An Industrial Internet of Things (IIoT) Intelligent Decision-Making Method, System, and Equipment Based on a Large Model

Abstract

This invention belongs to the field of Industrial Internet of Things (IIoT) technology, specifically relating to an IIoT intelligent decision-making method, system, and device based on a large model. The method includes: generating an adaptive dynamic threshold using a residual compensation algorithm based on real-time device data, operating parameters, and historical data; calculating anomaly scores based on the dynamic threshold and real-time device data; inputting the anomaly scores and the structured context of multimodal data into a large model to generate decision instructions with confidence assessment, and recording audit logs; the decision instructions include operational actions, operational parameters, and constraints; when the confidence level of the decision instruction reaches a confidence threshold or the anomaly score exceeds a set threshold, triggering automated operation, and performing closed-loop feedback based on the execution effect of the automated operation to dynamically update the confidence level, confidence threshold, residual compensation coefficient, and anomaly score weight. Through the residual compensation algorithm and adaptive dynamic threshold adjustment, the system's false alarm rate is reduced.

Description

Technical Field

[0001] This invention belongs to the field of industrial Internet of Things (IoT) technology, specifically relating to an industrial IoT intelligent decision-making method, system, and device based on a large model. Background Technology

[0002] In the field of Industrial Internet of Things (IIoT), alarm mechanisms for industrial equipment generally rely on fixed thresholds. This static setting method cannot be dynamically adjusted according to changes in actual operating conditions. For example, under different load rates, ambient temperatures, and operating durations, the range of parameters for normal operation of equipment will vary. However, fixed thresholds cannot adapt to these changes, leading to reduced alarm accuracy and the possibility of false alarms or missed alarms, which affects the normal operation of the equipment and production efficiency.

[0003] With the rapid development of artificial intelligence and big data modeling technology, many enterprises are actively exploring intelligent decision-making solutions. However, in industrial applications, current big data modeling technology is mostly limited to data analysis and has failed to form an effective closed loop with automated operations. While big data models can perform in-depth analysis of large amounts of industrial data, the analysis results often cannot be directly translated into specific automated operation instructions, leading to a disconnect between decision-making and execution. Furthermore, the lack of standardized big data modeling instruction interaction templates makes it difficult for different systems to conduct effective information exchange and collaborative work, further hindering the widespread application of intelligent decision-making technology in the Industrial Internet of Things (IIoT) field. Summary of the Invention

[0004] In view of the above-mentioned shortcomings of the prior art, the present invention provides an industrial Internet of Things intelligent decision-making method, system and device based on a large model to solve the above-mentioned technical problems.

[0005] In a first aspect, the present invention provides an intelligent decision-making method for the Industrial Internet of Things based on a large model, comprising:

[0006] S1. Based on real-time equipment data, operating parameters, and historical data, an adaptive dynamic threshold is generated using a residual compensation algorithm. Real-time equipment data includes current, vibration amplitude, and temperature. Operating parameters include load rate, ambient temperature, and runtime. Historical data consists of normal / abnormal time-series data from the past set time period.

[0007] S2. Based on dynamic thresholds and real-time device data, calculate an anomaly score, which is a weighted sum of the proportion of single-parameter exceedances, the correlation of multiple-parameter anomalies, and historical similarity multiplied by sensitivity.

[0008] S3. Input real-time equipment data, anomaly scores, and multimodal structured context into the large model to generate decision instructions with confidence assessment and record audit logs; the decision instructions include operation actions, operation parameters, and constraints; the structured context includes maintenance records, production plans, spare parts inventory status, and safe operating procedures.

[0009] S4. When the confidence level of the decision instruction reaches the confidence level threshold, an automated operation is triggered, and closed-loop feedback is performed based on the execution effect of the automated operation to dynamically update the confidence level, confidence level threshold, residual compensation coefficient, and anomaly score weight. When the anomaly score exceeds the set threshold but the confidence level does not reach the confidence level threshold, an early warning notification is triggered.

