A factory intelligent management method and system based on internet of things

Abstract

The application discloses a kind of factory intelligent management method and system based on Internet of Things, it is related to factory intelligent management technical field.A kind of factory intelligent management system based on Internet of Things, including have: measuring point acquisition module, semantic model module, event encapsulation module, strategy issuing module, strategy execution module, network outage autonomy module and conservative strategy module.The present application is by based on the current context of equipment semantic model and real-time event data Dynamic utility evaluation is carried out to strategy data, and in the strategy data of conflict, the comprehensive evaluation and conflict arbitration are carried out to the industrial equipment set corresponding to the linkage action and respective utility evaluation results, so that the strategy to be executed selects no longer only rely on fixed version number and priority, but fully consider field working condition, strategy necessity and conflict relationship between equipment, to reduce the probability of occurrence of strategy conflict and invalid execution.

Description

Technical Field

[0001] This invention relates to the field of intelligent factory management technology, and in particular to an intelligent factory management method and system based on the Internet of Things. Background Technology

[0002] With the development of IoT, Industrial Internet, and edge computing technologies, more and more factories are connecting field devices, sensors, and actuators to cloud platforms via edge gateways to achieve remote monitoring, alarm linkage, and optimized scheduling of production processes. A typical approach involves field devices connecting to a gateway via standard or proprietary protocols. The gateway collects various measurement data and uploads it to the cloud. The cloud then configures rules or policies based on this data and issues linkage commands when trigger conditions are met, used for starting or stopping equipment, adjusting process parameters, or performing safety protection controls. This type of solution has already found some application in industries such as energy and chemical engineering, and discrete manufacturing, in scenarios with stable network environments and minimal changes in process conditions.

[0003] However, existing factory IoT intelligent management systems generally suffer from several technical limitations: First, the semantics of measurement points often rely on manually configured fixed labels or simple rule mappings, making it difficult to reflect changes in the meaning of measurement points after equipment modifications or process adjustments. This results in the gateway failing to correctly understand the "business meaning" of the same sensor at different stages or modes. Second, cloud-based control policies typically rely on fixed version numbers and priorities for conflict arbitration, lacking dynamic evaluation of the current field context, equipment operating conditions, and policy execution effects. This can easily lead to multiple policies conflicting on the same device or execution results that do not meet safety expectations. Third, in the face of common weak network and network outage environments in industrial settings, existing solutions often rely solely on unified cloud-based decision-making or simple local caching. When the network is down, they can only maintain the last policy or degrade to manual intervention, lacking the ability to govern under complex conditions such as data anomalies, unreliable measurement points, or network interruptions. Furthermore, many systems only handle data reliability by simple detections such as missing or out-of-bounds errors on the acquisition side, failing to consider the reliability of "measurement point semantics" in conjunction with the reliability of "data quality." Summary of the Invention

[0004] This invention proposes an intelligent factory management method and system based on the Internet of Things. It is an intelligent management mechanism that simultaneously considers device semantic understanding, policy conflict arbitration, and data credibility assessment at the edge. This enables the edge gateway to dynamically evaluate and adaptively adjust the control strategy based on real-time context under conditions of changing operating conditions, data anomalies, and network instability.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows: An IoT-based intelligent factory management method includes: Industrial equipment is discovered and accessed through an edge gateway, and the communication protocol is determined and measurement point data is collected based on the communication characteristics of the industrial equipment. Perform probabilistic semantic inference based on multi-source context on the measurement point data, generate unified semantic labels with semantic confidence, and dynamically update the device semantic model; The measurement point data is encapsulated into event data containing event identifiers and sequence numbers, written to the local event cache queue, and stored after deduplication. Generate and distribute strategy data in the cloud, including version number, priority, trigger conditions, linkage actions and rollback actions; After receiving policy data, the edge gateway performs dynamic utility evaluation and conflict arbitration on the policy data based on the current context composed of the device semantic model and real-time event data, determines the policy to be executed and executes the linkage action, and executes the rollback action when it fails. When the edge gateway fails to connect to the cloud, it enters the autonomous mode of disconnection and continuously caches event data. After the connection is restored, it is replayed and uploaded according to the sequence number and the event data consistency is verified. A comprehensive credibility score is calculated by combining semantic confidence and data quality indicators of measurement point data. When the comprehensive credibility score is lower than the threshold, the corresponding measurement point data is isolated, and redundant measurement point data or estimated data are used to replace the measurement point data to participate in the trigger judgment. Dynamic utility evaluation and conflict arbitration are triggered to generate or select a conservative control strategy.

[0006] As a preferred technical solution of the present invention, the probabilistic semantic inference of the measurement point data based on multi-source context includes: the edge gateway acquiring the device communication characteristics of the industrial equipment and the spatiotemporal pattern characteristics of the measurement point data, and retrieving the policy execution feedback; generating the current multi-source context based on the device communication characteristics, spatiotemporal pattern characteristics and policy execution feedback; performing probabilistic semantic inference on the measurement point data based on the multi-source context to generate unified semantic labels and corresponding semantic confidence scores; and dynamically updating the device semantic model based on the semantic confidence scores.

[0007] As a preferred technical solution of the present invention, the dynamic updating of the device semantic model includes: when the semantic confidence meets the preset update conditions, the edge gateway writes the unified semantic label into the device semantic model to complete the update; when the semantic confidence does not meet the preset update conditions, the edge gateway keeps the device semantic model unchanged and performs probabilistic semantic inference based on the subsequently collected measurement point data to update the semantic confidence.

[0008] As a preferred embodiment of the present invention, the semantic confidence is adaptively corrected based on policy execution feedback. The policy execution feedback includes the execution result of the linkage action or the execution result of the rollback action. When the policy execution feedback indicates that the linkage action fails and triggers the rollback action, the semantic confidence of the corresponding unified semantic tag is reduced and the unified semantic tag is prohibited from being written into the device semantic model until the semantic confidence meets the preset update conditions. When the policy execution feedback indicates that the linkage action is executed successfully and the rollback action is not triggered, the semantic confidence of the corresponding unified semantic tag is increased, and the device semantic model is dynamically updated based on the increased semantic confidence.

[0009] As a preferred embodiment of the present invention, the dynamic utility evaluation includes: the edge gateway determines a unified semantic label corresponding to the triggering condition from the device semantic model based on the triggering condition in the policy data, and obtains measurement point data corresponding to the unified semantic label from the real-time event data to form the current context; when the measurement point data meets the triggering condition, the edge gateway evaluates the utility of the linkage action corresponding to the policy data in the current context based on the deviation of the measurement point data from the threshold in the triggering condition and the semantic confidence of the unified semantic label, and obtains a utility evaluation result for conflict arbitration.

[0010] As a preferred embodiment of the present invention, the conflict arbitration includes: when the triggering conditions of multiple policy data are met, the edge gateway determines the set of industrial equipment affected by each policy data based on the linkage action and device semantic model corresponding to each policy data, and when the sets of industrial equipment of any two policy data intersect, the corresponding policy data is determined as conflicting policy data; among the conflicting policy data, the edge gateway determines their respective comprehensive evaluation values ​​based on their utility evaluation results and the utility evaluation results of other policy data that conflict with them, and selects the policy data with the largest comprehensive evaluation value as the candidate policy data to be executed; when the candidate policy data to be executed is not unique, it further compares the version number and priority of the candidate policy data to be executed to determine the policy to be executed.

[0011] As a preferred embodiment of the present invention, the event data consistency verification includes: during the replay upload of event data, the edge gateway detects whether there are missing, duplicate, or out-of-order sequence numbers in the event data uploaded to the cloud, based on the sequence number of the event data recorded when it is written to the local event cache queue; when a missing sequence number is detected, the event data corresponding to the missing sequence number is retrieved from the local event cache queue and re-uploaded; when a duplicate or out-of-order sequence number is detected, the redundant event data is discarded and the event data is rearranged according to the sequence number order.

