Off-line virtual laboratory and digital intelligent application iteration method and system based on intelligent agent

Abstract

The invention discloses an off-line virtual laboratory and digital intelligent application iteration method and system based on an intelligent agent. The method comprises the following steps: constructing an off-line virtual laboratory consistent with a site in a non-production environment, and keeping the consistency of data and configuration through a synchronization mechanism; performing automatic scheduling on working condition arrangement, simulation deduction and rule checking by utilizing an intelligent agent; carrying out process deduction by adopting a mode of combining data driving and similar working condition retrieval, and completing index calculation and safety verification; and full-life-cycle management is carried out on the digital intelligent application, and cross-site rapid copying is realized through template packaging. According to the system, low-risk and high-efficiency iterative verification and popularization of digital intelligent application can be realized on the premise of not interfering field production, and the safety and engineering level of industrial intelligence landing are remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of industrial digital technology, and in particular to an offline virtual laboratory and a digital application iteration method and system based on intelligent agents. Background Technology

[0002] Existing industrial site digital intelligence systems, control strategies, and intelligent applications have long faced the following contradictions in engineering implementation: While on-site operating conditions offer the most realistic and complete constraints, any trial and error can lead to production fluctuations, downtime losses, and safety risks; offline environments are safer and more controllable, but often deviate from the on-site environment due to difficulties in continuously synchronizing configurations, point tables, version dependencies, and typical operating condition data, making offline test results unrepresentative and difficult to transfer to guide on-site decision-making. Simultaneously, the iteration of digital intelligence applications and control strategies often exhibits a fragmented construction and decentralized operation and maintenance pattern, lacking replicable, auditable, and quantifiable iterative methodologies and system support. This results in inconsistent evaluation criteria, incomplete verification evidence chains, high promotion and replication costs, and difficulty in achieving a closed-loop effect after deployment.

[0003] Specifically, existing technologies have shortcomings in the following aspects: Current technologies lack offline testing platforms that can be safely isolated and closely connected to the field for extended periods. In current practices, new processes, parameters, and control strategies often have to be verified in the field or semi-field, which can easily disrupt continuous production and introduce uncontrollable safety risks. Existing offline simulation or testing environments are mostly one-time setups or only synchronize partial historical time-series data, failing to systematically replicate the field software versions, application components, configuration parameters, and point table relationships. Furthermore, they drift with continuous changes in the field, ultimately making it difficult to apply offline verification conclusions to field decision-making.

[0004] Existing technologies lack a systematic synchronization and consistency maintenance mechanism that covers the historical statistical rules and logic of typical operating conditions in the configuration point table. To ensure the representativeness of offline virtual laboratories, it is not enough to focus solely on curve data playback. It is also necessary to include key assets such as system configuration, point table structure, unit range, sampling and quality identification, parameter templates, application versions and dependency lists, interface definitions, alarm rules, interlocking logic and emergency procedures in synchronization and version management, and to provide consistency verification, difference detection, change impact analysis and rollback mechanisms. Otherwise, the offline environment will inevitably have structural deviations from the field, resulting in unreproducible experiments, untraceable results, and unassessable risks.

[0005] Existing technologies struggle to implement simulations when mechanistic models are incomplete. Many sites or devices lack sophisticated mechanistic models that can be maintained long-term. Relying heavily on complex mechanistic equations for full-process simulations leads to problems such as long modeling cycles, difficulty in parameter identification, high maintenance costs, and difficulties in cross-site migration, making it difficult to scale up and reuse offline capabilities.

[0006] Existing technologies struggle to provide cost-effective and repeatable verification for high-risk, extreme operating conditions and failure scenarios. Industrial safety boundaries are often manifested in scenarios such as extreme liquid levels, extreme pressures, equipment failures, and frequent start-ups and shutdowns. However, conducting on-site verification carries high risks, complex approval processes, and significant organizational costs and downtime consequences. Consequently, protection strategies often remain at the level of document review or limited sampling, making it difficult to achieve system coverage and quantifiable conclusions.

[0007] Existing technological research experiments and scheme comparisons lack a unified standard indicator system and automated comparison capabilities. New processes and strategies often require parallel testing of multiple schemes, but traditional methods rely on personal experience and scattered tools, resulting in inconsistent indicator standards, non-standardized test package organization, and high costs for data preparation and reproduction. This leads to incomparability between schemes, significant review disputes, and difficulty in consolidating conclusions.

[0008] Existing technologies and digital applications lack a unified ledger and lifecycle governance for iteration. Most existing applications exist in a decentralized manner with partial deployment and scattered maintenance, lacking a unified application list and configuration ledger. Application versions, dependency points, applicable working condition boundaries, interface constraints, running status, audit and rollback information are difficult to manage systematically, resulting in high iteration risks, high handover costs, and slow problem localization.

[0009] Existing technologies lack standardized interfaces and automated dependency verification for rapid integration and testing. Application deployment is often slowed down by interface integration, point mapping, configuration generation, and conflict troubleshooting, and is highly dependent on human experience, which can easily introduce configuration errors and increase deployment risks.

[0010] Existing technologies lack quantifiable effects and a closed-loop system for continuous optimization after deployment. Many systems lack continuous evaluation mechanisms after deployment, making it difficult to prove contributions with data and to detect strategy degradation or deviations in applicability in a timely manner. This leads to a "deploy and it's over" situation, hindering continuous iteration.

[0011] The current technology is costly and error-prone to cross-site replication and promotion. When mature strategies and applications are promoted to different sites, a large amount of manual work is often required to complete site mapping, parameter migration, working condition adaptation checks and safety verification. This process is time-consuming and prone to omissions or errors, resulting in low replication efficiency.

[0012] The introduction of intelligent agents into existing technologies presents a challenge in balancing automation efficiency gains with security boundaries. Direct intervention of intelligent agents in the real-time control execution chain introduces uncontrollable risks, while remaining merely at the presentation layer fails to create a closed-loop engineering value. Summary of the Invention

[0013] The purpose of this invention is to address the shortcomings of existing technologies by providing an offline virtual laboratory and a method and system for iterative digital applications based on intelligent agents. This method constructs an offline virtual laboratory in a non-production environment that is consistent with or highly similar to the field, without accessing or interfering with the real-time control execution chain. It introduces intelligent agents as a collaborative hub for orchestration and analysis, forming a comprehensive engineering capability encompassing data and configuration synchronization, operational organization, simulation, scripted verification, indicator evaluation, application management, rapid integration and debugging, pre- and post-deployment comparative evaluation, and template-based replication and promotion. This allows the offline environment to maintain long-term representativeness of the field while achieving secure isolation, thereby significantly reducing the risks and costs of field verification and improving the iterative efficiency and scalable diffusion capabilities of digital intelligence.

[0014] In a first aspect, the present invention provides an offline virtual laboratory and digital application iteration system based on intelligent agents, comprising: The synchronization and isolation interface layer is used to obtain real-time data, historical data, system configuration, rule logic and business ledger from the field data source layer, and to perform de-identification and permission labeling on the obtained data to form data and configuration resources that can be safely used in offline environments. The data foundation and consistency governance layer connect to the synchronization and isolation interface layer. It is used to receive and store the data and configuration resources, and to perform version management, difference detection and data governance on the data and configuration resources based on the unified object model, so as to generate and maintain a consistent data foundation containing versioned data, configuration, rules, working condition profiles and dependency graphs. The simulation verification and experiment execution layer connects the data base and the consistency governance layer. It is used to arrange experimental test cases based on the working condition profile in the consistent data base, and call the simulation simulation executor. It uses data-driven simulation model, similar working condition retrieval and rule constraint verification to perform simulation simulation and verification of digital intelligent applications and strategies, including security policies, in an offline environment, and generate indicator calculation results and verification reports. The application iteration and delivery layer connects the data foundation, the consistency governance layer, and the simulation verification and experiment execution layer. It is used to manage the entire lifecycle of digital applications, perform application dependency parsing and point mapping based on the dependency graph in the consistent data foundation, configure and integrate applications through standard interfaces in the joint debugging sandbox, and generate application launch evaluation and optimization suggestions based on the indicator calculation results and verification reports generated by the simulation verification and experiment execution layer. The intelligent agent orchestration and process governance layer connects the synchronization and isolation interface layer, the data foundation and consistency governance layer, the inference verification and experiment execution layer, and the application iteration and delivery layer. It is used to automatically orchestrate and schedule the entire process of data synchronization, consistency governance, experiment orchestration, inference verification, application integration and evaluation delivery through intelligent agents, and to perform process governance based on the unified object model and the dependency graph. A unified portal and role workbench connect to the agent orchestration and process governance layer, providing a human-computer interaction interface to drive and monitor the automated processes of the agent orchestration and process governance layer. Through the approval and distribution interface of the synchronization and isolation interface layer, applications or policies that have been verified offline are securely distributed to the field after approval.

