Mobile terminal meteorological observation quality auditing system based on Internet

Abstract

The invention discloses a mobile terminal meteorological observation quality auditing system based on the Internet, and relates to the technical field of meteorological observation, and the system comprises a multi-modal data collection unit which is used for collecting meteorological parameters in real time, and carrying out the fusion processing of cloud data and local data; the data auditing unit is used for authenticating the meteorological sensor to ensure the reliability and authenticity of the data; the data preprocessing unit is used for performing data verification on the collected meteorological data to improve the quality of the data; the data missing value filling unit is used for predicting missing meteorological data based on the LSTM neural network and filling missing values; the network self-adaptive switching unit is used for carrying out network switching to ensure stable and safe transmission of data in a complex environment; and the data security control unit is used for ensuring the security of the data. Through the application technology of the mobile terminal, an optimized data processing mode and efficient system integration, the working efficiency and the data reliability are improved.

Description

Technical Field

[0001] This invention relates to the field of meteorological observation technology, specifically to a mobile meteorological observation quality review system based on the Internet. Background Technology

[0002] Meteorological data is the core basis for extreme weather warnings. For example, the prediction of strong wind disaster risks needs to integrate geographical location, environmental data (vegetation index, terrain type), and real-time meteorological data (wind speed, air pressure, temperature), and generate dynamic risk scores through quantitative models to provide scientific support for emergency response. In 2023, global losses due to meteorological disasters exceeded US$300 billion, highlighting the crucial role of high-precision data in reducing casualties and economic losses.

[0003] Many industries also rely on accurate meteorological data. For example, in power dispatching, load fluctuations in new energy power plants (wind / solar) are directly affected by weather conditions. In traffic management, airports and ports rely on ground-based cloud image segmentation technology to identify cloud height and type, avoiding flight delays or accidents caused by low visibility. In agricultural production, portable weather stations can monitor temperature, humidity, and optical rainfall in real time, providing data support for irrigation and pest and disease control.

[0004] For example, Chinese patent CN108984580B discloses a dynamic management system and method for meteorological station network information. Based on the dynamic management system for meteorological station network information, it proposes a two-level review and management mechanism, a version management mechanism, and a detection mechanism for erroneous information of meteorological station network information. This effectively solves the problems caused by the lack of quality control mechanism in the business management flow path of existing meteorological station network information management methods, as well as the inconsistency of station network information caused by the lack of unified management and standardization due to the flow of information through two different paths.

[0005] The above patents suffer from the problems described in the background technology: the models of various weather stations are different, and the data formats are significantly different, requiring manual conversion for fusion and analysis, which is inefficient and prone to errors; on-site audits are usually conducted in scenarios with unstable network environments, making it difficult to guarantee data synchronization and real-time performance. Summary of the Invention

[0006] The purpose of this invention is to provide a mobile-based quality management system audit software that solves the problems of delay and inconsistency in the traditional on-site audit data recording, transmission and synchronization process by using mobile application technology, optimized data processing methods and efficient system integration, thereby improving work efficiency and data reliability.

[0007] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A mobile meteorological observation quality review system based on the Internet, comprising: The multimodal data acquisition unit integrates voice input, image capture, and OCR text recognition functions to collect meteorological parameters in real time and perform fusion processing on cloud data and local data. The data verification unit is used to ensure the reliability and authenticity of the data by certifying the meteorological sensors; The data preprocessing unit is used to verify the collected meteorological data and improve its quality. The missing value imputation unit is used to predict missing meteorological data based on LSTM neural network and fill in the missing values; The network adaptive switching unit dynamically selects the optimal transmission network by constructing a network scoring function and performs network switching to ensure stable and secure data transmission in complex environments. The data security control unit is used to ensure data security through data encryption, deployment of intrusion detection systems and firewalls, and implementation of access control.

[0008] Furthermore, in the data review unit, each meteorological sensor device is pre-programmed with a globally unique device ID and digital certificate at the time of manufacture. The digital certificate is metadata, including: device model, production batch, and calibration validity period. When a device connects to the system, the mobile device reads the device's digital certificate via NFC or Bluetooth and initiates an online verification request to the CA center. This includes: certificate chain integrity check, certificate revocation list query, and signature algorithm strength verification.

