Whole-life-cycle big data management method and system for meteorological engineering project

Abstract

The invention relates to the technical field of computers, discloses a full-life-cycle big data management method and system for a meteorological engineering project, and aims to solve the problems of data islands, different standards, processing splitting, decision lag and the like in the prior art. The method comprises the following steps: collecting multi-source heterogeneous original data; performing standardization processing to form a data set with unified semantics and space-time reference; classifying and storing according to the full life cycle stage of the project, and constructing a project entity association index; calling a stage adaptation model based on the data set to execute prediction, diagnosis or optimization; generating visual early warning and operation and maintenance suggestions; and pushing a decision instruction to realize closed-loop management. Through adoption of the technical scheme, unified convergence, efficient retrieval, intelligent analysis and closed-loop decision of meteorological engineering project data can be realized, and data availability, early warning timeliness and management automation level are remarkably improved.

Description

Technical Field

[0001] This invention belongs to the field of computer technology, specifically relating to a method and system for big data management throughout the entire lifecycle of meteorological engineering projects. Background Technology

[0002] With the continuous advancement of meteorological observation technology and the increasing demand for meteorological services, meteorological engineering projects, as crucial infrastructure supporting national disaster prevention and mitigation, climate prediction, and public services, have become significantly more complex and data-intensive. Modern meteorological engineering projects encompass the entire chain from the construction of observation network, data acquisition and transmission, numerical simulation calculations to product release and application, involving multi-source, multi-dimensional, and high-frequency meteorological data streams, which places extremely high demands on the full lifecycle management of data.

[0003] The big data management of the entire lifecycle of meteorological engineering projects aims to achieve unified aggregation, efficient processing, and intelligent application of data from all stages, including project planning, construction, operation and maintenance, and decommissioning assessment. This management approach not only needs to be compatible with multiple data sources such as ground observation, satellite remote sensing, and radar detection, but also needs to support the integrated storage, dynamic updating, and cross-stage traceability of structured and unstructured data to ensure the continuity of meteorological operations and the scientific nature of decision-making.

[0004] Existing technologies for data management in meteorological engineering projects still have significant shortcomings: First, data management is often limited to single business processes or isolated systems, lacking a unified architecture that spans the entire project lifecycle, leading to data fragmentation, inconsistent standards, and difficulty in data sharing. Second, faced with massive amounts of heterogeneous meteorological data, existing systems have limited capabilities in real-time access, efficient storage, and rapid retrieval, making it difficult to meet the high-timeliness requirements of business operations. Third, data quality control and metadata management mechanisms are weak, failing to effectively support data traceability, version control, and reliability assessment. Finally, there is a lack of dynamic data governance strategies oriented towards project evolution, making it difficult to adaptively adjust data organization and service models according to changes in business scenarios. These problems severely restrict the release of data value and the improvement of intelligence levels in meteorological engineering projects, urgently requiring a big data management method and system that can achieve collaborative, efficient, and intelligent management throughout the entire lifecycle. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of existing technologies and provide a big data management method and system for the entire lifecycle of meteorological engineering projects, which can effectively solve the problems in the background technology. Currently, meteorological engineering projects suffer from structural problems throughout their entire lifecycle, including planning, construction, operation, and maintenance. These problems include dispersed multi-source heterogeneous data, inconsistent standards, fragmented processing workflows, isolated analytical models, and lagging decision support. This results in the ineffective release of data value, low project management efficiency, insufficient risk warning capabilities, and an inability to support the refined and intelligent operation and decision-making needs of meteorological engineering projects.

