Multi-protocol fusion intelligent gateway cooperative control method and system for industrial Internet of Things

Abstract

The invention relates to the technical field of industrial Internet of Things, and discloses a multi-protocol fusion intelligent gateway cooperative control method and system for the industrial Internet of Things, and the method comprises the steps: recognizing an industrial equipment communication protocol through a protocol reconnaissance module, and converting the industrial equipment communication protocol into intermediate data in a unified format; cleaning, integrating and recombining the intermediate data to generate a standardized data stream; performing first-stage early warning based on a static threshold value, calculating a correlation function value between parameters through multi-source collaborative analysis, and comparing the correlation function value with a normal coupling reference interval to trigger second-stage early warning; generating a pre-judgment type cooperative control instruction according to the early warning type and grade; and adaptively adjusting a communication strategy according to the network state and the early warning level, and reporting data. The system correspondingly comprises a protocol reconnaissance and adaptation module, a data management pipeline module, a multi-stage anomaly detection module, a pre-judgment type cooperative control module and a self-adaptive communication reporting module. According to the invention, the problems of industrial equipment protocol isomerism, weak data processing capability and insufficient system response real-time performance are solved.

Description

Technical Field

[0001] This application relates to the field of industrial Internet of Things (IoT) technology, and more specifically, to a collaborative control method and system for a multi-protocol fusion smart gateway for industrial IoT. Background Technology

[0002] In modern industrial production and management, Industrial Internet of Things (IIoT) technology is widely used to digitize and intelligentize production processes. The core of this technology lies in sensing the on-site situation through various devices and aggregating, processing, and controlling this information. However, in practical applications, industrial sites involve a wide variety of devices, often from different manufacturers and following different communication protocols, posing significant challenges to interoperability. Existing IoT gateways often struggle to handle these heterogeneous protocols, understanding only a few, resulting in numerous devices being unable to connect and forming isolated "protocol islands" that severely restrict the system's openness and scalability.

[0003] Furthermore, traditional IoT gateways have limited data processing capabilities, mostly performing only simple data transmission and lacking the ability to clean, integrate, analyze, and execute intelligent algorithms (such as fault diagnosis and early warning prediction) at the edge. This results in a large influx of raw data into the central control system, increasing its burden and affecting the speed of emergency response and the real-time nature of system actions. Simultaneously, the complex industrial environment places high demands on the stability and reliability of IoT gateways; data transmission errors and connection interruptions can affect the accuracy of system control. These issues collectively hinder the full potential of Industrial IoT technology in broader applications. Summary of the Invention

[0004] This application discloses a collaborative control method and system for multi-protocol fusion smart gateways for the Industrial Internet of Things (IIoT), aiming to solve technical problems such as heterogeneous device protocols, limited data processing capabilities, and insufficient real-time system response in the IIoT.

[0005] This invention is achieved through the following technical solution: Upon receiving an access request from an industrial device, the system identifies the communication protocol type of the industrial device through a protocol detection module, loads the corresponding protocol parser and data conversion rules, and converts the original protocol data of the industrial device into intermediate data in a unified format. Establish a data pipeline to clean, integrate, and restructure the intermediate data in the unified format to generate a standardized data stream; The system monitors the value and rate of change of individual data points in the standardized data stream in real time. When the value or rate of change exceeds a preset static threshold, a first-level warning is triggered. Simultaneously, multi-source collaborative analysis is performed on the standardized data stream. By calculating the correlation function values ​​between at least two key parameters from different industrial equipment, the dynamic coupling relationship is quantified in real time. The currently calculated correlation function values ​​or trends are compared with the normal coupling benchmark interval obtained by training or statistics based on historical normal data. The normal coupling benchmark interval is used to characterize the range of correlation strength between parameters under normal conditions. When the correlation function value continuously deviates from the benchmark interval to a first time threshold, or shows a trend of evolving into an abnormal domain, a second-level warning is triggered. Based on the preset linkage control strategy, and combined with the type and level of the first-level warning and / or the second-level warning, predictive control commands are sent to one or more corresponding actuators. Based on real-time monitoring of network load, communication link quality, and the current alert level, the system adaptively adjusts the size, sending frequency, and communication mode of data packets, and reports the alert information, control command execution records, and processed data to the upper-level control platform.

[0006] A multi-protocol fusion intelligent gateway collaborative control system for the Industrial Internet of Things, the system comprising: The protocol detection and adaptation module is used to identify industrial equipment protocols and complete the unified data conversion; The data governance pipeline module is used to clean, integrate, and reorganize data in a unified format. The multi-level anomaly detection module further includes a threshold comparison unit and a multi-source collaborative analysis unit, which are used to execute the detection logic of the first-level warning and the second-level warning, respectively. The predictive collaborative control module has a built-in knowledge base or rule engine for the linkage control strategy, which is used to generate and issue linkage control instructions based on the output of the multi-level anomaly detection module. The adaptive communication reporting module is equipped with a network status monitor and a reporting strategy scheduler, which are used to optimize the reporting strategy according to the network conditions and warning levels.

[0007] The present invention has the following beneficial effects: The method of this application effectively overcomes the shortcomings of existing industrial IoT gateways in terms of interconnection, data processing and real-time response through innovative technical solutions such as multi-protocol fusion, edge-side intelligent data processing, multi-level anomaly detection, predictive collaborative control and adaptive communication reporting, and provides a more efficient, stable and reliable solution for the digitalization and intelligentization of industrial production. Attached Figure Description

[0008] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0009] Figure 1 This is a schematic diagram of the method flow of the present invention. Detailed Implementation

[0010] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0011] To better understand the multi-protocol fusion smart gateway collaborative control method for the Industrial Internet of Things proposed in this application, it is necessary to explain some of the key terms involved.

[0012] The "protocol reconnaissance module" refers to a software or hardware unit integrated within a smart gateway. Its function is to automatically identify the communication protocol type of connected industrial equipment, such as Modbus, PROFIBUS, and OPC UA. This module typically achieves this by analyzing the handshake signals, data frame structures, or specific protocol identifiers sent by the device.

[0013] A "protocol parser" is a software component written for a specific communication protocol, responsible for decoding raw binary or text protocol data into structured data that can be understood by the gateway.

[0014] The “data transformation rules” define how to map and convert structured data output by different protocol parsers into a unified intermediate data format, such as JSON, XML, or Protobuf, for subsequent general processing.

[0015] A "data pipeline" is a logical concept that represents the entire process of data from receipt to processing and then to output, which includes a series of operations such as data cleaning, integration, and structured reorganization.

[0016] "Standardized data flow" refers to a data sequence that has been processed through a data pipeline, has a unified format, clear semantics, and a regular structure, and can be directly used for analysis and control.

[0017] The "static threshold" is a pre-set fixed value used to determine whether a single data point exceeds the normal range.

