Dynamic simulation method and system of industrial gateway based on digital twinning

Abstract

The application discloses a kind of based on digital twinning industrial gateway's dynamic simulation method and system, the present application relates to the technical field of digital twinning, the joint state of industrial gateway is input into the preset digital twinning space, and dynamic twin information is determined, so as to extract corresponding twin state feature, twin state feature is mapped to the preset gateway operation phase space, and corresponding dynamic simulation excitation sequence is generated in combination with time sequence disturbance factor, improve the accuracy of dynamic simulation excitation sequence.Dynamic simulation excitation sequence is input into dynamic twin information, and ahead time sequence deduction is carried out, so as to determine the running track of industrial gateway in future time period, and determine the resource allocation deviation degree of the deviation of gateway virtual and real;If resource allocation deviation degree exceeds preset dynamic threshold, then generate reverse tuning control instruction based on the strategy optimization of running track, along the execution of reverse tuning control instruction, realize the reverse modification of dynamic twin information.

Description

Technical Field

[0001] This invention relates to the technical field of digital twins, and more particularly to a dynamic simulation method and system for industrial gateways based on digital twins. Background Technology

[0002] As the core hub at the edge of the Industrial Internet, industrial gateways undertake critical tasks such as heterogeneous protocol conversion, edge computing, and data aggregation. Due to the harsh industrial environment and variable business loads, industrial gateways often face complex operating conditions such as computing power congestion, communication bottlenecks, and hidden hardware degradation. To ensure the high reliability of gateway operation, digital twin technology has been gradually introduced into the monitoring and simulation of industrial gateways, realizing the perception and mapping of the physical entity's state by constructing a virtual mapping model.

[0003] Current simulation methods mostly rely directly on multi-source raw operational data collected by physical sensors as the simulation driving source. On the one hand, the raw data is mixed with a large amount of environmental noise and occasional disturbances, and fails to combine the gateway's mechanistic constraint information to achieve state coupling driven by both data and mechanism. On the other hand, existing methods lack in-depth manifold analysis and phase space mapping of dynamic twin information, and only remain at the driving of surface data. They fail to remove redundant noise and extract twin state features that characterize the essential operating mode of the gateway, resulting in low accuracy of dynamic simulation excitation sequences. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of the prior art. This invention provides a dynamic simulation method and system for industrial gateways based on digital twins.

[0005] This invention provides a dynamic simulation method for an industrial gateway based on digital twins, comprising: The system acquires multi-source operational data from industrial gateways and determines the joint state of industrial gateways by combining the mechanism constraint information of industrial gateways. The joint state of industrial gateways is then input into a preset digital twin space, and dynamic twin information that is spatiotemporally synchronized with industrial gateways and continuously evolves is determined. Based on the analysis of dynamic twin information, the corresponding twin state features are extracted, the twin state features are mapped to the preset gateway operation phase space, and the corresponding dynamic simulation excitation sequence is generated by combining the timing disturbance factor. The dynamic simulation excitation sequence is input into the dynamic twin information and advanced time series extrapolation is performed to determine the operating trajectory of the industrial gateway in the future time period. Based on the identification of the operating trajectory, the resource allocation deviation degree characterizing the virtual-real deviation of the gateway is determined. If the resource allocation deviation exceeds the preset dynamic threshold, a reverse optimization control command is generated based on the strategy optimization of the running trajectory. By executing the reverse optimization control command, the response residual of the industrial gateway is determined, and then the dynamic twin information is adaptively reverse corrected.

[0006] This invention provides a dynamic simulation system for an industrial gateway based on digital twins. This system is applied to the aforementioned dynamic simulation method for an industrial gateway based on digital twins. The dynamic simulation system for the industrial gateway based on digital twins includes: The dynamic twin module is used to acquire multi-source operational data of the industrial gateway, and determine the joint state of the industrial gateway by combining the mechanism constraint information of the industrial gateway. The joint state of the industrial gateway is input into the preset digital twin space, and dynamic twin information that is synchronized with the industrial gateway in time and space and continuously evolves is determined. The dynamic simulation module is used to extract the corresponding twin state features based on the analysis of dynamic twin information, map the twin state features to the preset gateway running phase space, and generate the corresponding dynamic simulation excitation sequence in combination with the timing perturbation factor. The operation trajectory module is used to input the dynamic simulation excitation sequence into the dynamic twin information and perform advanced time series extrapolation to determine the operation trajectory of the industrial gateway in the future time period, and to determine the resource allocation deviation degree characterizing the virtual-real deviation of the gateway based on the identification of the operation trajectory. The correction module is used to generate reverse optimization control instructions based on the strategy optimization of the running trajectory if the resource allocation deviation exceeds the preset dynamic threshold. By executing the reverse optimization control instructions, the response residual of the industrial gateway is determined, and then the dynamic twin information is adaptively reverse corrected.

[0007] Compared with the prior art, the beneficial effects of the present invention are: (1) Obtain multi-source operation data of industrial gateways and determine the joint state of industrial gateways in combination with the mechanism constraint information of industrial gateways. Input the joint state of industrial gateways into the preset digital twin space and determine the dynamic twin information that is spatiotemporally synchronized with industrial gateways and continuously evolves. Extract the corresponding twin state features based on the analysis of dynamic twin information, map the twin state features to the preset gateway operation phase space, and generate the corresponding dynamic simulation excitation sequence in combination with the time-series disturbance factor. The introduction of dynamic twin information further controls the twin state features and improves the accuracy of dynamic simulation excitation sequence.

[0008] (2) The dynamic simulation excitation sequence is input into the dynamic twin information and advanced timing simulation is performed to determine the operating trajectory of the industrial gateway in the future time period. Based on the identification of the operating trajectory, the resource allocation deviation degree representing the virtual-real deviation of the gateway is determined. If the resource allocation deviation degree exceeds the preset dynamic threshold, the reverse optimization control instruction is generated based on the strategy optimization of the operating trajectory. The response residual of the industrial gateway is determined by following the execution of the reverse optimization control instruction. Then, the dynamic twin information is adaptively reverse corrected to further control the resource allocation deviation degree. The reverse optimization control instruction is fully considered, and the reverse correction of the dynamic twin information is realized. Attached Figure Description

[0009] Figure 1 This is a flowchart illustrating the dynamic simulation method for an industrial gateway based on digital twins in an embodiment of the present invention. Figure 2 This is a flowchart illustrating step S11 in the dynamic simulation method for an industrial gateway based on digital twins in this embodiment of the invention. Figure 3 This is a flowchart illustrating step S12 in the dynamic simulation method for an industrial gateway based on digital twins in this embodiment of the invention. Figure 4 This is a flowchart illustrating step S13 in the dynamic simulation method for an industrial gateway based on digital twins in an embodiment of the present invention. Figure 5 This is a flowchart illustrating step S14 in the dynamic simulation method for an industrial gateway based on digital twins in this embodiment of the invention. Figure 6 This is a schematic diagram of the structural composition of a dynamic simulation system for an industrial gateway based on digital twins, as described in an embodiment of the present invention. Detailed Implementation

[0010] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0011] Please see Figures 1 to 6 A dynamic simulation method for industrial gateways based on digital twins is proposed and applied to digital twin scenarios. The dynamic simulation method for industrial gateways based on digital twins includes: Step S11: Obtain multi-source operation data of the industrial gateway, and determine the joint state of the industrial gateway in combination with the mechanism constraint information of the industrial gateway. Input the joint state of the industrial gateway into the preset digital twin space, and determine the dynamic twin information that is synchronized with the industrial gateway in time and space and continuously evolves. Step S12: Extract the corresponding twin state features based on the analysis of dynamic twin information, map the twin state features to the preset gateway running phase space, and generate the corresponding dynamic simulation excitation sequence in combination with the timing perturbation factor; Step S13: Input the dynamic simulation excitation sequence into the dynamic twin information and perform advanced time series extrapolation to determine the operating trajectory of the industrial gateway in the future time period, and determine the resource allocation deviation degree characterizing the virtual-real deviation of the gateway based on the identification of the operating trajectory. Step S14: If the resource allocation deviation exceeds the preset dynamic threshold, a reverse optimization control command is generated based on the strategy optimization of the running trajectory. By following the execution of the reverse optimization control command, the response residual of the industrial gateway is determined, and then the dynamic twin information is adaptively reverse corrected.

[0012] refer to Figure 2 In step S11, the specific steps are as follows: S111: Obtain multi-source operation data of industrial gateway under heterogeneous communication protocols, extract protocol feature vectors from multi-source operation data, construct state coupling content driven by data and mechanism in combination with the mechanism constraint information of industrial gateway, and determine the joint state of industrial gateway containing implicit degradation state and explicit load state based on the multi-level iteration of this state coupling content. S112: Input the joint state into the preset digital twin space, use the joint state as the initial boundary condition, match the spatiotemporal evolution information corresponding to the industrial gateway in the digital twin space, and continuously evolve it in combination with the spatiotemporal change items that fluctuate with the on-site physical environment, thereby determining the dynamic twin information that is spatiotemporally synchronized with the industrial gateway and continuously evolves. This dynamic twin information includes multidimensional physical field distribution content and logical topology mapping content.