[0010] Based on real-time equipment data, operating parameters, and historical data, an adaptive dynamic threshold is generated using a residual compensation algorithm. This algorithm fully considers the equipment's operating characteristics under different load rates, ambient temperatures, and operating durations, allowing the threshold to dynamically adjust with changing conditions. This effectively avoids false alarms or missed alarms caused by traditional fixed thresholds failing to adapt to changes in operating conditions, thus improving the accuracy of equipment alarms. By calculating anomaly scores, comprehensively considering the proportion of single parameter exceeding limits, multi-parameter correlation anomalies, and historical similarity, and multiplying by sensitivity, the algorithm can comprehensively and accurately assess the degree of equipment anomalies. It not only focuses on whether a single parameter exceeds the limit but also considers the correlation between parameters and the similarity to historical faults, providing a more reliable basis for subsequent decision-making.

[0011] By inputting anomaly scores and structured contextualized multimodal data into a large model, decision instructions with confidence assessments are generated and audit logs are recorded. When conditions are met, automated operations are triggered, and closed-loop feedback is provided based on the execution results, dynamically updating relevant parameters. This closed-loop mechanism tightly integrates decision-making and execution, achieving end-to-end automation from data acquisition and analysis to decision execution, thus improving the intelligence level and response speed of industrial IoT systems.

[0012] As a further limitation of the technical solution of the present invention, the dynamic threshold includes an upper limit and a lower limit, wherein:

[0013]

[0014]

[0015] This is the load compensation coefficient. This is the environmental compensation coefficient. This refers to three standard deviations from the baseline threshold. Frequency band energy weight.

[0016] The formula incorporates parameters such as load compensation coefficient, environmental compensation coefficient, three times the standard deviation of the threshold baseline, and frequency band energy weight. It comprehensively considers the influence of various factors such as load, environment, historical data fluctuations, and vibration signal characteristics on the threshold, making the generated dynamic threshold more scientific and reasonable, and better able to adapt to complex industrial environments.

[0017] As a further limitation of the technical solution of the present invention, the calculation steps of the baseline value are as follows:

[0018] Obtain the rated power of the equipment and real-time power And calculate the load rate;

[0019] Outlier values ​​are removed from the current, vibration amplitude, and temperature data collected by the equipment under normal conditions, excluding data collected during equipment maintenance or when sensors malfunction.

[0020] The data is divided into different operating condition segments based on the load rate. For the normal data within each operating condition segment, a time window is set, and the arithmetic mean of the data within the time window is calculated to obtain the segmental baseline value. , among which, the k The segment baseline value for each working condition segment is ;

[0021] The global baseline value, i.e., the baseline value, is obtained by weighted summation of the baseline values ​​of different segments. ;

[0022] in, , , , ;

[0023] In the formula, N The number of data points within the window. This is the real-time data of the normal devices within the window. M The total number of operating condition segments. For the current working conditions, the first k The weight of each working condition segment r The current load rate, For the first k The average load rate of each operating condition segment This is the sharpening factor.

[0024] This series of steps fully considers the equipment's operation under different working conditions, eliminates the influence of abnormal data and irrelevant factors, and makes the baseline values ​​more accurate and reliable, providing a solid foundation for the generation of dynamic thresholds. Dividing the working conditions into segments according to the load rate and calculating the baseline values ​​separately can better adapt to the equipment's operating characteristics under different load conditions, enabling the dynamic thresholds to more accurately reflect the normal operating range of the equipment under different working conditions and improving the accuracy of anomaly detection.

[0025] As a further limitation of the technical solution of the present invention, the method for calculating frequency band energy includes:

[0026] Vibration signals After applying the Hanning window, an FFT transform is performed to obtain the spectrum. ;

[0027] The frequency band energy is obtained by extracting the spectrum of the target frequency band and calculating the energy proportion of the target frequency band; where, the target frequency band ;

[0028] .

[0029] By applying a Hanning window to the vibration signal and performing an FFT transform to obtain the spectrum, the energy percentage of the target frequency band is calculated, yielding the band energy. This method effectively extracts energy information from specific frequency bands in the vibration signal, aiding in the analysis of equipment vibration characteristics and providing crucial reference for anomaly detection and fault diagnosis. The calculation of band energy considers the frequency characteristics of the vibration signal, providing a more comprehensive reflection of the equipment's operating status. Compared to traditional detection methods based solely on time-domain parameters, it can more accurately identify potential equipment faults, improving the sensitivity and accuracy of anomaly detection.

