Cloud edge-end cooperative intelligent control method and platform for chemical water treatment black light factory

Abstract

The invention relates to the technical field of chemical water treatment, in particular to a cloud side-end cooperative intelligent control method and platform for a chemical water treatment black light factory, and the method comprises the steps: constructing a digital twin model fusing equipment topology and operation parameters, and simulating and optimizing the optimal operation state of a system in an unattended and multi-unit cooperative scene; the edge sensing terminal is used for collecting water quality, energy consumption and equipment operation parameters of each process unit under dynamic working condition disturbance in real time, and the actual operation state of the system is accurately recognized through feature extraction and data fusion analysis; by comparing actual and optimal operation states, extracting deviation parameters, performing multi-dimensional analysis, identifying an operation instability mode and key influence factors, generating a self-adaptive cloud side-end cooperative control strategy and driving an intelligent execution device to realize closed-loop regulation, the response capability, the operation stability and the energy efficiency level of the system to complex disturbance are improved, and the energy efficiency of the system is improved. And unmanned, intelligent and collaborative optimization operation of the chemical water treatment process is effectively supported.

Description

A cloud-edge-device collaborative intelligent control method and platform for a lights-out water treatment plant Technical Field

[0001] This invention relates to the field of water treatment technology, and in particular to a cloud-edge-device collaborative intelligent control method and platform for a light-out water treatment plant. Background Technology

[0002] With the deep integration of industrial automation and information technology, the water treatment industry is rapidly developing towards intelligent and unmanned operation. As a key supporting facility in water-intensive industries such as power, chemical, and metallurgy, the operational stability, energy efficiency, and effluent quality of water treatment systems directly affect the safety and environmental compliance of the entire production system. Traditional water treatment systems rely heavily on manual experience for operation and control, resulting in problems such as slow response, poor coordination, high energy consumption, and weak fault warning capabilities, making it difficult to meet the comprehensive demands of modern industry for efficient, low-carbon, and reliable operation. Especially in the context of building "lights-out factories" (i.e., unattended, fully automated smart factories), achieving autonomous perception, intelligent decision-making, and collaborative control throughout the entire process of water treatment systems has become a significant challenge for the industry's technological upgrade.

[0003] In recent years, the development of cloud computing, edge computing, and the Internet of Things (IoT) has provided new technological pathways for the intelligentization of water treatment systems. The cloud-edge-device collaborative architecture organically combines the powerful data storage and model computing capabilities of the cloud, the real-time sensing and rapid response capabilities of the edge, and the precise execution capabilities of terminal devices, providing a feasible solution for the intelligent control of complex industrial processes. Meanwhile, digital twin technology, by constructing a virtual mapping of the physical system, can simulate, predict, and optimize the process in virtual space, providing decision support for system trial-and-error adjustments. However, existing technologies still generally suffer from problems such as a disconnect between models and actual operation, insufficient utilization of sensing data, and poor adaptability of control strategies. Especially when facing dynamic operating disturbances such as fluctuations in influent water quality, equipment aging, and multi-unit coupling, it is difficult to achieve a balance between global optimization and rapid response.

[0004] The above content is only used to help understand the technical solution of the present invention and does not represent an admission that the above content is prior art. Summary of the Invention

[0005] The main objective of this invention is to provide a cloud-edge-device collaborative intelligent control method and platform for a lights-out water treatment plant. This aims to solve the technical problems of existing water treatment systems, which are unable to achieve accurate perception of operating status, closed-loop collaboration between models and reality, and multi-unit adaptive optimization control under dynamic operating conditions, thus failing to meet the comprehensive needs of lights-out plants for unattended operation, high efficiency, stability, and intelligent decision-making.

[0006] To achieve the above objectives, this invention provides a cloud-edge-device collaborative intelligent control method for a lights-out water treatment plant. The method includes: constructing a digital twin model based on the equipment topology and operating parameters of the water treatment process; determining the optimal operating state of the lights-out plant under unattended and multi-unit collaborative scenarios based on the digital twin model; collecting process parameter data of each process unit under dynamic operating condition disturbances based on deployed edge sensing terminals; performing feature extraction and fusion analysis on the process parameter data to determine the actual operating state of the water treatment system; determining the deviation parameters between the actual operating state and the optimal operating state; performing multi-dimensional analysis on the deviation parameters to generate a cloud-edge-device collaborative control strategy; and performing adaptive collaborative control of intelligent actuators based on the cloud-edge-device collaborative control strategy.

[0007] Optionally, the construction of a digital twin model based on the equipment topology and operating parameters of the water treatment process includes: determining the calling dimensions of the equipment topology and operating parameters of the water treatment system based on digital twin modeling specifications, and retrieving process flow diagrams, equipment characteristic parameters, and historical operating data from the cloud platform database based on the calling dimensions; classifying the equipment topology data and operating parameters hierarchically based on functional requirements to obtain system network topology parameters and unit dynamic characteristic parameters; constructing a process flow diagram of the water treatment system based on the network topology parameters, and labeling the dynamic response attributes of each treatment unit and connecting pipeline in the flow diagram based on the unit dynamic characteristic parameters; simultaneously, obtaining the target application scenario of the digital twin model based on the cloud platform, and determining the accuracy grading standard of the digital twin model based on the target application scenario; and labeling the regional functions of the digital twin model based on the accuracy grading standard to obtain a digital twin model covering the pretreatment unit, membrane separation unit, and water quality monitoring unit.

[0008] Optionally, determining the optimal operating state of the "lights-out" factory under unattended and multi-unit collaborative scenarios based on the digital twin model includes: determining the boundary constraints of the water treatment system based on the digital twin model, and simultaneously determining the dynamic operating condition disturbance types based on process characteristics and environmental standards; determining disturbance injection nodes based on the dynamic operating condition disturbance types and boundary constraints, and determining the optimal operating range based on the disturbance injection nodes and unit dynamic characteristics; injecting multiple types of dynamic operating condition disturbances into the pretreatment unit, membrane separation unit, and water quality monitoring unit in the digital twin model based on the optimal operating range; selecting key monitoring nodes in the digital twin model based on the control target, and tracking the injection of multiple types of dynamic operating condition disturbances throughout the entire process based on the key monitoring nodes and the digital twin model; determining the theoretical thresholds of water quality parameters, energy consumption parameters, and equipment status parameters at each key monitoring node under multiple types of dynamic operating condition disturbances based on the results of the entire process tracking; and obtaining the optimal operating state of the "lights-out" factory under unattended and multi-unit collaborative scenarios based on the theoretical thresholds.

[0009] Optionally, after obtaining the optimal operating state of the dark factory in unattended and multi-unit collaborative scenarios based on theoretical thresholds, the method further includes: acquiring the obtained optimal operating state and establishing a dynamic mapping relationship between dynamic operating condition disturbance types and the optimal operating state; binding the dynamic operating condition disturbance types with the optimal operating state based on the dynamic mapping relationship, and filling records of the dynamic operating condition disturbance types and the optimal operating state in the cloud-edge collaborative database in real time based on the state binding results; and constructing a cloud-edge-device collaborative control decision knowledge base based on the real-time filling records.

[0010] Optionally, before the deployed edge sensing terminal collects process parameter data of each process unit under dynamic operating condition disturbances, the process includes: acquiring the equipment topology of the water treatment system based on the cloud platform, and identifying key processing units and parameter-sensitive areas in the system based on the topology; using the key processing units and parameter-sensitive areas as a first set of sensing points; determining the environmental disturbance factors of the water treatment system based on process specifications, and analyzing the influence path of environmental disturbance factors on process parameters in conjunction with the equipment topology; using the nodes on the influence path of environmental disturbance factors on process parameters as a second set of sensing points; and optimizing and guiding the deployment of the edge sensing terminal based on the first and second sets of sensing points.

[0011] Optionally, the process parameter data of each process unit under dynamic operating condition disturbances is collected based on the deployed edge sensing terminals, including: monitoring the operating condition disturbance of the water treatment system for a target period and identifying the time-frequency characteristics of the dynamic operating condition disturbance based on the monitoring results; determining the data sampling rate configuration range of the edge sensing terminals based on the time-frequency characteristics and adaptively configuring the parameters of the edge sensing terminals based on the configuration range; driving the edge sensing terminals to perform wide-area synchronous monitoring of the system based on the parameter configuration results and collecting water quality parameters, equipment operating parameters and energy consumption parameters of each process unit based on the monitoring results; and aggregating the collected process parameter data to the cloud platform in real time based on the 5G low-latency communication link pre-established between each edge sensing terminal and the cloud platform.

[0012] Optionally, the step of performing feature extraction and fusion analysis on process parameter data to determine the actual operating status of the water treatment system includes: acquiring collected process parameter data and performing spatiotemporal alignment of the data on a digital twin model; dynamically rendering water quality fluctuations and energy consumption changes at each sensing point on the digital twin model based on the spatiotemporal alignment results, and identifying material flow characteristics and energy coupling modes between different units in the system based on the rendering results; analyzing the coupling relationships between units based on the material flow characteristics and energy coupling modes, and dynamically compensating and correcting the process parameter data at each sensing point based on the coupling relationships; and aggregating and statistically analyzing the actual water quality, energy consumption, and equipment status data at each sensing point based on the correction results to generate the actual operating status of the water treatment system.

[0013] Optionally, determining the deviation parameter between the actual operating state and the optimal operating state, and performing multi-dimensional analysis on the deviation parameter to generate a cloud-edge-device collaborative control strategy, includes: acquiring the actual operating state of the water treatment system, and quantitatively comparing the actual operating state with the corresponding optimal operating state to obtain dynamic deviation parameters; when the amplitude of the dynamic deviation parameter is within a preset safety threshold, determining that the actual operating state of the system is in a stable range, and continuously tracking disturbances in the system based on the determination result; otherwise, determining that the actual operating state of the system exceeds the stable range; dividing the water treatment system into cloud-edge collaborative control areas based on the determination result and the deployment location of the edge sensing terminals, and performing spatiotemporal correlation mining on the process parameter data at each sensing point based on the area division result; and quantifying the material loss and energy of each area based on the spatiotemporal correlation mining result. The system measures water loss and performs dynamic pattern recognition on the loss measures to determine the system's operational instability modes, including water quality exceeding standards, abnormal energy consumption, and equipment failure. Based on the loss measures under these instability modes, the system identifies potential faulty units in key equipment and performs intelligent diagnosis on these units. After confirming that no abnormalities exist in the potential faulty units, the cloud-edge-device collaborative control mechanism is activated. Based on the activation result, a pre-trained water quality-energy consumption coupling model is retrieved from the decision knowledge base, and the effluent water quality compliance rate is used as the core control indicator. The model is used to assess the safety boundary of process parameter data at each sensing point. Based on the safety boundary assessment results, the control quantities corresponding to adjusting the water quality, energy consumption, and equipment status at each sensing point to the optimal operating range are determined. The control quantities of each region are optimized and integrated to generate a cloud-edge-device collaborative control strategy.