[0018] "Correlation function value" is a mathematical quantity that measures the degree of interdependence or influence between key parameters of different industrial equipment, such as correlation coefficient, mutual information, or Granger causality value.

[0019] The “normal coupling baseline interval” is obtained by statistical analysis or training based on historical normal operation data, and it characterizes the range of fluctuations in the correlation strength between parameters under normal conditions.

[0020] The "first time threshold" and "second time threshold" are time length settings used to determine the persistence of abnormal states.

[0021] The "linkage control strategy" is a set of preset rules that defines what control actions should be taken and which actuators will be affected under different warning types and levels.

[0022] An actuator is a device that receives control commands and performs physical operations, such as a valve, motor, or circuit breaker.

[0023] "Network load" refers to the amount of data transmitted and the bandwidth usage in a network.

[0024] "Communication link quality" measures the reliability and efficiency of data transmission, such as packet loss rate and latency.

[0025] Please see Figure 1 As shown, a collaborative control method for a multi-protocol fusion smart gateway for the Industrial Internet of Things includes the following steps: Upon receiving an access request from an industrial device, the system identifies the communication protocol type of the industrial device through a protocol detection module, loads the corresponding protocol parser and data conversion rules, and converts the original protocol data of the industrial device into intermediate data in a unified format. Establish a data pipeline to clean, integrate, and restructure the intermediate data in the unified format to generate a standardized data stream; The system monitors the value and rate of change of individual data points in the standardized data stream in real time. When the value or rate of change exceeds a preset static threshold, a first-level warning is triggered. Simultaneously, multi-source collaborative analysis is performed on the standardized data stream. By calculating the correlation function values ​​between at least two key parameters from different industrial equipment, the dynamic coupling relationship is quantified in real time. The currently calculated correlation function values ​​or trends are compared with the normal coupling benchmark interval obtained by training or statistics based on historical normal data. The normal coupling benchmark interval is used to characterize the range of correlation strength between parameters under normal conditions. When the correlation function value continuously deviates from the benchmark interval to a first time threshold, or shows a trend of evolving into an abnormal domain, a second-level warning is triggered. Based on the preset linkage control strategy, and combined with the type and level of the first-level warning and / or the second-level warning, predictive control commands are sent to one or more corresponding actuators. Based on real-time monitoring of network load, communication link quality, and the current alert level, the system adaptively adjusts the size, sending frequency, and communication mode of data packets, and reports the alert information, control command execution records, and processed data to the upper-level control platform.

[0026] As one implementation, upon receiving an industrial device's access request, the smart gateway first activates a protocol reconnaissance module. This module can identify the communication protocol type of the industrial device in various ways. For example, one implementation involves the protocol reconnaissance module analyzing the header information, port number, or specific protocol handshake sequence of the data packets sent during the device's first communication to infer the device's protocol type. Another implementation involves the gateway maintaining a database of known protocol features. When a device accesses the network, the protocol reconnaissance module performs pattern matching between the received data stream and the features in the database to determine the most matching protocol. Once the protocol type is identified, the gateway dynamically loads pre-stored protocol parsers and data conversion rules corresponding to that protocol type. For example, if the Modbus TCP protocol is identified, the Modbus TCP parser and rules for converting Modbus register data into a unified JSON format are loaded. These parsers are responsible for decoding the device's raw protocol data (e.g., binary data frames) into structured, readable intermediate data.

[0027] Secondly, after the data unification transformation, a data pipeline is established to perform in-depth processing on the intermediate data in a unified format. This data pipeline can be built using stream processing frameworks such as Apache Flink or Kafka Streams. Within the data pipeline, data cleaning is performed first, for example, by filtering out invalid, duplicate, or malformed data using set rules. Next, data integration is performed, merging logically related data from the same or different devices to form a more complete dataset.

[0028] Finally, the data is restructured, for example, by transforming flat data structures into nested, more semantically meaningful structures, or by aggregating time-series data at a specific granularity, ultimately generating a standardized data stream. For instance, raw data from temperature sensors, pressure sensors, and flow meters, after protocol parsing and standardized format conversion, may exist as independent JSON objects. The data pipeline integrates these into a standardized data stream containing fields such as "device ID," "timestamp," "temperature," "pressure," and "flow," and may perform unit conversions or fill in missing values.

[0029] Secondly, multi-level anomaly detection is performed on the generated standardized data stream. The first-level warning focuses on anomalies at individual data points. The value and rate of change of individual data points in the standardized data stream are monitored in real time. For example, a static threshold of 80°C can be set for a temperature sensor; when the real-time temperature exceeds 80°C, the first-level warning is triggered. Simultaneously, the rate of temperature change can also be monitored; for example, if the temperature rises by more than 10°C within one minute, the first-level warning is also triggered. The second-level warning focuses on multi-source collaborative anomalies. Multi-source collaborative analysis is performed on the standardized data stream, quantifying the dynamic coupling relationship in real time by calculating the correlation function values ​​between at least two key parameters from different industrial equipment. For example, the correlation coefficient between boiler fuel consumption and steam output can be calculated. The currently calculated correlation function value or trend is compared with a normal coupling benchmark interval trained or statistically obtained based on historical normal data. For example, if historical data shows that the correlation coefficient between fuel consumption and steam output is typically between 0.8 and 0.9 under normal circumstances, a second-level warning may be triggered if the real-time calculated correlation coefficient consistently falls below 0.7 or rises above 0.95 and reaches the first time threshold. Furthermore, a second-level warning may also be triggered if the correlation function value shows a trend towards the outlier range (e.g., a continuous decline but not yet completely outside the baseline range).

[0030] Next, based on the preset linkage control strategy and considering the type and level of the first-level and / or second-level warnings, predictive control commands are sent to one or more corresponding actuators. For example, if a first-level warning indicates that the temperature of a certain device is too high, the linkage control strategy may instruct the cooling fan to start or increase its speed. If a second-level warning indicates an abnormal coordination relationship between multiple devices, such as a mismatch between the flow rate of an upstream pump and the opening of a downstream valve, the linkage control strategy may instruct the adjustment of the pump speed or the valve opening to avoid potential malfunctions. These control commands are predictive and designed to intervene before a malfunction occurs, thereby preventing production interruptions or equipment damage.

[0031] Finally, based on real-time monitoring of network load, communication link quality, and the currently triggered warning level, the gateway adaptively adjusts the packet size, transmission frequency, and communication mode, and reports the warning information, control command execution records, and processed data to the upper-level control platform. For example, when network load is high or communication link quality deteriorates (e.g., packet loss rate increases), the gateway may reduce packet size, decrease data transmission frequency, or switch from TCP mode to UDP mode to ensure timely transmission of critical warnings and control commands. When a high-level warning is triggered, the gateway may prioritize reporting the warning information and increase the reporting frequency to ensure that the upper-level control platform can obtain critical information and make decisions in a timely manner.