[0030] As a further limitation of the technical solution of the present invention, the steps for calculating the anomaly score based on dynamic thresholds and real-time device data include:

[0031] For each parameter, count the number of times it exceeds the dynamic threshold within the sliding time window; use the proportion of these exceeding times to the total number of sampling points as the exceedance ratio for that parameter; calculate the arithmetic mean of the exceedance ratios for all parameters to obtain the single-parameter exceedance ratio. The parameters include temperature, vibration amplitude, and current.

[0032] Calculate the Pearson correlation coefficient matrix among the parameters; take the mean of the absolute values ​​of the off-diagonal elements as the synchronicity score. Use 1 minus the synchronicity score as the associated outlier. ;

[0033]

[0034] ,

[0035] In the formula, n is the number of monitoring parameters. These are the elements of the correlation coefficient matrix. For parameters and covariance, The standard deviation is used; the time series data within the sliding time window, with each column representing a parameter and each row representing the sampled values ​​at the same time point, constructs a parameter data matrix X.

[0036] Extract the multidimensional feature vector of the current device; calculate its normalized Euclidean distance with the historical fault feature vector; use 1 minus the normalized distance as the historical similarity;

[0037] Extract the current device feature vector (Including average temperature, vibration frequency energy, current fluctuation range, etc.), retrieve the most similar fault feature vector from the historical fault database. ;

[0038] Calculate the normalized Euclidean distance

[0039] Historical similarity

[0040] These are the maximum and minimum value vectors of all feature vectors in the historical fault feature database, respectively.

[0041] The anomaly score is obtained by weighting the proportion of single-parameter out-of-range values, the correlation of multiple parameters and historical similarity, and multiplying it by the sensitivity.

[0042] The formula for calculating the anomaly score is as follows:

[0043]

[0044] In the formula, As a weighting factor, The sensitivity is dynamically adjusted based on equipment operating time and environmental conditions.

[0045] By calculating the proportion of single-parameter outliers, the correlation of multiple parameters, and historical similarity, and then comprehensively considering their weighted sum and multiplied by sensitivity, an anomaly score is obtained. This allows for a comprehensive assessment of the equipment's anomaly severity from multiple perspectives. It not only focuses on the anomalies of individual parameters but also considers the correlation between parameters and their similarity to historical faults, making the anomaly score more accurate and reliable. The sensitivity is dynamically adjusted based on equipment operating time and environmental conditions, adapting to changes in different operating stages and environments. This enables the anomaly score to more accurately reflect the actual anomaly situation of the equipment, improving the adaptability and accuracy of anomaly detection.

[0046] As a further limitation of the technical solution of the present invention, the generation of large model decision instructions in S3 includes:

[0047] It receives anomaly scores, real-time equipment data, maintenance records, production plans, and spare parts inventory status, and outputs decision options with confidence levels, including operation actions, confidence levels, operation parameters, and constraints. The confidence level threshold is initialized based on the equipment type and dynamically updated according to the execution effect of automated operations.

[0048] The large model takes into account structured contextual data, including maintenance records, production plans, spare parts inventory status, and safe operating procedures. It comprehensively considers multiple factors, providing a complete basis for generating decision-making instructions and making decisions more scientific and rational. The output includes decision options with confidence levels, such as operational actions, confidence levels, operational parameters, and constraints. This provides operators with clearer and more reliable decision-making suggestions. Furthermore, the confidence threshold is initialized based on equipment type and dynamically updated according to execution results, continuously improving the accuracy and reliability of decision-making.

[0049] As a further limitation of the technical solution of the present invention, the content of the automated operation includes: triggering one of the following automated operations according to the type of decision instruction:

[0050] Emergency shutdown, reduced load operation, automatic generation of maintenance work orders and push to the MES system, integration with the ERP system to reserve spare parts inventory, and sending early warning notifications to remind responsible persons.

[0051] Different automated operations are triggered based on the type of decision command, such as emergency shutdown, load reduction operation, automatic generation of maintenance work orders and push to the MES system, integration with the ERP system to reserve spare parts inventory, and sending early warning notifications to remind responsible persons. This enables corresponding measures to be taken according to different equipment anomalies, improving the automation level and emergency response capabilities of the industrial IoT system. The implementation of automated operations allows for rapid response to equipment anomalies, reducing the time and error of manual intervention, improving production efficiency, and simultaneously taking timely measures to prevent further escalation of equipment failures, thus ensuring production safety.