[0014] Optionally, the adaptive collaborative control of intelligent actuators based on the cloud-edge-device collaborative control strategy includes: acquiring the cloud-edge-device collaborative control strategy and decoding the strategy to obtain a collaborative control instruction set for each intelligent actuator; applying dynamic amplitude limiting constraints to the execution units of each intelligent actuator based on the collaborative control instruction set, and arranging the timing logic of each execution unit based on the constraint results to generate a collaborative control sequence for the intelligent actuators; and performing closed-loop regulation of each intelligent actuator based on the collaborative control sequence to achieve unattended operation and real-time optimization of multi-unit collaboration in a light-out water treatment plant.

[0015] Furthermore, to achieve the above objectives, the present invention also provides a cloud-edge-device collaborative intelligent control platform for a water treatment darkroom, the platform comprising: a memory, a processor, and a cloud-edge-device collaborative intelligent control program for a water treatment darkroom stored in the memory and executable on the processor, the cloud-edge-device collaborative intelligent control program for a water treatment darkroom configured to implement the steps of the cloud-edge-device collaborative intelligent control method for a water treatment darkroom as described above.

[0016] This invention provides a cloud-edge-device collaborative intelligent control method for a lights-out water treatment plant. The method constructs a digital twin model integrating equipment topology and operating parameters, enabling accurate prediction and simulation optimization of the optimal operating state of the water treatment system in unattended and multi-unit collaborative scenarios, thus enhancing the system's forward-looking and scientific nature. By deploying edge sensing terminals to collect multi-source process parameter data under dynamic operating disturbances in real time, and combining feature extraction and fusion analysis techniques, it achieves high-precision dynamic perception and state identification of the system's actual operating state. Furthermore, by performing multi-dimensional analysis of the deviation between the actual and optimal operating states, an adaptive cloud-edge-device collaborative control strategy is generated, driving intelligent actuators to achieve closed-loop collaborative regulation, significantly enhancing the system's response capability and control robustness to complex disturbances. Overall, this method achieves a deep integration of model-driven and data-driven approaches, constructing a full-link intelligent control closed loop from state perception and intelligent decision-making to precise execution. It effectively improves the automation, intelligence, and energy efficiency optimization level of the lights-out water treatment plant, reduces reliance on manual labor and operational risks, and has good application and promotion value and engineering practical prospects. Attached Figure Description

[0017] Figure 1 is a flowchart illustrating an embodiment of the cloud-edge-device collaborative intelligent control method for a light-out water treatment plant according to the present invention.

[0018] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0019] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0020] Referring to Figure 1, which is a flowchart illustrating an embodiment of the cloud-edge-device collaborative intelligent control method for a light-out water treatment plant according to the present invention, an embodiment of the cloud-edge-device collaborative intelligent control method for a light-out water treatment plant according to the present invention is proposed.

[0021] In one embodiment, the cloud-edge-device collaborative intelligent control method for the water treatment "lights-out" factory includes: step S100, constructing a digital twin model based on the equipment topology and operating parameters of the water treatment process, and determining the optimal operating state of the "lights-out" factory in unattended and multi-unit collaborative scenarios based on the digital twin model.

[0022] The digital twin model can be a virtual mapping system that integrates the topology and operating parameters of water treatment process equipment. It is used to reproduce the dynamic behavior of the physical system in digital space and can be used to simulate, predict, and optimize the optimal operating state of a "lights-out" factory under unattended and multi-unit collaborative scenarios. In this embodiment, the digital twin model can be a hybrid of a mechanistic model and a data-driven model built based on equipment, process logic, and historical / real-time operating parameters. For example, the digital twin model can include, but is not limited to, one or more of the following: mechanistic-driven digital twin, data-driven digital twin, and hybrid-driven digital twin.

[0023] Building a digital twin model based on the equipment topology and operating parameters of a water treatment process can integrate equipment, process logic, and operating parameters to establish a virtual mapping model. Furthermore, this operation can be achieved by using graph neural networks to model the equipment topology relationships and combining them with physical equations to describe unit behavior, or by using historical operating data to train an LSTM or Transformer model and embedding process constraints. This allows for high-fidelity prediction and simulation optimization of the system's optimal operating state.

[0024] Step S200: Based on the deployed edge sensing terminals, process parameter data of each process unit under dynamic operating condition disturbances are collected, and feature extraction and fusion analysis are performed on the process parameter data to determine the actual operating status of the water treatment system.

[0025] The edge sensing terminal can be a sensing and data acquisition device with local computing and communication capabilities deployed at various process units in water treatment. It can be used to collect multi-source process parameter data in real time under dynamic operating conditions, supporting low-latency status perception. In an exemplary embodiment, the edge sensing terminal can be integrated with an edge computing module through embedded sensors to realize data acquisition, preprocessing, and uploading. For example, the edge sensing terminal can be one or more of the following: online water quality monitoring terminal, equipment status sensing terminal, flow and pressure sensing terminal, etc.

[0026] Process parameter data can be a set of multidimensional real-time measurements reflecting the operating status of each process unit in a water treatment system, and can be used as the basic input for identifying the actual operating status of the system. For example, process parameter data may include, but is not limited to, one or more of the following: influent water quality parameters, membrane flux parameters, and regenerant consumption parameters. Feature extraction and fusion analysis of process parameter data to determine the actual operating status of the water treatment system can be performed by extracting key features from multi-source heterogeneous data and performing spatiotemporal fusion to form a unified state representation. Furthermore, this operation can be achieved by using wavelet transform and time-frequency analysis to extract dynamic features and fusing multi-sensor data through an attention mechanism, or by using principal component analysis for dimensionality reduction and inputting the data into a clustering model to identify the current operating mode, thereby achieving high-precision, low-latency identification of the actual operating status of the system.

[0027] Step S300: Determine the deviation parameters between the actual operating state and the optimal operating state, perform multi-dimensional analysis on the deviation parameters, generate a cloud-edge-device collaborative control strategy, and perform adaptive collaborative control of the intelligent actuator based on the cloud-edge-device collaborative control strategy.

[0028] The deviation parameter can be a quantitative indicator of the difference between the actual operating state and the optimal operating state predicted by the digital twin model. It can be used as the basis for generating control strategies, characterizing the degree and direction of the system's deviation from ideal operating conditions. For example, the deviation parameter may include, but is not limited to, one or more of water quality deviation parameters, energy consumption deviation parameters, and equipment coordination deviation parameters. The cloud-edge-device collaborative control strategy can be an adaptive control instruction set composed of global optimization in the cloud, local response at the edge, and execution instructions at the terminal. It can be used to drive intelligent actuators to achieve rapid and robust closed-loop control of complex disturbances. In a specific embodiment, the cloud-edge-device collaborative control strategy can generate optimization targets on the cloud side, generate local actions on the edge side, and execute specific adjustments on the terminal side based on the multi-dimensional analysis results of the deviation parameter. For example, the cloud-edge-device collaborative control strategy may include, but is not limited to, one or more of cloud scheduling strategies, edge response strategies, and terminal execution strategies.

[0029] Intelligent actuators can be process control equipment with automatic adjustment capabilities, capable of receiving and executing control commands, and used to precisely regulate key components such as valves, pumps, and dosing systems. For example, intelligent actuators can include, but are not limited to, one or more of intelligent regulating valves, variable frequency pumps, and automatic dosing devices. Multi-dimensional analysis of deviation parameters generates a cloud-edge-device collaborative control strategy. This can involve analyzing the causes of deviations from multiple dimensions, such as water quality, energy consumption, and equipment coordination, to generate hierarchical control commands. Furthermore, this operation can generate long-term cloud optimization strategies and immediate edge response strategies through a combination of rule engines and reinforcement learning, or by using multi-objective optimization algorithms to solve the Pareto front and allocate control weights and actions to the cloud, edge, and device, thereby forming an adaptive collaborative control strategy and improving system robustness. Adaptive collaborative control of intelligent actuators based on the cloud-edge-device collaborative control strategy can involve issuing hierarchical control commands to the corresponding execution units to achieve closed-loop regulation. Furthermore, this operation can be achieved by sending scheduling instructions from the cloud to edge nodes, which then decompose the instructions into device-level instructions and drive the execution device, or by edge nodes adjusting execution parameters in real time according to local disturbances and simultaneously feeding back the execution results to the cloud to update the model. This can complete a precise closed loop from decision-making to execution and effectively cope with complex disturbances.

[0030] Taking the regulation of a reverse osmosis system under sudden changes in influent water quality as an example, the cloud-edge-end collaborative intelligent control method of the water treatment "lights-out" plant in this embodiment can be as follows: When the turbidity of the raw water rises sharply, the edge sensing terminal captures changes in parameters such as influent SDI, pressure, and flow rate in real time, and identifies the system as being in an abnormal operating condition through feature fusion; the digital twin model predicts the ideal dosage and membrane cleaning cycle based on the current equipment topology and historical best operating data; the deviation parameter shows that the actual dosage is insufficient and the pressure difference rises too quickly; the cloud-edge-end collaborative control strategy is generated: the cloud adjusts the overall dosing plan, the edge node immediately increases the frequency of scale inhibitor dosing, and the metering pump and regulating valve in the intelligent actuator act synchronously to achieve rapid steady-state recovery.

