Integrated production operation platform based on digital twin

Abstract

The application discloses an integrated production operation platform based on digital twin, relates to the technical field of production operation, and establishes a digital twin operation model of a physical execution layer and a virtual mapping layer by constructing state characteristic parameters of operation nodes and fusing multi-source time sequence data, and introduces a state consistency constraint function to represent the synchronization deviation between the two layers; on the basis, real-time iterative calculation is performed on a target operation execution model to obtain state deviation values of each operation node, and state consistency is checked based on signal time sequence logic and a robust online monitoring algorithm to generate an abnormal state judgment result; further, according to the abnormal type, instruction delay, path re-planning or state rollback are executed to realize dynamic correction and closed-loop control of the operation process; the application can effectively improve the accuracy of abnormal identification and the timeliness of processing in complex production operation, and enhance the stability and safety of the overall operation of the system.

Description

Technical Field

[0001] This invention relates to the field of production operation technology, and more specifically to an integrated production operation platform based on digital twins. Background Technology

[0002] In port storage, transportation, and loading / unloading operations, multi-stage collaborative operations involving ships, vehicles, and storage tanks are required. Existing technologies largely rely on discrete information systems and manual scheduling. Although digital twins and process management have been introduced, there are still issues of timing discrepancies and semantic inconsistencies between fine-grained operational states and virtual models under complex operating conditions. In particular, during the fusion of multi-source heterogeneous data (such as PID process data, equipment status data, and positioning data), it is difficult to achieve millisecond-level synchronization and consistency verification, leading to a "state drift" phenomenon in local operation nodes. This means that the virtual model has been updated while the actual equipment has not yet responded or vice versa, resulting in false command triggering, path conflicts, or hidden safety risks. Such problems are highly concealed and difficult to identify through traditional monitoring methods, becoming a key technical bottleneck restricting the reliability of integrated intelligent operation systems. Summary of the Invention

[0003] The purpose of this invention is to provide an integrated production operation platform based on digital twins to address the shortcomings of the prior art.

[0004] To achieve the above objectives, the present invention provides the following technical solution: an integrated production operation platform based on digital twins, comprising:

[0005] Data acquisition module: Acquires initial process flow data and equipment status data of each work unit in the target production operation system, constructs corresponding PID structured models, and extracts status feature parameters of each node;

[0006] State mapping module: Based on the state feature parameters, it fuses positioning data and real-time sensing data to generate a multi-source time series data set, and establishes a mapping relationship set B=(b1,b2,…,bi,…,bn) of the state of the work node, where bi represents the correspondence between the physical state and the virtual state of the i-th work node.

[0007] Consistency constraint module: Based on the mapping relationship set B, a digital twin operation model is constructed, wherein the model includes a physical execution layer and a virtual mapping layer, and a state consistency constraint function is introduced to characterize the synchronization deviation between the two layers;

[0008] Execution control module: Based on actual work instructions and scheduling strategies, it applies dynamic work load and process drive signals to the digital twin work model to obtain the target work execution model;

[0009] Deviation analysis module: Utilizes the multi-source time-series data set to perform real-time iterative calculations on the target job execution model, identifies the state deviation values ​​of each job node, verifies them according to the state consistency constraint function, and generates abnormal state judgment results;

[0010] Correction module: Based on the abnormal state determination result, execute correction strategies for the corresponding job nodes, including instruction delay, path replanning, or state rollback.

[0011] Preferably, the extraction of state feature parameters of each node includes the following steps: performing structured parsing of the process flow diagram in the target production operation system, extracting pipeline connection relationships, equipment types and process parameters, and generating PID topology data; based on the PID topology data, uniquely identifying and encoding each operation node, and establishing an association mapping relationship between node and equipment status data; acquiring the operating status data of each device, and performing timestamp alignment and data cleaning processing on the operating status data to form a unified format status dataset; and extracting state feature parameters representing the operating status of nodes from the status dataset according to preset feature extraction rules.

[0012] Preferably, the construction of the digital twin operation model based on the mapping relationship set B includes the following steps: based on the correspondence between the physical state and virtual state of each operation node in the mapping relationship set B, construct the node state set of the physical execution layer and the simulation state set of the virtual mapping layer respectively, and establish a node-to-node mapping structure between the two layers according to the PID topology data; based on the node-to-node mapping structure, perform unified dimension normalization processing on the state vectors of each corresponding node to form comparable standard state vector pairs; construct a state consistency constraint function based on the standard state vector pairs; embed the state consistency constraint function into the digital twin operation model to form a digital twin operation model that simultaneously includes the physical execution layer, the virtual mapping layer, and the state consistency constraint relationship.

[0013] Preferably, applying dynamic workload and process-driven signals to the digital twin operation model includes the following steps: acquiring actual operation instructions, parsing and prioritizing the operation instructions according to a preset scheduling strategy to generate an operation instruction sequence and corresponding time constraint parameters; mapping each operation instruction to a corresponding operation node in the digital twin operation model based on the operation instruction sequence and time constraint parameters, allocating dynamic workload parameters to each operation node, and constructing process-driven signals according to the connection relationship between operation nodes and the order of operation processes.

[0014] Preferably, the dynamic workload parameters include workload intensity, execution duration, and resource utilization ratio.

[0015] Preferably, the process driving signal is applied sequentially to each work node according to the time constraint parameters to drive the state evolution of the digital twin work model; under the combined effect of dynamic work load and process driving signal, the state of the digital twin work model is updated and the time sequence is advanced to generate the target work execution model.

[0016] Preferably, the method of using the multi-source time-series data set to perform real-time iterative calculations on the target job execution model and identify the state deviation values ​​of each job node includes the following steps: serializing the multi-source time-series data set according to a unified time step and aligning it with the node state vectors at corresponding times in the target job execution model to form a node comparison data sequence; based on the node comparison data sequence, constructing a difference calculation model between the predicted node state value and the actual observed value, and iteratively calculating each job node within a continuous time window to obtain the initial value of the node state deviation; and dynamically weighting and smoothing the initial value of the node state deviation according to the historical operating data of each job node to obtain the node state deviation value.

[0017] Preferably, the verification based on the state consistency constraint function to generate an abnormal state judgment result includes the following steps: constructing a corresponding time-series signal sequence based on the state consistency constraint function value of each job node, and establishing a signal time-series logic judgment formula by combining preset job process constraints, state duration constraints, and dependencies between preceding and subsequent nodes; inputting the time-series signal sequence into a robust online monitoring algorithm to calculate the signal time-series logic judgment formula at each time point to obtain the time-series robustness value of the corresponding job node; identifying whether the job node has state mismatch, execution lag, or process out-of-order anomalies based on the comparison result between the time-series robustness value and the preset anomaly judgment threshold; and associating and summarizing the anomaly identification results of each job node according to the job process sequence to generate an abnormal state judgment result.

