Cloud-edge collaborative control system for industrial parks based on digital twins

Abstract

This invention relates to the field of industrial park control technology, and discloses a cloud-edge collaborative control system for industrial parks based on digital twins, including a cloud engine and multiple edge control nodes. The cloud engine offline constructs a historical steady-state snapshot library, while the edge control nodes retrieve historical steady states locally based on the library and real-time operating conditions. Furthermore, the edge control nodes use the deviation dynamics of the current state and the health characteristic values ​​of the controlled equipment as feedback to dynamically modulate and compensate for the retrieved historical steady states in order to generate the final control command. This invention avoids system regulation oscillations caused by command delays by changing the control method from online optimization calculation that relies on communication to distributed local reproduction based on shared history. At the same time, it gives the system elastic adaptability in the face of unknown shocks and realizes self-sensing and closed-loop compensation of the physical state of the equipment, ensuring the accurate achievement of the control effect.

Description

Technical Field

[0001] This invention relates to a cloud-edge collaborative control system for industrial parks based on digital twins, belonging to the field of industrial park control technology. Background Technology

[0002] In real-world industrial park scenarios where power generation and other factors fluctuate rapidly, the ideal communication assumptions of low latency and high bandwidth between the cloud, edge, and device, upon which the aforementioned architecture relies, are difficult to uphold in reality. This exposes an architectural control mismatch problem. Specifically, when the cloud completes minute-level optimization calculations based on past aggregated data and issues instructions, the physical conditions of the park may have changed again, causing the cloud instructions to become outdated. This exacerbates the chain reaction of system control fluctuations and turns the theoretical benefits of energy saving into physical wear and tear and maintenance costs caused by frequent start-ups and shutdowns of equipment.

[0003] To address this challenge, the industry has attempted to compensate by increasing edge computing power or optimizing communication protocols. However, these improvements still operate within the existing control framework based on communication latency and have failed to change the situation where a tightly coupled physical system with a millisecond-level response is constrained by a minute-level decision cycle.

[0004] Specifically, existing technologies suffer from the following shortcomings: 1. The decision-making basis of the control system lags behind the real-time state of the physical world in the time dimension, leading to a mismatch risk between control behavior and control objectives; 2. In the absence of real-time global information, the local response behaviors of each edge node are fragmented, making it difficult to form effective coordination, and may even conflict with each other; 3. To avoid system instability, engineering practice requires sacrificing system operating efficiency for basic operational stability, resulting in the system being unable to operate within a more efficient range. Therefore, the technical problem to be solved by this invention is how to construct a control system that can enable all distributed edge nodes in the entire region to quickly converge to a verified operating state based solely on locally available information, without relying on high-cost real-time communication and centralized online optimization calculations. Furthermore, based on coordination, it can dynamically adjust the response intensity according to disturbances to adapt to impacts and compensate for control based on equipment health status. Summary of the Invention

[0005] This invention provides a cloud-edge collaborative control system for industrial parks based on digital twins. Its main purpose is to solve the problems of existing control architectures, which are unable to achieve instantaneous communication between distributed nodes and lack the ability to cope with unknown disturbances and equipment aging due to communication delays and computational lags.

[0006] To achieve the above objectives, the present invention provides a cloud-edge collaborative control system for industrial parks based on digital twins, comprising:

[0007] A cloud engine is configured to analyze historical operating data of an industrial park offline to extract and build a steady-state snapshot library containing multiple steady-state snapshots. Each steady-state snapshot consists of a working condition fingerprint vector representing the macroscopic working condition and an associated device state vector that records the target state of each key controllable device under that working condition.

[0008] Multiple edge control nodes are deployed within the industrial park. Each edge control node pre-stores a steady-state snapshot library and is configured to: collect current environmental parameters in real time to construct a current operating condition fingerprint vector, and calculate the minimum deviation distance between the current operating condition fingerprint vector and all operating condition fingerprint vectors in the steady-state snapshot library. ;

[0009] Furthermore, a spectrum analysis is performed on the load current waveform collected during the control of a controlled electromechanical device to extract health characteristic values; subsequently, based on the minimum deviation distance... Retrieve the steady-state snapshot library to determine the target state adjustment for the baseline, and then compare the target state adjustment for the baseline with a value determined by the minimum deviation distance. acceleration as a function of time The dynamic resilience adjustment coefficient and a compensation coefficient determined by the health characteristic value are calculated to generate the final control command to control the device, so that all edge control nodes can achieve distributed self-consistent collaborative control based on a shared steady-state snapshot library.