[0006] To achieve the above objectives, the present invention provides the following technical solution: On one hand, a big data management system for the entire lifecycle of meteorological engineering projects. This system comprises the following components: a data acquisition and access module, used to acquire multi-source heterogeneous raw data in real time or in batches from meteorological observation equipment, engineering sensors, business information systems, and external environment databases; a data standardization and fusion module, used to perform format conversion, semantic alignment, spatiotemporal benchmark unification, and quality verification on the raw data to generate a standardized dataset; a lifecycle data warehouse module, used to divide and store structured, semi-structured, and unstructured data according to the meteorological engineering project stages, and establish an association index between project entities and data objects; an intelligent analysis and modeling module, used to construct prediction, diagnosis, and optimization models for different engineering stages based on the standardized dataset; a dynamic decision support module, used to generate visual reports, risk warnings, and operation and maintenance suggestions based on the model output results; and a system collaborative control module, used to issue decision instructions to relevant execution units or business systems to achieve closed-loop management. On the other hand, a big data management method for the entire lifecycle of a meteorological engineering project is provided. The method comprises the following steps: Step S110, collecting raw data covering meteorological elements, equipment status, construction progress, operation and maintenance records, and the external environment through various types of sensors and information systems deployed at the meteorological engineering site and remote platforms; Step S120, performing standardization processing on the raw data, including timestamp alignment, coordinate system unification, unit conversion, missing value imputation, and outlier removal, to form a standardized dataset with consistent semantics and spatiotemporal reference; Step S130, organizing the standardized dataset according to the entire lifecycle stages of the meteorological engineering project—including planning and project initiation, design and construction, and operation. Monitoring, maintenance, updates, and decommissioning assessment—classifying and storing data, and constructing a multi-dimensional association index between project entities and data objects; Step S140, based on the standardized dataset, calling or training corresponding machine learning or physics-driven models according to different stage business needs, and performing state prediction, fault diagnosis, performance evaluation, or resource optimization tasks; Step S150, converting the model analysis results into structured decision information, generating a visual dashboard, risk level warning, and specific operation and maintenance suggestions; Step S160, pushing the decision information to relevant business systems or control terminals through standard interfaces to trigger automated responses or assist manual decision-making, realizing data-driven closed-loop management; Preferably, the data acquisition and access module supports multi-source data access, including ground automatic weather stations, radiosonde radar, satellite remote sensing, IoT sensors, BIM model data, engineering management software logs, and public meteorological databases. It adopts an asynchronous transmission mechanism based on message queues to ensure data throughput stability in high-concurrency scenarios, with data access latency not exceeding 500 milliseconds. Preferably, the data standardization and fusion module has a built-in meteorological ontology knowledge base, which includes meteorological element coding standards (such as WMO No. 306), engineering equipment classification system and project phase definition specifications. Semantic mapping is achieved through a combination of rule engine and deep learning, with a semantic alignment accuracy of no less than 95%. The spatiotemporal reference is uniformly adopted using the WGS-84 coordinate system and UTC time standard, with spatial interpolation error controlled within 10 meters and time synchronization accuracy reaching the millisecond level. Furthermore, the full lifecycle data warehouse module adopts a layered storage architecture, with hot data stored in a distributed in-memory database (such as Redis or Apache Ignite), warm data stored in a columnar storage system (such as Apache Parquet), and cold data archived to object storage. The data access response time is less than 100 milliseconds in hot data scenarios. The project entity association index is built based on a graph database, which supports dynamically associating all data objects generated throughout the project's lifecycle with the project ID as the root node, and the query path depth can reach 10 levels. In addition, the intelligent analysis and modeling module integrates a variety of algorithm models, including an LSTM time series model for equipment failure prediction, a Kriging method for spatial interpolation of meteorological elements, a constraint satisfaction problem (CSP) solver for construction progress optimization, and a multi-objective optimization model for energy efficiency assessment; the model training adopts an online learning mechanism, and is incrementally updated every 24 hours based on new data, and the model prediction accuracy is maintained at over 90%. Furthermore, the visualization reports generated by the dynamic decision support module support multi-terminal adaptive rendering, and the risk warning adopts a three-level threshold mechanism (low, medium, and high). The warning triggering conditions are dynamically adjusted based on historical data statistical distribution and real-time model output. The operation and maintenance suggestions are generated using a hybrid strategy based on rules and case reasoning (CBR), and the suggestion adoption rate has been measured to be no less than 85%. Furthermore, the system's collaborative control module interfaces with external systems through a dual-channel approach of RESTful API and MQTT protocol, achieving a command issuance success rate of no less than 99.9%. The command execution status is transmitted back to the data warehouse in real time, forming a complete operational closed loop. The system supports seamless integration with mainstream engineering management systems such as SCADA, CMMS, and ERP. Compared with the prior art, the present invention has the following beneficial effects: It has enabled unified access, standardized processing, and structured storage of multi-source heterogeneous data throughout the entire lifecycle of meteorological engineering projects, from planning to decommissioning, solving the data silo problem and improving data availability to over 98%. By building a phase-aware data warehouse and project entity association index, data retrieval efficiency and contextual integrity were significantly improved, and the response time for complex association queries was reduced to 1 / 5 of the original system. By introducing a phase-adaptive intelligent analysis model and online learning mechanism, the accuracy and timeliness of state prediction and risk warning are greatly improved, with an average warning lead time of 72 hours. A closed-loop management process from data collection and analysis to decision execution has been established, realizing the automation and intelligence of meteorological engineering project management, reducing the frequency of manual intervention by 60% and reducing operation and maintenance costs by 25%. The system has good scalability and compatibility, and can be quickly adapted to different types of meteorological engineering projects, reducing the deployment cycle to one-third of the original solution. Attached Figure Description

[0007] Figure 1 This is a schematic diagram of the overall technical architecture of the big data management method and system for the entire life cycle of meteorological engineering projects proposed in this invention. Detailed Implementation

[0008] Please refer to Figure 1 To further illustrate the technical means and effects of the present invention in order to achieve the intended purpose, the following detailed description of the specific implementation methods, structures, features and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0009] Example 1 This embodiment applies the big data management method and system for the entire lifecycle of meteorological engineering projects proposed in this invention to the operation and maintenance scenario of a large regional meteorological radar station. This radar station is a core meteorological observation infrastructure, and its stable operation is crucial for regional weather forecasting and disaster prevention and mitigation. The radar station's lifecycle covers multiple stages, from initial planning and design, infrastructure construction, daily operation monitoring, equipment maintenance and updates, to final decommissioning assessment, involving massive amounts of multi-source heterogeneous data, including the radar's own observation data, site environmental monitoring data, power supply system operation data, cooling system performance data, server rack status data, and various engineering logs and maintenance records.

[0010] First, driven by the data acquisition and access module, the system comprehensively and in real-time acquires data from diverse data sources both inside and outside the meteorological radar station. Specifically, this module is responsible for collecting data from the following core sources: Meteorological observation equipment data primarily includes raw I / Q (In-phase / Quadrature) signal data from meteorological radar, processed basic data (such as reflectivity, radial velocity, and spectral width), and higher-level product data (such as combined reflectivity, VCP scan mode information, and Doppler velocity field). This data is transmitted in real-time via a dedicated high-bandwidth fiber optic network in standard formats such as binary streams or HDF5 / NetCDF. The system employs an asynchronous transmission mechanism based on message queues (e.g., an Apache Kafka cluster) to ensure stable and low-latency ingestion of high-concurrency data streams generated at the radar's high scanning frequency. Redundant links and data verification (e.g., CRC) are configured along the data transmission path to ensure data integrity and reliability, with the average data access latency strictly controlled to within 500 milliseconds.