[0018] Preferably, instruction delay is prioritized for low-severity anomalies, path replanning is prioritized for medium-severity anomalies, and state rollback is prioritized for high-severity anomalies.

[0019] The technical effects and advantages provided by the present invention in the above technical solution are as follows:

[0020] 1. This invention constructs a real-time iterative calculation mechanism based on a multi-source time-series data set and a target job execution model. It unifies the time alignment and differential calculation between the predicted state and the actual observed state, and combines this with dynamic weighted smoothing processing to achieve continuous quantitative characterization of the state deviation of job nodes. Compared to traditional methods based on single-moment threshold judgment, this invention can comprehensively reflect the state evolution trend within a time window, effectively suppressing the interference of instantaneous fluctuations on anomaly identification. This significantly improves the accuracy of identifying hidden deviations and gradual anomalies, thereby enhancing the stability and reliability of the production process.

[0021] 2. This invention introduces a signal timing logic judgment method based on state consistency constraint functions, and combines it with a robust online monitoring algorithm to jointly verify job flow constraints, state duration constraints, and dependencies between preceding and subsequent nodes. The timing robustness value is used to quantitatively evaluate the degree of anomalies, enabling not only the identification of state mismatch anomalies but also accurate differentiation of complex anomaly types such as execution lag and process out-of-order execution. Furthermore, through process-level correlation analysis of anomaly results, the anomaly propagation path can be characterized, achieving an improvement from single-node anomaly identification to full-process anomaly perception, thereby significantly enhancing the system's perception and decision support capabilities for complex job scenarios. Attached Figure Description

[0022] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0023] Figure 1 This is a flowchart of the integrated production operation platform based on digital twins according to the present invention.

[0024] Figure 2 This is a flowchart of the method for generating the target job execution model of the present invention.

[0025] Figure 3 This is a flowchart of the method for generating abnormal state determination results according to the present invention. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0027] For examples, please refer to Figure 1 , 2 As shown in Figure 3, the integrated production operation platform based on digital twins described in this embodiment includes:

[0028] In this embodiment, the data acquisition module is used to acquire the initial process flow data and equipment status data of each work unit in the target production operation system, construct the corresponding PID structured model, and extract the status feature parameters of each node.

[0029] First, the process flow diagram in the target production operation object is structured and parsed to extract pipeline connection relationships, equipment types and process parameters, and generate PID topology data. Specifically, the process flow diagram files corresponding to ship loading / unloading operations, truck loading / unloading operations, and tank transfer operations are first obtained. The graphic objects in the process flow diagrams are then classified according to preset layer attributes to obtain equipment elements, pipeline elements, and text elements. Subsequently, the elements to be identified are compared with the pre-established standard element library for contour similarity. When the contour similarity is greater than 0.92, the element to be identified is recognized as a valid equipment element, thus completing the identification of storage tanks, valves, pumps, loading / unloading arms, flow meters, and level gauges. After completing the equipment element identification, endpoint aggregation processing is performed on pipeline elements. When the distance between the endpoints of two line segments is no more than 3 pixels and the included angle is no more than 5°, the corresponding two line segments are merged into the same continuous pipeline. Based on this, pressure parameters, flow parameters, level parameters, and medium flow direction parameters are extracted according to the spatial adjacency relationship between text elements and equipment and pipeline elements, thereby forming PID topology data composed of node sets and connection sets. The node sets are used to represent equipment nodes, and the connection sets are used to represent medium flow paths.

[0030] After obtaining the PID topology data, each work node is uniquely identified and coded, and an association mapping relationship between work nodes and equipment status data is established. Specifically, a unique identifier code corresponding to each work node is generated according to the coding rule of "work area number - equipment category number - sequence number". The work area number is used to distinguish between docks, tank areas, and loading positions; the equipment category number is used to distinguish between storage tanks, valves, pumps, loading arms, flow meters, and level gauges; and the sequence number is generated sequentially according to the arrangement order of equipment of the same category in the process flow diagram. Subsequently, the unique identifier code is written into the data acquisition address table of the corresponding equipment, so that each work node is associated with the corresponding data acquisition address and process position, forming a ternary association table of "unique identifier code - acquisition address - process position". Further, the upstream and downstream nodes corresponding to each work node are determined according to the connection set, thereby forming a node association mapping relationship. Based on the ternary association table and the node association mapping relationship, a one-to-one correspondence between work nodes, process positions, and equipment status data is realized, providing a clear data association basis and topology constraint basis for subsequent extraction of status feature parameters.

[0031] After establishing the association mapping relationship, the operating status data of each device is acquired through the real-time acquisition interface, and the operating status data is timestamped and cleaned to form a status dataset in a unified format. Specifically, valve opening, pump start / stop status, flow rate, pressure, liquid level, and positioning coordinates are collected at a 1-second sampling period. After collection, using whole seconds as a unified time reference, data with a deviation of no more than 0.5 seconds from the collected time are mapped to the current sampling time, while data with a deviation greater than 0.5 seconds are mapped to the next sampling time, thus achieving timestamp alignment of multi-source data. After timestamp alignment, missing data is filled in using linear interpolation between two adjacent valid sample values, and data with more than three consecutive missing sampling times is marked as invalid data. For abnormal data, it is removed according to pre-set physical boundary thresholds, where the valve opening boundary is set to 0 to 100%, the liquid level boundary is set to 0 to the tank's calibrated height, the pressure boundary is set to 0 to 2.5 MPa, and the flow rate boundary is set to 0 to 2000 cubic meters per hour. After removing abnormal values, the remaining valid data is processed by median filtering based on five consecutive sample values ​​to obtain a unified format state dataset.

[0032] After obtaining the state dataset, state feature parameters representing the operating status of the work nodes are extracted from the state dataset according to preset feature extraction rules. Specifically, using five consecutive sampling times as a feature window, the steady-state value, rate of change, fluctuation amplitude, and connectivity status are calculated for each work node. The steady-state value is the arithmetic mean of the sampled values ​​within the feature window; the rate of change is the difference between the current sampled value and the previous sampled value divided by 1 second; the fluctuation amplitude is the difference between the maximum and minimum values ​​within the feature window; and the connectivity status is 1 when both upstream and downstream nodes have valid data, otherwise it is 0. Based on this, corresponding state feature parameters are extracted according to the type of work node. For valve nodes, the steady-state value and rate of change of opening are extracted; for pump nodes, the start / stop status and outlet pressure fluctuation amplitude are extracted; for storage tank nodes, the steady-state value and rate of change of liquid level are extracted; and for pipeline nodes, the steady-state value of flow rate and pressure fluctuation amplitude are extracted. Finally, the state feature parameters corresponding to each work node are collected according to a unique identifier to form a node state feature parameter set, which can be directly called for subsequent multi-source data fusion processing.

