Digital drainage twinborn model full life cycle management system

Abstract

The invention discloses a digital drainage twin model full life cycle management system comprising a model registration module configured to establish a unified model metadata specification and perform standardized registration of multiple types of models; the configuration management module is configured to provide a visual parameter configuration interface and perform online configuration and storage of the model; the operation monitoring module is configured to acquire operation state data of the monitoring model in real time; the calibration optimization module is configured to automatically adjust model parameters based on the deviation between the measured data and the simulated data; the version control module is configured to record a model version change record and control version switching; and the decommissioning archiving module is configured to perform archiving processing on the models meeting the decommissioning conditions. The method is suitable for resource management and control and efficient application of the multi-type twinborn model in the scenes of pipe network monitoring, risk early warning, emergency scheduling and the like.

Description

Technical Field

[0001] This invention relates to the fields of digital twin, model management, and urban drainage network operation and maintenance technology, and in particular to a digital drainage twin model full life cycle management system. Background Technology

[0002] With the deep penetration of digital twin technology into urban drainage pipe networks, various twin models have become key technological carriers supporting core businesses such as pipe network monitoring, risk early warning, and emergency dispatch. The full release of their application value directly affects the level of intelligent management and control of urban drainage. However, the current management system of digital drainage twin models still has many prominent technical bottlenecks and industry pain points, which seriously restrict the efficient utilization of model resources and business support capabilities. Specifically, these are reflected in the following five aspects: First, the management process is fragmented, resulting in extremely low model reuse rates. Current technologies lack a unified, integrated management platform and standardized operating procedures for key management aspects such as model registration, parameter configuration, operational monitoring, calibration and optimization, and decommissioning archiving. These processes are disconnected, leading to scattered model resource storage and disorganized management. Statistics show that under traditional management models, model reuse rates are generally less than 30%, resulting in a significant waste of human and computing resources due to repetitive modeling work. Simultaneously, poor data flow between processes creates "information silos," further reducing the overall efficiency of model management.

[0003] Secondly, the lack of a version management mechanism and insufficient traceability capabilities are problematic. Due to the absence of a standardized version control system, model versions from different development stages and application scenarios coexist, and crucial information such as version numbering rules, parameter change records, and accuracy iteration history are not retained by the system. This makes it impossible for model users to accurately identify the model version suitable for their target scenario, and when model accuracy anomalies occur, it is difficult to trace parameter changes and iteration processes, making it impossible to pinpoint the root cause of the problem and severely impacting the reliability and stability of model applications.

[0004] Third, the calibration process relies heavily on manual labor, resulting in low efficiency and low accuracy. The current model calibration process is highly dependent on manual operation by technicians, requiring manual adjustment of dozens of model parameters and repeated comparison of simulated and measured data. This is not only cumbersome and time-consuming (the calibration cycle for a single model usually exceeds 72 hours), but also susceptible to large calibration errors (generally ≥ ±8cm) due to differences in human experience. This makes it difficult to meet the stringent requirements of real-time monitoring and dynamic early warning of urban drainage pipe networks for model accuracy and update efficiency.

[0005] Fourth, there is insufficient adaptation to edge-cloud collaboration and a lack of iteration mechanisms. Existing management systems are mostly based on a single architecture design and have not been effectively adapted to the HarmonyOS edge-cloud collaborative architecture, making it impossible to achieve collaborative model management between edge nodes (on-site monitoring terminals) and cloud nodes (computing centers). Real-time operational data from edge models is difficult to synchronize quickly to the cloud, and parameters optimized after cloud training cannot be promptly distributed to the edge, resulting in delayed model iteration and an inability to dynamically adapt to the dynamic changes in the actual operating status of the pipeline network.

[0006] Fifth, the technical solutions are incomplete and lack a unified entry point. Existing technologies mostly focus on a single link or part of the model management process, failing to form a complete management solution covering the entire lifecycle of "registration-configuration-running-calibration-optimization-retirement," and lacking standardized process systems and toolchain support. At the same time, various model resources such as AI models and hydraulic models are managed in a decentralized manner, lacking a unified management entry point and collaborative mechanism, resulting in high difficulty in model integration, low iteration efficiency, and poor practical implementation of technical solutions.