[0009] Furthermore, the review unit also includes constructing a dynamic credibility score for each meteorological sensor data, with the score ranging from 0 to 1 and updated every 5 minutes; Among them, when the dynamic reliability score is greater than or equal to 0.9, the meteorological sensor data is directly approved and marked as highly reliable; When the dynamic reliability score is greater than or equal to 0.7 and less than 0.9, the meteorological sensor data needs to be manually reviewed and marked as medium reliability. When the dynamic reliability score is less than 0.7, the data acquisition of the meteorological sensor is automatically frozen and an equipment maintenance work order is generated.

[0010] Furthermore, the data preprocessing unit specifically includes the following steps for verifying the collected meteorological data: Define a meteorological data regular expression template library to verify the data format. When the data is in the wrong format, it will automatically return an error code and push it to the mobile device to request re-collection. It has a built-in meteorological and physical rule base to perform logical verification on the collected meteorological data. When the logical verification fails, the collected meteorological data is invalidated and re-collected. Calculate the spatial and temporal deviations of meteorological data and perform spatiotemporal correlation verification. When the spatial deviation is greater than 3 and the temporal deviation is greater than 2.5, the verification fails. Spatial interpolation is used to repair the meteorological data that fails the verification. The repair value is calculated as 0.5 × the value of the nearest station + 0.3 × the value of the second nearest station + 0.2 × the historical average value of this station. The repair result is marked with a status code, and the original value before repair is recorded.

[0011] Furthermore, the meteorological physics rule base includes: single-element threshold rules, inter-element coupling rules, and special rules for extreme events; The single-element threshold rules include setting rule thresholds for temperature, relative humidity, precipitation, wind speed, and air pressure. The coupling rules between the elements include: temperature-humidity coupling: when the temperature is greater than 35 degrees, the relative humidity must be less than or equal to 95%; precipitation-humidity coupling: when the precipitation is greater than 0.1 mm, the relative humidity must be greater than or equal to 85%; wind speed-pressure coupling: when the wind speed is greater than 17 m / s, the air pressure must be less than 1010 hPa; visibility-humidity coupling: when the relative humidity is greater than 90%, the visibility must be less than 10 km. The special rules for extreme events include: Typhoon scenario: When the wind speed is greater than 32.7 m / s and the air pressure is less than 970 hPa, the precipitation is allowed to exceed the threshold by 300%; Sandstorm scenario: When the visibility is less than 1 km and the relative humidity is less than 30%, the temperature is allowed to drop sharply by 8℃ / hour; Cold wave scenario: When the temperature drops by more than 10℃ in 24 hours, the air pressure is allowed to rise by 20 hPa in 24 hours.

[0012] Furthermore, in the data missing value filling unit, the data missing detection conditions include: marking missing when no data is received for two consecutive acquisition cycles, marking missing when the data value exceeds the sensor range, and marking missing when the device dynamic reliability score is less than 0.7; when data missing is triggered, LSTM neural network prediction is immediately started and a missing alarm is pushed to the mobile terminal, along with the site ID and parameter type.

[0013] Furthermore, the LSTM neural network specifically includes: an input layer, a spatiotemporal attention layer, an LSTM hidden layer, and an output layer; The input layer inputs 7-dimensional features, which specifically include: standardized timestamp data, latitude coordinates of the geographical location, longitude coordinates of the geographical location, distance-weighted observations of three neighboring meteorological stations, and the moving average of the historical data of the local station over 24 consecutive hours. The spatiotemporal attention layer dynamically weights features through a dual-path attention mechanism, wherein the spatial attention path calculates the spatial association weights between the current station and neighboring stations, and the temporal attention path extracts key temporal features from historical time series. The LSTM hidden layer adopts a two-layer gated recurrent unit structure. The first hidden layer is configured with 128 LSTM neuron units and outputs the feature sequence of the complete time step. The second hidden layer is configured with 64 LSTM neuron units and outputs the hidden state of the final time step. The output layer is configured with a dual-branch parallel output structure. The main prediction branch outputs the predicted values ​​of missing meteorological parameters through a linear activation function; the confidence evaluation branch outputs the confidence score of the predicted values ​​through a sigmoid activation function.

[0014] Furthermore, in the data missing value imputation unit, when the confidence score output by the LSTM neural network is less than 0.85, secondary prediction based on satellite data is invoked, specifically including: A pre-trained ResNet-34 convolutional neural network model was used to process satellite imagery. The input layer receives a 256×256 pixel three-channel composite image; Spatial features are extracted step by step through 5 convolutional blocks, where each convolutional kernel has a size of 3×3 and a stride of 2; Finally, a 16-dimensional feature vector is output and input into the fully connected layer; The satellite inversion results are output by linear activation of the regression layer; A joint prediction model is constructed to update the final prediction value, where the final prediction value = 0.7 × LSTM prediction value + 0.3 × satellite inversion value.

