Digital-twin-based driving energy efficiency and safety virtual-real fusion closed-loop management method

Abstract

The present application relates to the technical field of equipment energy efficiency management, and discloses a driving energy efficiency and safety virtual-real fusion closed-loop management method based on digital twinning. The method comprises the following steps: acquiring multi-source heterogeneous data of a physical environment, and constructing a digital twinning model of a physical space; then extracting equipment operation and environment state characteristic parameters, and constructing an information space behavior simulation model by fusing historical operation data; performing dynamic situation awareness simulation on equipment operation behavior based on the model, and generating a virtual behavior simulation result. The result is mapped to the digital twinning model, and a static safety evaluation value is obtained in combination with real-time safety monitoring data; an environment disturbance variable is acquired, and a dynamic safety prediction value is obtained by inputting the prediction model. Driving energy efficiency optimization parameters and safety control instructions are generated based on the fusion result of the two, the physical space equipment operation state is corrected in a closed loop, and effective linkage between the physical and virtual spaces is realized.

Description

Technical Field

[0001] This invention relates to the field of equipment energy efficiency management technology, specifically to a closed-loop management method for drive energy efficiency and safety based on digital twins, integrating virtual and real data. Background Technology

[0002] In industrial production and energy supply, the energy efficiency and operational safety of equipment drive systems are directly related to overall operating costs and production stability. Currently, equipment drive management relies heavily on real-time physical monitoring and manual experience-based adjustments. This approach has significant limitations when facing complex operating conditions. Equipment operation data in the physical environment is scattered, encompassing various parameters such as temperature, pressure, and speed. Inconsistent data source formats and collection frequencies make data integration difficult and fail to comprehensively reflect the actual operating status of the equipment. Traditional management methods often rely on static analysis of historical data, failing to capture dynamic changes during equipment operation in real time. When faced with environmental fluctuations such as voltage fluctuations and sudden load changes, it is difficult to quickly predict potential safety risks, often resulting in reactive measures only after a failure occurs. This not only affects production continuity but may also lead to greater economic losses due to the escalation of the failure.

[0003] In terms of energy efficiency optimization, existing methods lack precise simulation of equipment operating behavior and cannot accurately assess the impact of different operating parameters on energy efficiency. This results in energy efficiency optimization measures lacking specificity and making it difficult to achieve coordinated management of energy efficiency and safety. For example, in a motor drive system, adjusting the operating state based solely on a single parameter may improve short-term energy efficiency while ignoring the risk of temperature accumulation during long-term operation, potentially leading to equipment overload failure. Conversely, overemphasizing safety redundancy can result in conservative operating parameters and energy waste. Furthermore, in traditional management systems, there is a lack of effective linkage between the operating status of physical equipment and the virtual analysis model. Virtual simulation results cannot be fed back to the physical system for real-time adjustments, creating a management gap between "data acquisition - simulation analysis - decision execution." This leads to low management efficiency and fails to meet the comprehensive demands of modern industrial production for high efficiency, safety, and energy conservation. Summary of the Invention

[0004] The purpose of this invention is to provide a closed-loop management method for drive energy efficiency and safety based on digital twins, thereby solving the problems mentioned in the background art.

[0005] To achieve the above objectives, this invention provides a closed-loop management method for drive energy efficiency and safety based on digital twins, the method comprising:

[0006] Acquire multi-source heterogeneous data of the physical environment, and construct a digital twin model of the physical space based on the multi-source heterogeneous data;

[0007] Based on the digital twin model, the equipment operation characteristic parameters and environmental state characteristic parameters are extracted, and historical operation data is integrated to construct a behavioral simulation model of the information space;

[0008] Based on the aforementioned behavior simulation model, dynamic situational awareness simulation of equipment operation behavior is performed to generate behavior simulation results in virtual space.

[0009] The behavioral simulation results are mapped to the digital twin model, and a security situation assessment is performed in conjunction with the real-time collected security monitoring data to obtain a static security assessment value.

[0010] Obtain environmental disturbance variables, input the environmental disturbance variables into the prediction model to predict the evolution of safety risks, and obtain dynamic safety prediction values;

[0011] Based on the fusion result of the static safety assessment value and the dynamic safety prediction value, drive energy efficiency optimization parameters and safety control instructions are generated.

[0012] Based on the aforementioned drive energy efficiency optimization parameters and safety control instructions, the operating status of equipment in the physical space is corrected in a closed loop.

[0013] Preferably, the step of acquiring multi-source heterogeneous data of the physical environment and constructing a digital twin model of the physical space based on the multi-source heterogeneous data includes:

[0014] The multi-source heterogeneous data is spatiotemporally aligned to obtain a spatiotemporally synchronized sensor data stream;

[0015] Based on the device topology and spatial location constraints, the spatiotemporally synchronized sensor data stream is mapped to form device entity nodes.

[0016] Based on the device entity nodes, extract the energy transfer paths and material flow channels between devices, and construct physical connection edges;

[0017] Based on the device entity nodes and physical connection edges, environmental spatial mesh information is integrated to generate a digital twin model of the physical space.

[0018] Preferably, the behavioral simulation model for constructing the information space by integrating historical operational data includes:

[0019] Extract equipment operation command sequences and operating condition switching records from historical operation data;

[0020] Based on the sequence of equipment operation instructions, classify the operation behavior modes and label the corresponding operating status tags;

[0021] Based on the aforementioned operating condition status labels, a mapping rule base for operation behavior patterns and equipment operating status is established.

[0022] By combining the operational characteristic parameters of the device entity nodes, the mapping rule base is injected into the digital twin model to generate a behavioral simulation model of the information space.

[0023] Preferably, the step of performing dynamic situational awareness simulation of device operation behavior to generate behavioral simulation results in virtual space includes:

[0024] The operation command stream in the physical space is captured in real time, and the operation command stream is input into the behavior simulation model;

[0025] Based on the mapping rule base, the execution process of the operation instruction flow in the digital twin model is simulated;

[0026] The response logic of the device entity nodes is dynamically adjusted based on the real-time changes in environmental state characteristic parameters.

[0027] Output the simulated running trajectory and state transition sequence of the device entity nodes to generate the behavioral simulation results of the virtual space.