[0010] Preferably, the cloud engine is configured to extract steady-state snapshots by performing K-means clustering analysis on historical operating data, where each cluster center corresponds to a working condition fingerprint vector, and the mean of the device state in the dataset associated with the cluster center is determined as the device state vector.

[0011] Preferably, the operating condition fingerprint vector includes at least two of the following physical quantities: outdoor temperature, outdoor humidity, solar irradiance, time stamp, and electricity price stamp.

[0012] Preferably, when the edge control node adjusts the device state according to the final control command, it also includes a smooth transition module, which is configured to gradually bring the device state to the target state along a trajectory defined by a ramp function within a defined time period.

[0013] Preferably, the edge control node is further configured to: when the calculated minimum deviation distance If the value continuously exceeds a security threshold stored in the edge control node, the controlled device will be switched to a conservative operating mode that does not rely on the steady-state snapshot library, and an alarm message will be sent to the cloud engine.

[0014] Preferably, the dynamic toughness adjustment coefficient is generated by calculating the minimum deviation distance in real time. acceleration as a function of time absolute value Furthermore, based on acceleration absolute value Real-time generation of dynamic toughness adjustment coefficient The dynamic toughness adjustment coefficient The calculation method is as follows: ,in, It is a positive real number stored in the edge control node to adjust the system response sensitivity.

[0015] Preferably, the method for edge control nodes to acquire load current waveforms is as follows: reuse the power monitoring module in the industrial park, and activate high-frequency sampling of the load current waveform of the electromechanical equipment only when the edge control node executes control commands to change the operating state of the electromechanical equipment and during the subsequent stabilization period.

[0016] Preferably, the edge control node extracts health feature values ​​by performing a fast Fourier transform on the acquired load current waveform to obtain its spectrum data, extracting one or more energy values ​​or energy proportions within the characteristic harmonic frequency bands predetermined by the cloud engine from the spectrum data, and using the energy value or energy proportion as the health feature value.

[0017] Preferably, the compensation coefficient is generated as follows: the cloud engine is also configured to build a device health baseline library, which stores the baseline health feature values ​​corresponding to the health devices when performing different actions; the edge control node compares the real-time extracted health feature values ​​with the corresponding baseline health feature values ​​in the device health baseline library to determine the compensation coefficient.

[0018] Preferably, the edge control node further includes a decision arbitration module, which stores a rule set containing multiple arbitration rules based on physical common sense or security procedures; and is configured to use the arbitration rule set to perform a reasonableness check on the final control command before sending the final control command to the device, and only allow the command to be executed if the command passes the check, otherwise reject the command and force the system to enter a safe state.

[0019] Compared with the prior art, the beneficial effects of the present invention are:

[0020] 1. The system control method of this invention avoids the instruction latency problem caused by traditional centralized optimization calculation. The core task of its cloud engine is no longer online solving, but offline construction of a historical steady-state snapshot library covering various operating conditions. The edge control nodes distributed in various locations independently query and approach the same historically verified collaborative operating state based on the shared steady-state snapshot library and the same environmental data collected in real time. This makes the collaboration of the entire system no longer dependent on the cloud real-time communication and instruction issuance that are susceptible to latency interference, but is realized in a distributed architecture, thereby avoiding the control oscillation and equipment wear caused by the instruction lagging behind the changes in operating conditions.

[0021] 2. This invention also provides a system resilience adjustment mechanism to cope with unknown disturbances. When the system encounters a severe shock that has never occurred before, the edge control node not only focuses on the deviation distance between the current state and the historical steady state, but also on the acceleration of this deviation distance over time. This acceleration is used as a direct basis for judging the severity of the external disturbance. Based on the magnitude of this acceleration, the control adjustment amount of its output is dynamically converged in real time without nonlinearity. This means that when the system senses a severe shock, it will automatically slow down the magnitude of its adjustment action and absorb the shock energy in a flexible way, rather than resisting it with a fixed response logic. This allows the system to switch from a rigid safety mode to a flexible adaptation with self-protection capabilities when facing unknown risks.