[0007] In summary, existing technologies cannot effectively solve the core technical bottlenecks in the management of digital drainage twin models, such as "non-standard processes, untraceable versions, inefficient calibration, asynchronous collaboration, and unsmooth implementation." There is an urgent need for a standardized management system and method that covers the entire lifecycle, adapts to the end-to-cloud collaborative architecture, and supports automated iteration, in order to fill the industry's technical gap and improve the management efficiency and application value of digital drainage twin models. Summary of the Invention

[0008] To address the aforementioned shortcomings of existing technologies, the core objective of this invention is to provide a digital drainage twin model full lifecycle management system and method, specifically including: 1. Construct a full lifecycle management platform encompassing "registration-configuration-running-calibration-optimization-retirement," establish standardized management processes and operating procedures, achieve integrated and standardized management of model resources, and significantly improve model reuse rate; 2. Develop functions for automatic model calibration, intelligent version control, and multi-scheme comparison to replace traditional manual operations, shorten the model iteration cycle, reduce calibration errors, and improve model management efficiency and accuracy; 3. Achieve deep adaptation with HarmonyOS edge-cloud collaborative architecture, establish a collaborative iteration mechanism for edge-cloud models, ensure real-time synchronization of model parameters and efficient version updates, and dynamically adapt to the network operation status.

[0009] This invention provides a digital drainage twin model full lifecycle management system, comprising: The model registration module is configured to establish a unified model metadata specification and perform standardized registration of multiple types of models. The configuration management module is configured to provide a visual parameter configuration interface for online configuration and saving of models; The monitoring module is configured to collect and monitor the running status data of the monitoring model in real time. The calibration and optimization module is configured to automatically adjust model parameters based on the deviation between measured and simulated data. The version control module is configured to record model version change history and control version switching; and The retirement archiving module is configured to archive models that meet the retirement criteria.

[0010] In one embodiment of the present invention, the metadata includes: Basic information includes the model's unique identifier ID, name, version number, model type, development team, and creation time; Technical parameters include the algorithm framework, input parameter list, output parameter list, and runtime environment; Application attribute classes include applicable region, applicable scenario, accuracy index, and validity period; Related information includes related data sources, related models, and software copyright binding numbers.

[0011] In one embodiment of the present invention, the calibration and optimization module is configured to perform the following operations: The model parameters are encoded into chromosomes to construct the initial population; A fitness function is constructed with the goal of minimizing the root mean square error between measured and simulated data. The next generation of the population is generated through selection, crossover, and mutation operations; When the number of iterations reaches the preset threshold or the fitness function value converges, the iteration stops and the optimal parameter combination is output. Based on different initial populations or parameter ranges, 3-5 sets of optimization schemes are generated and quantitatively evaluated.

[0012] In one embodiment of the present invention, the initial population size N is 50 to 100; The crossover probability Pc is 0.6 to 0.8, and the mutation probability Pm is 0.01 to 0.05. The preset threshold is 100 to 200 generations, and the fitness function value converges to an error of ≤ ±5cm.

[0013] In one embodiment of the present invention, the version control module is configured to perform one or more of the following operations: Each time a version is updated, the changes, reasons for the changes, the person who performed the change, and the time of the change are automatically recorded, generating a version change log. Search historical versions, revert to a specified version with one click and run it, while recording the version switching trajectory; Automatically detect and mark compatibility between new and old versions.

[0014] In one embodiment of the present invention, the calibration and optimization module achieves automated calibration of model parameters through the following formula: ; in ; x is the model parameter vector. k is the number of parameters; n is the number of measured data samples; For the i-th measured value, This is the i-th simulated value; The fitness value ranges from 0 to 1, with values ​​closer to 1 indicating higher precision.

[0015] In one embodiment of the present invention, the decommissioning and archiving module calculates the model accuracy attenuation rate using the following formula: ; in For accuracy attenuation rate, This represents the root mean square error of the current model. The root mean square error during the initial registration of the model, when They must meet the retirement requirements at that time.

[0016] This invention also provides a method for full lifecycle management of a digital drainage twin model based on the above system, comprising: Perform system deployment and environment configuration; Standardized model registration is performed through the model registration module; The model is configured and run through the configuration management module; The mechanical energy model is automatically calibrated and optimized through the calibration and optimization module. Version control and traceability are performed through the version control module; Perform edge-cloud collaborative iteration; and Model retirement and archiving are performed through the retirement and archiving module.