[0033] State mapping module: Based on the state feature parameters, it fuses positioning data and real-time sensing data to generate a multi-source time-series data set, and establishes a mapping relationship set B=(b1,b2,…,bi,…,bn) of the state of the work nodes, where bi represents the correspondence between the physical state and the virtual state of the i-th work node.

[0034] In this embodiment, after obtaining the state characteristic parameters of each work node, multi-source data fusion processing is performed on the state characteristic parameters to generate a multi-source time-series data set, and a mapping relationship set B=(b1,b2,…,bi,…,bn) for the state of the work nodes is established. Specifically, firstly, the state characteristic parameters corresponding to each work node are indexed according to a unique identifier, and data alignment processing is performed with the positioning data and real-time sensing data of the corresponding node. The positioning data includes spatial coordinate information obtained based on BeiDou positioning or Global Positioning System, and the real-time sensing data includes real-time monitoring values ​​of flow rate, pressure, temperature, and liquid level. During the alignment process, a unified timestamp sequence is used as a benchmark, and data from different sources are resampled in chronological order. When data corresponding to a certain timestamp is missing, a forward hold method is used for compensation to ensure the consistency of various types of data in the time dimension.

[0035] After time alignment, feature fusion processing is performed on the state feature parameters, positioning data, and real-time sensor data to construct a multi-source time-series data set. Specifically, the state feature parameters, spatial coordinates, and sensor values ​​of the same working node within a continuous time window are concatenated in chronological order to form a feature vector sequence. Weighting coefficients are introduced to weight different data sources, with the weight of the state feature parameters set to 0.5, the weight of the positioning data set to 0.2, and the weight of the real-time sensor data set to 0.3. A unified node state vector is generated through weighted summation, thus forming a multi-source time-series data set.

[0036] Further, a mapping relationship set B for the state of the work nodes is established based on the multi-source time-series data set. Specifically, for any work node i, its physical state vector and virtual state vector are constructed respectively. The physical state vector is composed of real-time acquired sensor data and positioning data, and the virtual state vector is composed of the simulation state parameters of the corresponding node in the PID topology data. Then, a state matching function is used to calculate the similarity between the two types of state vectors. The state matching function is defined as the weighted Euclidean distance of the difference between the corresponding dimensions of the two vectors. When the calculation result is less than a preset threshold ε, it is determined that the physical state and virtual state of the work node are consistent; otherwise, it is determined that they are inconsistent. The threshold ε is obtained statistically based on historical operation data, and its value is 1.5 times the average historical deviation of each node.

[0037] Finally, the correspondence between the physical state and virtual state of each job node is represented by a set, forming a mapping relationship set B=(b1,b2,…,bi,…,bn), where bi is used to characterize the correspondence between the physical state and virtual state of the i-th job node at the current moment and its consistency result, providing basic data support for subsequent job model construction and state consistency verification.

[0038] Consistency Constraint Module: Based on the mapping relationship set B, a digital twin operation model is constructed, which includes a physical execution layer and a virtual mapping layer, and a state consistency constraint function is introduced to characterize the synchronization deviation between the two layers.

[0039] In this embodiment, based on the aforementioned mapping relationship set B, the physical and virtual states of each operation node are hierarchically organized to construct a digital twin operation model. Specifically, firstly, each mapping unit bi in the mapping relationship set B is read. The mapping unit bi contains the node identifier, physical state characteristic parameters, virtual state characteristic parameters, and corresponding time information of the i-th job node. Then, using the node identifier as an index, all mapping units bi are arranged according to the job flow sequence, and a physical execution layer is constructed based on the execution attributes of the job nodes in the actual production process. Each job node in the physical execution layer corresponds to the current state of the actual equipment or the actual job object. The current state includes at least the running state, position state, process parameter state, and timing state. After constructing the physical execution layer, a virtual mapping layer is further constructed based on the mapping results of the same node identifier in the virtual space. Each job node in the virtual mapping layer corresponds to the virtual expression state of the same node identifier in the physical execution layer. The virtual expression state includes at least the virtual position, virtual action, virtual process parameters, and virtual timing parameters. Then, using the same node identifier as the association condition, a node-to-node mapping structure is established between the physical execution layer and the virtual mapping layer, thereby forming the initial structure of the digital twin job model containing a set of nodes, a set of connection relationships, and a set of inter-layer mapping relationships. The node set consists of all work nodes, the connection relationship set is constructed based on the sequential execution relationship and material flow relationship in the work process, and the inter-layer mapping relationship set is directly generated from the mapping relationship set B.

[0040] After completing the initial structure construction of the digital twin operation model, the physical state vectors in the physical execution layer and the virtual state vectors in the virtual mapping layer are processed to unify their dimensions, forming standard state vector pairs that can be compared item by item. Specifically, for each operation node, the physical state vector Pi in the physical execution layer and the virtual state vector Vi in the virtual mapping layer are extracted, where the physical state vector Pi = [pi1, pi2, ..., pim] and the virtual state vector Vi = [vi1, vi2, ..., vim], and m represents the number of uniformly selected state feature dimensions. The number of state feature dimensions is determined according to the state feature items common to all operation nodes, including at least position coordinates, opening value, flow rate value, pressure value, liquid level value, and action timing value. For state features with different dimensions, an interval normalization method is used for unified processing and normalization. The k-th dimension state value is defined as the current value minus the minimum reference value of that dimension, divided by the difference between the maximum and minimum reference values ​​of that dimension. The maximum and minimum reference values ​​are obtained from 30 consecutive days of stable operation data. If a certain dimension state feature is discrete, the discrete state is mapped to a preset integer value, where the on state is mapped to 1, the off state is mapped to 0, and the fault state is mapped to -1. After normalization of all dimensions, the standard physical state vector P′i and the standard virtual state vector V′i corresponding to the i-th operation node are obtained, which are used as input data for the subsequent state consistency constraint function.