[0011] Engineering sensor data: Environmental sensors (such as temperature, humidity, wind speed, and air pressure sensors) deployed in the radar station equipment room, inside the radome, power supply room, cooling system, and server rack upload real-time data at a rate of seconds via Modbus TCP / IP or SNMP protocols. The Power Monitoring Unit (PMU) collects voltage, current, power factor, and energy consumption data and transmits it via Ethernet interface. Vibration sensors are installed on the radar antenna rotation mechanism and cooling pumps to monitor mechanical operating status. All sensor data is configured with edge computing nodes for initial data aggregation and timestamping, reducing transmission load and ensuring time synchronization accuracy.

[0012] Business information system data includes system logs, operator event records, and alarm information generated by radar control and monitoring software; maintenance work orders, spare parts inventory, and repair history records from the facility management system (CMMS); and performance metrics such as CPU utilization, memory usage, and disk I / O from the local high-performance computing (HPC) cluster. This data is accessed minute-by-minute or event-triggered via Syslog protocol, RESTful API calls, or database replication.

[0013] External Environment Database: This database acquires numerical weather prediction (NWP) model output data (e.g., WRF model results), high-resolution satellite remote sensing imagery, and data from other ground observation stations from regional meteorological centers. This data is acquired in batches via the OGC WCS / WFS service or a dedicated FTP channel to supplement internal radar station data and provide more comprehensive environmental context information.

[0014] The data acquisition and access module is configured with an independent adapter for each data source. It is responsible for initially encapsulating raw data of different protocols and formats and injecting it into a unified message queue, laying the foundation for subsequent standardized processing.

[0015] Next, the data standardization and fusion module performs refined processing on the collected multi-source heterogeneous raw data to eliminate differences between data and form a standardized dataset with consistent semantics and spatiotemporal reference. This process mainly includes the following key steps: Timestamp alignment: All incoming data is synchronized with a unified time server via Network Time Protocol (NTP) to ensure millisecond-level time synchronization accuracy for all time series data. For event data without an explicit timestamp, the system automatically assigns a timestamp based on its reception time or log generation time and performs deduplication.

[0016] Coordinate System 1: The geographical location, observation area, and related geographic information data of radar stations all adopt the Global Geodetic System (WGS-84) coordinate system. For data involving spatial interpolation or geographic matching, the system performs coordinate transformation and projection through a Geographic Information System (GIS) database, with spatial interpolation errors controlled within 10 meters.

[0017] Unit conversion: Standardize the measurement units from different sensors, such as temperature to degrees Celsius, wind speed to meters per second, and voltage to volts, to avoid analysis errors caused by inconsistent units.

[0018] Missing value imputation: For missing data caused by sensor failure, network interruption, or abnormal data transmission, the system employs an adaptive imputation algorithm. For short-term, continuously missing time-series data, linear interpolation or cubic spline interpolation based on spline curves is used. For periodic data, predictive imputation is performed by combining historical trends and neighboring data points. The system records the imputation operation and its confidence level for subsequent source tracing.

[0019] Outlier removal: Statistical methods (such as box plot detection based on IQR, Z-score) and machine learning methods (such as Isolation Forest or One-Class SVM) are used to detect outliers in the data. Detected outliers are marked, isolated, or corrected according to preset strategies to prevent them from negatively impacting subsequent analysis models. For example, when a temperature sensor reading suddenly exceeds the physically possible range, the value will be marked as an outlier and removed, and the system will trigger an alarm.

[0020] Semantic Alignment: The module incorporates a meteorological ontology knowledge base. This base includes not only the coding standards for meteorological elements defined in World Meteorological Organization (WMO) Publication No. 306 (e.g., reflectivity factor "Z", radial velocity "V", spectral width "W"), but also radar equipment component classification systems (e.g., magnetrons, klystrons, feeders, radomes), and project-phase definition specifications. By combining a rule engine with deep learning (e.g., a semantic mapping model based on the Transformer architecture), heterogeneous text descriptions, structured fields, and ontology knowledge are matched to achieve conceptual unification. For example, different descriptions of "Klystron failure" in equipment logs from different manufacturers are uniformly mapped to the standard ontology concept of "klystron failure," achieving a semantic alignment accuracy of no less than 95%.

[0021] After standardization and fusion, the original data is transformed into a standardized dataset with a unified format, semantics, and spatiotemporal benchmark. This dataset contains clear metadata labels and quality assessment information.

[0022] Subsequently, the full lifecycle data warehouse module is responsible for classifying and storing standardized datasets and building multi-dimensional association indexes between project entities and data objects. This module adopts a hierarchical storage architecture to optimize data access performance and storage costs. Hot data storage: For data with extremely high real-time requirements, such as real-time baseline data of the current radar scan, the latest environmental sensor readings, and system alarm information, a distributed in-memory database (such as an Apache Ignite cluster) is used. This data resides directly in memory, providing millisecond-level access response time (less than 100 milliseconds), supporting high-concurrency read and write operations, and is used to support real-time monitoring and rapid response applications.

