parameter settings, including weights and biases. Shared weights allow images to be directly used as network input, so that the artificial neural network, as a multi-layer perceptron specially designed for recognizing two-dimensional shapes, automatically recognizes image features. This network structure has a high degree of invariance to translation, scaling, tilt or other forms of deformation, which can improve the accuracy and efficiency of image recognition. The downsampling effect in the spatial domain is that in the process of load training, from sensor data to the final curve calibration process, a complete calibration curve can be obtained by sampling less data.

[0053] S5. Input the test set data into the trained artificial neural network model for verification, calculate the prediction accuracy, and optimize the parameters of the artificial neural network through the test set to improve the detection accuracy. The final adjusted network model is used as the final calibration model to form the sensor calibration curve (such as Figure 3 As shown), it is saved in the corresponding user table. If the calculated accuracy is 0.97, the loop operation is repeated 100 times, and the parameters of the artificial neural network need to be optimized and adjusted through the test set to improve the detection accuracy. When the accuracy reaches 0.99, the last adjusted network model is used as the final calibration model and saved in the corresponding user table, such as the data in the following table:

[0054]

[0055] S6. When the user sends a request to the cloud service platform, the cloud service platform decodes and inputs the calculated data into the calibration curve to obtain the current load capacity and presents it to the mobile device. For example, the decoded data 710456.34, which is the decimal data of the sensor measured by the user's vehicle, can be used to obtain the current load capacity of 7.8 tons in the calibration curve. The cloud service platform presents the load capacity data of 7.8 tons to the mobile client.

[0056] The present invention can be calibrated for different vehicle models. The calibration results are stored in the cloud through a cloud service platform, which can reduce the recalibration process when the vehicle replaces the sensor, reduce the maintenance cost, and make the calibration simple and fast.

[0057] Figure 3 The calibration curve diagram has the vertical axis representing the detection data and the horizontal axis representing the load. It can be seen from the figure that the final calibration curve of the load is basically consistent with the monitoring data of the sensor detection device, which meets the basic requirements for truck load detection. It can be seen from the curve in the figure that when the truck is in the process of loading, the error between the test point and the calibration curve is small. Therefore, the final calibration curve can meet the user's needs for real-time acquisition of current load data.

[0058] The present invention also provides an architecture of a vehicle-loaded weight cloud service platform calibration method based on an artificial neural network, which is a microservice architecture that can be applied to a mobile client of an Android system, and enables users to view the vehicle's self-load information in real time through the mobile client of the Android system, thereby solving the problem of poor real-time performance of two-way transmission of information between the server and the user. Compared with traditional monitoring systems, the microservice architecture can cope with higher concurrency capabilities and is more conducive to realizing multiple users online at the same time.

[0059] The sensors for measuring the self-load of the vehicle terminal are selected, installed and tested according to the vehicle structure. The functionality and non-functionality of the monitoring platform are then designed according to the business requirements of the system to form a system monitoring platform with a microservice architecture.

[0060] like Figure 4 As shown in FIG. 1 , the microservice architecture includes a mobile device and a cloud service platform. The mobile device is an Android system mobile client. The Android system mobile client uses the communication service to enable relevant personnel to perform corresponding calibration operations and receive and store the input and output relationship data of the sensor. The communication service is implemented through the Netty framework. The communication service can achieve simultaneous concurrent access of 10,000 terminals and respond to users within 2 seconds (such as Figure 5 As shown). The cloud service platform can realize the cloud calibration service of the self-loading sensor. The cloud service platform includes: a user service module, a calibration module and a vehicle information query module. The user service module, the calibration module and the vehicle information query module are implemented through the application service Spring-boot framework according to the MVC design method and the modular method, and complete the functions of user service, calibration service and query service, so that users can understand the vehicle load status in real time.

[0061] Figure 5The figure below is a throughput test chart of the cloud service platform. The vertical axis is throughput and the horizontal axis is time. Throughput refers to the number of concurrent requests processed by the server in a unit of time. Throughput can describe the ability of a system to process concurrent requests. The larger the throughput, the greater the amount of concurrent requests processed by the system, and the higher the concurrency of the system. The test results show that the throughput of the test environment can reach a maximum of 4500req / s, which means that 4500 concurrent requests can be processed per second, and nearly 10,000 requests can be processed in two seconds. At the same time, due to the gap between the performance of the server configured in the test environment and the performance of the formally running server, when the running server reaches 32 cores and 128G memory, it is easy to process concurrent requests from tens of thousands of terminals, so the throughput of the cloud service platform can meet the user's request response.