[0041] After obtaining the standard physical state vector P′i and the standard virtual state vector V′i, a state consistency constraint function is constructed to characterize the synchronization deviation between the physical execution layer and the virtual mapping layer. Specifically, for the i-th job node, the normalized difference of its state features in each dimension is calculated, and weight coefficients are set according to the degree of influence of different state features on the actual operation safety and scheduling accuracy. The weight coefficients for position coordinates are 0.20, opening values ​​are 0.20, flow rates are 0.15, pressure values ​​are 0.15, liquid levels are 0.20, and action timing values ​​are 0.10. The sum of all weight coefficients is 1. Then, a state consistency constraint function Ci is constructed for the i-th job node. The state consistency constraint function Ci is defined as the sum of the products of the absolute values ​​of the normalized differences of each state feature and their corresponding weight coefficients. Where wk represents the weight coefficient of the k-th dimension state feature, p′ik represents the k-th dimension value of the standard physical state vector P′i, and v′ik represents the k-th dimension value of the standard virtual state vector V′i; thus, the smaller the value of Ci, the more consistent the physical state and virtual state of the i-th operation node are, and the larger the value of Ci, the more obvious the synchronization deviation between the two is; after the state consistency constraint function Ci of all operation nodes is calculated, each Ci is combined according to the operation process order to obtain the set of state consistency constraint functions C=[C1, C2, ..., Cn] of the entire digital twin operation model.

[0042] After constructing the set of state consistency constraint functions C, the state consistency constraint functions are embedded in the digital twin operation model to form a complete digital twin operation model that includes the physical execution layer, the virtual mapping layer, and the inter-layer deviation constraint relationship. Specifically, the state consistency constraint function values ​​from 30 consecutive days of stable operational data are first used as sample data. The mean μi and standard deviation σi of the state consistency constraint function for each operational node are calculated, and the anomaly judgment threshold Ti is defined as μi+2σi. Then, the state consistency constraint function Ci of the i-th operational node is compared with the corresponding anomaly judgment threshold Ti. When Ci≤Ti, the i-th operational node is determined to be in a synchronized state; when Ci>Ti, the i-th operational node is determined to be in a synchronized deviation state. Based on this, the synchronized state or synchronized deviation state is written into the inter-layer mapping relationship of the corresponding node, so that the inter-layer mapping relationship of each node simultaneously includes the node identifier, physical state vector, virtual state vector, state consistency constraint function value, and consistency judgment result. Finally, according to the sequential dependency relationship in the operation process, the inter-layer mapping relationships of all nodes are connected in series to form a complete digital twin operation model that reflects the synchronization degree between the actual production operation process and the virtual mapping process. This provides a unified data and constraint basis for subsequent dynamic operation load application, process drive signal input, and anomaly state judgment.

[0043] Execution control module: Based on actual work instructions and scheduling strategies, it applies dynamic work load and process drive signals to the digital twin work model to obtain the target work execution model.

[0044] In this embodiment, dynamic workload and process-driven signals are applied to the digital twin operation model based on actual operation instructions and scheduling strategies to obtain the target operation execution model. First, actual operation instructions are acquired and parsed and prioritized according to a preset scheduling strategy to generate an operation instruction sequence and corresponding time constraint parameters. Specifically, the operation type, operation object, start position, target position, planned start time, planned end time, material category, and safety control requirements in the current production operation are collected and written into a unified format actual operation instruction table. Then, each actual operation instruction is parsed to extract the instruction number, node path, execution conditions, and completion conditions. Based on this, a priority scoring function is constructed, defined as a weighted sum of safety weight value, timeliness weight value, resource conflict weight value, and process dependency weight value. The safety weight value is divided into three levels: 5, 4, and 3, according to safety level. The timeliness weight value is calculated as the reciprocal of the time difference between the current time and the planned end time. The resource conflict weight value is based on the number of instructions competing for the same equipment at the same time. The calculations are as follows: the process dependency weight value is calculated based on the number of incomplete preceding nodes; the weight coefficients for the safety weight value, timeliness weight value, resource conflict weight value, and process dependency weight value are set to 0.40, 0.30, 0.20, and 0.10, respectively. Each actual work instruction is sorted from highest to lowest priority score to form a work instruction sequence. Then, based on the planned start time, planned end time, and standard operation duration, the time constraint parameters for each actual work instruction are calculated. These time constraint parameters include the earliest allowed start time, the latest allowed end time, and the allowed execution duration. The standard operation duration is determined by the median execution duration of similar operations over the past 30 days.

[0045] After obtaining the job instruction sequence and time constraint parameters, each job instruction is mapped to a corresponding job node in the digital twin job model based on the job instruction sequence and time constraint parameters, and dynamic job load parameters are assigned to each job node. Specifically, firstly, based on the node path and node identifier in the job instruction, the corresponding physical execution layer node and virtual mapping layer node are found in the digital twin job model, and a unique association is established between each actual job instruction and the corresponding job node; when an actual job instruction corresponds to multiple consecutive job nodes, the instruction decomposition result is established according to the order of the node paths to obtain a set of node-level job instructions; subsequently, dynamic job load parameters are assigned to each node-level job instruction. The dynamic job load parameters include job intensity, execution duration, and resource occupancy ratio, where job intensity is defined as the ratio of the node's processing volume to the node's rated processing volume per unit time, and execution duration is defined as the ratio of the node's processing volume to the node's rated processing volume per unit time. The planned execution time of the work instruction is defined as the ratio of the equipment capacity occupied by the node in the current operation to the total equipment capacity. The node processing capacity is determined based on the planned transport volume, planned loading and unloading volume, or planned inspection frequency. The rated processing capacity of the node is determined based on the equipment nameplate parameters and the average stable operation value over the past 30 days. When the work intensity is greater than 1, it is limited to 1. When the resource occupancy ratio is greater than 0.85, the node is determined to be in a high-load state, and this state is written into the node status data. Finally, the dynamic work load parameters are loaded onto the corresponding work nodes, so that each work node in the digital twin work model has a load representation corresponding to the current actual work.