[0015] Furthermore, the synchronization and isolation interface layer is used for: Establish a synchronization strategy that combines periodic synchronization and on-demand synchronization. The synchronization objects include historical time-series data, system configuration parameters, point table structure, alarm rules, interlocking logic, and emergency procedures. The synchronized data is uniformly encoded, time-aligned, and difference-detected, and a version number and checksum are generated for each synchronization. Sensitive data is anonymized and access rights are assigned.

[0016] Furthermore, the data foundation and consistency governance layer include: The unified object model module is used to assign offline object identifiers to measurement points, devices, rule entries, and application instances; The version management and difference detection module is used to store configuration snapshots, point tables and rules in a versioned manner and compare differences. The dependency graph module is used to maintain the dependencies between objects and generate regression verification plans when changes occur.

[0017] Furthermore, the deduction verification and experiment execution layer includes: The data-driven simulation model module uses a combination of data models, statistical laws, and similar working conditions retrieval and replay to perform process simulation. The rules and constraints verification module is used to verify the feasibility of the simulation results and identify risks. The security policy verification module is used to perform scripted verification of alarm rules and interlocking logic.

[0018] Furthermore, the application iteration and delivery layer includes: The application ledger and lifecycle management module is used to record application versions, dependency points, and running status; The dependency resolution and point mapping module is used to resolve application dependencies and perform point mapping verification. The integration sandbox module is used to perform interface integration testing and compatibility verification of applications in an offline environment.

[0019] Furthermore, the agents in the agent orchestration and process governance layer are used for: The tool interface layer calls data query, work condition arrangement, and simulation execution tools. Generate a task plan that includes data version, configuration version, and model version; When outputting conclusions, bind evidence chain elements such as data windows, key curves, and rule trigger records.

[0020] Furthermore, the system also includes: The security permissions and auditing module is used to provide identity authentication, role-based access control, and full-process auditing and traceability. The operation monitoring and maintenance module is used for task queue monitoring, resource quota management, and service health checks.

[0021] Furthermore, the deduction verification and experiment execution layer is also used for: Establish a unified indicator system, which should at least include indicators related to output, energy consumption, volatility, and alarm triggering. Comparative analysis of the results of parallel simulations of multiple scenarios is conducted to generate an assessment report that includes a comparison of differences and a risk list.

[0022] Furthermore, the application iteration and delivery layer is also used for: Encapsulate mature applications into template packages that include parameterized configurations and dependency declarations; Perform site mapping verification and operating condition adaptation checks during cross-site migration; Complete regression testing before the new site goes live.

[0023] Secondly, this invention provides an offline virtual laboratory and digital intelligence application iteration method based on intelligent agents, comprising the following steps: Step 1, Demand Triggering and Target Definition: Receive trigger requests from scientific research experiments, technical modification verification, strategy parameter tuning, security verification, application access, or field change regression; determine the target site or device scope, target problem, expected output type, and constraints; the expected output type includes reports, suggestions, configuration drafts, or regression conclusions; the constraints include security boundaries, data range, and time windows. Step 2, Version Selection and Synchronization Preparation: Based on the objectives determined in Step 1, select the configuration version, point table version, rule version, model version, and indicator caliber version, and generate a synchronization list and synchronization strategy; the synchronization strategy includes periodic synchronization or on-demand synchronization, full synchronization or incremental synchronization, and synchronization scope pruning, and establish the version binding relationship for this process. Step 3, Data and Configuration Synchronization and Consistency Verification: Based on the synchronization list and synchronization strategy generated in Step 2, synchronize time-series data, event data, configuration snapshots, point table metadata, rule logic, and emergency process assets from the field domain; perform difference detection, patch generation, and consistency verification on the synchronization results; update the offline environment version after the verification passes; if missing or inconsistent data is found, output repair suggestions and return to this step to re-execute until the missing data is found. Step 4, Data Governance, Object Modeling, and Operating Condition Profile Update: Perform time alignment, unit dimension consistency processing, quality identification, and source traceability on the data synchronized and verified in Step 3; update object mapping relationships and dependency graphs; update operating condition profiles and typical operating condition indexes based on historical operating data; Step 5, Operational Condition Orchestration and Test Case Generation: Receive key parameters and strategies input by the user or agent, generate operational condition scripts and convert them into standardized test cases; define the baseline, evaluation index set, and pass criteria; store the test cases in the test case library and bind version information. Step Six, Simulation Execution and Rule or Constraint Verification: Based on the test cases generated in Step Five, select at least one simulation route from the data model, statistical regularity, similar working condition replay, or combination route, and execute time progression and disturbance injection; during the simulation, call the rule and constraint verification, mark risks and provide correction prompts for out-of-bounds behavior, abnormal fluctuations, and infeasible outputs, and generate process variable trajectories, event sequences, and operation logs; Step 7: Calculation of indicators, comparison of schemes and output of conclusions: Based on the simulation results generated in Step 6, calculate output, energy consumption, volatility, stability, number of alarms or interlocking triggers and risk exposure indicators according to a unified indicator system; if there are multiple schemes simulated in parallel, output the horizontal comparison results and recommended conclusions, and clarify the recommendation premises, applicable boundaries and risk list, and form a standardized assessment report and evidence chain for archiving. Step 8, Security Policy Script Verification and Regression Verification: When security verification is required, import and load alarm rules, interlocking logic and emergency procedures, construct extreme condition or fault scenario scripts and execute verification; output protection action sequence, response time, coverage and completeness conclusions; if defects are found, generate candidate improvement schemes and trigger regression test case set re-verification until the criteria are met or risk acceptance recommendations are formed. Step 9, Application Integration, Delivery, and Launch Closed Loop: When application iteration is involved, complete dependency resolution, point mapping verification, and configuration package generation. After offline integration sandbox verification, generate launch materials and enter the approval process. After launch, continuously evaluate the application effect based on monitoring indicators and output optimization suggestions. When the application effect is stable and meets preset conditions, package it into a template package and start the cross-site replication process.

