A meteorological data transmission monitoring method and system

Abstract

The application discloses a kind of meteorological data transmission monitoring method and system, it is related to meteorological data monitoring transmission technical field, including obtaining the original meteorological data stream of multiple monitoring nodes containing time stamp, latitude and longitude coordinates, meteorological element measured value and node identifier, effective meteorological data set is generated by integrity check and outlier detection, extract adjacent time point element change rate, trend time sequence characteristics and adjacent node meteorological element gradient distribution, spatial autocorrelation spatial correlation characteristics, two kinds of characteristics are input transmission quality evaluation model, obtain each node data stream instant quality score and potential error probability, according to this mark transmission state, generate monitoring report with state label.This method can capture meteorological data time sequence and spatial implicit characteristics, realize transmission quality quantitative evaluation and transmission state fine marking, improve meteorological data transmission monitoring accuracy.

Description

Technical Field

[0001] This invention belongs to the field of meteorological data monitoring and transmission technology, specifically a meteorological data transmission and monitoring method and system. Background Technology

[0002] Current meteorological data transmission and monitoring largely rely on basic field verification, format validation, and single numerical threshold filtering of raw meteorological data streams for data processing. This only removes data records with missing fields, incorrect formats, and those exceeding conventional thresholds, and only addresses obvious data anomalies. Existing monitoring methods only complete basic effective data filtering and do not conduct in-depth analysis based on the characteristics of meteorological data itself; the monitoring logic remains at the surface-level verification stage of raw data.

[0003] Existing meteorological data transmission and monitoring schemes cannot uncover the temporal and spatial correlations of meteorological data, nor can they identify hidden anomalies during transmission. They can only make a single judgment on data validity and cannot quantify the transmission quality of the data stream. Current technologies cannot combine the temporal and spatial characteristics of meteorological data for monitoring, making it difficult to predict potential errors during data stream transmission, and resulting in monitoring results lacking accuracy and reference value.

[0004] Existing monitoring methods cannot score the data stream transmission quality of monitoring nodes, nor can they classify the transmission status in detail, nor can they generate monitoring reports with accurate status indication labels, making it difficult to meet the actual needs of accurate monitoring during meteorological data transmission. Summary of the Invention

[0005] This invention aims to solve at least one of the technical problems existing in the prior art;

[0006] Therefore, this invention proposes a meteorological data transmission and monitoring method, comprising:

[0007] Acquire raw meteorological data streams from multiple monitoring nodes, the raw meteorological data streams including timestamps, latitude and longitude coordinates, meteorological element measurements and node identifiers;

[0008] The original meteorological data stream is subjected to real-time integrity verification and outlier detection processing to remove data records with missing fields, format errors, and those exceeding reasonable threshold ranges, thereby generating a valid meteorological dataset.

[0009] Key temporal features and spatial correlation features are extracted from the effective meteorological dataset. The key temporal features include the rate of change and trend of change of elements at adjacent time points, and the spatial correlation features include the gradient distribution and spatial autocorrelation of the same meteorological element between adjacent monitoring nodes.

[0010] The key temporal features and spatial correlation features are input into a preset transmission quality assessment model to calculate the instantaneous quality score and potential error probability of the data stream corresponding to each monitoring node.

[0011] Based on the instantaneous quality score and potential error probability, the transmission status of the effective meteorological dataset is classified and labeled, and a data transmission monitoring report containing status labels is generated. The status labels are used to indicate whether the data stream is normal, the data stream is suspicious, or the data stream is erroneous.

[0012] Furthermore, the original meteorological data stream undergoes real-time integrity verification and outlier detection processing to remove data records with missing fields, format errors, and those exceeding reasonable threshold ranges, generating a valid meteorological dataset, including:

[0013] Check whether each data record in the original meteorological data stream contains a timestamp field, latitude and longitude coordinate field, all preset meteorological element measurement value fields, and node identifier field, and discard data records that are missing any one or more required fields;

[0014] For each data record containing complete fields, verify whether its timestamp conforms to the time progression logic, whether its latitude and longitude coordinates are within the preset geographic service area, and whether its node identifier exists in the registered node list.

[0015] For verified data records, read their meteorological element measurement values, compare each meteorological element measurement value with its predefined reasonable value range, and remove records where the meteorological element measurement value is lower than the minimum value or higher than the maximum value;

[0016] For meteorological element measurements within a reasonable range, further consistency analysis is performed between these measurements and historical adjacent values ​​at the same monitoring node. The magnitude of change is calculated, and data records whose magnitude of change exceeds the physically reasonable limit of abrupt change are removed.

[0017] The remaining data records that have passed all verification steps are packaged and marked as the valid meteorological dataset.

[0018] Furthermore, the measurement values ​​of meteorological elements within a reasonable range are further subjected to consistency analysis with historical adjacent values ​​at the same monitoring node, including:

[0019] The historical meteorological dataset of the monitoring node is obtained from the local cache, and the meteorological element measurement values ​​of multiple historical time points closest to the current time are extracted from the historical meteorological dataset.

[0020] Calculate the difference between the current meteorological element measurement value and the average value of the meteorological element measurement values ​​at the nearest multiple historical time points;

[0021] The difference is compared with the allowable change threshold preset based on the meteorological element type. The allowable change threshold reflects the maximum reasonable change that different meteorological elements can undergo in adjacent time periods.

[0022] If the difference exceeds the preset allowable change threshold for the corresponding type of meteorological element, the current data record is marked as a suspected anomaly.

[0023] For data records marked as suspected anomalies, further check the transmission signal quality logs of the monitoring nodes near the data generation time. If the signal quality is lower than the set standard, it is determined to be a transmission error. Otherwise, spatial consistency is checked in conjunction with data from neighboring monitoring nodes, and a final decision is made to retain or remove the record.

[0024] Furthermore, key temporal and spatial correlation features are extracted from the effective meteorological dataset, including:

[0025] For the time-series data sequence of each monitoring node in the effective meteorological dataset, the difference between the measured values ​​of meteorological elements between adjacent sampling times is calculated, and the difference is divided by the time interval to obtain the rate of change of the element.

[0026] Statistical analysis is performed on the sequence of the rate of change of the elements to identify whether they show an increasing, decreasing or stable trend within a certain time window, and a trend description is generated. The trend description is used to characterize the temporal change pattern of the corresponding meteorological elements.

[0027] At the same time, multiple spatially adjacent monitoring nodes are searched, the differences in the measured values ​​of the same meteorological element between the monitoring nodes are compared, and the difference between the maximum and minimum values ​​is calculated as a representation of the gradient distribution.

[0028] Based on the principles of geostatistics, the correlation coefficient of the same meteorological element measurements at the same time among all node pairs in the entire monitoring network is calculated. By analyzing the pattern of correlation coefficient decay with distance, the spatial autocorrelation is quantified.

