Weighing apparatus calibration method, apparatus and system, and storage medium

Abstract

The invention discloses a weighing apparatus calibration method, device and system and a storage medium, which are used for calibrating weighing apparatus errors. The method comprises the following steps: extracting measuring weighing apparatuses of all stations in a coal transportation system as nodes, extracting transportation paths among all stations as directed edges, constructing a coal transportation system diagram framework according to the nodes and the directed edges, endowing the coal transportation system diagram framework with global dependency relationships and context information of the nodes, and determining the coal transportation system diagram framework according to the global dependency relationships and the context information of the nodes. Obtaining a coal transportation system diagram network; predicting the error of each node weighing apparatus and the confidence of the error according to the coal transportation system graph network; and according to the error of each node weighing apparatus and the confidence coefficient of the error, calibrating the error of the node weighing apparatus. According to the scheme, the global dependency relationship and the context information are considered, the problems of transmission and accumulation of metering errors in the coal transportation process are solved, the consistency of upstream and downstream metering data is ensured, and the accuracy of coal transportation metering is improved.

Description

Technical Field

[0001] This application relates to the field of weighing instrument calibration technology, and in particular to a weighing instrument calibration method, apparatus, system and storage medium. Background Technology

[0002] As a vital energy source, the accuracy of coal measurement during transportation directly impacts the interests of both trading partners. However, due to the complex and variable transportation environment and inconsistent accuracy of measuring equipment, weight differences often occur between upstream and downstream measurement points, leading to the propagation and accumulation of measurement errors. Traditional independent calibration methods are insufficient for global optimization and cannot effectively address the problem of error propagation.

[0003] Therefore, how to provide a weighing instrument calibration method to improve measurement accuracy and data consistency has become an urgent technical problem to be solved. Summary of the Invention

[0004] This application provides a weighing instrument calibration method, apparatus, system, and storage medium to improve measurement accuracy and data consistency.

[0005] This application provides a method for calibrating a weighing instrument, including:

[0006] The weighing instruments at all stations in the coal transportation system are extracted as nodes. The node attributes include at least one of the following: equipment accuracy, historical measurement readings, geographical location, and environmental parameters.

[0007] Extract the transportation routes between all stations as directed edges, where the directed edge attributes include at least one of the following: transportation mode, transportation distance, loss rate, transportation time, and historical statistical data related to the transportation route;

[0008] Construct a coal transportation system graph framework based on the nodes and directed edges;

[0009] Assign global dependencies and context information to the nodes of the coal transportation system graph framework to obtain the coal transportation system graph network;

[0010] The errors of the weighing instruments at each node and the confidence level of the errors are predicted based on the coal transportation system diagram network.

[0011] The errors of the node weighing instruments are calibrated based on the errors of each node weighing instrument and the confidence level of the errors.

[0012] The beneficial effects of this application are as follows: This application extracts the weighing instruments at all stations in the coal transportation system as nodes, extracts the transportation paths between all stations as directed edges, constructs a coal transportation system graph framework based on the nodes and directed edges, and assigns global dependencies and contextual information to the nodes of the coal transportation system graph framework to obtain a coal transportation system graph network; predicts the error of each node's weighing instrument and the confidence level of the error based on the coal transportation system graph network; and calibrates the error of the node's weighing instrument based on the error of each node's weighing instrument and the confidence level of the error. By constructing a graph network, not only are the attributes of the nodes themselves considered, but global dependencies and contextual information are also incorporated, effectively solving the problem of the transmission and accumulation of measurement errors in the coal transportation process. This achieves global dependency modeling and error prediction, realizes dynamic capture and accurate calibration of measurement errors in the coal transportation process, ensures the consistency of upstream and downstream measurement data, and improves the accuracy of coal transportation measurement.

[0013] In one embodiment, the global dependencies and context information of nodes are assigned to the coal transportation system graph framework, including:

[0014] Obtain the global dependencies of nodes using a graph neural network;

[0015] The context data acquisition mechanism collects information that affects measurement errors, as well as specific error information, as context information.

[0016] The global dependencies are assigned to the nodes in the coal transportation system graph framework, and the information affecting the measurement error and the specific error information are integrated into the node attributes and / or the edge attributes of the directed edges.

[0017] In one embodiment, obtaining the global dependencies of nodes through a graph neural network includes:

[0018] Graph neural networks are used to exchange information and features between each node and its neighboring nodes in multiple rounds.

[0019] During multiple rounds of information exchange, the embedded representation of each node is continuously updated, so that the information of each node records the characteristics and information of some nodes or global nodes, in order to establish the dependency relationship between each node.

[0020] Once the dependencies between nodes are established, the graph neural network outputs node representation vectors containing the dependency information of the global network.

[0021] In one embodiment, the method further includes:

[0022] Determine the distance between the node representation vector of each node and the preset normal pattern, wherein the distance includes at least one of Euclidean distance, cosine similarity and Mahalanobis distance;

[0023] When the distance between the node representation vector of a target node and the preset normal mode is greater than a preset distance threshold, it is determined that the weighing instrument corresponding to the target node has a measurement deviation.

[0024] In one embodiment, predicting the error of each node's weighing instrument and the confidence level of the error based on the coal transportation system network includes:

[0025] Obtain a pre-established error prediction model;

[0026] Preset information containing global dependencies is obtained from the coal transportation system graph network and input into the error prediction model;

[0027] Obtain the error prediction values ​​of each node output by the error prediction model.

[0028] In one embodiment, calibrating the error of each node weighing instrument based on the error of each node weighing instrument and the confidence level of the error includes:

[0029] The calibration amount of the weighing instrument is determined based on the error of each node and the confidence level of the error;

[0030] A corresponding calibration factor is generated based on the calculated adjustment amount for each node, and the calibration factor includes an additive calibration factor and a multiplicative calibration factor.

[0031] The above calibration factors are applied to the original readings of each weighing instrument in the current batch to obtain the calibrated measurement values.

[0032] In one embodiment, the method further includes:

[0033] Verify whether the adjusted data closure error meets the standard.

[0034] When the adjusted data closure error exceeds the preset threshold, it is determined that the adjusted data closure error has not met the standard, and the manual verification mechanism is triggered.

[0035] In one embodiment, the coal transportation system network is dynamically updated based on the actual coal transportation process, and the method further includes:

[0036] When the data increment in the coal transportation system network reaches the preset increment threshold, the error prediction model is incrementally updated based on the new data to update the prediction model parameters.

[0037] The node confidence weights are dynamically adjusted based on the updated prediction model parameters.

[0038] This application also provides a weighing instrument calibration device, comprising:

[0039] The first extraction module is used to extract the weighing instruments of all stations in the coal transportation system as nodes. The node attributes include at least one of the following: equipment accuracy, historical measurement readings, geographical location, and environmental parameters.

[0040] The second extraction module is used to extract the transportation paths between all stations as directed edges, wherein the directed edge attributes include at least one of the following: transportation mode, transportation distance, loss rate, transportation time, and historical statistical data related to the transportation path.

[0041] The construction module is used to construct a coal transportation system graph framework based on the nodes and directed edges;

[0042] The assignment module is used to assign global dependencies and context information to the nodes of the coal transportation system graph framework to obtain the coal transportation system graph network.

[0043] The prediction module is used to predict the error of the weighing instruments at each node and the confidence level of the error based on the coal transportation system diagram network.

[0044] The calibration module is used to calibrate the error of each node weighing instrument based on the error of each node weighing instrument and the confidence level of the error.

[0045] In one embodiment, the assigning module includes:

[0046] The first acquisition submodule is used to obtain the global dependencies of nodes through a graph neural network;

[0047] The data acquisition submodule is used to collect information that affects measurement errors and specific error information as context information through a context data acquisition mechanism.

[0048] The submodule is used to assign the global dependency to the nodes in the coal transportation system graph framework, and to integrate the information affecting the measurement error and the specific error information into the node attributes and / or the edge attributes of the directed edges.