[0017] In one embodiment of the present invention, the step of performing edge-cloud collaborative iteration includes: Edge nodes collect and monitor data and model operation status in real time, and upload them to the cloud in encrypted form via the MQTT protocol; After receiving the data, the cloud node performs incremental training of the model, optimizes the parameters, and then sends the updated parameters to the edge nodes. Edge nodes automatically update model configurations after receiving parameters, enabling collaborative iteration; The system dynamically adjusts running nodes and optimizes resource allocation.

[0018] In one embodiment of the present invention, the data synchronization latency for edge-cloud collaboration is calculated using the following formula: ; Where T is the total synchronization delay, T trans For data transmission delay, T proc To reduce data processing latency, edge-cloud collaborative iteration ensures that T≤1s.

[0019] The present invention has the following beneficial effects: (1) A full life cycle management standard for digital drainage twin models is proposed, covering six core links: registration, configuration, operation, calibration, optimization and decommissioning. Standardized processes and metadata specifications are established, and the model reuse rate is increased from less than 30% to more than 80%, significantly reducing the cost of repetitive modeling. (2) An improved genetic algorithm is used to realize automatic model calibration, replacing the traditional manual calibration method. The calibration efficiency is increased by 5 times (the calibration cycle of a single model is shortened to within 12 hours), the calibration error is reduced to ≤±5cm, and the accuracy is improved by 40% compared with manual calibration. At the same time, it supports the comparison of multiple schemes and improves the scientific nature of model optimization. (3) Achieve deep adaptation with HarmonyOS edge-cloud collaborative architecture, build a two-way collaborative iteration mechanism for edge-cloud models, with data synchronization latency ≤1s and parameter update time ≤3s, solve the problem of model iteration lag under traditional single architecture, and realize dynamic matching between model and actual pipeline network operation status. Attached Figure Description

[0020] Figure 1 A block diagram of a digital drainage twin model full lifecycle management system according to an embodiment of the present invention is shown; Figure 2 A flowchart of a digital drainage twin model full lifecycle management method according to an embodiment of the present invention is shown. Detailed Implementation

[0021] In the following description, the invention is described with reference to various embodiments. However, those skilled in the art will recognize that the embodiments may be practiced without one or more specific details or with other alternatives and / or additional methods, materials, or components. In other instances, well-known structures, materials, or operations are not shown or described in detail so as not to obscure the inventive points of the invention. Similarly, for illustrative purposes, specific quantities, materials, and configurations are set forth to provide a comprehensive understanding of embodiments of the invention. However, the invention is not limited to these specific details.

[0022] In this invention, the various embodiments are merely intended to illustrate the solutions of the invention and should not be construed as limiting.

[0023] In this specification, references to "an embodiment" or "this embodiment" mean that a particular feature, structure, or characteristic described in connection with that embodiment is included in at least one embodiment of the invention. The phrase "in one embodiment" appearing throughout this specification does not necessarily refer to the same embodiment in all instances.

[0024] Furthermore, the numbering of the steps in the methods of the present invention does not limit the execution order of the method steps. Unless otherwise specified, the method steps may be executed in different orders.

[0025] The present invention will be further described below with reference to the accompanying drawings and specific embodiments.

[0026] Figure 1 A block diagram of a digital drainage twin model full lifecycle management system according to an embodiment of the present invention is shown.

[0027] In this embodiment, the hardware environment is shown in the table below: In this embodiment, the software environment is as follows: Operating systems: HarmonyOS 6.0 Industrial Edition (management server), HarmonyOS 6.0 Embedded Edition (edge ​​node), HarmonyOS 6.0 Server Edition (cloud node); Dependency tools: PostgreSQL 14 + PostGIS 3.3: Responsible for the storage and management of model metadata and spatial data; Git 2.40: Provides distributed control capabilities for model versions; MindSpore 3.0: Supports cloud-based model training and incremental iteration; Docker 24.0: Enables containerized deployment and environment isolation of system modules.

[0028] Core libraries: Scikit-learn 1.2: Used for the development of genetic algorithms and automatic calibration of model parameters; Flask 2.3: A web-based visual interactive interface for building a full lifecycle management platform; MQTT-C1.3.10: Enables low-latency data communication between edge nodes and cloud nodes; GitPython 3.1.31: Provides a code-level API for version control. NumPy 1.24: Supports efficient numerical computation during model calculation; Pandas 2.0: Completes the cleaning, processing, and analysis of monitoring and simulation data.