[0029] The correlation coefficient was calculated using the Pearson correlation coefficient formula. For two spatial points i and j, the meteorological element measurements were respectively... and The correlation coefficient between node i and node j The following formula can be used to calculate:

[0030]

[0031] in, This represents the correlation coefficient between node i and node j. Let represent the covariance of the meteorological element measurements between node i and node j. This represents the variance of the meteorological element measurements at node i. This represents the variance of the meteorological element measurements at node j.

[0032] Furthermore, the calculation of the correlation coefficient between all node pairs in the entire monitoring network at the same time for the same meteorological element measurement includes:

[0033] Obtain the valid meteorological dataset of all monitoring nodes at the current moment to form a spatial data point set, where each point contains latitude and longitude coordinates and at least one meteorological element measurement value;

[0034] Calculate the distance between all pairs of points in the spatial data point set, and the covariance of the meteorological element measurements of these two points;

[0035] Based on the covariance and the variance of each meteorological element measurement, the correlation coefficient between each pair of nodes is calculated;

[0036] All the calculated correlation coefficients are grouped according to their corresponding distances between nodes, and the average correlation coefficient within each distance group is calculated.

[0037] Plot the curve of the average correlation coefficient as a function of distance, fit the mathematical model parameters of the curve, and use the mathematical model parameters as the feature vector for quantifying the spatial autocorrelation.

[0038] Furthermore, the key temporal features and spatial correlation features are input into a preset transmission quality assessment model to calculate the instantaneous quality score and potential error probability of the data stream corresponding to each monitoring node, including:

[0039] The transmission quality assessment model is constructed based on historical transmission stability data and the physical constraint relationship of meteorological elements. It includes multiple parallel processing feature analysis branches, wherein the first feature analysis branch receives the key time series features and generates a time series anomaly index by analyzing whether the rate and trend of element change conflict with meteorological physical laws.

[0040] The second feature analysis branch receives the spatial correlation features and generates a spatial consistency deviation by analyzing whether the corresponding node data and the surrounding node data deviate from a preset normal threshold in terms of gradient distribution and spatial autocorrelation pattern.

[0041] The temporal anomaly index and spatial consistency deviation are fused and input into a comprehensive evaluation sub-model. The comprehensive evaluation sub-model learns the patterns of historically labeled fault data and normal data and outputs an original quality score.

[0042] The original quality score is normalized and mapped to a predetermined scoring range to obtain the instantaneous quality score;

[0043] Simultaneously, the activation values ​​of the intermediate layers of the comprehensive evaluation sub-model are extracted and input into the auxiliary classifier. The auxiliary classifier is used to output the probability values ​​of various types of errors in the corresponding data stream, and the maximum value among the probability values ​​of each category is taken as the potential error probability.

[0044] Furthermore, the transmission quality assessment model is trained through the following steps before deployment:

[0045] Collect historical meteorological data records and their corresponding manually verified quality labels, wherein the quality labels indicate whether the corresponding data record is valid or what type of error it belongs to;

[0046] Feature extraction is performed on the historical meteorological data records to generate key historical temporal features and historical spatial correlation features;

[0047] A neural network model is constructed as the initial transmission quality assessment model. Its input layer corresponds to the dimensions of the historical key temporal features and historical spatial correlation features, and its output layer includes a quality score output and an error probability output.

[0048] Historical data with the quality labels are used as training samples and input into the initial transmission quality assessment model. The internal parameters of the model are adjusted through the backpropagation algorithm to minimize the difference between the quality score output by the model and the human score, while minimizing the cross-entropy loss between the error probability distribution output by the model and the actual error type.

[0049] The model performance is evaluated using an independent validation dataset. The model structure or hyperparameters are adjusted until the preset accuracy requirements are met, resulting in the trained preset transmission quality evaluation model.

[0050] Furthermore, based on the instantaneous quality score and potential error probability, the transmission status of the effective meteorological dataset is classified and labeled, including:

[0051] Set high, medium, and low thresholds for quality scores, as well as confidence thresholds for the probability of potential errors;

[0052] For each data stream in the effective meteorological dataset, its instantaneous quality score is compared with the high, medium, and low thresholds of the quality score, and its potential error probability is compared with the confidence threshold.

[0053] If the instantaneous quality score is higher than the high threshold and the potential error probability is lower than the confidence threshold, the transmission status of the corresponding data stream will be marked as "data stream normal".

[0054] If the instantaneous quality score is between the medium and high thresholds, or the potential error probability is close to but does not exceed the confidence threshold, it is marked as "data stream suspicious" and the specific suspicious characteristics are recorded.

[0055] If the instantaneous quality score is below the low threshold, or the potential error probability exceeds the confidence threshold, it is marked as a "data flow error," and an attempt is made to infer the error type based on intermediate features of the model.

[0056] After all data streams are marked, a data transmission monitoring report is generated. The data transmission monitoring report lists the node identifier, corresponding status label, real-time quality score, potential error probability, and suspicious or erroneous description in tabular form.

[0057] Furthermore, for data streams marked as "suspicious," an automatic review step is also included:

[0058] Automatically trigger a retransmission request command for the monitoring node marked as "data flow suspicious", and receive supplementary or duplicate data records from the corresponding monitoring node;

[0059] Real-time integrity verification and outlier detection are performed on the supplementary or duplicate data records to form a valid dataset for review;

[0060] Extract review temporal features and review spatial correlation features from the valid review dataset;

[0061] The review time-series features and review spatial correlation features are input into a preset transmission quality assessment model to calculate the review instant quality score and the review potential error probability.

[0062] Based on the review results, if the immediate quality score improves to the normal range and the probability of potential errors decreases, the status label is corrected to "data flow normal"; otherwise, it is confirmed as "data flow error" and the data transmission monitoring report is updated.

[0063] Furthermore, the present invention also includes a meteorological data transmission monitoring system, the system comprising a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor, when executing the computer program, implements the steps of the meteorological data transmission monitoring method described above.

[0064] Compared with the prior art, the beneficial effects of the present invention are:

[0065] Real-time integrity verification and outlier detection are performed on the raw meteorological data stream. After removing data records with missing fields, format errors, and exceeding reasonable threshold ranges, the rate of change and trend of meteorological elements at adjacent time points are extracted to form key temporal features. The gradient distribution and spatial autocorrelation of the same meteorological element between adjacent monitoring nodes are extracted to form spatial correlation features. This can fully characterize the continuous change attributes of meteorological data in the time dimension and the correlation distribution attributes in the spatial dimension. It can accurately identify the hidden anomalies generated by meteorological data during transmission, distinguish between natural changes in meteorological data and data deviations caused by transmission anomalies, and break through the limitations of only performing surface field verification on raw data. It can achieve in-depth mining of the inherent transmission characteristics of meteorological data, fully preserve the temporal change patterns and spatial correlation characteristics of meteorological data, and accurately capture abnormal features that cannot be identified by conventional verification in the transmission process.

