CN-122018962-A - Digital drainage twin model full life cycle management system

CN122018962ACN 122018962 ACN122018962 ACN 122018962ACN-122018962-A

Abstract

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

Inventors

WANG HAO
ZHANG DEQUAN
WANG HAO

Assignees

上海中井汉鼎数字技术有限公司

Dates

Publication Date: 20260512
Application Date: 20260209

Claims (10)

1. A digital drainage twinning model full life cycle 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 and perform online configuration and storage of the model; The operation monitoring module is configured to collect and monitor the operation state data of the model in real time; A calibration optimization module configured to automatically adjust model parameters based on a deviation of the measured data from the simulated data; a version control module configured to record model version change records and control version switching, and And the retirement archiving module is configured to archive the model meeting the retirement condition.
2. The system of claim 1, wherein the metadata comprises: basic information class including model unique identification ID, name, version number, model type, development team, creation time; technical parameter classes, including an algorithm framework, an input parameter list, an output parameter list and an operating environment; The application attribute class comprises an application area, an application scene, a precision index and an expiration date; the association information class comprises an association data source, an association model and a soft-copy binding number.
3. The system of claim 1, wherein the rating optimization module is configured to: encoding model parameters into chromosomes to construct an initial population; Constructing an adaptability function by taking the root mean square error minimization of the measured data and the simulated data as a target; Generating a next generation population through selection, crossing and mutation operations; Stopping iteration when the iteration times reach a preset threshold or the fitness function value converges, and outputting an optimal parameter combination; based on different initial populations or parameter ranges, 3-5 groups of optimization schemes are generated and quantitatively evaluated.
4. The system of claim 1, wherein the initial population size N is 50 to 100; The crossover probability Pc is 0.6 to 0.8, and the variation probability Pm is 0.01 to 0.05; the preset threshold is 100 to 200 generations, and the fitness function value is converged to be less than or equal to +/-5 cm.
5. The system of claim 1, wherein the version control module is configured to perform one or more of the following: Each time the version is updated, automatically recording the change content, change reason, operator and operation time, and generating a version change log; Retrieving the historical version, backtracking to the appointed version by one key and running, and recording the version switching track; And automatically detecting the compatibility between the new version and the old version, and marking.
6. The system of claim 1, wherein the rating optimization module implements automated rating of model parameters by: ; Wherein the method comprises the steps of ; X is the vector of the parameters of the model, K is the number of parameters; n is the number of actual measurement data samples; as an i-th actual measurement value, the value of the current value, Is the i-th analog value; For the fitness value, the value range is 0 to 1, and the closer to 1, the higher the accuracy is.
7. The system of claim 1, wherein the retirement archive module calculates a model accuracy decay rate by: ; Wherein the method comprises the steps of In order to achieve a precise decay rate, As the root mean square error of the current model, For the root mean square error at initial registration of the model, when The retirement condition is satisfied.
8. A digital drainage twinning model full life cycle management method of the system of claims 1 to 7, comprising: Performing system deployment and environment configuration; Performing model standardized registration through a model registration module; Model configuration and operation are carried out through a configuration management module; Automatically calibrating and optimizing the mechanical energy model by a calibrating and optimizing module; version control and tracing are carried out through a version control module; performing end cloud collaborative iteration, and Model retirement and archiving are performed through a retirement archiving module.
9. The method of claim 8, wherein the performing end-cloud collaborative iterations comprises: The edge node collects monitoring data and the running state of the model in real time, encrypts and uploads the monitoring data and the running state of the model to the cloud through an MQTT protocol; After receiving the data, the cloud node performs model incremental training, optimizes parameters and then transmits updated parameters to the edge node; the edge node automatically updates the model configuration after receiving the parameters, so as to realize collaborative iteration; the system dynamically adjusts the operation node and optimizes the resource allocation.
10. The method of claim 8, wherein the end cloud collaborative data synchronization delay is calculated by the formula: ; T is total synchronous delay, T trans is data transmission delay, T proc is data processing delay, and T is less than or equal to 1s through end cloud cooperative iteration.

Description

Digital drainage twin model full life cycle management system Technical Field The invention relates to the technical field of digital twinning, model management and urban drainage pipe network operation and maintenance, in particular to a digital drainage twinning model full life cycle management system. Background Along with the deep penetration of the digital twin technology in the field of urban drainage pipe networks, various twin models become key technical carriers for supporting core services such as pipe network monitoring, risk early warning, emergency dispatching and the like, and the full release of the application value of the digital twin model directly influences the intelligent management and control level of urban drainage. However, the management system of the current digital drainage twin model still has a plurality of outstanding technical bottlenecks and industry pain points, so that the efficient utilization of model resources and service supporting capability are severely restricted, and the management system is specifically characterized in the following five aspects: Firstly, the management flow is fragmented, and the model multiplexing rate is extremely low. In the prior art, key management links such as model registration, parameter configuration, operation monitoring, rating optimization, retirement filing and the like lack a unified integrated management platform and standardized operation flow, and each link is in a splitting state, so that the model resources are stored in a scattered mode and are managed unordered. According to statistics, the model multiplexing rate in the traditional management mode is generally less than 30%, a large number of repeated modeling works cause serious waste of manpower and computational resources, meanwhile, data flow of each link is unsmooth, an 'information island' is formed, and the overall efficiency of model management is further reduced. Secondly, the version management mechanism is missing, and the traceability is insufficient. Because a normalized version control system is not established, the model versions corresponding to different development stages and different application scenes coexist in a mixed mode, and key information such as version number rules, parameter change records, precision iteration processes and the like are not reserved by the system. The model version of the adaptive target scene cannot be accurately identified by a model user, when the model has abnormal accuracy, the parameter change track and the iteration process are difficult to trace, the problem root cannot be positioned, and the reliability and the stability of the model application are seriously affected. Thirdly, the calibrating process depends on manual work, and the efficiency and the precision are low. The existing model calibration link is highly dependent on manual operation of technicians, tens of model parameters are required to be manually adjusted, simulation data and actual measurement data are repeatedly compared, the operation is complex, the time consumption is long (the single model calibration period is usually more than 72 hours), the calibration error is large (generally more than or equal to +/-8 cm) due to the influence of human experience difference, and the severe requirements of real-time monitoring and dynamic early warning of urban drainage pipe network on model precision and updating efficiency are difficult to meet. Fourth, the end cloud cooperative adaptation is insufficient, and the iteration mechanism is absent. The existing management system is based on a single architecture design, is not effectively adapted to a Hongshan Meng Duanyun cooperative architecture, and cannot realize the model cooperative management of an edge node (on-site monitoring end) and a cloud node (computing center). Real-time operation data of the edge end model are difficult to quickly synchronize to the cloud end, parameters after cloud training optimization cannot be issued to the edge end in time, so that iteration of the model is delayed, and dynamic changes of the actual operation state of the pipeline cannot be dynamically adapted. Fifthly, the technical scheme is not covered fully, and a unified inlet is lacked. The prior art focuses on a single link or part of functions of model management, does not form a complete management scheme covering the whole life cycle of registration, configuration, operation, calibration, optimization and retirement, lacks a standardized flow system and tool chain support, and simultaneously, the AI model, the hydraulic model and other various model resources are subjected to decentralized management, and lacks a unified management entrance and cooperation mechanism, so that the model integration difficulty is high, the iteration efficiency is low, and the actual floor property of the technical scheme is poor. In summary, the prior art cannot effectively solve the core technical bottlenecks of "nonstandard flow,