[0013] In the embodiments of this application, multi-source operation data of industrial gateway under heterogeneous communication protocols is obtained, and protocol feature vectors are extracted from the multi-source operation data. Combined with the mechanism constraint information of industrial gateway, a state coupling content driven by both data and mechanism is constructed. Based on the multi-level iteration of the state coupling content, a joint state of implicit degradation state and explicit load state in industrial gateway is determined, thus introducing a joint state of implicit degradation state and explicit load state.

[0014] At this time, multi-source operational data of the industrial gateway under heterogeneous communication protocols is acquired. The multi-source operational data covers physical layer signal quality, link layer transmission indicators, and application layer load messages. Based on protocol parsing rules and sliding window statistical models, the multi-source operational data is mined for time-series and frequency-domain features to extract protocol feature vectors that characterize the operational state of each protocol channel. The protocol feature vectors include, but are not limited to, protocol message arrival rate jitter, check retransmission rate, periodic command deviation, and link-level throughput overload rate.

[0015] The mechanism constraint information of the industrial gateway is retrieved, which includes the computing power conservation constraint, bus bandwidth allocation upper limit constraint, and heat dissipation power boundary constraint of the gateway processor. The protocol feature vector is mapped to the mechanism constraint space to construct a state coupling content driven by both data and mechanism. At this time, the computing power conservation constraint and the bus bandwidth allocation upper limit constraint are used as physical boundary penalty terms for state coupling. The message processing requirements and link-level throughput requirements in the protocol feature vector are used as driving sources. By constructing a directed acyclic graph containing resource allocation nodes and protocol processing nodes, and solving the dynamic occupancy weight of each node under the constraint of the physical boundary penalty terms, a state coupling content representing the mutual constraint and coupling relationship between data flow and physical resources is formed.

[0016] Using the aforementioned state coupling content as the initial value for iteration, a hidden Markov-Kalman joint filter iterator containing a state transition matrix and an observation matrix is ​​constructed. Multiple iterations are performed to decouple and determine the joint state of the industrial gateway, which includes both implicit degradation states and explicit load states. During these iterations, the explicit load states are obtained directly from the protocol feature vectors, such as CPU utilization and port queue length, through direct correction of the observation matrix. The implicit degradation states, using the explicit load states as input, are combined with the physical performance degradation rate model defined in the mechanistic constraint information, such as the cumulative effect of silicon wafer thermal migration and the effect of increased equivalent series resistance of electrolytic capacitors. Irreversible time accumulation is extrapolated in the state transition matrix, and the implicit degradation states are optimally estimated using Kalman gain and residual feedback. Finally, the joint state of the industrial gateway, composed of the concatenation of the explicit load state vector and the implicit degradation state vector, is output.

[0017] For implicit degradation states, based on the explicit load states collected by the industrial gateway during operation, including observables such as CPU utilization, memory usage, and port queue length, and combined with the pre-set physical degradation models in the mechanistic constraint information, such as the silicon wafer thermal migration accumulation model or the electrolytic capacitor equivalent series resistance growth model, a state estimation framework driven by time series and with degradation parameters as implicit variables is constructed.

[0018] The framework is based on the extended Kalman filter algorithm: at each time step, the system first uses the implicit degradation state estimated in the previous time step and the current explicit load state to predict the current degradation state through the state transition matrix; using the residual between the actual observed explicit load state and the predicted value, the prediction result is weighted and corrected through Kalman gain to obtain the optimal estimate of the implicit degradation state at the current time step; by iteratively executing the above prediction and correction steps, the implicit degradation state vector, including the degree of heat accumulation and power supply aging factor, can be continuously output without relying on additional sensors.

[0019] Specifically, in the dynamic simulation scenario of the industrial gateway, the industrial gateway simultaneously connects 64 Modbus RTU temperature transmitters and 8 OPCUA vision inspection devices, and aggregates and reports to the edge cloud via the MQTT protocol; it acquires multi-source operating data of the industrial gateway and extracts protocol feature vectors: for Modbus RTU slaves, it extracts the CRC check retransmission rate Vcrc and polling cycle jitter Vjitter caused by electromagnetic interference; for OPCUA nodes, it extracts the subscription callback delay Vdelay and session packet loss rate Vloss caused by image data bursts; for uplink MQTT, it extracts the publish queue backlog depth Vqueue, thus forming a multi-dimensional protocol feature vector Vproto=[Vcrc,Vjitter,Vdelay,Vloss,Vqueue].

[0020] Retrieve the mechanistic constraints of the industrial gateway: The computing power limit of the ARM Cortex-A72 processor of the industrial gateway is 20,000 DMIPS, and the gigabit Ethernet bus limit is 1Gbps; when constructing the state coupling content, the polling retransmission requirements of Modbus and the burst throughput requirements of OPCUA are used as the driving source input to the directed acyclic graph; when the burst of image data in OPCUA causes a surge in Vqueue, due to the upper limit constraint of gigabit Ethernet bandwidth allocation, the available bandwidth of the Modbus channel is squeezed, which leads to further deterioration of Vjitter. At the same time, the CPU computing power conservation constraint forces the interrupt handling weight to tilt towards OPCUA. This mutual constraint and dynamic allocation process of data flow and physical computing power and bandwidth resources constitutes the current state coupling content of the industrial gateway.

[0021] Based on the above state coupling content, multi-level iterations are performed: the dominant load state is determined, that is, the current CPU interrupt utilization rate reaches 85% and the Ethernet port buffer queue length reaches 800KB; then the implicit degradation state is determined, the industrial gateway has been running at a high load of 85% for a long time, causing the main chip junction temperature to remain at a high level of 75℃. According to the silicon wafer thermal migration accumulation effect and the capacitor high temperature ESR increase effect defined in the mechanism constraints, the implicit degradation state of the motherboard power supply module is deduced in the state transition matrix. For example, the capacitor ESR has degraded from the initial 10mΩ to 25mΩ, resulting in a 15% increase in the ripple coefficient. This degradation state cannot be directly measured by sensors, but it is captured and optimally estimated by the Kalman filter through the occasional bit flip rate residual feedback under dominant load. [Interrupt utilization 85%, buffer queue 800KB] and [ESR degradation 25mΩ, thermal migration accumulation 0.15] are concatenated to form a joint state that accurately represents the current actual operation and health degradation degree of the industrial gateway.

[0022] Furthermore, the joint state is input into a preset digital twin space, using the joint state as the initial boundary condition. The spatiotemporal evolution information corresponding to the industrial gateway is matched in the digital twin space, and continuously evolved in combination with the spatiotemporal change items that fluctuate with the on-site physical environment. This determines the dynamic twin information that is spatiotemporally synchronized with the industrial gateway and continuously evolves. This dynamic twin information includes multidimensional physical field distribution content and logical topology mapping content, and is compatible with the overall consideration of spatiotemporal change items that fluctuate with the physical environment, ensuring the accuracy of the dynamic twin information that is spatiotemporally synchronized with the industrial gateway and continuously evolves.

[0023] At this point, the aforementioned determined joint state, including the implicit degradation state and the explicit load state, is input into the preset digital twin space, and the joint state is used as the initial boundary condition to drive the digital twin space to start the deduction. At this point, the explicit load state is mapped to the initial occupancy of logical computing resources and communication bandwidth, and the implicit degradation state is mapped to the initial degradation coefficient of physical material parameters and the initial thermal resistance offset. In the multidimensional evolution model library preset in the digital twin space, the corresponding spatiotemporal evolution information is matched according to the numerical characteristics of the initial boundary conditions. The spatiotemporal evolution information includes a physical heat transfer evolution kernel based on finite element method and a logical task scheduling evolution kernel based on discrete event system specification, thereby establishing the evolution starting point for the alignment of the virtual entity and the physical entity with the spatiotemporal reference at the current moment.

[0024] The system retrieves spatiotemporal variation items that fluctuate with the physical environment on site. These items include the diurnal periodic slow-change component of ambient temperature, the electromagnetic pulse transient component caused by the start-up and shutdown of the workshop's high-power equipment, and the alternating fluctuation component of the cabinet's microenvironment humidity. These spatiotemporal variation items are coupled as external disturbance sources and input into the spatiotemporal evolution information of the digital twin space for continuous evolution. During this continuous evolution, the ambient temperature slow-change component and the thermal resistance offset in the latent degradation state are superimposed to correct the boundary convection coefficient of the physical heat transfer evolution kernel. The electromagnetic pulse transient component is superimposed on the interruption processing link of the logical task scheduling evolution kernel to simulate transient interference, thereby driving the digital twin to break away from the static snapshot on the time axis and form a dynamic temporal progression that fluctuates with the physical environment.