[0028] Preferably, the step of performing a security situation assessment by combining real-time collected security monitoring data to obtain a static security assessment value includes:

[0029] Extract abnormal equipment operation indicators and environmental risk factors from safety monitoring data;

[0030] Map the abnormal equipment operation indicators to the corresponding equipment entity nodes in the digital twin model;

[0031] Based on the spatial distribution of the environmental risk factors, the risk level of the environmental spatial grid in the digital twin model is calculated;

[0032] Based on the correlation of the operating status of the device entity nodes and the risk level of the environmental spatial grid, a multi-dimensional security situation fusion calculation is performed to generate the static security assessment value.

[0033] Preferably, the step of inputting the environmental disturbance variables into the prediction model to predict the evolution of safety risks and obtain dynamic safety prediction values ​​includes:

[0034] Construct a time-series-based risk propagation network, wherein the nodes of the risk propagation network correspond to the device entity nodes;

[0035] The environmental disturbance variables are decomposed into input vectors that are transmitted to each node of the risk propagation network;

[0036] Risk transmission weights are trained based on historical accident chain data, and the diffusion path of the input vector in the risk propagation network is calculated.

[0037] Output the risk probability distribution of each device entity node within a preset future time period to generate the dynamic security prediction value.

[0038] Preferably, the step of generating drive energy efficiency optimization parameters and safety control instructions based on the fusion result of the static safety assessment value and the dynamic safety prediction value includes:

[0039] Calculate the deviation matrix between the static security assessment value and the dynamic security prediction value in the spatiotemporal dimension;

[0040] Based on the abnormal regions in the deviation matrix that exceed a preset threshold, locate the target device entity node and the associated environmental space grid;

[0041] Extract the historical optimal values ​​of the operating characteristic parameters of the target device entity node to generate the drive energy efficiency optimization parameters;

[0042] The safety control instructions are generated based on the risk level change trend of the associated environmental spatial grid.

[0043] Preferably, before calculating the deviation matrix between the static security assessment value and the dynamic security prediction value in the spatiotemporal dimension, the method further includes:

[0044] The static safety assessment values ​​are discretized and encoded according to the environmental spatial grid to obtain the first assessment matrix;

[0045] The dynamic security prediction values ​​are normalized according to the same spatial grid and time slice to obtain the second prediction matrix;

[0046] The deviation matrix is ​​obtained by calculating the element-by-element difference between the first evaluation matrix and the second prediction matrix.

[0047] Preferably, the closed-loop correction of the device operating status in the physical space includes:

[0048] The device operating parameter settings in the drive energy efficiency optimization parameters are analyzed and sent to the actuators in the physical space;

[0049] The safety control commands are converted into equipment control protocol command streams to dynamically adjust the operating logic of the actuators;

[0050] The status correction values ​​fed back by the actuator are collected in real time, and the operating characteristic parameters of the corresponding device entity nodes in the digital twin model are updated.

[0051] Preferably, after updating the operational characteristic parameters of the corresponding device entity node in the digital twin model, the method further includes:

[0052] Based on the updated device entity node operating characteristic parameters, recalculate the energy transfer efficiency of the associated physical connection edges;

[0053] Based on the recalculated energy transfer efficiency, the mapping rule base in the behavioral simulation model of the information space is iteratively optimized.

[0054] Compared with the prior art, the beneficial effects of the present invention are:

[0055] By constructing a digital twin model of the physical space, effective integration of multi-source heterogeneous data from the physical environment is achieved. This breaks down the barriers of traditional data fragmentation and incompatible formats, enabling a comprehensive and realistic mapping of the operating status of physical equipment. Managers can intuitively grasp complete information about the equipment, from macroscopic operating parameters to microscopic component states, avoiding biases in equipment status judgments caused by missing or fragmented data. Based on the equipment operating characteristic parameters and environmental state characteristic parameters extracted from the digital twin model, combined with a behavioral simulation model constructed from historical operating data, the model can accurately reproduce the equipment's operating behavior under different working conditions. Instead of relying on single static data for analysis, it uses dynamic situational awareness simulation to capture subtle changes in the equipment's operation in real time, such as parameter fluctuation trends and component coordination states. This allows for more accurate identification of abnormal precursors in equipment operation, early detection of potential problems, and a shift away from the traditional reactive approach to fault response in management.

[0056] By mapping behavioral simulation results to a digital twin model and combining them with real-time safety monitoring data to perform a safety situation assessment, the resulting static safety assessment value reflects the current stable state of the equipment. Meanwhile, the dynamic safety prediction value obtained by introducing environmental disturbance variables and using a predictive model to predict the evolution of safety risks compensates for the inability of static assessment to cope with environmental changes. The integration of the two achieves a comprehensive assessment of the equipment's safety status, considering both the safety of the current state and the potential risks brought about by future environmental changes, such as the impact of disturbances like voltage fluctuations and load changes on equipment safety. This makes the safety assessment more comprehensive and accurate, avoiding safety management loopholes caused by focusing only on static safety while neglecting dynamic risks.

[0057] In terms of coordinated energy efficiency and safety management, the system generates driving energy efficiency optimization parameters and safety control commands based on the fusion of static safety assessment values ​​and dynamic safety prediction values. This achieves a balance between the two, preventing situations where excessive pursuit of energy efficiency sacrifices safety, or excessive emphasis on safety wastes energy. For example, when adjusting equipment operating parameters, it can improve energy utilization efficiency based on energy efficiency optimization parameters while ensuring that the equipment operates within safety thresholds through safety control commands, achieving simultaneous progress in energy efficiency improvement and safety assurance. Furthermore, by performing closed-loop correction of the physical equipment operating status based on driving energy efficiency optimization parameters and safety control commands, a complete management cycle of "physical equipment - digital twin - simulation analysis - decision execution - physical correction" is constructed. This allows virtual analysis results to be promptly transformed into actual operation commands, adjusting the physical equipment operating status in real time. This eliminates the disconnect between physical and virtual systems in traditional management, ensuring that management decisions are quickly implemented, continuously optimizing equipment operating status, and improving overall management efficiency. It is suitable for various scenarios with high requirements for equipment driving energy efficiency and safety, such as industrial production lines and energy supply systems, and can better meet the comprehensive needs of modern production for high efficiency, safety, and energy saving. Attached Figure Description

[0058] Figure 1 This is a schematic diagram illustrating the working principle of the digital twin-based closed-loop management method for driving energy efficiency and safety, which integrates virtual and real elements, as described in this invention.