[0012] As a preferred technical solution of the present invention, the calculation of the comprehensive credibility score includes: the edge gateway performing statistical analysis on the measurement point data within a preset time window based on the missing, out-of-bounds, abrupt changes, and data quality indicators that do not change within a preset time to obtain a first credibility score; and obtaining a second credibility score based on the semantic confidence of the corresponding unified semantic tag and the degree of deviation of the current measurement point data from the statistical characteristics of the historical measurement point data of the unified semantic tag in the device semantic model; when the first credibility score is higher than a first threshold, the comprehensive credibility score is calculated according to a preset first combination rule; otherwise, the comprehensive credibility score is calculated according to a preset second combination rule; the contribution coefficient of the first credibility score in the first combination rule is less than the contribution coefficient of the second credibility score, and the contribution coefficient of the first credibility score in the second combination rule is greater than the contribution coefficient of the second credibility score.

[0013] As a preferred technical solution of the present invention, the generation or selection of conservative control strategies includes: when the comprehensive credibility score is lower than a threshold, the edge gateway generates new strategy data based on the strategy data corresponding to the current strategy to be executed, keeps the original triggering conditions unchanged, adjusts the linkage action corresponding to the strategy data to act on the same industrial equipment, and causes the relevant measurement point data determined by the equipment semantic model to change towards the safety side, and sets the priority of the new strategy data to be higher than the priority of the original strategy data, and participates in the subsequent dynamic utility evaluation and conflict arbitration as a conservative control strategy; when there is strategy data that meets the above conditions in the issued strategy data, the edge gateway selects the strategy data as a conservative control strategy to participate in the dynamic utility evaluation and conflict arbitration.

[0014] An IoT-based intelligent factory management system includes: Measurement point acquisition module: Discovers and connects to industrial equipment through the edge gateway, determines the communication protocol based on the communication characteristics of the industrial equipment, and collects measurement point data; Semantic Model Module: Performs probabilistic semantic inference on measurement point data based on multi-source context, generates unified semantic labels with semantic confidence, and dynamically updates the device semantic model; Event encapsulation module: Encapsulates the measurement point data into event data containing event identifiers and sequence numbers, writes it to the local event cache queue, and stores it after deduplication; Policy delivery module: Generates and delivers policy data in the cloud, including version number, priority, trigger conditions, linkage actions and rollback actions; Policy execution module: After receiving policy data, the edge gateway performs dynamic utility evaluation and conflict arbitration on the policy data based on the current context composed of the device semantic model and real-time event data, determines the policy to be executed and executes the linkage action, and executes the rollback action when it fails. The offline self-governance module: When the connection between the edge gateway and the cloud is abnormal, it enters the offline self-governance mode and continuously caches event data. After the connection is restored, it replays and uploads the data according to the sequence number and performs event data consistency verification. Conservative Strategy Module: Calculates the overall credibility score. When the overall credibility score is lower than the threshold, the corresponding measurement point data is isolated. A conservative control strategy is generated or selected using redundant measurement point data or estimated data.

[0015] The present invention has the following advantages: This invention introduces a mechanism for adaptive correction of semantic confidence based on policy execution feedback. When the linkage action fails and triggers a rollback operation, the semantic confidence of the corresponding unified semantic label is automatically reduced and writing to the device semantic model is prohibited. When the linkage action succeeds and does not trigger a rollback operation, the semantic confidence is increased and the device semantic model is updated. This enables the semantic model to continuously self-correct in combination with the actual control effect, enhancing its self-correction capability for abnormal configurations and erroneous semantics.

[0016] This invention performs dynamic utility evaluation of strategy data based on the current context composed of device semantic model and real-time event data. In the case of conflicting strategy data, it combines the set of industrial devices corresponding to the linkage action and their respective utility evaluation results for comprehensive evaluation and conflict arbitration. This makes the selection of the strategy to be executed no longer dependent on fixed version number and priority, but fully considers the on-site working conditions, the necessity of the strategy and the conflict relationship between devices, thereby reducing the probability of strategy conflict and invalid execution.

[0017] This invention encapsulates measurement point data into event data containing event identifiers and sequence numbers, and writes it into a local event cache queue for deduplication storage. It continuously caches event data in the autonomous mode when the network is down, and replays and uploads it according to the sequence number after the network is restored, while performing event data consistency verification. This ensures the order and integrity of events in the cloud and edge under weak network and network downtime scenarios, realizes event-level data traceability and consistency control, and improves reliability in unstable network environments.

[0018] This invention obtains a first credibility score by statistically analyzing the measurement point data within a preset time window based on data quality indicators such as missing, out-of-bounds, abrupt changes, and long-term invariance. It then obtains a second credibility score by combining the semantic confidence of the unified semantic label and the deviation of the current measurement point data from historical statistical characteristics. Finally, it calculates a comprehensive credibility score according to different combination rules, so that the comprehensive credibility score reflects both the quality of the signal itself and the semantic rationality.

[0019] This invention proactively isolates the corresponding measurement point data when the overall credibility score is below a threshold, and uses redundant or estimated measurement point data to replace the original measurement point data in the trigger judgment. At the same time, it triggers the dynamic utility evaluation and conflict arbitration on the edge gateway side to generate or select a conservative control strategy. Under the premise of keeping the original triggering conditions unchanged, the linkage action is adjusted to a control method that makes the relevant measurement point data change towards the safety side. In the case of unreliable data or semantic uncertainty, it automatically retreats to a safer control mode, which significantly improves the self-healing capability and intrinsic safety level under abnormal operating conditions. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings in the following description are only schematic diagrams of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort. Figure 1 This is a schematic diagram of the structure of an IoT-based intelligent factory management system used in an embodiment of the present invention. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0022] Example 1: A factory intelligent management method based on the Internet of Things (IoT) includes the following steps: In this embodiment, the relative degree terms such as "higher," "larger," "lower," "smaller," "stronger," and "weaker" are not general subjective descriptions, but rather level mappings based on preset quantitative ranges or grading rules. Specifically, during the deployment phase, the edge gateway sets several numerical ranges or empirical steps for relevant parameters (including but not limited to trigger strength, semantic confidence, utility evaluation results, and data quality indicators) based on process requirements and historical operating data. For example, the numerical range is divided into three levels: low, medium, and high. During operation, continuous values ​​are classified into the corresponding "lower," "medium," and "higher" level descriptions according to this mapping relationship. The aforementioned relative degree terms appearing in this embodiment should be understood as quantitative level results obtained based on such preset ranges or grading rules, rather than unconstrained qualitative descriptions.

[0023] Step S1: Discover and access industrial equipment through the edge gateway, determine the communication protocol based on the communication characteristics of the industrial equipment, and collect measurement point data; In this embodiment, the factory production site includes several industrial devices (reactor A for synthesis reaction, circulating pump P1 for material conveying, cooling water valve V1 for cooling control, and workshop temperature and humidity acquisition device for environmental monitoring). These industrial devices are connected to the edge gateway via fieldbus interfaces (such as RS485). During power-on or periodic scanning, the edge gateway sequentially initiates probe requests to network addresses and serial port channels within its configured range, and collects the basic message information returned by the devices. It extracts the device IP address or slave address, device manufacturer identifier, device model identifier, response delay, message length, message header feature bytes, and function code distribution as communication characteristics of the industrial devices for subsequent communication protocol determination.

[0024] The edge gateway pre-stores feature templates for various industrial communication protocols. For example, the first protocol template corresponds to a fixed message length and a specific function code combination, while the second protocol template corresponds to a specific message header identifier and CRC check method. When the edge gateway acquires the communication characteristics of an industrial device, it matches these characteristics with the various protocol feature templates, comparing message length, function code sequence, response delay distribution, and vendor identifier. Based on the matching results, it selects the communication protocol type corresponding to the industrial device. For reactor A, which uses Ethernet TCP / IP communication, the edge gateway determines that it uses the first industrial protocol based on the vendor identifier and function code distribution in the handshake message. For circulating pump P1, which uses an RS485 bus, it determines that it uses the second industrial protocol based on the serial port baud rate, frame structure, and slave address response mode. After the communication protocol is determined, the edge gateway configures the measurement point polling cycle and register address table based on the corresponding protocol description.