[0066] By inputting the extracted key temporal and spatial correlation features into a preset transmission quality assessment model, the instantaneous quality score and potential error probability of the data stream corresponding to each monitoring node can be directly calculated. Based on the instantaneous quality score and potential error probability, the transmission status of the effective meteorological dataset is classified and labeled, generating a data transmission monitoring report containing labels for normal, suspicious, and error statuses. This enables quantitative characterization of meteorological data stream transmission quality, refines the judgment level of transmission status, accurately distinguishes different degrees of transmission anomalies, and intuitively presents the real-time transmission status of the data stream. It avoids judgment bias caused by single threshold verification, improves the accuracy and refinement of transmission status judgment, and forms standardized, directly applicable transmission monitoring results, achieving full-dimensional and accurate monitoring of the meteorological data transmission process. Attached Figure Description

[0067] Figure 1 This is a flowchart illustrating the steps of a meteorological data transmission and monitoring method according to the present invention.

[0068] Figure 2 A flowchart for generating valid meteorological datasets for real-time integrity verification and outlier detection and processing;

[0069] Figure 3 This is a flowchart for extracting key temporal and spatial correlation features from an effective meteorological dataset. Detailed Implementation

[0070] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0071] See Figure 1This invention provides a meteorological data transmission monitoring method. The overall implementation scheme is as follows: Raw meteorological data streams from multiple monitoring nodes are acquired. These raw data streams include timestamps, latitude and longitude coordinates, meteorological element measurements, and node identifiers. Real-time integrity verification and outlier detection processing are performed on the raw meteorological data streams, removing data records with missing fields, format errors, and those exceeding reasonable threshold ranges to generate a valid meteorological dataset. Key temporal features and spatial correlation features are extracted from the valid meteorological dataset. The key temporal features include the rate and trend of element change at adjacent time points, and the spatial correlation features include the gradient distribution and spatial autocorrelation of the same meteorological element between adjacent monitoring nodes. The key temporal features and spatial correlation features are input into a preset transmission quality assessment model to calculate the instantaneous quality score and potential error probability of the data stream corresponding to each monitoring node. Based on the instantaneous quality score and potential error probability, the transmission status of the valid meteorological dataset is classified and labeled, generating a data transmission monitoring report containing status labels. These status labels indicate whether the data stream is normal, suspicious, or erroneous.

[0072] In one embodiment of the present invention, the raw meteorological data stream undergoes real-time integrity verification and outlier detection processing, including multiple levels of inspection steps. See also... Figure 2The process involves checking each data record in the original meteorological data stream to ensure it contains a timestamp field, latitude and longitude coordinates, all preset meteorological element measurement values, and a node identifier field. Data records missing any one or more required fields are discarded. For data records containing complete fields, each record's timestamp is verified to conform to time progression logic, its latitude and longitude coordinates are verified to be within a preset geographic service area, and its node identifier is verified to exist in the registered node list. For verified data records, their meteorological element measurement values ​​are read, and each meteorological element measurement value is compared with its predefined reasonable value range. Records with meteorological element measurement values ​​below the minimum or above the maximum value are discarded. For meteorological element measurement values ​​within the reasonable value range, a consistency analysis is performed with the historical adjacent values ​​of the same monitoring node to calculate the magnitude of change. Data records with magnitudes exceeding the physically reasonable limit of abrupt change are discarded. The measurement values ​​of meteorological elements within a reasonable range are further analyzed for consistency with historical values ​​from adjacent times at the same monitoring node. Specifically, this involves: retrieving the historical meteorological dataset of the monitoring node from the local cache; extracting the meteorological element measurement values ​​from the historical meteorological dataset that are closest to the current time; calculating the difference between the current meteorological element measurement value and the average of the meteorological element measurement values ​​from the closest historical time points; comparing the difference with a preset allowable change threshold based on the meteorological element type, which reflects the maximum reasonable change that different meteorological elements can undergo in adjacent time periods; if the difference exceeds the preset allowable change threshold for the corresponding type of meteorological element, the current data record is marked as a suspected anomaly. For data records marked as suspected anomalies, the transmission signal quality log of the monitoring node near the data generation time point is further checked. If the signal quality is lower than the set standard, it is determined to be a transmission error; otherwise, spatial consistency is checked in conjunction with data from neighboring monitoring nodes, and a final decision is made to retain or remove the data. The remaining data record set that passes all verification steps is packaged and marked as the valid meteorological dataset.

[0073] In practice, each record in the acquired raw meteorological data stream must include a timestamp field, latitude and longitude coordinates, all preset meteorological element measurement values, and a node identifier field. Preset meteorological elements include temperature, air pressure, humidity, and wind speed. During the integrity verification phase, the processing system scans the input raw meteorological data stream line by line, discarding data records missing any one or more required fields. In practice, a data record from node "S001" that only contains a timestamp, longitude, and latitude but lacks a wind speed measurement value will be directly discarded and not included in subsequent processing. For data records that pass the integrity check and contain all fields, the processing system initiates a verification process. This verifies whether the timestamp conforms to the time progression logic, i.e., checking whether the timestamp of the record is before the current time of the processing system and not earlier than the timestamp of the most recently received record at that node. The system verifies whether the latitude and longitude coordinates are within a preset geographical service area. For example, the service area boundary of a meteorological monitoring network is 110.0°E to 120.0°E and 20.0°N to 30.0°N. If a record's coordinates are located at 130.0°E, the record will be deemed geographically invalid. The system also verifies whether the node identifier exists in a predefined list of registered nodes. This list contains unique identifiers for all authorized monitoring nodes. If a record's node identifier is not in the list, the record will be considered of unknown origin and removed. For data records that have passed time, geographical, and node authentication, the system reads their meteorological element measurements. Each meteorological element measurement has a predefined reasonable value range. In practice, the processing system compares each meteorological element measurement in the record with its corresponding reasonable value range. If a measurement value is lower than the minimum or higher than the maximum value of its preset range, the complete data record containing this out-of-range measurement value will be removed. For example, if a record's air pressure measurement is 800 hPa, lower than the preset minimum of 850 hPa, the record will be marked as abnormal and removed.

[0074] For meteorological element measurements within a reasonable range, the system will further perform a physical consistency analysis. This consistency analysis compares the current meteorological element measurement with historical values ​​from adjacent times at the same monitoring node, calculates the magnitude of change, and removes data records whose magnitude exceeds the physically reasonable limit for abrupt changes. In practice, for the temperature element, a physically reasonable limit for abrupt changes within a continuous 5-minute sampling interval is set at 5°C. In practice, the system retrieves the historical meteorological dataset of the monitoring node from its local cache and extracts the meteorological element measurements from the dataset that are closest to the current time. The difference between the current meteorological element measurement and the average of these closest historical measurements is calculated. A weighted average method can be used to calculate the difference, with higher weights for historical data closer to the current time. In practice, the weighted average is calculated using data from the previous three historical times (T-1, T-2, T-3). The calculation formula is as follows:

[0075]

[0076] in: , , These represent the meteorological element measurements, such as temperature, at the previous three moments, respectively. , , These are preset weighting coefficients that satisfy... Then, calculate the measurement value at the current time. with weighted average The absolute value of the difference .