[0059] Figure 2 Flowchart for building a digital twin model;

[0060] Figure 3 Flowchart for constructing a behavioral simulation model;

[0061] Figure 4 A flowchart for a security situation assessment;

[0062] Figure 5 A flowchart for generating energy efficiency and safety control commands. Detailed Implementation

[0063] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0064] Please see Figure 1 This invention provides a closed-loop management method for drive energy efficiency and safety based on digital twins, the method comprising:

[0065] Multi-source heterogeneous data of the physical environment is acquired, including equipment operating parameters, environmental status information, and safety monitoring data. A digital twin model of the physical space is constructed based on this multi-source heterogeneous data. This model forms a dynamic mapping of the physical space by integrating equipment entities and their connections. Subsequently, equipment operating characteristic parameters and environmental status characteristic parameters are extracted from the digital twin model and fused with historical operating data to construct a behavioral simulation model of the information space. Based on the behavioral simulation model, dynamic situational awareness simulation of equipment operation behavior is performed, generating behavioral simulation results for the virtual space. The behavioral simulation results are mapped back to the digital twin model, and a safety situation assessment is performed in conjunction with real-time acquired safety monitoring data to obtain a static safety assessment value. Simultaneously, environmental disturbance variables are acquired and input into a prediction model to predict the evolution of safety risks, obtaining dynamic safety prediction values. Based on the fusion result of the static safety assessment value and the dynamic safety prediction value, drive energy efficiency optimization parameters and safety control commands are generated. Finally, based on the drive energy efficiency optimization parameters and safety control commands, a closed-loop correction is performed on the equipment operating status of the physical space to achieve synergistic optimization of energy efficiency and safety.

[0066] Example 1: See Figure 2 The heterogeneous data from multiple sources in the physical environment originates from the following systems: First, equipment operation data acquired from programmable logic controllers (PLCs) and distributed control systems (DCS), including parameters such as motor current, voltage, power factor, speed, outlet pressure, flow meter readings, and bearing temperature. This data is uploaded in real time using different sampling frequencies and communication protocols (such as ModbusTCP and PROFIBUS). Second, environmental status data is collected by IoT sensor nodes deployed in different locations within the plant, including ambient temperature and humidity sensors, atmospheric pressure sensors, and gas concentration sensors used to detect potential hazardous gas leaks. This data is transmitted via a wireless LoRaWAN network, and its timestamps have an inherent microsecond delay compared to equipment data collected via wired networks. Finally, work order information, equipment maintenance logs, and manually entered operational condition notes from the Manufacturing Execution System (MES) constitute the system operation log data. This data is more macroscopic in terms of time sequence and is mainly unstructured text information.

[0067] Spatiotemporal alignment of the aforementioned multi-source heterogeneous data is a prerequisite for building an accurate model. Timestamp synchronization processing is applied to all incoming data streams, using a Network Time Protocol (NTP) server to calibrate the clocks of all data acquisition terminals. Data with different sampling frequencies are uniformly resampled to a common time reference point using interpolation algorithms (such as linear interpolation), for example, one data point per second. Minor jitter delays caused by network transmission are compensated for using data caching and time-series alignment algorithms to ensure that all sensor readings represent the system state at the same moment in the time dimension. Spatial coordinate alignment strongly correlates each data point with its specific location in physical space. The plant area defines a precise coordinate system using computer-aided design (CAD) drawings, where each piece of equipment and each sensor has unique spatial coordinates (X, Y, Z). The data processing flow binds each piece of equipment operation data record to the coordinates of its source equipment; similarly, each environmental sensor reading is also bound to the physical coordinates of that sensor. Through this process, the originally scattered and heterogeneous data streams are integrated into a spatiotemporally synchronized sensor data stream, in which each data record has a unified timestamp and a clear spatial attribute.

[0068] Based on the equipment topology and spatial constraints, the spatiotemporally synchronized sensor data streams are mapped to entities. The equipment topology is derived from the Piping and Instrumentation Diagram (P&ID) and the plant's single-line diagram. These engineering design drawings clearly define the functional connections between equipment. For example, the outlet of pump A is connected to the shell-side inlet of heat exchanger E-201 via pipe P-101, while the tube-side outlet of heat exchanger E-201 is connected to cooling tower C-401 via valve V-301. Spatial constraints are derived from the plant's 3D layout model, which specifies the exact placement and geometry of these devices and pipes in the actual physical space. The data processing system analyzes these drawings and models to construct a topology network containing all equipment, pipes, and valves as entities. Subsequently, the aforementioned spatiotemporally aligned data streams are injected into this network. For example, all current, pressure, and temperature data identified as "pump A" are associated with the entity node representing "pump A" in the topology network; all temperature, pressure, and flow data located on pipe P-101 are associated with that pipe entity. This process creates device entity nodes rich in real-time data. Each node not only contains its static attributes (such as device model and design parameters), but also continuously aggregates its dynamic operating parameters.

[0069] Based on these physical nodes, it is necessary to extract the energy transfer paths and material flow channels between devices. The analysis of energy transfer paths focuses on tracing the flow and conversion of electrical and thermal energy. For example, electrical energy flows from the distribution cabinet to the motor, driving the pump, converting electrical energy into mechanical energy; the pump propels the fluid (material) through the pipes, converting mechanical energy into pressure and kinetic energy; the fluid enters the heat exchanger for heat exchange, transferring the heat energy it carries to the process fluid on the other side. This series of energy conversions and transfers constitutes the energy transfer path from the power source to the final process system. Material flow channels trace the physical flow trajectory of the working medium (such as water, compressed air, chemical solutions), from the source (water tank, air compressor), through processing equipment (filters, heat exchangers), transport equipment (pumps, compressors), and finally to the point of use or discharge. By analyzing the topology network, the system automatically identifies these paths and channels and constructs physical connection edges. Each physical connection edge has directionality (indicating the direction of energy or matter flow) and type attributes (such as "electrical power supply", "hydraulic connection", "thermal coupling"), and carries the current energy or matter flow data.