[0046] After completing the dynamic workload parameter allocation, process-driven signals are constructed based on the connection relationships between work nodes and the work flow sequence, and then applied sequentially to each work node according to time constraint parameters. Specifically, firstly, based on the connection relationship set in the digital twin work model, the predecessor and successor nodes of each work node are extracted, and a node-driven sequence table is generated by combining the node-level work instruction set; subsequently, a process-driven signal is constructed for each work node, which consists of a start flag, a duration flag, a switch flag, and a completion flag. The start flag indicates that the node enters the execution state, the duration flag indicates that the node maintains the execution state, the switch flag indicates that the current node passes execution authority to the successor node, and the completion flag indicates that the current node has met the completion conditions. The completion conditions are: the valve reaches the target opening degree, the pump reaches the target operating state, and the flow rate reaches the planned level. The corresponding conditions for the value range and the liquid level reaching the target liquid level range are determined; the allowable deviation of the target opening degree is set to ±2%, the allowable deviation of the planned flow rate is set to ±3%, and the allowable deviation of the target liquid level is set to ±1.5%; based on this, the earliest allowed start time in the time constraint parameters is used as the trigger time of the start flag, the allowed execution duration is used as the maintenance duration of the continuity flag, the time when the completion condition is met is used as the generation time of the switching flag, and the latest allowed end time is used as the final determination time of the completion flag; when a certain preceding node has not generated a completion flag, the subsequent nodes connected to that preceding node must not receive the start flag, thus forming a process drive signal sequence with sequential dependencies.

[0047] After the process-driven signals are applied to each work node in sequence, the digital twin work model is updated in state and advanced in sequence under the combined effect of dynamic work load parameters and process-driven signals, and the target work execution model is generated. Specifically, the physical execution layer state and virtual mapping layer state of each work node are synchronously updated with an evolution step of 1 second. For work nodes under the start flag, their running state is updated to the execution state, and their dynamic work load parameters are written into the current state vector. For work nodes under the continuous flag, the flow rate, pressure, liquid level, and position status are recursively calculated according to the work intensity. The current state value is equal to the previous state value plus the product of the work intensity and the state change benchmark, which is determined by the average change rate of similar work over the past 30 days. For work nodes under the switching flag, the completion status of the current node is written into the pre-completion conditions of its subsequent nodes to activate the start judgment of the subsequent nodes. After all nodes have completed the state update, the deviation between the actual execution time and the allowed execution time of each work node is calculated. When the deviation is greater than 10% of the allowed execution time, the node is marked as a time-offset node. Subsequently, the state vectors, process drive signals, dynamic work load parameters, and time-offset marks of all nodes at each evolution moment are combined in chronological order to form the target work execution model. The target job execution model is used to characterize the execution process of actual job instructions under the current scheduling strategy, the node state evolution process, and the job process advancement process, and provides basic data for subsequent state deviation identification and abnormal state determination.

[0048] Deviation analysis module: It uses the multi-source time series data set to perform real-time iterative calculations on the target job execution model, identifies the state deviation values ​​of each job node, verifies them according to the state consistency constraint function, and generates abnormal state judgment results.

[0049] In this embodiment, a multi-source time-series data set is used to perform real-time iterative calculations on the target job execution model to identify the state deviation values ​​of each job node. First, the multi-source time-series data set is serialized according to a unified time step and time-aligned with the node state vectors at corresponding times in the target job execution model to form a node comparison data sequence. Specifically, the state feature parameters, positioning data, and real-time sensing data in the multi-source time-series data set are first categorized according to node identifiers, and a time series index is constructed using 1 second as the unified time step. For any job node, the physical observation data corresponding to that job node is extracted at each time step to form an actual observation vector Ai(t), where Ai(t) = [ai1(t), ai2(t), ..., aim(t)], and m represents the number of state feature dimensions. Simultaneously, the predicted state vector Yi(t) at the same job node and the same time step is extracted from the target job execution model, where Yi(t) = [yi1(t), yi2(t), ..., yim(t)]. Then, the actual... The observation vector Ai(t) and the predicted state vector Yi(t) are time-aligned. If the time deviation between the actual acquisition time and the current time step is no more than 0.5 seconds, the actual acquisition value is mapped to the current time step. If the time deviation is greater than 0.5 seconds, the actual acquisition value is mapped to the next time step. If an actual observation value is missing at a certain time step, the valid value of the previous time step is used for one hold compensation. After completing the data mapping of each time step, the actual observation vector Ai(t) and the predicted state vector Yi(t) of the same working node at consecutive time steps are combined in chronological order to obtain the node comparison data sequence Di={Ai(t), Yi(t)}, which is used for subsequent difference calculation.

[0050] After forming the node comparison data sequence, a difference calculation model between the predicted and actual observed node states is constructed based on this data sequence. Iterative calculations are performed on each work node within a continuous time window to obtain the initial value of the node state deviation. Specifically, for the i-th work node, a calculation window is set with five consecutive time steps. Within each calculation window, the difference between the predicted state vector Yi(t) and the actual observed vector Ai(t) in each dimension of the state features is calculated to obtain the single-time difference vector Ei(t), where... Based on this, a difference calculation model Fi is constructed, which is defined as a weighted accumulation function of the difference vectors at each time step within the window, i.e. Where L represents the calculation window length, with a value of 5. The difference weight coefficients represent the k-th dimension state features; The time step index variable is used to identify the current sampling time. The differential weighting coefficients are set according to the degree of influence of state characteristics on operational safety and continuity. Specifically, the differential weighting coefficient for position characteristics is 0.20, for opening characteristics it is 0.20, for flow characteristics it is 0.15, for pressure characteristics it is 0.15, for liquid level characteristics it is 0.20, and for action timing characteristics it is 0.10. The sum of all differential weighting coefficients is 1. Subsequently, the calculation proceeds step-by-step along the time axis using a sliding window. Each time step is advanced, the value of the differential calculation model Fi is recalculated to obtain the initial value Oi(t) of the node state deviation for the i-th operational node at each time step. The initial value Oi(t) of the node state deviation characterizes the original deviation between the predicted state and the actual observed state of the operational node within the current calculation window.

[0051] After obtaining the initial value of the node state deviation, the initial value is dynamically weighted and smoothed based on the historical operating data of each work node to obtain the node state deviation value. Specifically, the historical operating data of the corresponding work node under stable operating conditions for the past 30 days is retrieved first, and the historical deviation sequence is calculated according to the same difference calculation model; then, the mean μi and standard deviation σi of the historical deviation sequence are used as the deviation fluctuation reference for the i-th work node, and a dynamic weight function Wi(t) is constructed. The dynamic weight function Wi(t) is defined as the normalized reciprocal function of the difference between the current node state deviation initial value Oi(t) and the historical deviation mean μi, i.e. The value 0.001 is used to avoid a denominator of zero. Therefore, when the initial value of the current node state deviation is close to the historical deviation mean, the dynamic weight function Wi(t) takes a larger value; when the initial value of the current node state deviation deviates significantly from the historical deviation mean, the dynamic weight function Wi(t) takes a smaller value. After obtaining the dynamic weight function Wi(t), the initial values ​​of the node state deviation at the current time and the two previous time points are weighted and smoothed to obtain the node state deviation value Zi(t), which is calculated as follows: Where β1 is 0.50, β2 is 0.30, and β3 is 0.20, and the sum of β1, β2, and β3 is 1; through the above processing, the influence of occasional measurement fluctuations on the current deviation judgment is weakened, while persistent deviations are retained, thereby improving the ability of the node state deviation value Zi(t) to represent the actual abnormal state. Finally, the node state deviation values ​​Zi(t) of each work node at each time step are collected according to the node order and time order to form the deviation data basis required for subsequent verification.