[0077] In one embodiment of the present invention, key temporal features and spatial correlation features are extracted from the effective meteorological dataset, including two aspects: temporal feature extraction and spatial feature extraction. (See also...) Figure 3For each monitoring node in the effective meteorological dataset, the difference between meteorological element measurements at adjacent sampling times is calculated. This difference is then divided by the time interval to obtain the element's rate of change. Statistical analysis is performed on the element's rate of change sequence to identify whether it exhibits an increasing, decreasing, or stable trend within a certain time window, generating a trend description that characterizes the temporal variation pattern of the corresponding meteorological element. For the same moment, multiple spatially adjacent monitoring nodes are located, and the differences in the same meteorological element's measurements between the monitoring nodes are compared. The difference between the maximum and minimum values ​​is calculated as a representation of the gradient distribution. Based on the principles of geostatistics, the correlation coefficient between all node pairs in the entire monitoring network and the same meteorological element measurements at the same time is calculated. By analyzing the decay pattern of the correlation coefficient with distance, the spatial autocorrelation is quantified. The specific steps for calculating the correlation coefficient between all node pairs in the entire monitoring network and the same meteorological element measurements at the same time are: acquiring the effective meteorological dataset of all monitoring nodes at the current moment, forming a spatial data point set, where each point contains latitude and longitude coordinates and at least one meteorological element measurement value. Calculate the distance between all pairs of points in the spatial data set, and the covariance of the meteorological element measurements between these two points. Based on the covariance and the variance of each meteorological element measurement, calculate the correlation coefficient between each pair of nodes. Group all calculated correlation coefficients according to their corresponding node distances, and calculate the average correlation coefficient within each distance group. Plot the curve of the average correlation coefficient as a function of distance, fit the mathematical model parameters of the curve, and use these mathematical model parameters as the feature vector quantifying the spatial autocorrelation.

[0078] In practice, the extraction operation is performed based on the generated valid meteorological dataset, which is a data set ordered by timestamp and node identifier after integrity verification and outlier detection. For the time-series data sequence of each monitoring node in the valid meteorological dataset, the system processes the meteorological element measurements node by node. During processing, the system reads the meteorological element data of a monitoring node at consecutive time points. The difference between the meteorological element measurements between adjacent sampling times is calculated, and the difference is divided by the time interval to obtain the element's rate of change. In practice, assuming that the temperature measurement value of monitoring node A at time t1 is T1, and the temperature measurement value at the next sampling time t2 is T2, with a time interval Δt = t2 - t1, the element's rate of change V1 from time t1 to t2 can be calculated using the formula V1 = (T2 - T1) / Δt. For the extraction of spatial correlation features, the operation is performed on the valid meteorological data of all monitoring nodes at the same time. For the same time, the system searches for multiple spatially adjacent monitoring nodes based on their latitude and longitude coordinates. In some embodiments, the K nearest neighboring nodes are determined for each node by calculating the spherical distance between all node pairs. The differences in the measured values ​​of the same meteorological element among the monitoring nodes are compared. The comparison operation involves calculating the maximum and minimum values ​​of the same meteorological element measured among these neighboring nodes, and then calculating the difference between the maximum and minimum values. This difference is used as a representation of the gradient distribution. In a specific implementation, at time T, for monitoring node B and its four neighboring nodes, their air pressure measurements are 1010 hPa, 1012 hPa, 1009 hPa, 1013 hPa, and 1011 hPa, respectively. The difference between the maximum and minimum values ​​is 1013 hPa - 1009 hPa = 4 hPa. This 4 hPa difference is used as a quantitative representation of the local air pressure gradient distribution centered on node B. It can be understood that the gradient distribution reflects the degree of non-uniformity of meteorological elements within a local spatial range.

[0079] Based on the principles of geostatistics, the correlation coefficient of the same meteorological element measurements at the same time among all pairs of nodes in the entire monitoring network is calculated. The spatial autocorrelation is quantified by analyzing the decay pattern of the correlation coefficient with distance. The calculation of the correlation coefficient of the same meteorological element measurements at the same time among all pairs of nodes in the entire monitoring network includes multiple calculation steps. First, the effective meteorological dataset of all monitoring nodes at the current time is obtained, forming a spatial data point set. Each point in the spatial data point set contains latitude and longitude coordinates and at least one meteorological element measurement value. Second, the distance between all pairs of points in the spatial data point set, and the covariance of the meteorological element measurements of these two points are calculated. Third, based on the covariance and the variance of each meteorological element measurement value, the correlation coefficient between each pair of nodes is calculated. The formula used to calculate the correlation coefficient is the Pearson correlation coefficient formula. In a specific calculation formula, for two spatial points i and j, their meteorological element measurements are respectively... and Their covariance is Measured value The variance is Measured value The variance is Then the correlation coefficient between node i and node j The following formula can be used to calculate:

[0080]

[0081] in, This represents the correlation coefficient between node i and node j. Let represent the covariance of the meteorological element measurements between node i and node j. This represents the variance of the meteorological element measurements at node i. This represents the variance of the meteorological element measurements at node j. In practical implementation, and It can be the temperature measurement values ​​of all nodes at a certain moment.

[0082] All calculated correlation coefficients are grouped according to their corresponding distances between nodes. Distance grouping is based on preset distance intervals, for example, 0-10 km, 10-20 km, 20-30 km, etc., are set as different distance groups. The average correlation coefficient within each distance group is calculated. In a specific implementation, the arithmetic mean of the correlation coefficients of all node pairs whose distances fall within the 0-10 km range is taken to obtain the average correlation coefficient R1 for this distance group. A curve of the average correlation coefficient changing with distance is plotted, and the mathematical model parameters of the fitted curve are used as the feature vector quantifying the spatial autocorrelation. In some embodiments, an exponential decay model is used. The average correlation coefficient-distance curve is fitted, where d represents the distance, R(d) represents the model-predicted average correlation coefficient at distance d, and a and b are the parameters to be fitted. The fitting process is completed using the least squares method or other optimization algorithms. The values ​​of parameters a and b obtained constitute a two-dimensional feature vector describing the spatial autocorrelation of the meteorological element at the current moment. It can be understood that the spatial autocorrelation feature vector reflects the overall correlation structure of the meteorological element in space. The value of a can be interpreted as the theoretical correlation at zero distance, while the value of b reflects the rate at which the correlation decays with distance.

[0083] In one embodiment of the present invention, the key temporal features and spatial correlation features are input into a preset transmission quality assessment model. This model is constructed based on historical transmission stability data and the physical constraints of meteorological elements, and includes multiple parallel feature analysis branches. The first feature analysis branch receives the key temporal features and generates a temporal anomaly index by analyzing whether the rate and trend of element change conflict with meteorological physical laws. The second feature analysis branch receives the spatial correlation features and generates a spatial consistency deviation by analyzing whether the corresponding node data and surrounding node data deviate from a preset normal threshold in gradient distribution and spatial autocorrelation patterns. The temporal anomaly index and spatial consistency deviation are fused and input into a comprehensive assessment sub-model. This sub-model learns the patterns of historically labeled faulty and normal data and outputs a raw quality score. The raw quality score is normalized and mapped to a predetermined scoring interval to obtain the instantaneous quality score. Simultaneously, the activation values ​​of the intermediate layers of the comprehensive assessment sub-model are extracted and input into an auxiliary classifier. This auxiliary classifier outputs the probability values ​​of various types of errors in the corresponding data stream, and the maximum value among the probability values ​​of each category is taken as the potential error probability.

