Intelligent factory management method and system based on industrial internet

Abstract

The invention provides an intelligent factory management method and system based on the industrial internet, and the method comprises the steps: obtaining original data of a multi-source heterogeneous device, and generating a semantic event structure through preprocessing; the industrial event sequence is mapped into nodes, candidate edges are generated in combination with the physical topology connection relation of the equipment, and a target causal map is constructed; according to the target causal atlas, calculating a node risk score of each event node, and performing balance adjustment on the score by using a graph-level centrality value to generate a normalized node risk score of each event node; mapping the node risk scores to devices, and calculating a risk average value of each device; and converting a scoring matrix of the current task to all candidate devices into a scheduling strategy value, executing a scheduling strategy and collecting task execution feedback, and dynamically adjusting a target causal atlas and a node risk score according to the difference between an actual result and a predicted value to realize dual optimization of the target causal atlas and the scheduling strategy.

Description

Technical Field

[0001] The present invention belongs to the field of industrial Internet, and particularly relates to an intelligent factory management method and system based on the industrial Internet. Background Art

[0002] With the development of the new generation of information technology, the industrial Internet is becoming an important supporting foundation for the intelligent transformation of the manufacturing industry. By integrating technologies such as perception, communication, computing, and control, the industrial Internet builds the interconnection between multiple devices, production lines, personnel, and information systems in the factory, promoting the leap from "automation" to "intelligence". In this context, as the core platform for realizing efficient coordination in the manufacturing process, optimal allocation of resources, and refined control, the technical system of the intelligent factory management system is gradually evolving from the previous static rule-driven to data-driven and algorithm-driven. However, there are still many limitations in the current mainstream intelligent factory management methods, making them face key problems such as efficiency bottlenecks and response delays in practical applications.

[0003] First of all, when dealing with complex situations such as equipment anomalies, production interruptions, or environmental disturbances, the existing factory management systems still rely heavily on rule-based static response mechanisms or single-point prediction models. These methods may be effective in the face of typical abnormal events, but in scenarios with strong multi-device coupling, heterogeneous system collaboration, and non-linear fluctuations, their strain capacity is significantly insufficient. Especially in complex production lines, an equipment anomaly often triggers a series of chain reactions. The existing models lack the ability to model the causal relationships between events, resulting in the inability to accurately identify the abnormal conduction path and key impact nodes, thereby increasing the uncertainty of management intervention and response delay.

[0004] Secondly, in terms of the resource scheduling mechanism, the current scheduling systems mostly construct the task execution order based on heuristic rules, priority sorting, or central control strategies. Although this method has a certain degree of generality and real-time performance, it often ignores the risk factors and potential interference variables behind task execution. For example, when allocating tasks between high-energy-consuming devices and low-energy-consuming devices, if the current health status and risk level of the devices are not fully perceived, it is easy to cause equipment overload, frequent anomalies, and even shutdown accidents. In addition, the current scheduling systems usually cannot automatically adjust the scheduling strategy according to the real-time changing environment and state on the production site, lacking the closed-loop ability of "perception - judgment - execution - feedback", with rigid strategies and being difficult to adapt to the highly dynamic manufacturing environment.

[0005] Again, in terms of information system integration, the data relied on by intelligent factories comes from a large number of heterogeneous devices and systems, such as PLC controllers, industrial cameras, sensor networks, MES systems, etc. These data have characteristics such as diverse structures, inconsistent time, and uneven quality. Traditional data fusion mechanisms often cannot process these complex data streams in real time, resulting in a one-sided and unstable understanding of the overall factory state by the management system. In addition, the lack of a unified data-driven modeling framework also makes it difficult for intelligent decision-making systems to be scalable and interpretable.

[0006] Therefore, although the current industrial Internet and intelligent factory systems have made progress in underlying communication, edge collection, and partial prediction capabilities, there are still significant shortcomings in the abnormal response ability under multi-factor coupling, the adaptability of scheduling strategies, and the intelligent closed-loop of data-driven decision-making. Facing the current management requirements of the manufacturing industry for highly coordinated, real-time response, and dynamic optimization, there is an urgent need for a management method and system that can unify the perception, analysis, and decision-making processes and have the ability of cross-entity and cross-system intelligent collaboration to support the daily operation and emergency response of intelligent factories in a more efficient and reliable manner. Summary of the Invention

[0007] To overcome the above technical problems, the present invention proposes an intelligent factory management method and system for an industrial Internet environment, aiming to build an intelligent management mechanism with real-time abnormal perception, risk-driven scheduling decision-making, and multi-dimensional resource collaboration capabilities. Compared with existing methods, the present invention has carried out systematic reconstruction in key links such as event modeling, decision logic, policy execution, and feedback optimization.