[0049] In one embodiment, the acquisition submodule is further configured to:

[0050] Graph neural networks are used to exchange information and features between each node and its neighboring nodes in multiple rounds.

[0051] During multiple rounds of information exchange, the embedded representation of each node is continuously updated, so that the information of each node records the characteristics and information of some nodes or global nodes, in order to establish the dependency relationship between each node.

[0052] Once the dependencies between nodes are established, the graph neural network outputs node representation vectors containing the dependency information of the global network.

[0053] In one embodiment, the acquisition submodule is further configured to:

[0054] Determine the distance between the node representation vector of each node and the preset normal pattern, wherein the distance includes at least one of Euclidean distance, cosine similarity and Mahalanobis distance;

[0055] When the distance between the node representation vector of a target node and the preset normal mode is greater than a preset distance threshold, it is determined that the weighing instrument corresponding to the target node has a measurement deviation.

[0056] In one embodiment, the prediction module includes:

[0057] The second acquisition submodule is used to acquire the pre-established error prediction model;

[0058] The input submodule is used to obtain preset information containing global dependencies from the coal transportation system graph network and input it into the error prediction model;

[0059] The third acquisition submodule is used to acquire the error prediction values ​​of each node output by the error prediction model.

[0060] In one embodiment, the calibration module includes:

[0061] The determination submodule is used to determine the calibration amount of the weighing instrument based on the error of each node weighing instrument and the confidence level of the error;

[0062] A generation submodule is used to generate corresponding calibration factors based on the calculated adjustment amount of each node. The calibration factors include additive calibration factors and multiplicative calibration factors.

[0063] The calibration submodule is used to apply the above calibration factors to the original readings of each weighing instrument in the current batch to obtain the calibrated measurement value.

[0064] In one embodiment, the apparatus further includes:

[0065] The verification module is used to verify whether the adjusted data closure error meets the standard.

[0066] The trigger module is used to determine that the adjusted data closure error has not met the standard when it exceeds the preset threshold, and to trigger the manual verification mechanism.

[0067] In one embodiment, the coal transportation system network is dynamically updated based on the actual coal transportation process, and the device further includes:

[0068] The update module is used to incrementally update the error prediction model based on the newly added data when the data increment in the coal transportation system network reaches a preset increment threshold, so as to update the prediction model parameters.

[0069] The adjustment module is used to dynamically adjust the node confidence weights based on the updated prediction model parameters.

[0070] This application also provides a weighing instrument calibration system, including:

[0071] At least one processor; and,

[0072] A memory communicatively connected to the at least one processor; wherein,

[0073] The memory stores instructions that can be executed by the at least one processor to implement the weighing instrument calibration method described in any of the above embodiments.

[0074] This application also provides a computer-readable storage medium, which, when the instructions in the storage medium are executed by the processor corresponding to the weighing calibration system, enables the weighing calibration system to implement the weighing calibration method described in any of the above embodiments.

[0075] Other features and advantages of this application will be set forth in the following description and will be apparent in part from the description or may be learned by practicing the application. The objectives and other advantages of this application may be realized and obtained by means of the structures particularly pointed out in the written description, claims, and drawings.

[0076] The technical solution of this application will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0077] The accompanying drawings are provided to further illustrate the present application and form part of the specification. They are used together with the embodiments of the present application to explain the application and do not constitute a limitation thereof. In the drawings:

[0078] Figure 1 This is a flowchart of a weighing instrument calibration method according to an embodiment of this application;

[0079] Figure 2 This is a schematic diagram of the structure of a weighing instrument calibration device according to an embodiment of this application;

[0080] Figure 3 This is a schematic diagram of the hardware structure of a weighing instrument calibration system according to an embodiment of this application. Detailed Implementation

[0081] The preferred embodiments of this application are described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit this application.

[0082] Figure 1This is a flowchart of a weighing instrument calibration method according to an embodiment of this application, as follows: Figure 1 As shown, the method can be implemented as follows: S101-S106:

[0083] In step S101, the weighing instruments at all stations in the coal transportation system are extracted as nodes, wherein the node attributes include at least one of the following: equipment accuracy, historical measurement readings, geographical location, and environmental parameters.

[0084] In step S102, the transportation paths between all stations are extracted as directed edges, wherein the attributes of the directed edges include at least one of the following: transportation mode, transportation distance, loss rate, transportation time, and historical statistical data related to the transportation path.

[0085] In step S103, a coal transportation system graph framework is constructed based on the nodes and directed edges;

[0086] In step S104, global dependencies and context information are assigned to the nodes of the coal transportation system graph framework to obtain the coal transportation system graph network.

[0087] In step S105, the error of each node weighing instrument and the confidence level of the error are predicted based on the coal transportation system diagram network.

[0088] In step S106, the error of each node weighing instrument is calibrated based on the error of each node weighing instrument and the confidence level of the error.

[0089] In this application, all weighing instruments at various stations in the coal transportation system are extracted as nodes, such as mine track scales and power plant weighbridges. Each node attribute includes at least one of the following key information: equipment accuracy, historical measurement readings, historical calibration coefficients, geographical location, and environmental parameters (such as temperature and humidity range). These attributes describe the state of the weighing instrument itself and its context, providing fundamental data support for subsequent error analysis and calibration. The transportation paths between all stations are extracted as directed edges to represent the actual transportation process of coal from one metering point (e.g., the shipping end) to another (e.g., the receiving end). Each edge has a weight and attributes to quantify the impact of the transportation process on measurement. The directed edge attributes include at least one of the following: transportation mode (truck, rail, conveyor belt, etc.), transportation distance, normal transportation loss rate, transportation time, and historical statistical data related to the transportation path (e.g., average weight per batch, typical error level). These attributes quantify the impact of the transportation process on measurement, providing key inputs for constructing the error propagation model. Based on the above definitions of nodes and directed edges, a coal transportation system graph framework is constructed. The coal transportation system diagram framework illustrates the relationships between various measurement points in the coal transportation measurement network and the error propagation paths.

[0090] Furthermore, the coal transportation system graph framework is assigned global dependencies and contextual information to its nodes to obtain the coal transportation system graph network. Within this graph model framework, the system needs to collect, process, and extract features from multi-source data from actual operations to provide training and real-time input for machine learning error prediction models. The main data sources and their processing are as follows:

[0091] (1) Measurement readings: This includes the weighing records of each weighing instrument during each batch of coal transportation. For all relevant weighing instruments involved in a single transportation, an ordered measurement sequence is collected (e.g., a batch of coal weighs 80 tons at weighing instrument A upstream, 79 tons at weighing instrument B midway, and 78 tons at weighing instrument C downstream). These data are used to calculate the actual upstream and downstream differences, serving as the basis for error analysis.

[0092] (2) Transportation route information: Data such as the route, distance, time and transportation mode of each batch of transportation are obtained from the transportation management system. Some information (such as distance and mode) has been fixed in the edge attributes of the graph model and can be obtained by association through path ID; dynamic information such as actual transportation time and stops along the way are also collected through IoT devices.

[0093] (3) Environment and Equipment Status: Collect environmental sensor data (temperature, humidity, weather conditions) during transportation and weighing, as well as the status of the weighing equipment itself (sensor zero drift, equipment usage time, last calibration date, daily maintenance records, etc.). This data is acquired through sensing devices installed on-site or the equipment self-testing system and uploaded periodically.

[0094] (4) Transportation Batch Attributes: Obtain information related to transportation batches, such as coal type (calorific value, moisture content, particle size, etc.), vehicle type and number (e.g., truck model, cargo box volume), and loading volume of each batch. These attributes are recorded through the logistics management system and used to analyze the impact of certain batch characteristics on measurement errors.