[0023] Warm data storage: Frequently accessed but not requiring ultra-real-time batch data, such as historical sensor data, daily or hourly radar product summaries, engineering logs, and maintenance work order records, are stored in high-performance columnar storage systems (e.g., Apache Parquet format files stored on HDFS or object storage services). Columnar storage optimizes analytical query performance, making it particularly suitable for OLAP (Online Analytical Processing) scenarios.

[0024] Cold data archiving: Long-term archived data, such as raw radar I / Q data, historical design documents, annual operation and maintenance reports, and large datasets that have exceeded their expiration date, are archived to cost-effective object storage services (such as AWS S3 or MinIO). This data is accessed infrequently but needs to be stored long-term to meet regulatory compliance or later traceability requirements.

[0025] The data warehouse is logically divided according to the entire lifecycle stages of meteorological engineering projects, including: Planning and project initiation stage: Store project proposals, feasibility study reports, preliminary design plans, and budget documents.

[0026] Design and construction phase: Store detailed design drawings (CAD / BIM model data), equipment and material lists (BOM), construction logs, quality inspection reports, and final acceptance documents.

[0027] Operation monitoring phase: Store real-time radar observation data, station environment and equipment operation status data, system performance indicators, and output data products.

[0028] Maintenance and update phase: Store fault reports, repair records, spare parts replacement information, software upgrade logs, and performance tuning reports.

[0029] Decommissioning assessment phase: storage device life assessment report, environmental impact assessment report, dismantling plan.

[0030] To enable rapid data retrieval and cross-stage traceability, the system constructs an index linking project entities and data objects based on a graph database (such as Neo4j). Using each meteorological radar project ID as the root node, it dynamically associates all data objects generated throughout the project's lifecycle. For example, a "Radar Station Project" node can be associated with the "Klystron_SN001" component node, the "MaintenanceTask_ID123" maintenance task node, and the "FaultEvent_ID456" fault event node. Nodes are connected through various relationship types such as "HAS_COMPONENT," "PERFORMED_ON," and "RELATED_TO_FAULT," supporting complex multi-level depth (up to 10 levels of query path) association queries starting from the project ID. This allows for quick retrieval of all maintenance records, associated fault events, and their impact on radar performance for a specific radar component within a specific lifecycle stage, significantly improving data retrieval efficiency and contextual integrity.

[0031] Building upon this foundation, the intelligent analysis and modeling module, based on standardized datasets, invokes or trains corresponding machine learning or physics-driven models to perform tasks such as state prediction, fault diagnosis, performance evaluation, or resource optimization, tailored to the business needs of different engineering stages. This module integrates various advanced algorithm models: Equipment Failure Prediction: For failure prediction of core components of the radar station (such as the klystron, transmitter, and antenna rotation mechanism), a Long Short-Term Memory (LSTM) time series model is employed. This model uses the time series of historical operating parameters of the equipment (such as output power, operating voltage, temperature, vibration frequency, and current fluctuations) as input features. For example, for the klystron, the input feature vector... It includes indicators such as power, voltage, and current over a continuous 48-hour period. The LSTM model can capture time dependencies and output the probability of a failure occurring within the next 24 or 72 hours. The model training adopts an online learning mechanism, and is incrementally updated every 24 hours based on new operating data and maintenance records, ensuring that the model adapts to equipment aging and environmental changes, and maintaining a prediction accuracy of over 90%.

[0032] Spatial interpolation of meteorological elements: Kriging is used to spatially interpolate data from sparsely distributed ground meteorological observation points around the radar station, generating high-resolution meteorological element fields (such as temperature, humidity, and wind field) to assist in radar data quality control and environmental impact assessment. This method considers spatial autocorrelation and provides an error estimate for the interpolation results.

[0033] Construction schedule optimization: During the radar station infrastructure construction and large equipment installation phases, a constraint satisfaction problem (CSP) solver is used to optimize the construction plan. This solver takes construction tasks, resources (manpower, equipment), time windows, and task dependencies as inputs to find the optimal or near-optimal schedule that satisfies all constraints, thereby shortening the construction period or reducing costs.

[0034] Energy efficiency assessment and optimization: Based on real-time energy consumption data of the radar station, equipment operating modes, and environmental parameters, a multi-objective optimization model is constructed. This model aims to minimize total energy consumption and maximize radar data quality by adjusting radar scanning strategies (such as PRT, scanning angle, and pulse width) and cooling system operating parameters to find the optimal operating mode.

[0035] The module supports model version management and A / B testing, ensuring that new models undergo rigorous verification before going live.

[0036] Subsequently, the dynamic decision support module generates visual reports, risk warnings, and operation and maintenance suggestions based on the output of the intelligent analysis and modeling module.

[0037] Visualized Reports: The system generates multi-terminal adaptive rendering visualization dashboards covering the radar station's real-time operating status (such as scanning mode and data output), core equipment health indicators (such as klystron life prediction), environmental monitoring data, historical fault trends, maintenance plans, and spare parts inventory. Managers can view these dashboards anytime via PC, tablet, or mobile phone to quickly understand the overall situation of the radar station.

[0038] Risk warning: A three-level threshold mechanism (low, medium, and high) is used for risk warning. The warning triggering conditions are dynamically adjusted based on the historical statistical distribution of data (e.g., the 95% confidence interval of a certain parameter) and the real-time model output (e.g., the failure probability predicted by LSTM).

[0039] Low-level warning: For example, a reading from an auxiliary sensor deviates slightly from the normal range, but does not affect core functions. The system automatically records this and triggers routine checks.