[0025] After completing protocol identification, the edge gateway polls and reads the equipment according to the preset acquisition cycle for each industrial device, using the corresponding communication protocol to obtain raw measurement point data. Taking reactor A as an example, the edge gateway reads information such as the inlet temperature measurement point, the internal pressure measurement point, and the stirring motor current measurement point of reactor A every 5 seconds; taking circulating pump P1 as an example, the edge gateway reads measurement point data such as the pump outlet pressure, operating status word, and fault status word every 10 seconds. For each measurement point data, the edge gateway records the measurement point identifier ("Reactor A_Inlet Temperature", "Circulating Pump P1_Operating Status"), acquisition timestamp, measurement value, and physical unit during acquisition, forming a structured measurement point data record.

[0026] Step S2: Perform probabilistic semantic inference on the measurement point data based on multi-source context, generate unified semantic labels with semantic confidence, and dynamically update the device semantic model; In this embodiment, multi-source context refers to the process by which the edge gateway interprets a single measurement point data point, considering the current value of the measurement point itself, the communication characteristics of the devices related to the measurement point, the temporal change characteristics of the measurement point over a period of time, and the historical policy execution feedback information related to the measurement point. Taking the "Reactor A_Inlet Temperature" measurement point as an example, the multi-source context includes: the device model and manufacturer characteristics of Reactor A as a continuous heating device, the typical change curve of this temperature measurement point in several batches of production in the past, and the linkage cooling or shutdown policies and their execution results that the system has executed when the temperature measurement point enters a certain value range. Based on the above multi-source context, the edge gateway performs probabilistic semantic inference on the measurement point data at the current moment, determines whether the measurement point should be understood as "preheating temperature", "constant temperature reaction temperature" or "cooling stage temperature" in the current process stage, and assigns a semantic confidence score to each candidate semantic. Finally, the semantic with the highest semantic confidence score is selected as the current unified semantic label for the measurement point, and the semantic confidence score is recorded in the device semantic model. In this embodiment, the equipment semantic model is used to maintain the mapping relationship between industrial equipment and unified semantic tags, the semantic confidence of each semantic tag, and the statistical characteristics of historical measurement points corresponding to the semantic tag.

[0027] The probabilistic semantic inference of the measurement point data based on multi-source context includes: the edge gateway acquiring the device communication characteristics of the industrial equipment and the spatiotemporal pattern characteristics of the measurement point data, and retrieving the policy execution feedback; generating the current multi-source context based on the device communication characteristics, spatiotemporal pattern characteristics and policy execution feedback; performing probabilistic semantic inference on the measurement point data based on the multi-source context to generate unified semantic labels and corresponding semantic confidence scores; and dynamically updating the device semantic model based on the semantic confidence scores.

[0028] Specifically, the edge gateway first classifies reactor A, circulating pump P1, and cooling water valve V1 (industrial equipment) into different equipment types and manufacturer categories based on device communication characteristics, providing prior knowledge on the device side for subsequent semantic inference. Simultaneously, the edge gateway constructs spatiotemporal pattern features for each measurement point within a preset time window. For example, for the "reactor A_inlet temperature" measurement point, it statistically analyzes the average, maximum, minimum, and rate of change over the past 30 minutes, as well as its correlation with the "reactor A_internal pressure" measurement point, forming a spatiotemporal pattern feature vector for that measurement point. For policy execution feedback, the edge gateway extracts from the policy execution process records the triggered actions, whether the actions were successful, whether a rollback action was triggered, and the changes in the device state after execution when the measurement point was within a certain value range. This information is then organized into historical control effect features related to the measurement point. The aforementioned device communication features, measurement point spatiotemporal pattern features, and policy execution feedback are combined to form the multi-source context input for the current moment, which is processed by probabilistic semantic inference logic.

[0029] In this embodiment, the probabilistic semantic inference logic matches and scores multi-source contexts using a preset set of rules and a statistical model. For example, when reactor A is in the preheating stage of a certain formula and the inlet temperature shows a monotonically increasing trend, and the "heating power limit" strategy has been successfully triggered in this stage in the past, the confidence level of the "preheating temperature" semantic is increased. When the inlet temperature is close to the historical reaction temperature range and the stirring motor has been running stably, the confidence level of the "reaction temperature" semantic is increased. Finally, the edge gateway selects the one with the highest confidence level from multiple candidate semantics as a unified semantic label and outputs the unified semantic label and its corresponding semantic confidence level to the device semantic model for subsequent updates. Based on the new unified semantic label and semantic confidence level, the device semantic model adjusts the original mapping relationship from measurement points to semantics, so that the semantic description in the model gradually conforms to the actual operating conditions.

[0030] The dynamic updating of the device semantic model includes: when the semantic confidence meets the preset update conditions, the edge gateway writes the unified semantic label into the device semantic model to complete the update; when the semantic confidence does not meet the preset update conditions, the edge gateway keeps the device semantic model unchanged and performs probabilistic semantic inference based on the subsequently collected measurement point data to update the semantic confidence.

[0031] In this embodiment, the preset update conditions include two parts: a semantic confidence threshold and a duration condition. For example, for the semantic label corresponding to "Reactor A_Inlet Temperature", only when the semantic confidence score is consistently higher than 0.8 for multiple inferences is the semantic label considered reliable in the current process stage, and then it is written into the device semantic model as the currently effective semantic mapping. When the factory production formula or process flow is adjusted, and the inlet temperature measurement point of Reactor A gradually changes from the original role of "preheating temperature" to "cooling temperature", the device semantic model will not immediately change its semantics due to short-term fluctuations. Instead, it accumulates semantic confidence through multiple probabilistic inferences, and writes it uniformly after meeting the preset update conditions, ensuring the stability of the model update process. When the semantic confidence score obtained from a certain inference is lower than the preset update conditions, the edge gateway retains the original semantic mapping relationship in the device semantic model unchanged, while continuing to perform probabilistic semantic inference on the subsequently collected measurement point data, and updates the semantic confidence score according to the new multi-source context results, until the semantic confidence score meets the preset update conditions again or a new semantic label becomes dominant.

[0032] The semantic confidence is adaptively corrected based on policy execution feedback, which includes the execution result of the linkage action or the execution result of the rollback action. When the policy execution feedback indicates that the linkage action failed and triggered the rollback action, the semantic confidence of the corresponding unified semantic label is reduced and the unified semantic label is prohibited from being written into the device semantic model until the semantic confidence meets the preset update conditions. When the policy execution feedback indicates that the linkage action was successful and did not trigger the rollback action, the semantic confidence of the corresponding unified semantic label is increased, and the device semantic model is dynamically updated based on the increased semantic confidence.

[0033] In this embodiment, the strategy execution feedback is generated by the strategy execution process in S5 and returned to the semantic model update link in the form of recorded entries. For a linkage cooling strategy triggered by "reactor A_inlet temperature", when probabilistic semantic inference interprets the measurement point as "cooling stage temperature" and selects a linkage strategy accordingly, if the strategy execution fails and the edge gateway triggers the corresponding rollback operation, the strategy execution feedback will mark the linkage action as failed and triggered a rollback in the feedback record. When the semantic model module receives such feedback, it will consider that the previous judgment on the unified semantic label "cooling stage temperature" was biased, and the corresponding semantic confidence will decrease according to the preset decrease range. The semantic label will be temporarily prohibited from being written into the device semantic model, even if probabilistic semantic inference continues to give the same candidate semantic in the short term. Only in the subsequent production process, when the linkage action is successfully executed again without triggering a rollback operation through multiple strategy execution feedbacks, will the semantic confidence gradually increase until the preset update conditions are met again and writing into the model is allowed. Conversely, when the linkage action is successfully executed and no rollback action is triggered, the strategy execution feedback will mark the control behavior as a valid control result. Based on this, the semantic model module increases the semantic confidence of the corresponding unified semantic label, thereby increasing the trust of the device semantic model in the semantic interpretation. When the semantic confidence reaches the preset update condition, the semantic label is written into the device semantic model, so that subsequent strategy evaluation and trigger judgment are based on the updated semantics.