[0032] In some embodiments described above in this application, a multi-source collaborative analysis of standardized data streams is proposed. This involves calculating the correlation function values ​​between at least two key parameters from different industrial devices, quantifying their dynamic coupling relationship in real time, and comparing them with a normal coupling benchmark interval to trigger a second-level early warning. Specifically, the steps of performing multi-source collaborative analysis on standardized data streams, calculating dynamic coupling relationships in real time, and triggering a second-level early warning can be further refined into the following process.

[0033] Multiple operating parameters from various industrial devices are collected in real time and synchronized in time to form a synchronization parameter set; Based on the set of synchronous parameters collected by the industrial equipment under historical normal operating conditions, a normal interaction pattern library reflecting the intrinsic correlation between multi-source parameters is established through at least one of mutual information calculation, Granger causality test or neural network model training. During real-time operation, the real-time interaction characteristics reflected by the current synchronization parameter set are calculated in the same way as those established in normal mode. Calculate the similarity or distance metric between the real-time interaction feature and the corresponding pattern in the normal interaction pattern library; if the metric value is continuously lower than a similarity threshold dynamically adjusted according to the current operating conditions and reaches a second time threshold, it is determined that there is a hidden collaboration anomaly, triggering the second level warning.

[0034] Specifically, real-time acquisition of multi-source operating parameters from multiple industrial devices synchronized over time refers to the synchronous acquisition of key operating parameters, such as temperature, pressure, flow rate, rotational speed, current, and voltage, from multiple industrial devices connected to a smart gateway. These parameters are assigned precise timestamps during acquisition to ensure temporal synchronization, thus forming a synchronized parameter set. This synchronized parameter set is the foundation for multi-source collaborative analysis, and its accuracy directly affects the subsequent anomaly detection results.

[0035] Based on the set of synchronous parameters collected from industrial equipment under historical normal operating conditions, a normal interaction pattern library reflecting the intrinsic correlations among multi-source parameters is established through at least one of the following methods: mutual information computation, Granger causality test, or neural network model training. Historical normal operating conditions refer to the stable operating period of industrial equipment under fault-free and abnormal conditions. Mutual information computation is used to quantify the degree of nonlinear dependence between two or more variables, capturing associations more complex than simple correlations. Granger causality test is used to determine whether one time series can predict another, thus revealing the causal relationship between parameters. Neural network model training, such as using autoencoders or Long Short-Term Memory (LSTM) networks, can learn and encode complex temporal patterns and interaction laws of multi-source parameters under normal conditions. The normal interaction pattern library established through these methods stores the baseline characteristics of the dynamic coupling relationships between various parameters of industrial equipment under healthy conditions.

[0036] During real-time operation, the same methods used to establish the normal mode are employed to calculate the real-time interaction characteristics reflected by the current synchronization parameter set. This means that during actual device operation, the system continuously collects the real-time synchronization parameter set and uses the same methods as those used when building the normal interaction mode library, such as mutual information calculation, Granger causality tests, or neural network models, to extract the real-time correlation characteristics between the current parameters. These real-time interaction characteristics reflect the dynamic coupling of various parameters under the current operating state of the device.

[0037] Further, the similarity or distance metric between the real-time interaction feature and the corresponding pattern in the normal interaction pattern library is calculated. The similarity or distance metric can be Euclidean distance, cosine similarity, Mahalanobis distance, or probability distance based on a specific model output, used to quantify the deviation of the real-time interaction feature from the baseline pattern in the normal pattern library. If this metric value continuously falls below a similarity threshold dynamically adjusted according to the current operating conditions, reaching a second time threshold, a hidden coordination anomaly is determined, triggering the second-level warning. The similarity threshold dynamically adjusted according to the current operating conditions can be adjusted in real time based on factors such as equipment load, production cycle time, and ambient temperature to adapt to normal fluctuations caused by changes in operating conditions. The second time threshold is used to avoid false alarms caused by instantaneous fluctuations, ensuring that only persistent deviations are considered abnormal. A hidden coordination anomaly refers to a situation where a single parameter may not exceed a static threshold, but the dynamic coupling relationship between multiple parameters has changed abnormally, indicating a potential fault or performance degradation.

[0038] In some embodiments described above, this application proposes a scheme to detect covert cooperative anomalies by establishing a normal interaction pattern library and comparing it with real-time interaction features using similarity or distance metrics. However, in actual industrial operation, equipment conditions are often dynamically changing. Simple similarity or distance metrics may be insufficient to accurately capture persistent, gradual deviations caused by multi-factor coupling. Especially when the macroscopic operating state changes, the preset normal pattern library may not be fully adapted to the current operating conditions, thus affecting the accuracy and robustness of anomaly detection. To address this, this application further proposes to optimize the steps of calculating real-time interaction features and comparing them with the normal interaction pattern library. By introducing macroscopic state parameters, calculating nonlinear coupling degree, and combining dynamic time warping techniques, cooperative anomalies under complex operating conditions can be identified more accurately.

[0039] In this regard, this application further proposes the following steps for calculating real-time interaction features and comparing them with a normal interaction mode library: Perform filtering and normalization preprocessing on the current synchronization parameter set; Introduce macroscopic state parameters that characterize the overall load or productivity of the equipment. Based on the current value of these macroscopic state parameters, interpolate or select modes in the normal interaction mode library to obtain an adaptive reference mode that matches the current macroscopic state. The nonlinear coupling degree of the preprocessed synchronization parameter set within the sliding time window is calculated using methods such as transfer entropy, correlation coefficient time-shift analysis, or recursion graph-based methods. The calculated real-time nonlinear coupling degree sequence is compared with the coupling degree benchmark interval corresponding to the adaptive reference mode by performing distance calculation or trend correlation analysis based on dynamic time warping (DTW) to detect whether there is a persistent, gradual deviation caused by multi-factor coupling.

[0040] Specifically, filtering and normalization preprocessing of the current synchronization parameter set refers to cleaning and standardizing the multi-source operating parameters acquired in real time before subsequent coupling degree calculations. Filtering removes sensor noise and transient interference; methods such as moving average filtering, Kalman filtering, or wavelet denoising can be used to improve data purity and reliability. Normalization preprocessing unifies parameters with different dimensions and numerical ranges to the same scale; for example, Z-score normalization or Min-Max normalization can be used. This eliminates the influence of dimensional differences on coupling degree calculation results, ensuring comparability between different parameters.