[0040] Intermediate warning: For example, if the klystron output power exhibits continuous fluctuations and the LSTM model predicts a failure probability exceeding 50% within the next 72 hours, the system will immediately notify maintenance personnel via email and SMS and recommend scheduling preventative maintenance.

[0041] Advanced warning: For example, if the radar antenna rotation mechanism vibrates violently and real-time vibration data exceeds the safety threshold, the intelligent analysis module determines that it may cause structural damage. The system triggers the highest priority alarm and recommends immediate shutdown for inspection to prevent further equipment damage or safety accidents.

[0042] Operations and Maintenance Recommendations: Operations and maintenance recommendations are generated using a hybrid strategy based on rules and case-based reasoning (CBR). When a specific fault is detected or a potential problem is predicted, the system first provides preliminary recommendations based on a predefined rule base (e.g., "Klystron power decreases and temperature increases => check coolant circulation"). If the rule base fails to provide a clear solution, the case-based reasoning module retrieves the most similar resolved issues from the historical fault case database and provides corresponding solutions and operational steps. For example, for a novel vibration anomaly, the system searches for similar vibration cases in motor-type equipment and provides possible diagnostic directions. In practice, the adoption rate of recommendations generated by this strategy is no less than 85%.

[0043] Finally, the system collaborative control module sends the decision instructions generated by the dynamic decision support module to the relevant execution units or business systems to achieve data-driven closed-loop management.

[0044] Command issuance: Connects with external systems via dual channels, including RESTful API and MQTT protocol.

[0045] RESTful API: Used for command interaction with upper-level management systems (such as SCADA, CMMS, ERP). For example, when the system generates an operation and maintenance suggestion to "replace Klystron", a maintenance work order is automatically created in the CMMS via the API and assigned to the corresponding maintenance team.

[0046] MQTT protocol: Used for lightweight, low-latency command transmission with field control terminals or automation equipment. For example, when the system detects that the temperature in the computer room is too high but has not reached the advanced warning level, it can directly adjust the operating power of the air conditioner or fan via MQTT commands.

[0047] The success rate of command issuance has been rigorously verified and is no less than 99.9%.

[0048] Status Feedback and Closed-Loop Management: All issued instructions and their execution status (such as "Work order created," "Equipment adjusted," "Maintenance completed") are fed back to the full lifecycle data warehouse in real time via standard interfaces. This feedback status information is used to update the equipment status model and maintenance records in the system, and serves as feedback data for the intelligent analysis and modeling modules, used for model retraining and optimization. For example, after maintenance personnel complete maintenance, they close the work order in the CMMS; this status is simultaneously fed back, updating the equipment's historical maintenance records and used to evaluate the effectiveness of decision recommendations. This complete closed-loop operation mechanism ensures automation and intelligence throughout the entire process from data collection, analysis, decision-making to execution. The system supports seamless integration with mainstream engineering management systems such as SCADA (Supervisory Control and Data Acquisition), CMMS (Computerized Maintenance Management System), and ERP (Enterprise Resource Planning), ensuring data flow and collaborative operation between different systems, significantly reducing the frequency of manual intervention and improving overall operation and maintenance efficiency.

[0049] In the data standardization and fusion module, when performing semantic alignment, if a semantic mapping model based on the Transformer architecture is used, its core lies in mapping text descriptions from different sources and with different expressions to predefined meteorological domain ontology concepts. This process typically involves an encoder-decoder structure, which learns semantic relationships in massive amounts of labeled data to transform non-standardized text input into a standardized representation in the ontology knowledge base. For example, for input text... (Representing a non-standardized device description or fault report fragment), the model will embed it into a high-dimensional vector space and perform similarity matching with predefined concept vectors in the ontology knowledge base.

[0050] In the semantic mapping process, a simplified attention mechanism can be expressed as: in, It is a query vector. It is a key vector. It is a value vector. This is the dimension of the key vector. Through this mechanism, the model can focus on the most relevant semantic fragments in the input text and align them with ontology concepts. For example, when the system receives the log information "radar transmission power abnormal", the model can accurately identify the association between "transmission power" and the concept of "radar transmitter output power" in the ontology knowledge base, and map "abnormal" to the ontology category of "fault state", thereby achieving semantic standardization.

[0051] Example 2 This embodiment applies the big data management method and system for the entire lifecycle of meteorological engineering projects of the present invention to the construction, operation, and maintenance of a large-scale wind power plant meteorological engineering project. Unlike Embodiment 1, which mainly focuses on the operation of a single radar station, this embodiment focuses on a complex engineering project that includes a large number of wind turbines, power transmission and transformation equipment, and supporting infrastructure. It particularly emphasizes the progress management and quality control during the construction phase, the energy efficiency optimization during the operation phase, and the technical differences in integration with traditional engineering management methods.

[0052] Firstly, regarding the data acquisition and access module, in addition to conventional meteorological elements (wind speed, wind direction, temperature, humidity) and equipment status (wind turbine operating parameters, bearing temperature, vibration, pitch angle) data, this embodiment particularly strengthens the acquisition and access of heterogeneous data from the construction site: Data from the wind turbine itself and power transmission and transformation equipment: The SCADA system acquires real-time data on the blade angle, generator speed, power output, grid connection status, and operating parameters of transformers and combiner boxes, such as temperature, voltage, and current. This data is transmitted using the OPC UA protocol or Modbus TCP / IP protocol, and the edge computing gateway performs protocol conversion and initial data packet encapsulation.