[0015] Furthermore, in the network adaptive switching unit, the calculation formula for the network scoring function is as follows: in, Indicates online ratings, Indicates bandwidth. Indicates time delay. Indicates packet loss rate. Indicates the power consumption coefficient. , , and These represent the weights of bandwidth, latency, packet loss rate, and power consumption, respectively.

[0016] Furthermore, the network adaptive switching rules in the network scoring function specifically include: when the system detects that the score of a candidate network exceeds the score of the currently used network by more than 25%, network switching is automatically triggered; if the packet loss rate of the current network exceeds the 5% threshold for three consecutive sampling periods, network switching to the candidate network with the highest score is forcibly initiated regardless of the score comparison result; when the system detects more than 3 network switchings within 5 minutes, anti-oscillation protection is automatically activated, and the scoring gain threshold is increased to 40%, and this protection state lasts for 10 minutes; for networks that fail two consecutive switching verifications, they are automatically marked as unavailable and isolated for 10 minutes, during which time the system excludes the network from participating in the scoring and ranking.

[0017] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. The composite deviation model of this invention reduces the false alarm rate, the scoring function optimization reduces the switching delay, and the spatiotemporal attention mechanism improves the accuracy of missing value prediction.

[0018] 2. The LSTM prediction of this invention only initiates satellite verification when the model itself lacks confidence, thus avoiding resource waste; a 5-minute timeout mechanism is set for satellite data access, and the system automatically reverts to the LSTM result after the timeout; the fusion weights are adjusted in real time according to the quality of the satellite data to improve the robustness of the results.

[0019] 3. The network scoring function constructed in this invention does not require preset scenario rules for weights. It is autonomously optimized based on real-time data, identifies sudden changes in network state through a sliding window, and uses the tanh function to avoid weight jumps, thus ensuring connection stability.

[0020] 4. This invention achieves intelligent handover decisions through a triple mechanism of dynamic thresholds, service awareness, and stability protection. It maximizes network utilization efficiency while ensuring data transmission continuity.

[0021] 5. This invention optimizes data storage and transmission: It utilizes efficient indexing, partitioned storage, and data compression technologies to ensure rapid data processing and transmission while reducing network bandwidth consumption; it also enhances data security and reliability: A VPN encrypted channel ensures secure data transmission, avoiding the risk of information leakage. Attached Figure Description

[0022] Other features, objects, and advantages of the invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings: Figure 1 This is a schematic diagram of the system units according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the system architecture according to an embodiment of the present invention; Figure 3 This is a schematic diagram of the mobile terminal technical architecture according to an embodiment of the present invention. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0024] like Figure 1 As shown, a mobile meteorological observation quality review system based on the Internet includes: The multimodal data acquisition unit integrates voice input, image capture, and OCR text recognition functions to collect meteorological parameters in real time and perform fusion processing on cloud data and local data. The data verification unit is used to ensure the reliability and authenticity of the data by certifying the meteorological sensors; The data preprocessing unit is used to verify the collected meteorological data and improve its quality. The missing value imputation unit is used to predict missing meteorological data based on LSTM neural network and fill in the missing values; The network adaptive switching unit dynamically selects the optimal transmission network by constructing a network scoring function and performs network switching to ensure stable and secure data transmission in complex environments. The data security control unit is used to ensure data security through data encryption, deployment of intrusion detection systems and firewalls, and implementation of access control.

[0025] In the data auditing unit, each meteorological sensor device is pre-programmed with a globally unique device ID and digital certificate at the factory. The digital certificate is metadata, including: device model, production batch and calibration validity period. When a device connects to the system, the mobile device reads the device's digital certificate via NFC or Bluetooth and initiates an online verification request to the CA center. This includes: certificate chain integrity check, certificate revocation list query, and signature algorithm strength verification.