[0070] Based on the physical nodes and physical connection edges of the equipment, environmental spatial mesh information is integrated. The three-dimensional space of the entire plant area is divided into regular cubic mesh units. The size of each mesh unit can be set according to accuracy requirements, for example, 1 meter × 1 meter × 1 meter per unit. Each mesh unit stores environmental state data obtained from the aggregation or interpolation calculation of all environmental sensors (temperature, humidity, gas concentration) within its spatial range. In the digital twin model, these environmental spatial meshes exist as a background field, fused with the equipment physical nodes located in specific meshes and the physical connection edges (pipes) that cross the meshes. For example, the node data of a pump located in a high-temperature mesh area will incorporate the environmental temperature attribute of that mesh; the connection edge data of a pipe crossing a high-humidity mesh will also consider this environmental factor. Through this fusion, a digital twin model of the physical space is generated. This model is no longer an isolated network of equipment, but a virtual mapping that deeply integrates the equipment operating status, system energy / material flow, and precise spatiotemporal environmental background, comprehensively and dynamically reflecting the real state of the physical space, laying a solid foundation for subsequent simulation, evaluation, and optimization.

[0071] Example 2: See Figure 3On the ion-exchange membrane electrolysis production line of a chlor-alkali chemical plant, an information space behavioral simulation model and dynamic situational awareness simulation were implemented. This production line includes rectifier units, electrolyzer arrays, brine circulating pumps, chlorine treatment towers, and supporting automatic control valve systems, with historical operating data accumulated over three years. The information space behavioral simulation model was constructed by integrating this historical operating data. Equipment operation command sequences and operating condition switching records were extracted from the historical operating database. The operation command sequences originated from the operation logs of the distributed control system (DCS), recording all manual operations and automatic control commands with millisecond-level timestamps, such as "open the feed valve of electrolyzer A to 65%" and "start the current ramp-up program of rectifier unit B". Operating condition switching records were extracted from the alarm and event history database, marking key transition points in equipment operating modes, such as "electrolyzer switches from start-up mode to constant current operation mode" and "circulating pump switches to standby pump due to low flow alarm". The data processing system cleaned and analyzed these records, removing invalid operations and false alarm entries.

[0072] Based on the sequence of equipment operation commands, operation behavior patterns were categorized, and density clustering algorithms (such as DBSCAN) were used for pattern recognition. Analysis revealed that operation commands often appear in combinations, forming reproducible patterns. For example, when increasing the output of the electrolyzer, the operator typically executes the command sequence "increase rectifier current setpoint → fine-tune feed valve opening → start auxiliary circulation pump," which is identified as the "production increase mode." Similarly, "rectifier unit soft start procedure" and "emergency load reduction procedure" are also identified as independent modes. Each identified operation behavior pattern category is labeled with a corresponding operating condition label. The operating condition label is derived from a real-time database snapshot of the DCS, recording the steady-state operating parameters reached by the equipment after the mode execution. For example, the label for "production increase mode" is "high current density operation" and "feed flow rate > 120m³ / h". 3 / h".

[0073] Based on operating condition labels, a rule base for mapping operating behavior patterns to equipment operating states is established. Each rule uses an "IF-THEN" structure to define the causal relationship between the operating mode and the state change. For example, rule R1: "IF execute the production increase mode instruction sequence THEN increase the electrolytic cell current density to 4.5kA / m²". 2 ±0.1kA / m 2 "The feed valve opening is increased to 115% of the set value." Rule parameters are obtained by statistically analyzing the distribution characteristics of equipment parameters under the same patterns in historical data. The rule base contains hundreds of such rules, covering all identified operating patterns and their impact on critical equipment such as rectifiers, electrolyzers, pumps, and valves.

[0074] By combining the operational characteristic parameters of the equipment entity nodes, a mapping rule base is injected into the digital twin model. In the constructed digital twin model of the electrolysis production line, each equipment entity node (such as an electrolytic cell unit or rectifier transformer) continuously receives real-time operational characteristic parameters from the physical space, including temperature, pressure, current, and voltage. The rule injection process binds mapping rules to specific equipment nodes. For example, all rules involving changes in electrolytic cell current are only associated with the node representing the electrolytic cell in the digital twin model. The rule engine is integrated into the digital twin platform, enabling the model to parse operation commands and trigger the execution of corresponding rules. This generates a behavioral simulation model of the information space, which has the ability to predict changes in equipment state based on input commands.

[0075] Dynamic situational awareness simulation of equipment operation behavior is performed to generate behavioral simulation results in a virtual space. The operation command flow in the physical space is captured in real time, and millisecond-updated operation command events are obtained from the DCS control layer via the OPCUA interface. For example, the control system issues a command to "increase the current setpoint of electrolyzer C to 12kA". This command flow is immediately input into the behavioral simulation model. Based on a mapping rule base, the execution process of the operation command flow in the digital twin model is simulated. The rule engine parses the command and matches it to the "electrolyzer current adjustment" rule subset. The simulation engine executes the rules in the digital twin model: first, it calculates the impact of a step change in current on the thermal balance of the electrolyzer and predicts the temperature rise curve; then, it simulates the correlation effect, such as the increase in chlorine production due to the increase in current, which triggers a pressure chain response in the chlorine treatment tower. The entire process is accelerated in the virtual space, simulating the time history of the actual equipment response.

[0076] The response logic of the equipment entity nodes is dynamically adjusted based on real-time changes in environmental state characteristic parameters. These parameters are updated in real-time through an environmental grid in the digital twin model. For example, when the environmental temperature and humidity sensor detects a sudden increase in workshop temperature, this change is mapped to the environmental grid cell containing the electrolyzer. The behavioral simulation model dynamically corrects relevant rule parameters: In high-temperature environments, the electrolyzer's heat dissipation efficiency decreases, therefore the predicted temperature rise curve in the original rule is corrected, with the temperature rise slope increasing by 15%. Similarly, if the environmental grid detects an abnormal hydrogen concentration, rules related to the electrolyzer's safety interlock will be activated, predicting the possible emergency shutdown sequence.