[0034] Step S3: Encapsulate the measurement point data into event data containing event identifiers and sequence numbers, write it to the local event cache queue, and store it after deduplication; In this embodiment, event data in a unified format is constructed based on the measurement point data and its semantic information for subsequent real-time processing and offline replay. The event data refers to a record abstracted from a single or group of measurement point acquisition actions, with time as the axis, and each record is assigned a unique identifier and sequence information. Specifically, after each acquisition of measurement point data such as the inlet temperature of reactor A, the pressure inside the reactor, the outlet pressure of circulating pump P1, and the opening degree of cooling water valve V1, the edge gateway combines this measurement point data with the corresponding acquisition timestamp, unified semantic tag, and semantic confidence level to form the event subject. Furthermore, an event identifier and sequence number are added to this subject to constitute complete event data.

[0035] Event identifiers are used to logically uniquely mark the occurrence of an event, facilitating tracking and classification management across the cloud and edge. An event identifier is composed of a device identifier, a measurement point category, and an event type. For example, for a periodic sampling event of the inlet temperature of reactor A, "EV_REACTOR_A_INLET_TEMP_SAMPLE" is used as the event identifier; for an over-temperature alarm event triggered by the inlet temperature of reactor A exceeding the process limit, "EV_REACTOR_A_INLET_TEMP_OVERLIMIT" is used as the event identifier. Through this combination, the event identifier indicates both the industrial equipment from which the event originates and the measurement points and business semantics involved, facilitating subsequent aggregation and analysis in the cloud by device, process segment, or event type.

[0036] The sequence number is used to characterize the relative order of event data over time and provides a basis for network replay and consistency verification. In this embodiment, the edge gateway assigns a monotonically increasing integer sequence number to each event data item. The sequence number increments according to the event generation time and continues to use the latest persistently stored sequence number after the edge gateway restarts, to avoid sequence number conflicts between different operating cycles. Specifically, the edge gateway maintains a global sequence counter. Each time a new event data item is generated, the current value of the counter is written as the sequence number into the event data, and then the counter is incremented by one.

[0037] In this embodiment, the local event cache queue is used to store event data that has not yet been uploaded to the cloud or is waiting to be replayed. The queue organizes the event data in ascending order using the sequence number as the sorting key and stores it in the local storage medium of the edge gateway. After generating new event data, the edge gateway writes it to the tail of the local event cache queue and performs deduplication storage before writing. Deduplication storage means that when writing new event data, it checks whether there is already a record with the same event identifier and the same sequence number in the queue, or whether duplicate events with completely identical content have been stored within a short time window. Event data determined to be duplicated will not be written again, thereby avoiding a large amount of redundant data in the local cache due to device jitter or duplicate reporting.

[0038] Specifically, in a periodic sampling scenario, if the inlet temperature of reactor A remains exactly the same across two consecutive sampling periods, and the corresponding unified semantic label and semantic confidence remain unchanged, the edge gateway will compare the event data content with the previous event with the same event identifier before generating the second sampling event. If the sampled value and semantic information are confirmed to be completely consistent, the second sampling event can be optionally marked as a duplicate event, not written to the local event cache queue, and only the most recent sampling time is updated in the internal state for subsequent statistical analysis. For critical process measurement points that require strict recording of each sample, the edge gateway does not perform content-level deduplication, but only uses the sequence number as the deduplication criterion. If a duplicate write request occurs due to network jitter or abnormal retries, and the sequence number of the event data to be written is the same as that of the event data at the end of the queue, the duplicate event data is discarded.

[0039] Step S4: Generate and distribute strategy data in the cloud, including version number, priority, triggering conditions, linkage actions and rollback actions; In this embodiment, policy data is used to describe the control rules configured in the cloud for the factory production process. Its core function is to structure information such as "under what conditions, what control actions are performed on which industrial devices, and how to recover in case of control failure" into records and send them to the edge gateway. Policy data is configured based on unified semantic tags, enabling the same policy to be portable across different field device instances and working in conjunction with the device semantic model.

[0040] Specifically, a policy configuration library is maintained in the cloud. Process engineers create multiple policies based on the business requirements of reactor A, circulating pump P1, and cooling water valve V1 in the policy configuration interface. Each policy is assigned a unique policy identifier upon generation and includes fields for version number, priority, trigger condition, linkage action, and rollback action, forming a complete policy data structure. The version number is used to distinguish the revision status of the same business policy at different configuration stages. For example, the initial version is set to "V1", and after adjusting the trigger threshold or linkage logic, it is upgraded to "V2", which facilitates subsequent version comparison and conflict arbitration on the edge gateway side. The priority is used to express the importance and execution order of the policy. The higher the value, the higher the priority of the policy when multiple candidate policies exist. For example, safety interlock policies are set to high priority, and energy consumption optimization or equipment maintenance policies are set to medium to low priority.

[0041] The trigger condition field defines the on-site conditions upon which the strategy takes effect. It describes a unified semantic label and its corresponding measurement point data conditions. Taking reactor A as an example, when configuring the "over-temperature interlock strategy" in the cloud, the trigger condition is defined as "the unified semantic label is 'reactor A_inlet temperature' and the corresponding measurement point data continuously exceeds 120℃ for 300 seconds." Here, "reactor A_inlet temperature" is the unified semantic label formed in step S2, and "exceeds 120℃ for 300 seconds" represents the numerical and time constraints on the measurement point data under this label. For the protection strategy of circulating pump P1, the trigger condition can be defined as "the unified semantic label is 'circulating pump P1_outlet pressure' and the corresponding measurement point data is below the set lower limit for several sampling cycles," thus characterizing the risk of insufficient flow or idling. The trigger conditions are stored in the cloud as structured condition expressions and are sent to the edge gateway for judgment by the strategy execution logic in conjunction with real-time event data.

[0042] The "Action Linkage" field describes the specific actions the cloud expects the edge gateway to perform on industrial equipment when trigger conditions are met. Action linkage is associated with specific equipment and control variables based on unified semantic tags. For example, for the strategy "Reactor A_Inlet Temperature Exceeds Limit," the linkage actions include "Reduce the opening of the heating steam regulating valve to below 50%", "Keep the stirring motor running", and "Send an alarm message to the operator terminal." For the strategy "Circulating Pump P1 Outlet Pressure is Low," the linkage actions include "Reduce the opening of the upstream feed valve" and "Limit the feeding rate of Reactor A if the pressure has not recovered." Action linkage is recorded in the strategy data as an action list. Each action includes a target equipment identifier, a target control variable identifier, and a target control value or control mode, used to drive the corresponding equipment to execute.

[0043] The rollback action field defines the reverse operation to restore the current state when a linked action fails or produces an undesirable effect. When configuring the strategy in the cloud, a corresponding rollback action is assigned to each linked action. For example, for the linked action of "reducing the opening of the heating steam regulating valve," the rollback action is set to "restore the opening of the heating steam regulating valve to the set value recorded before execution"; for the linked action of "limiting the feeding rate of reactor A," the rollback action is set to "restore the feeding rate to the formula set value before strategy execution." Rollback actions and linked actions are recorded one-to-one in the strategy data and used to indicate whether rollback is triggered in the strategy execution feedback. The adaptive correction of semantic confidence relies on the execution results of the linked actions and rollback actions for correction. Therefore, the rollback actions in the strategy data not only serve the function of control recovery but also provide an important source of feedback information for the semantic model.

[0044] During the policy distribution process, the cloud selects a set of policy data related to the target factory or production line based on the current process formula, production schedule, and equipment list, and packages and sends it to the edge gateway through a secure communication channel. After receiving the policy data, the edge gateway stores the policy data in its local policy storage area and records the version number and priority of each policy, providing a policy basis for dynamic utility evaluation and conflict arbitration based on the current context.