[0024] The beneficial effects of this invention are as follows: By constructing an offline virtual experimental environment consistent with the field in a non-production environment and maintaining physical or logical isolation from the real-time control link, this invention achieves zero-operation on field equipment during the verification process. It is particularly suitable for high-risk scenarios such as extreme liquid levels, pressures, and equipment failures, significantly reducing production disturbances and safety risks caused by on-site trial and error, and improving engineering safety. By incorporating configuration parameters, point table structures, model parameters, typical operating conditions, and historical statistics into the periodic or on-demand synchronization scope, and implementing version management and consistency maintenance mechanisms, the offline environment can maintain long-term representativeness of on-site operating conditions. This effectively solves the problem of non-transferable conclusions caused by data drift in traditional offline simulations, improving the credibility and guiding value of simulation results. Employing a combination of data models, statistical laws, empirical rules, and similar operating condition retrieval, this invention can quickly build usable simulation capabilities even when the mechanistic model is missing or incomplete. This significantly reduces the modeling threshold and maintenance costs, shortens the construction cycle, and better meets the needs of rapid engineering implementation. By structurally importing interlocking logic, alarm rules, and emergency procedures into the offline environment, and utilizing scripts to drive simulations of extreme operating conditions and fault scenarios, repeatable and quantifiable verification of safety strategies is achieved. This allows for the systematic output of results such as action sequence, response time, and process completeness, enhancing the depth and credibility of safety verification. A unified indicator system supports parallel comparison of multiple solutions, automatically generating evaluation reports and optimization suggestions including indicators such as output, energy consumption, fluctuations, alarm interlocking triggers, and risk exposure. This solves the problems of inconsistent indicator definitions and difficulty in reproduction in traditional solution comparisons, significantly improving the efficiency of research and technological upgrades. Through unified management, standardized interfaces, and automated dependency verification and joint debugging assistance, application silos and redundant development are reduced, lowering integration costs and deployment risks, shifting application iteration from decentralized operation and maintenance to systematic governance. A mechanism for comparing pre- and post-deployment effects and a continuous optimization loop are established. Data-driven evaluation and parameter tuning suggestions avoid the dilemma of "deployment and then stagnation," enhancing the long-term operational value of the application. Mature applications are encapsulated into parameterizable templates, and combined with point mapping, operational condition adaptation checks, and regression verification, enabling rapid and reliable replication and promotion across sites, significantly improving the scalability of the results. The intelligent agent, acting as the collaborative hub of the offline domain, is responsible for auxiliary decision-making tasks such as operational condition orchestration, simulation scheduling, and report generation. Its output is limited to interpretable suggestions rather than direct control commands, ensuring engineering safety and compliance while improving automation levels, thus forming a complete, scalable, and maintainable technical system. Attached Figure Description

[0025] Figure 1 This is a diagram of the architecture of the agent-based offline virtual laboratory and digital application iterative system of the present invention; Figure 2 This is a flowchart of the agent-based offline virtual laboratory and digital application iterative method of the present invention. Detailed Implementation

[0026] The following non-limiting embodiments are intended to enable those skilled in the art to more fully understand the present invention, but do not limit the present invention in any way. It should be noted that the following embodiments should not be construed as limiting the scope of protection of the present invention. If those skilled in the art make some non-essential improvements and adjustments to the present invention based on the above description, they shall still fall within the scope of protection of the present invention.

[0027] I. Overall Technical Solution: This invention addresses the construction and continuous iteration needs of digital and intelligent systems at the industrial site or device level, proposing an integrated technical solution based on intelligent agents for offline virtual laboratories and digital and intelligent application iteration methods and systems. This solution integrates offline virtual experiment capabilities and digital and intelligent application iteration capabilities within the same engineering framework. Using a non-production environment as its platform, it constructs a standardized path from experimentation to deployment and then to replication and promotion through design principles such as offline isolation but long-term consistency, data-driven deduction without relying on complex mechanisms, extreme condition script verification without touching the on-site execution chain, unified application management but support for rapid integration, quantifiable deployment effects that form a continuous optimization loop, and mature solutions that are templated but can be migrated and reused across sites. The system uses intelligent agents as the coordinating hub, responsible for transforming requirement expressions into executable condition orchestrations, test cases, verification scripts, evaluation reports, and configuration drafts. Through tool interfaces, permission boundaries, and auditing mechanisms, the capabilities of intelligent agents are limited to the offline domain and suggestion output domain, architecturally preventing them from directly intervening in real-time on-site control execution, thereby achieving an engineering balance between efficiency improvement and safety compliance.

[0028] II. Technical solutions for the construction and operation of offline virtual laboratories.

[0029] (a) Offline deployment and security isolation.

[0030] Deploy an offline virtual experimental environment on a non-production network or independent computing resource. This environment replicates or is compatible with the version system and runtime dependencies of relevant on-site components at the software stack level, and includes at least data access and storage components, simulation and deduction execution components, rule and constraint verification components, test case management components, indicator calculation and report generation components, and visualization and interaction components.

[0031] The offline virtual experimental environment is strongly isolated from the real-time control link on site. Isolation methods may include physical isolation, logical isolation, one-way data transmission, or controlled read-only channels. By default, only the field side is allowed to output data and configuration snapshots to the offline side. The offline side does not have a channel to directly connect to the field control execution. If a closed loop of suggestion approval and issuance is required, the offline side only outputs the configuration draft and change summary. It can only be issued through the controlled channel after entering the independent approval link, and the issuance action must meet the requirements of hierarchical authorization and audit trail.

[0032] The offline virtual experimental environment is divided into tenant spaces or project spaces according to sites or devices, forming resource quotas and data isolation, avoiding the risk of unavailability caused by data crosstalk, confused conclusions or resource contention when batch experiments or multiple teams use it concurrently.

[0033] (ii) Scope and strategy for synchronizing on-site data and configuration.

[0034] Establish a synchronization strategy that combines periodic synchronization and on-demand synchronization. Periodic synchronization is used to maintain the long-term representativeness of the offline environment, while on-demand synchronization is used to deal with scenarios such as field changes, special tests, or regression verification before deployment. The synchronization strategy supports selecting synchronization sets by site, device, system, or application dependency scope to reduce unnecessary data transfer costs.

[0035] Synchronization objects include not only historical time-series data, but also configuration data and semantic data, including at least system configuration parameters, parameter templates, point table structure and measurement point metadata, unit dimensions and range information, sampling period and quality identification rules, typical working condition segments and historical statistical profiles, start and stop records and key event sequences, alarm rules and interlocking logic, emergency procedures and handling templates, as well as interface definitions and version information related to application access.

[0036] During the synchronization process, data is uniformly encoded, time-aligned, missing and anomaly labeled, quality-layered, and source-traceable. Difference detection is performed on configuration and point tables to identify difference types such as additions, deletions, renamings, range changes, unit changes, sampling changes, and quality code rule changes. Version numbers, verification summaries, and rollback points are generated for each synchronization to support the reproducibility and audit traceability of offline experiments.

[0037] During the synchronization process, it supports data anonymization and access control, allowing sensitive fields, sensitive ranges, or sensitive business information to be anonymized or made visible in a hierarchical manner according to rules, ensuring that offline experiments can be conducted without breaching data security boundaries.

[0038] (III) Construction and object-oriented management of experimental data base.

[0039] A unified data foundation for testing is built in the offline domain, integrating time-series data, event data, configuration data, point table semantics, and rule logic into an object-oriented management system; stable offline object identifiers are assigned to measurement points, equipment, process units, rule entries, emergency nodes, test cases, and application instances, and the mapping relationship between objects and field identifiers is maintained.

[0040] The data foundation maintains the dependency graph of the measurement point configuration rules model and use case applications. When the field point table or configuration changes and is synchronized offline, the system automatically completes the impact scope analysis, outputs the list of affected use cases, models, rules and applications, and generates the regression verification plan that needs to be executed, so as to avoid the failure of offline test or the expansion of the risk of going online due to the change of implicit dependency.

[0041] The data base maintenance condition profile library groups historical operations according to dimensions such as seasonal temperature zones, load ranges, process status, start-up and shutdown stages, and equipment combination status, forming profile indicators such as baseline statistics, fluctuation bands, quantile intervals, out-of-bounds probability, and event frequency. The profiles are then linked with typical operating condition segments to support subsequent similar operating condition retrieval, comparative evaluation, and consistency of indicator standards.

[0042] The data platform provides a unified data retrieval interface and unified query capabilities, supporting data acquisition by object, time window, working condition group, and event trigger window. It also transmits quality codes and source traceability information at the interface layer, enabling subsequent analysis and evaluation to explicitly process data quality rather than implicitly ignore it.

[0043] (iv) Process simulation and deduction technology route that does not rely on complex mechanism equations.

[0044] The simulation and deduction of this invention does not rely on solving complex mechanistic equations. Instead, it adopts a combination of data model, statistical regularity, empirical rules / constraints, and similar working condition retrieval and replay. This approach lowers the modeling threshold while ensuring usability, and allows for gradual improvement in accuracy and coverage based on site maturity.

[0045] The data model is used to characterize the mapping relationship between key input parameters and process variables and energy consumption indicators. The model types can include regression models, piecewise models, baseline offset models based on operating condition clusters, and interpretable nonlinear mapping models, etc. The system supports hierarchical organization of models at the equipment level, unit level, and station level, so that local usable capabilities can be quickly implemented, while reserving expansion space for larger-scale coupled inference.