[0077] The simulation engine outputs the simulated operating trajectory and state transition sequence of the physical nodes of the output equipment. During the simulated current increase in the electrolyzer, the simulation engine records the state changes of the virtual electrolyzer nodes: the initial state is "stable operation, current 11.8kA, temperature 85℃"; the predicted state 3 seconds after command execution is "current rises to 11.95kA, temperature 86.2℃"; the predicted state 8 seconds later is "current 12kA, temperature 87.5℃ tending to stabilize". Simultaneously, the state transition sequence is recorded: "normal → current increase → temperature compensation → new steady state". For the affected chlorine treatment tower nodes, the engine outputs their pressure fluctuation trajectory and the transition sequence "equilibrium → pressure increase → regulating valve response → restoration of equilibrium". These simulated operating trajectories and state transition sequences constitute the behavioral simulation results in the virtual space, stored in the twin platform database as time-series data. This behavioral simulation result is continuously updated; whenever a new operating command is input or environmental parameters change, the simulation model immediately recalculates and outputs the latest prediction. For example, when the actual ambient temperature change exceeds a threshold, even without a new operating command, the model will automatically rerun the simulation and update the predicted equipment state trajectory.

[0078] Example 3: See Figure 4 A safety situation assessment is performed by combining real-time collected safety monitoring data to obtain a static safety assessment value. The safety monitoring data comes from multiple independent safety instrumented systems (SIS) and gas detection systems. Equipment operation anomaly indicators and environmental risk factors are extracted from this data. Equipment operation anomaly indicators include, but are not limited to: cell voltage deviation of the electrolyzer (the difference between actual and theoretical voltage), harmonic distortion rate of the rectifier cabinet, effective value of bearing vibration acceleration of the circulating pump, and difference between the motor winding temperature and its rated value. Environmental risk factors include: hydrogen concentration sensor readings installed around the electrolyzer, trace leak detection values ​​at chlorine pipeline flange connections, percentage of oxygen content in the ambient air, and coordinates and temperature values ​​of overheated hotspots on the equipment surface monitored by thermal imaging cameras.

[0079] The abnormal operating indicators of equipment are mapped to corresponding equipment entity nodes in the digital twin model. This is a precise matching process based on unique equipment identifiers. For example, the cell voltage deviation value of an electrolytic cell numbered "EC-102" is associated with the virtual entity node "Electrolytic Cell-102" with the same ID in the digital twin model, serving as a dynamic risk attribute for that node. Similarly, the harmonic distortion rate of the rectifier cabinet "RMU-01" is bound to its corresponding virtual node. In this way, abnormal states in the physical world are accurately projected into the virtual model, enabling the digital twin to reflect the health and abnormal conditions of the physical entities in real time.

[0080] Based on the spatial distribution of environmental risk factors, the risk level of the environmental spatial grid in the digital twin model is calculated. The digital twin model divides the factory workshop into 1m x 1m x 1m three-dimensional grid cells. The readings of each hydrogen concentration sensor and gas leak detector are assigned to one or more grid cells according to their physical coordinates. For multiple sensors within the same grid, the maximum value or weighted average of their readings is taken as the representative value for that grid. The risk level is calculated based on a preset threshold range. For example, for hydrogen concentration, Level 0 (Safe): <25% LEL (Lower Explosive Limit); Level 1 (Caution): 25%~50% LEL; Level 2 (Warning): 50%~75% LEL; Level 3 (Danger): ≥75% LEL. Each grid cell is assigned a discrete risk level value based on the range in which the sensor readings within it fall.

[0081] Based on the operational status correlations of equipment nodes and the risk levels of the environmental spatial grid, a multi-dimensional security posture fusion calculation is performed to generate a static security assessment value. Equipment operational status correlations describe the functional dependencies and physical proximity relationships between nodes. For example, an electrolyzer node has a strong correlation with the rectifier cabinet node supplying its power, the circulating pump node supplying brine, and the environmental grid where its physical location is situated. The fusion calculation is not a simple weighted average but considers the propagation and superposition effects of correlation strength. The system defines an influence weight wij for each type of correlation, representing the degree of influence of node i's risk on node j. The static security assessment value S_j for a given equipment node or grid j can be expressed as the aggregation of its own risk and the risks of all its associated nodes:

[0082]

[0083] in: Represents a node The static security assessment value is a comprehensive risk score. Represents a node Its own risk value comes from the value converted from its abnormal indicators or environmental risk level. Represents nodes The set of all other related nodes. Represents a node The risk is transmitted to the node The weighting coefficients are predefined based on the type and strength of the connections between nodes (such as electrical connections, material flow connections, physical proximity). Indicates from node The risk value that is being transmitted. and It is a coefficient that balances the importance of its own risks and the risks of transmission, satisfying... In this way, an anomaly in a rectifier cabinet not only raises its own assessment value but also, according to weights, affects the assessment values ​​of all electrolyzer nodes it supplies; similarly, a high hydrogen risk level in a grid will increase the assessment values ​​of all device nodes within and adjacent to it. Ultimately, each device node and grid cell calculates a quantified S_j value, and the set of all these values ​​constitutes a static assessment of the current safety state of the entire system.

[0084] Environmental disturbance variables are acquired and input into a prediction model to predict the evolution of safety risks, resulting in dynamic safety prediction values. Environmental disturbance variables refer to internal or external event parameters that may cause significant changes in the system state. They are characterized by their potential for sudden occurrence and the spread of their impact over time. In this embodiment, environmental disturbance variables include: strong wind and pressure change data for the next two hours provided by weather forecasts; power grid frequency fluctuation warnings for the next one hour issued by the power grid dispatch center; and estimates of the location and diffusion rate of suspicious smoke outside the plant area detected through image recognition.

[0085] A time-series-based risk propagation network is constructed, where each node corresponds to an entity node in the digital twin model. Edges in the network represent possible risk propagation paths between nodes and their time delay attributes. For example, the risk of grid frequency fluctuations first propagates to the rectifier cabinet node, and after a very short delay, further propagates to the electrolytic cell node; the risk of building swaying caused by strong winds propagates to all elevated tank and pipe support nodes; peripheral smoke indicates a potential fire risk, and its propagation path is modeled along the distribution of combustibles and wind direction.

[0086] Environmental disturbance variables are decomposed into input vectors that propagate to each node in the risk propagation network. Each disturbance variable is transformed into an initial risk input for one or more nodes. For example, power grid frequency fluctuation warnings are transformed into initial risk values ​​for all rectifier cabinet nodes, the magnitude of which is proportional to the predicted amplitude of the frequency deviation. Strong wind data are transformed into initial risk values ​​for all open-air equipment nodes and tall building nodes. Suspicious smoke information is transformed into initial risk values ​​for all grids and equipment within a certain fan-shaped area downwind of the suspected smoke.