[0041] Furthermore, macroscopic state parameters characterizing the overall load or productivity of the equipment are introduced, such as the equipment's production speed, total power consumption, and product output. The purpose is to reflect the current overall operating status of the equipment. Based on the current value of these macroscopic state parameters, patterns in the normal interaction pattern library are interpolated or selected. For example, if the pattern library stores normal patterns under three typical operating conditions—low load, medium load, and high load—when the current macroscopic state parameter is between medium and high load, an adaptive reference pattern between the two can be generated through linear or nonlinear interpolation; alternatively, a pre-stored pattern closest to the current macroscopic state can be directly selected as a reference. Thus, an adaptive reference pattern matching the current macroscopic state can be obtained, aiming to ensure a high degree of consistency between the comparison benchmark and the actual operating conditions, thereby improving the accuracy of anomaly detection.

[0042] Among these methods, transfer entropy, correlation coefficient time-shift analysis, or recursion graph-based methods are employed to calculate the nonlinear coupling degree of the preprocessed synchronization parameter set within a sliding time window. Transfer entropy quantifies the directionality and intensity of information flow from one variable to another, and is particularly suitable for detecting nonlinear causal relationships. Correlation coefficient time-shift analysis reveals the dynamic correlation between parameters by calculating the correlation coefficient at different time lags. Recursion graph-based methods analyze the dynamic behavior and coupling characteristics of complex systems by visualizing and quantifying the reproducibility of the system's phase space trajectory. These methods can capture nonlinear and dynamic parameter coupling relationships that are difficult to detect using traditional linear methods, aiming to more comprehensively characterize the real-time interaction features between multi-source parameters. The use of a sliding time window allows the coupling degree calculation to reflect the instantaneous changes in the correlation between parameters, adapting to the dynamic characteristics of industrial processes.

[0043] In practical applications, the calculated real-time nonlinear coupling degree sequence is compared with the coupling degree benchmark interval corresponding to the adaptive reference mode using distance calculation or trend correlation analysis based on Dynamic Time Warping (DTW). The DTW algorithm can effectively handle nonlinear distortions and stretching of time series along the time axis. Even if there are temporal misalignments or velocity differences between the real-time sequence and the reference sequence, DTW can find the optimal alignment path and calculate the true distance, aiming to more robustly assess the similarity between two coupling degree sequences. Trend correlation analysis focuses on comparing whether the changing trends of the sequences are consistent, for example, by calculating the Pearson correlation coefficient or the Spearman rank correlation coefficient. These methods can detect whether there are persistent, gradual deviations caused by multi-factor coupling. For example, when the nonlinear coupling degree sequence of multiple parameters continuously deviates from the normal benchmark interval over a period of time, even if the numerical change of a single parameter is not significant, it may indicate the occurrence of a hidden anomaly.

[0044] In some embodiments described above in this application, multi-source collaborative analysis of standardized data streams is performed, and the results are compared with normal coupling benchmark intervals obtained through training or statistics based on historical normal data to detect dynamic coupling anomalies between devices. However, in actual industrial production environments, the operating modes of equipment are not static and may change significantly due to production task switching, process parameter adjustments, or equipment maintenance. If such changes in operating modes are not effectively identified and adapted to, the established normal interaction mode library may become mismatched with the current actual operating mode, leading to false alarms or missed alarms and reducing the accuracy and reliability of anomaly detection.

[0045] In response, this application further proposes a step for adaptive operation mode, which includes: Continuously monitor one or more indicator parameters that characterize the overall operating mode of the equipment; When the change in the indicated parameter exceeds the set range, the local mode relearning process is initiated. During the relearning process, a set of synchronization parameters for a sufficient duration is collected under the new mode. Using the same method as described above, the dynamic coupling reference mode under this mode is recalculated and established, and then added to or updated to the normal interaction mode library. The nonlinear coupling degree calculated in real time is compared with the updated pattern library to achieve anomaly detection under adaptive operation mode.

[0046] Specifically, the indicated parameters can be understood as key variables that reflect the current operating status of the equipment, production tasks, or process flow. For example, these parameters may include the product model of the production line, the production speed setpoint, the equipment load rate, the on / off status of key valves or pumps, and control commands issued by the host computer. By continuously monitoring these indicated parameters, potential changes in the equipment's operating mode can be effectively detected. The set range refers to the threshold or interval used to determine whether a significant change in the operating mode has occurred. When the value of the indicated parameter or its changing trend exceeds the preset set range, it indicates that the equipment may have entered a new operating mode, at which point a mode relearning process needs to be initiated.

[0047] The local pattern relearning process aims to re-establish or update the corresponding normal interaction pattern for a new operating mode. During the relearning process, a sufficient duration of synchronization parameter set generated by the device operating in the new mode needs to be collected. "Sufficient duration" here refers to the time length sufficient to fully capture the stable correlation characteristics between parameters in the new mode, ensuring the representativeness and robustness of the learned pattern. In this invention, methods such as mutual information computation, Granger causality testing, or neural network model training can be used to learn and establish a dynamic coupling reference pattern under the new mode from the newly collected synchronization parameter set. After establishment, the new dynamic coupling reference pattern will be added to the existing normal interaction pattern library, or updated if the pattern already exists but needs updating.

[0048] Therefore, during subsequent real-time operation, the system compares the real-time calculated nonlinear coupling degree with the updated pattern library. This means that when the device is in a specific operating mode, its nonlinear coupling degree will be compared with the corresponding dynamic coupling reference mode under that mode, thereby achieving anomaly detection under adaptive operating mode.

[0049] In some embodiments described above in this application, a scheme is proposed to continuously monitor indicator parameters representing the overall operating mode of the equipment and initiate a local mode relearning process when the change exceeds a set range in order to achieve adaptive adjustment of the operating mode. However, in actual industrial scenarios, the switching of operating modes may not always be instantaneous and singular, but may involve signals from multiple data sources, which may have conflicts, delays, or gradual changes. If only a single or simple threshold judgment is relied upon, it may lead to misjudgment or missed judgment of mode switching, thereby affecting the accuracy and timeliness of local mode relearning, and thus reducing the robustness and adaptability of the entire system's anomaly detection.

[0050] To address this, this application further proposes steps for monitoring operating mode indicator parameters and confirming mode switching, thereby improving the accuracy and reliability of mode switching determination. These steps include: Operating mode indication parameters are collected from multiple independent data sources, including equipment host computer control commands, key process parameter set values, and production line scheduling system signals. The indicator parameters from different data sources are fused and inferred based on voting or DS evidence theory to form preliminary operational mode inference results; In particular, linear or nonlinear regression extrapolation is performed on the indicator parameters that show a gradual trend, and consistency verification is performed with other instantaneous switching signals to form the gradual pattern inference results; A mode switching confidence assessment mechanism is established to quantify and fuse the uncertainty of the preliminary operation mode inference result and the gradual mode inference result, and output the final operation mode determination result with confidence. Only when the confidence is higher than the preset switching action threshold is the operation mode confirmed to have changed and the local mode relearning is triggered.