[0026] The review unit also includes building a dynamic credibility score for each meteorological sensor data, with the score ranging from 0 to 1 and updated every 5 minutes; The formula for calculating the dynamic credibility score is as follows: in, express, , and These represent the corresponding weights. Indicates the calibration status (0 / 1). Indicates the number of days since the last calibration (valid for ≤30 days). Indicates the signal strength ratio (measured value / theoretical value); Weight: =0.6, =0.3, =0.1; Among them, when the dynamic reliability score is greater than or equal to 0.9, the meteorological sensor data is directly approved and marked as highly reliable; When the dynamic reliability score is greater than or equal to 0.7 and less than 0.9, the meteorological sensor data needs to be manually reviewed and marked as medium reliability. When the dynamic reliability score is less than 0.7, the data acquisition of the meteorological sensor is automatically frozen and an equipment maintenance work order is generated.

[0027] The data preprocessing unit specifically includes the following steps for verifying the collected meteorological data: Define a meteorological data regular expression template library to verify the data format. When the data is in the wrong format, it will automatically return an error code and push it to the mobile device to request re-collection. It has a built-in meteorological and physical rule base to perform logical verification on the collected meteorological data. When the logical verification fails, the collected meteorological data is invalidated and re-collected. Calculate the spatial and temporal deviations of meteorological data and perform spatiotemporal correlation verification. When the spatial deviation is greater than 3 and the temporal deviation is greater than 2.5, the verification fails. Spatial interpolation is used to repair the meteorological data that fails the verification. The repair value is calculated as 0.5 × the value of the nearest station + 0.3 × the value of the second nearest station + 0.2 × the historical average value of this station. The repair result is marked with a status code, and the original value before repair is recorded.

[0028] Wherein, spatial deviation = |current value - mean value of neighboring 50km stations| / regional standard deviation, and temporal deviation = |current value - mean value of the same period in history| / (3 × historical standard deviation); Composite alarm conditions: spatial deviation greater than 3.0 and temporal deviation greater than 2.5.

[0029] The meteorological and physical rule base includes: single-element threshold rules, inter-element coupling rules, and special rules for extreme events; The single-element threshold rules include setting rule thresholds for temperature, relative humidity, precipitation, wind speed, and air pressure, as shown in Table 1. Table 1 The coupling rules between the elements include: temperature-humidity coupling: when the temperature is greater than 35 degrees, the relative humidity must be less than or equal to 95%; precipitation-humidity coupling: when the precipitation is greater than 0.1 mm, the relative humidity must be greater than or equal to 85%; wind speed-pressure coupling: when the wind speed is greater than 17 m / s, the air pressure must be less than 1010 hPa; visibility-humidity coupling: when the relative humidity is greater than 90%, the visibility must be less than 10 km. The special rules for extreme events include: Typhoon scenario: When the wind speed is greater than 32.7 m / s and the air pressure is less than 970 hPa, the precipitation is allowed to exceed the threshold by 300%; Sandstorm scenario: When the visibility is less than 1 km and the relative humidity is less than 30%, the temperature is allowed to drop sharply by 8℃ / hour; Cold wave scenario: When the temperature drops by more than 10℃ in 24 hours, the air pressure is allowed to rise by 20 hPa in 24 hours.

[0030] The data missing value filling unit includes the following data missing detection conditions: marking missing data when no data is received for two consecutive acquisition cycles, marking missing data when the data value exceeds the sensor range, and marking missing data when the device dynamic reliability score is less than 0.7. When data missing is triggered, the LSTM neural network prediction is started immediately and a missing alarm is pushed to the mobile terminal, along with the site ID and parameter type.

[0031] The LSTM neural network specifically includes: an input layer, a spatiotemporal attention layer, an LSTM hidden layer, and an output layer; The input layer inputs 7-dimensional features, which specifically include: standardized timestamp data, latitude coordinates of the geographical location, longitude coordinates of the geographical location, distance-weighted observations of three neighboring meteorological stations, and the moving average of the historical data of the local station over 24 consecutive hours. The spatiotemporal attention layer dynamically weights features through a dual-path attention mechanism, wherein the spatial attention path calculates the spatial association weights between the current station and neighboring stations, and the temporal attention path extracts key temporal features from historical time series. The LSTM hidden layer adopts a two-layer gated recurrent unit structure. The first hidden layer is configured with 128 LSTM neuron units and outputs the feature sequence of the complete time step. The second hidden layer is configured with 64 LSTM neuron units and outputs the hidden state of the final time step. The output layer is configured with a dual-branch parallel output structure. The main prediction branch outputs the predicted values ​​of missing meteorological parameters through a linear activation function; the confidence evaluation branch outputs the confidence score of the predicted values ​​through a sigmoid activation function.