[0095] Data Preprocessing and Feature Construction: The collected raw data is often messy and of varying dimensions, requiring cleaning and transformation to construct the feature vectors needed for model training. First, based on a unique transport batch identifier, the measurement records of upstream and downstream weighing instruments for the same batch of coal are correlated and matched to ensure that they correspond to the same transport event for analyzing differences in upstream and downstream readings. Then, combining business experience and error influencing factors, a multi-dimensional feature set is designed, for example:

[0096] (1) Measurement difference characteristics: The total error or loss of the batch is measured by directly calculating the difference between the readings of the upstream and downstream weighing instruments (such as the upstream weight minus the downstream weight); and the relative error rate (difference divided by the upstream weight) is used to standardize the error range of different scales of transportation.

[0097] (2) Transportation route characteristics: These include the distance of the shipment (continuous numerical value), the duration of the shipment (which can be discretized into time periods or directly used as continuous values), and the type of transportation mode (using one-hot encoding or numerical encoding: e.g., highway truck = 1, train = 2, conveyor belt = 3, etc.). These characteristics reflect the typical impact patterns of the transportation process on errors.

[0098] (3) Environmental operating conditions: Continuous environmental data such as temperature and humidity during transportation / weighing are used as feature inputs (normalization can be performed as needed to prevent inconsistencies in dimensions), and discrete information such as weather conditions is encoded (e.g., sunny day = 0, rainy day = 1, snowy day = 2) to quantify the impact of extreme weather on measurement.

[0099] (4) Equipment status characteristics: Relevant indicators of the weighing instrument, such as the time interval (in days) since the last calibration, equipment model or sensor type (which can be categorized and coded), current sensor temperature drift value, power supply voltage fluctuations, etc. These characteristics are used to capture the impact of equipment performance changes on reading accuracy. During data processing, numerical equipment characteristics are standardized to avoid imbalance of feature weights caused by different units of measurement.

[0100] (5) Coal batch characteristics: The properties of coal itself may affect measurement differences. For example, high moisture content in coal increases volatile matter loss, and particle size affects loading and unloading losses. Information such as coal type, moisture percentage, and particle size distribution are encoded as features to enrich the model's ability to predict errors for special batches.

[0101] After extraction and encoding, the aforementioned features are concatenated in a predetermined order to form the model input vector. To improve the effectiveness of model training, continuous numerical features are normalized or standardized (e.g., scaling values ​​to the 0-1 range, or subtracting the mean and dividing by the standard deviation to make its distribution mean 0) to accelerate convergence and avoid bias caused by differences in units of measurement. Classification features are represented using one-hot encoding or learned embedding vectors so that the model can understand discrete category information. Finally, the system constructs a feature matrix X (each row corresponding to a feature vector of a transportation batch or measurement event) and a corresponding label vector y (each row corresponding to the actual error or calibration deviation "true value" of that batch). This paired (X,y) dataset will be used for subsequent training and validation of the error prediction model.

[0102] Through comprehensive multi-source data integration and feature engineering, the input to the error prediction model is ensured to possess comprehensive information and high quality. Compared to traditional methods that rely solely on single weighing instrument readings for simple calibration, this solution's data processing module fully leverages historical big data to extract deep-seated correlation features, significantly improving the model's ability to understand the error generation mechanism. This data-driven feature extraction process lays a solid foundation for subsequent machine learning predictions.

[0103] By introducing a Graph Neural Network (GNN), the system's ability to process graph-structured data is enhanced, enabling it not only to store network relationships but also to learn the implicit error propagation patterns between weighing instruments. This is impossible to achieve in traditional linear calibration methods. Through GNN, the measurement points are no longer isolated but can "sense" each other's data changes, significantly improving the accuracy and robustness of error identification from an engineering perspective. The application of this method in complex industrial metrology networks represents an innovative attempt, effectively distinguishing it from existing calibration methods that rely on empirical formulas.

[0104] After collecting and processing the data, global dependencies are assigned to the nodes of the coal transportation system graph framework to capture the correlations between multiple weighing instrument values. For example, global dependencies can be obtained through graph neural networks. Specifically, graph neural networks represent the practical problem as an iterative update and message passing process of node features. Based on its own attributes and information from neighboring nodes, the graph neural network facilitates multiple rounds of information and feature exchange between nodes and their neighbors. The information and feature exchange rules are as follows: each node is assigned an initial feature vector containing its attribute information (such as equipment accuracy, historical metering readings, etc.); in each round of exchange, each node sends its current feature vector to all its neighboring nodes and receives feature vectors from its neighboring nodes; each node updates its own feature vector based on the received neighboring node feature vectors and its own current feature vector using an aggregation function (such as mean aggregation, maximum aggregation, or attention mechanism); the above neighboring node information aggregation and feature update steps are repeated for multiple iterations to gradually capture the complex dependencies between nodes. When the maximum number of iterations is reached, iteration stops to prevent overfitting or wasting computational resources due to excessive iterations. Simultaneously, the changes in node feature vectors after each iteration are monitored. When the change in feature vectors is less than a certain preset threshold, the network is considered to have converged, and iteration stops. The change can be measured by calculating the Euclidean distance or cosine similarity between feature vectors. During multiple rounds of information exchange, the embedding representation of each node is continuously updated, so that the information of each node records the features and information of some nodes or global nodes, thereby establishing the dependencies between nodes.

[0105] Once the dependencies between nodes are established, the graph neural network outputs node representation vectors containing dependency information for the entire network. These vectors encapsulate the location information of each node within the global network and its association with other nodes, which can be used for subsequent error analysis and anomaly detection. Furthermore, the distance between each node's representation vector and a preset normal pattern can be determined. This distance includes at least one of Euclidean distance, cosine similarity, and Mahalanobis distance, and the difference is quantified by calculating the distance between the node representation vector and the preset normal pattern vector. When the distance between the node representation vector of a target node and the preset normal pattern is greater than a preset distance threshold, it is determined that the weighing instrument corresponding to the target node has a measurement deviation. If the embedding of a node deviates from the normal pattern, it may indicate an abnormal deviation in the weighing instrument. The node representations output by the graph neural network can also serve as input features for subsequent error prediction or calibration decisions, thereby improving the model's ability to characterize complex relational structures. Furthermore, through a contextual data acquisition mechanism, information affecting measurement errors and specific error information are collected as contextual information. This information includes, but is not limited to, transportation path characteristics (such as stations along the route and slope), climate environment (such as temperature, humidity, and rainfall), operating conditions (such as loading and unloading methods and time consumption), and equipment operating status (such as sensor temperature and power supply voltage stability). The global dependencies are assigned to nodes in the coal transportation system graph framework, and the information affecting measurement errors and specific error information are integrated into node attributes and / or the edge attributes of directed edges to enrich the dimensions of error modeling and improve the accuracy of error prediction. In model implementation, there are two ways to integrate this data into node / edge attributes: one is to directly feed it as input feature vectors into the machine learning model; the other is to use it as additional features in message passing within the GNN to influence node state updates.

[0106] The system predicts the errors of weighing instruments at each node and their confidence levels based on the coal transportation system graph network. Specifically, a pre-established error prediction model is acquired; this model can be pre-established using algorithms such as XGBoost or LSTM. A graph neural network is used to learn relationships and output enhanced features. Based on the constructed graph network and feature data, the system needs to establish a measurement error prediction model to dynamically model and predict the reading errors of upstream and downstream weighing instruments during each transportation process, so that real-time correction can be achieved through a dynamic calibration algorithm. For example, mature supervised machine learning algorithms can be used to train and model historical measurement data to form the ability to predict errors. Candidate models include:

[0107] (1) Gradient Boosting Decision Tree Model (XGBoost): This decision tree ensemble algorithm performs exceptionally well in processing structured data, automatically capturing the nonlinear relationships and high-order interactions between features and targets. Utilizing the aforementioned extracted multidimensional features, XGBoost can fit the complex mapping relationship between upstream and downstream reading errors and influencing factors. Its advantages include fast training speed, insensitivity to missing and outliers, and provision of feature importance metrics, which is helpful in understanding which factors contribute significantly to the error.