[0051] Specifically, collecting operating mode indication parameters from multiple independent data sources refers to obtaining comprehensive information about the equipment's operating mode by integrating different information sources, such as equipment control commands from the host computer, key process parameter setpoints, and production line scheduling system signals. These data sources each have different characteristics and reliability. For example, host computer control commands typically directly reflect the operator's intentions, while key process parameter setpoints reflect the actual goals of the production process, and production line scheduling system signals provide more macro-level production planning information. Multi-source acquisition avoids the limitations or malfunctions that may exist with a single data source, improving the comprehensiveness of pattern recognition.

[0052] This process involves fusing and inferring indication parameters from different data sources using either a voting method or the DS evidence theory to form preliminary operational pattern inferences. The aim is to comprehensively utilize multi-source information and resolve potential conflicts or uncertainties between different data sources. While the voting method can simply determine the pattern through majority decision, the DS evidence theory provides a more rigorous framework for handling uncertainty and conflicting evidence. It assigns a confidence interval to each possible operational pattern, resulting in a more robust preliminary inference.

[0053] In practical applications, linear or nonlinear regression extrapolation is performed on indicator parameters exhibiting gradual trends, and consistency verification is performed with other instantaneous switching signals to form gradual mode inference results. The purpose is to identify operating mode switches that do not occur instantaneously but evolve gradually. For example, during the process of equipment slowly increasing from low load to high load, its key parameters will show a gradual trend. Through regression extrapolation, future mode states can be predicted, and cross-validation with instantaneous switching signals can ensure accurate identification of gradual modes, avoiding missing early signs of mode switching due to simple threshold judgments.

[0054] Furthermore, a mode switching confidence assessment mechanism is established to quantify and fuse the uncertainty of the preliminary operating mode inference results and the gradual mode inference results, outputting a final operating mode determination result with confidence. This mechanism aims to evaluate the reliability of different inference results and fuse them to provide a quantified confidence level. Uncertainty quantification can employ methods such as fuzzy logic, probability theory, or DS evidence theory to model the uncertainty of each inference result. The fusion process then combines these uncertain results to generate a more convincing final determination. Only when the confidence level is higher than a preset switching action threshold is the operating mode change confirmed and the local mode relearning triggered. This ensures that the resource-intensive relearning process is only initiated when the mode switching has sufficiently high confidence, avoiding system instability or resource waste due to misjudgment.

[0055] Furthermore, the execution steps of the mode switching confidence assessment mechanism include: Receive multiple pattern inference results that may conflict and their source identifiers; According to the preset priority rules, which are pre-set based on the inherent reliability level and historical accuracy of the data source, the conflicting inference results are arbitrated to obtain the arbitration result; Based on the real-time signal-to-noise ratio, packet loss rate, and recent warning status related to the data source on which each inference result depends, the weights of each inference result in the fusion evaluation are dynamically calculated and assigned to obtain the weighted inference result. The arbitration result and the weighted inference result are then fused with information based on Bayesian inference or weighted average to generate the final mode switching confidence.

[0056] Specifically, receiving multiple potentially conflicting mode inference results and their source identifiers means that the system receives inference information about the current operating mode from multiple independent data sources, such as equipment host computer control commands, key process parameter settings, and production line scheduling system signals. Each inference result is accompanied by its source identifier for subsequent differentiation and processing. These inference results may indicate different operating modes or have different levels of confidence in the same mode, thus creating conflicts.

[0057] The arbitration of conflicting inferences based on preset priority rules, resulting in an arbitration outcome, means that upon receiving conflicting pattern inferences, the system makes a judgment according to pre-defined priority rules. These priority rules can be preset based on the inherent reliability level of the data source (e.g., host computer control commands typically have the highest priority) and historical accuracy (e.g., a sensor has historically performed more accurately under specific operating conditions). The purpose of arbitration is to select the most credible or logically consistent inference from multiple conflicting pieces of information as the preliminary arbitration result.

[0058] In practical applications, the weights of each inference result in the fusion evaluation are dynamically calculated and assigned based on the real-time signal-to-noise ratio, packet loss rate, and the most recent warning status associated with that data source from which the inference results depend, resulting in a weighted inference result. This means that the contribution of each data source is not fixed but dynamically adjusted according to its current real-time data quality and system status. For example, when the signal-to-noise ratio of a data source decreases or the packet loss rate increases, its weight will be adjusted accordingly; if a data source frequently triggers warnings recently, it may also indicate a decline in its data reliability, and its weight will also decrease. Through this dynamic weighting, the reliability of each data source at the current moment can be more accurately reflected.

[0059] Furthermore, the arbitration result and the weighted inference result are fused using Bayesian inference or weighted averaging to generate the final mode-switching confidence level. This means that after obtaining the arbitration result and the weighted inference result, the system employs advanced information fusion algorithms, such as Bayesian inference or weighted averaging, to integrate this information. Bayesian inference updates the posterior probability based on prior probabilities and observed evidence, thus more accurately quantifying uncertainty; the weighted averaging method performs simple and effective aggregation based on the weights of each inference result. Through this fusion, a comprehensive final mode-switching determination result with higher confidence can be generated.

[0060] This application further proposes a scheme for data consistency preprocessing before information fusion to ensure that the data participating in the fusion have high consistency, thereby improving the accuracy and reliability of mode switching confidence assessment.

[0061] Before information fusion, a data consistency preprocessing step is also included: Check the consistency of the data format, timestamp, and semantic expression between the arbitration result and the weighted inference result; If the data formats are inconsistent, the preset format conversion rules will be invoked to unify them; If timestamps are misaligned, time synchronization is performed using an interpolation alignment algorithm based on a high-precision clock source or a buffer waiting strategy. If semantic conflicts exist, different expressions are mapped to unified concept nodes and then parsed based on a predefined industrial IoT ontology library.

[0062] Specifically, data consistency preprocessing refers to a comprehensive check and correction performed on the arbitration results and weighted inference results before inputting them into the information fusion module, in order to eliminate potential data heterogeneity. The primary step is to check the consistency of data format, timestamps, and semantic representation, aiming to identify and locate potential mismatches in the data. When data format inconsistencies are detected, such as one data source providing data in JSON format while another provides it in XML format, or mismatched numeric types (e.g., integers and floating-point numbers), the system will invoke preset format conversion rules for unification. These rules can be pre-configured to support common industrial data format conversions, ensuring that all data maintains structural consistency for easier subsequent processing.

[0063] Furthermore, if timestamp misalignment is detected—that is, data from different data sources describe the same event or state but their recording times differ—this application employs an interpolation alignment algorithm based on a high-precision clock source or a buffer waiting strategy for time synchronization. The interpolation alignment algorithm aligns data points to a unified time axis by mathematically interpolating the time series data; the buffer waiting strategy allows data to wait briefly in a buffer until the timestamps of all relevant data points meet the synchronization requirements, thus ensuring that the fused data are precisely corresponding in the time dimension. In addition, if semantic conflicts exist, such as different devices or systems using different terms for the same concept (e.g., "motor speed" vs. "motor RPM"), this application uses a predefined industrial IoT ontology library to map different expressions to a unified concept node before parsing. The industrial IoT ontology library is a knowledge graph containing industrial terminology, concepts, and their interrelationships. By unifying heterogeneous semantic expressions into standard concepts within the ontology library, semantic ambiguity can be eliminated, ensuring the logical accuracy of information fusion.