[0032] In the missing value imputation unit, when the confidence score output by the LSTM neural network is less than 0.85, secondary prediction based on satellite data is invoked. Specifically, this includes: A pre-trained ResNet-34 convolutional neural network model was used to process satellite imagery. The input layer receives a 256×256 pixel three-channel composite image; Spatial features are extracted step by step through 5 convolutional blocks, where each convolutional kernel has a size of 3×3 and a stride of 2; Finally, a 16-dimensional feature vector is output and input into the fully connected layer; The satellite inversion results are output by linear activation of the regression layer; A joint prediction model is constructed to update the final prediction value, where the final prediction value = 0.7 × LSTM prediction value + 0.3 × satellite inversion value.

[0033] The 16-dimensional feature vector specifically includes: Infrared brightness temperature: Temperature value (K) of radiance conversion for the 10.8μm channel. Visible light reflectance: Surface reflectance (%) of 0.65μm channel Cloud top height: calculated based on infrared brightness temperature and standard atmospheric profile (km) Cloud optical thickness: differential inversion of visible and infrared channels (dimensionless) Surface temperature gradient: standard deviation (K) of a 3×3 pixel window Vegetation index: Normalized difference of (NIR-Red) / (NIR+Red) Water vapor index: ratio of 6.7μm to 10.8μm channels Cloud phase indicators: 0 = Clear sky, 1 = Liquid cloud, 2 = Ice crystal cloud Surface albedo: Shortwave broadband albedo (0~1) Aerosol optical thickness: 0.55 μm; aerosol extinction coefficient. Boundary layer height: retrieved by fusion of infrared and microwave data (m) Earth's surface thermal inertia: diurnal brightness temperature difference inversion (J·m -2 ·K -1 ·s -0 · 5 ) Cloud motion vector: Displacement (m / s) calculated by cross-correlation of consecutive frame images. Precipitation probability: Estimated using the multispectral threshold method (0~1) Terrain shading factor: Solar shading rate calculated by the digital elevation model (0~1). Data freshness: The reciprocal of the time difference (in minutes) between the time of data collection and the current time.

[0034] Data source for the fill-in results: PRED_LSTM (pure model prediction), PRED_SAT (satellite-assisted prediction).

[0035] In the network adaptive handover unit, the network scoring function is calculated using the following formula: in, Indicates online ratings, Indicates bandwidth. Indicates time delay. Indicates packet loss rate. Indicates the power consumption coefficient. , , and These represent the weights of bandwidth, latency, packet loss rate, and power consumption, respectively. The power consumption coefficients are: 4G = 1.0, 5G = 1.2, Wi-Fi = 0.8, and private network = 0.6. The weights for bandwidth, latency, packet loss rate, and power consumption are dynamic weights, and the specific calculation formula is as follows: in, This represents the dynamic weights, where i=1, i=2, i=3, and i=4 represent the bandwidth, latency, packet loss rate, and power consumption coefficients, respectively. The basic weights representing bandwidth, latency, packet loss rate, and power consumption are 0.4, 0.3, 0.2, and 0.1, respectively. This represents the sensitivity coefficient, with a default value of 0.3. This represents the hyperbolic tangent function, with an output range of [-1, 1]. This represents the non-linear adjustment factor, with a default value of 2.0. This indicates the current measured values, including bandwidth / Mbps, latency / ms, packet loss rate / %, and power consumption coefficient. This represents the sliding window mean, which is the exponentially weighted average of the most recent 10 measurements. This represents the standard deviation of the sliding window (the 10 most recent measurements).

[0036] The network scoring function includes the following adaptive switching rules: when the system detects that the score of a candidate network exceeds the score of the currently used network by more than 25%, network switching is automatically triggered; if the packet loss rate of the current network exceeds the 5% threshold for three consecutive sampling periods, network switching to the candidate network with the highest score is forcibly initiated regardless of the score comparison result; when the system detects more than three network switching events within 5 minutes, anti-vibration protection is automatically activated, increasing the score gain threshold to 40%, and this protection state lasts for 10 minutes; for networks that fail two consecutive switching verifications, they are automatically marked as unavailable and isolated for 10 minutes, during which time the system excludes the network from participating in the scoring and ranking.