[0029] like Figure 1 As shown, in this embodiment, the digital drainage twin model full lifecycle management system includes: The model registration module 100 is used to establish a unified model metadata specification, supporting the standardized registration of various types of models such as AI models (rainfall prediction models, urban flooding identification models), hydraulic models (SWMM, MIKE models), and hydrological models. It records core metadata such as model name, version number, runtime environment dependencies, input and output field definitions, accuracy indicators, and applicable scenarios.

[0030] The metadata specifications for the digital drainage twin model clarify the metadata classification and field definitions: Basic information includes the model's unique identifier ID, name, version number, model type, development team, and creation time; Technical parameters include the algorithm framework, input parameter list (name, data type, value range), output parameter list (name, data type, precision requirements), and operating environment (hardware configuration, software dependencies, computing power requirements). Application attributes include applicable area, applicable scenario (daily monitoring, rainstorm warning, emergency dispatch), accuracy indicators (historical simulation error, confidence interval), and validity period; Related information includes related data sources (monitoring sites, database tables), related models (upstream and downstream dependency models), and software copyright binding numbers.

[0031] Standardized registration enables unified management and rapid retrieval of model information, providing a data foundation for subsequent configuration, operation, and calibration.

[0032] The configuration management module 200 provides a visual parameter configuration interface, supports online configuration and saving of model running parameters, boundary conditions, and data interfaces, binds configuration information to model versions, and supports one-click reuse of historical configuration schemes.

[0033] The 300-level monitoring module is used to collect real-time model running status data (CPU / memory usage, runtime, data processing volume), monitor the rationality and stability of the model output results, and automatically trigger alarms and record logs when running abnormalities occur or results exceed thresholds.

[0034] The calibration and optimization module 400 integrates an automatic calibration algorithm and multi-scheme comparison functionality. Based on the deviation between measured and simulated data, it automatically adjusts model parameters, generates multiple sets of optimized schemes, and quantifies and evaluates their accuracy. It supports one-click application of the optimal scheme. An improved genetic algorithm is used to automate the calibration and optimization of model parameters. The core process is as follows: Step 1: Parameter encoding. Encode key model parameters (such as roughness, permeability coefficient, and convergence time) into chromosomes and construct an initial population (population size N=50-100). Step 2: Define the fitness function. Construct the fitness function with the goal of minimizing the root mean square error (RMSE) between the measured data and the simulated data. Step 3: Genetic operations, through selection, crossover (crossover probability P_c = 0.6-0.8), and mutation (mutation probability P_m = 0.01-0.05), to generate the next generation population; Step 4: Iteration termination. When the number of iterations reaches the preset threshold (100-200 generations) or the fitness function value converges (error ≤ ±5cm), the iteration stops and the optimal parameter combination is output. Step 5: Compare multiple schemes. Based on different initial populations or parameter ranges, generate 3-5 sets of optimization schemes and conduct quantitative evaluation from dimensions such as accuracy, operating efficiency, and stability. Support users to select or the system to automatically recommend the optimal scheme.

[0035] The calibration and optimization module 400 automates the calibration of model parameters using the following formula: ; in ; x is the model parameter vector. k is the number of parameters; n is the number of measured data samples; For the i-th measured value, This is the i-th simulated value; The fitness value ranges from 0 to 1, with values ​​closer to 1 indicating higher precision.

[0036] The version control module 500 employs a distributed version management mechanism to record model version change records (parameter modifications, algorithm optimizations, accuracy iterations), version creator, creation time, compatibility information, etc., supporting multiple versions coexisting, historical version rollback, and version switching. The model version management mechanism is built based on Git distributed version control technology. The version numbering rule adopts the format of "major version number.minor version number.revision number" (e.g., V1.2.3). The major version number corresponds to major feature iterations, the minor version number corresponds to parameter optimizations or local improvements, and the revision number corresponds to bug fixes. Each time a version is updated, the changes (parameter modification details, algorithm adjustment, new function addition), reasons for the changes, the person who performed the change, and the time of the change are automatically recorded, and a version change log is generated. It supports searching historical versions by version number, change time, operator, etc., and can revert to a specified version with one click and run it. At the same time, it records the version switching trajectory to ensure that the model application is traceable. Automatically detect compatibility between new and old versions (parameter compatibility, interface compatibility, runtime environment compatibility) and clearly mark it in the version description to avoid runtime abnormalities caused by version switching.