[0064] The step of dynamically calculating and allocating weights further includes: Maintain a dynamic reliability score for each data source, which is updated based on its recent data quality metrics, historical failure frequency, and suitability record under the current specific operating mode; When the dynamic reliability score of a data source falls below the first threshold, the weight of its corresponding inference result is reduced linearly. When a data source provides consistent and ultimately confirmed correct inferences over multiple consecutive judgment periods, its weight is increased exponentially. Set a minimum guarantee weight greater than zero for each data source to ensure that its information can still participate in the fusion with a minimum, even when reliability is low.

[0065] Specifically, the dynamic reliability score can be understood as a quantitative assessment of the trustworthiness of each data source in providing pattern inference information. This score is not fixed but continuously updated based on the real-time performance and historical records of the data source. "Recent data quality indicators" may include, but are not limited to, the real-time signal-to-noise ratio of data packets, the integrity of data transmission (such as packet loss rate), data freshness, and the reasonableness of data values. For example, when the real-time signal-to-noise ratio of a data source decreases or the packet loss rate increases, its recent data quality indicators will deteriorate accordingly, thus affecting its dynamic reliability score. "Historical failure frequency" refers to the number or proportion of times the data source has provided erroneous or inconsistent inferences in the past, reflecting its long-term stability. "Applicability record under the current specific operating mode" considers the differences that different data sources may exhibit under different industrial operating modes; for example, some sensors may experience decreased accuracy in high-temperature modes but perform well in normal-temperature modes. By integrating these factors, the reliability of the data source can be comprehensively and dynamically assessed.

[0066] When the dynamic reliability score of a data source falls below a preset "first threshold," it indicates that the reliability of that data source has dropped to an unacceptable level. At this point, to prevent its low-quality information from negatively impacting the final fusion result, the weight of its corresponding inference result in the information fusion process will be "linearly reduced." This linear reduction mechanism makes the weight adjustment smooth and controllable, avoiding system oscillations that might result from a sudden drop in weight.

[0067] Conversely, when a data source consistently provides "consistent and ultimately confirmed correct inferences" across "multiple consecutive judgment periods," this indicates high stability and accuracy. To reward this sustained good performance and fully utilize its high-quality information, its weight will be "exponentially increased." This exponential increase allows reliable data sources to play a greater role in the fusion process more quickly, thereby accelerating the system's adoption of high-quality information.

[0068] Furthermore, to avoid the loss of potentially critical information due to a data source's extremely low reliability score causing its weight to drop to zero, this application sets a "minimum guaranteed weight greater than zero" for each data source. This means that even if a data source has a very low dynamic reliability score, its information can still participate in the fusion process with a minimum. The purpose is that even if a data source performs poorly most of the time, its weak signals may still have important reference value in certain specific or emergency situations and should not be completely ignored.

[0069] The information fusion step specifically includes: Analyze the chronological order, duration overlap, and statistical correlation of the various inference results in the time series to identify potential causal chains or collaborative change patterns among them; Based on the identified causal or collaborative relationships, a weighted directed graph is constructed to dynamically plan the path and order of fusion, and the initial weights calculated based on the reliability of the data source are fine-tuned according to the importance of the graph nodes. Following the planned fusion path and the fine-tuned weights, sequential Bayesian updates or neural network fusion machines are used to complete the final information fusion and confidence calculation.

[0070] Specifically, before performing information fusion, the received multiple pattern inference results are first analyzed in depth. This analysis aims to reveal the inherent connections between these inference results over time, including their chronological order, degree of overlap on the timeline, and statistical correlations. Through this analysis, potential causal chains between inference results can be identified—that is, a change in one inference result may lead to a change in another—or cooperative change patterns can be identified, meaning that multiple inference results tend to change simultaneously or in a specific pattern.

[0071] Identifying causal chains or collaborative change patterns is crucial for constructing fusion strategies. Based on these relationships, a weighted directed graph can be built. In this graph, each node can represent a pattern inference result or its source, edges represent causal or collaborative relationships between them, and edge weights reflect the strength or confidence level of these relationships. This weighted directed graph is used to dynamically plan the path and sequence of information fusion, ensuring that information is effectively integrated according to its inherent logical relationships. For example, if inference result A is the cause of inference result B, then during fusion, information from A might be processed first, and its influence might be passed on to B.

[0072] Furthermore, based on this weighted directed graph, the initial weights calculated according to the reliability of the data source can be fine-tuned. This fine-tuning is based on the importance of the graph nodes; for example, nodes that are upstream in a causal chain or have higher centrality may be assigned higher weights to their corresponding inference results to reflect their key role in the overall pattern determination.

[0073] Finally, following the planned fusion path and fine-tuned weights, sequential Bayesian update or a neural network fusion engine is used to complete the final information fusion and confidence calculation. Sequential Bayesian update allows information to be gradually incorporated, with each update based on the latest evidence and prior knowledge. Neural network fusion engines, on the other hand, can learn and capture the complex nonlinear relationships between inference results, thereby generating more accurate and robust mode-switching confidence scores.

[0074] In some preferred embodiments, it is assumed that when monitoring the equipment's operating mode, the system receives mode inference results from the equipment's host computer control commands, key process parameter setpoints, and production line scheduling system signals. For example, the host computer command may instruct the equipment to enter "high-speed production mode," while at the same time, the setpoint of a certain key process parameter also begins to adjust towards the range corresponding to "high-speed production," but the production line scheduling system signal may lag slightly.

[0075] At this point, the fusion method of this application first analyzes these inference results. It identifies that the host computer command is usually the direct cause of the mode switch, while the change in the process parameter setpoint is its result or coordination signal, and the production line scheduling signal may be a confirmation signal, but with a time delay. Based on this causal and timing relationship, a directed graph can be constructed: host computer command -> process parameter setpoint -> production line scheduling signal.

[0076] During the fusion process, the inference results from the host computer instructions may be assigned a higher initial weight, and this weight may be fine-tuned to be even higher based on its upstream position in the causal chain. When the process parameter settings also confirm the "high-speed production mode," the information enhances the overall confidence level due to its coordination with the host computer instructions. Even if there is a slight delay in the production line scheduling signal, the system can further confirm and increase the confidence level after receiving the scheduling signal through sequential Bayesian updates, rather than simply waiting for all signals to synchronize. If a signal does not conform to the causal chain or the coordination mode, its weight may be dynamically reduced to avoid interference from erroneous information in the final decision. In this way, the system can determine mode switching earlier and more accurately, and confirm it with higher confidence.