[0037] Switching decision-making processes: Scans available networks (4G / 5G / Wi-Fi / weather network / satellite) every 30 seconds. Calculate and rank the network scores. Execution switching conditions: Optimal network score > Current score × 1.25 The current network packet loss rate is greater than 10% for three consecutive times.

[0038] like Figure 2 As shown, this illustrates the communication and data flow process between the mobile terminal, PC terminal, DingTalk platform, meteorological network, and cloud server.

[0039] This invention utilizes an H5 page to implement the review function on mobile devices, integrating it with the DingTalk platform for user authentication and information push. During on-site review, reviewers collect data via mobile devices, supporting voice input, image input (such as photos and videos), and OCR technology for automatic text recognition. All data can be uploaded to a PC database in real time, ensuring the completeness and timeliness of the review information.

[0040] The mobile application uses a local SQLite database as its core data caching solution, fully leveraging its lightweight, embedded, and high-performance advantages to achieve fast data storage and retrieval without relying on external servers. SQLite's ACID transaction characteristics ensure data consistency and reliability, making it particularly suitable for the complex and ever-changing network environment of mobile devices. To further improve data query efficiency, the software employs a multi-level indexing optimization strategy, including creating B-tree indexes for frequently queried fields, using composite indexes for complex query scenarios, and combining this with database partitioning technology to logically partition data by time, category, or user behavior. For example, time-series data (such as chat logs or logs) is partitioned by day or month; user behavior data is partitioned by user ID or geographic location. This partitioning strategy significantly reduces the data scanning range during queries, improves response speed, and reduces memory consumption.

[0041] To ensure the security of audit data during transmission, the Advanced Encryption Standard (AES) is used to encrypt the data. AES provides strong encryption capabilities, effectively protecting data from unauthorized access. Furthermore, a digital signature is added to each data packet to verify the authenticity of the data source and ensure data integrity. Asymmetric encryption algorithms (such as RSA) are used to securely exchange the keys used for AES encryption, thereby enhancing the security of the entire encryption process.

[0042] Throughout the transmission process, a multi-layered security strategy is implemented to protect data security. In addition to the aforementioned data encryption, an intrusion detection system (IDS) and firewall are deployed to prevent malicious attacks. Simultaneously, a role-based access control (RBAC) system is used to manage user permissions, ensuring that only authorized personnel can access sensitive information. Regular network security audits are conducted to promptly identify and resolve potential security threats.

[0043] In addition, the software incorporates data encryption and privacy protection mechanisms, encrypting locally stored sensitive information (such as user credentials and personal data) to ensure data security. During data transmission, TLS / SSL protocols are used for encrypted transmission to prevent data theft or tampering. Furthermore, the software supports data backup and recovery functions, allowing users to save important data via cloud or local backups to avoid data loss due to device loss or damage.

[0044] like Figure 3 As shown, this invention deeply integrates with the DingTalk platform, implementing Single Sign-On (SSO) via the OAuth 2.0 protocol to ensure seamless switching between multiple systems. Furthermore, it utilizes WebSocket technology to deliver task notifications in real-time, ensuring timely information delivery. The system also employs a VPN encrypted channel to ensure secure data transmission and prevent data leakage.

[0045] This invention deeply integrates with the DingTalk platform, fully leveraging its powerful enterprise-level ecosystem capabilities to achieve efficient and secure Single Sign-On (SSO) functionality through the OAuth 2.0 protocol. As an industry-standard authorization framework, OAuth 2.0 allows users to directly access various functions of the system after authentication through the DingTalk platform without exposing their passwords. This mechanism not only simplifies the user login process but also significantly improves system security, avoiding the management complexity and potential security risks associated with multiple account systems. After user authentication on the DingTalk platform, the system automatically generates and manages access tokens and refresh tokens, ensuring seamless switching between multiple systems and supporting automatic renewal after session timeout, further enhancing the user experience.

[0046] Regarding task notifications, this invention employs WebSocket technology to implement real-time push functionality. Compared to traditional HTTP polling, WebSocket, based on a full-duplex communication protocol, enables the establishment of persistent connections between the client and server, achieving low-latency, high-efficiency data transmission. When a new task or status update occurs in the system, the server immediately pushes the notification to the client via the WebSocket channel, ensuring users receive critical information promptly. For example, when a user's task status changes (such as task completion, task extension, or new task assignment), the system pushes a notification to the user's DingTalk client in real time, combined with DingTalk's message reminder functions (such as pop-ups, sound alerts, etc.) to ensure timely information delivery. Furthermore, the system supports categorized management and personalized settings for notifications, allowing users to choose to receive specific types of notifications according to their needs, avoiding information overload.