[0045] Step S5: After receiving the policy data, the edge gateway performs dynamic utility evaluation and conflict arbitration on the policy data based on the current context composed of the device semantic model and real-time event data, determines the policy to be executed and executes the linkage action, and executes the rollback action when it fails. In this embodiment, a closed-loop process of "strategy selection—decision-execution—feedback" is implemented at the edge. After completing S1 to S4, the edge gateway locally stores multiple strategy data related to reactor A, circulating pump P1, and cooling water valve V1, and continuously obtains real-time event data to continuously update the device semantic model. The current context is composed of the device semantic model and real-time event data. The device semantic model provides unified semantic labels, semantic confidence, and historical statistical characteristics, while the real-time event data provides the collected values ​​and event sequences of each measuring point at the current moment. Within each strategy evaluation cycle, the edge gateway reads all or all strategy data related to the current process stage from the local strategy storage area, performs dynamic utility evaluation on each of these strategy data, and performs conflict arbitration based on the conflict relationship between the evaluation results and the policy's target, thereby selecting a single strategy to be executed and calling the communication protocol to issue linkage control commands to the corresponding industrial equipment. When the linkage action fails, the device is restored according to the rollback action field in the strategy data, and a strategy execution feedback record is generated to provide a data source for adaptive correction of semantic confidence.

[0046] The dynamic utility evaluation includes: the edge gateway determines the unified semantic label corresponding to the trigger condition from the device semantic model based on the trigger condition in the policy data, and obtains the measurement point data corresponding to the unified semantic label from the real-time event data to form the current context; when the measurement point data meets the trigger condition, the edge gateway evaluates the utility of the linkage action corresponding to the policy data in the current context based on the deviation of the measurement point data from the threshold in the trigger condition and the semantic confidence of the unified semantic label, and obtains the utility evaluation result for conflict arbitration.

[0047] Specifically, at the start of a policy evaluation cycle, the edge gateway iterates through its local policy list. For each policy data entry, it reads the unified semantic tag referenced in its trigger condition. For example, for the "Reactor A Inlet Temperature Over-Temperature Interlock Policy," the unified semantic tag referenced by the trigger condition is "Reactor A_Inlet Temperature"; for the "Circulating Pump P1 Low Pressure Protection Policy," the unified semantic tag referenced by the trigger condition is "Circulating Pump P1_Outlet Pressure." The edge gateway retrieves these unified semantic tags in the device semantic model, obtaining the semantic confidence corresponding to the tag and the historical statistical characteristics associated with that tag. Based on this, the edge gateway searches for the latest event data record within the current cycle from the real-time event data, whose event identifier corresponds to the unified semantic tag, and extracts the measurement point acquisition timestamp and measurement point value, thus constructing the current context sample used to determine whether to trigger the policy.

[0048] When the measurement point data meets the triggering conditions, the edge gateway further refines the analysis of the "trigger strength," that is, it calculates and compares the degree of deviation of the measurement point data from the threshold in the triggering conditions. For example, when the triggering condition for "Reactor A_Inlet Temperature" is "temperature above 120℃ for 300 seconds," if the current measurement point value is 135℃ and has been above the threshold for the past 300 seconds, the deviation is large in both numerical and temporal dimensions, and the trigger strength of this strategy is calibrated as high. If the current measurement point value is 122℃ and the duration has just reached 300 seconds, the deviation is relatively small, and the trigger strength is calibrated as medium. For the "Circulating Pump P1_Outlet Pressure Low Pressure Protection Strategy," when the outlet pressure is significantly lower than the lower limit for a long period of time, the trigger strength is also calibrated as high.

[0049] After obtaining the trigger strength, the edge gateway evaluates the policy utility by combining the semantic confidence of the unified semantic label. Semantic confidence reflects the reliability of the semantic interpretation upon which the current trigger condition depends. For example, when "reactor A_inlet temperature" is stably interpreted as "preheating temperature" by the device semantic model and the semantic confidence is high, the utility evaluation result of the preheating stage temperature control policy configured based on this label tends to be high. Conversely, if the device semantic model has recently undergone a semantic switch and the semantic confidence of this label is at a low level, the dynamic utility evaluation gives greater weight to semantic confidence when calculating the policy utility, thus weakening the policy utility evaluation result. The edge gateway maps the trigger strength and semantic confidence to a utility evaluation result for conflict arbitration using preset rules, such as converting it into a numerical score within a preset range. Finally, for each policy data where the trigger condition is met, the edge gateway generates a utility evaluation result corresponding to the current context, which serves as input for subsequent conflict arbitration.

[0050] The conflict arbitration includes: when the triggering conditions of multiple policy data are met, the edge gateway determines the set of industrial devices affected by each policy data based on the linkage action and device semantic model corresponding to each policy data, and when the sets of industrial devices of any two policy data have an intersection, the corresponding policy data is determined to be conflicting policy data; among the conflicting policy data, the edge gateway determines their comprehensive evaluation value based on their respective utility evaluation results and the utility evaluation results of other policy data that conflict with them, and selects the policy data with the largest comprehensive evaluation value as the candidate policy data to be executed; when the candidate policy data to be executed is not unique, it further compares the version number and priority of the candidate policy data to be executed to determine the policy to be executed.

[0051] In this embodiment, the coordinated actions of different strategy data may simultaneously act on the same equipment such as reactor A or circulating pump P1, or even make contradictory requests for the same control variable on the same equipment. For example, an energy optimization strategy may require reducing the heating power of the reactor, while a safety protection strategy may require rapidly reducing the heating power or directly shutting off the heating source when the temperature approaches the upper limit. To avoid executing contradictory coordinated actions on the same equipment within the same cycle, the edge gateway determines the industrial equipment set of each strategy data based on the coordinated actions and the device semantic model during the conflict arbitration phase. Strategy data that acts on the same equipment set and whose control objectives are contradictory are identified as conflicting strategy data.

[0052] For a set of strategy data identified as conflicting, the edge gateway comprehensively considers the utility evaluation results of each strategy and the conflict with other strategies to calculate a comprehensive evaluation value for each strategy. For example, if a safety interlock strategy has a high utility evaluation result in the current context, while the conflicting energy optimization strategy only has a medium utility evaluation result, then when calculating the comprehensive evaluation value, the comprehensive evaluation value of the safety interlock strategy is set higher than that of the energy optimization strategy, making the safety strategy a priority candidate for execution. If multiple safety protection strategies act simultaneously on reactor A but in different directions, such as one strategy controlling heating power and another controlling material feeding rate, the edge gateway further distinguishes the comprehensive evaluation values ​​of the two safety strategies based on the number of conflicting strategies and their utility evaluation results, prioritizing the strategy with a more direct impact on the overall operating conditions and more significant safety benefits as the candidate for execution.

[0053] When candidate policies to be executed are not unique, the edge gateway enters the version number and priority comparison stage. For candidate policies with different version numbers, the edge gateway selects the newest policy as the final policy to be executed based on the version number recorded in the policy data, ensuring that the latest policy issued after the process adjustment takes effect first. When there are multiple candidate policies with the same version number, their priority fields are compared. The policy with the higher priority value is selected as the policy to be executed, and the policy with the lower priority value is not executed in this round of evaluation. Through the above dynamic utility evaluation and conflict arbitration process, the edge gateway determines only one policy to be executed in each policy evaluation cycle, thereby avoiding the execution of multiple mutually interfering control actions on the same device at the same time.

[0054] After selecting the strategy to be executed, the edge gateway interacts with the industrial equipment via a communication protocol according to the linkage action fields defined in the strategy data. It sends the setpoints or control commands of the target control variables to the corresponding devices, such as writing a new opening percentage to the heating steam regulating valve of reactor A, or writing a new operating mode or speed setting to circulating pump P1. Simultaneously, the edge gateway activates linkage action result monitoring, collecting relevant measurement point data and device status words within a preset time window, and determining whether the linkage action was successfully executed based on predefined success criteria. If the linkage action does not achieve the desired effect within the specified time, or the device status word returns a fault flag, the edge gateway performs a recovery operation according to the rollback operation recorded in the strategy data. For example, it restores the opening of the heating steam valve of reactor A to the previous setpoint, or restores circulating pump P1 to its original operating mode. The execution result of each linkage action and rollback operation is marked in the corresponding field of the strategy execution feedback record and used as strategy execution feedback input.