[0046] Statistical patterns are used to provide feasible extrapolation capabilities when there are insufficient training samples or stable input features. For example, quantile bands, fluctuation bands, threshold boundaries, trend drift estimation, periodic features, and statistical patterns of start and stop phases can be used to generate reference output intervals and mark uncertainties so that the evaluation report can distinguish between deterministic conclusions and interval conclusions.

[0047] Empirical rules and constraints are used to verify the feasibility and label the risks of the simulation results. The rules can include the allowable range of equipment, the limit of the rate of change, the process forbidden zone, the operation boundary, the safety redundancy requirements, etc. The constraint verification can be performed by time step or by event node during the simulation process, and output the out-of-bounds type, duration, triggering condition and risk level, so as to establish an explicit relationship between the simulation results and the safety boundary.

[0048] Similar operating condition retrieval and replay is used to preserve the true dynamic form and the true coupling relationship. The system constructs operating condition feature vectors based on features such as influent volume, pressure, temperature, water mixing ratio, start-stop rhythm, key valve position or process status, retrieves the most similar operating condition segments in the historical database, and generates simulation curves through scaling transformation, time stretching, interpolation mapping and boundary correction. The system records the retrieved segments and mapping parameters, so that the report can explain which historical segments were referenced and how they were mapped to obtain the current output, thereby enhancing interpretability and traceability.

[0049] Based on the above combined approach, the system allows the introduction of a small number of simplified conservation constraints or simplified mechanisms as consistency check items when necessary. However, these check items are only used for verification and correction and do not constitute a dependency on the solution of complex mechanisms, thereby avoiding long-cycle modeling, parameter identification and continuous maintenance costs.

[0050] (v) Operating condition arrangement, parameter customization and test case standardization.

[0051] The system provides operational condition orchestration capabilities, allowing users to customize key production parameters and operating strategies in the offline virtual experiment interface or through open interfaces. These include continuous parameters such as influent volume, pressure, temperature, and water mixing ratio, as well as discrete strategies such as start-up and shutdown strategies, switching rhythms, chemical dosing strategies, hot washing strategies, and bypass strategies.

[0052] The system supports various operating condition input formats, including constant operating condition, step disturbance operating condition, slope change operating condition, periodic disturbance operating condition, combined operating condition, and operating condition with fault injection; for each operating condition input, the duration, sampling granularity, baseline, set of evaluation indicators, and pass criteria can be specified.

[0053] The system supports natural language input, and the intelligent agent parses the user's intent into a standardized operating condition script. The script includes at least the parameter sequence, disturbance injection method, fault triggering condition, duration, sampling and output requirements, etc. After the script is generated, it is entered into the test case library and bound to the version number, so that the same test can be repeatedly executed in the future for regression verification or cross-site comparison.

[0054] The system uniformly converts operating condition inputs into test case objects. Each test case includes at least a test case identifier, applicable site / device scope, dependent configuration version, input parameter definition, execution boundary, evaluation index, pass criterion, expected output, and audit information. Test cases support reuse, derivation, parameter scanning, and batch execution to support the large-scale development of scientific research experiment design and scheme comparison.

[0055] (vi) Simulation execution, indicator system and report generation.

[0056] The system provides simulation executor and task scheduling capabilities. The simulation executor reads input according to the use case definition and calls the inference model, statistical law, similarity retrieval and rule verification components to generate process variable trajectories and event sequences. The task scheduler supports parallel execution of multiple use cases, multiple scenarios and parameter scanning tasks, and allocates computing resources according to resource quotas and priorities to avoid system unavailability caused by high concurrency.

[0057] The system establishes a unified indicator system and indicator library, maintaining the definition, unit, calculation method, time window, abnormal data processing rules and statistical methods for each indicator to ensure comparability across personnel, sites and versions; the indicators should at least cover categories such as output-related, energy consumption-related, fluctuation and stability-related, start-stop and recovery-related, alarm and interlock trigger-related, risk exposure and out-of-bounds duration-related.

[0058] The system outputs a standardized report for each simulation execution. The report includes at least a description of the input conditions, configuration and data versions, a description of the simulation route, a summary of key process variable curves, a summary table of indicators, a list of risks and out-of-bounds events, a comparison with the baseline, and interpretable attribution suggestions. For multi-scheme comparison scenarios, the report provides a comparison of differences, a risk-benefit trade-off analysis, and a recommended solution, and clarifies the prerequisites and applicable boundaries of the recommendation to avoid misapplying conclusions from local conditions to incompatible conditions.

[0059] The system supports auditing and recording reports, and archives reports along with use cases, model versions, rule versions, and data versions to ensure that subsequent reviews can be replayed, verified, and explained, forming a valuable experimental asset.

[0060] (vii) Scripted verification of extreme operating conditions and safety strategies.

[0061] The system imports alarm rules, interlocking logic, and emergency procedures into the offline virtual experimental environment in a structured form, forming a computable rule graph, state machine, or process node network, and semantically binds them with an objectified point table, enabling the rules to be triggered and recorded during simulation.

[0062] The system provides a script verification mechanism that combines time-based progression and event-triggered execution. Time-based progression drives the evolution of process variables over time, while event-triggered execution triggers alarms, interlocking actions, and emergency process nodes when conditions are met. The script can construct high-risk scenarios such as container liquid level limits, pressure limits, equipment failures, sensor drift, frequent start-stops, communication interruptions, and valve position jamming, and specify the trigger time, disturbance amplitude, duration, recovery conditions, and verification targets.

[0063] During the verification process, the system records the timing of rule triggering, action sequence, response time, coverage, conflict and competition situations, as well as the completion status of emergency process nodes, and outputs the conclusions of coverage, completeness, and consistency checks. When problems such as unreasonable thresholds, gaps in the action chain, missing process nodes, or delayed responses are found, the system outputs the location of weak links, risk level, and improvement suggestions, and can solidify the improvement suggestions into candidate rule versions to enter the review and regression verification cycle.

[0064] The entire verification process is completed in an offline environment without operating on-site equipment or triggering on-site execution links. This transforms the originally high-risk, high-cost, and difficult-to-cover on-site tests into repeatable, quantifiable, and traceable offline script verification, significantly reducing the need for dangerous tests and production stoppage verification.

[0065] III. Technical solutions for iterative digital applications and systematic governance.

[0066] (a) List of digital applications and lifecycle ledger.

[0067] The system establishes a unified list of digital applications and a lifecycle ledger, bringing existing and new applications under unified management. The ledger records at least the application identifier, version number, dependency points and dependent services, applicable working condition boundaries, key parameter items, operating status, change records, rollback strategies, summary of evaluation conclusions, and audit information, thus achieving a governance foundation that is searchable, controllable, and traceable.

[0068] The system establishes a dependency graph for each application, which manages the relationships between the application and test point objects, rule entries, model versions, and interface permissions in a graphical way. When the test point table or configuration changes, it automatically generates prompts and regression verification lists to avoid application unavailability or misjudgment due to dependency breakage after deployment.

[0069] The system supports the entire lifecycle of an application, from project initiation, development, offline integration testing, offline verification and review, online release, operation monitoring, evaluation and optimization to decommissioning. It generates evidence materials and audit records at each status node, making application iteration change from human-to-human monitoring to process control and traceable evidence.

[0070] (ii) Standardized interface system and rapid integration mechanism.

[0071] The system defines standardized data interfaces, control interfaces, and interface embedding specifications. The data interfaces are used to uniformly acquire real-time and historical data, simulation data, and profiling data. The control interfaces are used to output suggestions, control drafts, or policy candidates and clarify the default boundaries that cannot be directly executed. The interface embedding specifications are used to unify the interaction methods of menu permissions, message notifications, alarm display, and report presentation.

[0072] When a new application is integrated, it should be submitted in accordance with the specifications, including application description and dependency declaration, configuration template, logic description, and necessary model or rule package. The application management component will automatically parse the dependencies and complete point mapping suggestions, unit dimension verification, sampling granularity matching checks, permission requirement list generation, and conflict detection. For missing points or inconsistent items, no fabrication will be made. Instead, a missing point mark + candidate replacement + manual confirmation method will be used to ensure that the integration results are feasible and auditable.