[0087] Risk propagation weights are trained based on historical accident chain data, and the diffusion path of the input vector in the risk propagation network is calculated. Historical accident chain data records various abnormal events and their eventual failure or accident sequences that occurred in the factory area in the past. Machine learning algorithms (such as graph neural networks based on time series graphs, i.e., time series graph neural networks (GNN4TS), which can be divided into "graph neural networks for different time series analysis tasks," "GNNs for time series prediction," "GNNs for time series anomaly detection," "GNNs for time series classification," and "GNNs for missing data imputation," primarily used in time series analysis to overcome the limitations of traditional methods in capturing complex interactions between variables by integrating time information and graph structure relationships) analyze these historical chains, training the propagation weight wij and propagation delay parameter τij for each edge in the risk propagation network. During prediction, the initial risk input vector is injected into the network nodes, and then the propagation process of risk along the network edges with time steps is simulated. At each time step, the risk value of node j is calculated and updated based on its current value, the risk value of the incoming connection, and the corresponding weights and delays. This process simulates the dynamics of risk spreading from its source to related systems.

[0088] The system outputs the risk probability distribution of each device node within a preset future time period, generating dynamic safety prediction values. After the risk propagation network simulation is completed, for each future time point (e.g., every 5-minute interval), each device node will output a risk value, representing the probability or degree that the node is in a high-risk state at that moment. All these time-series arranged node risk values ​​constitute the dynamic safety prediction value. It describes how the system's future safety state might evolve under the influence of environmental disturbance variables, for example, "It is predicted that in 15 minutes, the risk probability of rectifier cabinet RMU-01 will rise to 0.7, potentially triggering an early warning; it is predicted that in 40 minutes, the risk level of the electrolysis workshop grid in the downwind area will generally increase."

[0089] Example 4: See Figure 5 The static safety assessment values ​​reflect the system's current safety status, and the data is organized according to an environmental spatial grid. The environmental spatial grid divides the plant area into rectangular cells on a two-dimensional plane, each cell having a unique row and column coordinate identifier. The static safety assessment values ​​are discretized and encoded according to the environmental spatial grid to obtain the first assessment matrix. The rows and columns of this matrix correspond to the coordinates of the grid cells, and the value of each element in the matrix represents the comprehensive safety assessment score of that grid cell at the current moment. The score is usually a normalized numerical value, for example, ranging from 0 to 1, with higher values ​​indicating greater risk. This encoding process transforms continuous assessment results into structured matrix data, facilitating subsequent quantitative comparison and analysis.

[0090] Dynamic security predictions reflect the trend of the system's security status over a future period. Their data is also based on the same environmental spatial grid structure, but with an added time dimension. The dynamic security predictions are normalized according to the same spatial grid and time slices to obtain the second prediction matrix. Time slices divide the future prediction period into several equally spaced time points, for example, one slice every 5 minutes. For each time slice and each grid cell, the dynamic prediction provides a risk probability value. Normalization ensures that these probability values ​​are scaled in accordance with the evaluation score of the first evaluation matrix, allowing for direct comparison. Therefore, the second prediction matrix is ​​a three-dimensional data structure, but its slices at specific time slices can be considered a two-dimensional matrix with the exact same dimensionality and grid alignment as the first evaluation matrix.

[0091] The deviation matrix is ​​obtained by calculating the element-wise differences between the first evaluation matrix and the second prediction matrix at the same time (usually the current time or the nearest future time). The deviation matrix has the same size as the evaluation and prediction matrices, and the value of each element represents the difference between the evaluation based on the current measured state and the state based on the future prediction for that grid cell. A significant positive deviation may indicate that the risk of the current assessment is higher than that of the trend-based prediction, suggesting the existence of sudden risks that have not been captured by the prediction model. A significant negative deviation may indicate that the prediction model predicts a significant increase in risk, but the current assessment has not yet reflected this trend, which has important early warning significance.

[0092] Based on anomalies exceeding preset thresholds in the deviation matrix, the system locates target equipment entities and their associated environmental spatial grids. The preset thresholds are determined based on historical data and operational experience, with thresholds set for both positive and negative deviations. The system iterates through all elements in the deviation matrix, marking grid cells whose absolute values ​​exceed the thresholds. These cells constitute spatial anomalies. Subsequently, the system queries the digital twin model to identify equipment entities entirely or primarily located within these anomaly grid regions; these nodes are identified as target equipment entities. Simultaneously, these anomaly grid cells are themselves marked as associated environmental spatial grids. For example, if a grid region has severely excessive deviations and contains a critical chlorine compressor unit, that compressor unit is identified as a target equipment entity, and its associated grid region is identified as an associated environmental spatial grid.

[0093] The system extracts the historical optimal values ​​of the operating characteristic parameters of the target equipment entity nodes to generate drive energy efficiency optimization parameters. The historical optimal value does not refer to the theoretical maximum or minimum value, but rather the average or mode of the key operating parameters of the equipment under safe, stable, and efficient operating conditions over a past period (e.g., the past three months). These parameters are recorded in a historical database. For each located target equipment entity node, the system retrieves the historical optimal values ​​of its key parameters from its historical operating data. For example, for the chlorine compressor, its historical optimal values ​​might include: inlet pressure maintained at 0.12 MPa, outlet temperature not exceeding 85°C, and current fluctuation less than 5% of the rated value. These retrieved optimal parameter values ​​are packaged into a set of setpoint recommendations, i.e., drive energy efficiency optimization parameters. The purpose of these parameters is to guide the equipment's operating state back to a known safe and efficient range in the event of safety risk deviations.