[0047] To ensure data security during transmission, this invention introduces VPN (Virtual Private Network) encrypted tunnel technology. VPN establishes an encrypted tunnel over a public network to encrypt and protect data transmission between the user and the system, effectively preventing data theft or tampering during transmission. The system employs industry-leading encryption algorithms (such as AES-256) and a secure key management mechanism to ensure adequate data protection even in public network environments. Furthermore, VPN technology supports data transmission path optimization, dynamically selecting the optimal path based on network conditions to further improve transmission efficiency.

[0048] In addition to the core functions mentioned above, this invention leverages the open capabilities of the DingTalk platform to achieve more value-added features. For example, by integrating with DingTalk's calendar and schedule management functions, the system can automatically synchronize tasks to the user's DingTalk calendar, facilitating unified management of time and tasks. Simultaneously, the system supports deep integration with DingTalk's group chat and document collaboration functions, allowing users to directly initiate group chats or share documents from the task details page, improving team collaboration efficiency. Furthermore, the system provides rich API interfaces, supporting seamless integration with other internal enterprise systems, further expanding the system's application scenarios.

[0049] The examples described herein are merely preferred embodiments of the invention and are not intended to limit the concept and scope of the invention. Any modifications and improvements made by those skilled in the art to the technical solutions of the invention without departing from the design concept of the invention should fall within the protection scope of the invention.

Claims

1. A mobile meteorological observation quality review system based on the Internet, characterized in that, include: The multimodal data acquisition unit integrates voice input, image capture, and OCR text recognition functions to collect meteorological parameters in real time and perform fusion processing on cloud data and local data. The data verification unit is used to ensure the reliability and authenticity of the data by certifying the meteorological sensors; The data preprocessing unit is used to verify the collected meteorological data and improve its quality. The missing value imputation unit is used to predict missing meteorological data based on LSTM neural network and fill in the missing values; The network adaptive switching unit dynamically selects the optimal transmission network by constructing a network scoring function and performs network switching to ensure stable and secure data transmission in complex environments. The data security control unit is used to ensure data security through data encryption, deployment of intrusion detection systems and firewalls, and implementation of access control.

2. The system according to claim 1, characterized in that, In the data auditing unit, each meteorological sensor device is pre-programmed with a globally unique device ID and digital certificate at the factory. The digital certificate is metadata, including: device model, production batch and calibration validity period. When a device connects to the system, the mobile device reads the device's digital certificate via NFC or Bluetooth and initiates an online verification request to the CA center. This includes: certificate chain integrity check, certificate revocation list query, and signature algorithm strength verification.

3. The system according to claim 2, characterized in that, The review unit also includes building a dynamic credibility score for each meteorological sensor data, with the score ranging from 0 to 1 and updated every 5 minutes; Among them, when the dynamic reliability score is greater than or equal to 0.9, the meteorological sensor data is directly approved and marked as highly reliable; When the dynamic reliability score is greater than or equal to 0.7 and less than 0.9, the meteorological sensor data needs to be manually reviewed and marked as medium reliability. When the dynamic reliability score is less than 0.7, the data acquisition of the meteorological sensor is automatically frozen and an equipment maintenance work order is generated.

4. The system according to claim 3, characterized in that, The data preprocessing unit specifically includes the following steps for verifying the collected meteorological data: Define a meteorological data regular expression template library to verify the data format. When the data is in the wrong format, it will automatically return an error code and push it to the mobile device to request re-collection. It has a built-in meteorological and physical rule base to perform logical verification on the collected meteorological data. When the logical verification fails, the collected meteorological data is invalidated and re-collected. Calculate the spatial and temporal deviations of meteorological data and perform spatiotemporal correlation verification. When the spatial deviation is greater than 3 and the temporal deviation is greater than 2.5, the verification fails. Spatial interpolation is used to repair the meteorological data that fails the verification. The repair value is calculated as 0.5 × the value of the nearest station + 0.3 × the value of the second nearest station + 0.2 × the historical average value of this station. The repair result is marked with a status code, and the original value before repair is recorded.