[0055] Step S6: When the edge gateway fails to connect to the cloud, it enters the autonomous mode and continuously caches event data. After the connection is restored, the data is replayed and uploaded according to the sequence number, and the consistency of the event data is verified. In this embodiment, the edge gateway periodically sends heartbeat messages or status synchronization requests to the cloud. If no response is received from the cloud within a preset timeout period, the current connection status is determined to be abnormal. Specifically, the edge gateway maintains a "cloud connection status" flag. When multiple consecutive heartbeat timeouts or communication handshake failures occur, this flag is set to a disconnected state, triggering the network-disconnected autonomous mode. In this embodiment, the network-disconnected autonomous mode refers to a working state that operates independently based on local data and issued policies. In this state, the edge gateway stops sending real-time event data and status synchronization messages to the cloud, but continues to execute the measurement point acquisition, semantic inference, policy evaluation, and linkage control logic described in S1 to S5 to maintain basic control over reactor A, circulating pump P1, and cooling water valve V1. At the same time, in the network-disconnected autonomous mode, the edge gateway no longer attempts to upload newly generated event data immediately, but instead writes it all into the local event cache queue for unified replay and uploading after the network is restored.

[0056] During the continuous caching of event data, the edge gateway continues to assign a unique sequence number to each event data item as described in S3, and writes it to the local event cache queue in sequence according to the sequence number. For the reactor A inlet temperature sampling event, the circulating pump P1 outlet pressure sampling event, and the over-temperature or under-pressure alarm events triggered by these measuring points, the edge gateway serializes and records them in the order of their generation time. To avoid local storage resources being occupied by a large number of low-value events, the edge gateway appropriately adjusts the event caching strategy in the offline autonomous mode. For example, it retains all alarm events and critical process quantity sampling events, and uses a length-limited queue or periodic archiving method to process some non-critical monitoring events. However, regardless of the caching strategy used, the event data written to the local event cache queue retains the complete event identifier, sequence number, and event body content, providing a basis for subsequent replay uploads and consistency checks.

[0057] When the edge gateway detects that communication with the cloud has returned to normal, for example, if the cloud handshake process is successfully re-established, the edge gateway resets the cloud connection status flag to a normal connection state and exits the network outage autonomous mode, entering the replay upload phase. During the replay upload phase, the edge gateway first determines the range of event sequences to be replayed. Based on the locally recorded "most recently successfully uploaded event sequence number," it starts reading event data sequentially from the first unuploaded event after that sequence number, constructing a replay queue in ascending order of sequence numbers. Subsequently, the edge gateway sends the event data in the replay queue to the cloud in batches, recording the highest successfully uploaded sequence number after each batch is uploaded.

[0058] The event data consistency verification includes: during the replay upload of event data, the edge gateway checks whether there are missing, duplicate, or out-of-order sequence numbers in the event data uploaded to the cloud, based on the sequence number of the event data recorded when it is written to the local event cache queue; when a missing sequence number is detected, the event data corresponding to the missing sequence number is retrieved from the local event cache queue and re-uploaded; when a duplicate or out-of-order sequence number is detected, the redundant event data is discarded and the event data is rearranged according to the sequence number order.

[0059] In this embodiment, the specific process of event data consistency verification is as follows: Before replaying the upload, the edge gateway performs a pre-check on the event data to be uploaded in the local event cache queue, checking whether the sequence numbers in the queue are consecutive and whether there are duplicate or reverse-order inserted entries. When a gap in the sequence numbers is found in the local queue, the edge gateway records the gap in the local log and marks the missing sequence number segment as an unrecoverable historical gap, so that the cloud can consider the data incompleteness of that time period when performing data analysis. Under normal circumstances, the local event cache queue is arranged in ascending order of sequence numbers.

[0060] During the replay upload process, the edge gateway checks the sequence number of the event data sequence to be uploaded each time it sends a batch of event data. It compares the minimum sequence number of the current batch of event data with the maximum sequence number of the previously confirmed uploaded batch, ensuring that the minimum sequence number of the current batch equals the maximum sequence number of the previous batch plus one. This guarantees that the uploaded sequence maintains a continuously increasing order from the cloud's perspective. If, during replay upload, an event data entry is found to have a sequence number less than that of a previously uploaded event, the event data is considered out of order and is not considered a valid entry for this round of upload; it is temporarily stored locally for further processing. If two or more event data entries have the same sequence number, they are considered duplicate event data; only one is retained as a valid event, and the remaining event data with the same sequence number are removed from the upload list for this batch.

[0061] When the cloud receives replay-uploaded event data, it verifies the receiving order of the event data according to the sequence number attached to the edge gateway and returns an acknowledgment message containing the largest successfully received sequence number. Upon receiving this acknowledgment message, the edge gateway checks it against its locally recorded event cache queue. If it confirms that the cloud has successfully received a specific sequence number, the edge gateway marks the event data in its local event cache queue with sequence numbers less than or equal to that sequence number as deletable and cleans them up in batches, freeing up storage resources, provided local storage space allows.

[0062] Step S7: Calculate the comprehensive credibility score by combining semantic confidence and data quality indicators of the measurement point data. When the comprehensive credibility score is lower than the threshold, isolate the corresponding measurement point data and replace the measurement point data with redundant measurement point data or estimated data to participate in the trigger judgment. This triggers dynamic utility evaluation and conflict arbitration to generate or select a conservative control strategy.

[0063] In this embodiment, step S7 is used to quantitatively evaluate the reliability of data from a single measuring point, and to isolate and replace the data of that measuring point when the reliability is insufficient, thereby preventing abnormal data or semantic understanding errors from directly participating in the policy triggering judgment and thus affecting the linkage control of reactor A and circulating pump P1. To this end, the edge gateway introduces a "comprehensive reliability score" index, which comprehensively reflects the quality status of the measuring point data itself and the degree of consistency with historical behavior under the current semantic interpretation. The higher the comprehensive reliability score, the more reliable the current reading of the measuring point is at both the signal and semantic levels; when the comprehensive reliability score is lower than a preset threshold, it indicates that the measuring point data has significant uncertainty. In this embodiment, the measuring point data is considered unsuitable for direct participation in the policy triggering judgment, and is replaced by redundant measuring points or estimated data, while switching to a more conservative control strategy.

[0064] The calculation of the comprehensive credibility score includes: within a preset time window, the edge gateway performs statistical analysis on the measurement point data based on data quality indicators such as missing, out-of-bounds, abrupt changes, and data that remains unchanged within a preset time period to obtain a first credibility score; and obtains a second credibility score based on the semantic confidence of the corresponding unified semantic tag and the degree of deviation of the current measurement point data from the statistical characteristics of historical measurement point data of the unified semantic tag in the device semantic model; when the first credibility score is higher than a first threshold, the comprehensive credibility score is calculated according to a preset first combination rule; otherwise, the comprehensive credibility score is calculated according to a preset second combination rule; in the first combination rule, the contribution coefficient of the first credibility score is less than the contribution coefficient of the second credibility score, and in the second combination rule, the contribution coefficient of the first credibility score is greater than the contribution coefficient of the second credibility score.