[0022] 3. This invention integrates control and equipment health status perception without increasing additional hardware costs. When the edge control node executes control commands, it reuses the existing power monitoring module on site to collect the load current waveform of the controlled electromechanical equipment and performs spectrum analysis to extract health feature values ​​that can characterize its physical health status. These health feature values ​​are not only used for fault alarms, but also serve as a key compensation coefficient, directly participating in the generation and calculation of the final control commands. This provides feedforward compensation for response deviations caused by equipment wear and aging. Its ultimate goal is to ensure that the physical effect output by the equipment is consistent with the system expectation, rather than just the precise issuance of electrical commands. This extends the control closed loop from the command level to the effect level, solving the gap between the ideal control model and the real physical world caused by equipment degradation. Attached Figure Description

[0023] Figure 1 This is a schematic diagram of the operation flow of the control system of the present invention;

[0024] Figure 2 This is a schematic diagram comparing the effects of the compensation mechanism of the present invention;

[0025] Figure 3 This is a schematic diagram of the architecture deployment of the control system of the present invention. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Of course, the described embodiments are only some embodiments of the present invention, and not all embodiments. It should be understood that the following embodiments are intended to explain the present invention, and not to limit the scope of protection of the present invention.

[0027] This invention provides a cloud-edge collaborative control system for industrial parks based on digital twins. The system mainly consists of a cloud engine and multiple edge control nodes located within the industrial park. The core operation mode is as follows: the cloud engine is configured to analyze historical data offline to build a shared steady-state snapshot library, while the multiple edge control nodes, without relying on real-time communication with the cloud, query this library locally and combine it with real-time operating conditions to achieve a distributed, self-consistent collaborative control. In physical systems like industrial parks, which have multivariable and tightly coupled characteristics, control architectures that rely on cloud-based online optimization calculations at the minute level before issuing commands often suffer from communication and computation delays, causing control commands to lag behind the instantaneous changes in physical operating conditions. To address this challenge, the cloud engine in this invention is configured to offline build a historical steady-state snapshot library. This cloud engine collects historical operating data from the industrial park and executes... For example, in K-means clustering analysis, each cluster center is determined as a condition fingerprint vector representing a specific operating condition. Simultaneously, the mean of the equipment states in the dataset associated with that cluster center is determined as the equipment state vector associated with that condition fingerprint vector. Thus, each steady-state snapshot consists of a condition fingerprint vector and an equipment state vector, and the collection of all these steady-state snapshots constitutes a lightweight steady-state snapshot library. The condition fingerprint vector can specifically include at least two of the following: outdoor temperature, outdoor humidity, solar irradiance, time stamp, and electricity price stamp. The equipment state vector records the target state of each key controllable device under that operating condition. By shifting the core task in the cloud from online solving to offline construction of a historical knowledge base, the basis for control decisions shifts from online computation relying on high-latency communication to local reproduction of verified global collaborative states.

[0028] Correspondingly, in the absence of real-time global information, the local response behaviors of each edge control node within the park are prone to fragmentation. Therefore, each edge control node is configured to pre-store the same steady-state snapshot library built by the cloud engine. During system operation, each edge control node collects current environmental parameters in real time to construct a current operating condition fingerprint vector with the same dimension as the historical operating condition fingerprint vectors in the library. Then, the edge control node uses vector distance calculation, such as Euclidean distance calculation, to retrieve the edge control node from the locally stored steady-state snapshot library that has the minimum deviation distance from the current operating condition fingerprint vector. The historical operating condition fingerprint vector is retrieved, and the device state vector associated with the retrieved historical operating condition fingerprint vector is read, which is used as the target state for controlling the connected device. Since all edge control nodes share the same steady-state snapshot library and query based on the same real-time collected environmental data, they can converge to the same historically verified coordinated global operating state at the same time without communicating with each other, thus forming a distributed self-consistent cooperative control.