[0073] The system provides a configuration generator that generates site-specific configuration packages based on application templates and mapping results. The configuration package includes default parameter values, adjustable ranges, dependency verification results, change summaries, and rollback points. The configuration package is also bound to an offline verification report to form a chain of evidence explaining why the configuration is the way it is, reducing communication costs and the risk of misunderstandings during the online review.

[0074] The system provides a joint debugging sandbox, which allows applications to run and receive simulated inputs in an offline virtual experimental environment. Through consistent data and event interfaces, scenarios can be reproduced to verify application logic, boundary conditions, exception handling, concurrent behavior, and interface compatibility. During the joint debugging process, the system records call logs, key outputs, and exception stack information to provide a basis for subsequent repair and regression testing.

[0075] (iii) Closed loop of pre-launch verification, post-launch evaluation and continuous optimization.

[0076] Before going live, the system conducts comparative verification of the application based on an offline virtual laboratory. This includes at least a comparison of the application under the same working conditions with the application turned on and off, sensitivity analysis of different parameter combinations, robustness testing under boundary conditions and abnormal data quality, and outputs expected benefits, a risk list, applicable boundaries, and monitoring indicator suggestions to provide a quantitative basis for the go-live decision.

[0077] After going live, the system continuously collects operational data and performance indicators, and conducts comparative evaluations according to working conditions. Evaluation methods may include comparisons before and after going live, comparisons of similar working conditions, horizontal comparisons of similar equipment, and critical event window analysis. The system outputs changes in indicators such as output, energy consumption, fluctuations, alarm interlocks, safety events, stability, and number of manual interventions, and provides attribution clues when degradation occurs, such as working conditions exceeding applicable boundaries, data quality degradation, parameter drift, changes in equipment status, or changes in external constraints.

[0078] The system transforms the evaluation conclusions into an iterative task list, which includes suggestions for parameter optimization, threshold and rule adjustment, expansion or convergence of the applicable scope, strategy downgrade or deactivation, and submits the suggestions and evidence chain to the review process. This ensures that the application iteration forms a closed loop of online evaluation, optimization, verification and online deployment, rather than ending once online.

[0079] (iv) Templated encapsulation and cross-site copying and promotion.

[0080] The system encapsulates applications that meet preset evaluation conditions into reusable template packages. Each template package includes at least the following: application package or component reference, configuration template and parameterization items, dependency declaration and interface specifications, applicable working condition boundary description, offline regression test case set, evaluation baseline and monitoring indicator suggestions, as well as risk warnings and rollback strategies. This ensures that the template package can not only be installed, but also verified, evaluated, and rolled back.

[0081] When a template is migrated to a new site, the system guides the completion of point mapping and unit dimension consistency verification, automatically identifies missing items and provides candidate measurement points or candidate indicators, and at the same time builds a working condition profile of the new site's historical data and compares it with the template's applicable boundaries, outputting a list of working conditions that need to be verified and risk warnings to avoid misusing the template to sites with significant differences.

[0082] Before a template goes live on a new site, it must undergo regression testing in an offline virtual lab. Only after successful testing can it enter the controlled deployment process. After deployment, it continues to be included in the effect evaluation and optimization loop, thereby realizing a large-scale promotion mechanism that allows a single site to run successfully, multiple sites to replicate, and continuous evolution, while significantly reducing manual adaptation costs and configuration error risks.

[0083] IV. Technical solutions for intelligent agent collaborative hub and tool-based orchestration.

[0084] (a) Agent role localization and input / output boundaries.

[0085] In this invention, the intelligent agent is positioned as a collaborative hub and automated assistant in the offline domain. Its core responsibility is to organize requirements, data, rules, models, use cases, reports, and configuration drafts into an executable workflow and generate interpretable suggestions and evidence chains. By default, the intelligent agent does not have direct authority to execute real-time control on-site. Its output to the production side is limited to suggestions, reports, risk warnings, and configuration drafts. Execution must go through a hierarchical authorization and approval process.

[0086] The input to the intelligent agent can come from natural language requirements, structured test requests, application access declarations, change orders, alarms, and event triggers. The intelligent agent transforms the input into standardized objects, such as work scripts, test cases, evaluation tasks, regression plans, configuration drafts, risk lists, and approval materials, thereby reducing the cost of manually translating ideas into executable actions.

[0087] (ii) Tool interface layer and controllable calling mechanism.

[0088] The system encapsulates capabilities such as data query, work condition orchestration, similarity retrieval, simulation execution, rule verification, indicator calculation, report generation, dependency resolution, configuration generation, joint debugging and startup, and template packaging into controlled tools. The intelligent agent calls the tools through the tool interface layer to complete the task. The tool interface layer performs input verification, permission verification, quota restriction and auditing records for each call to prevent the intelligent agent from causing resource loss of control or unauthorized access in batch tasks.

[0089] The tool interface layer supports secondary confirmation or manual review of sensitive operations, such as generating configuration drafts that can be distributed, generating cross-site template packages, and releasing new indicator versions, to ensure that key actions are controllable.

[0090] The tool interface layer provides a structured summary of the output, including key parameters, discrepancies, scope of impact, risk warnings, and rollback points, making the agent's output engineering readable and facilitating its entry into the approval process and the formation of audit evidence.

[0091] (III) Intelligent agent workflow orchestration and human-machine collaboration.

[0092] After receiving a task, the agent generates a task plan. The task plan includes at least the data version, configuration version, rule version, and model version to be used, the set of use cases to be executed, the baseline selection, the indicator caliber selection, and the expected report type. After the task plan is generated, the agent enters the execution phase, calling tools according to the plan and monitoring the execution status.

[0093] If data loss, inconsistency in measurement points, rule conflicts, or model unavailability are found during execution, the agent outputs actionable remedial suggestions, such as supplementing the synchronization range, selecting alternative measurement points, adjusting the working condition boundaries, reverting to the statistical regularity route, enabling the replay route of similar working conditions, or triggering manual confirmation. After the remedial suggestions are confirmed, they form new executable use cases or new configuration drafts, thus forming a closed loop.

[0094] When an intelligent agent outputs a conclusion, it must be bound to elements of the evidence chain, including at least a data window, key curves or key indicators, rule trigger records, baseline descriptions, and binding information with the version number, so as to ensure that the recommendations are interpretable, verifiable, and traceable, and meet the requirements of engineering management and auditing.

[0095] V. Technical solutions for security boundaries, access control, and audit traceability.

[0096] (a) Domain isolation and least privilege.

[0097] The system isolates the production control domain from the offline test domain. The offline domain is read-only and receives field data and configuration snapshots by default, and does not have direct distribution permissions. If a closed-loop distribution is required, it must be achieved through an independent approval link and a controlled channel, and one-click rollback is supported.

[0098] The system adopts the principle of least privilege, configuring role permissions for actions such as data access, test case creation, model training, rule publishing, configuration generation, and template publishing. It also supports fine-grained authorization by site, device, and application to avoid cross-site misoperation and unauthorized access.

[0099] For critical operations, support multi-factor confirmation, dual-person review, or tiered approval strategies to ensure that risky operations are controllable.

[0100] (ii) Full-process audit and evidence chain archiving.

[0101] The system audits and records the entire process, including synchronization, test cases, simulation, verification, evaluation, configuration, deployment, rollback, and template migration. The audit records include at least the operator, time, target object, input summary, output summary, associated version number, and key evidence links, so that every conclusion and every change is traceable.

[0102] The system supports replaying the entire process by event number, use case number, application version, or deployment batch, reproducing the data window, key rule triggers, and indicator calculation methods at that time, thereby supporting review, training, and accountability.

[0103] (III) Availability assurance and degradation strategies.

[0104] The system's critical services support redundant deployment and failure failover, as well as breakpoint resume and failure retries, ensuring the stable execution of batch testing and evaluation tasks.