[0052] After obtaining the state deviation values ​​of each work node, the results are verified according to the state consistency constraint function to generate an abnormal state determination result. First, a corresponding time-series signal sequence is constructed based on the state consistency constraint function values ​​of each work node, and a signal timing logic determination formula is established by combining preset work process constraints, state duration constraints, and dependencies between preceding and subsequent nodes. Specifically, the state consistency constraint function values ​​Ci(t) calculated for the i-th job node at continuous time steps are arranged in chronological order to form a time-series signal sequence Si={Ci(1), Ci(2), ..., Ci(T)}. Then, a preset job process constraint is constructed based on the actual job process. The preset job process constraint is used to limit the subsequent job node to enter the execution state only after the preceding job node has completed. A state duration constraint is constructed based on the standard job duration. The state duration constraint is used to limit the abnormal deviation state of a certain job node to be determined as a continuous abnormality only after it has continuously exceeded the preset duration. A node dependency constraint is constructed based on the node connection relationship. The node dependency constraint is used to limit the downstream node from entering the target execution state in advance when the upstream node has not met the completion condition. Afterwards, the above three types of constraints are written into the signal timing logic judgment formula. Specifically, if "the state consistency constraint function value Ci(t) exceeds the deviation limit value θi" is defined as the atomic proposition Pi(t), then "the work node is in an abnormal deviation state for a continuous time interval Δt" can be expressed as a persistence proposition under the action of the start-stop operator, and "the subsequent work node can only start after the previous work node is completed" can be expressed as a dependency proposition under the action of the finish-stop operator; where the deviation limit value θi is determined by 95% of the state consistency constraint function value in the stable work data of the past 30 days, and the duration Δt is taken as 3 consecutive time steps. In the above way, the work logic relationship is transformed into a computable signal timing logic judgment formula.

[0053] After establishing the signal timing logic determination formula, the timing signal sequence is input into the robust online monitoring algorithm to calculate the signal timing logic determination formula at each time point, thereby obtaining the timing robustness value of the corresponding work node. Specifically, the robust online monitoring algorithm is executed using a time-by-time recursive calculation method; for the atomic proposition Pi(t), the difference between the current state consistency constraint function value Ci(t) and the deviation limit value θi is first calculated to obtain the atomic proposition robustness. When ρi(t) is positive, it indicates that the security requirements corresponding to the atomic proposition are met at the current moment; when ρi(t) is negative, it indicates that the security requirements corresponding to the atomic proposition are violated at the current moment. For persistent propositions, the minimum robustness of the atomic proposition is taken as the robustness of the persistent proposition within the time interval corresponding to the preset duration Δt. For dependent propositions, the time difference between the completion time of the preceding task node and the start time of the subsequent task node is used as the basis for dependency verification. When the subsequent task node starts before the completion of the preceding task node, the negative value of this time difference is taken as the robustness of the dependent proposition. Based on the above calculations, the minimum value of the robustness of multiple propositions corresponding to each task node at each moment is taken as the temporal robustness value Ri(t) of the task node at that moment. The larger the temporal robustness value Ri(t), the more the current task node satisfies the preset task process constraints, state duration constraints, and dependencies between preceding and following nodes; the smaller the temporal robustness value Ri(t), the more the current task node deviates from the preset temporal requirements.

[0054] After obtaining the temporal robustness value of the corresponding job node, the system identifies whether the job node exhibits state mismatch, execution lag, or process out-of-order anomalies based on the comparison between the temporal robustness value and the preset anomaly judgment threshold. Specifically, based on the temporal robustness value samples of each job node under stable operating conditions over the past 30 days, the historical mean ηi and historical standard deviation δi of the i-th job node are calculated respectively, and the preset anomaly judgment threshold Γi is defined as... Subsequently, the current time-series robustness value Ri(t) is compared with the corresponding preset anomaly judgment threshold Γi. When Ri(t) ≥ Γi, the work node is determined to be in a normal state; when Ri(t) < Γi, the work node is determined to have an anomaly risk. Further, if the state consistency constraint function value continuously exceeds the deviation limit value θi, and the actual observed state deviates from the predicted state in at least two dimensions of position, opening degree, flow rate, pressure, or liquid level, it is identified as a state mismatch anomaly. If the current work node satisfies the process dependency relationship, but the completion time is later than the latest allowed end time by more than two time steps, it is identified as an execution lag anomaly. If a subsequent work node enters the execution state before the preceding work node meets the completion conditions, it is identified as a process out-of-order anomaly. Through the above classification and judgment method, the anomaly results can be correlated with specific anomaly types, which facilitates subsequent correction and processing.

[0055] After completing the anomaly identification of each work node, the anomaly identification results of each work node are associated and summarized according to the work process sequence to generate an anomaly status judgment result. Specifically, firstly, the anomaly identification results of all work nodes are arranged according to the work process sequence in the target job execution model to form an anomaly result sequence. Then, the identification results of the same anomaly type occurring consecutively within adjacent time steps are merged to form anomaly event segments. Each anomaly event segment includes at least a node identifier, anomaly type, anomaly start time, anomaly end time, corresponding node state deviation value, and corresponding temporal robustness value. After forming the anomaly event segments, the anomaly event segments with direct process associations are chained together according to the dependencies between preceding and following nodes to obtain an anomaly propagation chain. If the initial anomaly event segment in an anomaly propagation chain is a state mismatch anomaly, and its subsequent anomaly event segments are execution lag anomalies or process out-of-order anomalies, then the anomaly propagation chain is marked as a critical anomaly chain. Finally, all anomaly event segments and anomaly propagation chains are collected in chronological order to form an anomaly state judgment result, which is used to characterize the location of anomaly occurrence, duration of anomaly, anomaly evolution path, and severity of anomaly in the current work process of the target job execution model.

[0056] Correction module: Based on the abnormal state determination result, execute correction strategies for the corresponding job nodes, including instruction delay, path replanning, or state rollback.