[0029] When the system encounters a severe shock unprecedented in history, if the minimum deviation distance between the current operating condition fingerprint and all historical fingerprints in the database... If the deviation consistently exceeds a preset safety threshold, the system switches to a conservative operating mode that does not adjust based on real-time conditions, thus limiting its adjustment capabilities. To enhance the system's adaptability to unknown disturbances, the edge control nodes are also configured to consider this deviation distance when making control decisions. The dynamic characteristics that change over time, specifically, the edge control node through its deviation distance calculated in real time. The time series is subjected to two consecutive difference operations to calculate the deviation acceleration. absolute value Furthermore, based on acceleration absolute value A dynamic toughness adjustment coefficient is generated in real time. Dynamic toughness adjustment coefficient The calculation method is as follows ,in, This is a preset positive real number, the value of which can be determined through offline simulation after weighing the system response speed and overshoot suppression effect; this coefficient It is used to dynamically modulate the adjustment amount of the retrieved target state. When the system is stable... Approaching zero, coefficient It approaches 1, but when subjected to severe shocks, Increase, coefficient The corresponding decrease occurs, and the system automatically adjusts the amplitude of its adjustment action, thus achieving an adjustment mechanism that adapts the response intensity to the severity of the disturbance.

[0030] Traditional control models do not consider response deviations caused by physical wear and tear on equipment, potentially leading to a disconnect between control commands and physical effects. To compensate for this deviation, edge control nodes are configured to reuse existing power monitoring modules within the industrial park when executing control commands. High-frequency sampling of the load current waveform is activated only when the edge control node executes control commands to change the operating state of the electromechanical equipment and during a defined stabilization period thereafter. After acquiring the load current waveform, the edge control node performs a Fast Fourier Transform to obtain its spectral data and extracts energy values ​​or energy percentages within one or more pre-defined characteristic harmonic frequency bands as health characteristic values. Correspondingly, the cloud engine is configured to build an equipment health baseline library, storing baseline health characteristic values ​​corresponding to different actions performed by healthy equipment. The edge control node compares the real-time extracted health characteristic values ​​with the corresponding baseline health characteristic values ​​in the equipment health baseline library to determine a compensation coefficient. This compensation coefficient directly... The edge control node participates in the generation and calculation of the final control command, and performs feedforward compensation for response deviations caused by equipment performance degradation, so that the physical effect output by the equipment approaches the system's target. To further improve the system's operational stability and decision-making security, the edge control node also includes a smooth transition module when adjusting the equipment state according to the final control command. This module is configured to gradually bring the equipment state to the target state along a trajectory defined by, for example, a ramp function within a defined time period. At the same time, the edge control node also includes a decision arbitration module. This module stores a rule set containing multiple arbitration rules based on physical common sense or safety procedures, and is configured to use the arbitration rule set to perform a rationality check on the final control command before sending it to the equipment. For example, it checks whether cooling and heating commands conflict. The command is only allowed to be executed if it passes the check. If it fails the check, the command is rejected and the system is forced to enter a safe state. This serves as a protective mechanism to prevent unreasonable decisions caused by upstream information contamination.

[0031] Example 1: In a specific industrial park application, the photovoltaic power generation drops instantaneously in the afternoon due to rapid cloud movement, while a critical production line stops operating due to unplanned maintenance, resulting in a sharp decrease in the overall electricity load of the park. The superposition of these two events poses a direct challenge to the stability of the park's power grid and the economic operation of the energy system. When this situation occurs, all edge control nodes deployed in the park, including nodes controlling photovoltaic inverters, central air conditioning systems, and energy storage units, collect parameters characterizing changes in the external environment in real time and independently generate a current operating condition fingerprint vector corresponding to this situation locally. By calculating the deviation distance between this current operating condition fingerprint vector and all historical operating condition fingerprint vectors in the locally stored steady-state snapshot library, these edge control nodes retrieve the same historical steady-state snapshot associated with low solar irradiance and low production load. This mechanism provides all distributed nodes with a unified global target operating state that is verified by historical operating data as coordinated, and transforms the control problem from how to calculate an unknown optimal cooperative strategy online to how to approach a known optimal cooperative history.

[0032] Furthermore, since the changes in operating conditions are sudden, the minimum deviation distance calculated by each edge control node... Its time series shows an accelerating trend of increase, which makes the distance... The acceleration determined by the second derivative absolute value Instantaneous increase, based on dynamic toughness adjustment coefficient The generation method, i.e. A larger one The value will result in the coefficient The value instantly converges to a value less than 1. At this point, although each edge control node has retrieved the target device state vector, the baseline target state adjustment amount generated based on it is multiplied by the dynamic resilience adjustment coefficient before being sent to the device. As a result, the actual execution magnitude of a series of adjustment actions, such as reducing the power of the photovoltaic inverter and reducing the power of the air conditioning system, is instantaneously converged, and the system responds to the impact in a more moderate way, thereby achieving a balance between the response speed to the disturbance and maintaining the stability of the system.