[0105] The system provides a degradation strategy. When the learning model is unavailable or the data quality is insufficient, it can automatically revert to the path of statistical regularity or similar working conditions and mark the reason for degradation and the range of uncertainty in the report to avoid outputting seemingly accurate but unbelievable conclusions.

[0106] The system provides resource isolation and quota control to prevent batch scanning tasks from a single team from consuming all resources and causing other critical verification tasks to fail.

[0107] VI. System Architecture Description (Interpretation of the top-down architecture diagram).

[0108] (a) Unified portal and role-based workbench (top layer) Role-based workbench: Provides a unified entry point for different roles such as research, technology, operation and maintenance, and management, and displays functional views such as experiment arrangement, verification and execution, application management, evaluation dashboard, template promotion, and audit traceability according to role.

[0109] Experiment and Use Case Workbench: Provides capabilities for selecting work conditions, editing parameters, searching the use case library, initiating batch tasks, comparing results, downloading reports, and archiving them.

[0110] Application Iteration Workbench: Provides an application list, version status, dependencies, integration testing entry, deployment evaluation, optimization suggestions, and template release entry.

[0111] Audit and Retrospective Workbench: Provides query and playback entry points by site / version / use case / application / deployment batch, supporting rapid location of evidence chain and retrospective reproduction.

[0112] (ii) Intelligent agent orchestration and process governance layer (the central layer below the portal).

[0113] Central intelligent agent (orchestrator): Receives natural language or structured requests, generates execution plans, and calls lower-level tools to complete tasks such as synchronization, orchestration, deduction, verification, evaluation, joint debugging, configuration draft generation, and report archiving.

[0114] Workflow / Orchestration Engine: Decomposes a task into controllable process nodes (e.g., first synchronize, then model, then extrapolate, then evaluate), and supports conditional branching, failure rollback, and retry strategies.

[0115] Tool Interface Gateway: Encapsulates lower-level capabilities into callable tools, providing input validation, output structuring, call quotas, concurrency limits, double confirmation for sensitive operations, and audit logging.

[0116] Strategy and rule scheduler: Selects appropriate inference routes for agents or human tasks, verifies rule sets, indicator versions and comparison baselines, and uniformly manages degradation strategies (e.g., reverting to the statistical / similar retrieval route when the model is unavailable).

[0117] Evidence Chain Assembler: Combines key inputs, data windows, hit condition segments, rule trigger records, indicator calculation methods, and output conclusions into a traceable evidence chain for use in review, audit, and debriefing.

[0118] (III) Application Iteration and Delivery Layer (the business layer that undertakes the entire lifecycle before and after the launch).

[0119] Application ledger and lifecycle management module: records application identifier, version, dependency points and dependent services, applicable working condition boundaries, running status, change records, rollback policies, evaluation summary, etc.

[0120] Dependency resolution and point mapping module: parses the measurement point / event / service dependencies declared by the application, and completes point mapping suggestions, unit dimension verification, sampling granularity matching checks, and missing item prompts.

[0121] Standard Interfaces and Adaptation Modules: Provides unified data interfaces, suggested output interfaces, event interfaces, and interface embedding specifications, and performs interface consistency and permission boundary verification for new applications.

[0122] Configuration generation and difference review module: Generates site-specific configuration packages from templates and mapping results, and automatically generates difference summaries, scope of impact, risk warnings and rollback points.

[0123] Integration Sandbox Module: Runs applications in an offline environment, providing simulated input, exception injection, and interface integration capabilities, and outputs integration logs, conflict lists, and repair suggestions.

[0124] The online evaluation and optimization suggestion module supports pre-launch evaluation and post-launch continuous evaluation, generates indicator comparisons, attribution clues and optimization suggestions, and transforms the suggestions into the next round of iteration tasks.

[0125] Template repository and replication / promotion module: Mature applications are encapsulated into template packages, including parameterized configuration, dependency declarations, applicable boundaries and regression test cases, supporting cross-site migration verification and regression validation.

[0126] (iv) Deduction, verification and experimental execution layer (the core capability layer that undertakes simulation, verification and evaluation).

[0127] The job condition orchestration and test case management module converts user-input or agent-generated job condition scripts into standard test cases, and manages test case versions, execution boundaries, indicator sets, and pass / fail criteria.

[0128] Simulation and deduction executor module: responsible for time progression, variable updates, disturbance injection and parallel task execution, outputting process variable trajectories, event sequences and running logs.

[0129] Data-driven inference model module: Provides statistical regularity models, data models and combined inference capabilities, and supports model version management and data version binding.

[0130] Similar working conditions retrieval and replay module: provides feature construction, similarity retrieval, fragment replay and mapping transformation, preserves the real dynamic form and outputs hit evidence.

[0131] Rules and Constraints Verification Module: Provides empirical rules, safety boundaries and feasibility constraints, and performs boundary violation judgment, risk classification and correction annotation on the simulated trajectory.

[0132] Security policy verification module: Import alarm rules, interlocking logic and emergency procedures to form an executable rule diagram / state machine / process network, support extreme condition script verification and output action sequence, response time and completeness conclusion.

[0133] Indicator Calculation and Report Generation Module: Based on a unified indicator library, this module calculates indicators such as output, energy consumption, fluctuation, alarm / interlock triggering, risk exposure, and stability, and generates comparison reports and scheme selection conclusions.

[0134] (v) Data foundation and consistency governance layer (the root layer that is like the site, reproducible, and traceable).

[0135] Unified Object Model Module: Incorporates measurement points, equipment, process units, rule entries, use cases, and application instances into a unified object system, and maintains object identification and mapping relationships.

[0136] Version management and difference detection module: Versioned storage of configuration snapshots, point tables, rules and models, supporting incremental patches, difference comparison, impact range analysis and rollback point generation.

[0137] Data Governance and Quality Identification Module: Completes time alignment, unit dimension consistency, missing / abnormal / drift identification, quality code transmission and source traceability.

[0138] The Operating Condition Database and Profile Database modules: These modules store typical operating condition segments, operating condition grouping rules, and statistical profiles, providing a unified reference for similarity retrieval, comparative evaluation, and indicator baselines.

[0139] Dependency Graph and Regression Plan Module: Maintains the dependency graph of object rule model use case applications, and automatically generates regression verification checklists and execution plans when changes occur.

[0140] (vi) Synchronization and Isolation Interface Layer (Boundary Layer between Offline Domain and Field Domain).

[0141] Data synchronization channel module: supports one-way or controlled read-only synchronization, supports periodic synchronization and on-demand synchronization, and supports tailoring the synchronization set by site / device / dependency scope.

[0142] Configuration and rule import module: Imports assets such as point tables, configuration parameters, alarm interlocks and emergency procedures into the offline domain in a structured manner, and completes format verification and consistency checks.

[0143] The data masking and permission labeling module: It masks sensitive fields and sensitive ranges according to rules, and adds permission labels to data and configurations to ensure offline availability and compliance.

[0144] Approval and Issuance Interface Module (Optional but Controlled): Only carries the material transmission of offline suggestions or configuration drafts into the approval chain, and does not provide the ability for direct connection and execution by intelligent agents.

[0145] (vii) On-site data source layer (bottom layer, actual production side).

[0146] Real-time data acquisition and control system: provides real-time measurement points, status variables, operating modes, alarm / interlock status, etc.

[0147] Historical database and data platform: Provides historical curves, statistical indicators, event logs, start and stop records, etc.

[0148] Business systems and ledger data: Provide equipment ledgers, maintenance records, operation records, process data, etc.

[0149] Rule configuration and emergency process sources: Provides assets such as alarm rules, interlocking logic, emergency processes and handling templates.

[0150] The on-site side is responsible for actual production and control execution, while the offline side is responsible for verification and evaluation. The two are separated by an isolation interface layer to achieve read-only synchronization / controlled approval, ensuring clear security boundaries.

[0151] Security and Access Control Module: Provides identity authentication, role-based access control, least privilege control, secondary confirmation of key actions, and full audit trail and playback capabilities.

[0152] Operation monitoring and maintenance support module: Provides task queue monitoring, resource quotas, service health checks, failure retry, degradation strategies and capacity warnings to ensure stable operation of batch testing and evaluation.