[0031] In one embodiment, a digital twin model is constructed based on the equipment topology and operating parameters of the water treatment process. This includes: determining the calling dimensions of the equipment topology and operating parameters of the water treatment system based on digital twin modeling specifications, and retrieving process flow diagrams, equipment characteristic parameters, and historical operating data from a cloud platform database based on these calling dimensions. Specifically, determining the calling dimensions of the equipment topology and operating parameters based on digital twin modeling specifications can be done by defining the dimensions (such as time granularity, spatial granularity, and parameter type) for extracting equipment topology and operating parameters from the cloud platform database according to standardized modeling specifications. Furthermore, this operation can be achieved by defining parameter calling interfaces and metadata structures using industrial digital twin standards such as ISO 23247, or by defining the semantic dimensions of equipment attributes and operating parameters based on a domain ontology model, thereby ensuring the integrity and consistency of the data required for model construction.

[0032] Based on functional requirements, equipment topology data and operating parameters are hierarchically categorized to obtain system network topology parameters and unit dynamic characteristic parameters. The unit dynamic characteristic parameters can be a set of time-varying parameters describing the dynamic response behavior of each process unit and connecting pipeline in water treatment under different operating conditions. These parameters can be used to label the dynamic response attributes of treatment units and pipelines in the digital twin model, enabling the model to reflect the time-varying characteristics of the real system. In this embodiment, unit dynamic characteristic parameters can be extracted from historical operating data in the cloud platform database, combined with equipment physical characteristics and control logic, and dynamic response features can be extracted through system identification or data-driven methods. For example, unit dynamic characteristic parameters may include, but are not limited to, one or more of the following: flow-pressure difference response parameters, water quality-flux coupling parameters, regeneration cycle-performance degradation parameters, etc. Furthermore, hierarchically categorizing equipment topology data and operating parameters based on functional requirements can involve structurally classifying the raw data according to two categories of requirements: system-level and unit-level dynamic behavior. In one exemplary embodiment, this operation can be achieved by using graph database storage devices as network topology parameters, time-series database storage unit input and output responses as dynamic characteristic parameters, or by using knowledge graph technology to extract entity-relationships from device attributes and operation logs and automatically classify them. This can separate static topology information from dynamic behavioral information and support modular construction of the model.

[0033] A process flow diagram of a water treatment system is constructed based on network topology parameters, and dynamic response attributes of each treatment unit and connecting pipeline in the flow diagram are labeled based on unit dynamic characteristic parameters. This labeling of dynamic response attributes based on unit dynamic characteristic parameters can be achieved by binding the extracted dynamic characteristic parameters to the corresponding unit and pipeline nodes in the digital twin flow diagram. In a specific embodiment, this operation can be achieved by embedding state-space equations or transfer functions into the flow diagram nodes to describe their dynamic characteristics, or by using configurable response templates (such as first-order inertia + delay) and injecting the identified parameters, thereby endowing the model with the ability to reflect the time-varying response of the real physical system.

[0034] Simultaneously, the target application scenarios of the digital twin model are obtained based on the cloud platform, and the accuracy grading standards for the digital twin model are determined based on these target application scenarios. These accuracy grading standards can be categorized according to the model fidelity and computational complexity set for the target application scenarios of the digital twin model. This can be used to match model complexity with control requirements, avoiding over-modeling or under-modeling. In this embodiment, the accuracy grading standards can be based on the differences in simulation accuracy, response speed, and computational resource requirements of the target application scenarios (such as fault diagnosis, energy efficiency optimization, and collaborative scheduling), formulating multi-level modeling specifications. For example, accuracy grading standards can include, but are not limited to, high-fidelity simulation level, real-time control level, and trend prediction level. Furthermore, determining the accuracy grading standards for the digital twin model based on the target application scenarios can be achieved by setting different modeling accuracy requirements according to the model's purpose (such as real-time control, long-term optimization, and fault early warning). In one exemplary embodiment, this operation can be achieved by setting a simplified model with low latency and medium accuracy for real-time control scenarios, setting a high-precision mechanistic model for offline optimization, or by mapping the scenario requirement matrix to the enable / disable rules and parameter update frequency of model components, thereby achieving a reasonable match between modeling resources and control objectives.

[0035] Based on the accuracy grading standard, the digital twin model is labeled with regional functions to obtain a digital twin model covering the pretreatment unit, membrane separation unit and water quality monitoring unit.

[0036] Regional functional labeling can be a method of structurally marking different regions in a digital twin model according to their process functions, which can be used to support subsequent multi-unit collaborative optimization and hierarchical control strategy generation. For example, regional functional labeling can include, but is not limited to, labeling of pretreatment functional areas, membrane separation functional areas, and water quality monitoring functional areas. Furthermore, regional functional labeling of the digital twin model based on accuracy grading standards can be achieved by structurally marking preset functional areas (such as pretreatment, membrane separation, and water quality monitoring) in the model and associating them with corresponding accuracy levels. In a specific embodiment, this operation can be achieved by labeling the functional category and accuracy level of each subsystem in the model metadata for use by edge or cloud schedulers, or by differentiating functional areas and accuracy levels in the visualization interface through color coding or layer isolation, thereby supporting subsequent regional and accuracy-based collaborative simulation and control strategy generation.

[0037] Taking the use of a digital twin model in a rapid response scenario to membrane fouling as an example, the cloud-edge-end collaborative intelligent control method of the water treatment "lights-out" plant in this embodiment can be as follows: When the system detects an abnormal increase in the pressure difference of the reverse osmosis unit, the digital twin model identifies the unit as belonging to the "membrane separation functional area" based on the regional functional labeling, and calls the labeled unit dynamic characteristic parameters (such as the coupling relationship between flux-pressure difference-recovery rate) according to its "real-time control level" accuracy classification standard to quickly simulate the recovery effect under different cleaning strategies; at the same time, since the pretreatment unit belongs to the "trend prediction level", only a simplified model is used to participate in the collaborative analysis, thereby maintaining global coordination while ensuring response speed.

[0038] In one embodiment, determining the optimal operating state of a lights-out factory under unattended and multi-unit collaborative scenarios based on a digital twin model includes: determining the boundary constraints of the water treatment system based on the digital twin model; and determining the types of dynamic operating condition disturbances based on process characteristics and environmental standards. The boundary constraints can be a set of system operation restrictions composed of equipment physical limits, process safety specifications, and environmental emission standards. These constraints can provide feasible domain boundaries for simulation and optimization in the digital twin model, ensuring the generated optimal operating state is engineering-feasible. In this embodiment, the boundary constraints may include, but are not limited to, one or more of equipment safety constraints, water quality discharge constraints, and energy consumption upper limit constraints. The types of dynamic operating condition disturbances can be classification identifiers reflecting non-steady-state external or internal disturbances that may occur during actual operation. These can be used to guide the selection of disturbance injection types, enabling the digital twin simulation to cover typical anomalies and changing scenarios. For example, dynamic operating condition disturbance types may include influent water quality abrupt changes, equipment performance degradation disturbances, and load fluctuation disturbances.

[0039] The disturbance injection nodes are determined based on the dynamic operating condition disturbance type and boundary constraints, and the optimal operating range is determined based on the disturbance injection nodes and the dynamic characteristics of the units. The disturbance injection nodes can be pre-defined process units or connection points in the digital twin model used to apply dynamic operating condition disturbances. They can serve as the starting point for disturbance simulation, controlling the propagation path and impact range of the disturbance in the system. Furthermore, the disturbance injection nodes can be selected based on the matching of the process flow topology and the disturbance type, choosing locations that are sensitive to and representative of the system state. In an exemplary embodiment, the disturbance injection nodes may include raw water inlet nodes, membrane module front-end nodes, regenerant dosing nodes, etc. The optimal operating range can be the parameter feasible region where the system can maintain stable, efficient, and compliant operation under various dynamic disturbance conditions while satisfying boundary constraints. It can be used as the target area for controlling the operating state of a "lights-out" plant, replacing the traditional single optimal setpoint. In a specific embodiment, the optimal operating range may include a water quality stability range, an energy consumption economic range, and an equipment lifespan guarantee range.

[0040] Determining disturbance injection nodes based on dynamic operating condition disturbance types and boundary constraints can be achieved by matching disturbance types with system topology and filtering injectable locations based on boundary constraints. Furthermore, this operation can be implemented by locating input ports associated with disturbance types in the device topology graph using graph traversal algorithms, or by using historical fault data clustering analysis to identify high-frequency disturbance locations as candidate injection nodes. This ensures that disturbance injection conforms to actual operating conditions without violating system safety boundaries. Determining the optimal operating range based on disturbance injection nodes and unit dynamic characteristics can be achieved by simulating disturbance propagation in a digital twin model and delineating feasible operating domains based on the dynamic characteristics of each unit, such as response speed and recovery capability. Further, this operation can be achieved by extracting a subset of parameters satisfying all constraints from the disturbance response surface using interval analysis, or by generating numerous disturbance scenarios through Monte Carlo simulation and statistically analyzing the distribution of operating parameters satisfying boundary conditions. This generates an operating range with disturbance resistance capabilities, improving system stability under disturbances.

[0041] Based on the optimal operating range, various types of dynamic operating condition disturbances are injected into the pretreatment unit, membrane separation unit, and water quality monitoring unit in the digital twin model. This injection can be achieved by applying time-varying signals at designated injection nodes according to preset disturbance types to drive the digital twin model's operation. Furthermore, this operation can simulate water quality fluctuations by superimposing noise signals conforming to a specific probability distribution at the model input, or by modifying the unit aging coefficient or efficiency decay curve to simulate equipment performance degradation. This proactively stimulates the system's dynamic response under complex coupled disturbances, enhancing the realism of the model's predictions.

[0042] Based on the control objectives, key monitoring nodes are selected in the digital twin model, and the entire process of multiple types of dynamic operating condition disturbance injections is tracked based on these key monitoring nodes and the digital twin model. Key monitoring nodes can be core observation points in the digital twin model used to track the propagation effect of disturbances and assess the system response. They can support the quantitative analysis of the impact of disturbances throughout the entire process, providing a data foundation for the generation of theoretical thresholds. Furthermore, key monitoring nodes can be selected from pretreatment, membrane separation, and water quality monitoring units based on process sensitivity analysis and control target requirements. For example, key monitoring nodes may include reverse osmosis feed nodes, mixed bed effluent nodes, and concentrate discharge nodes. Tracking the entire process of multiple types of dynamic operating condition disturbance injections based on key monitoring nodes and the digital twin model can involve continuously recording the parameter evolution trajectory of key monitoring nodes after disturbance injection. Furthermore, this operation can be achieved by enabling the event-triggered recording mechanism of the digital twin model to save high-frequency data only when parameters exceed limits or undergo abrupt changes, or by constructing a disturbance-response causal graph to track the temporal dependencies between parameters at each node, thereby fully capturing the propagation path and cumulative effect of disturbances in the system.