[0033] Meanwhile, after the control command is issued, an edge control node controlling the central air conditioning compressor collects the compressor's load current waveform through a multiplexed power monitoring module and extracts its health characteristic value via Fast Fourier Transform. This node compares this health characteristic value with the baseline value in the equipment health baseline library and finds that the compressor has experienced slight performance degradation due to long-term operation. Based on this, a compensation coefficient is generated. This compensation coefficient is related to the dynamic resilience adjustment coefficient. The target state adjustment amount, acting together on the baseline, generates the final control command. This mechanism ensures that the adjustment of equipment state not only considers drastic changes in external operating conditions but also feeds forward to take into account the physical health of the equipment itself. The synergistic effect of the two mechanisms adjusts both the overall response behavior of the system to external shocks and the execution effect of commands specific to each equipment. Ultimately, the overall operating state of the park smoothly transitions to a new equilibrium point that matches the historically efficient operating state. This process avoids system oscillations and equipment losses that may be caused by over-adjustment. The coordinated control and adaptive adjustment of the entire system are achieved through shared historical data and unified dynamic response rules, without any communication interaction among all edge nodes.

[0034] Example 2: To quantitatively verify the system response characteristics and stability of the technical solution of the present invention in response to sudden disturbances, a hardware-in-the-loop simulation test platform was built. This platform reproduces an industrial park power grid model including photovoltaic power generation units, variable industrial loads, and energy storage units using a real-time simulator. The physical edge control node hardware running the control logic of the present invention is connected to this closed-loop system. Simultaneously, a control group using a traditional hierarchical optimization control method is set up, with a 60-second communication and computation delay between the cloud and edge nodes in the control group. During the experiment, the dynamic resilience adjustment coefficient within the edge control node of the technical solution of the present invention is... Key parameters The value is set to balance the speed and stability of the system response. The specific value is determined through an offline calibration procedure. This procedure applies a series of standardized step perturbations to the simulation model to minimize key system state variables, such as the fluctuation integral of the power exchange between the industrial park and the main power grid, as the optimization objective, and iteratively seeks the best value. In this experiment, the constructed industrial park model is used. The value of is determined to be 0.5.

[0035] After the experiment started, at the 100th second, a disturbance simulating a 50% step decrease in solar irradiance due to cloud cover and a 40% step decrease in load simulating a critical production line shutdown were simultaneously applied to the simulation system. In the control group, the power exchange between the park and the main power grid experienced a reverse power surge after the disturbance occurred because the local response lagged behind the load change. After the cloud optimization command was issued, a large overshoot occurred, and the system power fluctuations continued for a long time before converging. In the experimental group running the technical solution of this invention, when the disturbance occurred, all edge control nodes obtained the coordinated target state almost simultaneously based on the local query steady-state snapshot library, and calculated the dynamic resilience adjustment coefficient according to the severity of the disturbance, adjusting the control output. The fluctuation of the system power exchange was suppressed to a small range and quickly converged to a new steady state.

[0036] Table 1: Comparison of key performance indicators between the experimental group and the control group after the disturbance occurred.

[0037]

[0038] Referring to Table 1, the peak power overshoot and power fluctuation convergence time of the experimental group were significantly lower than those of the control group. This is because the control commands in the control group had a minute-level delay, causing the commands to act on a system state that had already changed, thus exacerbating system oscillations. The experimental group avoided this delay by locally querying the steady-state snapshot library, achieving near real-time response. Simultaneously, the deviation acceleration in the experimental group... The peak value is much larger than that of the control group, reflecting the system's rapid detection of disturbance changes, and thus generating a dynamic resilience coefficient less than 1. The intensity of the control actions was actively reduced, which is also confirmed by the fact that the total variance of the key equipment adjustment actions in the experimental group was much smaller than that in the control group.