[0108] (2) Long Short-Term Memory Neural Network (LSTM): If the error exhibits temporal correlation or a long-term trend, a recurrent neural network like LSTM, suitable for time series data, can be introduced. By incorporating time sequence into the training data (e.g., sorting by shipping batch time), LSTM can learn the long-term dependency pattern of the error over time, capturing potential seasonal variations or system shifts caused by equipment aging. The error prediction output by LSTM, combined with the current context, can improve the estimation accuracy of errors in future batches.

[0109] (3) Graph Neural Network Models: Given the characteristics of graph structures, dedicated graph learning models can be explored to predict the deviations of each node. For example, graph convolutional networks can be used to directly regress the offset of each weighing node relative to a certain benchmark, or graph attention mechanisms can be used to identify the links most likely to produce abnormal errors. These models can complement the aforementioned XGBoost or LSTM, providing a prediction scheme that better fits the graph structure when the amount of data is large enough and the relationships are complex.

[0110] During the training phase, the prepared feature matrix X is input into the selected model, and the corresponding label y is used as the supervision signal. The model parameters are continuously adjusted to make the predicted output approximate the true error value. The key is to clearly define the model's prediction target (label) to ensure that the output can be used to guide calibration decisions. Possible target definitions include:

[0111] (1) Absolute Error Prediction: The total difference between upstream and downstream measurements of a batch of coal is used as the label. That is, absolute error = upstream measurement value - downstream measurement value (assuming no loss under ideal conditions, this difference is regarded as the total error). Based on this, the model learns to predict how much weight will be lost or gained along this path under given conditions.

[0112] (2) Single-point deviation prediction: The reading offset of a specific weighing instrument node is used as a label. For example, a reference weighing instrument reading that has been validated over a long period of time in the network is selected as the benchmark, and the deviation value of each other weighing instrument relative to the benchmark is calculated. That is, single-point deviation = current weighing instrument measurement value - reference weighing instrument measurement value. After the model is trained, it can predict the degree to which the reading of each weighing instrument deviates from the normal value under different conditions.

[0113] (3) Error Confidence Prediction: The labels are designed as classification or scoring problems to determine whether the measurement difference of a batch is within the normal range. For example, when the upstream and downstream differences are within the historical statistical range, they are marked as 0 (normal), and when they exceed the threshold, they are marked as 1 (abnormal); or the error magnitude is mapped to a confidence score between 0 and 1, indicating the degree to which the batch readings are considered reliable. The model output can then be used as the basis for confidence weights (see the calibration strategy in the next section).

[0114] Model training can employ supervised learning, adjusting model parameters by minimizing prediction errors (such as mean squared error or log loss). Training data should cover various transportation routes, environmental conditions, and equipment statuses to ensure good generalization ability. After training, the model can be used for online prediction: whenever a batch of transport occurs and relevant weighing instruments generate readings, the system extracts relevant features in real time and inputs them into the model to obtain a predicted value for the measurement error of that batch. For example, the model might output a result indicating "the estimated upstream total weight deviates from the downstream by +1.5 tons" (a positive value indicates that the upstream reading is too high), or "the confidence level of this batch of data is only 70%, and there may be abnormal deviations." These prediction results will serve as an important basis for the next step of dynamic calibration decisions.

[0115] After training, the error prediction model can output the predicted error values ​​for each node. Therefore, after model construction is complete, preset information containing global dependencies is obtained from the coal transportation system graph network and input into the error prediction model. This information includes node attributes, directed edge attributes, and node representation vectors calculated through a graph neural network. This allows the acquisition of the predicted error values ​​for each node output by the error prediction model, reflecting the potential measurement errors at each weighing node under given conditions, providing a quantitative basis for subsequent dynamic calibration.

[0116] The system calibrates the errors of each node weighing instrument based on their errors and confidence levels. With the estimation results and confidence information provided by the error prediction model, the system enters the core dynamic calibration decision-making stage. Specifically, the calibration amount of the weighing instrument is determined based on the errors and confidence levels of each node weighing instrument; a corresponding calibration factor is generated based on the calculated adjustment amount for each node. This calibration factor includes additive and multiplicative calibration factors, depending on the calibration requirements. The additive calibration factor is used to correct zero-point deviation; that is, when the weighing instrument reading is systematically too high or too low, the reading is adjusted by adding or subtracting a fixed value. For example, the additive calibration factor is determined using the following formula:

[0117] F add =-E;

[0118] Among them, F addis the additive calibration factor; E is the predicted error value of the weighing instrument. If the predicted error value is positive, it means the weighing instrument reading is too high, and the error value needs to be subtracted; if the predicted error value is negative, it means the weighing instrument reading is too low, and the absolute value of the error value needs to be added.

[0119] The multiplicative calibration factor is used to correct proportional errors. It primarily corrects for proportional errors in weighing instruments; that is, when there is a proportional deviation between the weighing instrument reading and the true value, the reading is adjusted by multiplying it by a coefficient. For example, the additive calibration factor can be determined using the following formula:

[0120]

[0121] Among them, F mul Here, E is the multiplication calibration factor; E is the predicted error value of the weighing instrument; and M is the nominal full-scale value or reference value of the weighing instrument. This formula assumes that the error is proportional to the reading, and the reading is corrected by adjusting the proportionality coefficient. When the error is positive, it indicates that the reading is too high, and the multiplication factor is less than 1 to lower the reading; when the error is negative, it indicates that the reading is too low, and the multiplication factor is greater than 1 to raise the reading.

[0122] The above calibration factors are applied to the original readings of each weighing instrument in the current batch to obtain the calibrated measurement values. In practical applications, additive calibration factors, multiplicative calibration factors, or a combination of both can be selected based on the specific characteristics and error mode of the weighing instrument. For example, the calculated calibration factors can be applied to the original readings of each weighing instrument in the current batch to obtain the calibrated measurement values:

[0123] M c =M r ×F mul +F add

[0124] Among them, M r The original reading of the weighing instrument; M c The calibrated measurement value; F mul F is the multiplication calibration factor; add This is the additive calibration factor.

[0125] Then check whether the corrected upstream and downstream readings tend to be consistent, that is, whether the difference is within the allowable error range. If there is still an error, it may be necessary to iterate or introduce constraints to redistribute the error until the consistency requirements are met.

[0126] During execution, the network relationships in the graph model and the historical reliability of each weighing instrument can be used to automatically calculate the calibration factor that needs to be applied to each node, thereby achieving the unification of measurement readings from multiple locations and error cancellation. The execution of the dynamic calibration algorithm includes the following steps:

[0127] (1) Error Aggregation Analysis: After a batch of coal has been weighed by multiple weighing instruments in the network, the system first summarizes the error prediction results of each point in the batch and analyzes the overall closure error. Assume the model predicts that the upstream weighing instrument A reading is ΔA tons higher and the downstream weighing instrument B reading is ΔB tons lower (positive values ​​indicate overweight, negative values ​​indicate underweight), and the estimated normal transportation loss along the route is ε tons. These predicted values ​​are combined to calculate the total closure error Δ = ΔA + ΔB - ε. Ideally, if the model is accurate and the transportation loss estimate is reasonable, Δ ≈ 0 (meaning the excess weight upstream is exactly equal to the underweight weight downstream plus the loss). If the calculation shows that Δ is significantly non-zero, it indicates that a systematic bias still exists: this may be due to a bias in the model prediction or additional errors caused by unconsidered factors. At this point, the next step is needed: combining confidence weights to identify and correct the source of error.