[0025] During the continuous evolution of the spatiotemporal evolution information, constrained by initial boundary conditions and driven by spatiotemporal change items, dynamic twin information that is spatiotemporally synchronized with and continuously evolving with the industrial gateway is synchronously extracted and determined. The dynamic twin information is composed of multidimensional physical field distribution content and logical topology mapping content. Among them, the multidimensional physical field distribution content is the output of the physical heat transfer evolution kernel, which represents the thermal stress field distribution, electromagnetic field distribution, and power ripple field distribution inside the gateway entity in a three-dimensional mesh form. The logical topology mapping content is the output of the logical task scheduling evolution kernel, which represents the protocol message parsing flow path, thread scheduling dependency relationship, and memory buffer pool occupancy topology in a directed graph topology form. The two are aligned through a spatiotemporal reference mechanism to realize the state association and joint output of the physical field mesh and logical topology nodes under the premise of timestamp synchronization.

[0026] For dynamic twin information, the generation of dynamic twin information is based on the deep fusion of joint state and spatiotemporal evolution information. Specifically, the joint state determined in step S111 is used as the initial boundary condition input to the preset digital twin space. The explicit load state in the joint state, such as CPU interrupt utilization and buffer queue length, is directly mapped to the initial computing power load parameters and initial queue depth values ​​in the logical task scheduling evolution kernel. The implicit degradation state, such as the increment of the equivalent series resistance of the capacitor and the thermal migration accumulation, is mapped to the thermal resistance offset coefficient of the power module and the thermal conductivity attenuation factor of the main chip packaging layer in the physical thermal transfer evolution kernel.

[0027] After completing the initial condition configuration, the system calls the spatiotemporal evolution information matching the current working condition from the preset evolution model library. Among them, the physical heat transfer evolution kernel uses the finite element method to perform gridded iterative calculations on the temperature field, electromagnetic field and ripple field inside the gateway, while the logical task scheduling evolution kernel performs discrete event-driven simulation and deduction on the parsing and flow of protocol messages, thread scheduling dependencies and memory buffer pool occupancy status based on the discrete event system specification.

[0028] During the aforementioned evolutionary calculations, the system simultaneously receives spatiotemporal changes from the on-site physical environment, such as the diurnal periodic gradual variation curve of ambient temperature, the transient amplitude and duration of electromagnetic pulses in the workshop, and the alternating waveforms of cabinet humidity. These spatiotemporal changes are treated as external disturbance sources and are superimposed onto the boundary convection heat transfer coefficient of the physical heat transfer evolutionary core and the interrupt response link of the logical task scheduling evolutionary core in a time-synchronized manner. As the evolutionary calculations progress, the system outputs the current three-dimensional gridded physical field distribution and the logical topology mapping in the form of a directed graph within each discrete time step, and stores the two in association using a unified timestamp label, thereby forming a dynamic twin information that is synchronized with the physical gateway in terms of time axis and spatial state and is continuously updated.

[0029] Specifically, the aforementioned determined industrial gateway joint state is input into the digital twin space as the initial boundary conditions: in the explicit load state, "CPU interrupt utilization rate of 85% and Ethernet buffer queue of 800KB" is mapped to the initial computing power load and queue depth boundary of the logical task scheduling evolution core; in the implicit degradation state, "capacitor ESR degrades to 25mΩ and thermal migration accumulation of 0.15" is mapped to the initial attenuation coefficient of the power module voltage regulation and filtering performance and the initial thermal resistance increment of the main chip thermal grease in the physical thermal transfer evolution core; based on the above severe initial boundary conditions, the digital twin space matches the spatiotemporal evolution information corresponding to high load and high degradation conditions, and establishes the current evolution benchmark point for virtual-real synchronization of the industrial gateway.

[0030] The spatiotemporal changes in the industrial gateway deployment workshop were captured: the workshop's high daytime temperatures caused a diurnal temperature variation component ranging from 25°C to 40°C; the starting of a high-power motor nearby generated electromagnetic pulse transients; and the humidity inside the cabinet fluctuated. These spatiotemporal changes were input into the twin space for continuous evolution. The ambient temperature rising to 40°C, coupled with the main chip's thermal migration accumulation of 0.15, caused a sharp deterioration in the convective heat dissipation boundary of the physical heat transfer evolution core. At the same time, the electromagnetic pulse transients were superimposed on the transmission link of high-frequency image data from 8 OPCUA vision devices, simulating the transient impact of external electromagnetic fields on the message verification process in the logical task scheduling evolution core, driving the twin to evolve synchronously with the workshop environment.

[0031] The system continuously evolves and generates dynamic twin information: the multidimensional physical field distribution outputs a three-dimensional temperature field cloud map inside the industrial gateway, clearly showing that under the dual effects of an ambient temperature of 40℃ and an ESR degradation of 25mΩ, a local thermal stress accumulation spot of 115℃ appears in the PCB area below the main chip, and the peak-to-peak value of the power module output ripple field expands to 120mV; the logical topology mapping outputs a directed graph inside the current industrial gateway, showing that the low-priority polling task nodes of the 64 Modbus RTU transmitters are blocked and suspended due to CPU interrupts occupying 85%, while the 8 OPCUA image data streams occupy 70% of the node weight in the buffer pool topology and form a critical path; thus, the three-dimensional hot spot physical field distribution and the protocol message blocking logical topology are locked together at microsecond timestamps, accurately determining the spatiotemporally synchronized and continuously evolving dynamic twin information of the industrial gateway's "high temperature physical degradation induced logical scheduling congestion".

[0032] refer to Figure 3 In step S12, the specific steps are as follows: S121: Perform high-dimensional manifold analysis on dynamic twin information to remove redundant environmental noise and extract twin state features that characterize the essential operating mode of the gateway. Project the twin state features onto the preset gateway operating phase space through nonlinear manifold mapping. S122: Reconstruct the dynamic trajectory of the industrial gateway in the gateway operation phase space, capture the temporal perturbation factor at the trajectory bifurcation point in the phase space, and tensor-quantize and fuse the temporal perturbation factor with the state evolution direction in the gateway operation phase space to generate a dynamic simulation excitation sequence containing multi-scale excitation components and prediction attributes.

[0033] In the embodiments of this application, high-dimensional manifold analysis is performed on the dynamic twin information to remove redundant environmental noise and extract twin state features that characterize the essential operating mode of the gateway. The twin state features are then projected onto a preset gateway operating phase space through nonlinear manifold mapping.

[0034] At this point, the dynamic twin information containing multidimensional physical field distribution and logical topological mapping is analyzed using high-dimensional manifold analysis. The dynamic twin information is constructed into a high-dimensional tensor manifold, and the geodesic distance and intrinsic dimension on the manifold are calculated using local tangent space arrangement or diffusion mapping algorithms. Based on the characteristic that environmental noise in manifold space is usually high-frequency scattered and lacks topological continuity, an adaptive frequency domain-topology joint filter based on manifold curvature is designed. Isolated scattered points and discontinuous perturbation components with abrupt changes in manifold curvature are identified as redundant environmental noise and stripped away, thereby filtering out non-essential interference caused by spatiotemporal changes that fluctuate with the on-site physical environment, while preserving the main topological structure of the manifold.

[0035] Optionally, performing high-dimensional manifold analysis and removing redundant environmental noise from the dynamic twin information is achieved by constructing a high-dimensional manifold dimensionality reduction framework with the local tangent space arrangement algorithm as the core and introducing an adaptive filtering mechanism based on manifold curvature. Specifically, the grid node data representing the multidimensional physical field distribution in the dynamic twin information generated in step S112 is concatenated with the edge weight sequence data representing the logical topological mapping to form a high-dimensional observation vector sequence. Each time segment corresponds to a data point in a high-dimensional space, and all time segments together span a high-dimensional tensor manifold.

[0036] The system employs a local tangent space permutation algorithm to calculate the tangent space coordinates within the neighborhood of each data point, and maps the high-dimensional manifold to a low-dimensional embedding space by minimizing the global permutation error. During this process, the system simultaneously calculates the local manifold curvature at each data point, which is quantitatively characterized by the rate of change of the principal directions of the tangent space within the neighborhood. For isolated data points where the manifold curvature abruptly changes, and for perturbation trajectory segments lacking temporal continuity due to external electromagnetic pulses or temperature fluctuations, the system identifies them as redundant environmental noise.

[0037] To achieve automatic noise removal, the system designs an adaptive threshold filter: this filter uses the median value of the global manifold curvature as a reference and sets a dynamically adjusted curvature tolerance bandwidth. Any data point whose curvature exceeds this bandwidth is identified as a noise point and removed from the manifold backbone topology. After removing noise points, the system re-performs a local tangent space arrangement on the remaining data points, forming a smooth, continuous manifold backbone topology in a low-dimensional embedding space that reflects the inherent operating rules of the gateway. Based on this, the system extracts twin state features representing the essential operating mode along the principal tangent vector direction of the manifold backbone, such as the concentrated anisotropic gradient of the thermal stress field or the congestion inertia direction of the communication link.