[0008] In the first aspect, an embodiment of the present invention provides an intelligent factory management method based on the industrial Internet, and the method includes:

[0009] Obtain the original data of multi-source heterogeneous devices, generate a semantic event structure through preprocessing, construct an industrial event sequence, and screen out valid events through stability verification; wherein, the original data includes: original sampling values, device reporting timestamps, and data source device numbers; the preprocessing is to perform time alignment processing on the original data to generate an aligned timestamp, and perform event semantic encoding based on the aligned timestamp to obtain a semantic event structure; the semantic event structure includes: event type, aligned timestamp, and data source device number; the industrial event sequence includes several semantic event structures;

[0010] Map the industrial event sequence to nodes, generate candidate edges in combination with the physical topology connection relationship of the devices, and construct a target causal graph;

[0011] According to the target causal graph, calculate the node risk score of each event node, and use the graph-level centrality value to balance and adjust the score to generate the normalized node risk score of each event node;

[0012] Obtain the running requirements of the current task, map the node risk scores to devices, calculate the average risk of each device, and calculate based on the average risk of each device combined with the running requirements of the current task to obtain the scoring matrix of the current task for all candidate devices;

[0013] Convert the scoring matrix of the current task for all candidate devices into a scheduling policy for execution, execute the scheduling policy and collect task execution feedback, and dynamically adjust the target causal graph and node risk scores according to the difference between the actual result and the predicted value to achieve double optimization of the target causal graph and the scheduling policy.

[0014] Further, perform time alignment processing on the original data to generate aligned timestamps, and perform event semantic encoding based on the aligned timestamps to obtain a semantic event structure, specifically including:

[0015] Perform unified time axis alignment processing based on the device reported timestamps to obtain aligned timestamps;

[0016] Use the aligned timestamps and the original sampling values to generate events to generate a semantic event structure;

[0017] Among them, the screening of valid events through stability verification, that is, the screening of the semantic event structure, includes:

[0018] Obtain the original sampling values, calculate the change rate of the original sampling values, calculate the absolute value of the average acceleration of the original sampling values based on the change rate, and compare the calculated absolute value of the average acceleration with a preset threshold, and only retain the semantic event structure corresponding to the absolute value of the average acceleration less than the preset threshold.

[0019] Further, the unified time axis alignment processing based on the device reported timestamps to obtain aligned timestamps specifically includes:

[0020] Collect the timestamps reported by all devices;

[0021] Construct a drift optimization model with the goal of minimizing the time correction value, use each timestamp as input, and output the aligned timestamp to minimize the loss function.

[0022] Further, mapping the industrial event sequence to nodes, combining the device physical topology connection relationship to generate candidate edges, and constructing a target causal graph specifically includes:

[0023] Obtain the duration of the current time, estimate it by counting whether the same event occurs continuously within a certain number of seconds after it, and use it as the propagation credibility factor; based on the node set, when the aligned timestamp of the first node is less than the timestamp of the second node and there is a connection relationship between the data source device numbers of the first node and the second node in the physical topology, then it is used as the candidate edge between the first node and the second node;

[0024] Calculate the edge weight according to the candidate edge;

[0025] Retain incoming edges with the highest edge weights for each node to construct a complete graph.

[0026] Furthermore, the calculating the edge weight according to the candidate edge , is calculated as:

[0027]

[0028] Wherein, is the conditional probability that occurs after the event occurs; is the marginal probability of for normalization; is the normal propagation delay of device ; is the standard deviation of the allowable propagation window; is the duration of the event to determine whether it is a systematic and slowly developing anomaly; is the system maximum anomaly duration benchmark; is the anomaly suppression coefficient to control the impact of chronic events; is the structure dependence factor, indicating the communication credibility between devices, derived from the confidence of the process network structure; ; and respectively represent the aligned timestamps of devices and .