[0037] The retirement archiving module 600 establishes model retirement evaluation standards (such as accuracy decay rate ≥20% or disappearance of applicable scenarios), archives models that meet the retirement conditions, retains complete model files, running logs and iteration records, and supports subsequent traceability and query.

[0038] The retirement archiving module 600 calculates the model accuracy attenuation rate using the following formula: ; in For accuracy attenuation rate, This represents the root mean square error of the current model. The root mean square error during the initial registration of the model, when They must meet the retirement requirements at that time.

[0039] In this embodiment, the digital drainage twin model full lifecycle management system is also adapted to the HarmonyOS edge-cloud collaborative architecture to establish a two-way collaborative iteration mechanism between the edge and cloud models. The steps are as follows: Edge nodes collect real-time monitoring data (water level, flow velocity, flow rate) and model operation status data of the pipeline network, and transmit them to the cloud in encrypted form via the MQTT protocol, with a data synchronization latency of ≤1s; The cloud utilizes the computing power of the Ascend 910 cluster to perform incremental training and parameter optimization on the model based on real-time and historical data uploaded from the edge. The cloud will distribute the optimized model parameters or updated model version to the edge nodes through HarmonyOS distributed soft bus technology, supporting batch distribution and incremental updates, with an update time of ≤3 seconds; Based on the pipeline network operation status (such as rainfall and pipeline load), the division of labor between edge and cloud models is dynamically adjusted. In daily monitoring scenarios, the edge nodes run lightweight models independently, while in computing-intensive scenarios such as rainstorm warnings, the cloud undertakes the main computing tasks, and the edge nodes are responsible for data collection and result display, thereby achieving optimized resource allocation.

[0040] In the bidirectional collaborative iteration mechanism, the edge-cloud collaborative data synchronization latency is calculated using the following formula: ; Where T is the total synchronization delay, T trans For data transmission delay, T procTo reduce data processing latency, edge-cloud collaborative iteration ensures that T≤1s.

[0041] Figure 2 A flowchart of a digital drainage twin model full lifecycle management method according to an embodiment of the present invention is shown.

[0042] In another embodiment of the present invention, a method for full lifecycle management of a digital drainage twin model based on the above system is also provided, the steps of which include: S1. System Deployment and Environment Configuration: Install the corresponding versions of the operating system and dependent tools on the management server, edge nodes, and cloud nodes respectively, and configure network communication (ensure edge-cloud network connectivity). Deploy a PostgreSQL+PostGIS database and create database table structures such as model metadata table, version information table, and runtime log table; Deploy the web interface of the full lifecycle management platform and configure the end-to-cloud communication interface (MQTT server address, port, authentication information).

[0043] S2. Model Standardization Registration: Model developers can submit model registration applications through a web interface, fill in metadata information (basic information, technical parameters, application attributes, and association information), and upload model files. The system automatically verifies the integrity and legality of metadata. Once the verification is successful, a unique model ID is assigned, registration is completed, and the metadata is stored in the database. After registration, the system generates standardized model information cards, which can be searched by model type, applicable scenario, accuracy index and other conditions.

[0044] S3. Model Configuration and Operation: Operation and maintenance personnel can use the configuration management module to select the target model and version, and configure the running parameters (such as boundary conditions and calculation step size) and data interface (associating with monitoring data sources). Submit a task to run, and the system will automatically allocate a running node (edge / cloud) based on the task type (daily monitoring / emergency dispatch) and start the model running; The operation monitoring module collects real-time operation status data, monitors CPU / memory usage, operation progress and output results, and triggers alarms when abnormalities occur.

[0045] S4. Automatic Model Calibration and Optimization: The system periodically (or manually triggered by the user) starts an automatic calibration task to read measured data (pipeline monitoring data for the past 7 days) and model simulation data from the database; Initialize the genetic algorithm parameters (population size, number of iterations, crossover / mutation probability), calculate the fitness value, and complete parameter optimization; Three optimal parameter schemes are generated and quantitatively evaluated from the dimensions of accuracy (RMSE), running efficiency (time consumption), and stability (standard deviation of error in multiple runs). The evaluation results are displayed for users to choose from. After user confirmation, the system will bind the optimal parameters to the current model version, generate a new model version, and record the change log.