[0065] Specifically, data quality indicators are used to reflect the health status of the acquisition link at the signal layer. In this embodiment, the edge gateway configures a preset time window for each measurement point, such as the most recent 10 minutes or the most recent N sampling cycles. Within this time window, it counts cases of data loss, values ​​exceeding reasonable physical ranges, sudden value changes, and prolonged inactivity at that measurement point. Data loss reflects communication interruption or acquisition error; values ​​exceeding limits reflect sensor saturation or range errors; sudden changes reflect sensor jitter or hardware failure; and inactivity reflects sensor jamming or long-term failure. For example, for the "Reaction Vessel A_Inlet Temperature" measurement point, if multiple sampling failures occur within the most recent 10 minutes, or the temperature value suddenly jumps to a value far exceeding the limit temperature defined in the device semantic model, or remains completely unchanged for a long time during a process stage where temperature fluctuations are expected, the corresponding item in the data quality indicator is set to an abnormal state. The first confidence score decreases as the number and severity of abnormal items increase.

[0066] The second credibility score reflects the reasonableness of the semantic meaning. In step S2, the edge gateway maintains the semantic confidence score and historical statistical characteristics for each unified semantic label, such as the normal operating range, mean, and fluctuation range of "Reactor A_Inlet Temperature" under the semantic meaning of "Preheating Temperature". In step S7, the edge gateway reads the unified semantic label and semantic confidence score of the measurement point at the current moment and calculates the deviation of the current measurement point value from the historical statistical characteristics. When the semantic confidence score is high and the current value is within the allowable fluctuation range of the historical statistical characteristics, the second credibility score is high; when the semantic confidence score is low or the current value deviates significantly from the historical statistical range, such as a value of 200℃ appearing under the unified semantic label interpreted as "Ambient Temperature", the second credibility score decreases. In this way, the second credibility score unifies "whether the current semantic interpretation is reliable" and "whether the current value is reasonable under this semantic meaning" into a single metric.

[0067] The comprehensive credibility score is synthesized from a first credibility score and a second credibility score according to a preset combination rule. In this embodiment, when the first credibility score is higher than a first threshold, it indicates that the measurement point is in a healthy state in terms of data quality. At this time, the comprehensive credibility score focuses more on the second credibility score to reflect the importance of semantic confidence and consistency of historical behavior. When the first credibility score is lower than or equal to the first threshold, it indicates that the measurement point has significant problems in terms of data quality. At this time, the comprehensive credibility score focuses more on the first credibility score to avoid over-reliance on semantic judgment when the signal itself is obviously abnormal. Through this segmented combination method, this embodiment enables the comprehensive credibility score to simultaneously consider data quality and semantic rationality, and automatically adjust the focus under different abnormal conditions.

[0068] After calculating the overall credibility score, the edge gateway compares it with a pre-set threshold. When the overall credibility score is higher than or equal to the threshold, it indicates that the credibility of the current measurement point data meets the requirements for the execution of the control strategy, and the measurement point data continues to be used as the trigger input in the dynamic utility evaluation and conflict arbitration in step S5. When the overall credibility score is lower than the threshold, it indicates that the measurement point data is not suitable as a direct basis for the strategy trigger condition at the current moment, and step S7 performs isolation and substitution processing on the measurement point data. Isolation means that the original data of the measurement point is no longer used in the trigger condition calculation within the current trigger judgment period, but instead, alternative data is selected from redundant measurement points or estimation results.

[0069] In this embodiment, redundant measurement point data sources include parallel sensors configured for the same physical quantity. For example, two temperature sensors are installed on the inlet pipe of reactor A and connected to the edge gateway respectively, or equivalent physical quantities are derived by different combinations of measurement points on the same process equipment. Estimation data sources include empirical models or simple regression models established based on relevant measurement points. For example, based on measurement points such as "reactor A_inlet temperature", "heating steam valve opening", and "circulating pump P1 outlet pressure", a reasonable range for "reactor A_inlet temperature" is estimated using a pre-trained linear or rule-based model. When the overall credibility score of measurement point data with the unified semantic label "reactor A_inlet temperature" is lower than a threshold, the edge gateway in this embodiment prioritizes checking for redundant temperature measurement points. If a measurement point "reactor A_inlet temperature_redundancy 1" exists and its overall credibility score is higher than the threshold, then this redundant measurement point data is used to replace the original measurement point data in triggering the judgment. If no credible redundant measurement point data exists, the configured estimation model is called to calculate the estimated temperature value based on the relevant measurement point data, and this estimated value is used as the data for triggering the judgment.

[0070] After the above isolation and substitution are completed, dynamic utility evaluation and conflict arbitration are triggered to generate or select a conservative control strategy. The generation or selection of a conservative control strategy includes: when the overall credibility score is below a threshold, the edge gateway generates new strategy data based on the strategy data corresponding to the current strategy to be executed, keeping the original triggering conditions unchanged, adjusting the linkage action corresponding to the strategy data to act on the same industrial equipment, and causing the relevant measurement point data determined by the equipment semantic model to change towards the safety side, and setting the priority of the new strategy data to be higher than that of the original strategy data, participating as a conservative control strategy in subsequent dynamic utility evaluation and conflict arbitration; when there is strategy data in the issued strategy data that meets the above conditions, the edge gateway selects that strategy data as a conservative control strategy to participate in dynamic utility evaluation and conflict arbitration.

[0071] In this embodiment, the meaning of "change towards the safety side" is determined based on the physical meaning of the specific measurement point. For example, for the measurement point "Reactor A_Inlet Temperature," which is closely related to safety, a change towards the safety side refers to changing the temperature value towards a lower value within the safe process range; for the measurement point "Reactor A_Internal Pressure," a change towards the safety pressure range refers to reducing the internal pressure; for the measurement point "Circulating Pump P1_Outlet Pressure," a change towards the safety side can be defined as reducing the flow rate to reduce the mechanical impact on the equipment without affecting the minimum process requirements. When the overall reliability score is below the threshold and the current strategy to be executed involves the above measurement points, the edge gateway generates a conservative control strategy based on the target device and control quantity of the original strategy without changing the original triggering conditions. For example, the linkage action of "increasing the heating power according to the original set value" is adjusted to "limiting the heating power to a safe upper limit below the current value," or the linkage action of "increasing the speed of circulating pump P1" is adjusted to "maintaining the current speed or slightly reducing the speed." Meanwhile, the conservative control strategy is given a higher priority than the original strategy, so that in the process of dynamic utility evaluation and conflict arbitration, the conservative control strategy is selected as the strategy to be executed first in scenarios where the credibility of the measurement data is insufficient.

[0072] When the cloud has already pre-configured a conservative control policy for abnormal operating conditions during the policy distribution phase, and the triggering conditions of the conservative control policy are consistent with the current policy to be executed, and the linkage action meets the requirements for change to the security side, the edge gateway does not need to generate new policy data when the comprehensive trust score is lower than the threshold. Instead, it directly selects the policy that meets the above conditions from the distributed policy data as the conservative control policy and introduces dynamic utility evaluation and conflict arbitration.

[0073] Example 2, an IoT-based intelligent factory management system, see [link / reference] Figure 1 As shown, it includes the following modules: Measurement point acquisition module: Discovers and connects to industrial equipment through the edge gateway, determines the communication protocol based on the communication characteristics of the industrial equipment, and collects measurement point data; Semantic Model Module: Performs probabilistic semantic inference on measurement point data based on multi-source context, generates unified semantic labels with semantic confidence, and dynamically updates the device semantic model; Event encapsulation module: Encapsulates the measurement point data into event data containing event identifiers and sequence numbers, writes it to the local event cache queue, and stores it after deduplication; Policy delivery module: Generates and delivers policy data in the cloud, including version number, priority, trigger conditions, linkage actions and rollback actions; Policy execution module: After receiving policy data, the edge gateway performs dynamic utility evaluation and conflict arbitration on the policy data based on the current context composed of the device semantic model and real-time event data, determines the policy to be executed and executes the linkage action, and executes the rollback action when it fails. The offline self-governance module: When the connection between the edge gateway and the cloud is abnormal, it enters the offline self-governance mode and continuously caches event data. After the connection is restored, it replays and uploads the data according to the sequence number and performs event data consistency verification. Conservative Strategy Module: Calculates the overall credibility score. When the overall credibility score is lower than the threshold, the corresponding measurement point data is isolated. A conservative control strategy is generated or selected using redundant measurement point data or estimated data.