[0029] Furthermore, according to the target causal graph, calculate the node risk score of each event node, specifically including:

[0030] Combining the duration and the edge weight, through node risk assessment, determine the node risk score of each event node, calculated as:

[0031]

[0032] Wherein, is the node risk score, is the anomaly intensity weight, For the duration of an event and for the device control ability of event node where i and j represent event nodes i and j; E represents the set of nodes, represents the amplification factor control coefficient, and is the dangerous path coefficient of event node

[0033] Furthermore, obtaining the running requirements of the current task, mapping the node risk scores to devices, calculating the risk average value of each device, and calculating based on the risk average value of each device in combination with the running requirements of the current task to obtain the scoring matrix of the current task for all candidate devices, specifically including:

[0034] Summing and averaging the risk assessment values of all devices based on the node risk scores to obtain the risk average value of each device;

[0035] According to the risk average value, comprehensively considering the risk situation of resources, the characteristics of resources themselves, and the urgency of tasks, generating the device scheduling score corresponding to the task; among them, the task is most urgent and has the highest score; the device is most flexible and has the best match;

[0036] Aggregating the device scheduling scores of the current task to obtain the scoring matrix of the current task for all candidate devices.

[0037] In a second aspect, an embodiment of the present invention provides an intelligent factory management system based on industrial Internet, and the system includes:

[0038] A data processing module: used to obtain the raw data of multi-source heterogeneous devices, generate a semantic event structure through preprocessing, construct an industrial event sequence, and screen out valid events through stability verification; wherein, the raw data includes: raw sampling values, device reporting timestamps, and data source device numbers; the preprocessing is to perform time alignment processing on the raw data to generate an aligned timestamp, and perform event semantic encoding based on the aligned timestamp to obtain a semantic event structure; the semantic event structure includes: event type, aligned timestamp, and data source device number; the industrial event sequence includes several semantic event structures;

[0039] A causal modeling module: used to map the industrial event sequence to nodes, generate candidate edges in combination with the device physical topology connection relationship, and construct a target causal graph;

[0040] A risk assessment module: used to calculate the node risk score of each event node according to the target causal graph, and use the graph-level centrality value to balance and adjust the score to generate the normalized node risk score of each event node;

[0041] The scheduling policy module: It is used to obtain the running requirements of the current task, map the node risk scores to devices, calculate the average risk of each device, and calculate based on the average risk of each device combined with the running requirements of the current task to obtain the scoring matrix of the current task for all candidate devices;

[0042] The feedback optimization module: It is used to convert the scoring matrix of the current task for all candidate devices into a scheduling policy for execution, execute the scheduling policy and collect task execution feedback, and dynamically adjust the target causal graph and node risk scores according to the difference between the actual result and the predicted value, so as to realize the dual optimization of the target causal graph and the scheduling policy.

[0043] The beneficial technical effects of the present invention are at least as follows:

[0044] By introducing an event correlation mechanism with the ability of causal relationship modeling, the system of the present invention can model the event linkage between multiple devices and multiple processes, so as to identify the propagation chain and key nodes of abnormal events, and realize the accurate perception and active response to complex production anomalies. At the same time, in order to enhance the intelligence and robustness of resource scheduling, the present invention proposes a strategy decision-making mechanism that integrates risk perception and multi-objective trade-off, so that the scheduling process is not only planned based on resource capabilities and task priorities, but also can real-time perceive the potential risks in the current system state, thereby dynamically adjusting the scheduling path to avoid system instability. In addition, the present invention constructs a closed-loop control system that supports feedback learning, which can continuously optimize model parameters and policy selection according to actual production feedback, realize the ability leap of the intelligent factory from "passive scheduling" to "active collaboration", and truly meet the intelligent management requirements of the industrial Internet in a complex and dynamic environment. Description of the Drawings

[0045] The present invention will be further described with reference to the drawings, but the embodiments in the drawings do not constitute any limitation to the present invention. For those of ordinary skill in the art, other drawings can be obtained according to the following drawings without creative efforts.

[0046] Figure 1 It is a flowchart of an intelligent factory management method based on the industrial Internet of the present invention. Detailed Embodiments

[0047] The embodiments of the present invention will be described in detail below. The examples of the embodiments are shown in the drawings, where the same or similar reference numerals represent the same or similar elements or elements with the same or similar functions from beginning to end. The embodiments described below by referring to the drawings are exemplary and are only used to explain the present invention and should not be construed as a limitation to the present invention.

[0048] In one embodiment, as Figure 1As shown, an intelligent factory management method based on the industrial Internet is provided, including the following steps:

[0049] S1. Obtain the original data of multi-source heterogeneous devices, generate a semantic event structure through preprocessing, construct an industrial event sequence, and screen out valid events through stability verification; wherein, the original data includes: original sampling values, device reporting timestamps, and data source device numbers; the preprocessing is to perform time alignment processing on the original data to generate an aligned timestamp, and perform event semantic encoding based on the aligned timestamp to obtain a semantic event structure; the semantic event structure includes: event type, aligned timestamp, and data source device number; the industrial event sequence includes several semantic event structures.