[0039] Example 3: This example combines Figures 1 to 3 A description of the cloud-edge collaborative control system for industrial parks based on digital twins, such as... Figure 1 As shown, the process starts with historical operating data and performs offline analysis and modeling through a cloud engine. Specifically, the cloud engine uses historical data such as park temperature, load, and equipment status to perform offline analysis using algorithms such as K-means clustering, thereby constructing a steady-state snapshot library and an equipment health baseline library. The steady-state snapshot library contains a series of snapshots consisting of operating condition fingerprints and associated equipment status vectors, while the equipment health baseline library stores the feature values ​​of healthy equipment under different operating conditions. During the real-time operation phase of the system, edge control nodes collect real-time environmental parameters, such as temperature, humidity, and irradiance, to generate a current operating condition fingerprint vector with the same dimension as the historical library. Based on this, historical steady-state snapshots are retrieved, and the minimum deviation distance from the best matching snapshot in the library is calculated. Meanwhile, the system also continuously calculates this deviation distance. acceleration as a function of time To generate a dynamic resilience adjustment coefficient for shock response, the system also collects load current waveforms and performs spectrum analysis to extract current health characteristic values ​​when controlling electromechanical equipment such as air conditioners, inverters, and energy storage units. These characteristic values ​​are then compared with the equipment health baseline library to generate a compensation coefficient. Finally, the final control command, which integrates dynamic adjustment and health compensation, is generated and sent to the electromechanical equipment after being verified for rationality and safety by the decision arbitration module based on the rule set and progressively executed by the smooth transition module using ramp functions.

[0040] like Figure 2As shown in the figure, the horizontal axis represents the degree of equipment performance degradation, increasing from 0% to 30%, while the vertical axis represents the equipment's output efficiency. The figure contains three curves, representing the actual output without compensation and the actual output with compensation. The target output curve is a horizontal line at 100% efficiency, representing the system's expected performance. The actual output curve without compensation is a dashed line that drops sharply as equipment performance degrades, indicating that without compensation, the actual output efficiency of the equipment will be far lower than the target. In contrast, the actual output curve with compensation is a solid line that remains close to the target output throughout the entire equipment performance degradation range. This intuitively demonstrates that the compensation coefficient mechanism of this invention can effectively offset the performance degradation caused by physical aging and wear of the equipment, ensuring that the final physical execution effect is consistent with the expected target of the control system.

[0041] like Figure 3 As shown in the diagram, the architecture clearly illustrates the two core components of the system: CloudSide and EdgeSide. In the cloud, a cloud server cluster provides computing power, on which a cloud engine, serving as the core logical unit, is deployed, along with a historical database for storing historical data and a device health baseline library for storing device health records. The core task of the cloud is offline analysis and modeling. At the edge, multiple edge control gateways are deployed, such as Edge Control Gateway #1 and Edge Control Gateway #2 shown in the diagram. Each gateway deploys local edge control logic and stores a copy of the steady-state snapshot library distributed from the cloud. The edge control gateways connect directly to and control on-site physical devices, such as central air conditioning units and photovoltaic inverters, via real-time control buses such as Modbus or OPCUA. This distributed architecture enables each edge node to make independent and rapid local decisions, achieving decoupling from real-time communication with the cloud.

[0042] Example 4: When the control system of this invention is first deployed in an industrial park or when a key electromechanical device is added to the park, in order to ensure that the system's operating parameters have deterministic basis based on field data, an initialization and parameter calibration procedure needs to be performed. For the safety threshold used in the edge control node to determine unknown operating conditions, the calibration process is as follows: First, using the cloud engine, historical operating data from at least one complete operating year of the park is processed to extract a set of operating condition fingerprint vectors covering all time periods. Second, for each operating condition fingerprint vector in this set, its minimum deviation distance from all other vectors in the set is calculated. Finally, for all the calculated... The values ​​are statistically distributed, and the value corresponding to the 99.9th percentile of the distribution is set as the safety threshold stored in the edge control node for triggering the conservative operation mode.