[0128] (2) Calculation of Confidence Level of Weighing Instrument Nodes: The system assesses the measurement reliability of each relevant weighing instrument based on its historical performance and converts the confidence level into a quantitative weighting coefficient. Specifically, a weight w is assigned to each node i participating in the batch measurement. i (Values ​​range from 0 to 1), reflecting the reliability of the reading. The weight design ensures that the sum of the weights of all nodes is 1. Weighing instruments with high confidence (such as those with long-term stable measurements and minimal differences from other points) are assigned higher weights, indicating that their readings are more reliable; conversely, weighing instruments with large historical error fluctuations have lower weights. The weight calculation method can be based on statistical indicators, such as letting w i Proportional to the reciprocal of the node's error variance or the historical score of calibration accuracy. Through standardization, the current set of weight assignments for upstream and downstream (or points in a multi-node link) is obtained {w}. i}

[0129] (3) Error Allocation and Adjustment Calculation: For the total predicted deviation Δ, error is allocated and readings are adjusted according to the weight of each node. The principle is "the lower the confidence level, the larger the adjustment." Taking a simple upstream and downstream node as an example, if Δ = upstream measurement value - downstream measurement value - ε represents the difference between the upstream and downstream readings (by directional agreement, if the upstream is higher, then Δ > 0; if the upstream is lower, then Δ < 0), then the upstream adjustment amount ΔA and the downstream adjustment amount ΔB are calculated as follows:

[0130] ΔA=-Δ*w B ; ΔB=Δ*w A .

[0131] The negative sign is used to achieve hedging, for example, when the upstream reading is higher than the downstream (Δ>0) and the upstream weighing instrument A is relatively reliable (w A Larger, w B If the value is relatively small, then the upstream reading only needs to be slightly reduced by ΔA (due to w). B (small), the downstream reading needs to be adjusted upwards by ΔB accordingly (because w)A Larger values ​​make ΔB close to most of Δ. Conversely, if the downstream weighing instrument is more reliable, then w B If the error is large, the formula will result in ΔA being close to Δ (significant upstream adjustment) while ΔB is small (minor downstream adjustment). For transportation routes involving multiple nodes (such as intermediate links), the error allocation can be generalized as: let the adjustment amount for each node be Δi = -Δ*w 其他 The specific allocation ratio is determined according to the relative size of each node's weight across all nodes. This inverse confidence allocation method ensures that the error is mainly borne by unreliable readings, thus making the entire network more consistent after adjustment.

[0132] (4) Calibration Factor Generation: Based on the calculated adjustment amount Δi for each node, a corresponding calibration factor is generated and prepared for application to the original readings. The calibration factor can take either an additive offset or a multiplicative coefficient form, depending on the calibration requirements: for most deviations such as zero-point drift in weighing instrument readings, an additive calibration factor is more direct (e.g., subtracting 2 tons); for measurement scaling errors, a multiplicative factor can also be introduced (e.g., multiplying the reading by 0.99). In this scheme, additive factors are mostly used to eliminate fixed deviations between batches, and multiplicative factors are combined to correct systematic errors when necessary. Let the adjustment amount for a node be Δi, then the corresponding additive calibration factor f... i =-Δ i (The reason for using a negative value is that if the calculated Δi is positive, it means the original reading needs to be increased by that amount. However, the calibration factor is defined as original reading + factor = correction value, therefore the factor should be a negative adjustment.) For example, if the upstream weighing instrument A is too high and needs to be reduced by 1 ton, then a calibration factor f is generated. A =-(-1 ton)=+1 ton (meaning that a correction of +1 ton should be added to subsequent readings, which is essentially equivalent to reducing the original reading by 1 ton). The system outputs such calibration parameters for each weighing instrument that needs adjustment, preparing for automatic calibration.

[0133] (5) Data Adjustment and Consistency Check: Apply the above calibration factors to the original readings of each weighing instrument in the current batch to obtain the corrected measurement value. For example, if the original record of upstream A is 80.0 tons, calculate f A = -0.5 tons, then the adjusted measurement value is 79.5 tons; downstream B's original record is 78.0 tons, f B= +0.5 tons, the adjusted value is 78.5 tons. After adjustment, the system rechecks the consistency of upstream and downstream readings: ideally, after allocation calibration, the difference between relevant readings should be within the allowable error range (e.g., ±0.1 tons or a threshold determined according to the accuracy class). If residual differences exceeding the allowable range are found after correction, iterative calibration (recalculating error allocation or adjusting confidence weights) can be considered until convergence. In addition, constraints can be introduced to prevent oscillations, such as limiting the number of iterations, or triggering anomaly handling procedures if the residuals are still large after adjustment (see next step). This check ensures that the data after calibration reaches a closed equilibrium, providing a unified and reliable metrological result for subsequent data reporting and settlement.

[0134] (6) Handling of Abnormal Situations: During dynamic calibration, if the model prediction results or adjusted data show abnormalities, the protection mechanism should be triggered promptly. Abnormal situations include: the required adjustment amount is too large and exceeds the reasonable range (for example, a batch requires an adjustment of more than 5% of the weight, far exceeding the normal measurement error level), or multiple iterations fail to converge, or the confidence level given by the model is extremely low. In response to abnormalities, the system takes two-level measures: First, it limits the single correction amplitude and sets a threshold to ensure that any calibration factor does not exceed the upper limit of the equipment's allowable adjustment, so as to avoid over-correction and the introduction of new errors; second, it marks the batch or related weighing instruments, indicating that manual intervention is required for investigation. At this time, the personnel can check whether there are non-measuring reasons (such as actual loss of coal during transportation, equipment failure) that cause abnormal readings, and repair or recalibrate the equipment. If the equipment is confirmed to be abnormal, the system can temporarily remove the data of that node and calibrate other nodes, or use weight reduction to skip the trust of abnormal nodes. Through the above abnormal handling mechanism, the stability and security of the system during automatic operation are ensured, and blind adjustments will not lead to greater deviations when encountering extreme situations.

[0135] During dynamic calibration, if the model prediction results or adjusted data are abnormal (such as the required adjustment amount being too large and exceeding the reasonable range), the system should promptly trigger the protection mechanism. The protection mechanism includes limiting the single calibration amplitude to not exceed a certain threshold, marking the batch as abnormal and prompting manual intervention for investigation, etc. Through the manual verification mechanism, it is possible to check whether there are non-measuring error reasons (such as special circumstances such as lost coal during the journey), so as to ensure the accuracy and reliability of the calibration results.

[0136] Furthermore, the adjusted data closure error is verified to ensure it meets the standard. The adjusted data includes the calibrated measurement values. In the upstream and downstream metering scenarios of coal transportation, theoretically, from the starting point to the end point, after metering at each stage, the final total metering result should satisfy a certain logical relationship, i.e., the data should be closed. For example, starting from the loading metering at the coal mine, passing through multiple transfer stations, and finally arriving at the power plant for unloading metering, if the metering at each stage is accurate, then the total loading volume minus reasonable losses during transportation should equal the total unloading volume. The adjusted data closure error is this difference value actually calculated. Taking a simple linear transportation process as an example, suppose coal is transported from coal mine A to power plant B, passing through transfer station C. The loading metering value at coal mine A is M. A The measurement value of transfer station C is M. C The unloading measurement value of power plant B is M. B If losses are not considered during transportation, theoretically M A =M C =M B Data closure error ΔM=|M A -M B If transportation losses are considered, assuming a loss rate of r and a transportation distance of d, then the theoretical unloading measurement value M of power plant B is... B理论 =M A ×(1-r×d), data closure error ΔM=|M B理论 -M B When calculating data closure error, it is necessary to fully consider all factors that may affect the measurement results. Besides transportation losses, these include the accuracy error of the measuring equipment, environmental factors (such as the effects of temperature and humidity on the measuring equipment), and coal spillage or contamination during transportation. For example, in high-temperature environments, coal may lose weight due to moisture evaporation, but this reduction is not necessarily due to transportation losses but rather environmental factors, and must be properly distinguished and handled when calculating data closure error.