[0052] As a further limitation of the technical solution of the present invention, in S4, the step of dynamically updating the instruction parameters based on the execution effect of the automated operation after the automated operation is triggered includes:

[0053] Collect actual operating status data of the equipment after executing decision commands, including equipment fault repair time, downtime, maintenance costs, and operation success rate;

[0054] The confidence level of a single decision instruction in the large model is updated based on the success rate of the last N operations; whereby... ; All are constants;

[0055] The performance evaluation index of the closed-loop feedback mechanism is calculated based on the collected data; the performance evaluation index is a reward function.

[0056]

[0057] in,

[0058]

[0059] In the formula, For MTBF improvement rate, To achieve cost savings, This represents the number of erroneous operations. These are the weighting coefficients;

[0060] like If the score falls below the set threshold for T consecutive times, the confidence threshold, residual compensation coefficient, and anomaly score weight of the large model decision-making process will be dynamically adjusted.

[0061] Data on the actual operating status of equipment after executing decision commands is collected, including equipment fault repair time, downtime, maintenance costs, and operation success rate. Based on this data, an evaluation index for the execution effect of the closed-loop feedback mechanism is calculated. This objectively assesses the execution effect of automated operations and provides a basis for subsequent parameter adjustments. The confidence level of a single decision command in the large model is updated using a dynamic confidence update strategy based on the success rate of the last N operations. If the confidence level falls below a set threshold for T consecutive operations, the confidence threshold, residual compensation coefficient, and anomaly scoring weight of the large model's decision-making process are dynamically adjusted. This closed-loop feedback mechanism continuously optimizes the decision-making process, improves the accuracy and reliability of decisions, and enables the industrial IoT system to better adapt to the ever-changing industrial environment.

[0062] Secondly, the present invention also provides an intelligent decision-making method for the Industrial Internet of Things based on a large model, comprising:

[0063] The dynamic threshold calculation module is used to generate adaptive dynamic thresholds based on real-time equipment data, operating parameters, and historical data using a residual compensation algorithm. The real-time equipment data includes current, vibration amplitude, and temperature; the operating parameters include load rate, ambient temperature, and runtime; and the historical data consists of normal / abnormal time-series data from the past set time period.

[0064] An anomaly detection and scoring module is used to calculate an anomaly score based on dynamic thresholds and real-time device data. The anomaly score is a weighted sum of the proportion of single-parameter exceedances, the correlation of multiple-parameter anomalies, and historical similarity, multiplied by sensitivity.

[0065] The large model decision instruction generation module is used to input real-time equipment data, anomaly scores, and multimodal structured context into the large model, generate decision instructions with confidence assessment, and record audit logs; the decision instructions include operation actions, operation parameters, and constraints; the structured context includes maintenance records, production plans, spare parts inventory status, and safety operation procedures;

[0066] The automated execution engine module is used to trigger automated operations when the confidence level of a decision instruction reaches the confidence level threshold, and to provide closed-loop feedback based on the execution effect of the automated operations, dynamically updating the confidence level, confidence level threshold, residual compensation coefficient, and anomaly score weight; when the anomaly score exceeds the set threshold but the confidence level does not reach the confidence level threshold, an early warning notification is triggered.

[0067] As a further limitation of the technical solution of the present invention, the dynamic threshold includes an upper limit and a lower limit, wherein:

[0068]

[0069]

[0070] This is the load compensation coefficient. This is the environmental compensation coefficient. This refers to three standard deviations from the baseline threshold. Frequency band energy weight.

[0071] As a further limitation of the technical solution of the present invention, the dynamic threshold calculation module performs the following steps for calculating the baseline value:

[0072] Obtain the rated power of the equipment and real-time power And calculate the load rate;

[0073] Outlier values ​​are removed from the current, vibration amplitude, and temperature data collected by the equipment under normal conditions, excluding data collected during equipment maintenance or when sensors malfunction.