[0050] Specifically, the goal of this step is to collect on-site operation data from multi-source heterogeneous devices in the industrial Internet environment and uniformly convert it into an industrial event sequence with standardized structure, time synchronization, and clear semantics for subsequent causal graph modeling (S2). The entire processing flow is concatenated with time alignment as the main line, and semantic encoding and stability judgment as supporting processing modules. The three act on the same input data set and are processed in sequence.

[0051] The input data from the industrial site is in a unified structure:

[0052]

[0053] Among them, is the original sampling value, such as temperature, current, alarm code; is the device reporting timestamp; is the data source device number. These original data are discrete time series without synchronization and explicit semantic definition.

[0054] The first step: Time alignment and drift correction between devices (must be carried out first)

[0055] Due to the local clock differences or network delays of different devices, the timestamp error may reach the second level. If not corrected, it will lead to confusion in the causal relationship between events. Therefore, cross-device unified time axis alignment is carried out first.

[0056] Adopt a drift optimization model:

[0057]

[0058] Among them, is the time correction value of device ; is the position of the system reference time axis (discrete grid points with a set time step such as 1s); is a regularization coefficient that restricts excessive offsets to avoid misalignment. This alignment process is applied to all in to correct and generate a new timestamp for subsequent unified processing.

[0059] Step 2: Event semantic encoding (based on the aligned time series)

[0060] Use the aligned and to generate events. The semantic event generation method is as follows:

[0061] For continuous variables (such as temperature, current): Detect jumps or abnormal slopes;

[0062] For discrete code values (such as alarm codes): Parse through the code table; all events are uniformly encoded as . Generate the semantic event structure:

[0063] This structure is the node prototype for all subsequent graph construction processes (i.e., the nodes in the causal graph ).

[0064] Step 3: Event stability verification:

[0065] Since industrial signals may experience severe short-term jitters that cause false triggers, the present invention introduces a local stability function to determine whether the event is "stably generated", i.e., whether it can be used as an effective event node in the subsequent model. The judgment function is as follows:

[0066]

[0067] Where is the window length (e.g., 3 seconds); is the absolute value of the average acceleration; is the change rate of the signal within the past window; is the stability threshold, and the event is only accepted when the change is small. This function is used for retention determination after event generation to prevent "false trigger nodes" caused by jitters or electromagnetic interference from affecting subsequent causal graph reasoning. After obtaining

[0068] we get an event sequence that has passed alignment, semantic clarity, and stability screening, which serves as the basis for subsequent causal graph modeling.

[0069] S2. Map industrial event sequences to nodes, combine the physical topology connection relationships of devices to generate candidate edges, and construct the target causal graph.

[0070] Specifically, the goal of this step is based on the industrial event sequence output by S1 Construct a causal graph that reflects the true propagation path between abnormal events . This graph not only describes the temporal logical relationship between events, but also reflects their industrial semantic relevance, and considers key characteristics such as system structure topology and abnormal duration. Traditional causal modeling methods often use black-box algorithms (such as neural network-based structure learning or GNN inference), lacking the ability of industrial semantic interpretation and unable to meet the requirements of deployability and real-time performance.

[0071] At the same time, this step adopts a graph modeling method with transparent structure and causal interpretability, and introduces multiple innovations designed specifically for industrial scenarios (such as abnormal propagation time-delay window, prior of communication path between devices, event persistence suppression term, etc.) to construct a highly reliable graph structure for risk assessment and scheduling control.

[0072] This step realizes the construction of the graph structure through the following three stages:

[0073] ① Node initialization and event normalization: Map each piece of data in to a candidate node , forming a node set . Each node is bound with the following attributes: : Event type;

[0074] : Timestamp; : Data source device number; : The duration of this event, estimated by counting whether the same event occurs continuously within a certain number of seconds after it, as the propagation credibility factor.

[0075] ② Candidate edge generation and screening logic: For each pair of node pairs that meet , if and have a connection relationship in the physical topology (i.e., can communicate or there is a dependency in the upstream and downstream process chains), then they are used as candidate edges , and enter the edge weight calculation stage.

[0076] ③ Edge weight scoring function: The core candidate edge weight scoring function is as follows:

[0077]

[0078] Among them, is the conditional probability that occurs after the event occurs, counted from ; is The marginal probability for normalization; is the device of the normal propagation delay (which can be obtained from system physical transmission monitoring or process time); is the standard deviation of the allowable propagation window; is the event duration, used to determine whether it is a systematic and slowly developing anomaly (such as environmental temperature rise); is the system's maximum anomaly duration benchmark; is the anomaly suppression coefficient, controlling the impact of chronic events; is the structure dependence factor, indicating the communication credibility between devices, derived from the confidence of the process network structure (such as the topology trust score generated by the operation and maintenance system); and respectively represent the time stamps of the devices and after alignment.