[0038] On the manifold backbone topology after stripping away redundant environmental noise, a criterion for determining the essential operating mode is defined based on the operating mechanism of industrial gateways. The essential operating mode characterizes the steady-state or critical-state characteristics of the gateway in the dimensions of computing, communication, and heat dissipation. Using the tangent vector field and principal curvature direction on the manifold, twin state features characterizing the essential operating mode of the gateway are extracted. The twin state features include computing power saturation features, communication link congestion inertia features, and thermal field aggregation anisotropy features. These features effectively eliminate the masking of occasional transient interference and accurately reflect the intrinsic evolution trend of gateway resource allocation and physical degradation.

[0039] A pre-defined gateway operating phase space is constructed. This phase space is a three-dimensional orthogonal space spanned by the computational load axis, the communication load axis, and the thermal stress load axis. Its coordinate basis vectors correspond to the core performance degradation index of the gateway. A distance-preserving nonlinear manifold mapping algorithm, such as the isomap or an extended form of Locally Linear Embedding (LLE), is used to embed the twin state features on the high-dimensional manifold backbone topology into the gateway operating phase space in a low-dimensional way. During the projection process, the geodesic distance between high-dimensional manifold data points and the local neighborhood topology remain unchanged. This reduces the dimensionality of complex dynamic twin information into an analyzable phase point trajectory and phase plane distribution in the phase space, thus completing the geometric representation of the gateway's operating state.

[0040] Specifically, high-dimensional manifold analysis is performed on the dynamic twin information of the industrial gateway: the dynamic twin information containing "the physical field of a local hot spot at 115℃ and the OPCUA message node occupying 70% of the buffer pool" is constructed into a high-dimensional tensor manifold; at this time, the transient components of electromagnetic pulses generated by the start and stop of high-power motors next to the workshop are represented as isolated high-frequency scattered points deviating from the main topology on the manifold, and the global thermal background translation caused by the day and night temperature fluctuations of the cabinet is represented as a low-frequency slowly varying base. By calculating the manifold curvature, an adaptive frequency domain-topology joint filter is used to filter out the above-mentioned isolated scattered points with curvature abrupt changes and the low-frequency slowly varying base, that is, the redundant environmental noise caused by motor electromagnetic interference and day and night temperature difference is stripped away, while the manifold main topology formed by the internal operating state of the gateway is retained.

[0041] Twin state features are extracted from the manifold backbone topology after noise removal: Based on the industrial gateway mechanism, the tangent vector field is calculated on the manifold backbone to extract features that characterize the essential operating mode. Specifically, these include the "communication link congestion inertia feature" caused by the continuous high-frequency image transmission of the 8-channel OPCUA vision device, which is manifested as a tangent vector component that grows unidirectionally along the communication dimension in high-dimensional space, and the "thermal field aggregation anisotropy feature" caused by the ESR degrading to 25mΩ and the main chip being fully loaded, which is manifested as the gradient feature of the local curvature of the manifold shrinking sharply in the thermal stress dimension. This feature accurately reflects the intrinsic degradation trend of the industrial gateway's computing power being occupied by communication tasks and the deterioration of heat dissipation.

[0042] The extracted twin state features are projected onto a preset gateway operating phase space: a three-dimensional operating phase space of the industrial gateway is constructed, spanned by "ARM core computing power load rate", "uplink gigabit Ethernet bandwidth utilization rate" and "main chip junction temperature over-temperature ratio". Using a distance-preserving nonlinear manifold mapping algorithm, the "communication link congestion inertia characteristics" and "thermal field aggregation anisotropy characteristics" on the high-dimensional manifold backbone are projected onto this phase space. During the projection process, the geodesic distance of the high load of OPCUA on Modbus low priority tasks in the original high-dimensional topology is strictly kept unchanged. The complex dynamic twin information of the industrial gateway is transformed into a cluster of phase points in the phase space located in the quadrant of "high communication load - high computing power load - high thermal stress". The cluster of phase points exhibits a geometric trajectory approaching the boundary of the phase space, realizing an intuitive geometric representation of the operating state.

[0043] Furthermore, the dynamic trajectory of the industrial gateway is reconstructed in the gateway's operating phase space, and the temporal perturbation factor at the trajectory bifurcation point in the phase space is captured. The temporal perturbation factor is then tensor-fused with the state evolution direction in the gateway's operating phase space to generate a dynamic simulation excitation sequence containing multi-scale excitation components and prediction attributes. Dynamic twin information is introduced, and the twin state characteristics are further controlled, improving the accuracy of the dynamic simulation excitation sequence.

[0044] At this point, in the preset gateway operation phase space, based on the discrete sequence of phase point evolution over time, the differential geometric approximation algorithm in phase space reconstruction theory is used to fit and generate a continuous industrial gateway operation dynamic trajectory; the vector field Jacobian matrix of the dynamic trajectory along each orthogonal coordinate axis is calculated, and the sign flipping state of the eigenvalues ​​of the Jacobian matrix is ​​monitored in real time; when the sign of the eigenvalue is detected to be flipped and the local curvature of the dynamic trajectory exceeds the set bifurcation threshold, the spatiotemporal node is determined to be the trajectory bifurcation point. The trajectory bifurcation point indicates that the gateway operation state is about to undergo a qualitative change, such as sliding from a steady state to a collapse state; at this trajectory bifurcation point, the small initial value sensitive fluctuation that causes the trajectory topological change is extracted as a temporal disturbance factor. The temporal disturbance factor includes micro-task scheduling lock-up pulse, sudden protocol retransmission storm induced sequence, and transient thermal stress breakdown critical ripple.

[0045] To address temporal perturbation factors, after projecting the twin state characteristics onto the gateway's operational phase space, the system continuously tracks the position changes of phase points at fixed time steps and uses differential geometry to fit a smooth dynamic trajectory curve from the discrete phase point sequence. At each trajectory point, the system calculates the Jacobian matrix along each coordinate axis, which quantitatively describes the local scaling and rotation characteristics of the trajectory in its neighborhood at that point.

[0046] The system monitors the real part sign of all eigenvalues ​​of the Jacobian matrix in real time: when the real part of all eigenvalues ​​is negative, it indicates that the trajectory is in a locally stable convergent state; when the real part of a certain eigenvalue flips from negative to positive and its magnitude exceeds the preset bifurcation sensitivity threshold, the system determines that the spatiotemporal node is the trajectory bifurcation point, indicating that the gateway's operating status is about to undergo a qualitative change.

[0047] At the bifurcation point, the specific steps for the system to capture the time-series perturbation factor are as follows: calculate the difference vector between the actual observed dynamic trajectory at the bifurcation point and the unperturbed reference trajectory extrapolated from historical data; orthogonally decompose the difference vector along each coordinate axis of the phase space to obtain the deviation components in each dimension; perform local fluctuation analysis on the time series of each deviation component to extract the spike pulse component with a frequency higher than the gateway's conventional dynamic response bandwidth and the transient coupling component exhibiting a decaying oscillation pattern; encapsulate these components according to their role in the trajectory bifurcation process to form the time-series perturbation factor. The tangent vector of the dynamic trajectory in the gateway's operational phase space at the current moment is extracted and used as a first-order directed tensor representing the macroscopic evolution direction of the gateway state. The aforementioned captured microscopic temporal perturbation factors are constructed as a second-order covariant perturbation tensor, which represents the nonlinear coupling impact effect between perturbation components in each dimension. Based on tensor outer product and contraction operations, the first-order directed tensor and the second-order covariant perturbation tensor are tensor-fused, that is, the macroscopic evolution direction is used to directionally modulate the microscopic perturbation tensor, while the microscopic perturbation tensor is used to correct the local curvature and anisotropy characteristics of the macroscopic evolution tensor, thereby generating a high-order fused evolution tensor that combines macroscopic trend inertia and microscopic mutation excitation characteristics.

[0048] Based on the aforementioned high-order fusion evolution tensor, a progressive numerical expansion and dimensionality reduction analysis of the phase space state is performed on the time axis to generate a dynamic simulation excitation sequence. The dynamic simulation excitation sequence is decoupled into multi-scale excitation components, which include low-frequency macroscopic trend excitation components expanded by the first-order evolution tensor, and high-frequency burst pulse excitation components and transient coupled oscillation excitation components obtained by the analysis of the second-order perturbation tensor. At the same time, based on the dynamic characteristics of eigenvalue reversal at the trajectory bifurcation point, predictive attributes are assigned to each sequence node in the excitation sequence. The predictive attributes include bifurcation timestamp, critical phase transition probability, and instability propagation direction, thereby generating a dynamic simulation excitation sequence that presents both the normal decay of the gateway and accurately embeds extreme degradation mutation points.

[0049] Specifically, in the three-dimensional operating phase space of the industrial gateway spanned by "ARM core computing power load rate", "uplink gigabit Ethernet bandwidth utilization rate" and "main chip junction temperature over-temperature ratio", the dynamic trajectory is fitted and reconstructed. When the industrial gateway simultaneously bears the peak of 8 OPCUA image data and 64 ModbusRTU polling requests, the dynamic trajectory approaches the dual saturation boundary of computing power and bandwidth. At this time, the eigenvalue of the Jacobian matrix flips from negative to positive, and the local curvature suddenly increases, which is determined to be the trajectory bifurcation point. At this bifurcation point, the timing disturbance factor that caused this change is captured, namely the "micro-task scheduling deadlock pulse" generated by the competition for CPU time slices between the OPCUA subscription callback thread and the ModbusRTU polling interrupt handling thread, and the "sudden protocol retransmission storm induced sequence" caused by the overflow of the network port PHY chip's transmit and receive FIFO due to high temperature.