[0084] In practice, key temporal features and spatial correlation features are input into a pre-defined transmission quality assessment model, and an instantaneous quality score and potential error probability are calculated. The pre-defined transmission quality assessment model is constructed based on historical transmission stability data and the physical constraints of meteorological elements, and includes multiple parallel feature analysis branches. In practice, key temporal features include the sequence of element change rates and a description of change trends extracted from the past hour's temperature data of a specific monitoring node. Spatial correlation features include a representation of the local pressure gradient distribution centered on that node and a model parameter vector describing spatial autocorrelation. The first feature analysis branch receives the key temporal features and is a small neural network module. The first feature analysis branch generates a temporal anomaly index by analyzing whether the element change rates and trends conflict with meteorological physical laws.

[0085] The second feature analysis branch receives the spatial correlation features and is another parallel neural network module. This branch generates a spatial consistency deviation score by analyzing whether the corresponding node data deviates from a preset normal threshold in terms of gradient distribution and spatial autocorrelation patterns compared to surrounding node data. The preset normal threshold is a feature distribution range obtained based on historical normal data statistics. For example, under stable weather conditions, the local pressure gradient distribution is typically less than 5 hPa, and the spatial autocorrelation model parameter b is typically between 20 km and 50 km. If the input pressure gradient distribution is 10 hPa, or parameter b is 5 km, it indicates that the spatial characteristics of that node significantly deviate from the normal pattern, and the second feature analysis branch will output a high spatial consistency deviation score. The spatial consistency deviation score is also a value between 0 and 1. In specific implementations, the temporal anomaly index and the spatial consistency deviation score are concatenated into a combined feature vector.

[0086] The temporal anomaly index and spatial consistency deviation are fused and input into a comprehensive evaluation sub-model. This sub-model learns patterns from historically labeled fault and normal data to output a raw quality score. The comprehensive evaluation sub-model is a multilayer perceptron. The feature fusion operation directly concatenates the temporal anomaly index and spatial consistency deviation into a two-dimensional vector. This serves as the input to the comprehensive evaluation sub-model. After processing this input vector, the comprehensive evaluation sub-model produces a raw quality score at the output layer. Original mass fraction The range of values ​​is a continuous range defined by the model, for example... Regarding the original mass fraction The score is then normalized and mapped to a predetermined scoring interval to obtain the instantaneous quality score. The normalization process employs a linear transformation method, for example, using the formula: Will From the scope Mapped to a predetermined rating range ,in and These are the theoretical minimum and maximum values ​​(-5 and 5) of the original quality fraction, respectively, obtained after mapping. This is the instantaneous quality score. Refer to Table 1, which shows the correspondence between input features, intermediate indicators, and output scores.

[0087] Table 1: Input and Output Table of Transmission Quality Assessment Model

[0088] Node ID Time series anomaly index Spatial consistency deviation Original mass fraction Instant Quality Rating N001 0.15 0.08 3.2 78.8 N002 0.65 0.12 -2.1 33.4 N003 0.02 0.45 1.8 66.4 N004 0.80 0.90 -4.7 12.4

[0089] Simultaneously, the intermediate layer activation values ​​of the comprehensive evaluation sub-model are extracted and input into an auxiliary classifier, which outputs the probability values ​​of various types of errors in the corresponding data stream. The intermediate layer activation values ​​of the comprehensive evaluation sub-model refer to the output values ​​of the neurons in its last hidden layer. The auxiliary classifier is a softmax classifier, whose input is the extracted intermediate layer activation values. The auxiliary classifier outputs the probability values ​​of various types of errors in the corresponding data stream, and the error categories may include "sensor drift," "communication interference," "complete failure," etc. The maximum value among the probability values ​​of each category is taken as the potential error probability. In some embodiments, for node N002, the auxiliary classifier may output the probability distribution as: "sensor drift": 0.70, "communication interference": 0.25, "complete failure": 0.05, then the maximum value of 0.70 is taken as the potential error probability of this node. It can be understood that the instantaneous quality score provides a quantitative assessment of the overall quality of the data stream, while the potential error probability and the error category corresponding to its maximum probability provide an indication of possible causes of failure.

[0090] In one embodiment of the present invention, the transmission quality assessment model needs to acquire assessment capabilities through a training process before deployment. Historical meteorological data records and their corresponding manually verified quality labels are collected; the quality labels indicate whether the corresponding data record is valid or belongs to a certain error type. Feature extraction is performed on the historical meteorological data records to generate historical key temporal features and historical spatial correlation features. A neural network model is constructed as the initial transmission quality assessment model, with its input layer corresponding to the dimensions of the historical key temporal features and historical spatial correlation features, and its output layer including a quality score output and an error probability output. Historical data with the quality labels is used as training samples and input into the initial transmission quality assessment model. The model's internal parameters are adjusted using a backpropagation algorithm to minimize the difference between the model's output quality score and the manual score, while simultaneously minimizing the cross-entropy loss between the model's output error probability distribution and the actual error type. The model performance is evaluated using an independent validation dataset, and the model structure or hyperparameters are adjusted until the preset accuracy requirements are met, resulting in the trained preset transmission quality assessment model.

[0091] In practice, the training process begins with collecting historical meteorological data records and their corresponding manually verified quality labels. The collected historical meteorological data records are raw data streams acquired from multiple monitoring nodes over a past period, containing timestamps, latitude and longitude coordinates, meteorological element measurements, and node identifiers. The manually verified quality labels are the result of meteorological data experts annotating each historical data record according to data quality assessment standards. The quality labels indicate whether the corresponding data record is valid or what type of error it belongs to. Error types can include, but are not limited to, "transmission timeout," "value exceeding limits," "spatiotemporal inconsistency," "sensor malfunction," and "communication interference." In practice, a set containing 100,000 historical meteorological data records is compiled, with each record accompanied by a quality label, such as "valid," "sensor malfunction," or "communication interference." See Table 2, which shows a simplified mapping table of historical data records and quality labels.