[0079] Among them, is the frequency and time delay perception term; is the anomaly intensity suppression term;

[0080] It should be noted that the core innovation points of this formula include:

[0081] Introducing a time dynamic Gaussian term to strengthen the real-time conduction;

[0082] Adding an anomaly persistence suppression term to prevent chronic anomalies from overly dominating the causal path;

[0083] The structure dependence factor strengthens the deployability of the edges and industrial credibility, avoiding false dependencies brought by pure statistical co-occurrence.

[0084] To prevent the graph structure from being too dense, further retain the highest-weight incoming edges for each node:

[0085]

[0086] Finally, retain the edge set to construct a complete graph, is the effective causal edge set of the event ; represents an operation, that is, selecting the first K from the given elements. Obtain : the graph structure, where is the event node set, is the effective causal edge set, is the edge weight set; Each edge in has multi-dimensional features such as time, structure, and frequency, which are directly called by the risk scoring mechanism in S3 later.

[0087] S3. Calculate the node risk score for each event node according to the target causal graph, and use the graph-level centrality value to balance and adjust the score to generate the normalized node risk score for each event node.

[0088] Specifically, the goal of this step is to calculate the risk score for each event node based on the causal graph constructed in S2. Different from the traditional scoring method that only relies on local anomaly detection, the present invention proposes a graph-structure-aware risk scoring model that combines graph structure features, propagation paths, anomaly intensity, and factory control capabilities to more realistically reflect the "systemic risk" that an event may cause in the entire industrial system.

[0089] Among them, considering that anomalies in industrial scenarios often exhibit characteristics such as "chain diffusion", "failure of key bottleneck nodes", and "local anomalies are prone to trigger cascades", the risk scoring model designed in this step introduces the following structure modeling ideas unique to industrial scenarios in the scoring logic: based on the causal propagation intensity represented by the edge weights in S2 ; combined with event persistence and the adjustability of control nodes ; introduce "hazard path density" and "downstream amplification factor" in the graph topology as scoring gain terms; design a normalized scoring mechanism with a graph regularization term to prevent the amplification of risk judgment errors caused by structural deviations.

[0090] Input corresponding to the output of S2:

[0091] Graph structure , where: is the set of event nodes; is the set of edges, representing causal relationships; is the set of causal edge weights; each node has the following attributes: : The aligned timestamp in S1; : The data source device number; : The industrial event label; : The event duration, statistically obtained in S2; : The control adjustability factor, given by the device control level table, with a range , and the smaller the value, the more difficult it is to be regulated (such as old equipment).

[0092] Furthermore, the present invention defines the risk score of each event node as the following composite formula:

[0093]

[0094] The first term Represents the "abnormality intensity" of the node itself, taking into account its duration and the device control ability ; The longer it is, the more intense the abnormal development is; The smaller it is, the more difficult it is for the device to adjust quickly; overall, it reflects the "controllability risk" of the abnormality.

[0095] The second item is the "structural propagation risk", considering the influence intensity of all downstream nodes that the node may activate: is the edge weight of the causal edge, originating from S2; is the node 's "hazard path coefficient", defined as the mean value of all its outgoing edges i.e., its "downstream amplification ability"; is the amplification factor control coefficient, a system parameter.

[0096] Among them, the model has the following two innovative points: combining the abnormality ontology, propagation ability and downstream diffusion potential to comprehensively characterize the systemic risk of the node; introducing the device regulation ability as a constraint factor , preventing the misidentification of high-frequency adjustable points as high-risk sources by mistake.

[0097] Furthermore, in order to enhance the stability and interpretability of the model among different structures, the present invention further designs a graph structure-aware regularization normalization mechanism, using the graph-level centrality value to balance and adjust the scores, preventing the overestimation of the risks of some central nodes due to overly concentrated structures:

[0098]

[0099] Among them, is the PageRank centrality score of the node , reflecting its "degree of propagation center" in the graph; is the adjustment factor, used to control the influence of the regularization term on the final risk value; this operation can suppress the structural scoring deviation caused by overly concentrated topology, and is a very crucial stability optimization strategy in industrial deployment.