[0043] For the functional modules related to equipment health status perception in the edge control node, the initialization and calibration process is illustrated by a newly added variable frequency water pump. When the new equipment is installed and in its initial health state, the system enters an equipment baseline construction mode. In this mode, the edge control node controls the water pump to operate stably at a set of preset speed points covering its commonly used operating range. At each speed point, the edge control node collects the load current waveform of the water pump at high frequency through a reused power monitoring module and performs rapid transformation on the collected waveform data. By comparing the current spectrum obtained at different speeds, the system identifies those harmonic frequency bands that have concentrated and stable energy at specific speeds but change regularly with the speed. These frequency bands are then determined as characteristic harmonic frequency bands used to characterize the health status of the water pump. Subsequently, the system uses the average energy integral value within these characteristic harmonic frequency bands when the water pump is running stably at each reference speed point as the baseline health feature value, which is stored together with the corresponding speed information in the equipment health baseline library of the edge control node.

[0044] Furthermore, the compensation coefficients used to compensate for control commands are also deterministically defined in their generation method. During the daily operation of the water pump, the edge control node extracts its health characteristic values ​​in real time. It also retrieves the baseline health characteristic value corresponding to the current operating status from the equipment health baseline database. In this specific configuration, health trait values ​​are positively correlated with equipment operating efficiency, and the compensation coefficient... Defined as This coefficient is then used to adjust the control commands issued to the water pump. If the real-time health characteristic value decreases, the compensation coefficient will be greater than 1, thereby increasing the output strength of the command to compensate for the deviation in physical effect caused by the degradation of equipment performance.

[0045] Example 5: When the system of the present invention is deployed in a newly built industrial park lacking long-term historical operating data, in order to achieve immediate system activation and subsequent self-optimization, the cloud engine first issues a general steady-state snapshot library to all edge control nodes in the park. This general library is constructed by integrating historical data from multiple industrial parks that are already operating stably and have similar industrial characteristics through cluster analysis. This provides the new park with an initial operating basis, enabling it to have basic collaborative control capabilities. On this basis, the cloud engine is configured to continuously collect the actual operating data of the new park itself, and when the accumulated data reaches a preset threshold, it automatically triggers a cluster analysis of the park's own data to generate a steady-state snapshot library based on the park's own data, and uses it to replace the original general library in the edge control nodes.

[0046] While implementing the above deployment, to ensure that control behavior complies with the safety procedures of the specific site, the decision arbitration module in the edge control node needs to be configured. Each edge control node is pre-installed with a template library containing multiple basic physical and electrical safety arbitration rules. During the on-site commissioning phase, engineering technicians can select and activate applicable rules from the template library through the configuration interface, based on the specific model of the equipment installed in the newly built park, the constraints of the process flow, and the safety management regulations. They can also add additional arbitration rules for special operating conditions of the site. This ability to add rules allows specific operating boundary conditions for specific equipment, such as large-capacity battery energy storage systems, such as the rule that charging is prohibited when the grid voltage is abnormal, to be included in the verification scope of the decision arbitration module. After this configuration, the decision arbitration module has the ability to verify deterministic instructions that comply with the safety boundaries of the specific site.

[0047] Example 6: In the process of constructing a steady-state snapshot library using K-means clustering analysis, the cloud engine first performs data preprocessing and model hyperparameter optimization steps. In the data preprocessing stage, the cloud engine verifies the integrity and consistency of the collected historical operating data and removes invalid data segments where sensor readings exceed their physical range or have remained unchanged for a long time. In the hyperparameter optimization stage, to determine the number of clusters... Value, the cloud engine is for a preset The K-means clustering algorithm is executed multiple times within a given range, and the silhouette coefficient is calculated for each clustering result. Finally, the value corresponding to the maximum global silhouette coefficient is selected. The value represents the number of clusters used to build the final steady-state snapshot library.

[0048] Similarly, before the edge control node collects current environmental parameters in real time to construct the current operating condition fingerprint vector, it also performs the aforementioned validity check on the sensor data. When an anomaly is detected in the data of a key sensor, the node will temporarily use a moving average based on the valid data of that sensor over a short period of time as a substitute to maintain the continuity of the control function; in addition, by comparing health characteristic values... Compared with baseline health characteristics To generate compensation coefficients Furthermore, its generation method is configured such that, in the equipment baseline construction mode, simulated efficiency decay is injected at multiple operating points to calibrate the nonlinear relationship between the health characteristic value and the required compensation force that describes the entire cycle from health to decline of the equipment, and the model parameters of this relationship are stored in the equipment health baseline library for real-time calculation. Called at any time.