[0062] The present invention splits the entire system into multiple microservices according to the needs, breaks the whole into parts and implements them one by one, and finally coordinates to complete the functions of the entire system. The modular design of the system enables each part of the system to have its own independent function, making the system easy to maintain, expandable and highly reusable. In the process of realizing data generation and transmission, the JT / T808 protocol is adopted, and its own message body is formulated according to the protocol to realize the design of its own communication protocol. Based on the collection of data, the Netty framework is adopted to realize the collection service system. Using the distributed system mode, the real-time collection and processing of massive data is realized, and a set of high-availability, expandable and easy-to-maintain application monitoring system is designed; the user's real-time viewing of the vehicle's self-load information is realized through the mobile client of the Android system, which solves the problem of poor real-time performance of the two-way transmission of information between the server and the user. Compared with the traditional monitoring system, the present invention can cope with higher concurrency capabilities and is more conducive to realizing multiple users online at the same time.

[0063] It is to be understood that the present invention is described by some embodiments, and it is known to those skilled in the art that various changes or equivalent substitutions may be made to these features and embodiments without departing from the spirit and scope of the present invention. In addition, under the teachings of the present invention, these features and embodiments may be modified to adapt to specific circumstances and materials without departing from the spirit and scope of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed herein, and all embodiments falling within the scope of the claims of this application are within the scope of protection of the present invention.

Claims

1. A vehicle weight calibration method based on artificial neural network, characterized in that: The method comprises: Obtain actual weight data of the goods loaded at each stage of the loading process; Collect sensor detection data at various stages of the vehicle's loading process; Input the actual weight data and the sensor detection data corresponding to each stage of the loading process into the final calibration model to output sensor calibration relationship data; Analyze the sensor calibration relationship data to form a calibration curve and send it to the terminal to present it to the user; The final calibration model is trained in the following way: Acquire multiple sets of training data, the training data including actual weight data for training, sensor detection data for training, and corresponding calibration relationship data; Construct an artificial neural network structure, input training data for training and obtain the final calibration model; The acquiring of multiple sets of training data, wherein the training data includes actual weight data for training, sensor detection data for training, and corresponding calibration relationship data, includes: Obtain training actual weight data for each extreme of the multi-pack loading process and training sensor detection data corresponding to each stage; Obtain multiple sets of calibration relationship data for training by calculating actual weight data for training and sensor detection data for training; Forming training data including actual weight data for training, sensor detection data for training, and corresponding calibration relationship data; The input training data is trained to obtain the final calibration model, including: Dividing the training data into training set data and test set data; Construct an artificial neural network structure and input training set data for training to form an intermediate model; Input the test set into the intermediate model for verification, calculate the prediction accuracy, and use the test set to optimize the parameters of the artificial neural network, and use the final adjusted network model as the final calibration model; The artificial neural network structure can regularly scan the input features, perform matrix element multiplication and summation of the input features, and add the deviation: where (i,j)∈{0,1,...L l+1 }, The summation part in the formula is equivalent to solving a cross correlation; b is the deviation, Z l and Z l+1 represents the artificial input and output of the l+1th layer, also known as the feature map, L l+1 Z l+1 The size of the eigenvalue is assumed to be the same length and width; Z(i,j) corresponds to the length of the eigenvalue, and K is the number of channels of the eigenvalue.

2. The vehicle weight calibration method according to claim 1, characterized in that: The actual weight data and the sensor detection data are uploaded to a cloud platform for storage, and the analysis of the sensor calibration relationship data to form a calibration curve is completed on the cloud platform and sent to the user.

3. The vehicle weight calibration method according to claim 2, characterized in that: The cloud platform is provided with a forward proxy program Netty program and a backend processing program developed by Spring-boot.

4. The vehicle weight calibration method according to any one of claims 1 to 3, characterized in that: The artificial neural network structure adopts a local perception area and / or adopts a shared weight method so that repeated units use the same parameter settings and / or perform downsampling in the spatial domain.

5. A vehicle-mounted weight calibration cloud service system based on artificial neural network, characterized in that: The cloud service system adopts a microservice architecture, including mobile devices and cloud service platforms; The mobile device enables relevant personnel to perform corresponding calibration operations through communication services, and receives and stores data on the input-output relationship of the sensor; The cloud service platform is used to implement the calibration method described in any one of claims 1-4.

6. The vehicle-mounted weight calibration cloud service system as claimed in claim 5, characterized in that: The cloud service platform includes: a user service module, a calibration module and a vehicle information query module to realize the cloud calibration service of the self-loading sensor; the user service module, the calibration module and the vehicle information query module are implemented through the application service Spring-boot framework according to the MVC design method and the modular method, and complete the functions of user service, calibration service and query service, so that users can understand the vehicle load status in real time.