[0100] S4. Obtain the running requirements of the current task, map the node risk scores to the devices, calculate the average risk of each device, and calculate based on the average risk of each device in combination with the running requirements of the current task to obtain the scoring matrix of the current task for all candidate devices.

[0101] Specifically, the goal of this step is based on the risk scores generated by S3 , construct a resource scheduling scoring function integrating risk awareness to guide the generation of task resource matching strategies. The core of the scheduling strategy is to construct a scoring function between tasks and allocable resources , which needs to comprehensively consider: the risk level of resource entities (given by ); the adjustable ability of resources (given by ); the running requirements of the current task (such as urgency, required resources);

[0102] The first step is to map the node risk score to the device dimension. Since multiple event nodes may correspond to the same device , the present invention calculates the average risk of each device:

[0103]

[0104] Where represents the set of all event nodes mapped to device .

[0105] Then, construct a scheduling scoring function for each task and device :

[0106]

[0107] Where is the risk score of the device; is the adjustment response ability of the device; is the task urgency (the larger the integer, the more urgent); is the weight factor set by the system to balance adjustability and urgency.

[0108] Among them, the core of the design of this scoring function lies in:

[0109] Risk aversion: The higher the risk of the device, the smaller it is, and the lower the score;

[0110] Task matching: The more urgent the task, the higher the score; the more flexible the device, the better the matching;

[0111] The function structure is easy to deploy and interpretable, and is suitable for direct use in the MES system scorekeeper.

[0112] For example: If the task has , and the optional device has , , , , then:

[0113] This score can be used for sorting / filtering in the task scheduling system.

[0114] S5. Convert the score matrix of the current task for all candidate devices into a scheduling policy for execution, execute the scheduling policy, and collect task execution feedback. Dynamically adjust the target causal graph and node risk scores based on the difference between the actual result and the predicted value to achieve double optimization of the target causal graph and the scheduling policy.

[0115] Specifically, this step is the closed-loop convergence link of the solution of the present invention. Its core task is to convert the scheduling score vector generated in S4 into specific scheduling policy execution actions, record and feedback on the scheduling results, and introduce a feedback-driven mechanism to achieve dynamic correction and learning of the causal structure graph and risk scores, thereby forming a dual-path optimization closed-loop of structure - policy. Traditional intelligent factory scheduling systems often have the problem of "static" policies, that is, the scheduling model is difficult to respond to changes in execution results, resulting in risks being repeatedly activated and the policy converging to a sub-optimal value.

[0116] This step proposes an optimization method that combines policy execution trajectory feedback and structure weight adjustment mechanism. Based on the actual scheduling behavior, construct an execution feedback record set, improve the edge weights of the graph structure through the structure loss backpropagation mechanism, and update the propagation parameters in the risk score calculation function at the same time, truly realizing that "the model continuously evolves as the system runs".

[0117] First, the system executes scheduling according to the score result, completes the task running on the device and records the corresponding results . The execution feedback can be obtained through multi-source collection. For example:

[0118] Extract the task execution status code from the MES system;

[0119] Analyze the timeout / failure label from the device log;

[0120] Use sensor data to determine task completion (e.g., a temperature drop represents cooling completion).

[0121] Next, the system takes the task - resource feedback as input, traces back its influence path, and updates the weighted edge weights of the graph structure The update formula is as follows:

[0122]

[0123] Among them, represents the edge weight after the th scheduling; is the step size for edge weight update; is to determine whether the current edge belongs to the device path; is the actual execution feedback; is based on the current predicted execution effect value (such as ); is the edge impact factor on the device's participation in the path (which can be taken as the standardized itself);

[0124] This update process reflects the structural feedback innovation mechanism of this step: The system uses the gap between the actual execution result and the risk prediction value to inversely adjust the edge weights of the corresponding paths in the graph structure to enhance the accuracy of the future causal graph.

[0125] At the same time, the propagation gain parameter in the risk scoring function can also be fine-tuned based on the result set:

[0126]

[0127] Among them, is the adjustment factor; when the actual result is worse than the prediction (i.e., ), it means that the current risk propagation factor is set too weak, and the penalty for risk diffusion needs to be enhanced; otherwise, it means that the current scoring model is too conservative, and the scheduling selection space can be appropriately relaxed.