[0043] Based on the full-process tracking results, theoretical thresholds for water quality parameters, energy consumption parameters, and equipment status parameters at each key monitoring node are determined under various types of dynamic operating condition disturbances. These theoretical thresholds can be the extreme allowable values ​​of water quality, energy consumption, and equipment status parameters at each key monitoring node under simulations of various dynamic disturbance conditions. They can be used to form a quantitative benchmark for the optimal operating state of a "lights-out" plant, for subsequent deviation identification and control strategy generation. In a specific embodiment, the theoretical thresholds may include water quality compliance thresholds, unit water production energy consumption thresholds, and membrane pressure difference safety thresholds. Determining the theoretical thresholds for water quality parameters, energy consumption parameters, and equipment status parameters at each key monitoring node based on the full-process tracking results can be achieved by extracting extreme parameter values ​​that satisfy boundary constraints from the disturbance simulation trajectory. Furthermore, this operation can be achieved by performing envelope analysis on the parameter extreme values ​​under each type of disturbance scenario to obtain conservative boundaries as thresholds, or by using quantile regression methods to dynamically adjust the threshold confidence level according to the disturbance intensity, thereby forming an operating benchmark with physical meaning and control guidance value.

[0044] The optimal operating state of a dark factory in unattended and multi-unit collaborative scenarios is obtained based on theoretical thresholds.

[0045] Taking the optimized operation of the membrane system in response to the seasonal increase in raw water turbidity as an example, the cloud-edge-end collaborative intelligent control method of the light-out water treatment plant in this embodiment can be as follows: The digital twin model first identifies boundary constraints (such as maximum transmembrane pressure difference and minimum recovery rate) and dynamic disturbance type (seasonal increase in raw water turbidity); based on this, the disturbance injection node is determined to be the inlet of the multi-media filter; combined with the fouling response characteristics of the ultrafiltration and reverse osmosis units, the optimal operating range is defined (such as backwashing frequency of 0.5–1.2 times / hour and scale inhibitor dosage of 3–5 mg / L); a turbidity step increase disturbance is injected into the model, and key monitoring nodes are tracked throughout the process (ultrafiltration permeate SDI, reverse osmosis inter-stage pressure difference, concentrate flow rate); finally, the theoretical thresholds are obtained: SDI≤4.5, inter-stage pressure difference increase≤15%, and unit permeate energy consumption≤2.8 kWh / m³; this threshold set constitutes the optimal operating state benchmark of the light-out plant under this disturbance.

[0046] In one embodiment, after obtaining the optimal operating state of a "lights-out" factory under unattended and multi-unit collaborative scenarios based on theoretical thresholds, the method further includes: acquiring the obtained optimal operating state and establishing a dynamic mapping relationship between dynamic operating condition disturbance types and the optimal operating state; wherein, the dynamic mapping relationship can be a structured association rule or index mechanism between dynamic operating condition disturbance types and their corresponding optimal operating states, which can be used to support rapid retrieval and matching of disturbance types to optimal operating states, improving control decision efficiency. In this embodiment, the dynamic mapping relationship can be implemented through association forms such as key-value pairs, graph nodes, or vector embeddings between disturbance type labels and optimal operating state parameter sets. For example, the dynamic mapping relationship can include, but is not limited to, one or more of rule mapping relationships, vector similarity mapping relationships, graph neural network embedding mapping relationships, etc. Establishing a dynamic mapping relationship between dynamic operating condition disturbance types and optimal operating states can be achieved by structurally associating the identified disturbance types with the optimal operating states obtained through digital twin simulation. Furthermore, this operation can be achieved by constructing a hash mapping table with perturbation type as key and optimal running state parameter set as value, or by embedding perturbation feature vector and optimal state vector together in a low-dimensional space and using cosine similarity to achieve nearest neighbor retrieval, thereby realizing semantic connection between perturbation scenario and ideal running benchmark and supporting fast matching.

[0047] Based on a dynamic mapping relationship, dynamic operating condition disturbance types are bound to optimal operating states, and the dynamic operating condition disturbance types and optimal operating states are recorded in real time in the cloud-edge collaborative database based on the state binding results. The state binding results can be a logical or data-level solidified association between a specific dynamic operating condition disturbance type and its corresponding optimal operating state, ensuring that disturbance-state pairs can be consistently invoked during system operation and avoiding redundant calculations. In an exemplary embodiment, the state binding results may include, but are not limited to, disturbance-threshold binding packages, disturbance-interval binding records, and disturbance-policy binding snapshots. Binding dynamic operating condition disturbance types to optimal operating states can be achieved by logically encapsulating or data packaging the disturbance-state pairs in the dynamic mapping relationship to form indivisible calling units. Furthermore, this operation can be implemented by generating a JSON binding object containing a disturbance ID, timestamp, optimal operating interval, and theoretical threshold, or by creating a "correspondence" triplet between disturbance entities and state entities in a knowledge graph, thereby ensuring that the complete operating state configuration can be directly invoked after disturbance identification, avoiding information fragmentation.

[0048] A cloud-edge collaborative database can be a distributed storage system deployed at both the cloud and edge, equipped with data synchronization and consistency guarantee mechanisms. It can support real-time filling, updating, and cross-level access of perturbation-state pairs. In one specific embodiment, the cloud-edge collaborative database can adopt an edge cache + cloud master database architecture, maintaining data consistency through an incremental synchronization protocol. For example, the cloud-edge collaborative database can include, but is not limited to, edge time-series databases, cloud knowledge graph libraries, and Hybrid Transaction Analysis and Processing (HTAP) databases. Real-time filling of records in the cloud-edge collaborative database can involve writing state binding results to the database while maintaining data synchronization between the edge and cloud. Furthermore, this operation can be achieved by triggering an asynchronous synchronization task to the cloud master database after local writing at the edge and handling concurrent writes using conflict resolution strategies, or by using a streaming write interface to append new perturbation-state pairs to the time-series database as event logs. This enables persistent knowledge storage and cross-level sharing, supporting online learning and historical backtracking.

[0049] A cloud-edge-device collaborative control decision knowledge base is built based on real-time record filling.

[0050] The cloud-edge-device collaborative control decision knowledge base can be a structured knowledge set built based on disturbance-state binding records to support control strategy generation. It can provide prior knowledge support for deviation analysis and adaptive control, reducing the computational burden of online optimization. In this embodiment, the cloud-edge-device collaborative control decision knowledge base can extract disturbance-state pairs from the cloud-edge collaborative database and organize them into queryable and reasonable knowledge units according to control objectives. For example, the cloud-edge-device collaborative control decision knowledge base may include, but is not limited to, a disturbance response rule base, a running state template base, and a multi-objective optimization strategy base. Constructing the cloud-edge-device collaborative control decision knowledge base can involve extracting disturbance-state binding records from the cloud-edge collaborative database and organizing them into callable knowledge units according to control logic. Furthermore, this operation can be implemented by building a strategy template library categorized by process unit and control objective, supporting fuzzy matching and interpolation, or by using a graph database to build a disturbance-state-action triplet network, supporting path reasoning and combined strategy generation. This provides structured prior knowledge for subsequent control strategy generation, improving decision-making speed and robustness.

[0051] Taking the automatic strategy invocation under equipment aging disturbance as an example, the cloud-edge-end collaborative intelligent control method of the water treatment "lights-out" plant in this embodiment can be as follows: When the edge sensing terminal detects a continuous decrease in reverse osmosis membrane flux and identifies it as "equipment performance degradation disturbance", the system retrieves the state binding result corresponding to the disturbance type in the cloud-edge-end collaborative control decision knowledge base, and obtains the pre-stored optimal operating range (such as increasing the flushing frequency and reducing the recovery rate) and theoretical threshold (such as the maximum allowable pressure difference of 1.8MPa). This knowledge originates from the previous injection of similar disturbances into the digital twin model and the completion of the whole process tracking, and is then written into the cloud-edge collaborative database after state binding. The control module directly calls this knowledge without re-simulation, quickly generates adjustment instructions to drive the intelligent execution device, and achieves efficient response.

[0052] In one embodiment, before collecting process parameter data of each process unit under dynamic operating condition disturbances based on the deployed edge sensing terminals, the process includes: acquiring the equipment topology of the water treatment system based on the cloud platform, and identifying key processing units and parameter-sensitive areas in the system based on the topology; using the key processing units and parameter-sensitive areas as a first set of sensing points; determining the environmental disturbance factors of the water treatment system based on process specifications, and analyzing the influence path of the environmental disturbance factors on the process parameters in conjunction with the equipment topology; using the nodes on the influence path of the environmental disturbance factors on the process parameters as a second set of sensing points; and optimizing and guiding the deployment of the edge sensing terminals based on the first and second sets of sensing points.

[0053] The equipment topology can be a structured map describing the physical and technological flow logic between various devices in a water treatment system. It can serve as a basis for identifying key treatment units, parameter-sensitive areas, and disturbance propagation paths. In this embodiment, the equipment topology can be extracted and structured from engineering design data or a SCADA system via a cloud platform. Key treatment units can be process equipment or subsystems in the water treatment process chain that have a decisive impact on effluent quality, system energy efficiency, or operational stability. They can be prioritized for sensing to ensure that the status of core components can be monitored in real time. For example, key treatment units may include, but are not limited to, one or more of reverse osmosis units, mixed-bed ion exchange units, and ultrafiltration pretreatment units. Parameter-sensitive areas can be process locations or equipment ranges in the system where even small changes in operating parameters can significantly affect overall performance or effluent quality. They can be used as high-value sensing points to improve the sensitivity and accuracy of status identification. In an exemplary embodiment, parameter-sensitive areas may include, but are not limited to, the membrane module inlet section, the vicinity of the regenerant dosing point, and the pH adjustment reaction zone.