[0049] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0050] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A cloud-edge collaborative control system for industrial parks based on digital twins, characterized in that, include: A cloud engine is configured to analyze historical operating data of an industrial park offline to extract and build a steady-state snapshot library containing multiple steady-state snapshots. Each steady-state snapshot consists of a working condition fingerprint vector representing the macroscopic working condition and an associated device state vector that records the target state of each key controllable device under that working condition. Multiple edge control nodes are set up in the industrial park. Each edge control node has a pre-stored steady-state snapshot library and is configured to: collect current environmental parameters in real time to form the current operating condition fingerprint vector, and calculate the minimum deviation distance between the current operating condition fingerprint vector and all operating condition fingerprint vectors in the steady-state snapshot library. Furthermore, a spectrum analysis is performed on the load current waveform collected during the control of a controlled electromechanical device to extract health characteristic values. Then, based on the minimum deviation distance, a steady-state snapshot library is retrieved to determine the target state adjustment amount of the benchmark. The target state adjustment amount of the benchmark is multiplied by a dynamic resilience adjustment coefficient determined by the acceleration of the minimum deviation distance changing with time and a compensation coefficient determined by the health characteristic value to generate the final control command to control the device. Thus, all edge control nodes achieve distributed self-consistent collaborative control based on the shared steady-state snapshot library. The cloud engine is configured to extract steady-state snapshots by performing K-means clustering analysis on historical operating data, where each cluster center corresponds to a working condition fingerprint vector, and the mean of the device state in the dataset associated with the cluster center is determined as the device state vector; The operating condition fingerprint vector includes at least two of the following physical quantities: outdoor temperature, outdoor humidity, solar irradiance, time stamp, and electricity price stamp; The dynamic toughness adjustment coefficient is generated by calculating the minimum deviation distance in real time. acceleration as a function of time The absolute value of; and, based on the absolute value of acceleration a. Real-time generation of dynamic toughness adjustment coefficient The dynamic toughness adjustment coefficient The calculation method is as follows: ,in, It is a positive real number stored in the edge control node to adjust the system response sensitivity; The edge control node extracts health feature values ​​by performing a fast Fourier transform on the acquired load current waveform to obtain its spectrum data, extracting one or more energy values ​​or energy proportions within the characteristic harmonic frequency bands predetermined by the cloud engine from the spectrum data, and using the energy value or energy proportion as the health feature value. The compensation coefficient is generated as follows: the cloud engine is also configured to build a device health baseline library, which stores baseline health feature values ​​corresponding to healthy devices performing different actions; the edge control node compares the real-time extracted health feature values ​​with the corresponding baseline health feature values ​​in the device health baseline library; the compensation coefficient... If the real-time health metric value decreases, the compensation coefficient will be greater than 1, thereby increasing the output strength of the instruction; among which, For health characteristics, These are baseline health characteristics.

2. The cloud-edge collaborative control system for industrial parks based on digital twins as described in claim 1, characterized in that, When adjusting the device state according to the final control command, the edge control node also includes a smooth transition module, which is configured to gradually bring the device state to the target state along a trajectory defined by a ramp function within a defined time period.

3. The cloud-edge collaborative control system for industrial parks based on digital twins as described in claim 1, characterized in that, The edge control node is also configured to switch the controlled device to a conservative operating mode that does not rely on the steady-state snapshot library and send an alarm message to the cloud engine when the calculated minimum deviation distance d is continuously greater than a safety threshold stored in the edge control node.

4. The cloud-edge collaborative control system for industrial parks based on digital twins as described in claim 1, characterized in that, The method for edge control nodes to acquire load current waveforms is as follows: reuse the power monitoring module in the industrial park, and activate high-frequency sampling of the load current waveform of the electromechanical equipment only when the edge control node executes control commands to change the operating status of the electromechanical equipment and during the subsequent stabilization period.

5. The cloud-edge collaborative control system for industrial parks based on digital twins as described in claim 1, characterized in that, The edge control node also includes a decision arbitration module, which stores a rule set containing multiple arbitration rules based on physical common sense or security procedures. It is configured to use the arbitration rule set to validate the rationality of the final control command before sending it to the device, and only allow execution if the command passes the validation. If the validation fails, the command is rejected and the system is forced into a secure state.

Images