[0057] In this embodiment, after obtaining the abnormal state determination result, a correction strategy is executed on the corresponding job node according to the abnormal state determination result to reduce the continuous impact of the abnormal state on the current production operation process and restore the target job execution model to an executable state that satisfies the job flow constraints and state consistency constraints. The correction strategy includes instruction delay, path replanning, and state rollback. Specifically, the abnormal node identifier, abnormal type, abnormal start time, abnormal duration, node state deviation value, temporal robustness value, and abnormal propagation chain information are first read from the abnormal state judgment results, and an abnormal handling input dataset is constructed based on this. Then, the target correction method for the corresponding operation node is determined according to the abnormal type and the abnormal severity. The abnormal severity is determined jointly by the node state deviation value and the temporal robustness value. Specifically: when the node state deviation value is higher than 1.5 times the historical average deviation value of the corresponding node, and the temporal robustness value is lower than 15% of the preset abnormal judgment threshold of the corresponding node, the abnormality is judged as a high-severity abnormality; when the node state deviation value is higher than the historical average deviation value of the corresponding node but not more than 1.5 times, and the temporal robustness value is lower than the preset abnormal judgment threshold of the corresponding node but not lower than 15%, the abnormality is judged as a medium-severity abnormality; when the node state deviation value only exceeds the historical average deviation value for a short time, and the temporal robustness value fluctuates around the preset abnormal judgment threshold, the abnormality is judged as a low-severity abnormality. Based on the above severity classification, instruction delay is prioritized for low-severity anomalies, path replanning is prioritized for medium-severity anomalies, and state rollback is prioritized for high-severity anomalies; if the anomaly propagation chain length is greater than 2 job nodes, the target correction method is directly upgraded by one level.

[0058] After determining the target correction method, if the execution instruction of the corresponding job node is delayed, the timing of the job instructions for the current node and its subsequent nodes is readjusted. Specifically, firstly, the job instruction, planned start time, planned end time, and latest allowed end time corresponding to the abnormal node are extracted, and the delay compensation amount is calculated. The delay compensation amount is defined as the sum of the duration of the abnormality and the estimated recovery time of the node status. The estimated recovery time of the node status is calculated using the average recovery time of similar abnormalities in the past 30 days. If the number of similar abnormality samples in the past 30 days is less than 10, the estimated recovery time of the node status is taken as 1.2 times the duration of the current abnormality. Subsequently, the delay compensation amount is added to the planned start time and planned end time of the current node's job instruction to obtain the delayed execution time interval. Based on the node dependency relationship in the target job execution model, the delayed execution time interval is sequentially passed to subsequent nodes, thereby updating the planned start time and planned end time of subsequent nodes. Subsequently, a feasibility check is performed on the updated execution time interval. This check includes two aspects: first, the updated plan's end time must not exceed the latest allowed end time of the corresponding node; second, the updated plan's start time must not be earlier than the completion time of the preceding node. When both conditions are met, the instruction delay correction is deemed valid, and the corrected time parameters are written into the target job execution model. If either condition is not met, the instruction delay is stopped, and path replanning is initiated. Through this process, timing conflicts caused by short-term anomalies can be eliminated without altering the job path.

[0059] If the corresponding job node performs path replanning, an alternative path search is performed on the job path containing the abnormal node to avoid the current abnormal node or abnormal connection path. Specifically, firstly, a path search graph is constructed based on the set of connection relationships in the digital twin job model, where each job node is a vertex in the graph and each executable connection relationship is an edge in the graph; for each edge, the edge weight is defined as the sum of the passage cost, job duration cost, and risk cost, where the passage cost is determined based on the material transfer distance or process jump distance between nodes, the job duration cost is determined based on the standard execution time corresponding to the edge, and the risk cost is determined based on the historical abnormality frequency of the node associated with the edge; the risk cost is specifically taken as the ratio of the number of abnormal occurrences of the corresponding node in the past 30 days to the total number of executions in the past 30 days. Subsequently, the edge weights corresponding to the abnormal node and abnormal connection path are increased to 3 times the original value. If the abnormality type is a process out-of-order abnormality, the connection edge is directly marked as an unselectable edge; on this basis, starting from the preceding normal node of the abnormal node and ending at the current job target node, an alternative path is re-searched using the minimum cumulative cost search method. The minimum cumulative cost search method is as follows: starting from the starting node, the cumulative edge weights from the starting node to adjacent nodes are calculated step by step, and at each step, the unprocessed node with the smallest cumulative edge weight is selected to continue the search until the target node is reached; finally, an alternative path with the smallest cumulative edge weight is obtained. After completing the alternative path search, the alternative path is further constrained and verified. The constraint verification includes that the resource occupancy ratio of each node in the path must not exceed 0.85, the state consistency constraint function value of each node in the path must not exceed the corresponding deviation limit value, and the total execution time of the alternative path must not exceed 1.2 times the total execution time of the original path. When the alternative path meets the above constraint verification, the alternative path is written into the target job execution model to replace the original abnormal path; when the alternative path does not meet the above constraint verification, the path replanning is stopped, and state rollback processing is initiated. Through the above processing, local abnormal nodes or abnormal connection paths can be bypassed while keeping the job objective unchanged, thereby maintaining job continuity.

[0060] If the execution status of the corresponding job node is rolled back, the execution status of the abnormal node and its associated nodes will be restored to the stable state corresponding to the preset rollback time to eliminate the impact of the abnormal propagation on subsequent operations. Specifically, firstly, the preset rollback time is determined. The preset rollback time is defined as the most recent time before the start of the abnormality, where the value of the corresponding node's state consistency constraint function does not exceed the deviation limit for three consecutive time steps. If no time meeting the above conditions exists, the preset rollback time is defined as the start time of the current job instruction. Subsequently, a node state snapshot corresponding to the preset rollback time is extracted from the target job execution model. The node state snapshot includes at least the running status, position status, opening value, flow rate value, pressure value, liquid level value, and action timing value of the abnormal node and its upstream and downstream nodes at the preset rollback time. Then, the state values ​​of the abnormal node and its associated nodes at the current time are replaced with the corresponding state values ​​in the node state snapshot, thereby completing the state restoration. After state recovery is completed, the recovery results need to be verified for consistency. This verification includes: recalculating the state consistency constraint function value of the recovered node, recalculating the temporal robustness value of the recovered node, and determining whether the recovered node satisfies the dependency constraints between preceding and following nodes and the state duration constraints. When the state consistency constraint function value of the recovered node does not exceed the corresponding deviation limit and the temporal robustness value is not lower than the corresponding preset anomaly judgment threshold, the state rollback is considered successful, and this recovered state is used as the starting point for new job execution. If the above conditions are not met after recovery, the search continues to move forward to the last stable moment that met the conditions, and the node state snapshot extraction and state replacement operations are repeated until a recovered state that meets the conditions is obtained. Through the above processing, abnormal nodes can be restored from an uncontrollable state to a verified stable state, thereby blocking the further spread of the anomaly propagation chain.