[0153] Please see Figure 2 The present invention provides an agent-based offline virtual laboratory and digital application iterative method, comprising: S1, Demand Triggering and Goal Definition.

[0154] Triggering sources include scientific research experiments, technical modification verification, strategy parameter tuning, safety verification, application access, and field change regression. In this module, the target site / device scope, target problem, expected output (report / recommendation / configuration draft / regression conclusion) and constraints (safety boundary, data range, time window) are determined.

[0155] S2, Version Selection and Synchronization Preparation.

[0156] Select the configuration version, point table version, rule version, model version, and indicator definition version used in this process, generate a synchronization list and synchronization strategy (periodic / on-demand, full / incremental, pruning scope), and establish the version binding relationship for this process to ensure reproducibility and auditability in the future.

[0157] S3, data and configuration synchronization and consistency verification.

[0158] Synchronize time-series data, event data, configuration snapshots, point table metadata, rule logic, and emergency procedures from the field domain; complete difference detection, patch generation, and consistency verification; update the offline environment version after verification; if missing or inconsistent data is found, output repair suggestions and return to this module to complete the missing data.

[0159] S4, data governance, object modeling, and operational profile updates.

[0160] Synchronous data is time-aligned, unit-scaled consistent, quality-identified, and source-traceable; object mapping and dependency graphs are updated; historical operation condition profiles and typical operation condition indexes are updated to provide a basis for subsequent similarity retrieval, baseline selection, and indicator comparison.

[0161] S5, operational condition arrangement and test case generation.

[0162] Users or intelligent agents input key parameters and strategies, generate operating condition scripts and transform them into standardized test cases; define the control baseline, evaluation index set and pass criteria; input test cases into the database and bind versions to form repeatable assets.

[0163] S6, Simulation Execution and Rule / Constraint Verification.

[0164] Select the simulation route based on the use case (data model / statistical regularity / replay of similar working conditions / combined route), advance the execution time and inject disturbances; call rules and check constraints during the process, mark risks and provide correction prompts for out-of-bounds, abnormal fluctuations and infeasible outputs, and generate trajectories, events and logs.

[0165] S7: Indicator calculation, scheme comparison and conclusion output.

[0166] Calculate indicators such as output, energy consumption, fluctuation, stability, alarm / interlock triggering, and risk exposure according to a unified indicator system; if multiple schemes are implemented in parallel, output horizontal comparison and recommendation conclusions, and clarify the recommendation premises, applicable boundaries and risk list, forming a standardized assessment report and evidence chain archive.

[0167] S8, security policy script verification and regression verification (can be performed in parallel or serially with module 6).

[0168] When safety verification is required, import and load alarm rules, interlocking logic and emergency procedures, construct extreme working conditions / failure scenario scripts and execute verification; output action sequence, response time, coverage and completeness conclusions; if defects are found, generate candidate improvement solutions and trigger regression test case sets for re-verification until the criteria are met or risk acceptance recommendations are formed.

[0169] S9, application integration, delivery, and online closed loop (end module).

[0170] If application iteration is involved, complete dependency resolution, point mapping verification and configuration package generation. After offline joint debugging sandbox verification, generate online materials for approval. After online, continuously evaluate the effect according to monitoring indicators and output optimization suggestions. When the effect is stable and meets the conditions, package it into a template package and enter the cross-site replication process. This completes the closed loop from requirements to delivery to replication and promotion.

[0171] In engineering implementation, the above technical solutions form a closed-loop system with offline virtual laboratories and iterative governance of digital applications. Through the tool-based orchestration of intelligent agents, the processes of work condition orchestration, simulation evaluation, script verification, dependency resolution, configuration generation, joint debugging assistance, evaluation optimization, and template migration are linked together as a standard process. At the same time, isolation, permission, auditing, and rollback mechanisms ensure that the real-time control execution link on site is not touched, thereby achieving low-risk, low-cost, and replicable iteration and promotion of digital applications under the premise of safety and controllability.

[0172] In summary, this invention constructs an offline virtual experimental environment in a non-production network or independent computing resources. Through network isolation, permission isolation, and interface boundary control, it ensures that the environment can be used for simulation, verification, evaluation, and joint debugging, but does not connect to the real-time control execution link on site. From an engineering architecture perspective, this achieves the safe and controllable characteristics of an environment that closely resembles the field and experiments that do not disturb the field.

[0173] This invention establishes a synchronization mechanism that combines periodicity and on-demand. The synchronization objects not only include historical time-series data, but also cover configuration parameters, point table structure, typical working condition segments, historical statistical profiles, and application-related version and dependency information. Furthermore, the synchronization results are managed in a versioned manner, with difference detection and consistency verification, so that the offline environment can maintain its representativeness of the field over a long period of time without drifting over time.

[0174] This invention performs unified encoding, time alignment, and unit dimension consistency processing on data and configuration assets entering the offline domain, and performs quality identification and source traceability marking for missing, abnormal, delayed, and drifting conditions, forming a calculable, reproducible, and auditable data foundation, providing consistent input for subsequent deduction, verification, and evaluation.

[0175] In the absence of a complete or incomplete mechanistic model, this invention uses a combination of data models, statistical laws, empirical rules or rule bases, similar working condition retrieval and replay, and constraint verification to complete the working condition simulation. It outputs the main process variables and energy consumption change results related to key production parameters, and explicitly labels uncertainties and boundary constraints to form a simulation route that can be implemented and continuously enhanced.

[0176] This invention supports custom settings for key production parameters and operation strategies in an offline virtual experiment interface or through an interface. It can organize various working conditions such as continuous changes, step disturbances, start-stop strategies and fault injection, and automatically condense each test into a repeatable working condition test case or script, so that scientific research experiments, technical transformation verification and regression testing have traceability and reproducibility.

[0177] This invention imports interlocking logic, alarm rules, and emergency procedures into an offline environment in a structured manner, and constructs a script simulation verification mechanism based on time progression and event triggering. It can conduct offline verification for high-risk scenarios such as extreme liquid levels, extreme pressures, equipment failures, and frequent start-stops, and output results such as the sequence of protection actions, response time, coverage, and process completeness. It replaces or significantly reduces on-site hazardous tests in an offline manner.

[0178] This invention focuses on new processes, new equipment parameters, or new control strategies, supports parallel simulation and comparative evaluation of multiple schemes, and automatically summarizes and compares outputs of output, energy consumption, fluctuations, alarm / interlock triggering, risk exposure, and stability according to a unified indicator system, forming a standardized basis for scheme optimization and technical decision-making.

[0179] This invention establishes a list of digital applications and a lifecycle ledger, uniformly recording information such as application version, dependent points, applicable working conditions, operating status and interface constraints, and reserving standardized data interfaces, control interfaces and interface embedding specifications, enabling new applications to be quickly connected according to the specifications. At the same time, the system automatically completes dependency checks, point mapping verification, configuration generation and offline debugging assistance, reducing the cost of repeated docking and the risk of going online.

[0180] This invention provides a quantifiable comparative evaluation of digital applications before and after their deployment. After deployment, it continuously summarizes key indicators and automatically generates evaluation reports and optimization suggestions. Furthermore, it encapsulates stable and qualified applications into parameterizable templates. Through point mapping, parameter migration, and working condition adaptation checks, it enables rapid cross-site replication and promotion, forming an engineering closed loop of deployment evaluation, optimization, and replication.

[0181] The embodiments of the present invention described above do not constitute a limitation on the scope of protection of the present invention.