[0137] Once it is determined that the adjusted data closure error fails to meet the standard, in order to ensure the accuracy and reliability of the measurement results and avoid losses to both trading parties due to data errors, the system will trigger a manual verification mechanism. In one embodiment, when the adjusted data closure error exceeds a preset threshold, it is determined that the adjusted data closure error fails to meet the standard, and the manual verification mechanism is triggered. The preset threshold is determined comprehensively based on factors such as actual coal transportation business needs, the accuracy of metering equipment, and historical data statistical analysis. It represents the maximum allowable range of data closure error under normal transportation and measurement conditions. The preset threshold is not static and a dynamic adjustment mechanism should be established. As coal transportation business changes, metering equipment is upgraded, and environmental factors change, the preset threshold should be evaluated and adjusted in a timely manner. For example, by collecting a large amount of historical coal transportation measurement data, including measurement data under different transportation routes, different seasons, and different transportation modes, statistical analysis is performed on this data to calculate the distribution of data closure error, such as the mean and standard deviation. Based on the statistical analysis results and combined with the business requirements for measurement accuracy, a reasonable preset threshold is determined. For example, if historical data shows that the data closure error is less than a certain value in 95% of cases, then this value can be used as a reference for a preset threshold. When the adjusted data closure error exceeds the preset threshold, it indicates that the calibrated data may still have a large error, or there may be other unknown anomalies, causing the data to fail to meet the expected consistency requirements. Therefore, it is determined that the adjusted data closure error has not met the standard.

[0138] The dynamic calibration network based on graph neural networks in this application has self-learning and evolution capabilities, and can continuously optimize its model and strategy over time and with data accumulation, enabling the system to maintain excellent performance under different transportation structures and operating conditions.

[0139] (1) Online Model Update: The system periodically retrains or incrementally updates the error prediction model using newly collected transportation batch data, adjusting its parameters according to the latest data distribution. For example, the model is refreshed after a certain number of batches or when significant seasonal changes occur, and the weights of the XGBoost decision tree or LSTM are fine-tuned using recent data. This ensures that the model always fits the current equipment status and environmental conditions, avoiding post-training aging.

[0140] (2) Adaptive Calibration Rules: As long-term operational data of each weighing instrument accumulates, the system dynamically adjusts the confidence assessment of each node. For example, if a weighing instrument's performance improves after repair and its error decreases in new data, the system will automatically increase its confidence weight to reflect its improved reliability. Conversely, if a node gradually ages, leading to increased deviation, its weight will be reduced, prompting the algorithm to increase the correction force for that node. The weight adjustment mechanism can be implemented through rolling calculations of error variance, recent calibration residuals, and other indicators, thereby ensuring that the calibration allocation strategy evolves synchronously with the actual equipment situation.

[0141] (3) Accumulation of anomaly experience: Each triggered anomaly handling (such as a batch being manually confirmed to have special losses) will also be recorded in the knowledge base for future model training and decision-making reference. By learning these anomaly patterns, the system can identify and adopt conservative calibration strategies or directly issue warnings for manual intervention when encountering batches with similar characteristics in the future, thereby continuously improving the accuracy of anomaly detection.

[0142] (4) Adaptability to Diverse Operating Conditions: Since coal transportation may involve changes such as new routes, vehicle replacements, and the introduction of new weighing instruments, the self-learning mechanism ensures that the network has scalability and adaptability. When a new weighing instrument node or path is added, the system can evaluate its error characteristics based on the initial calibration and a small amount of operational data and incorporate it into the global calibration framework; for extreme conditions that have not been experienced for a long time (such as rare weather or special coal quality), the model will adjust its parameters after the first experience, so that it can handle better when it is encountered again. In this way, the entire calibration network continuously improves itself and always maintains a high level of consistency.

[0143] This solution achieves globally synchronized multi-point calibration through dynamic calibration, a feat traditionally difficult to accomplish. In the past, each weighing instrument was typically calibrated independently at fixed intervals, often resulting in manual adjustments afterward when data inconsistencies arose. In contrast, this solution integrates the calibration process into the workflow in real-time through an error propagation model and confidence weights, automatically balancing upstream and downstream readings and ensuring data closure at the system level. A series of algorithmic steps ensure reasonable and stable calibration decisions and include fallback mechanisms for anomalies, enhancing the reliability of engineering applications. Particularly noteworthy is that each weighing instrument serves as a reference within this network, no longer relying on isolated standard weights for calibration, thus achieving intelligent collaboration within the metrology system. This method significantly reduces the frequency of manual intervention and downtime calibration, improving the automation and intelligence of cross-regional, multi-stage coal metering, representing a major improvement over existing technologies.

[0144] In one embodiment, the coal transportation system network is dynamically updated based on the actual coal transportation process. Specifically, when the data increment in the coal transportation system network reaches a preset increment threshold, the error prediction model is incrementally updated based on the new data to update the prediction model parameters. Incremental updates ensure that the model always fits the current equipment status and environmental conditions, avoiding aging problems after training. The confidence weights of nodes are dynamically adjusted based on the updated prediction model parameters. As long-term operating data of each weighing instrument accumulates, the system dynamically corrects the confidence assessment of each node. For example, if a weighing instrument's performance improves after repair and its error decreases in new data, the system will automatically increase its confidence weight; conversely, if a node gradually ages, leading to an increase in deviation, its weight will be reduced. The weight adjustment mechanism is implemented through rolling calculation of indicators such as error variance and recent calibration residuals, ensuring that the calibration allocation strategy evolves synchronously with the actual equipment situation.

[0145] The self-learning evolution mechanism upgrades this solution from a one-off solution to a continuously optimizing intelligent system. Traditional calibration techniques lack adaptability, while this invention, through data-driven model updates and rule optimization, enables the calibration network to "grow" alongside real-world scenarios. This ensures the system's effectiveness over long periods of operation, reduces repetitive manual calibration and parameter tuning, and demonstrates significant advantages in intelligence and automation. Compared to existing technologies, this mechanism endows the system with evolutionary and adaptive attributes, representing a crucial innovation in achieving stable metrology across various scenarios.

[0146] This application utilizes a dynamic calibration network construction method based on graph computing and machine learning. This method not only considers the attributes of the nodes themselves but also incorporates global dependencies and contextual information. It effectively solves the problem of the transmission and accumulation of measurement errors during coal transportation, realizes dynamic calibration of measurement data across links and regions, improves the accuracy and reliability of coal transportation measurement, and enhances measurement precision and data consistency.

[0147] The beneficial effects of this application are as follows: This application extracts the weighing instruments at all stations in the coal transportation system as nodes, extracts the transportation paths between all stations as directed edges, constructs a coal transportation system graph framework based on the nodes and directed edges, and assigns global dependencies and contextual information to the nodes of the coal transportation system graph framework to obtain a coal transportation system graph network; predicts the error of each node's weighing instrument and the confidence level of the error based on the coal transportation system graph network; and calibrates the error of the node's weighing instrument based on the error of each node's weighing instrument and the confidence level of the error. By constructing a graph network, not only are the attributes of the nodes themselves considered, but global dependencies and contextual information are also incorporated, effectively solving the problem of the transmission and accumulation of measurement errors in the coal transportation process. This achieves global dependency modeling and error prediction, realizes dynamic capture and accurate calibration of measurement errors in the coal transportation process, ensures the consistency of upstream and downstream measurement data, and improves the accuracy of coal transportation measurement.

[0148] In one embodiment, step S104 above can be implemented as steps A1-A3 as follows:

[0149] In step A1, the global dependencies of nodes are obtained through a graph neural network;

[0150] In step A2, information affecting measurement error and specific error information are collected as context information through a context data acquisition mechanism.

[0151] In step A3, the global dependency is assigned to the nodes in the coal transportation system graph framework, and the information affecting the measurement error and the specific error information are integrated into the node attributes and / or the edge attributes of the directed edges.

[0152] In one embodiment, step A1 above can be implemented as steps A11-A13 as follows:

[0153] In step A11, a graph neural network is used to exchange information and features between each node and its neighboring nodes in multiple rounds.

[0154] In step A12, during the multi-round information exchange process, the embedded representation of each node is continuously updated so that the information of each node records the characteristics and information of some nodes or global nodes, so as to establish the dependency relationship between each node.