[0054] The system acquires the equipment topology of the water treatment system via a cloud platform and identifies key processing units and parameter-sensitive areas based on this topology. This can be achieved by retrieving structured topology data from the cloud platform and combining it with process importance indicators and parameter sensitivity analysis to identify key areas. Furthermore, this operation can be implemented by using graph theory algorithms (such as betweenness centrality) to identify key nodes in the topology as key processing units, or by calculating the partial derivatives of parameters in each region with respect to effluent quality using historical operating data to identify highly sensitive areas. This provides a system-level structural basis for selecting sensing points, avoiding the blindness of experience-based point placement.

[0055] Using key processing units and parameter-sensitive areas as the first set of sensing points can be achieved by integrating the location information of the identified key processing units and parameter-sensitive areas into a unified set. Furthermore, this operation can be implemented by storing the set elements in the form of device IDs or spatial coordinates, or by weighting them by importance for subsequent resource allocation, thereby forming a sensing priority list oriented towards the core system performance. Environmental disturbance factors can be a set of uncontrollable variables from outside the system that may cause fluctuations in process parameters. These can be used to model disturbance propagation mechanisms and guide the anti-interference sensing layout. In a specific embodiment, environmental disturbance factors may include, but are not limited to, fluctuations in influent water quality, changes in ambient temperature, and fluctuations in water supply pressure.

[0056] An influence path can be a causal propagation link where environmental disturbances affect process parameters through the equipment topology. It can be used to reveal the transmission mechanism of disturbances in the system and locate critical nodes susceptible to impact. In this embodiment, the influence path can be derived from the topology and process mechanism model to determine the direction and intensity of disturbance propagation. Environmental disturbances in the water treatment system are identified based on process specifications, and the influence path of these disturbances on process parameters is analyzed in conjunction with the equipment topology. This can be achieved by listing typical disturbance sources according to industry standards and operating procedures, and simulating their propagation path in the system using the topology. Furthermore, this operation can be achieved by constructing a directed graph of disturbance propagation and using a path search algorithm (such as Dijkstra's algorithm) to determine the influence path, or by simulating parameter response trajectories under different disturbance inputs using a mechanism model. This establishes a disturbance-response mapping relationship, supporting anti-interference sensing design.

[0057] The nodes along the path of environmental disturbance factors affecting process parameters are used as the second set of sensing points. This can be achieved by extracting all intermediate nodes or key turning points along the path. Further, this operation can be implemented by selecting the node with the largest parameter change gradient along the path as a representative point, or by retaining the beginning, end, and highly coupled nodes in the middle of the path to form a minimal coverage set, thus enabling full observability of the disturbance propagation process. The first set of sensing points can be a set of locations for priority deployment of sensing terminals, consisting of key processing units and parameter-sensitive areas, which can be used to ensure high-density coverage of areas related to the core performance of the system. Further, the first set of sensing points can be combined with the second set of sensing points to form the optimization basis for the deployment of edge sensing terminals. The second set of sensing points can be a set of sensing locations consisting of key nodes along the path of environmental disturbance factors, which can be used to enhance the system's early perception and propagation tracking capabilities of dynamic disturbances. In an exemplary embodiment, the second set of sensing points can be merged with the first set of sensing points to guide the deployment of edge sensing terminals, forming complementary coverage.

[0058] Optimizing and guiding the deployment of edge sensing terminals based on the first and second sets of sensing points can be achieved by merging the two sets and removing duplicates to generate a final deployment suggestion that guides the selection of installation locations for the edge sensing terminals. Furthermore, this operation can be implemented by using set intersection and union operations to generate a comprehensive list of sensing points and adjusting it in conjunction with on-site wiring conditions, or by introducing coverage optimization algorithms (such as solving the maximum coverage problem) to select the optimal subset under budget constraints. This maximizes the value of sensing information with limited hardware resources, improving data representativeness and system robustness.

[0059] Taking the disturbance caused by algae blooms in raw water due to high summer temperatures as an example, the cloud-edge-device collaborative intelligent control method of the water treatment "lights-out" plant in this embodiment can be as follows: After the cloud platform analyzes the equipment topology, it identifies the ultrafiltration unit as the key treatment unit, and its inlet end as the parameter-sensitive area, forming the first set of sensing points; at the same time, according to the process specifications, "water temperature rise" and "algae concentration rise" are determined as environmental disturbance factors, and combined with topology analysis, their influence path is: raw water tank, multi-media filter, ultrafiltration inlet header, ultrafiltration membrane module, and each node on the path constitutes the second set of sensing points; after the two are integrated, edge sensing terminals are deployed at the raw water tank outlet, the multi-media filter differential pressure monitoring point, the ultrafiltration inlet header turbidity point, and the membrane module cross-membrane differential pressure point; when algae blooms, the system captures turbidity and UV254 anomalies at the raw water tank outlet in advance, triggers an early warning and starts pre-oxidation dosing to avoid ultrafiltration fouling.

[0060] In one embodiment, process parameter data of each process unit under dynamic operating condition disturbances are collected based on the deployed edge sensing terminals. This includes: monitoring the operating condition disturbances of the water treatment system during a target period, and identifying the time-frequency characteristics of the dynamic operating condition disturbances based on the monitoring results. The time-frequency characteristics of the dynamic operating condition disturbances can be a joint representation of the disturbances in the time and frequency dimensions, reflecting features such as the time of occurrence, duration, rate of change, and periodicity of the disturbance. This can be used as a basis for adaptive adjustment of the edge sensing terminal sampling strategy, improving the targeting and effectiveness of data collection. In an exemplary embodiment, the time-frequency characteristics of the dynamic operating condition disturbances can include, but are not limited to, one or more of the following: transient impact disturbance characteristics, slowly changing trend disturbance characteristics, and periodic oscillation disturbance characteristics. Identifying the time-frequency characteristics of the dynamic operating condition disturbances based on the monitoring results can involve performing time-frequency joint analysis of the original monitoring signals within the target period to extract the dynamic patterns of the disturbances. Furthermore, this operation can be achieved by using short-time Fourier transform or continuous wavelet transform to decompose water quality or pressure signals into time and frequency components, or by using a sliding window combined with a mutation detection algorithm to identify the start and end times of disturbances and the changing gradients, thereby avoiding information loss or redundancy caused by fixed sampling.

[0061] The data sampling rate configuration range for the edge sensing terminal is determined based on time-frequency characteristics, and adaptive parameter configuration is performed on the edge sensing terminal based on the configuration range. The data sampling rate configuration range can be an upper and lower limit range of the number of data points collected by the edge sensing terminal per unit time, dynamically set according to disturbance conditions. This range can guide the edge sensing terminal to dynamically adjust the sampling density under different disturbance intensities, balancing data fidelity and resource overhead. For example, the data sampling rate configuration range can include, but is not limited to, one or more of the following: high disturbance response sampling range, steady-state low-frequency sampling range, and transitional adaptive sampling range. Determining the data sampling rate configuration range for the edge sensing terminal based on time-frequency characteristics and performing adaptive parameter configuration based on the configuration range can be achieved by mapping the disturbance frequency components and change rates to sampling rate requirements and dynamically issuing configuration instructions to the edge terminal. In a specific embodiment, this operation can be achieved by establishing a disturbance frequency-sampling rate mapping table and issuing configuration parameters from the edge management node, or by having the edge terminal's built-in sampling strategy engine autonomously adjust the sampling rate based on locally identified time-frequency characteristics. This allows sensing resources to be allocated on demand, improving data resolution during critical disturbance periods and reducing communication and computing load during steady-state periods.

[0062] The system uses edge sensing terminals to perform wide-area synchronous monitoring of the system based on parameter configuration results. Based on these monitoring results, it collects water quality parameters, equipment operating parameters, and energy consumption parameters for each process unit. This wide-area synchronous monitoring can be achieved by each edge sensing terminal synchronously collecting data from multiple process units at a configured sampling rate under a unified time reference. Furthermore, this operation can achieve microsecond-level time synchronization via 5G network time synchronization or the PTP precision time protocol, or by employing a triggered synchronous acquisition mechanism where edge coordination nodes broadcast sampling start signals. This ensures strict alignment of cross-unit process parameters in the time dimension, supporting subsequent multi-source data fusion analysis.

[0063] Based on the 5G low-latency communication links pre-established between each edge sensing terminal and the cloud platform, the collected process parameter data is aggregated to the cloud platform in real time.

[0064] The 5G low-latency communication link can be a dedicated communication channel built on 5G network slicing and edge computing technologies, possessing deterministic low transmission latency characteristics. It can be used to ensure the real-time and reliable aggregation of process parameter data collected by multiple edge sensing terminals to the cloud platform, supporting rapid decision-making. In this embodiment, the 5G low-latency communication link can establish an end-to-end QoS-guaranteed connection through 5G ultra-reliable low-latency communication mode. For example, the 5G low-latency communication link can include, but is not limited to, one or more of ultra-reliable low-latency communication slicing links, MEC local backhaul links, and time-sensitive network converged 5G links.

[0065] Based on the pre-established 5G low-latency communication links between each edge sensing terminal and the cloud platform, the collected process parameter data is aggregated to the cloud platform in real time. This can be achieved by uploading synchronously collected multi-dimensional process parameters to the cloud through the configured 5G low-latency links. In one specific embodiment, this operation can be achieved by using 5G network slicing to allocate independent low-latency channels to the water treatment system to prioritize the transmission of key parameters, or by performing data compression and differential encoding at the edge and then pushing the data to the cloud platform in batches via the 5G uplink. This significantly reduces the data transmission latency for sensing and decision-making, supporting rapid updates of cloud-side models and strategy generation.

[0066] Taking multi-unit collaborative sensing under sudden fluctuations in raw water turbidity as an example, the cloud-edge-end collaborative intelligent control method of the water treatment "lights-out" plant in this embodiment can be as follows: When the turbidity of the upstream water source suddenly increases by 3 times within 10 minutes due to heavy rain, the edge sensing terminal first monitors the inlet at the basic sampling rate; the system identifies that the disturbance has high-frequency sudden change and time-varying non-stationary characteristics, and accordingly increases the sampling rate of the reverse osmosis front section, ultrafiltration unit and dosing system from 1 time / minute to 10 times / second; all terminals achieve microsecond-level synchronous acquisition through 5G time synchronization to obtain parameters such as water quality, pressure difference and flow rate; the data is uploaded to the cloud platform within 200ms via 5G low-latency link to update the state of the digital twin model and trigger the collaborative control strategy.