[0050] The tangent vector at the bifurcation point of the industrial gateway's dynamic trajectory is extracted. This tangent vector serves as a first-order directed tensor pointing to the macroscopic collapse evolution direction of "100% computing power full load - 100% bandwidth congestion - junction temperature exceeding limit". The captured "microscopic task scheduling deadlock pulse" and "protocol retransmission storm induced sequence" are constructed into a second-order covariant perturbation tensor to characterize the cross-coupling impact of concern and retransmission in the computing power and bandwidth dimensions. Through tensor fusion, the macroscopic collapse evolution direction amplifies the microscopic perturbation. At the same time, the microscopic perturbation tensor causes the originally smooth macroscopic evolution direction to be locally distorted, generating a high-order fused evolution tensor indicating that the industrial gateway is about to "crash".

[0051] The high-order fusion evolution tensor is progressively unfolded on the time axis to generate a dynamic simulation stimulus sequence. This sequence contains multi-scale stimulus components: the low-frequency macro-trend stimulus component simulates the continuous ramping process of gateway computing power and bandwidth as the OPCUA frame rate increases; the high-frequency burst pulse stimulus component accurately simulates the spike impact of CPU scheduling lockup at the bifurcation point; and the transient coupling oscillation stimulus component simulates the bandwidth oscillation of TCP retransmission storm caused by lockup. In addition, the sequence nodes are given predictive attributes, indicating that under the current stimulus injection, the critical phase transition probability of the industrial gateway experiencing "watchdog reset" at the subsequent 500-millisecond bifurcation timestamp is 85%, and the instability propagation direction is reverse propagation from the communication scheduling module to the power management module, thus providing high-fidelity stimulus input including micro-mutation prediction for subsequent advanced simulations.

[0052] refer to Figure 4 In step S13, the specific steps are as follows: S131: The dynamic simulation excitation sequence is used as the driving content and loaded into the dynamic twin information, thereby starting the advanced time series inference engine in the twin space. Based on the current boundary conditions of the dynamic twin information and the preset evolution constraints, recursive inference is performed to determine the operating trajectory of the industrial gateway in the future time period. S132: Based on the identification of the running trajectory, the corresponding topological manifold is extracted and compared with the ideal running manifold of the industrial gateway to identify the drift between the virtual and real in the dimensions of computing resources, communication bandwidth and processing latency. The drift is then mapped in a multi-dimensional space and the norm is solved to determine the resource allocation deviation degree that characterizes the virtual-real deviation of the gateway.

[0053] In the embodiments of this application, the dynamic simulation excitation sequence is used as the driving content and loaded into the dynamic twin information, thereby starting the advanced temporal inference engine in the twin space. Based on the current boundary conditions of the dynamic twin information and the preset evolutionary constraints, recursive inference is performed to determine the operating trajectory of the industrial gateway in the future time period. This approach is compatible with the overall consideration of the current boundary conditions of the dynamic twin information and the preset evolutionary constraints, ensuring the accuracy of the operating trajectory of the industrial gateway in the future time period.

[0054] At this point, the aforementioned generated dynamic simulation excitation sequence containing multi-scale excitation components and prediction attributes is used as driving content and loaded into the dynamic twin information through a spatiotemporal reference alignment mechanism. Then, the low-frequency macroscopic trend excitation components, high-frequency burst pulse excitation components, and transient coupled oscillation excitation components in the dynamic simulation excitation sequence are mapped and injected into the scheduling directed edges of the multidimensional physical field distribution grid nodes and logical topology mapping of the dynamic twin information according to their timestamps and prediction attributes. This breaks the historical following state of the digital twin space, activates the built-in advanced temporal inference engine, and switches the twin space from a passive mapping mode to an active prediction inference mode.

[0055] For the advanced time series extrapolation engine, the recursive extrapolation mechanism of the advanced time series extrapolation engine is based on the current boundary conditions of the dynamic twin information and the preset evolution constraints. It adopts a combination of explicit time advancement and stepwise constraint clamping. At this time, the dynamic simulation excitation sequence generated in step S122 is injected into the corresponding multidimensional physical field distribution grid nodes and directed edges of the logical topology mapping in the dynamic twin information according to its timestamp and prediction attributes. After loading, the system locks the specific values ​​of all state variables in the dynamic twin information at the current time, including but not limited to temperature, electric field strength, ripple voltage on the grid nodes, and thread occupancy weight, queue depth and message forwarding rate on the logical nodes. These state variables together constitute the current boundary conditions of the advanced extrapolation.

[0056] The joint state of the dynamic twin information at the current moment is locked as the current boundary condition for the advance simulation. At the same time, preset evolutionary constraints are retrieved. The evolutionary constraints include nonlinear constraints on processor computing power decay, physical limit truncation constraints on bus bandwidth, and transient resistance-capacitance delay constraints on heat dissipation. In the advance timing simulation engine, recursive simulation is performed with a set time step: within each simulation time step, the current boundary condition and the injected dynamic simulation excitation sequence are coupled by mechanics and logic, and the calculation result is forcibly clamped to the boundary envelope of the evolutionary constraints. The clamped output state is used as the updated boundary condition for the next simulation time step, thereby realizing recursive rolling and state transfer on the time axis and ensuring that the simulation trajectory does not deviate from the physically realizable domain.

[0057] As the recursive deduction mechanism iterates continuously along the time axis, the advanced time-series deduction engine gradually calculates the multidimensional physical field distribution state and logical topology mapping state of the industrial gateway at each discrete moment in the future time period. The state sequences at each discrete moment are then fitted into a line in the preset gateway operation phase space to determine the operating trajectory of the industrial gateway in the future time period. The operating trajectory not only includes the position information of the phase points, but also carries the trajectory tangent slope and curvature change information at each deduction moment, fully characterizing the phase space phase transition path and critical state approximation trend of the industrial gateway under the dynamic simulation excitation and evolutionary constraints within the future time window.

[0058] Specifically, the aforementioned generated dynamic simulation stimulus sequence is loaded into the dynamic twin information of the industrial gateway: the high-frequency burst pulse stimulus component simulating CPU scheduling lockup is precisely injected into the competing node of "OPCUA callback thread and Modbus polling interrupt thread" in the logical topology mapping content of the dynamic twin information; the transient coupling oscillation stimulus component simulating TCP retransmission storm is injected into the directed edge of "uplink MQTT publish queue"; at the same time, the low-frequency macro trend stimulus component is superimposed on the grid node of the physical field distribution of the main chip, thereby activating the advanced timing inference engine of the industrial gateway and entering the active prediction mode of future operating conditions.

[0059] The current joint state of the industrial gateway, namely "85% CPU interrupt usage, 800KB buffer queue, and 75℃ main chip junction temperature," is locked as the current boundary condition. Pre-defined evolution constraints are introduced: the computing power limit of the industrial gateway's ARM Cortex-A72 processor (20,000 DMIPS) is used as a non-linear constraint for computing power attenuation; the gigabit Ethernet limit of 1Gbps is used as a physical bandwidth truncation constraint; and the heat sink thermal resistance of 0.5℃ / W is used as a transient resistance-capacitance delay constraint for heat dissipation. In the inference engine, the inference is recursively performed in 10-millisecond steps. When the inference reaches the scheduling lockout pulse injection, the logical topology calculates that the CPU demand surges to 30,000 DMIPS, but is forcibly clamped to 20,000 DMIPS due to the computing power limit truncation constraint, causing the task to be suspended. The suspended state is superimposed with the transient oscillation excitation of the next step, causing the MQTT queue to overflow. This overflow state is updated as the boundary condition for the next step, and so on, recursively rolling.

[0060] The engine extrapolates the state within the next second in 100 steps, calculating the distribution sequence of the hot spot rapidly spreading from 75℃ to 95℃ in the physical field of the industrial gateway during this period, as well as the mapping state sequence of Modbus polling completely timeout and MQTT publish link disconnection in the logical topology. The above discrete state sequence is mapped to the three-dimensional phase space of "ARM core computing power load rate - uplink bandwidth utilization rate - main chip junction temperature overheat ratio", and the running trajectory of the industrial gateway is fitted and determined: the trajectory starts from the initial phase point, rises sharply on the surface of dual constraints of computing power and bandwidth, and the trajectory curvature changes abruptly at about 600 milliseconds, quickly penetrating the safety threshold surface of "junction temperature overheat ratio 1.5", showing a typical phase space divergence and collapse approach trend, thus accurately determining the complete running trajectory of the industrial gateway that will experience thermal shutdown and communication paralysis within the next second.