[0074] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A factory intelligent management method based on the Internet of Things, characterized in that, include: Industrial equipment is discovered and accessed through an edge gateway, and the communication protocol is determined and measurement point data is collected based on the communication characteristics of the industrial equipment. Perform probabilistic semantic inference based on multi-source context on the measurement point data, generate unified semantic labels with semantic confidence, and dynamically update the device semantic model; The semantic confidence is adaptively corrected based on the policy execution feedback. The policy execution feedback includes the execution result of the linkage action or the execution result of the rollback action. When the policy execution feedback indicates that the linkage action has failed and triggers the rollback action, the semantic confidence of the corresponding unified semantic tag is reduced and the unified semantic tag is prohibited from being written into the device semantic model until the semantic confidence meets the preset update conditions. When the strategy execution feedback indicates that the linkage action is successfully executed and no rollback action is triggered, the semantic confidence of the corresponding unified semantic tag is increased, and the device semantic model is dynamically updated based on the increased semantic confidence. The measurement point data is encapsulated into event data containing event identifiers and sequence numbers, written to the local event cache queue, and stored after deduplication. Generate and distribute strategy data in the cloud, including version number, priority, trigger conditions, linkage actions and rollback actions; After receiving policy data, the edge gateway performs dynamic utility evaluation and conflict arbitration on the policy data based on the current context composed of the device semantic model and real-time event data, determines the policy to be executed and executes the linkage action, and executes the rollback action when it fails. The dynamic utility evaluation includes: the edge gateway determines the unified semantic label corresponding to the trigger condition from the device semantic model based on the trigger condition in the policy data, and obtains the measurement point data corresponding to the unified semantic label from the real-time event data to form the current context; when the measurement point data meets the trigger condition, the edge gateway evaluates the utility of the linkage action corresponding to the policy data in the current context based on the deviation of the measurement point data from the threshold in the trigger condition and the semantic confidence of the unified semantic label, and obtains the utility evaluation result for conflict arbitration; When the edge gateway fails to connect to the cloud, it enters the autonomous mode of network outage and continuously caches event data. After the connection is restored, it is replayed and uploaded according to the sequence number and the event data consistency is verified. A comprehensive credibility score is calculated by combining semantic confidence and data quality indicators of measurement point data. When the comprehensive credibility score is lower than the threshold, the corresponding measurement point data is isolated, and redundant measurement point data or estimated data are used to replace the measurement point data to participate in the trigger judgment. Dynamic utility evaluation and conflict arbitration are triggered to generate or select a conservative control strategy. The generation or selection of conservative control strategies includes: when the comprehensive credibility score is lower than the threshold, the edge gateway generates new strategy data based on the strategy data corresponding to the current strategy to be executed, keeps the original triggering conditions unchanged, adjusts the linkage action corresponding to the strategy data to act on the same industrial equipment, and causes the relevant measurement point data determined by the equipment semantic model to change towards the safety side, and sets the priority of the new strategy data to be higher than the priority of the original strategy data, and participates in the subsequent dynamic utility evaluation and conflict arbitration as a conservative control strategy; when there is strategy data that meets the above conditions in the issued strategy data, the edge gateway selects the strategy data as a conservative control strategy to participate in the dynamic utility evaluation and conflict arbitration.

2. The factory intelligent management method based on the Internet of Things according to claim 1, characterized in that, The probabilistic semantic inference of the measurement point data based on multi-source context includes: the edge gateway acquiring the device communication characteristics of the industrial equipment and the spatiotemporal pattern characteristics of the measurement point data, and retrieving the policy execution feedback; generating the current multi-source context based on the device communication characteristics, spatiotemporal pattern characteristics and policy execution feedback; performing probabilistic semantic inference on the measurement point data based on the multi-source context to generate unified semantic labels and corresponding semantic confidence scores; and dynamically updating the device semantic model based on the semantic confidence scores.

3. The factory intelligent management method based on the Internet of Things according to claim 2, characterized in that, The dynamic updating of the device semantic model includes: when the semantic confidence meets the preset update conditions, the edge gateway writes the unified semantic label into the device semantic model to complete the update; when the semantic confidence does not meet the preset update conditions, the edge gateway keeps the device semantic model unchanged and performs probabilistic semantic inference based on the subsequently collected measurement point data to update the semantic confidence.

4. The factory intelligent management method based on the Internet of Things according to claim 1, characterized in that, The conflict arbitration includes: when the triggering conditions of multiple policy data are met, the edge gateway determines the set of industrial devices affected by each policy data based on the linkage action and device semantic model corresponding to each policy data, and when the industrial device sets of any two policy data have an intersection, the corresponding policy data is determined to be conflicting policy data; among the conflicting policy data, the edge gateway determines their comprehensive evaluation value based on their respective utility evaluation results and the utility evaluation results of other policy data that conflict with them, and selects the policy data with the largest comprehensive evaluation value as the candidate policy data to be executed; when the candidate policy data to be executed is not unique, it further compares the version number and priority of the candidate policy data to be executed to determine the policy to be executed.

5. The factory intelligent management method based on the Internet of Things according to claim 1, characterized in that, The event data consistency verification includes: during the replay upload of event data, the edge gateway checks whether there are missing, duplicate, or out-of-order sequence numbers in the event data uploaded to the cloud, based on the sequence number of the event data recorded when it is written to the local event cache queue; when a missing sequence number is detected, the event data corresponding to the missing sequence number is retrieved from the local event cache queue and re-uploaded; when a duplicate or out-of-order sequence number is detected, the redundant event data is discarded and the event data is rearranged according to the sequence number order.

6. The factory intelligent management method based on the Internet of Things according to claim 1, characterized in that, The calculation of the comprehensive credibility score includes: within a preset time window, the edge gateway performs statistical analysis on the measurement point data based on data quality indicators such as missing, out-of-bounds, abrupt changes, and data that remains unchanged within a preset time period to obtain a first credibility score; and obtains a second credibility score based on the semantic confidence of the corresponding unified semantic tag and the degree of deviation of the current measurement point data from the statistical characteristics of historical measurement point data of the unified semantic tag in the device semantic model; when the first credibility score is higher than a first threshold, the comprehensive credibility score is calculated according to a preset first combination rule; otherwise, the comprehensive credibility score is calculated according to a preset second combination rule; in the first combination rule, the contribution coefficient of the first credibility score is less than the contribution coefficient of the second credibility score, and in the second combination rule, the contribution coefficient of the first credibility score is greater than the contribution coefficient of the second credibility score.

7. A factory intelligent management system based on the Internet of Things, characterized in that, The system applies the IoT-based intelligent factory management method described in any one of claims 1 to 6, comprising: Measurement point acquisition module: Discovers and connects to industrial equipment through the edge gateway, determines the communication protocol based on the communication characteristics of the industrial equipment, and collects measurement point data; Semantic Model Module: Performs probabilistic semantic inference on measurement point data based on multi-source context, generates unified semantic labels with semantic confidence, and dynamically updates the device semantic model; Event encapsulation module: Encapsulates the measurement point data into event data containing event identifiers and sequence numbers, writes it to the local event cache queue, and stores it after deduplication; Policy delivery module: Generates and delivers policy data in the cloud, including version number, priority, trigger conditions, linkage actions and rollback actions; Policy execution module: After receiving policy data, the edge gateway performs dynamic utility evaluation and conflict arbitration on the policy data based on the current context composed of the device semantic model and real-time event data, determines the policy to be executed and executes the linkage action, and executes the rollback action when it fails. The offline self-governance module: When the connection between the edge gateway and the cloud is abnormal, it enters the offline self-governance mode and continuously caches event data. After the connection is restored, it replays and uploads the data according to the sequence number and performs event data consistency verification. Conservative Strategy Module: Calculates the overall credibility score. When the overall credibility score is lower than the threshold, the corresponding measurement point data is isolated. A conservative control strategy is generated or selected using redundant measurement point data or estimated data.

Images