[0053] Construction progress and quality data: BIM Model Data: During the wind farm construction phase, the system periodically exports information such as project progress, component status, and design changes from the BIM (Building Information Modeling) software, typically in IFC (Industry Foundation Classes) format or through a dedicated API interface for batch import. This data provides the digital skeleton of the project, containing the three-dimensional geometric information and attribute data of components such as wind turbine foundations, towers, nacelles, and blades.

[0054] Construction site sensors include temperature and humidity sensors deployed in the concrete pouring area, strain sensors at steel structure installation points, and GPS positioning and operational status sensors for construction vehicles (cranes, transport vehicles). These IoT sensors transmit data via LoRaWAN or NB-IoT networks to monitor the construction environment and key operational parameters in real time.

[0055] Drone imagery and laser point cloud data: Drones are regularly used to conduct aerial photography of the construction site, acquiring high-resolution orthophotos and 3D laser point cloud data to monitor construction progress, terrain changes, component installation accuracy, and safety hazards. The image data is pre-processed and then uploaded to cloud storage.

[0056] Project management software logs: Log data such as construction task allocation, material entry and exit, and personnel attendance from the Project Management Information System (PMIS), Material Management System (MMS), and Labor Management System (LMS) are accessed via RESTful API or file synchronization.

[0057] Meteorological and environmental data: In addition to the wind measurement tower data inside the wind field, it also accesses wind energy resource assessment data from the regional meteorological bureau, historical extreme weather event records, and topographic data.

[0058] Similar to the asynchronous transmission mechanism in Embodiment 1, this embodiment also uses a message queue (such as RabbitMQ) to buffer and transmit this data, ensuring the stability and real-time performance of data throughput under high concurrency, and controlling the data access latency to within 500 milliseconds.

[0059] Secondly, the data standardization and fusion module performs standardization processing on the above data.

[0060] Timestamp and coordinate system alignment: Similar to Example 1, NTP and WGS-84 standards are used.

[0061] Units and missing value handling: Similar interpolation and outlier removal algorithms are used, but optimized for the characteristics of engineering construction data. For example, for missing values ​​in the BIM model component attributes, it may be necessary to fill them in by combining design specifications or default values.

[0062] Semantic Alignment: The module incorporates a unique ontology knowledge base specific to the wind power industry, including wind turbine component classifications (such as tower segments, blade models, and hubs), construction task codes (such as foundation pouring, tower hoisting, and nacelle installation), and quality acceptance standards (such as concrete strength grade and bolt tightening torque). Semantic mapping is achieved through a combination of a rule engine and a graph neural network (GNN), achieving an accuracy of no less than 95%. For example, semantic association is established between "foundation grouting" in the construction log and "wind turbine foundation construction stage" in the BIM model. The spatiotemporal reference is uniformly based on the WGS-84 coordinate system and UTC time standard, with spatial interpolation errors controlled within 10 meters and time synchronization accuracy reaching the millisecond level.

[0063] Furthermore, the full lifecycle data warehouse module categorizes and stores data according to the entire lifecycle stages of wind power projects, from planning and design to decommissioning, and constructs a multi-dimensional association index between project entities and data objects.

[0064] Planning and design phase: storage wind resource assessment report, environmental impact assessment, site selection plan, and power grid connection plan.

[0065] Design and construction phase: storing BIM models, construction plans (Gantt charts, critical paths), equipment delivery lists, construction logs, quality inspection reports, and bills of quantities. This phase involves a massive amount of data with complex interrelationships.

[0066] Operation monitoring phase: Store SCADA data, meteorological observation data, power grid dispatch data, fault records, and power generation data for each wind turbine.

[0067] Maintenance and update phase: Stores regular maintenance plans, preventive maintenance records, fault repair reports, spare parts management information, and software update logs.

[0068] Decommissioning assessment phase: storage device life assessment, dismantling plan, and environmental restoration plan.

[0069] The data warehouse employs a layered storage architecture. Real-time wind turbine operation data is stored in a distributed in-memory database (such as Redis), while historical SCADA data, BIM component attributes, and construction logs are stored in a columnar storage system (such as Apache Parquet). Raw UAV imagery and historical design documents are archived in object storage. A graph database (such as ArangoDB) constructs an index linking project entities, using the wind farm project ID as the root node and dynamically associating entities such as wind turbine IDs, construction task IDs, equipment supplier IDs, fault event IDs, and personnel IDs. For example, a "wind farm project" node can be associated with multiple "wind turbine" nodes, and each "wind turbine" node is further associated with its sub-component nodes such as "tower section" and "blade assembly," and further associated with "construction tasks," "quality inspection records," and "operational fault events." The query path depth can reach 10 levels, enabling efficient retrieval of complex related information such as "which wind turbine a certain batch of blades from a certain supplier was installed on, its installation quality record, and what faults occurred within a year after installation."

[0070] In addition, the intelligent analysis and modeling module integrates a variety of algorithm models tailored to the characteristics of wind power projects: Construction Schedule Prediction and Optimization: Unlike the CSP solver in Example 1, this example employs a dynamic scheduling optimization model based on reinforcement learning (RL). This model treats construction tasks (such as wind turbine installation and cable laying) as actions of an agent within an environment (weather, resource availability, dependencies). Through algorithms such as Q-learning or Proximal Policy Optimization (PPO), it learns strategies to complete task sequences under different external conditions to minimize construction time and cost. Model inputs include real-time weather forecasts, equipment failure rates, personnel attendance rates, and material supply status. The output is the optimal task scheduling scheme for the next 3-7 days, dynamically responding to uncertainties during construction.