[0046] S5. Version Control and Tracking: Each time parameters are optimized or functions are improved, the system automatically generates a new version number, creates a new version, and records the changes and reasons. When it is necessary to revert to a previous version, the user selects the target version number through the web interface, and the system loads the historical model file and parameter configuration, supporting direct running or secondary optimization based on the historical version. Before switching versions, the system checks for compatibility. If incompatibility is found, a detailed compatibility report is generated.

[0047] S6, End-to-Cloud Collaborative Iteration: Edge nodes collect and monitor data and model operation status in real time, and upload them to the cloud in encrypted form via the MQTT protocol; After receiving the data, the cloud node performs incremental training of the model, optimizes the parameters, and then sends the updated parameters to the edge nodes. After receiving parameters, the edge nodes automatically update the model configuration without manual intervention, achieving collaborative iteration. The system dynamically adjusts running nodes and optimizes resource allocation.

[0048] S7. Model Retirement and Archiving: The system periodically (quarterly) checks the model status and generates a retirement assessment report when the retirement conditions are met (accuracy decay rate ≥20%, applicable scenarios disappear, or the model is replaced by a new version). After user confirmation, the complete information, including model files, metadata, running logs, and version records, is archived in the database and marked as "retired," supporting subsequent queries and traceability.

[0049] In one embodiment of the present invention, the system performance is verified through multi-dimensional testing, and the test data is as follows: Table 1. Model Management Efficiency Test Data As shown in Table 1, compared with traditional management methods, this invention achieves a leapfrog improvement in core management indicators: The model reuse rate increased significantly from 28% to 83%, an increase of 55 percentage points, completely changing the waste of resources in repetitive modeling; The registration time for a single model has been reduced from 30 minutes (manual entry) to 5 minutes (standardized form + automatic validation), a reduction of 83.3%. The success rate of version tracing has been increased from 35% (relying on manual recording) to 100% (automatic recording + retrieval), solving the version tracing problem; The efficiency of decommissioning and archiving has been significantly optimized, reduced from 2 hours (manual sorting) to 10 minutes (automatic archiving + classified storage), an efficiency improvement of 91.7%.

[0050] Table 2. Automatic calibration performance test data As shown in Table 2, the automated calibration scheme comprehensively surpasses traditional manual calibration: The single-model calibration cycle was shortened from 72 hours to 10 hours, improving efficiency by 86.1% and significantly reducing time costs; The calibration error (RMSE) was reduced from ±8.5cm to ±4.2cm, a reduction of 50.6%, significantly improving the accuracy assurance capability; The new feature allows for the generation of multiple solutions, enabling the output of 3-5 optimized solutions (compared to only 1 manually generated solution in the traditional method), providing a wider range of options. This invention significantly lowers the technical threshold; while traditional methods require professional personnel to operate the entire process, this invention only requires ordinary personnel to confirm the optimal solution.

[0051] Table 3. End-to-Cloud Collaborative Iterative Test Data As shown in Table 3, the test results related to edge-cloud collaboration all meet and exceed the target requirements: Data synchronization (edge ​​to cloud, 100MB of data) has a latency of only 0.7s (target ≤1s), resulting in highly efficient data flow; The parameter delivery (cloud to edge, 10KB parameter file) took 2.3 seconds (target ≤ 3 seconds), and the parameter updates were timely; The collaborative iteration cycle is significantly shortened, and the optimization of the rainstorm scenario model only takes 2 hours, which is 1 / 12 of the traditional method (24 hours); In scenarios involving sudden changes in pipeline traffic, the model achieves an accuracy rate of 98.5% (target ≥ 95%), demonstrating excellent adaptation stability.

[0052] Table 4. System Stability Test Data As shown in Table 4, after 72 hours of continuous operation testing, the stability of the system of this invention fully meets the standards: Platform stability: When managing 100 models concurrently, there are no crashes or deadlocks, and the average CPU utilization is ≤60%; Data storage stability: With a cumulative total of 500 models and 1000 versions registered, there was no data loss, and the query response time was ≤0.5s; Edge-to-cloud communication stability: During continuous data transmission between the edge and the cloud, the communication success rate is 100%, with no data packet loss.