[0092] Table 2: Historical Meteorological Data Records and Quality Labels

[0093] Record ID Node identifier Timestamp Temperature measurement (°C) Quality Label D001 Station_A 2023-08-0110:00:00 25.3 efficient D002 Station_B 2023-08-0110:00:00 999.9 Numerical limit exceeded D003 Station_A 2023-08-0110:05:00 25.5 efficient D004 Station_C 2023-08-0110:05:00 24.8 Sensor failure

[0094] Feature extraction is performed on the historical meteorological data records to generate historical key temporal features and historical spatial correlation features. Historical key temporal features include descriptions of the rate of change and trend of elements at adjacent time points calculated for the time-series data of each monitoring node. Historical spatial correlation features include gradient distribution representations and spatial autocorrelation feature vectors calculated for data from different nodes at the same time. In a specific implementation, for historical record D001, the system extracts its temperature change rate sequence and trend description within a specific time window and calculates its spatial features relative to neighboring nodes at the corresponding time. A neural network model is constructed as the initial transmission quality assessment model. Its input layer corresponds to the dimensions of the historical key temporal features and historical spatial correlation features, and its output layer includes a quality score output and an error probability output. In a specific implementation, the historical key temporal features and historical spatial correlation features are concatenated into a fixed-length feature vector. The number of neurons in the input layer is equal to the dimension of this concatenated feature vector. The output layer consists of two parts: the quality score output is a single neuron that predicts a continuous quality score; the error probability output is a softmax layer with the number of neurons equal to the number of predefined error types plus the number of "valid" categories, which outputs the probability distribution of each category.

[0095] Historical data with the aforementioned quality labels are used as training samples and input into the initial transmission quality assessment model. The quality labels need to be converted to fit the model's output. For the quality score output, a human score needs to be assigned to each training sample. The human score is based on the quality label; for example, a record labeled "valid" corresponds to a human score of 90 points (out of 100), and a record labeled "communication interference" corresponds to a human score of 30 points. For the error probability output, the quality label is converted into a one-hot vector. For example, if there are K error types (including the "valid" type), the label corresponding to "sensor failure" is a vector of length K, with 1 at the "sensor failure" category index and 0 at the remaining positions. The model's internal parameters are adjusted using a backpropagation algorithm to minimize the difference between the model's output quality score and the human score, while also minimizing the cross-entropy loss between the model's output error probability distribution and the true error type. In specific implementation, the total loss function... Defined as quality score regression loss Cross-entropy loss for misclassification The weighted sum, for example:

[0096]

[0097] Where: α is the preset weight coefficient. Regression loss Mean squared error and cross-entropy loss can be used. The standard multi-class cross-entropy formula is used for calculation. The model undergoes multiple iterations on the training set, and the backpropagation algorithm calculates the total loss. The weights and bias parameters in the model are updated so that the quality scores predicted by the model are closer to human scores, and the error probability distribution of the predictions is closer to the true one-hot label distribution.

[0098] The model performance is evaluated using an independent validation dataset. The model structure or hyperparameters are adjusted until the preset accuracy requirements are met, resulting in a pre-trained, preset transmission quality assessment model. The independent validation dataset is a reserved portion of historical data that was not used in training, and it also carries quality labels. In practice, the model's performance on the validation set is evaluated using multiple metrics, such as the root mean square error of quality score prediction, misclassification accuracy, and precision. It is understood that if the model's performance metrics on the validation set do not reach the preset threshold, the model structure or hyperparameters need to be adjusted. Model structure adjustments include increasing or decreasing the number of hidden layers in the neural network and changing the number of neurons in the hidden layers. Hyperparameter adjustments include changing the learning rate, batch size, and weight coefficients α in the loss function. In practice, after multiple rounds of adjustment and retraining, when the model's root mean square error of quality score prediction on the validation set is less than 5 points and the misclassification accuracy reaches 96%, the model is considered to meet the preset accuracy requirements. At this point, the model parameters are fixed, and this is considered the pre-trained, preset transmission quality assessment model, ready for deployment for online evaluation. In some embodiments, the trained model is saved as a file format that includes the network structure definition and all optimized parameter values.

[0099] In one embodiment of the present invention, the transmission status of the effective meteorological dataset is classified and labeled according to the instantaneous quality score and the potential error probability. High, medium, and low thresholds for the quality score, and a confidence threshold for the potential error probability are set. For each data stream in the effective meteorological dataset, its instantaneous quality score is compared with the high, medium, and low thresholds, and its potential error probability is compared with the confidence threshold. If the instantaneous quality score is higher than the high threshold and the potential error probability is lower than the confidence threshold, the transmission status of the corresponding data stream is marked as "data stream normal". If the instantaneous quality score is between the medium and high thresholds, or the potential error probability is close to but does not exceed the confidence threshold, it is marked as "data stream suspicious", and specific suspicious characteristics are recorded. If the instantaneous quality score is lower than the low threshold, or the potential error probability exceeds the confidence threshold, it is marked as "data stream error", and an attempt is made to infer the error type based on intermediate features of the model. After marking all data streams, a data transmission monitoring report is compiled, which lists the node identifier, corresponding status label, instantaneous quality score, potential error probability, and suspicious or error description in tabular form. For data streams marked as "suspicious," the process also includes an automatic review step: automatically triggering a retransmission request command for the monitoring nodes marked as "suspicious," and receiving supplementary or duplicate data records from the corresponding monitoring nodes. Real-time integrity verification and outlier detection processing are performed on the supplementary or duplicate data records to form a valid review dataset. Review temporal features and review spatial correlation features are extracted from the valid review dataset. These features are input into a preset transmission quality assessment model to calculate the immediate review quality score and the probability of potential errors. Based on the review results, if the immediate review quality score improves to the normal range and the probability of potential errors decreases, the status label is corrected to "data stream normal"; otherwise, it is confirmed as "data stream error," and the data transmission monitoring report is updated.

[0100] In the specific implementation, high, medium, and low thresholds for the quality score, as well as a confidence threshold for the potential error probability, are set. The quality score ranges from 10 to 100, with a high threshold of 80, a medium threshold of 60, and a low threshold of 40. The potential error probability ranges from 0 to 1, with a confidence threshold of 0.7. These thresholds are pre-set based on historical evaluation results before model deployment. For each data stream in the effective meteorological dataset, its instantaneous quality score is compared with the high, medium, and low thresholds of the quality score, and its potential error probability is compared with the confidence threshold. The comparison operations are sequential logical judgments.

[0101] If the instantaneous quality score is higher than the high threshold and the potential error probability is lower than the confidence threshold, the transmission status of the corresponding data stream is marked as "data stream normal". In specific implementation, if a data stream has an instantaneous quality score of 85 and a potential error probability of 0.3, since 85 > 80 (higher than the high threshold) and 0.3 < 0.7 (lower than the confidence threshold), the status of this data stream is marked as "data stream normal". If the instantaneous quality score is between the medium and high thresholds, or the potential error probability is close to but does not exceed the confidence threshold, it is marked as "data stream suspicious", and the specific suspicious characteristics are recorded. In specific implementation, "between the medium and high thresholds" means that the instantaneous quality score is greater than or equal to the medium threshold of 60 and less than or equal to the high threshold of 80. "Potential error probability close to but does not exceed the confidence threshold" can be defined as the potential error probability being within a preset close range. For example, if the confidence threshold is 0.7, the close range can be defined as [0.65, 0.7). It can be understood that a data stream that meets any of these conditions is marked as "data stream suspicious". For example, a data stream with an instantaneous quality score of 75 and a potential error probability of 0.2 is labeled as "suspicious" because 75 satisfies the condition of being greater than or equal to 60 and less than or equal to 80. Another data stream with an instantaneous quality score of 90 and a potential error probability of 0.68 is also labeled as "suspicious" because 0.68 falls within the range [0.65, 0.7). Specific suspicious features can be extracted from intermediate or input features of the transmission quality assessment model, such as recording a temporal anomaly index of 0.4 or a spatial consistency deviation of 0.6.