[0128] In one embodiment, an intelligent factory management system based on the industrial Internet is provided. The system includes:

[0129] Data processing module: used to obtain the raw data of multi-source heterogeneous devices, generate a semantic event structure through preprocessing, construct an industrial event sequence, and filter out valid events through stability verification; wherein, the raw data includes: raw sampling values, device reporting timestamps, and data source device numbers; the preprocessing is to perform time alignment processing on the raw data to generate an aligned timestamp, and perform event semantic encoding based on the aligned timestamp to obtain a semantic event structure; the semantic event structure includes: event type, aligned timestamp, and data source device number; the industrial event sequence includes several semantic event structures;

[0130] Causal modeling module: used to map the industrial event sequence to nodes, generate candidate edges in combination with the device physical topology connection relationship, and construct a target causal graph;

[0131] Risk assessment module: used to calculate the node risk score of each event node according to the target causal graph, and use the graph-level centrality value to balance and adjust the score to generate the normalized node risk score of each event node;

[0132] Scheduling strategy module: used to obtain the running requirements of the current task, map the node risk score to the device, calculate the average risk of each device, and calculate based on the average risk of each device combined with the running requirements of the current task to obtain the scoring matrix of the current task for all candidate devices;

[0133] Feedback optimization module: used to convert the scoring matrix of the current task for all candidate devices into a scheduling strategy for execution, execute the scheduling strategy and collect task execution feedback, and dynamically adjust the target causal graph and node risk score according to the difference between the actual result and the predicted value to achieve double optimization of the target causal graph and the scheduling strategy.

[0134] Unless otherwise specifically stated, the relative steps, numerical expressions, and numerical values of the components and steps set forth in these embodiments do not limit the scope of the present application.

[0135] If the above functions are implemented in the form of software function units and sold or used as independent products, they can be stored in a non-volatile computer-readable storage medium executable by a processor. Based on this understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the system described in various embodiments of the present application. The foregoing storage medium includes: USB flash drive, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), magnetic disk, or optical disk, etc., which can store program codes.

[0136] In the description of the present application, it should be noted that the orientation or positional relationship indicated by the terms "upper", "lower", etc. is based on the orientation or positional relationship shown in the drawings, or the orientation or positional relationship in which the inventive product is usually placed during use. It is only for the convenience of describing the present application and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the present application.

[0137] In the description of the present application, it should also be noted that, unless otherwise clearly specified and defined, the terms "arranged", "installed", "connected", and "coupled" should be understood in a broad sense. For example, it may be a fixed connection, a detachable connection, or an integral connection; it may be a mechanical connection or an electrical connection; it may be directly connected or indirectly connected through an intermediate medium, and it may be the communication inside two components. For those of ordinary skill in the art, the specific meanings of the above terms in the present application can be understood according to specific circumstances.

[0138] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present application.

Claims

1. An intelligent factory management method based on the industrial Internet, characterized in that, The method includes: Obtaining the original data of multi-source heterogeneous devices, generating a semantic event structure through preprocessing, constructing an industrial event sequence, and screening valid events through stability verification; wherein, the original data includes: original sampling values, device reported timestamps, and data source device numbers; the preprocessing is to perform time alignment processing on the original data to generate aligned timestamps, and perform event semantic encoding based on the aligned timestamps to obtain a semantic event structure; the semantic event structure includes: event type, aligned timestamp, and data source device number; the industrial event sequence includes several semantic event structures. Mapping the industrial event sequence to nodes, generating candidate edges in combination with the device physical topology connection relationship, and constructing a target causal graph. According to the target causal graph, calculating the node risk score of each event node, and using the graph-level centrality value to balance and adjust the score to generate the normalized node risk score of each event node. Obtaining the running requirements of the current task, mapping the node risk score to the device, calculating the risk average value of each device, and calculating based on the risk average value of each device in combination with the running requirements of the current task to obtain the scoring matrix of the current task for all candidate devices. Converting the scoring matrix of the current task for all candidate devices into a scheduling policy for execution, executing the scheduling policy and collecting task execution feedback, and dynamically adjusting the target causal graph and node risk score according to the difference between the actual result and the predicted value to achieve the dual optimization of the target causal graph and the scheduling policy.

2. The intelligent factory management method based on industrial Internet according to claim 1, characterized in that Performing time alignment processing on the original data to generate aligned timestamps, and performing event semantic encoding based on the aligned timestamps to obtain a semantic event structure, specifically including: Performing unified time axis alignment processing based on the device reported timestamp to obtain an aligned timestamp. Using the aligned timestamp and the original sampling value to generate a semantic event structure. Among them, the screening of valid events through stability verification is the screening of the semantic event structure, including: Obtaining the original sampling value, calculating the change rate of the original sampling value, calculating the absolute value of the average acceleration of the original sampling value based on the change rate, and comparing the calculated absolute value of the average acceleration with a preset threshold, and only retaining the semantic event structure corresponding to the absolute value of the average acceleration less than the preset threshold.