[0067] In one embodiment, feature extraction and fusion analysis are performed on process parameter data to determine the actual operating status of the water treatment system. This includes: acquiring collected process parameter data and spatiotemporally aligning the data on a digital twin model. Spatiotemporal alignment of the process parameter data on the digital twin model can be achieved by mapping process parameter data from different sensing points and sampling frequencies to corresponding nodes and time steps in the digital twin model based on their physical location and timestamps. Furthermore, spatiotemporal alignment of the process parameter data on the digital twin model can be achieved by establishing a spatial index based on the device topology map and combining it with timestamp interpolation for synchronous alignment, or by using a local clock synchronization protocol on the edge terminal to unify the time base before mapping according to the model node coordinates. This ensures that multi-source heterogeneous data and the virtual model remain consistent in spatial topology and time series, providing a foundation for subsequent fusion analysis.

[0068] Based on the spatiotemporal alignment results, water quality fluctuations and energy consumption changes at each sensing point are dynamically rendered on a digital twin model. The rendering results are then used to identify material flow characteristics and energy coupling patterns between different units in the system. Material flow characteristics can represent the transfer and distribution patterns of materials (such as raw water, intermediate product water, and regenerated wastewater) between process units in the water treatment system in terms of flow rate, concentration, and phase. These characteristics can characterize the dynamic behavior of material exchange between units, supporting coupled analysis of the system's operating status. In an exemplary embodiment, material flow characteristics can include, but are not limited to, one or more of continuous flow material characteristics, batch flow material characteristics, and reflux material characteristics. Energy coupling patterns can represent the mutual influence and synergistic relationships between different process units in terms of energy consumption, heat exchange, and pressure energy transfer. These patterns can reveal the system's energy efficiency correlation structure, providing a basis for energy consumption deviation identification and optimization. For example, energy coupling patterns can employ pump-valve power consumption coupling patterns, membrane separation pressure difference coupling patterns, and regeneration heat energy coupling patterns.

[0069] Dynamically rendering water quality fluctuations and energy consumption changes on a digital twin model can be achieved by overlaying aligned water quality and energy consumption data onto the corresponding units or connection paths of the digital twin model in the form of visualization or numerical fields. In a specific embodiment, dynamic rendering of water quality fluctuations and energy consumption changes on a digital twin model can be achieved by using color gradients or vector arrows to render water concentration and flow direction in real time in the 3D model, or by constructing a time-series heatmap to represent the energy consumption change trend of each unit and binding it to the model structure. This can explicitly present the propagation path and intensity of dynamic disturbances within the system, and help identify material and energy interaction patterns.

[0070] The coupling relationships between process units are analyzed based on material flow characteristics and energy coupling modes, and the process parameter data at each sensing point are dynamically compensated and corrected based on these coupling relationships. The coupling relationship can be a dynamic interaction mechanism between process units determined by both material flow characteristics and energy coupling modes, which can serve as the basis for dynamic compensation and correction of process parameter data, improving the consistency of the sensing data. Furthermore, the coupling relationship can include, but is not limited to, one or more of strong coupling relationships, weak coupling relationships, and time-varying coupling relationships.

[0071] Analyzing the coupling relationships between units based on material flow characteristics and energy coupling patterns can extract features such as the correlation of material flux and energy consumption response delay between units from dynamic rendering results, and deduce the coupling strength and direction. In a specific embodiment, analyzing the coupling relationships between units based on material flow characteristics and energy coupling patterns can identify causal coupling between units through Granger causality tests or mutual information analysis, or it can construct a graph neural network to learn the edge weights between nodes as a representation of coupling relationships using rendered data as input. This can quantify the interdependencies between process units and provide structured priors for data correction. Dynamically compensating and correcting process parameter data based on coupling relationships can correct the deviation of sensing point data affected by disturbances from neighboring units based on the analyzed coupling relationships. For example, dynamic compensation and correction of process parameter data based on coupling relationships can be achieved by constructing a Kalman filter observation model using coupling relationships to perform state estimation correction on the original data, or by aggregating neighboring node data using a graph convolutional network to generate corrected sensing values. This can eliminate data distortion caused by local disturbance propagation, sensor blind spots, or transmission delays, and improve the accuracy of state recognition.

[0072] Based on the correction results, the actual water quality, energy consumption and equipment status data at each sensing point are aggregated and statistically analyzed to generate the actual operating status of the water treatment system.

[0073] Taking the collaborative disturbance identification of multi-media filtration and reverse osmosis units as an example, the cloud-edge-end collaborative intelligent control method of the water treatment "lights-out" plant in this embodiment can be as follows: When the pressure difference of the multi-media filter increases abnormally, the edge sensing terminal collects the increase in its effluent turbidity and the fluctuation of the reverse osmosis feed water pressure; the process parameter data is mapped to the corresponding node of the digital twin model after spatiotemporal alignment; the dynamic rendering shows that the turbidity disturbance propagates downstream along the process, while the energy consumption of the reverse osmosis section increases synchronously; the system identifies the strong material flow characteristics and pressure-energy coupling mode between "filtration-membrane separation" accordingly; after analyzing the strong coupling relationship between the two, the system performs neighborhood compensation correction on the reverse osmosis feed water turbidity sensor data to eliminate instantaneous fluctuation interference; finally, the corrected water quality, pressure difference and energy consumption data are aggregated to accurately determine that the actual operating state of the system is "decreased pretreatment efficiency" rather than membrane fouling.

[0074] In one embodiment, the deviation parameters between the actual operating state and the optimal operating state are determined, and the deviation parameters are analyzed in multiple dimensions to generate a cloud-edge-device collaborative control strategy. This includes: acquiring the actual operating state of the water treatment system and quantitatively comparing the actual operating state with the corresponding optimal operating state to obtain dynamic deviation parameters. The dynamic deviation parameters can be quantitative differences between the actual and optimal operating states over time, used to characterize the dynamic characteristics of the system deviating from ideal operating conditions. The dynamic deviation parameters can be used as the basis for triggering a graded response mechanism to distinguish whether the system is in a stable or unstable state. In an exemplary embodiment, the dynamic deviation parameters may include, but are not limited to, one or more of the following: water quality dynamic deviation parameters, energy consumption dynamic deviation parameters, and equipment collaborative dynamic deviation parameters.

[0075] When the amplitude of the dynamic deviation parameter is within a preset safety threshold, the system is determined to be in a stable operating range, and continuous disturbance tracking is performed on the system based on the determination result; otherwise, the system is determined to be out of a stable operating range. Based on the determination result and the deployment location of the edge sensing terminals, the water treatment system is divided into cloud-edge collaborative control areas, and spatiotemporal correlation mining is performed on the process parameter data at each sensing point based on the area division result. The cloud-edge collaborative control area can be a functional division of the water treatment system based on the deployment location of the edge sensing terminals, used to achieve differentiated control by region. The cloud-edge collaborative control area can be used to support the design of control structures that support local rapid response and global coordinated optimization. For example, the cloud-edge collaborative control area can include, but is not limited to, one or more of the following: pretreatment control area, membrane treatment control area, and regeneration control area. The spatiotemporal correlation mining of the process parameter data at each sensing point based on the area division result can be performed within the cloud-edge collaborative control area, by performing cross-point correlation modeling on the time series data of multiple sensing points. Furthermore, this operation can be achieved by using graph attention networks to model the space and combining them with long short-term memory networks to capture temporal dependencies, or by using Granger causality tests and mutual information methods to construct a spatiotemporal correlation graph. This can break down data silos and identify the propagation paths and coupling effects of disturbances in the system. The spatiotemporal correlation mining results can be a set of correlation features obtained by modeling and extracting the coupling relationships of process parameter data at each sensing point in the temporal and spatial dimensions. The spatiotemporal correlation mining results can be used to reveal cross-unit disturbance propagation paths and impact mechanisms, supporting loss quantification. In a specific embodiment, the spatiotemporal correlation mining results may include, but are not limited to, one or more of the following: temporal causal correlation patterns, spatial topological correlation patterns, and multivariate coupling correlation patterns.

[0076] Based on spatiotemporal correlation mining results, the material imbalance and energy loss in each region are quantified, and dynamic pattern recognition is performed on the imbalance to determine the system's operational instability modes. These instability modes include water quality exceeding standards, abnormal energy consumption, and equipment failure. Each operational instability mode can be a categorizable abnormal behavior type exhibited by the system under material or energy imbalance conditions. These operational instability modes can provide a semantic anomaly identification basis for control strategies, enabling targeted responses. For example, operational instability modes may include, but are not limited to, one or more of the following: water quality exceeding standards, abnormal energy consumption, and equipment failure.

[0077] Based on the spatiotemporal correlation mining results, the material loss and energy loss in each region are quantified, and dynamic pattern recognition is performed on the loss to determine the system's operational instability mode. This can be achieved by calculating the material / energy balance residuals based on the correlation mining results and matching them with a preset instability mode template. Furthermore, this operation can be implemented by constructing feature vectors from the mass conservation equation residuals and energy balance deviations and inputting them into a support vector machine classifier to identify instability modes, or by using a hidden Markov model to perform state clustering on the temporal evolution of loss and mapping it to operational instability modes. This allows for semantic classification of system anomalies, improving the targeting of control strategies.

[0078] Based on the system's failure metrics during operational instability, potential faulty units in critical equipment are identified and intelligently diagnosed. Once no abnormalities are found in these units, a cloud-edge-device collaborative control mechanism is activated. These potential faulty units can be critical equipment or subsystems that may cause system anomalies, inferred from failure metrics and operational instability modes. They can serve as target objects for intelligent diagnosis, avoiding ineffective control. In one specific embodiment, potential faulty units may include, but are not limited to, one or more of high-pressure pump units, reverse osmosis membrane modules, and chemical dosing units. Identifying and intelligently diagnosing potential faulty units in critical equipment based on the system's failure metrics during operational instability modes can involve locating high-risk equipment based on instability modes and failure metric distributions, and then verifying their health status using a diagnostic model. Furthermore, this operation can be achieved through cross-validation using Bayesian network inference of equipment failure probabilities combined with auxiliary signals such as vibration and temperature, or by using a digital twin model to virtually reproduce potential faulty units and comparing the simulation and measured response differences. This avoids ineffective control due to misjudgment and improves system reliability.