[0077] In order to effectively deploy the above methods in complex industrial environments, a specific, modular system architecture is needed to carry and execute these complex logics, so as to ensure smooth data flow, accurate real-time response, and overall system reliability.

[0078] To address this, this application proposes a multi-protocol fusion intelligent gateway collaborative control system for the Industrial Internet of Things (IIoT). This system is used to implement the above method and specifically includes: The protocol detection and adaptation module is used to identify industrial equipment protocols and complete the unified data conversion; The data governance pipeline module is used to clean, integrate, and reorganize data in a unified format. The multi-level anomaly detection module further includes a threshold comparison unit and a multi-source collaborative analysis unit, which are used to execute the detection logic of the first-level warning and the second-level warning, respectively. The predictive collaborative control module has a built-in knowledge base or rule engine for the linkage control strategy, which is used to generate and issue linkage control instructions based on the output of the multi-level anomaly detection module. The adaptive communication reporting module is equipped with a network status monitor and a reporting strategy scheduler, which are used to optimize the reporting strategy according to the network conditions and warning levels.

[0079] The proposed solution breaks down the aforementioned methods into multiple functionally independent modules, achieving a clear system architecture and specialized functions. The protocol reconnaissance and adaptation module ensures seamless access and standardization of heterogeneous industrial equipment data, laying the foundation for subsequent processing. The data governance pipeline module performs high-quality preprocessing of the data, guaranteeing the accuracy and usability of the data input to the anomaly detection module. The multi-level anomaly detection module, through its threshold comparison unit and multi-source collaborative analysis unit, can comprehensively and accurately identify single-parameter anomalies and multi-parameter collaborative anomalies, thereby improving the sensitivity and accuracy of anomaly detection. The predictive collaborative control module transforms detected anomalies into specific control actions, realizing closed-loop management from monitoring to intervention. The adaptive communication reporting module ensures that critical information can be uploaded to the upper-level control platform in a timely and reliable manner under different network conditions, maintaining the overall communication efficiency and stability of the system. Thus, the entire system forms a complete chain from data acquisition, processing, analysis, decision-making to execution and reporting, effectively supporting the comprehensive implementation of the aforementioned methods.

[0080] Through the above technical solution, this application provides a system with a clear structure and complete functions, capable of efficiently implementing a multi-protocol fusion intelligent gateway collaborative control method for the Industrial Internet of Things (IIoT). The system's modular design significantly improves its maintainability, scalability, and robustness, making it possible to deploy and manage this method in complex and ever-changing industrial environments. Furthermore, the specialized division of labor among the modules helps optimize resource utilization and reduce internal coupling, thereby improving the overall system's operating efficiency and response speed. In particular, by encapsulating complex anomaly detection logic and control strategies in dedicated modules, the system can more flexibly adapt to different industrial scenarios and business needs, providing solid technical support for the intelligentization and automation of industrial production.

[0081] In summary, the multi-protocol fusion smart gateway collaborative control method for the Industrial Internet of Things disclosed in this application solves the problems of heterogeneous device protocols, limited data processing capabilities, and insufficient real-time system response in the prior art through the following means, and achieves significant and excellent technical results: First, the protocol reconnaissance module identifies the communication protocol type of industrial equipment and loads the corresponding protocol parser and data conversion rules, converting heterogeneous raw protocol data into intermediate data in a unified format. This effectively solves the "protocol island" problem caused by the wide variety of industrial field equipment and protocol incompatibility, greatly improving the openness and scalability of the industrial Internet of Things system.

[0082] Secondly, a data pipeline is established to clean, integrate, and restructure intermediate data in a unified format, generating a standardized data stream. Real-time monitoring and multi-source collaborative analysis are then performed at the edge. This edge-side data processing capability reduces the burden on the central control system and improves data processing efficiency. In particular, a multi-level anomaly detection mechanism is constructed by triggering a first-level early warning through real-time monitoring of the value and rate of change of individual data points, and by triggering a second-level early warning by calculating the correlation function values ​​between key parameters from multiple sources and comparing them with a normal coupling benchmark interval. This mechanism can not only detect anomalies in single parameters but also identify hidden collaborative anomalies and gradual anomalies that are difficult to detect using traditional methods, significantly improving the sensitivity and accuracy of anomaly detection.

[0083] Furthermore, based on the preset linkage control strategy and combined with the warning type and level, predictive control commands are sent to the actuators. This predictive control enables early intervention against potential risks, preventing further escalation of faults, thereby improving the real-time response and control accuracy of the system.

[0084] Finally, based on real-time monitoring of network load, communication link quality, and early warning level, the system adaptively adjusts the size, transmission frequency, and communication mode of data packets, and reports the early warning information, control command execution records, and processed data to the upper-level control platform. This adaptive communication reporting strategy optimizes network resource utilization, ensures timely and reliable transmission of critical information, and avoids a large amount of raw data flooding the upper-level platform, further reducing the burden on the central system.

[0085] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0086] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A collaborative control method for a multi-protocol fusion smart gateway for the Industrial Internet of Things, characterized in that, include: Upon receiving an access request from an industrial device, the system identifies the communication protocol type of the industrial device through a protocol detection module, loads the corresponding protocol parser and data conversion rules, and converts the original protocol data of the industrial device into intermediate data in a unified format. Establish a data pipeline to clean, integrate, and restructure the intermediate data in the unified format to generate a standardized data stream; The system monitors the value and rate of change of individual data points in the standardized data stream in real time. When the value or rate of change exceeds a preset static threshold, a first-level warning is triggered. Simultaneously, multi-source collaborative analysis is performed on the standardized data stream, and the dynamic coupling relationship is quantified in real time by calculating the correlation function values ​​between at least two key parameters from different industrial equipment. Compare the currently calculated correlation function value or trend with the normal coupling benchmark interval obtained based on training or statistics of historical normal data; The normal coupling reference interval is used to characterize the range of correlation strength between parameters under normal conditions; when the correlation function value continuously deviates from the reference interval to the first time threshold, or shows a trend of evolving into the abnormal domain, a second-level warning is triggered. Based on the preset linkage control strategy, and combined with the type and level of the first-level warning and / or the second-level warning, predictive control commands are sent to one or more corresponding actuators. Based on real-time monitoring of network load, communication link quality, and the current alert level, the system adaptively adjusts the size, sending frequency, and communication mode of data packets, and reports the alert information, control command execution records, and processed data to the upper-level control platform.