[0053] Although various embodiments of the invention have been described above, it should be understood that they are presented by way of example only and not as limitations. It will be apparent to those skilled in the art that various combinations, modifications, and alterations can be made without departing from the spirit and scope of the invention. Therefore, the breadth and scope of the invention disclosed herein should not be limited by the exemplary embodiments disclosed above, but should be defined solely by the appended claims and their equivalents.

Claims

1. A digital drainage twin model full lifecycle management system, characterized in that, include: The model registration module is configured to establish a unified model metadata specification and perform standardized registration of multiple types of models. The configuration management module is configured to provide a visual parameter configuration interface for online configuration and saving of models; The monitoring module is configured to collect and monitor the running status data of the monitoring model in real time. The calibration and optimization module is configured to automatically adjust model parameters based on the deviation between measured and simulated data. The version control module is configured to record model version change history and control version switching; and The retirement archiving module is configured to archive models that meet the retirement criteria.

2. The system according to claim 1, characterized in that, The metadata includes: Basic information includes the model's unique identifier ID, name, version number, model type, development team, and creation time; Technical parameters include the algorithm framework, input parameter list, output parameter list, and runtime environment; Application attribute classes include applicable region, applicable scenario, accuracy index, and validity period; Related information includes related data sources, related models, and software copyright binding numbers.

3. The system according to claim 1, characterized in that, The calibration and optimization module is configured to perform the following operations: The model parameters are encoded into chromosomes to construct the initial population; A fitness function is constructed with the goal of minimizing the root mean square error between measured and simulated data. The next generation of the population is generated through selection, crossover, and mutation operations; When the number of iterations reaches the preset threshold or the fitness function value converges, the iteration stops and the optimal parameter combination is output. Based on different initial populations or parameter ranges, 3-5 sets of optimization schemes are generated and quantitatively evaluated.

4. The system according to claim 1, characterized in that, The initial population size N is 50 to 100; The crossover probability Pc is 0.6 to 0.8, and the mutation probability Pm is 0.01 to 0.

05. The preset threshold is 100 to 200 generations, and the fitness function value converges to an error of ≤ ±5cm.

5. The system according to claim 1, characterized in that, The version control module is configured to perform one or more of the following operations: Each time a version is updated, the changes, reasons for the changes, the person who performed the change, and the time of the change are automatically recorded, generating a version change log. Search historical versions, revert to a specified version with one click and run it, while recording the version switching trajectory; Automatically detect and mark compatibility between new and old versions.

6. The system according to claim 1, characterized in that, The calibration and optimization module achieves automated calibration of model parameters using the following formula: ； in ; x is the model parameter vector. k is the number of parameters; n is the number of measured data samples; For the i-th measured value, This is the i-th simulated value; The fitness value ranges from 0 to 1, with values ​​closer to 1 indicating higher precision.

7. The system according to claim 1, characterized in that, The decommissioning and archiving module calculates the model accuracy attenuation rate using the following formula: ； in For accuracy attenuation rate, The root mean square error of the current model. The root mean square error during the initial registration of the model, when They must meet the retirement requirements at that time.

8. A method for full lifecycle management of a digital drainage twin model of the system according to claims 1 to 7, characterized in that, include: Perform system deployment and environment configuration; Standardized model registration is performed through the model registration module; The model is configured and run through the configuration management module; The mechanical energy model is automatically calibrated and optimized through the calibration and optimization module. Version control and traceability are performed through the version control module; Perform edge-cloud collaborative iteration; as well as Model retirement and archiving are performed through the retirement and archiving module.

9. The method according to claim 8, characterized in that, The aforementioned end-to-cloud collaborative iteration includes: Edge nodes collect and monitor data and model operation status in real time, and upload them to the cloud in encrypted form via the MQTT protocol; After receiving the data, the cloud node performs incremental training of the model, optimizes the parameters, and then sends the updated parameters to the edge nodes. Edge nodes automatically update model configurations after receiving parameters, enabling collaborative iteration; The system dynamically adjusts running nodes and optimizes resource allocation.

10. The method according to claim 8, characterized in that, The data synchronization latency for edge-cloud collaboration is calculated using the following formula: ； Where T is the total synchronization delay, T trans For data transmission delay, T proc To reduce data processing latency, edge-cloud collaborative iteration ensures that T≤1s.

Images