[0102] If the instantaneous quality score is below a low threshold, or the potential error probability exceeds a confidence threshold, it is marked as a "data stream error," and an attempt is made to infer the error type based on intermediate features of the model. In practice, below the low threshold means the instantaneous quality score is less than 40. Exceeding the confidence threshold means the potential error probability is greater than or equal to 0.7. For example, a data stream with an instantaneous quality score of 35 and a potential error probability of 0.5 is marked as a "data stream error" because 35 < 40. Another data stream with an instantaneous quality score of 65 but a potential error probability of 0.85 is also marked as a "data stream error" because 0.85 >= 0.7. Inferring the error type based on intermediate features of the model involves analyzing the output probability distribution of the auxiliary classifier and recording the error type with the highest probability as the inference result. After marking all data streams, a data transmission monitoring report is compiled, which lists each node identifier, corresponding status label, instantaneous quality score, potential error probability, and description of suspicious or erroneous data in tabular form. In some embodiments, the data transmission monitoring report is a table containing the following: node identifier, status label, instant quality score, potential error probability, suspicious feature description, and inferred error type.

[0103] For data streams marked as "suspicious," an automatic review step is also included. A retransmission request instruction is automatically triggered for the monitoring node marked as "suspicious," and supplementary or duplicate data records are received from the corresponding monitoring node. In specific implementation, the monitoring system sends a retransmission request to the monitoring node with node identifier "STN_007," requesting it to retransmit the meteorological data records of the most recent batch or a specified time point. After receiving the instruction, the "STN_007" node rereads and retransmits the requested data records from its local cache or sensor. Real-time integrity verification and outlier detection processing are performed on the supplementary or duplicate data records to form a valid review dataset. Review time-series features and review spatial correlation features are extracted from the valid review dataset. The review time-series features and review spatial correlation features are input into a preset transmission quality assessment model to calculate the real-time review quality score and the review potential error probability.

[0104] Based on the review results, if the immediate review quality score improves to the normal range and the potential error probability decreases, the status label is corrected to "Data Flow Normal"; otherwise, it is confirmed as "Data Flow Error," and the data transmission monitoring report is updated. In specific implementation, "improving to the normal range" can be defined as the immediate review quality score being greater than a preset normal threshold. "Decrease in the potential error probability" can be defined as a decrease in the potential error probability compared to the initial potential error probability by more than a certain percentage, such as 20%. For example, in the initial assessment, the data flow status of node "STN_007" is "Data Flow Suspicious," with an immediate quality score of 70 and a potential error probability of 0.68. After automatic review, the immediate review quality score calculated using supplementary data is 88, and the potential error probability is 0.15. Since the immediate review quality score 88 > 80, and the potential error probability 0.15 decreases by approximately 78% compared to the initial 0.68 (greater than 20%), the system automatically corrects the status label of node "STN_007" in the data transmission monitoring report to "Data Flow Normal." Understandably, if the immediate quality score during review is 50 and the potential error probability during review is 0.72, since the immediate quality score of 50 is not greater than 80, and the potential error probability of 0.72 is not significantly lower than the initial 0.68, the system will confirm the status label of node "STN_007" as "data flow error" and update the monitoring report. The updated data transmission monitoring report reflects the latest status after review and correction.

[0105] The above embodiments are only used to illustrate the technical methods of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical methods of the present invention without departing from the spirit and scope of the technical methods of the present invention.

Claims

1. A method of monitoring the transmission of meteorological data, characterized in that, The method includes: Acquire raw meteorological data streams from multiple monitoring nodes, the raw meteorological data streams including timestamps, latitude and longitude coordinates, meteorological element measurements and node identifiers; The original meteorological data stream is subjected to real-time integrity verification and outlier detection processing to remove data records with missing fields, format errors, and those exceeding reasonable threshold ranges, thereby generating a valid meteorological dataset. Key temporal features and spatial correlation features are extracted from the effective meteorological dataset. The key temporal features include the rate of change and trend of change of elements at adjacent time points, and the spatial correlation features include the gradient distribution and spatial autocorrelation of the same meteorological element between adjacent monitoring nodes. The key temporal features and spatial correlation features are input into a preset transmission quality assessment model to calculate the instantaneous quality score and potential error probability of the data stream corresponding to each monitoring node. Based on the instantaneous quality score and potential error probability, the transmission status of the effective meteorological dataset is classified and labeled, and a data transmission monitoring report containing status labels is generated. The status labels are used to indicate whether the data stream is normal, the data stream is suspicious, or the data stream is erroneous.

2. The method of claim 1, wherein, The raw meteorological data stream is subjected to real-time integrity verification and outlier detection processing to remove data records with missing fields, format errors, and those exceeding reasonable threshold ranges, generating a valid meteorological dataset, including: Check whether each data record in the original meteorological data stream contains a timestamp field, latitude and longitude coordinate field, all preset meteorological element measurement value fields, and node identifier field, and discard data records that are missing any one or more required fields; For each data record containing complete fields, verify whether its timestamp conforms to the time progression logic, whether its latitude and longitude coordinates are within the preset geographic service area, and whether its node identifier exists in the registered node list. For verified data records, read their meteorological element measurement values, compare each meteorological element measurement value with its predefined reasonable value range, and remove records where the meteorological element measurement value is lower than the minimum value or higher than the maximum value; For meteorological element measurements within a reasonable range, further consistency analysis is performed between these measurements and historical adjacent values ​​at the same monitoring node. The magnitude of change is calculated, and data records whose magnitude of change exceeds the physically reasonable limit of abrupt change are removed. The remaining data records that have passed all verification steps are packaged and marked as the valid meteorological dataset.

3. The method of claim 2, wherein, The measurement values ​​of meteorological elements within a reasonable range are further analyzed for consistency with historical adjacent values ​​at the same monitoring node, including: The historical meteorological dataset of the monitoring node is obtained from the local cache, and the meteorological element measurement values ​​of multiple historical time points closest to the current time are extracted from the historical meteorological dataset. Calculate the difference between the current meteorological element measurement value and the average value of the meteorological element measurement values ​​at the nearest multiple historical time points; The difference is compared with the allowable change threshold preset based on the meteorological element type. The allowable change threshold reflects the maximum reasonable change that different meteorological elements can undergo in adjacent time periods. If the difference exceeds the preset allowable change threshold for the corresponding type of meteorological element, the current data record is marked as a suspected anomaly. For data records marked as suspected anomalies, further check the transmission signal quality logs of the monitoring nodes near the data generation time. If the signal quality is lower than the set standard, it is determined to be a transmission error. Otherwise, spatial consistency is checked in conjunction with data from neighboring monitoring nodes, and a final decision is made to retain or remove the record.