2. The multi-protocol fusion intelligent gateway collaborative control method for industrial IoT according to claim 1, characterized in that, The steps of performing multi-source collaborative analysis on standardized data streams, calculating dynamic coupling relationships in real time, and triggering a second-level early warning include: Multiple operating parameters from various industrial devices are collected in real time and synchronized in time to form a synchronization parameter set; Based on the set of synchronous parameters collected by the industrial equipment under historical normal operating conditions, a normal interaction pattern library reflecting the intrinsic correlation between multi-source parameters is established through at least one of mutual information calculation, Granger causality test or neural network model training. During real-time operation, the real-time interaction characteristics reflected by the current synchronization parameter set are calculated in the same way as those established in normal mode. Calculate the similarity or distance metric between the real-time interaction feature and the corresponding pattern in the normal interaction pattern library; if the metric value is continuously lower than a similarity threshold dynamically adjusted according to the current operating conditions and reaches a second time threshold, it is determined that there is a hidden collaboration anomaly, triggering the second level warning.

3. The multi-protocol fusion intelligent gateway collaborative control method for industrial IoT according to claim 2, characterized in that, The steps of calculating real-time interaction features and comparing them with a normal interaction pattern library include: Perform filtering and normalization preprocessing on the current synchronization parameter set; Introduce macroscopic state parameters that characterize the overall load or productivity of the equipment. Based on the current value of these macroscopic state parameters, interpolate or select modes in the normal interaction mode library to obtain an adaptive reference mode that matches the current macroscopic state. The nonlinear coupling degree of the preprocessed synchronization parameter set within the sliding time window is calculated using methods such as transfer entropy, correlation coefficient time-shift analysis, or recursion graph-based methods. The calculated real-time nonlinear coupling degree sequence is compared with the coupling degree benchmark interval corresponding to the adaptive reference mode by performing distance calculation or trend correlation analysis based on dynamic time warping (DTW) to detect whether there is a persistent, gradual deviation caused by multi-factor coupling.

4. The multi-protocol fusion intelligent gateway collaborative control method for industrial IoT according to claim 3, characterized in that, The method also includes a step of adaptive operating mode: Continuously monitor one or more indicator parameters that characterize the overall operating mode of the equipment; When the change in the indicated parameter exceeds the set range, the local mode relearning process is initiated. During the relearning process, a set of synchronization parameters for a sufficient duration is collected under the new mode, and the dynamic coupling reference mode under the new mode is recalculated and established in the same manner as in claim 2, and added to or updated in the normal interaction mode library. The nonlinear coupling degree calculated in real time is compared with the updated pattern library to achieve anomaly detection under adaptive operation mode.

5. A multi-protocol fusion intelligent gateway collaborative control method for industrial IoT according to claim 4, characterized in that, The steps for monitoring the operating mode indicator parameters and confirming the mode switch include: Operating mode indication parameters are collected from multiple independent data sources, including equipment host computer control commands, key process parameter set values, and production line scheduling system signals. The indicator parameters from different data sources are fused and inferred based on voting or DS evidence theory to form preliminary operational mode inference results; Linear or nonlinear regression extrapolation is performed on the indicator parameters that show a gradual trend, and consistency verification is performed with other instantaneous switching signals to form the gradual pattern inference results; A mode switching confidence assessment mechanism is established to quantify and fuse the uncertainty of the preliminary operation mode inference result and the gradual mode inference result, and output the final operation mode determination result with confidence. Only when the confidence is higher than the preset switching action threshold is the operation mode confirmed to have changed and the local mode relearning is triggered.

6. The multi-protocol fusion intelligent gateway collaborative control method for industrial IoT according to claim 5, characterized in that, The execution steps of the mode switching confidence assessment mechanism include: Receive inference results from multiple modes that may conflict, along with their source identifiers; According to the preset priority rules, which are pre-set based on the inherent reliability level and historical accuracy of the data source, the conflicting inference results are arbitrated to obtain the arbitration result; Based on the real-time signal-to-noise ratio, packet loss rate, and the most recent warning status related to the data source on which each inference result depends, the weights of each inference result in the fusion evaluation are dynamically calculated and assigned to obtain the weighted inference result. The arbitration result and the weighted inference result are fused using information based on Bayesian inference or weighted average to generate the final mode switching confidence score.

7. A multi-protocol fusion intelligent gateway collaborative control method for industrial IoT according to claim 6, characterized in that, Before information fusion, a data consistency preprocessing step is also included: Check the consistency of the data format, timestamp, and semantic expression between the arbitration result and the weighted inference result; If the data formats are inconsistent, the preset format conversion rules will be invoked to unify them; If timestamps are misaligned, time synchronization is performed using an interpolation alignment algorithm based on a high-precision clock source or a buffer waiting strategy. If semantic conflicts exist, different expressions are mapped to unified concept nodes and then parsed based on a predefined industrial IoT ontology library.

8. A multi-protocol fusion intelligent gateway collaborative control method for industrial IoT according to claim 7, characterized in that, The step of dynamically calculating and allocating weights further includes: Maintain a dynamic reliability score for each data source, which is updated based on its recent data quality metrics, historical failure frequency, and suitability record under the current specific operating mode; When the dynamic reliability score of a data source falls below the first threshold, the weight of its corresponding inference result is reduced linearly. When a data source provides consistent and ultimately confirmed correct inferences over multiple consecutive judgment periods, its weight is increased exponentially. Set a minimum guarantee weight greater than zero for each data source to ensure that its information can still participate in the fusion with a minimum, even when reliability is low.

9. A multi-protocol fusion intelligent gateway collaborative control method for industrial IoT according to claim 8, characterized in that, The information fusion step specifically includes: Analyze the chronological order, duration overlap, and statistical correlation of the various inference results in the time series to identify potential causal chains or collaborative change patterns among them; Based on the identified causal or collaborative relationships, a weighted directed graph is constructed to dynamically plan the path and order of fusion, and the initial weights calculated based on the reliability of the data source are fine-tuned according to the importance of the graph nodes. Following the planned fusion path and the fine-tuned weights, sequential Bayesian updates or neural network fusion machines are used to complete the final information fusion and confidence calculation.

10. A multi-protocol fusion intelligent gateway collaborative control system for the Industrial Internet of Things, characterized in that, A collaborative control method for a multi-protocol fusion smart gateway for industrial IoT, as described in any one of claims 1 to 9, comprises: The protocol detection and adaptation module is used to identify industrial equipment protocols and complete the unified data conversion; The data governance pipeline module is used to clean, integrate, and reorganize data in a unified format. The multi-level anomaly detection module further includes a threshold comparison unit and a multi-source collaborative analysis unit, which are used to execute the detection logic of the first-level warning and the second-level warning, respectively. The predictive collaborative control module has a built-in knowledge base or rule engine for the linkage control strategy, which is used to generate and issue linkage control instructions based on the output of the multi-level anomaly detection module. The adaptive communication reporting module is equipped with a network status monitor and a reporting strategy scheduler, which are used to optimize the reporting strategy according to the network conditions and warning levels.

Images