Claims

1. An offline virtual laboratory and digital application iterative system based on intelligent agents, characterized in that, include: The synchronization and isolation interface layer is used to obtain real-time data, historical data, system configuration, rule logic and business ledger from the field data source layer, and to perform de-identification and permission labeling on the obtained data to form data and configuration resources that can be safely used in offline environments. The data foundation and consistency governance layer connect to the synchronization and isolation interface layer. It is used to receive and store the data and configuration resources, and to perform version management, difference detection and data governance on the data and configuration resources based on the unified object model, so as to generate and maintain a consistent data foundation containing versioned data, configuration, rules, working condition profiles and dependency graphs. The simulation verification and experiment execution layer connects the data base and the consistency governance layer. It is used to arrange experimental test cases based on the working condition profile in the consistent data base, and call the simulation simulation executor. It uses data-driven simulation model, similar working condition retrieval and rule constraint verification to perform simulation simulation and verification of digital intelligent applications and strategies, including security policies, in an offline environment, and generate indicator calculation results and verification reports. The application iteration and delivery layer connects the data foundation, the consistency governance layer, and the simulation verification and experiment execution layer. It is used to manage the entire lifecycle of digital applications, perform application dependency parsing and point mapping based on the dependency graph in the consistent data foundation, configure and integrate applications through standard interfaces in the joint debugging sandbox, and generate application launch evaluation and optimization suggestions based on the indicator calculation results and verification reports generated by the simulation verification and experiment execution layer. The intelligent agent orchestration and process governance layer connects the synchronization and isolation interface layer, the data foundation and consistency governance layer, the inference verification and experiment execution layer, and the application iteration and delivery layer. It is used to automatically orchestrate and schedule the entire process of data synchronization, consistency governance, experiment orchestration, inference verification, application integration and evaluation delivery through intelligent agents, and to perform process governance based on the unified object model and the dependency graph. A unified portal and role workbench connect to the agent orchestration and process governance layer, providing a human-computer interaction interface to drive and monitor the automated processes of the agent orchestration and process governance layer. Through the approval and distribution interface of the synchronization and isolation interface layer, applications or policies that have been verified offline are securely distributed to the field after approval.

2. The agent-based offline virtual laboratory and digital application iterative system as described in claim 1, characterized in that, The synchronization and isolation interface layer is used for: Establish a synchronization strategy that combines periodic synchronization and on-demand synchronization. The synchronization objects include historical time-series data, system configuration parameters, point table structure, alarm rules, interlocking logic, and emergency procedures. The synchronized data is uniformly encoded, time-aligned, and difference-detected, and a version number and checksum are generated for each synchronization. Sensitive data is anonymized and access rights are assigned.

3. The agent-based offline virtual laboratory and digital application iterative system as described in claim 1, characterized in that, The data foundation and consistency governance layer include: The unified object model module is used to assign offline object identifiers to measurement points, devices, rule entries, and application instances; The version management and difference detection module is used to store configuration snapshots, point tables and rules in a versioned manner and compare differences. The dependency graph module is used to maintain the dependencies between objects and generate regression verification plans when changes occur.

4. The agent-based offline virtual laboratory and digital application iterative system as described in claim 1, characterized in that, The deduction verification and experiment execution layer includes: The data-driven simulation model module uses a combination of data models, statistical laws, and similar working conditions retrieval and replay to perform process simulation. The rules and constraints verification module is used to verify the feasibility of the simulation results and identify risks. The security policy verification module is used to perform scripted verification of alarm rules and interlocking logic.

5. The agent-based offline virtual laboratory and digital application iterative system as described in claim 1, characterized in that, The application iteration and delivery layer includes: The application ledger and lifecycle management module is used to record application versions, dependency points, and running status; The dependency resolution and point mapping module is used to resolve application dependencies and perform point mapping verification. The integration sandbox module is used to perform interface integration testing and compatibility verification of applications in an offline environment.

6. The agent-based offline virtual laboratory and digital application iterative system as described in claim 1, characterized in that, The agents in the agent orchestration and process governance layer are used for: The tool interface layer calls data query, work condition arrangement, and simulation execution tools. Generate a task plan that includes data version, configuration version, and model version; When outputting conclusions, bind evidence chain elements such as data windows, key curves, and rule trigger records.

7. The agent-based offline virtual laboratory and digital application iterative system as described in claim 1, characterized in that, The system also includes: The security permissions and auditing module is used to provide identity authentication, role-based access control, and full-process auditing and traceability. The operation monitoring and maintenance module is used for task queue monitoring, resource quota management, and service health checks.

8. The agent-based offline virtual laboratory and digital application iterative system as described in claim 1, characterized in that, The deduction verification and experiment execution layer is also used for: Establish a unified indicator system, which should at least include indicators related to output, energy consumption, volatility, and alarm triggering. Comparative analysis of the results of parallel simulations of multiple scenarios is conducted to generate an assessment report that includes a comparison of differences and a risk list.

9. The agent-based offline virtual laboratory and digital application iterative system as described in claim 1, characterized in that, The application iteration and delivery layer is also used for: Encapsulate mature applications into template packages that include parameterized configurations and dependency declarations; Perform site mapping verification and operating condition adaptation checks during cross-site migration; Complete regression testing before the new site goes live.

10. An iterative method for offline virtual laboratories and digital applications based on intelligent agents, characterized in that, Includes the following steps: Step 1, Demand Triggering and Target Definition: Receive trigger requests from scientific research experiments, technical modification verification, strategy parameter tuning, security verification, application access, or field change regression; determine the target site or device scope, target problem, expected output type, and constraints; the expected output type includes reports, suggestions, configuration drafts, or regression conclusions; the constraints include security boundaries, data range, and time windows. Step 2, Version Selection and Synchronization Preparation: Based on the objectives determined in Step 1, select the configuration version, point table version, rule version, model version, and indicator caliber version, and generate a synchronization list and synchronization strategy; the synchronization strategy includes periodic synchronization or on-demand synchronization, full synchronization or incremental synchronization, and synchronization scope pruning, and establish the version binding relationship for this process. Step 3, Data and Configuration Synchronization and Consistency Verification: Based on the synchronization list and synchronization strategy generated in Step 2, synchronize time-series data, event data, configuration snapshots, point table metadata, rule logic, and emergency process assets from the field domain; Perform difference detection, patch generation, and consistency verification on the synchronization results. Once the verification is passed, update the offline environment version. If any missing or inconsistent information is found, a repair suggestion will be output and the process will be repeated until the missing information is found. Step 4, Data Governance, Object Modeling and Working Condition Profile Update: Perform time alignment, unit dimension consistency processing, quality identification and source traceability on the data synchronized and verified in Step 3; update object mapping relationships and dependency relationship graphs; Update the operating condition profile and typical operating condition index based on historical operating data; Step 5, Operational Condition Orchestration and Test Case Generation: Receive key parameters and strategies input by the user or agent, generate operational condition scripts and convert them into standardized test cases; define the baseline, evaluation index set, and pass criteria; store the test cases in the test case library and bind version information. Step Six, Simulation Execution and Rule or Constraint Verification: Based on the test cases generated in Step Five, select at least one simulation route from the data model, statistical regularity, similar working condition replay, or combination route, and execute time progression and disturbance injection; during the simulation, call the rule and constraint verification, mark risks and provide correction prompts for out-of-bounds behavior, abnormal fluctuations, and infeasible outputs, and generate process variable trajectories, event sequences, and operation logs; Step 7, Indicator Calculation, Scheme Comparison and Conclusion Output: Based on the simulation results generated in Step 6, calculate output, energy consumption, volatility, stability, alarm or interlock trigger times and risk exposure indicators according to the unified indicator system; If multiple scenarios are simulated in parallel, output the horizontal comparison results and recommended conclusions, and clarify the recommended premises, applicable boundaries and risk list, forming a standardized assessment report and evidence chain for archiving; Step 8, Security Policy Script Verification and Regression Verification: When security verification is required, import and load alarm rules, interlocking logic and emergency procedures, construct extreme condition or fault scenario scripts and execute verification; output protection action sequence, response time, coverage and completeness conclusions; If a defect is found, candidate improvement solutions are generated and the regression test case set is re-validated until the criteria are met or a risk acceptance recommendation is formed. Step 9, Application Integration, Delivery and Launch Closed Loop: When application iteration is involved, complete dependency resolution, point mapping verification and configuration package generation. After offline integration sandbox verification is passed, generate launch materials and enter the approval process. After launch, the application's effectiveness will be continuously evaluated based on monitoring metrics, and optimization suggestions will be provided. When the application performs stably and meets the preset conditions, it is packaged into a template package and the cross-site replication process is initiated.

Images