[0061] Furthermore, based on the identification of the operating trajectory, the corresponding topological manifold is extracted and compared with the ideal operating manifold of the industrial gateway to identify the drift between the virtual and real worlds in terms of computing resources, communication bandwidth, and processing latency. The drift is then mapped in a multidimensional space and its norm is solved to determine the resource allocation deviation degree that characterizes the virtual-real deviation of the gateway. This approach takes into account the overall consideration of identifying the drift between the virtual and real worlds in terms of computing resources, communication bandwidth, and processing latency, ensuring the accuracy of the resource allocation deviation degree that characterizes the virtual-real deviation of the gateway.

[0062] At this point, the geometric and dynamic identification of the operating trajectory of the aforementioned industrial gateway in the future time period is performed. A sliding window tangent space extraction algorithm is used to capture the topological manifold spanned by the operating trajectory in phase space over time, which is defined as the actual inferred manifold. At the same time, the ideal operating manifold of the industrial gateway under optimal resource allocation and no degradation conditions is retrieved. The ideal operating manifold represents the low curvature smooth manifold of the gateway under the equilibrium state of computation, communication and heat dissipation. The actual inferred manifold and the ideal operating manifold are registered and compared for differences under the same time stamp sequence. By calculating the geodesic distance and normal vector deflection angle between the corresponding grid nodes of the two manifolds, the manifold distortion characteristics between the two manifolds are identified.

[0063] Guided by the aforementioned manifold distortion characteristics, the difference between the actual projected manifold and the ideal operating manifold is orthogonally decomposed into the core operating dimensions of the gateway, thereby identifying the drift between the virtual and real manifolds in the dimensions of computing resources, communication bandwidth, and processing latency. At this time, the difference between the projected area of ​​the actual projected manifold on the computing power coordinate axis and the projected area of ​​the ideal operating manifold is calculated to obtain the computing resource drift; the difference in the integral of the bandwidth occupancy rate on the communication bandwidth coordinate axis is calculated to obtain the communication bandwidth drift; and the difference in the arc length and the phase shift of the tangent vector of the two manifolds along the time axis are calculated to obtain the processing latency drift. The drift in each dimension accurately reflects the degree of deviation of the gateway due to resource contention and degradation intensification in the future.

[0064] The calculated resource drift, communication bandwidth drift, and processing latency drift are constructed into a three-dimensional drift vector and mapped to a preset resource penalty space. The resource penalty space is assigned a weighted coefficient matrix based on the sensitivity of different gateway operating modes. In the resource penalty space, the norm of the weighted three-dimensional drift vector is solved, specifically using the weighted Hamming norm or the second-order norm to calculate its magnitude, thereby scalarizing the multi-dimensional drift and determining the resource allocation deviation degree that characterizes the virtual-to-real deviation of the gateway. The resource allocation deviation degree is a non-negative scalar, and its magnitude directly reflects the severity of the deviation of the actual resource allocation state of the gateway from the ideal security state at future moments, providing a quantitative basis for subsequent dynamic threshold determination and reverse optimization.

[0065] Specifically, the operational trajectory of the aforementioned industrial gateway, which is "approaching the collapse threshold" within the next second, is identified. The actual manifold spanned by this trajectory in the "computing power-bandwidth-latency" phase space is extracted. Due to the impact of sudden scheduling lock-up and retransmission storms, this manifold exhibits distortion characteristics of local severe distortion and high curvature folds. The ideal operational manifold of the industrial gateway when handling concurrent traffic of the same scale of OPCUA and Modbus, with each protocol thread evenly allocated time slices, is retrieved. This ideal manifold presents as a smooth and convergent two-dimensional manifold surface. The two are registered and compared, and it is found that the actual manifold undergoes a geometric topological abrupt change at 600 milliseconds on the time axis, which deviates significantly from the smooth surface of the ideal manifold.

[0066] The aforementioned manifold distortion features were orthogonally decomposed and identified along three core dimensions: In terms of computing power, the actual manifold experienced a persistent 100% CPU utilization due to scheduling deadlock, while the ideal manifold should have a stable oscillation of 65% CPU utilization during the same period. This resulted in a 35% persistent positive overload in computing resources. In terms of bandwidth, the actual manifold experienced a 50MHz frequency spike in uplink bandwidth utilization due to TCP retransmission storms, while the ideal manifold maintained a stable 70% bandwidth utilization. This resulted in a 30% transient congestion in communication bandwidth. In terms of latency, the actual manifold experienced a surge in Modbus polling response time from the ideal 10ms to 500ms due to queue overflow. This resulted in a 490ms severe lag in processing latency.

[0067] The calculations for "resource drift: 35% overload", "communication bandwidth drift: 30% congestion", and "processing latency drift: 490ms lag" are constructed into a three-dimensional drift vector Vdrift=[0.35,0.30,0.49], and mapped to the resource penalty space of the industrial gateway. Since the industrial gateway is extremely sensitive to the processing latency of downlink control commands, Modbus delays directly cause production line downtime. Therefore, this resource penalty space assigns a penalty weight of 0.5 to the latency dimension, and weights of 0.3 and 0.2 to computing power and bandwidth, respectively. The second norm of the weighted vector is then calculated. The output resource allocation deviation was 0.288, which is far greater than the deviation threshold for steady-state operation of industrial gateways, usually <0.1. This quantitatively confirms that the gateway will experience a serious risk of virtual-real resource allocation deviation and loss of control in the next 600 milliseconds.

[0068] refer to Figure 5 In step S14, the specific steps are as follows: S141: If the resource allocation deviation exceeds the preset dynamic threshold based on the current fault tolerance margin of the system, the strategy optimization mechanism of the running trajectory is triggered. In the feasible solution space of the gateway control parameters, a global adaptive content with the goal of minimizing the resource allocation deviation is constructed, and a reverse optimization control instruction is generated through heuristic optimization. S142: Mark the execution path of the reverse tuning control command and input it into the digital twin space. Synchronously collect the physical response data of the industrial gateway and determine the response residual between the virtual expected state and the physical actual state. If the response residual meets the convergence condition, use the response residual as a compensation factor to reverse correct the model parameters and state variables of the dynamic twin information. If the response residual does not meet the convergence condition, reconstruct the timing disturbance factor based on the response residual and re-trigger the advanced timing deduction until the virtual-real closed loop converges.

[0069] In the embodiments of this application, if the resource allocation deviation exceeds the preset dynamic threshold based on the current fault tolerance margin of the system, the strategy optimization mechanism of the running trajectory is triggered. A global adaptive content aimed at minimizing the resource allocation deviation is constructed in the feasible solution space of the gateway control parameters, and a reverse optimization control instruction is generated through heuristic optimization.

[0070] At this point, the resource allocation deviation degree, which represents the virtual-to-real deviation of the gateway, calculated above, is obtained and compared with a preset dynamic threshold. The dynamic threshold is not a fixed constant, but a variable that is dynamically adjusted based on the current fault tolerance margin of the industrial gateway system. At this point, a fault tolerance margin assessment model is constructed based on the severity of the gateway's current implicit degradation state and the thermal stress margin of the physical field. When the degradation degree deepens or the thermal margin is reduced, the system fault tolerance margin decreases, and the value of the dynamic threshold is adaptively lowered accordingly to improve the sensitivity to potential instability risks. If the resource allocation deviation degree exceeds the dynamically adjusted dynamic threshold, it is determined that the gateway will break through the safe operation boundary according to the current trajectory, and a strategy optimization mechanism for the operation trajectory is triggered.

[0071] After triggering the strategy optimization mechanism, the set of controllable parameters of the industrial gateway is extracted. The set of control parameters includes thread priority allocation weight, clock frequency reduction step, communication buffer queue ratio, and protocol message retransmission backoff time. The upper and lower bounds of each control parameter are determined according to the physical limits and protocol specifications, thereby constructing a feasible solution space for the gateway control parameters. Within the feasible solution space, with minimizing the aforementioned resource allocation deviation as the optimization objective, a global adaptive content is constructed by combining gateway mechanism constraints and logical scheduling rules. The global adaptive content is a multimodal fitness landscape function defined on the feasible solution space. It maps each set of candidate control parameters to the reciprocal of the resource allocation deviation under prediction and inference, and marks the penalty barrier generated by the constraint boundary in the fitness landscape, thereby characterizing the global adaptability of the control parameter combination to eliminate the virtual-real deviation.

[0072] To address the non-convex, multi-extremal, and high-dimensional coupling characteristics of the global adaptive content, a heuristic optimization algorithm, such as an improved genetic algorithm or particle swarm optimization algorithm, is employed to iteratively search the feasible solution space. During the search process, crossover, mutation, or velocity update operators are designed to guide the candidate parameter population to evolve along the gradient upward direction of the fitness landscape. Simultaneously, chaotic perturbations are introduced to escape local optima and gradually approach the globally optimal parameter combination that minimizes resource allocation deviation. This globally optimal parameter combination is extracted and inversely deduced into specific control actions that can be executed by the physical gateway. These actions are then encapsulated as reverse tuning control instructions. The execution direction of these reverse tuning control instructions is opposite to the initial evolution direction that caused the resource allocation deviation, aiming to forcibly pull the gateway's operating state back from the deviation trajectory to the ideal manifold envelope.