[0155] In step A13, when the dependency relationships of each node are established, the graph neural network outputs a node representation vector containing the dependency relationship information of the global network.

[0156] In one embodiment, the method may also be implemented as follows: steps A14-A15:

[0157] In step A14, the distance between the node representation vector of each node and the preset normal pattern is determined, wherein the distance includes at least one of Euclidean distance, cosine similarity and Mahalanobis distance;

[0158] In step A15, when the distance between the node representation vector of the target node and the preset normal mode is greater than a preset distance threshold, it is determined that the weighing instrument corresponding to the target node has a measurement deviation.

[0159] In one embodiment, step S105 above can be implemented as steps B1-B3 as follows:

[0160] In step B1, the pre-established error prediction model is obtained;

[0161] In step B2, preset information containing global dependencies is obtained from the coal transportation system graph network and input into the error prediction model;

[0162] In step B3, the error prediction values ​​of each node output by the error prediction model are obtained.

[0163] In one embodiment, step S106 above can be implemented as steps C1-C3 as follows:

[0164] In step C1, the calibration amount of the weighing instrument is determined based on the error of each node weighing instrument and the confidence level of the error;

[0165] In step C2, a corresponding calibration factor is generated based on the calculated adjustment amount for each node. The calibration factor includes an additive calibration factor and a multiplicative calibration factor.

[0166] In step C3, the above calibration factor is applied to the original readings of each weighing instrument in the current batch to obtain the calibrated measurement value.

[0167] In one embodiment, the method may also be implemented as follows: steps D1-D2:

[0168] In step D1, verify whether the adjusted data closure error meets the standard;

[0169] In step D2, when the adjusted data closure error exceeds the preset threshold, it is determined that the adjusted data closure error has not met the standard, and the manual verification mechanism is triggered.

[0170] In one embodiment, the coal transportation system network is dynamically updated based on the actual coal transportation process, and the method can also be implemented as the following steps E1-E2:

[0171] In step E1, when the data increment in the coal transportation system network reaches a preset increment threshold, the error prediction model is incrementally updated based on the new data to update the prediction model parameters.

[0172] In step E2, the node confidence weights are dynamically adjusted based on the updated prediction model parameters.

[0173] Figure 2 This is a schematic diagram of the structure of a weighing instrument calibration device according to one embodiment of this application, as shown below. Figure 2 As shown, the device includes:

[0174] The first extraction module 201 is used to extract the weighing instruments of all stations in the coal transportation system as nodes. The node attributes include at least one of the following: equipment accuracy, historical measurement readings, geographical location, and environmental parameters.

[0175] The second extraction module 202 is used to extract the transportation paths between all stations as directed edges, wherein the directed edge attributes include at least one of the following: transportation mode, transportation distance, loss rate, transportation time, and historical statistical data related to the transportation path.

[0176] Module 203 is used to construct a coal transportation system graph framework based on the nodes and directed edges;

[0177] The assignment module 204 is used to assign global dependencies and context information to the nodes of the coal transportation system graph framework to obtain the coal transportation system graph network.

[0178] The prediction module 205 is used to predict the error of the weighing instruments at each node and the confidence level of the error based on the coal transportation system diagram network.

[0179] The calibration module 206 is used to calibrate the error of each node weighing instrument based on the error of each node weighing instrument and the confidence level of the error.

[0180] In one embodiment, the assigning module includes:

[0181] The first acquisition submodule is used to obtain the global dependencies of nodes through a graph neural network;

[0182] The data acquisition submodule is used to collect information that affects measurement errors and specific error information as context information through a context data acquisition mechanism.

[0183] The submodule is used to assign the global dependency to the nodes in the coal transportation system graph framework, and to integrate the information affecting the measurement error and the specific error information into the node attributes and / or the edge attributes of the directed edges.

[0184] In one embodiment, the acquisition submodule is further configured to:

[0185] Graph neural networks are used to exchange information and features between each node and its neighboring nodes in multiple rounds.

[0186] During multiple rounds of information exchange, the embedded representation of each node is continuously updated, so that the information of each node records the characteristics and information of some nodes or global nodes, in order to establish the dependency relationship between each node.

[0187] Once the dependencies between nodes are established, the graph neural network outputs node representation vectors containing the dependency information of the global network.

[0188] In one embodiment, the acquisition submodule is further configured to:

[0189] Determine the distance between the node representation vector of each node and the preset normal pattern, wherein the distance includes at least one of Euclidean distance, cosine similarity and Mahalanobis distance;

[0190] When the distance between the node representation vector of a target node and the preset normal mode is greater than a preset distance threshold, it is determined that the weighing instrument corresponding to the target node has a measurement deviation.

[0191] In one embodiment, the prediction module includes:

[0192] The second acquisition submodule is used to acquire the pre-established error prediction model;

[0193] The input submodule is used to obtain preset information containing global dependencies from the coal transportation system graph network and input it into the error prediction model;

[0194] The third acquisition submodule is used to acquire the error prediction values ​​of each node output by the error prediction model.

[0195] In one embodiment, the calibration module includes:

[0196] The determination submodule is used to determine the calibration amount of the weighing instrument based on the error of each node weighing instrument and the confidence level of the error;

[0197] A generation submodule is used to generate corresponding calibration factors based on the calculated adjustment amount of each node. The calibration factors include additive calibration factors and multiplicative calibration factors.

[0198] The calibration submodule is used to apply the above calibration factors to the original readings of each weighing instrument in the current batch to obtain the calibrated measurement value.

[0199] In one embodiment, the apparatus further includes:

[0200] The verification module is used to verify whether the adjusted data closure error meets the standard.

[0201] The trigger module is used to determine that the adjusted data closure error has not met the standard when it exceeds the preset threshold, and to trigger the manual verification mechanism.

[0202] In one embodiment, the coal transportation system network is dynamically updated based on the actual coal transportation process, and the device further includes:

[0203] The update module is used to incrementally update the error prediction model based on the newly added data when the data increment in the coal transportation system network reaches a preset increment threshold, so as to update the prediction model parameters.

[0204] The adjustment module is used to dynamically adjust the node confidence weights based on the updated prediction model parameters.

[0205] Figure 3 This is a schematic diagram of the hardware structure of a weighing instrument calibration system according to one embodiment of this application, as shown below. Figure 3 As shown, the weighing instrument calibration system includes:

[0206] At least one processor 320; and,

[0207] Memory 304 communicatively connected to the at least one processor 320; wherein,

[0208] The memory 304 stores instructions that can be executed by the at least one processor 320 to implement the weighing instrument calibration method described in any of the above embodiments.

[0209] Reference Figure 3The weighing calibration system 300 may include one or more of the following components: a processing component 302, a memory 304, a power supply component 306, a multimedia component 308, an audio component 310, an input / output (I / O) interface 312, a sensor component 314, and a communication component 316.

[0210] Processing component 302 typically controls the overall operation of the weighing instrument calibration system 300. Processing component 302 may include one or more processors 320 to execute instructions to complete all or part of the steps of the method described above. Furthermore, processing component 302 may include one or more modules to facilitate interaction between processing component 302 and other components. For example, processing component 302 may include a multimedia module to facilitate interaction between multimedia component 308 and processing component 302.

[0211] Memory 304 is configured to store various types of data to support the operation of the weighing instrument calibration system 300. Examples of this data include instructions for any application or method operating on the weighing instrument calibration system 300, such as text, images, videos, etc. Memory 304 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0212] Power supply assembly 306 provides power to various components of the weighing calibration system 300. Power supply assembly 306 may include a power management system, one or more power supplies, and other components associated with generating, managing, and distributing power to the vehicle control system 300.