4. The method of claim 1, wherein, Key temporal and spatial correlation features are extracted from the effective meteorological dataset, including: For the time-series data sequence of each monitoring node in the effective meteorological dataset, the difference between the measured values ​​of meteorological elements between adjacent sampling times is calculated, and the difference is divided by the time interval to obtain the rate of change of the element. Statistical analysis is performed on the sequence of the rate of change of the elements to identify whether they show an increasing, decreasing or stable trend within a certain time window, and a trend description is generated. The trend description is used to characterize the temporal change pattern of the corresponding meteorological elements. At the same time, multiple spatially adjacent monitoring nodes are searched, the differences in the measured values ​​of the same meteorological element between the monitoring nodes are compared, and the difference between the maximum and minimum values ​​is calculated as a representation of the gradient distribution. Based on the principles of geostatistics, the correlation coefficient of the same meteorological element measurements at the same time among all node pairs in the entire monitoring network is calculated. By analyzing the pattern of correlation coefficient decay with distance, the spatial autocorrelation is quantified. The correlation coefficient is calculated by the Pearson correlation coefficient formula, for two spatial points i and j, with meteorological element measurements and , the correlation coefficient between node i and node j is calculated by the following formula: in, This represents the correlation coefficient between node i and node j. Let represent the covariance of the meteorological element measurements between node i and node j. This represents the variance of the meteorological element measurements at node i. This represents the variance of the meteorological element measurements at node j.

5. A meteorological data transmission monitoring method according to claim 4, characterized in that, The calculation of the correlation coefficient between all nodes in the entire monitoring network at the same time for the same meteorological element measurement includes: Obtain the valid meteorological dataset of all monitoring nodes at the current moment to form a spatial data point set, where each point contains latitude and longitude coordinates and at least one meteorological element measurement value; Calculate the distance between all pairs of points in the spatial data point set, and the covariance of the meteorological element measurements of these two points; Based on the covariance and the variance of each meteorological element measurement, the correlation coefficient between each pair of nodes is calculated; All the calculated correlation coefficients are grouped according to their corresponding distances between nodes, and the average correlation coefficient within each distance group is calculated. Plot the curve of the average correlation coefficient as a function of distance, fit the mathematical model parameters of the curve, and use the mathematical model parameters as the feature vector for quantifying the spatial autocorrelation.

6. The meteorological data transmission and monitoring method according to claim 1, characterized in that, The key temporal and spatial correlation features are input into a preset transmission quality assessment model to calculate the instantaneous quality score and potential error probability of the data stream corresponding to each monitoring node, including: The transmission quality assessment model is constructed based on historical transmission stability data and the physical constraint relationship of meteorological elements. It includes multiple parallel processing feature analysis branches, wherein the first feature analysis branch receives the key time series features and generates a time series anomaly index by analyzing whether the rate and trend of element change conflict with meteorological physical laws. The second feature analysis branch receives the spatial correlation features and generates a spatial consistency deviation by analyzing whether the corresponding node data and the surrounding node data deviate from a preset normal threshold in terms of gradient distribution and spatial autocorrelation pattern. The temporal anomaly index and spatial consistency deviation are fused and input into a comprehensive evaluation sub-model. The comprehensive evaluation sub-model learns the patterns of historically labeled fault data and normal data and outputs an original quality score. The original quality score is normalized and mapped to a predetermined scoring range to obtain the instantaneous quality score; Simultaneously, the activation values ​​of the intermediate layers of the comprehensive evaluation sub-model are extracted and input into the auxiliary classifier. The auxiliary classifier is used to output the probability values ​​of various types of errors in the corresponding data stream, and the maximum value among the probability values ​​of each category is taken as the potential error probability.

7. A meteorological data transmission monitoring method according to claim 6, characterized in that, The transmission quality assessment model is trained through the following steps before deployment: Collect historical meteorological data records and their corresponding manually verified quality labels, wherein the quality labels indicate whether the corresponding data record is valid or what type of error it belongs to; Feature extraction is performed on the historical meteorological data records to generate key historical temporal features and historical spatial correlation features; A neural network model is constructed as the initial transmission quality assessment model. Its input layer corresponds to the dimensions of the historical key temporal features and historical spatial correlation features, and its output layer includes a quality score output and an error probability output. Historical data with the quality labels are used as training samples and input into the initial transmission quality assessment model. The internal parameters of the model are adjusted through the backpropagation algorithm to minimize the difference between the quality score output by the model and the human score, while minimizing the cross-entropy loss between the error probability distribution output by the model and the actual error type. The model performance is evaluated using an independent validation dataset. The model structure or hyperparameters are adjusted until the preset accuracy requirements are met, resulting in the trained preset transmission quality evaluation model.

8. The meteorological data transmission and monitoring method according to claim 1, characterized in that, Based on the instantaneous quality score and potential error probability, the transmission status of the effective meteorological dataset is classified and labeled, including: Set high, medium, and low thresholds for quality scores, as well as confidence thresholds for the probability of potential errors; For each data stream in the effective meteorological dataset, its instantaneous quality score is compared with the high, medium, and low thresholds of the quality score, and its potential error probability is compared with the confidence threshold. If the instantaneous quality score is higher than the high threshold and the potential error probability is lower than the confidence threshold, the transmission status of the corresponding data stream will be marked as "data stream normal". If the instantaneous quality score is between the medium and high thresholds, or the potential error probability is close to but does not exceed the confidence threshold, it is marked as "data stream suspicious" and the specific suspicious characteristics are recorded. If the instantaneous quality score is below the low threshold, or the potential error probability exceeds the confidence threshold, it is marked as a "data flow error," and an attempt is made to infer the error type based on intermediate features of the model. After all data streams are marked, a data transmission monitoring report is generated. The data transmission monitoring report lists the node identifier, corresponding status label, real-time quality score, potential error probability, and suspicious or erroneous description in tabular form.

9. A meteorological data transmission monitoring method according to claim 8, characterized in that, For data streams marked as "suspicious", an automated review step is also included: Automatically trigger a retransmission request command for the monitoring node marked as "data flow suspicious", and receive supplementary or duplicate data records from the corresponding monitoring node; Real-time integrity verification and outlier detection are performed on the supplementary or duplicate data records to form a valid dataset for review; Extract review temporal features and review spatial correlation features from the valid review dataset; The review time-series features and review spatial correlation features are input into a preset transmission quality assessment model to calculate the review instant quality score and the review potential error probability. Based on the review results, if the immediate quality score improves to the normal range and the probability of potential errors decreases, the status label is corrected to "data flow normal"; otherwise, it is confirmed as "data flow error" and the data transmission monitoring report is updated.

10. A meteorological data transmission and monitoring system, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the meteorological data transmission monitoring method as described in any one of claims 1 to 9.

Images