[0073] Specifically, the aforementioned calculation shows that the resource allocation deviation of the industrial gateway is 0.288 in the next 600 milliseconds. At this time, the dynamic threshold of the industrial gateway is obtained. Since the junction temperature of the main chip of the industrial gateway has reached 75°C and the capacitor ESR has degraded to 25mΩ, the fault margin of system heat dissipation and voltage regulation has been significantly reduced. Based on this, the fault margin assessment model lowers the dynamic threshold to 0.15. It can be seen that the resource allocation deviation of 0.288 exceeds the dynamic threshold of 0.15, and it is determined that the industrial gateway is about to experience thermal failure and communication paralysis, and the strategy optimization mechanism for this crash operation trajectory is immediately triggered.

[0074] Adjustable parameters of the industrial gateway are extracted to construct a feasible solution space including "OPCUA callback thread weight: value 0.1-0.9", "ModbusRTU polling thread weight: value 0.1-0.9", "ARM core operating frequency reduction level: 1.5GHz / 1.2GHz / 0.8GHz", and "TCP retransmission backoff time exponent: value 1-5". With the goal of minimizing the resource allocation deviation of 0.288, combined with the upper limit of computing power and bandwidth allocation constraints, a global adaptive content is constructed. In the parameter domain of "high OPCUA weight + low frequency + long backoff time", a high fitness peak is shown, while in the conventional domain of "thread weight balance + high frequency", a low penalty barrier is formed due to the inability to eliminate latency drift.

[0075] A particle swarm optimization algorithm is used to search the feasible solution space. Guided by the fitness landscape, the particle swarm overcomes local extrema, such as the depletion of computing power caused by only reducing the frequency without adjusting the thread weights, and gradually converges to the globally optimal parameter combination. This optimal combination is extracted, and reverse tuning control instructions are generated. The instructions specifically include: "forcibly reducing the weight of the OPCUA callback thread from 0.5 to 0.2", "increasing the weight of the ModbusRTU polling thread from 0.3 to 0.6 to ensure low latency of downlink control instructions", "triggering the frequency reduction level from 1.5GHz to 1.2GHz to suppress thermal stress accumulation", and "increasing the TCP retransmission backoff index from 2 to 4 to alleviate bandwidth congestion". The regulation logic of this reverse tuning control instruction is completely opposite to the deviation direction caused by the current OPCUA preemption of computing power and high-frequency retransmission, aiming to forcibly block the collapse evolution trajectory of the industrial gateway.

[0076] Furthermore, the execution path of the reverse tuning control command is marked and input into the digital twin space. The physical response data of the industrial gateway is collected synchronously to determine the response residual between the virtual expected state and the physical actual state. If the response residual meets the convergence condition, the model parameters and state variables of the dynamic twin information are reverse-corrected using the response residual as a compensation factor. If the response residual does not meet the convergence condition, the timing disturbance factor is reconstructed based on the response residual and the advanced timing deduction is retried until the virtual-real closed loop converges. At the same time, the resource allocation deviation is further controlled, and the reverse tuning control command is fully considered to realize the reverse correction of the dynamic twin information.

[0077] At this point, the execution path of the aforementioned generated reverse tuning control instruction in the logical topology mapping content and multidimensional physical field distribution content is marked. The execution path represents the thread scheduling weight migration trajectory, frequency adjustment action response link, and buffer queue reallocation topology within the instruction's scope. The reverse tuning control instruction marked with the execution path is synchronously input into the digital twin space and the physical industrial gateway. In the digital twin space, the virtual expected state is obtained based on the current dynamic twin information. At the same time, on the physical gateway side, the physical response data after the industrial gateway executes the instruction is synchronously collected by the heterogeneous communication protocol acquisition engine and onboard sensors, and the actual physical state is extracted and generated, thereby constructing a virtual and real dual-end response state pair with spatiotemporal alignment characteristics.

[0078] The difference between the virtual expected state and the physical actual state in terms of computational resource allocation, communication bandwidth usage, and processing latency is calculated, and this difference is determined as the response residual. The second norm of the response residual is compared with a preset residual convergence threshold, which is adaptively set based on the uncertainty boundary of the gateway's current implicit degradation state. If the second norm of the response residual is less than or equal to the residual convergence threshold, the response residual is determined to meet the convergence condition, indicating that the twin model's prediction of physical regulation is highly consistent with the actual physical response. If the second norm of the response residual is greater than the residual convergence threshold, the response residual is determined to not meet the convergence condition, indicating that the twin model has unmodeled dynamics or mechanism drift, and there is still a significant deviation between the virtual and real worlds.

[0079] If the response residuals meet the convergence conditions, they are transformed into compensation factors to reverse-correct the model parameters and state variables of the dynamic twin information. At this time, the state variables are corrected in the gradient direction of the residual vector, and the model parameters in the mechanism constraint equation are updated using the recursive least squares method with the goal of minimizing the residuals, thereby eliminating the drift of the twin model accumulated over time. If the response residuals do not meet the convergence conditions, the high-frequency unmodeled dynamic and nonlinear distortion components in the response residuals are extracted, and the temporal perturbation factor is reconstructed based on the components. The reconstructed temporal perturbation factor is re-injected into the gateway running phase space, and the advanced temporal inference engine is re-triggered to perform iterative prediction and strategy optimization until the response residuals meet the convergence conditions, thereby realizing the virtual-real closed-loop convergence and model self-evolution between the digital twin system and the physical entity.

[0080] Specifically, the aforementioned reverse tuning control instruction, which includes "OPCUA weight reduced to 0.2, Modbus weight increased to 0.6, main frequency reduced to 1.2GHz, and TCP backoff index increased to 4," is marked with an execution path marker. The marker indicates that the instruction is mapped to "interrupt controller priority inversion path" and "TCP / IP protocol stack backoff timer reset path" in the logical topology, and to "ARM core power supply voltage reduction trajectory" in the physical field. The instruction is synchronously sent to the industrial gateway physical entity and digital twin space. After the physical gateway executes the instruction, it collects physical response data obtained by onboard sensors, such as the actual main frequency reduced to 1.18GHz, Modbus polling latency reduced to 15ms, and actual network port packet loss rate reduced to 2%, generating the actual physical state. The twin space then deduces the ideal virtual expected state where the main frequency is strictly reduced to 1.2GHz and the latency is reduced to 12ms.

[0081] Comparing the above virtual and real dual-end states, the response residuals are extracted: there is a frequency offset residual of 0.02GHz in the main frequency dimension, caused by the nonlinearity of the physical chip's buck transient response; there is a hysteresis residual of 3ms in the Modbus delay dimension, caused by the physical interrupt context switching overhead being greater than the ideal model assumption. The second norm of this response residual is calculated to be 0.035. However, the current industrial gateway has a large fault tolerance uncertainty due to ESR degradation, so the adaptive residual convergence threshold is set to 0.02. Since 0.035 is greater than 0.02, it is determined that the current response residual does not meet the convergence condition, indicating that the current siam model has an uncovered mechanism drift in its modeling of the chip's buck transient response and interrupt switching overhead.

[0082] Since the response residuals did not meet the convergence condition, the 3ms hysteresis distortion component and the 0.02GHz frequency offset distortion component in the residuals were extracted and encapsulated as a reconstructed timing perturbation factor characterizing the "chip buck transient frequency slippage and interrupt context overhead". This reconstructed timing perturbation factor was recoupled to the gateway operation phase space of the industrial gateway, and the advanced timing inference engine was restarted. Under the new perturbation factor, the engine predicted a longer latency degradation and a slower frequency drop, thereby regenerating more conservative reverse tuning instructions, such as further increasing the Modbus weight to 0.7, or directly setting the frequency reduction step to 0.8GHz to avoid the nonlinear slippage region, and issuing them again and collecting the residuals synchronously. This process was repeated until the virtual and real response residual norm was less than 0.02. At this point, the final small residual was used as a compensation factor to correct the voltage-frequency conversion equation parameters and interrupt switching time constant of the ARM core in the dynamic twin information, completing the adaptive reverse correction and self-evolution closed loop of the industrial gateway digital twin model.

[0083] Please see Figure 6 The dynamic simulation system for the digital twin-based industrial gateway is applied to the aforementioned dynamic simulation method for the digital twin-based industrial gateway; the dynamic simulation system for the digital twin-based industrial gateway includes: The dynamic twin module 21 is used to acquire multi-source operating data of the industrial gateway, determine the joint state of the industrial gateway in combination with the mechanism constraint information of the industrial gateway, input the joint state of the industrial gateway into the preset digital twin space, and determine the dynamic twin information that is synchronized with the industrial gateway in time and space and continuously evolves. The dynamic simulation module 22 is used to extract the corresponding twin state features based on the analysis of dynamic twin information, map the twin state features to the preset gateway running phase space, and generate the corresponding dynamic simulation excitation sequence in combination with the timing disturbance factor. The running trajectory module 23 is used to input the dynamic simulation excitation sequence into the dynamic twin information and perform advanced time series extrapolation to determine the running trajectory of the industrial gateway in the future time period, and to determine the resource allocation deviation degree characterizing the virtual-real deviation of the gateway based on the identification of the running trajectory. The correction module 24 is used to generate a reverse optimization control command based on the strategy optimization of the running trajectory if the resource allocation deviation exceeds the preset dynamic threshold. By executing the reverse optimization control command, the response residual of the industrial gateway is determined, and then the dynamic twin information is adaptively reverse corrected.