5. The system according to claim 4, characterized in that, The meteorological and physical rule base includes: single-element threshold rules, inter-element coupling rules, and special rules for extreme events; The single-element threshold rules include setting rule thresholds for temperature, relative humidity, precipitation, wind speed, and air pressure. The coupling rules between the elements include: temperature-humidity coupling: when the temperature is greater than 35 degrees, the relative humidity must be less than or equal to 95%; precipitation-humidity coupling: when the precipitation is greater than 0.1 mm, the relative humidity must be greater than or equal to 85%; wind speed-pressure coupling: when the wind speed is greater than 17 m / s, the air pressure must be less than 1010 hPa; visibility-humidity coupling: when the relative humidity is greater than 90%, the visibility must be less than 10 km. The special rules for extreme events include: Typhoon scenario: When the wind speed is greater than 32.7 m / s and the air pressure is less than 970 hPa, the precipitation is allowed to exceed the threshold by 300%; Sandstorm scenario: When the visibility is less than 1 km and the relative humidity is less than 30%, the temperature is allowed to drop sharply by 8℃ / hour; Cold wave scenario: When the temperature drops by more than 10℃ in 24 hours, the air pressure is allowed to rise by 20 hPa in 24 hours.

6. The system according to claim 5, characterized in that, The data missing value filling unit includes the following data missing detection conditions: marking missing data when no data is received for two consecutive acquisition cycles, marking missing data when the data value exceeds the sensor range, and marking missing data when the device dynamic reliability score is less than 0.

7. When data missing is triggered, the LSTM neural network prediction is started immediately and a missing alarm is pushed to the mobile terminal, along with the site ID and parameter type.

7. The system according to claim 6, characterized in that, The LSTM neural network specifically includes: an input layer, a spatiotemporal attention layer, an LSTM hidden layer, and an output layer; The input layer inputs 7-dimensional features, which specifically include: standardized timestamp data, latitude coordinates of the geographical location, longitude coordinates of the geographical location, distance-weighted observations of three neighboring meteorological stations, and the moving average of the historical data of the local station over 24 consecutive hours. The spatiotemporal attention layer dynamically weights features through a dual-path attention mechanism, wherein the spatial attention path calculates the spatial association weights between the current station and neighboring stations, and the temporal attention path extracts key temporal features from historical time series. The LSTM hidden layer adopts a two-layer gated recurrent unit structure. The first hidden layer is configured with 128 LSTM neuron units and outputs the feature sequence of the complete time step. The second hidden layer is configured with 64 LSTM neuron units and outputs the hidden state of the final time step. The output layer is configured with a dual-branch parallel output structure. The main prediction branch outputs the predicted values ​​of missing meteorological parameters through a linear activation function; the confidence evaluation branch outputs the confidence score of the predicted values ​​through a sigmoid activation function.

8. The system according to claim 7, characterized in that, In the missing value imputation unit, when the confidence score output by the LSTM neural network is less than 0.85, secondary prediction based on satellite data is invoked. Specifically, this includes: A pre-trained ResNet-34 convolutional neural network model was used to process satellite imagery. The input layer receives a 256×256 pixel three-channel composite image; Spatial features are extracted step by step through 5 convolutional blocks, where each convolutional kernel has a size of 3×3 and a stride of 2; Finally, a 16-dimensional feature vector is output and input into the fully connected layer; The satellite inversion results are output by linear activation of the regression layer; A joint prediction model is constructed to update the final prediction value, where the final prediction value = 0.7 × LSTM prediction value + 0.3 × satellite inversion value.

9. The system according to claim 8, characterized in that, In the network adaptive handover unit, the network scoring function is calculated using the following formula: in, Indicates online ratings, Indicates bandwidth. Indicates time delay. Indicates packet loss rate. Indicates the power consumption coefficient. , , and These represent the weights of bandwidth, latency, packet loss rate, and power consumption, respectively.

10. The system according to claim 9, characterized in that, The network scoring function includes the following adaptive switching rules: when the system detects that the score of a candidate network exceeds the score of the currently used network by more than 25%, network switching is automatically triggered; if the packet loss rate of the current network exceeds the 5% threshold for three consecutive sampling periods, network switching to the candidate network with the highest score is forcibly initiated regardless of the score comparison result; when the system detects more than three network switching events within 5 minutes, anti-vibration protection is automatically activated, increasing the score gain threshold to 40%, and this protection state lasts for 10 minutes; for networks that fail two consecutive switching verifications, they are automatically marked as unavailable and isolated for 10 minutes, during which time the system excludes the network from participating in the scoring and ranking.

Images