[0094] Based on the risk level change trends of associated environmental spatial grids, safety control instructions are generated. These risk level change trends are extracted from dynamic safety prediction values ​​and describe the direction and magnitude of the risk probability evolution for each associated environmental spatial grid over a future period. The system analyzes these trends and generates corresponding pre-control instructions. The instruction generation logic is based on a pre-set strategy library. For example, if the dynamic risk prediction of an associated grid shows that its hydrogen concentration risk probability will continuously rise from 0.3 to 0.8 within the next 15 minutes, the system will generate an instruction to "activate the enhanced ventilation system in this grid area, increasing the air exchange rate to 12 times per hour" according to the strategy. If the predicted trend of another grid shows an increasing fire risk probability, an instruction to "pre-open the water supply valves of the fire sprinkler system in this area to maintain pipeline pressure" will be generated. These instructions aim to proactively intervene in the environment to curb the evolution of predicted risks (see Table 1).

[0095] Table 1: Abnormal Area Location and Response Strategy Table.

[0096]

[0097] Example 5: Parsing and distributing equipment operating parameter settings from drive energy efficiency optimization parameters to actuators in the physical space. Drive energy efficiency optimization parameters are instruction packets transmitted to the control layer in a structured data format, containing specific parameter setting targets for specific equipment. For example, the optimization parameter packet for the chlorine compressor CP-101 includes the following key settings: inlet pressure target value is 0.12MPa (absolute pressure), outlet temperature target value is no more than 85℃, motor current fluctuation amplitude is controlled within ±5% of the rated value, and vibration velocity effective value is maintained below 2.1mm / s. The control system's instruction parsing module receives these parameter packets, verifies their equipment identifiers and numerical validity, and converts them into control signals that the underlying actuators can recognize. For the compressor CP-101, these settings are distributed to its associated anti-surge controller, lubricating oil cooling system actuators, and variable frequency drive unit. These actuators adjust their internal control logic according to the received settings, gradually approaching the target parameters with the equipment's operating state. The entire distribution process is completed through industrial Ethernet and real-time control bus, ensuring the accuracy and timeliness of instruction transmission.

[0098] The safety control commands are transformed into equipment control protocol command streams, dynamically adjusting the operating logic of the actuators. Safety control commands are higher-level control commands targeting environmental risks or equipment interlocking operations. For example, the commands for high-risk grid area D07 are "start the area cooling fan" and "switch the feed regulating valve HV-205 to the standby circuit". These text-described commands need to be transformed into specific, executable equipment control sequences. The command conversion engine first queries the equipment database of the plant control system to determine that "area cooling fan" corresponds to three independent fan devices (FN-01, FN-02, FN-03), and "switch to standby circuit" corresponds to a control command code for regulating valve HV-205. Subsequently, the engine generates a command stream conforming to the control bus protocol: first, it sends a command to switch the control of HV-205 from the current main controller to the standby controller, and executes the operation sequence of fully closing the valve and then gradually opening it to the preset safe opening degree; simultaneously, it sends a start command to the PLC of the three fans and sets the speed to 80% of the rated speed. This series of instructions is issued sequentially according to strict timing requirements, dynamically changing the operating logic and coordination methods of the relevant implementing agencies in order to address the identified risks.

[0099] The system collects real-time feedback on status corrections from the actuators and updates the operational characteristic parameters of the corresponding equipment entity nodes in the digital twin model. Once the actuators begin operation, a sensor network deployed on the equipment continuously monitors changes in their operational status. These changes, known as status corrections, represent the actual effects of executed commands. For example, the sensor on compressor CP-101 reports that its inlet pressure is gradually decreasing from 0.13 MPa to 0.122 MPa, and its outlet temperature is dropping from 88°C to 85.5°C. The valve position feedback signal from regulating valve HV-205 indicates that it is at 65% opening under standby circuit control. The current feedback from zone fans FN-01 / 02 / 03 indicates that it has started and is operating at 78% speed. This real-time data is transmitted back to the digital twin system via the data acquisition interface. Based on the equipment identifier, the system locates the corresponding equipment entity node in the digital twin model and updates the node's operational characteristic parameters with these latest parameter values ​​reflecting the state after command execution. For example, the attribute values ​​such as "inlet pressure" and "outlet temperature" under node CP-101 are updated. This allows digital twin models to synchronize the latest state of physical devices almost in real time after instruction intervention.

[0100] Based on the updated operational characteristic parameters of the equipment entity nodes, the energy transfer efficiency of the associated physical connection edges is recalculated. Changes in equipment status inevitably affect the energy flow of the entire system connected to it. The physical connection edges in the digital twin model represent these energy transfer relationships. For example, after the status of compressor CP-101 is updated, the energy transfer efficiency of the "compressor to chlorine drying tower" connection via pipeline needs to be recalculated. The calculation process is based on the updated compressor outlet pressure and temperature, as well as the measured flow rate of chlorine in the pipeline, combined with the inlet pressure of the drying tower. Using thermodynamic and fluid dynamic principles, the pressure loss and heat loss of this pipeline section under the current operating conditions are calculated, thus obtaining a new value for the energy transfer efficiency. Similarly, after the regulating valve HV-205 is switched to the standby circuit, the flow resistance characteristics of its pipeline change, and the hydraulic transport efficiency of the "brine preheater to electrolyzer" connection edge also needs to be recalculated. These recalculated efficiency values ​​update the attributes of the corresponding physical connection edges, more accurately reflecting the current energy consumption distribution and transmission performance of the system.

[0101] Based on the recalculated energy transfer efficiency, the mapping rule base in the behavioral simulation model of the information space is iteratively optimized. The behavioral simulation model relies on the mapping rule base to predict the impact of operational commands on the system state. The parameters and logic in the rule base need to be consistent with the actual characteristics of the system. When the energy transfer efficiency of physical connection edges changes significantly, it means that the dynamic response characteristics of the system may have changed. For example, if the calculation finds that the energy efficiency from the compressor to the drying tower has decreased by 5% compared to the historical average, this indicates that the system resistance may have increased. The mapping rule base in the behavioral simulation model needs to be iteratively optimized accordingly. The optimization process first analyzes the reasons for the efficiency change, and then adjusts the relevant rules in the rule base that are affected. For example, modifying the gain coefficient of the rule regarding the impact of "increasing the compressor speed" on "outlet pressure", or updating the parameters of the prediction effect of "turning on the cooling fan" on "ambient grid temperature". These adjustments enable the behavioral simulation model to more accurately predict the possible results of equipment operation behavior under the current actual system conditions in subsequent simulations, thereby maintaining the consistency between the digital twin virtual prediction and the real physical world.