[0071] Wind turbine fault diagnosis and remaining life prediction: In addition to the LSTM time series model, this embodiment also introduces a diagnostic model based on a combination of wavelet transform and convolutional neural network (CNN) to analyze vibration signals of key components such as wind turbine gearboxes and generators, and identify early fault characteristics. Time-frequency features are extracted by wavelet decomposition of the vibration signals and then input into the CNN for pattern recognition. Simultaneously, by combining physical driving models (such as fatigue life models) and machine learning models, the remaining life (RUL) of wind turbine components (such as bearings and blades) is predicted.

[0072] Wind energy resource assessment and power generation forecasting: A Gaussian process regression (GPR) model is used to fuse wind measurement tower data and NWP model output to perform high-precision spatial interpolation of wind speed fields. Combined with wind turbine power curves, the short-term (0-72 hours) and medium-to-long-term (monthly, annual) power generation of wind fields is predicted to optimize grid dispatch and trading strategies.

[0073] Construction quality anomaly detection: Using **computer vision models (such as YOLOv7)**, the construction site images captured by drones are analyzed to automatically detect quality defects such as concrete cracks, exposed rebar, and component installation deviations. The results are then compared with the BIM model to pinpoint the location of the deviations, thereby improving the efficiency of quality inspection.

[0074] The model is trained using an online learning mechanism, with incremental updates based on new data every 24 hours, and the model prediction accuracy is maintained above 90%.

[0075] Furthermore, the dynamic decision support module transforms model analysis results into structured decision information and provides customized visualization and early warning functions.

[0076] Visualized Reports: Generate customized visual dashboards for wind farm owners, project managers, construction supervisors, and operation and maintenance personnel. For example, project managers can view overall project progress and critical path deviations, construction supervisors can view heat maps of construction quality defects in specific areas, and operation and maintenance personnel can monitor the performance of individual wind turbines and receive fault warnings in real time. Visualized reports support multi-terminal adaptive rendering.

[0077] Risk warning: A three-level threshold mechanism (low, medium, and high) is adopted. The warning triggering conditions are based on the output of the RL model (such as the potential delay risk of the scheduling scheme), the RUL prediction value of the wind turbine components, and the construction quality inspection results.

[0078] Low-level warning: For example, a construction task is slightly behind schedule, but still within an acceptable range. The system automatically adjusts the priority of subsequent non-critical tasks.

[0079] Intermediate warning: For example, if the RL model predicts that the pouring of a key wind turbine foundation may be delayed by 50% due to weather conditions over the next three days, the system will notify the project manager via email and suggest that alternative construction teams be deployed in advance or that material supply plans be adjusted.

[0080] Advanced alert: For example, if the RUL (Range Limitless) prediction for the wind turbine main bearing is less than 30 days, or if drone imagery detects severe cracks in the wind turbine blades, the system will immediately trigger the highest level alarm, requiring immediate shutdown for inspection and the development of an emergency repair plan.

[0081] Operation and Maintenance Recommendations: Recommendations are generated using a hybrid strategy of rule-based and case-based reasoning (CBR). For example, when wind turbine blade icing is detected as causing a decrease in power generation, the system combines meteorological data and historical cases to recommend activating the blade de-icing system. When the RL model provides a new scheduling scheme, the system also provides the performance of that scheme under similar historical conditions as a reference. The recommendation adoption rate has been measured to be no less than 85%.

[0082] In addition, the system collaborative control module interfaces with external systems through a dual channel of RESTful API and MQTT protocol to send decision instructions to relevant execution units.

[0083] Integration with SCADA system: Send wind turbine operation mode adjustment commands to the wind farm's SCADA system via API, such as automatically entering standby mode when the wind speed is too high, or adjusting the power generation capacity according to the grid load demand.

[0084] Integration with BIM / PMIS systems: Information on construction schedule deviations and quality defects is transmitted back to the BIM model via API, and the model status is updated in real time; new construction tasks are created or the priority of existing tasks is modified in PMIS automatically.

[0085] Integration with mobile terminals: Push task instructions, safety alerts, and operation guides to the mobile terminals of construction or maintenance personnel via MQTT.

[0086] The system boasts a command issuance success rate of no less than 99.9%, and command execution status (such as "wind turbine has been shut down" and "construction task has been updated") is transmitted back to the data warehouse in real time, forming a complete operational loop. The system supports seamless integration with mainstream engineering management systems such as SCADA, PMIS, MMS, and LMS, enabling digital and intelligent management of wind power projects from design and construction to operation.

[0087] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A full life cycle big data management system for meteorological engineering projects, characterized in that, The application relates to a meteorological engineering project lifecycle management system. The data acquisition and access module is used for acquiring multi-source heterogeneous original data in real time or in batches from meteorological observation equipment, engineering sensors, business information systems and external environment databases; The data standardization and fusion module is used for performing format conversion, semantic alignment, space-time reference unification and quality checking on the multi-source heterogeneous original data, and generating a standardized data set; The full lifecycle data warehouse module is used for storing structured, semi-structured and unstructured data according to meteorological engineering project stages, and establishing an associated index of project entities and data objects; The intelligent analysis and modeling module is used for constructing prediction, diagnosis and optimization models for different engineering stages based on the standardized data set; The dynamic decision support module is used for generating a visual report, a risk warning and an operation and maintenance suggestion according to model output results; The system coordination control module is used for issuing decision instructions to relevant execution units or business systems, and realizing closed-loop management.