[0213] The multimedia component 308 includes a screen that provides an output interface between the weighing calibration system 300 and the user. In some embodiments, the screen may include a liquid crystal display (LCD) and a touch panel (TP). If the screen includes a touch panel, the screen may be implemented as a touchscreen to receive input signals from the user. The touch panel includes one or more touch sensors to sense touches, swipes, and gestures on the touch panel. The touch sensors may sense not only the boundaries of touch or swipe actions but also the duration and pressure associated with the touch or swipe operation. In some embodiments, the multimedia component 308 may also include a front-facing camera and / or a rear-facing camera. When the weighing calibration system 300 is in an operating mode, such as a shooting mode or a video mode, the front-facing camera and / or the rear-facing camera may receive external multimedia data. Each front-facing camera and rear-facing camera may be a fixed optical lens system or have focal length and optical zoom capabilities.

[0214] Audio component 310 is configured to output and / or input audio signals. For example, audio component 310 includes a microphone (MIC) configured to receive external audio signals when the weighing calibration system 300 is in an operating mode, such as alarm mode, recording mode, voice recognition mode, and voice output mode. The received audio signals may be further stored in memory 304 or transmitted via communication component 316. In some embodiments, audio component 310 also includes a speaker for outputting audio signals.

[0215] I / O interface 312 provides an interface between processing component 302 and peripheral interface modules, such as keyboards, click wheels, buttons, etc. These buttons may include, but are not limited to, home buttons, volume buttons, power buttons, and lock buttons.

[0216] Sensor assembly 314 includes one or more sensors for providing status assessments of various aspects of the weighing calibration system 300. For example, sensor assembly 314 may include a sound sensor. Additionally, sensor assembly 314 can detect the on / off state of the weighing calibration system 300, the relative positioning of components (e.g., the display and keypad of the weighing calibration system 300), and the operating state of the weighing calibration system 300 or one of its components, such as the operating state of the air distribution plate, structural state, the operating state of the discharge scraper, the orientation or acceleration / deceleration of the weighing calibration system 300, and temperature changes of the weighing calibration system 300. Sensor assembly 314 may include a proximity sensor configured to detect the presence of nearby objects without any physical contact. Sensor assembly 314 may also include an optical sensor, such as a CMOS or CCD image sensor, for use in imaging applications. In some embodiments, sensor assembly 314 may also include an accelerometer, a gyroscope, a magnetic sensor, a pressure sensor, a material buildup thickness sensor, or a temperature sensor.

[0217] Communication component 316 is configured to enable the weighing instrument calibration system 300 to provide wired or wireless communication capabilities with other devices and cloud platforms. The weighing instrument calibration system 300 can access wireless networks based on communication standards, such as WiFi, 2G, or 3G, or combinations thereof. In one exemplary embodiment, communication component 316 receives broadcast signals or broadcast-related information from an external broadcast management system via a broadcast channel. In one exemplary embodiment, communication component 316 also includes a near-field communication (NFC) module to facilitate short-range communication. For example, the NFC module may be implemented based on radio frequency identification (RFID) technology, Infrared Data Association (IrDA) technology, ultra-wideband (UWB) technology, Bluetooth (BT) technology, and other technologies.

[0218] In an exemplary embodiment, the weighing instrument calibration system 300 may be implemented by one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to perform the weighing instrument calibration method described in any of the above embodiments.

[0219] This application also provides a computer-readable storage medium, which, when the instructions in the storage medium are executed by the processor corresponding to the weighing calibration system, enables the weighing calibration system to implement the weighing calibration method described in any of the above embodiments.

[0220] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage and optical storage) containing computer-usable program code.

[0221] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0222] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0223] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0224] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. A method for calibrating a weighing instrument, characterized in that, include: The weighing instruments at all stations in the coal transportation system are extracted as nodes. The node attributes include at least one of the following: equipment accuracy, historical measurement readings, geographical location, and environmental parameters. Extract the transportation routes between all stations as directed edges, where the directed edge attributes include at least one of the following: transportation mode, transportation distance, loss rate, transportation time, and historical statistical data related to the transportation route; Construct a coal transportation system graph framework based on the nodes and directed edges; Assign global dependencies and context information to the nodes of the coal transportation system graph framework to obtain the coal transportation system graph network; The errors of the weighing instruments at each node and the confidence level of the errors are predicted based on the coal transportation system diagram network. The errors of the node weighing instruments are calibrated based on the errors of each node weighing instrument and the confidence level of the errors.

2. The method as described in claim 1, characterized in that, Assign global dependencies and context information to the nodes of the coal transportation system graph framework, including: Obtain the global dependencies of nodes using a graph neural network; The context data acquisition mechanism collects information that affects measurement errors, as well as specific error information, as context information. The global dependencies are assigned to the nodes in the coal transportation system graph framework, and the information affecting the measurement error and the specific error information are integrated into the node attributes and / or the edge attributes of the directed edges.

3. The method as described in claim 2, characterized in that, The process of obtaining the global dependencies of nodes through a graph neural network includes: Graph neural networks are used to exchange information and features between each node and its neighboring nodes in multiple rounds. During multiple rounds of information exchange, the embedded representation of each node is continuously updated, so that the information of each node records the characteristics and information of some nodes or global nodes, in order to establish the dependency relationship between each node. Once the dependencies between nodes are established, the graph neural network outputs node representation vectors containing the dependency information of the global network.

4. The method as described in claim 3, characterized in that, The method further includes: Determine the distance between the node representation vector of each node and the preset normal pattern, wherein the distance includes at least one of Euclidean distance, cosine similarity and Mahalanobis distance; When the distance between the node representation vector of a target node and the preset normal mode is greater than a preset distance threshold, it is determined that the weighing instrument corresponding to the target node has a measurement deviation.

5. The method as described in claim 1, characterized in that, The prediction of the error of each node weighing instrument and the confidence level of the error based on the coal transportation system network includes: Obtain a pre-established error prediction model; Preset information containing global dependencies is obtained from the coal transportation system graph network and input into the error prediction model; Obtain the error prediction values ​​of each node output by the error prediction model.

6. The method as described in claim 1, characterized in that, The calibration of the node weighing instrument errors based on the errors of each node weighing instrument and the confidence level of the errors includes: The calibration amount of the weighing instrument is determined based on the error of each node and the confidence level of the error; A corresponding calibration factor is generated based on the calculated adjustment amount for each node, and the calibration factor includes an additive calibration factor and a multiplicative calibration factor. The above calibration factors are applied to the original readings of each weighing instrument in the current batch to obtain the calibrated measurement values.

7. The method as described in claim 1, characterized in that, The coal transportation system network is dynamically updated based on the actual coal transportation process, and the method further includes: When the data increment in the coal transportation system network reaches the preset increment threshold, the error prediction model is incrementally updated based on the new data to update the prediction model parameters. The node confidence weights are dynamically adjusted based on the updated prediction model parameters.

8. A weighing instrument calibration device, characterized in that, include: The first extraction module is used to extract the weighing instruments of all stations in the coal transportation system as nodes. The node attributes include at least one of the following: equipment accuracy, historical measurement readings, geographical location, and environmental parameters. The second extraction module is used to extract the transportation paths between all stations as directed edges, wherein the directed edge attributes include at least one of the following: transportation mode, transportation distance, loss rate, transportation time, and historical statistical data related to the transportation path. The construction module is used to construct a coal transportation system graph framework based on the nodes and directed edges; The assignment module is used to assign global dependencies and context information to the nodes of the coal transportation system graph framework to obtain the coal transportation system graph network. The prediction module is used to predict the error of the weighing instruments at each node and the confidence level of the error based on the coal transportation system diagram network. The calibration module is used to calibrate the error of each node weighing instrument based on the error of each node weighing instrument and the confidence level of the error.

9. A weighing instrument calibration system, characterized in that, include: At least one processor; as well as, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to implement the weighing instrument calibration method as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, When the instructions in the storage medium are executed by the processor corresponding to the weighing instrument calibration system, the weighing instrument calibration system is able to implement the weighing instrument calibration method as described in any one of claims 1-7.

Images