3. The intelligent factory management method based on industrial Internet according to claim 2, wherein The performing unified time axis alignment processing based on the device reported timestamp to obtain an aligned timestamp specifically includes: Collecting the timestamps reported by all devices. Constructing a drift optimization model with the goal of minimizing the time correction value, taking each timestamp as input, and outputting an aligned timestamp to minimize the loss function.

4. The intelligent factory management method based on industrial Internet according to claim 1, wherein, The mapping the industrial event sequence to nodes, generating candidate edges in combination with the device physical topology connection relationship, and constructing a target causal graph specifically includes: Obtain the duration of the current time, estimated by counting whether the same event occurs continuously within a certain number of seconds after it, and use it as the propagation credibility factor; based on the node set, when the aligned timestamp of the first node is less than the timestamp of the second node and there is a connection relationship in the physical topology between the data source device numbers of the first node and the second node, it is used as the candidate edge between the first node and the second node; Calculate the edge weight according to the candidate edge; Retain the incoming edges with the highest edge weights for each node to construct a complete graph spectrum.

5. The intelligent factory management method based on industrial Internet according to claim 4, characterized in that Calculating the edge weight according to the candidate edge , which is calculated as: Among them, is the conditional probability that occurs after the event occurs; is 's marginal probability, which is used for normalization; is the normal propagation delay of device ; is the standard deviation of the allowed propagation window; is the duration of event , which is used to determine whether it is a systematic and slowly developing anomaly; is the system's maximum anomaly duration benchmark; is the anomaly suppression coefficient, which controls the impact of chronic events; is the structure dependence factor, which represents the communication credibility between devices and is derived from the confidence of the process network structure; and respectively represent the time stamps after alignment of devices and .

6. The intelligent factory management method based on industrial Internet according to claim 5, wherein, According to the target causal graph, calculate the node risk score of each event node, specifically including: Combining the duration and the edge weight, through node risk assessment, determine the node risk score of each event node, calculated as: Among them, is the node risk score, is the abnormal intensity weight, is the event duration, is the device control ability of the event node ; i and j represent event nodes i and j; E represents the node set, represents the amplification factor control coefficient, is the event node hazard path coefficient.

7. The intelligent factory management method based on industrial Internet according to claim 1, characterized in that, Obtain the running requirements of the current task, map the node risk score to the device, calculate the risk average value of each device, and calculate based on the risk average value of each device combined with the running requirements of the current task to obtain the scoring matrix of the current task for all candidate devices, specifically including: Based on the node risk score, sum and average the risk assessment values of all devices to obtain the risk average value of each device; According to the risk average value, comprehensively consider the risk situation of the resources, the characteristics of the resources themselves, and the urgency of the task, and generate the device scheduling score for the corresponding task; among them, the task is the most urgent, with the highest score; the device is the most flexible, with the best match; Aggregate the device scheduling scores of the current task to obtain the scoring matrix of the current task for all candidate devices.

8. An intelligent factory management system based on the industrial Internet, characterized in that, The system includes: Data processing module: used to obtain the raw data of multi-source heterogeneous devices, generate a semantic event structure through preprocessing, construct an industrial event sequence, and filter out valid events through stability verification; among them, the raw data includes: raw sampling values, device reported timestamps, and data source device numbers; the preprocessing is to perform time alignment processing on the raw data to generate an aligned timestamp, and perform event semantic encoding based on the aligned timestamp to obtain a semantic event structure; the semantic event structure includes: event type, aligned timestamp, and data source device number; the industrial event sequence includes several semantic event structures; Causal modeling module: used to map the industrial event sequence to nodes, generate candidate edges in combination with the device physical topology connection relationship, and construct a target causal graph; Risk assessment module: used to calculate the node risk score of each event node according to the target causal graph, and use the graph-level centrality value to balance and adjust the score to generate the normalized node risk score of each event node; Scheduling strategy module: used to obtain the running requirements of the current task, map the node risk score to the device, calculate the risk average value of each device, and calculate based on the risk average value of each device combined with the running requirements of the current task to obtain the scoring matrix of the current task for all candidate devices; Feedback optimization module: used to convert the scoring matrix of the current task for all candidate devices into a scheduling policy for execution, execute the scheduling policy and collect task execution feedback, and dynamically adjust the target causal graph and node risk scores according to the difference between the actual result and the predicted value, so as to achieve the dual optimization of the target causal graph and the scheduling policy.

Images