[0079] Based on the activation results, a pre-trained water quality-energy consumption coupling model is retrieved from the decision knowledge base, and the effluent quality compliance rate is used as the core control indicator. The model is used to assess the safety boundary of process parameter data at each sensing point. The water quality-energy consumption coupling model can be a multi-objective optimization model pre-trained in the decision knowledge base, describing the nonlinear trade-off between effluent quality and system energy consumption. This model can be used to guide energy efficiency optimization within the safety boundary, with the effluent quality compliance rate as the core constraint. In an exemplary embodiment, the water quality-energy consumption coupling model can be one or more of the following, including but not limited to a mechanism-data hybrid coupling model, a reinforcement learning coupling model, and a multi-objective surrogate model.

[0080] The system retrieves a pre-trained water quality-energy consumption coupling model from the decision knowledge base, using the effluent quality compliance rate as the core control indicator. Based on the model, it performs a safety boundary assessment of the process parameter data at each sensing point. This can be achieved by loading the coupling model, using the current process parameters as input, and calculating the feasible energy consumption control range to meet water quality standards. Furthermore, this operation can be performed by employing a constraint satisfaction problem solver to search for the parameter combination that minimizes energy consumption under water quality compliance constraints, or by using a surrogate model to quickly predict effluent quality under different control parameters and construct a safety boundary hyperplane. This ensures that all control actions achieve energy efficiency optimization while maintaining environmental compliance.

[0081] The safety boundary assessment result can be a quantitative output of the feasible controllable range of process parameters at each sensing point under the premise of meeting the effluent water quality standards, based on the water quality-energy consumption coupling model. The safety boundary assessment result can be used to determine the safe and feasible domain of control quantities in each area, preventing control actions from causing water quality violations. For example, the safety boundary assessment result may include, but is not limited to, one or more of the following: water quality safety boundary, energy consumption optimization boundary, and equipment operation safety boundary.

[0082] Based on the safety boundary assessment results, determine the control quantities corresponding to adjusting the water quality, energy consumption, and equipment status at each sensing point to the optimal operating range; optimize and integrate the control quantities of each area to generate a cloud-edge-device collaborative control strategy.

[0083] Optimizing and integrating control parameters from various regions to generate a cloud-edge-device collaborative control strategy can be achieved by integrating control parameter suggestions within the safety boundaries of each region and generating globally coordinated commands through multi-objective optimization. Furthermore, this operation can be accomplished by employing a distributed model predictive control framework to enable edge nodes in each region to negotiate and generate a consistent control strategy, or by constructing an integer programming model in the cloud with water quality compliance as a hard constraint and minimum energy consumption as a soft objective to solve for the globally optimal control parameter allocation. This allows for coordinated and consistent control actions across regions, avoiding local optimization conflicts.

[0084] Taking the abnormal energy consumption caused by membrane fouling in a reverse osmosis system as an example, the cloud-edge-end collaborative intelligent control method of the water treatment "lights-out" plant in this embodiment can be as follows: The edge sensing terminal detects an increase in inter-section pressure difference and a decrease in permeate rate, with dynamic deviation parameters exceeding the threshold; the system is divided into membrane treatment control areas, and spatiotemporal correlation mining reveals a strong correlation between pressure difference and influent turbidity and cleaning cycle; material loss measurement shows a decrease in recovery rate and an increase in energy loss, and dynamic pattern recognition determines it to be an "abnormal energy consumption" unstable operation mode; the potential fault unit is identified as the reverse osmosis membrane module, and intelligent diagnosis confirms that there is no mechanical fault but fouling exists; the cloud-edge-end collaborative mechanism is activated, and the water quality-energy consumption coupling model is called to evaluate the minimum acceptable flux and maximum pressure difference under the premise of maintaining the effluent water quality standard; the safety boundary assessment results provide the cleaning frequency and operating pressure adjustment range; finally, the control quantities of each area are integrated to generate a collaborative control strategy of reducing flux, initiating online cleaning, and fine-tuning the scale inhibitor dosage.

[0085] In one embodiment, adaptive collaborative control of intelligent execution devices based on a cloud-edge-device collaborative control strategy includes: acquiring the cloud-edge-device collaborative control strategy and decoding the strategy to obtain a collaborative control instruction set for each intelligent execution device. The collaborative control instruction set can be a structured set of control commands generated after decoding the cloud-edge-device collaborative control strategy, oriented towards each intelligent execution device. This set can provide a unified and parsable instruction foundation for multi-unit execution, ensuring accurate transmission of control intent. In this embodiment, the collaborative control instruction set can include, but is not limited to, one or more of device-level action instructions, process-level scheduling instructions, and safety boundary instructions. Decoding the strategy to obtain the collaborative control instruction set for each intelligent execution device can be achieved by converting high-level semantic instructions in the cloud-edge-device collaborative control strategy into low-level control commands recognizable by each execution device. Furthermore, this operation can be implemented by using a protocol parsing engine to parse the JSON / YAML format strategy into Modbus / TCP instruction frames, or by matching the strategy intent with the execution capability through device digital profiling to generate a customized instruction set. This enables semantic mapping from control strategy to specific execution actions, ensuring instruction executability.

[0086] Dynamic limiting constraints are applied to the execution units of each intelligent actuator based on a collaborative control instruction set. The timing logic of each execution unit is then orchestrated based on the constraint results to generate a collaborative control sequence for the intelligent actuators. Each execution unit can be the smallest controllable functional module directly involved in physical regulation within the intelligent actuator. It can receive and execute specific control signals to adjust parameters such as valve opening, pump frequency, and dosage. For example, execution units may include, but are not limited to, electric regulating valve actuators, frequency converters, and metering pump controllers. Dynamic limiting constraints can be boundary limits on the output capabilities of the execution units that are adjusted in real-time based on current operating disturbances and equipment status. This can prevent over-control or damage to the execution units due to instructions exceeding their physical or safety capabilities. In an exemplary embodiment, dynamic limiting constraints can dynamically calculate the limiting threshold based on equipment aging, media characteristics, and environmental parameters fed back from the edge sensing terminal. For example, dynamic limiting constraints may include, but are not limited to, mechanical stroke limiting, power output limiting, and response rate limiting. Dynamic limiting constraints are applied to the execution units of each intelligent actuator based on a collaborative control instruction set. This can be achieved by imposing dynamic upper and lower limits on the output range of the original instructions based on real-time operating conditions and equipment status. Furthermore, this operation can be implemented by using an edge-side equipment health assessment model to update the limiting parameters in real time and embedding them into the instruction preprocessing module, or by dynamically tightening the high-pressure pump frequency upper limit based on predictions of membrane pressure differential changes within the next 5 minutes based on influent water quality fluctuations. This can prevent execution units from losing control or causing equipment damage due to instruction exceeding limits, thus improving control safety.

[0087] Timing logic orchestration can organize the actions of multiple execution units according to the process flow topology and control objectives, arranging their temporal and logical relationships. This can ensure the coordination and process compliance of multiple unit actions in the spatiotemporal dimensions. In a specific embodiment, timing logic orchestration can combine the process flow diagram and timing constraint rules, using finite state machines or Petri nets to model the execution sequence. For example, timing logic orchestration can include, but is not limited to, sequential start orchestration, interlock protection orchestration, and redundancy switching orchestration. Based on the constraint results, timing logic orchestration of each execution unit generates a collaborative control sequence for the intelligent execution device. This can be achieved by arranging the action sequence and triggering conditions of the execution units according to the process logic relationships, based on the limited instructions. Furthermore, this operation can be implemented by defining the start-stop dependencies between units using an event-triggered finite state machine model, or by constructing a directed graph of the process flow and generating an acyclic execution sequence through topological sorting. This ensures that multiple units operate in an orderly and conflict-free manner under complex coupled conditions, enhancing system synergy.

[0088] A coordinated control sequence can be a sequence of execution unit actions arranged by timing logic, possessing temporal order and logical dependencies. It can be used to ensure that multiple units operate in coordination according to the correct timing and logic in complex processes. For example, a coordinated control sequence may include, but is not limited to, serial start-stop sequences, parallel adjustment sequences, and fault switching sequences.

[0089] Closed-loop control of each intelligent actuator is achieved based on the collaborative control sequence, realizing unmanned operation of the water treatment "lights-out" plant and real-time optimization of multi-unit collaboration.

[0090] Closed-loop control of each intelligent actuator based on a collaborative control sequence can be achieved by driving the actuators one by one according to a pre-arranged timing sequence and amplitude constraints, and feeding back the execution results for strategy iteration. Furthermore, this operation can send the sequence to the PLC controller via the OPCUA protocol and simultaneously collect execution feedback for edge verification, or embed an anomaly detection mechanism during execution. If the execution deviates from the expected trajectory, it triggers re-arrangement or a safe shutdown, thus achieving a reliable closed loop from strategy to physical action, supporting stable and optimized operation under unattended conditions.

[0091] Taking the coordinated start-up and shutdown of multiple units in a reverse osmosis system as an example, the cloud-edge-end coordinated intelligent control method of the water treatment "lights-out" plant in this embodiment can be as follows: when the digital twin model predicts that a backup RO unit needs to be started to cope with the increase in load, the cloud-edge-end coordinated control strategy issues a start command; after decoding, a coordinated control command set including a high-pressure pump, an inlet valve, and a flushing valve is generated; the edge side dynamically limits the maximum frequency of the high-pressure pump to 45Hz according to the current aging degree of the membrane module; the timing logic orchestration module, according to the process requirements, first opens the inlet valve for 3 seconds and then starts the high-pressure pump, while simultaneously delaying for 10 seconds to close the flushing valve; finally, each execution unit is driven according to this coordinated control sequence to achieve smooth and shock-free switching and avoid water hammer or membrane fouling.