[0061] After completing instruction delay, path replanning, or state rollback, the correction results are uniformly verified to generate the corrected job execution results. Specifically, firstly, the state deviation and timing robustness are recalculated on the corrected target job execution model. Then, the corrected node state deviation value is compared with the corresponding historical average deviation value, and the corrected timing robustness value is compared with the corresponding preset anomaly judgment threshold. If the corrected node state deviation value is not higher than 1.1 times the historical average deviation value, and the corrected timing robustness value is not lower than the preset anomaly judgment threshold, the correction is deemed valid. If the corrected node state deviation value is still higher than 1.1 times the historical average deviation value, or the corrected timing robustness value is still lower than the preset anomaly judgment threshold, the correction is deemed invalid, and the next target correction method is executed in the order of instruction delay, path replanning, and state rollback. Finally, the verified and valid correction results are written into the target job execution model to form the corrected job execution results, which are used for the continuous advancement of subsequent production operations. The above processing enables the abnormal state determination results to be directly converted into executable corrective actions, and through a quantitative verification process, it ensures that the corrected job execution process once again meets the job flow constraints, state consistency constraints, and timing safety requirements.

[0062] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. An integrated production operation platform based on digital twins, characterized in that: include: Data acquisition module: Acquires initial process flow data and equipment status data of each work unit in the target production operation system, constructs corresponding PID structured models, and extracts status feature parameters of each node; State mapping module: Based on the state feature parameters, it fuses positioning data and real-time sensing data to generate a multi-source time series data set, and establishes a mapping relationship set B=(b1,b2,…,bi,…,bn) of the state of the work node, where bi represents the correspondence between the physical state and the virtual state of the i-th work node. Consistency constraint module: Based on the mapping relationship set B, a digital twin operation model is constructed, wherein the model includes a physical execution layer and a virtual mapping layer, and a state consistency constraint function is introduced to characterize the synchronization deviation between the two layers; Execution control module: Based on actual work instructions and scheduling strategies, it applies dynamic work load and process drive signals to the digital twin work model to obtain the target work execution model; Deviation analysis module: Utilizes the multi-source time-series data set to perform real-time iterative calculations on the target job execution model, identifies the state deviation values ​​of each job node, verifies them according to the state consistency constraint function, and generates abnormal state judgment results, including the following steps: Based on the state consistency constraint function value of each work node, a corresponding time sequence signal sequence is constructed, and combined with the preset work process constraints, state duration constraints and previous and next node dependency constraints, a signal timing logic judgment formula is established. The time-series signal sequence is input into the robust online monitoring algorithm, and the signal timing logic judgment formula at each time point is calculated to obtain the timing robustness value of the corresponding work node. Based on the comparison between the time robustness value and the preset anomaly judgment threshold, identify whether the work node has state mismatch, execution delay or process out-of-order anomaly. The anomaly identification results of each work node are correlated and summarized according to the work process sequence to generate anomaly status judgment results; For low-severity anomalies, prioritize instruction delay; for medium-severity anomalies, prioritize path replanning; and for high-severity anomalies, prioritize state rollback. Correction module: Based on the abnormal state determination result, execute correction strategies for the corresponding job nodes, including instruction delay, path replanning, or state rollback.

2. The integrated production operation platform based on digital twins according to claim 1, characterized in that: The extraction of state feature parameters for each node includes the following steps: performing structured parsing of the process flow diagram in the target production operation system, extracting pipeline connection relationships, equipment types, and process parameters, and generating PID topology data; based on the PID topology data, uniquely identifying and encoding each operation node, and establishing an association mapping relationship between node and equipment status data; acquiring the operating status data of each device, and performing timestamp alignment and data cleaning processing on the operating status data to form a unified format status dataset; and extracting state feature parameters representing the operating status of nodes from the status dataset according to preset feature extraction rules.

3. The integrated production operation platform based on digital twins according to claim 1, characterized in that: Based on the mapping relationship set B, a digital twin operation model is constructed, including the following steps: Based on the correspondence between the physical and virtual states of each operation node in the mapping relationship set B, a node state set for the physical execution layer and a simulation state set for the virtual mapping layer are constructed respectively, and a node-to-node mapping structure between the two layers is established according to the PID topology data; based on the node-to-node mapping structure, the state vectors of each corresponding node are uniformly normalized to form comparable standard state vector pairs; a state consistency constraint function is constructed based on the standard state vector pairs; the state consistency constraint function is embedded in the digital twin operation model to form a digital twin operation model that simultaneously includes the physical execution layer, the virtual mapping layer, and the state consistency constraint relationship.

4. The integrated production operation platform based on digital twins according to claim 1, characterized in that: Applying dynamic workload and process-driven signals to the digital twin operation model includes the following steps: acquiring actual operation instructions, parsing and prioritizing the operation instructions according to a preset scheduling strategy, generating an operation instruction sequence and corresponding time constraint parameters; mapping each operation instruction to a corresponding operation node in the digital twin operation model based on the operation instruction sequence and time constraint parameters, allocating dynamic workload parameters to each operation node, and constructing process-driven signals according to the connection relationship between operation nodes and the order of operation processes.

5. The integrated production operation platform based on digital twins according to claim 4, characterized in that: The dynamic job load parameters include job intensity, execution duration, and resource utilization ratio.

6. The integrated production operation platform based on digital twins according to claim 4, characterized in that: The process-driven signals are applied sequentially to each work node according to the time constraint parameters to drive the state evolution of the digital twin work model; under the combined effect of dynamic work load and process-driven signals, the state of the digital twin work model is updated and the time sequence is advanced to generate the target work execution model.

7. The integrated production operation platform based on digital twins according to claim 1, characterized in that: The multi-source time series data set is used to perform real-time iterative calculations on the target job execution model to identify the state deviation values ​​of each job node. The steps include: serializing the multi-source time series data set according to a unified time step and aligning it with the node state vector at the corresponding time in the target job execution model to form a node comparison data sequence. Based on the node comparison data sequence, a difference calculation model is constructed between the predicted value and the actual observed value of the node state. The model is iteratively calculated for each work node within a continuous time window to obtain the initial value of the node state deviation. The initial value of the node state deviation is dynamically weighted and smoothed according to the historical operation data of each work node to obtain the node state deviation value.

Images