[0084] It should be noted that although multiple modules are mentioned in the detailed description above, this division is not mandatory; in fact, according to the embodiments of this disclosure, the features and functions of two or more modules or described above can be embodied in one module; conversely, the features and functions of one module described above can be further divided into multiple modules to be embodied.

[0085] Other embodiments of this disclosure will readily occur to those skilled in the art upon consideration of the specification and practice of the disclosure herein; this application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not disclosed herein; the specification and embodiments are to be considered exemplary only.

[0086] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A dynamic simulation method for industrial gateways based on digital twins, characterized in that, include: The system acquires multi-source operational data from industrial gateways and determines the joint state of industrial gateways by combining the mechanism constraint information of industrial gateways. The joint state of industrial gateways is then input into a preset digital twin space, and dynamic twin information that is spatiotemporally synchronized with industrial gateways and continuously evolves is determined. Based on the analysis of dynamic twin information, the corresponding twin state features are extracted, the twin state features are mapped to the preset gateway operation phase space, and the corresponding dynamic simulation excitation sequence is generated by combining the timing disturbance factor. The dynamic simulation excitation sequence is input into the dynamic twin information and advanced time series extrapolation is performed to determine the operating trajectory of the industrial gateway in the future time period. Based on the identification of the operating trajectory, the resource allocation deviation degree characterizing the virtual-real deviation of the gateway is determined. If the resource allocation deviation exceeds the preset dynamic threshold, a reverse optimization control command is generated based on the strategy optimization of the running trajectory. By executing the reverse optimization control command, the response residual of the industrial gateway is determined, and then the dynamic twin information is adaptively reverse corrected.

2. The dynamic simulation method for industrial gateways based on digital twins according to claim 1, characterized in that, The process of acquiring multi-source operational data from the industrial gateway, determining its joint state in conjunction with its mechanistic constraint information, inputting this joint state into a preset digital twin space, and determining dynamic twin information that is spatiotemporally synchronized and continuously evolving with the industrial gateway includes: The system acquires multi-source operational data of industrial gateways under heterogeneous communication protocols, extracts protocol feature vectors from the multi-source operational data, constructs state coupling content driven by both data and mechanism in combination with the mechanistic constraint information of industrial gateways, and determines the joint state of industrial gateways containing implicit degradation state and explicit load state based on the multi-level iteration of this state coupling content.

3. The dynamic simulation method for industrial gateways based on digital twins according to claim 2, characterized in that, The process of acquiring multi-source operational data of the industrial gateway, determining the joint state of the industrial gateway in conjunction with its mechanistic constraint information, inputting the joint state of the industrial gateway into a preset digital twin space, and determining dynamic twin information that is spatiotemporally synchronized and continuously evolving with the industrial gateway, further includes: The joint state is input into a preset digital twin space. Using the joint state as the initial boundary condition, the spatiotemporal evolution information corresponding to the industrial gateway is matched in the digital twin space. The spatiotemporal change items fluctuating with the physical environment on site are continuously evolved to determine the dynamic twin information that is spatiotemporally synchronized with the industrial gateway and continuously evolves. This dynamic twin information includes multidimensional physical field distribution content and logical topology mapping content.

4. The dynamic simulation method for industrial gateways based on digital twins according to claim 1, characterized in that, The process of extracting corresponding twin state features based on the analysis of dynamic twin information, mapping the twin state features to a preset gateway operating phase space, and generating corresponding dynamic simulation excitation sequences in conjunction with timing perturbation factors includes: High-dimensional manifold analysis is performed on dynamic twin information to remove redundant environmental noise and extract twin state features that characterize the essential operating mode of the gateway. The twin state features are then projected onto a preset gateway operating phase space through nonlinear manifold mapping.

5. The dynamic simulation method for industrial gateways based on digital twins according to claim 4, characterized in that, The process of extracting corresponding twin state features based on the analysis of dynamic twin information, mapping the twin state features to a preset gateway operating phase space, and generating corresponding dynamic simulation excitation sequences in conjunction with timing perturbation factors, further includes: The dynamic trajectory of the industrial gateway is reconstructed in the gateway operation phase space, and the temporal perturbation factor at the trajectory bifurcation point in the phase space is captured. The temporal perturbation factor is then fused with the state evolution direction in the gateway operation phase space through tensor quantization, thereby generating a dynamic simulation excitation sequence containing multi-scale excitation components and prediction attributes.

6. The dynamic simulation method for industrial gateways based on digital twins according to claim 1, characterized in that, The process of inputting dynamic simulation excitation sequences into dynamic twin information and performing advanced time-series extrapolation to determine the operating trajectory of the industrial gateway in the future time period, and determining the resource allocation deviation degree characterizing the virtual-to-real deviation of the gateway based on the identification of the operating trajectory, includes: The dynamic simulation excitation sequence is used as the driving content and loaded into the dynamic twin information, thereby starting the advanced time series inference engine in the twin space. Based on the current boundary conditions of the dynamic twin information and the preset evolution constraints, recursive inference is performed to determine the operating trajectory of the industrial gateway in the future time period.

7. The dynamic simulation method for industrial gateways based on digital twins according to claim 6, characterized in that, The step of inputting the dynamic simulation excitation sequence into the dynamic twin information and performing advanced time-series extrapolation to determine the operating trajectory of the industrial gateway in the future time period, and determining the resource allocation deviation degree characterizing the virtual-real deviation of the gateway based on the identification of the operating trajectory, further includes: Based on the identification of the operating trajectory, the corresponding topological manifold is extracted and compared with the ideal operating manifold of the industrial gateway to identify the drift between the virtual and real in terms of computing resources, communication bandwidth and processing latency. The drift is then mapped in a multidimensional space and the norm is solved to determine the resource allocation deviation degree that characterizes the virtual-real deviation of the gateway.

8. The dynamic simulation method for industrial gateways based on digital twins according to claim 1, characterized in that, If the resource allocation deviation exceeds a preset dynamic threshold, a reverse optimization control command is generated based on the strategy optimization of the running trajectory. Following the execution of the reverse optimization control command, the response residual of the industrial gateway is determined, and then adaptive reverse correction is performed on the dynamic twin information, including: If the resource allocation deviation exceeds the preset dynamic threshold based on the current fault tolerance margin of the system, the strategy optimization mechanism of the running trajectory is triggered. In the feasible solution space of the gateway control parameters, a global adaptive content with the goal of minimizing the resource allocation deviation is constructed, and reverse optimization control instructions are generated through heuristic optimization.

9. The dynamic simulation method for industrial gateways based on digital twins according to claim 8, characterized in that, If the resource allocation deviation exceeds a preset dynamic threshold, a reverse optimization control command is generated based on the strategy optimization of the running trajectory. Following the execution of the reverse optimization control command, the response residual of the industrial gateway is determined, and then adaptive reverse correction is performed on the dynamic twin information. This also includes: The execution path of the reverse tuning control command is marked and input into the digital twin space. The physical response data of the industrial gateway is collected synchronously to determine the response residual between the virtual expected state and the physical actual state. If the response residual meets the convergence condition, the model parameters and state variables of the dynamic twin information are reversely corrected using the response residual as a compensation factor. If the response residual does not meet the convergence condition, the timing disturbance factor is reconstructed based on the response residual and the advanced timing deduction is retried until the virtual-real closed loop converges.

10. A dynamic simulation system for an industrial gateway based on digital twins, characterized in that, The dynamic simulation system for the digital twin-based industrial gateway is applied to the dynamic simulation method for the digital twin-based industrial gateway as described in any one of claims 1-9; the dynamic simulation system for the digital twin-based industrial gateway includes: The dynamic twin module is used to acquire multi-source operational data of the industrial gateway, and determine the joint state of the industrial gateway by combining the mechanism constraint information of the industrial gateway. The joint state of the industrial gateway is input into the preset digital twin space, and dynamic twin information that is synchronized with the industrial gateway in time and space and continuously evolves is determined. The dynamic simulation module is used to extract the corresponding twin state features based on the analysis of dynamic twin information, map the twin state features to the preset gateway running phase space, and generate the corresponding dynamic simulation excitation sequence in combination with the timing perturbation factor. The operation trajectory module is used to input the dynamic simulation excitation sequence into the dynamic twin information and perform advanced time series extrapolation to determine the operation trajectory of the industrial gateway in the future time period, and to determine the resource allocation deviation degree characterizing the virtual-real deviation of the gateway based on the identification of the operation trajectory. The correction module is used to generate reverse optimization control instructions based on the strategy optimization of the running trajectory if the resource allocation deviation exceeds the preset dynamic threshold. By executing the reverse optimization control instructions, the response residual of the industrial gateway is determined, and then the dynamic twin information is adaptively reverse corrected.

Images