[0102] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0103] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A digital-twin-based driving energy efficiency and safety virtual-real fusion closed-loop management method, characterized in that, The method comprises the following steps: acquiring multi-source heterogeneous data of a physical environment, constructing a digital twin model of a physical space based on the multi-source heterogeneous data, including: performing spatio-temporal alignment processing on the multi-source heterogeneous data to obtain a spatio-temporal synchronized sensor data stream; performing entity correlation mapping on the spatio-temporal synchronized sensor data stream according to device topology relationship and spatial position constraints to form device entity nodes; extracting energy transmission paths and material flow channels between devices based on the device entity nodes to construct physical connection edges; integrating environmental space grid information according to the device entity nodes and the physical connection edges to generate the digital twin model of the physical space; extracting device operation characteristic parameters and environmental state characteristic parameters according to the digital twin model, and constructing a behavior simulation model of an information space by fusing historical operation data; performing dynamic situation awareness simulation on device operation behaviors based on the behavior simulation model to generate a behavior simulation result of a virtual space; mapping the behavior simulation result into the digital twin model, and performing safety situation assessment by combining real-time collected safety monitoring data to obtain a static safety assessment value; acquiring environmental disturbance variables, inputting the environmental disturbance variables into a prediction model to perform safety risk evolution prediction to obtain a dynamic safety prediction value; generating driving energy efficiency optimization parameters and safety control instructions based on the fusion result of the static safety assessment value and the dynamic safety prediction value; performing closed-loop correction on device operation states of the physical space according to the driving energy efficiency optimization parameters and the safety control instructions.

2. The digital-twin-based driving energy efficiency and safety virtual-real fusion closed-loop management method according to claim 1, characterized in that, The method of constructing the behavior simulation model of the information space by fusing historical operation data comprises the following steps: extracting device operation instruction sequences and working condition switching records from historical operation data; dividing operation behavior mode categories according to the device operation instruction sequences, and labeling corresponding working condition state tags; establishing a mapping rule library of operation behavior modes and device operation states based on the working condition state tags; injecting the mapping rule library into the digital twin model by combining the running characteristic parameters of the device entity nodes to generate the behavior simulation model of the information space.

3. The digital-twin-based driving energy efficiency and safety virtual-real fusion closed-loop management method according to claim 2, characterized in that, The method of performing dynamic situation awareness simulation on device operation behaviors to generate a behavior simulation result of a virtual space comprises the following steps: real-time capturing of operation instruction streams of the physical space, and inputting the operation instruction streams into the behavior simulation model; simulating the execution process of the operation instruction streams in the digital twin model based on the mapping rule library; dynamically adjusting the response logic of the device entity nodes according to real-time changes of environmental state characteristic parameters; outputting the simulated operation trajectories and state transition sequences of the device entity nodes to generate the behavior simulation result of the virtual space.

4. The digital-twin-based driving energy efficiency and safety virtual-real fusion closed-loop management method according to claim 1, characterized in that, The method of performing safety situation assessment by combining real-time collected safety monitoring data to obtain a static safety assessment value comprises the following steps: extracting device operation abnormality indicators and environmental risk factors from safety monitoring data; mapping the device operation abnormality indicators to corresponding device entity nodes of the digital twin model; calculating the risk levels of environmental space grids in the digital twin model according to the spatial distribution of the environmental risk factors; Based on the running state correlation of the device entity nodes and the risk level of the environmental space grid, multi-dimensional safety situation fusion calculation is performed to generate the static safety evaluation value.

5. The digital-twin-based driving energy efficiency and safety virtual-real fusion closed-loop management method according to claim 4, characterized in that, The safety risk evolution prediction of the environmental disturbance variable input prediction model includes: A risk propagation network based on time series is constructed, and the nodes of the risk propagation network correspond to the device entity nodes; The environmental disturbance variable is decomposed into an input vector conducted to each node of the risk propagation network; According to the historical accident chain data, the risk transmission weight is trained, and the diffusion path of the input vector in the risk propagation network is calculated; The risk probability distribution of each device entity node in the future preset period is output to generate the dynamic safety prediction value.

6. The digital-twin-based driving energy efficiency and safety virtual-real fusion closed-loop management method according to claim 1, characterized in that, Based on the fusion result of the static safety evaluation value and the dynamic safety prediction value, the driving energy efficiency optimization parameter and the safety control instruction are generated, including: The deviation matrix of the static safety evaluation value and the dynamic safety prediction value in the space-time dimension is calculated; According to the abnormal area exceeding the preset threshold in the deviation matrix, the target device entity node and the associated environmental space grid are located; The running feature parameter historical optimal value of the target device entity node is extracted to generate the driving energy efficiency optimization parameter; Based on the risk level change trend of the associated environmental space grid, the safety control instruction is generated.

7. The digital-twin-based driving energy efficiency and safety virtual-real fusion closed-loop management method according to claim 6, characterized in that, Before calculating the deviation matrix of the static safety evaluation value and the dynamic safety prediction value in the space-time dimension, it further includes: The static safety evaluation value is discretely coded according to the environmental space grid to obtain a first evaluation matrix; The dynamic safety prediction value is normalized according to the same space grid and time slice to obtain a second prediction matrix; The deviation matrix is obtained by calculating the element-by-element difference between the first evaluation matrix and the second prediction matrix.

8. The digital-twin-based driving energy efficiency and safety virtual-real fusion closed-loop management method according to claim 1, characterized in that, The closed-loop correction of the device running state of the physical space includes: The device running parameter setting value in the driving energy efficiency optimization parameter is analyzed and issued to the execution mechanism of the physical space; The safety control instruction is converted into a device control protocol instruction stream to dynamically adjust the running logic of the execution mechanism; The state correction amount fed back by the execution mechanism is collected in real time to update the running feature parameters of the corresponding device entity nodes in the digital twin model.

9. The digital-twin-based driving energy efficiency and safety virtual-real fusion closed-loop management method according to claim 8, characterized in that, After updating the running feature parameters of the corresponding device entity nodes in the digital twin model, it further includes: According to the updated device entity node running feature parameters, the energy transmission efficiency of the associated physical connection edge is recalculated; Based on the recalculated energy transmission efficiency, the mapping rule library in the behavior simulation model of the information space is iteratively optimized.

Images