2. The full lifecycle big data management system for meteorological engineering projects of claim 1, wherein, The full lifecycle data warehouse module adopts a hierarchical storage architecture, hot data is stored in a distributed memory database, warm data is stored in a columnar storage system, and cold data is archived to an object storage, and the data access response time is less than 100 milliseconds in a hot data scenario; the project entity associated index is constructed based on a graph database, takes a project ID as a root node, dynamically associates all data objects generated in the whole lifecycle of the project, and the query path depth can reach 10 layers.

3. The full lifecycle big data management system for weather engineering projects of claim 1, wherein, The intelligent analysis and modeling module integrates an LSTM time sequence model, a Kriging method, a constraint satisfaction problem solver and a multi-objective optimization model; the LSTM time sequence model takes a time sequence of historical operation parameters of equipment as input, and outputs a future fault probability; the Kriging method is used for meteorological element space interpolation; the constraint satisfaction problem solver is used for construction progress optimization; and the multi-objective optimization model is used for energy efficiency evaluation; model training adopts an online learning mechanism, and is incrementally updated based on new data every 24 hours, and the model prediction accuracy rate is maintained to be higher than 90%.

4. The full lifecycle big data management system for weather engineering projects of claim 1, wherein, The visual report generated by the dynamic decision support module supports multi-terminal adaptive rendering; the risk warning adopts a three-level threshold mechanism, and a warning triggering condition is dynamically adjusted based on historical data statistical distribution and real-time model output; operation and maintenance suggestion generation adopts a hybrid strategy based on rules and case reasoning, and the suggestion adoption rate is not less than 85%.

5. The full lifecycle big data management system for weather engineering projects of claim 1, wherein, The system coordination control module is connected with external systems through a RESTful API and a MQTT protocol double channel, the instruction issuing success rate is not less than 99.9%, instruction execution states are fed back to the data warehouse in real time, a complete operation closed loop is formed, and the system supports seamless integration with SCADA, CMMS and ERP systems.

6. The full lifecycle big data management system for weather engineering projects of claim 1, wherein, The data collection and access module supports multi-source data access of automatic weather stations, sounding radars, satellite remote sensing, Internet of Things sensors, BIM model data, engineering management software logs, and public weather databases, adopts an asynchronous transmission mechanism based on a message queue, and the data access delay is not more than 500 milliseconds; the data standardization and fusion module has a built-in meteorological ontology knowledge base, realizes semantic mapping through a rule engine and a deep learning joint method, the semantic alignment accuracy is not less than 95%, the time and space reference is unified in the WGS-84 coordinate system and the UTC time standard, the spatial interpolation error is controlled within 10 meters, and the time synchronization accuracy reaches the millisecond level.

7. A management method of the full life cycle big data management system for meteorological engineering projects according to any one of claims 1 to 6, characterized in that, The method comprises the following steps: In step S110, raw data covering meteorological elements, equipment status, construction progress, operation and maintenance records, and external environment are collected through multiple types of sensors and information systems deployed on meteorological engineering sites and remote platforms; in step S120, standardization processing is performed on the raw data, including time stamp alignment, coordinate system unification, unit conversion, missing value interpolation, and abnormal value elimination, to form a standardized data set with consistent semantics and space-time reference; in step S130, the standardized data set is classified and stored according to the whole life cycle stages of a meteorological engineering project, including planning and establishment, design and construction, operation and monitoring, maintenance and update, and retirement evaluation, and a multi-dimensional association index between project entities and data objects is constructed; in step S140, based on the standardized data set, corresponding machine learning or physical driven models are called or trained according to different stage business requirements, and state prediction, fault diagnosis, performance evaluation or resource optimization tasks are performed; in step S150, model analysis results are converted into structured decision information, and visual dashboards, risk level early warning and specific operation and maintenance suggestions are generated; in step S160, the decision information is pushed to related business systems or control terminals through a standard interface, triggering automated response or assisting manual decision making, to realize data-driven closed-loop management.

8. The management method of the full lifecycle big data management system for meteorological engineering projects according to claim 7, characterized in that, In step S120, semantic alignment is realized through a built-in meteorological ontology knowledge base, which includes WMO No. 306 meteorological element coding standards, engineering equipment classification systems and project stage definition specifications, and a rule engine and deep learning joint method are used for mapping, with a semantic alignment accuracy of not less than 95%. 9.The management method of the full lifecycle big data management system for meteorological engineering projects according to claim 7, characterized in that, In step S140, for device fault prediction, an LSTM time series model is used, with 48 hours of continuous device operation parameter time series as input and fault probability in the next 24 to 72 hours as output; for meteorological element spatial interpolation, a Kriging method is used; for construction progress optimization, a constraint satisfaction problem solver is used; and for energy efficiency evaluation, a multi-objective optimization model is used. 10.The management method of the full lifecycle big data management system for meteorological engineering projects according to claim 7, wherein, In step S150, the three-level threshold mechanism of risk early warning includes: low-level early warning corresponds to slight parameter deviation, triggering routine inspection; medium-level early warning corresponds to fault probability exceeding 50%, notifying maintenance personnel to arrange preventive maintenance; high-level early warning corresponds to safety threshold overrun, suggesting immediate shutdown inspection; and operation and maintenance suggestions are generated through a combination of rule base matching and historical case retrieval.

Images