[0092] Furthermore, to achieve the above objectives, the present invention also provides a cloud-edge-device collaborative intelligent control platform for a water treatment darkroom, the platform comprising: a memory, a processor, and a cloud-edge-device collaborative intelligent control program for a water treatment darkroom stored in the memory and executable on the processor, the cloud-edge-device collaborative intelligent control program for a water treatment darkroom configured to implement the steps of the cloud-edge-device collaborative intelligent control method for a water treatment darkroom as described above.

[0093] In addition, to achieve the above objectives, the present invention also provides a medium storing a cloud-edge-device collaborative intelligent control program for a water treatment darkroom, wherein when the cloud-edge-device collaborative intelligent control program for the water treatment darkroom is executed by a processor, the program implements the steps of the cloud-edge-device collaborative intelligent control method for the water treatment darkroom as described above.

[0094] Other embodiments or specific implementations of the cloud-edge-device collaborative intelligent control platform for the water treatment "lights-out" factory described in this invention can be found in the above-mentioned method embodiments, and will not be repeated here.

[0095] The above are merely preferred embodiments of the present invention and do not limit the scope of the patent. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.

Claims

1. A cloud-edge-device collaborative intelligent control method for a lights-out water treatment plant, characterized in that, The method includes: constructing a digital twin model based on the equipment topology and operating parameters of the water treatment process; determining the optimal operating state of the dark factory under unattended and multi-unit collaborative scenarios based on the digital twin model; collecting process parameter data of each process unit under dynamic operating condition disturbances based on deployed edge sensing terminals; performing feature extraction and fusion analysis on the process parameter data to determine the actual operating state of the water treatment system; determining the deviation parameters between the actual operating state and the optimal operating state; performing multi-dimensional analysis on the deviation parameters to generate a cloud-edge-device collaborative control strategy; and performing adaptive collaborative control of intelligent actuators based on the cloud-edge-device collaborative control strategy.

2. The cloud-edge-device collaborative intelligent control method for a light-out water treatment plant as described in claim 1, characterized in that, The construction of a digital twin model based on the equipment topology and operating parameters of the water treatment process includes: determining the calling dimensions of the equipment topology and operating parameters of the water treatment system based on digital twin modeling specifications, and retrieving process flow diagrams, equipment characteristic parameters, and historical operating data from the cloud platform database based on the calling dimensions; classifying the equipment topology data and operating parameters hierarchically based on functional requirements to obtain system network topology parameters and unit dynamic characteristic parameters; constructing a process flow diagram of the water treatment system based on the network topology parameters, and labeling the dynamic response attributes of each treatment unit and connecting pipeline in the flow diagram based on the unit dynamic characteristic parameters; simultaneously, obtaining the target application scenario of the digital twin model based on the cloud platform, and determining the accuracy grading standard of the digital twin model based on the target application scenario; and labeling the regional functions of the digital twin model based on the accuracy grading standard to obtain a digital twin model covering the pretreatment unit, membrane separation unit, and water quality monitoring unit.

3. The cloud-edge-device collaborative intelligent control method for a light-out water treatment plant as described in claim 1, characterized in that, The process of determining the optimal operating state of a "lights-out" factory under unattended and multi-unit collaborative scenarios based on a digital twin model includes: determining the boundary constraints of the water treatment system based on the digital twin model; simultaneously, determining the types of dynamic operating condition disturbances based on process characteristics and environmental standards; determining disturbance injection nodes based on the types of dynamic operating condition disturbances and boundary constraints; determining the optimal operating range based on the disturbance injection nodes and unit dynamic characteristics; injecting multiple types of dynamic operating condition disturbances into the pretreatment unit, membrane separation unit, and water quality monitoring unit in the digital twin model based on the optimal operating range; selecting key monitoring nodes in the digital twin model based on the control objectives; and tracking the injection of multiple types of dynamic operating condition disturbances throughout the entire process based on the key monitoring nodes and the digital twin model; determining the theoretical thresholds of water quality parameters, energy consumption parameters, and equipment status parameters at each key monitoring node under multiple types of dynamic operating condition disturbances based on the results of the entire process tracking; and obtaining the optimal operating state of the "lights-out" factory under unattended and multi-unit collaborative scenarios based on the theoretical thresholds.

4. The cloud-edge-device collaborative intelligent control method for a light-out water treatment plant as described in claim 3, characterized in that, After obtaining the optimal operating state of the dark factory in unattended and multi-unit collaborative scenarios based on theoretical thresholds, the method further includes: acquiring the obtained optimal operating state and establishing a dynamic mapping relationship between dynamic operating condition disturbance types and the optimal operating state; binding the dynamic operating condition disturbance types with the optimal operating state based on the dynamic mapping relationship, and filling records of the dynamic operating condition disturbance types and the optimal operating state in the cloud-edge collaborative database in real time based on the state binding results; and constructing a cloud-edge-device collaborative control decision knowledge base based on the real-time filling records.

5. The cloud-edge-device collaborative intelligent control method for a light-out water treatment plant as described in claim 1, characterized in that, Before the deployed edge sensing terminals collect process parameter data of each process unit under dynamic operating condition disturbances, the process includes: acquiring the equipment topology of the water treatment system based on the cloud platform, and identifying key processing units and parameter-sensitive areas in the system based on the topology; using the key processing units and parameter-sensitive areas as a first set of sensing points; determining the environmental disturbance factors of the water treatment system based on process specifications, and analyzing the influence path of environmental disturbance factors on process parameters in conjunction with the equipment topology; using the nodes on the influence path of environmental disturbance factors on process parameters as a second set of sensing points; and optimizing and guiding the deployment of the edge sensing terminals based on the first and second sets of sensing points.

6. The cloud-edge-device collaborative intelligent control method for a light-out water treatment plant as described in claim 1, characterized in that, The deployed edge sensing terminals collect process parameter data of each process unit under dynamic operating condition disturbances, including: monitoring the operating condition disturbances of the water treatment system for a target period and identifying the time-frequency characteristics of the dynamic operating condition disturbances based on the monitoring results; determining the data sampling rate configuration range of the edge sensing terminals based on the time-frequency characteristics and adaptively configuring the parameters of the edge sensing terminals based on the configuration range; driving the edge sensing terminals to perform wide-area synchronous monitoring of the system based on the parameter configuration results and collecting water quality parameters, equipment operating parameters, and energy consumption parameters of each process unit based on the monitoring results; and aggregating the collected process parameter data to the cloud platform in real time based on the 5G low-latency communication links pre-established between each edge sensing terminal and the cloud platform.

7. The cloud-edge-device collaborative intelligent control method for a light-out water treatment plant as described in claim 1, characterized in that, The process parameter data feature extraction and fusion analysis to determine the actual operating status of the water treatment system includes: acquiring collected process parameter data and aligning the data spatiotemporally on a digital twin model; dynamically rendering water quality fluctuations and energy consumption changes at each sensing point on the digital twin model based on the spatiotemporal alignment results, and identifying material flow characteristics and energy coupling modes between different units in the system based on the rendering results; analyzing the coupling relationships between units based on the material flow characteristics and energy coupling modes, and dynamically compensating and correcting the process parameter data at each sensing point based on the coupling relationships; and aggregating and statistically analyzing the actual water quality, energy consumption, and equipment status data at each sensing point based on the correction results to generate the actual operating status of the water treatment system.

8. The cloud-edge-device collaborative intelligent control method for a light-out water treatment plant as described in claim 1, characterized in that, The process of determining the deviation parameters between the actual operating state and the optimal operating state, and performing multi-dimensional analysis on the deviation parameters to generate a cloud-edge-device collaborative control strategy, includes: acquiring the actual operating state of the water treatment system, and quantitatively comparing the actual operating state with the corresponding optimal operating state to obtain dynamic deviation parameters; when the amplitude of the dynamic deviation parameters is within a preset safety threshold, determining that the actual operating state of the system is in a stable range, and continuously tracking disturbances in the system based on the determination result; otherwise, determining that the actual operating state of the system exceeds the stable range; dividing the water treatment system into cloud-edge collaborative control areas based on the determination result and the deployment location of the edge sensing terminals, and performing spatiotemporal correlation mining on the process parameter data at each sensing point based on the area division result; and quantifying the material loss and energy loss in each area based on the spatiotemporal correlation mining results. The system's operational instability modes are determined by dynamically identifying instability metrics, including water quality exceeding standards, abnormal energy consumption, and equipment failure. Based on the instability metrics under these modes, potential faulty units in key equipment are identified and intelligently diagnosed. Once no abnormalities are found in the potential faulty units, a cloud-edge-device collaborative control mechanism is activated. Based on the activation results, a pre-trained water quality-energy consumption coupling model is retrieved from the decision knowledge base, with the effluent quality compliance rate as the core control indicator. The model is used to assess the safety boundaries of process parameter data at each sensing point. Based on the safety boundary assessment results, the corresponding control quantities for adjusting the water quality, energy consumption, and equipment status at each sensing point to the optimal operating range are determined. The control quantities for each region are optimized and integrated to generate a cloud-edge-device collaborative control strategy.

9. The cloud-edge-device collaborative intelligent control method for a light-out water treatment plant as described in claim 1, characterized in that, The adaptive collaborative control of intelligent actuators based on the cloud-edge-device collaborative control strategy includes: acquiring the cloud-edge-device collaborative control strategy and decoding the strategy to obtain a collaborative control instruction set for each intelligent actuator; applying dynamic amplitude limiting constraints to the execution units of each intelligent actuator based on the collaborative control instruction set, and arranging the timing logic of each execution unit based on the constraint results to generate a collaborative control sequence for the intelligent actuators; and performing closed-loop regulation of each intelligent actuator based on the collaborative control sequence to achieve unattended operation and real-time optimization of multi-unit collaboration in the water treatment "lights-out" plant.

10. A cloud-edge-device collaborative intelligent control platform for a lights-out water treatment plant, characterized in that: The platform includes: a memory, a processor, and a cloud-edge-device collaborative intelligent control program for a water treatment darkroom stored in the memory and executable on the processor. The cloud-edge-device collaborative intelligent control program for the water treatment darkroom is configured to implement the steps of the cloud-edge-device collaborative intelligent control method for a water treatment darkroom as described in any one of claims 1 to 9.

Images