[0074] The data is divided into different operating condition segments based on the load rate. For the normal data within each operating condition segment, a time window is set, and the arithmetic mean of the data within the time window is calculated to obtain the segmental baseline value. , among which, the k The segment baseline value for each working condition segment is ;

[0075] The global baseline value, i.e., the baseline value, is obtained by weighted summation of the baseline values ​​of different segments. ;

[0076] in, , , , ;

[0077] In the formula, N The number of data points within the window. This is the real-time data of the normal devices within the window. M The total number of operating condition segments. For the current working conditions, the first k The weight of each working condition segment r The current load rate, For the first k The average load rate of each operating condition segment This is the sharpening factor.

[0078] As a further limitation of the technical solution of the present invention, the method for calculating frequency band energy by the dynamic threshold calculation module includes:

[0079] Vibration signals After applying the Hanning window, an FFT transform is performed to obtain the spectrum. ;

[0080] The frequency band energy is obtained by extracting the spectrum of the target frequency band and calculating the energy proportion of the target frequency band; where, the target frequency band ;

[0081] .

[0082] As a further limitation of the technical solution of the present invention, the anomaly detection and scoring module, based on dynamic thresholds and real-time device data, calculates anomaly scores by including the following steps:

[0083] For each parameter, count the number of times it exceeds the dynamic threshold within the sliding time window; use the proportion of these exceeding times to the total number of sampling points as the exceedance ratio for that parameter; calculate the arithmetic mean of the exceedance ratios for all parameters to obtain the single-parameter exceedance ratio. The parameters include temperature, vibration amplitude, and current.

[0084] Calculate the Pearson correlation coefficient matrix among the parameters; take the mean of the absolute values ​​of the off-diagonal elements as the synchronicity score. Use 1 minus the synchronicity score as the associated outlier. ;

[0085]

[0086] ,

[0087] In the formula, n is the number of monitoring parameters. These are the elements of the correlation coefficient matrix. For parameters and covariance, The standard deviation is used; the time series data within the sliding time window is constructed with each column representing a parameter and each row representing the sampled values ​​at the same time point to form a parameter data matrix X.

[0088] Extract the multidimensional feature vector of the current device; calculate its normalized Euclidean distance with the historical fault feature vector; use 1 minus the normalized distance as the historical similarity;

[0089] Extract the current device feature vector (Including average temperature, vibration frequency energy, current fluctuation range, etc.), retrieve the most similar fault feature vector from the historical fault database. ;

[0090] Calculate the normalized Euclidean distance

[0091] Historical similarity

[0092] These are the maximum and minimum value vectors of all feature vectors in the historical fault feature database, respectively.

[0093] The anomaly score is obtained by weighting the proportion of single-parameter out-of-range values, the correlation of multiple parameters and historical similarity, and multiplying it by the sensitivity.

[0094] The formula for calculating the anomaly score is as follows:

[0095]

[0096] In the formula, As a weighting factor, The sensitivity is dynamically adjusted based on equipment operating time and environmental conditions.

[0097] As a further limitation of the technical solution of the present invention, the large model decision instruction generation module is specifically used to receive input structured context data, including maintenance records, production plans, spare parts inventory status and safety operation procedures, and output decision options with confidence, including operation actions, confidence levels, operation parameters and constraints.

[0098] As a further limitation of the technical solution of the present invention, the automated operation of the automated execution engine module includes: triggering one of the following automated operations according to the type of decision instruction:

[0099] Emergency shutdown, reduced load operation, automatic generation of maintenance work orders and push to the MES system, integration with the ERP system to reserve spare parts inventory, and sending early warning notifications to remind responsible persons.

[0100] As a further limitation of the technical solution of the present invention, the step of dynamically updating the instruction parameters based on the closed-loop feedback of the execution effect of the automated operation after the automated operation is triggered includes:

[0101] Collect actual operating status data of the equipment after executing decision commands, including equipment fault repair time, downtime, maintenance costs, and operation success rate;

[0102] The confidence level of a single decision instruction in the large model is updated based on the success rate of the last N operations; whereby... ; All are constants;

[0103] The performance evaluation index of the closed-loop feedback mechanism is calculated based on the collected data; the performance evaluation index is a reward function.

[0104]

[0105] in,

[0106]

[0107] In the formula, For MTBF improvement rate, To achieve cost savings, This represents the number of erroneous operations. These are the weighting coefficients;

[0108] like If the score falls below the set threshold for T consecutive times, the confidence threshold, residual compensation coefficient, and anomaly score weight of the large model decision-making process will be dynamically adjusted.

[0109] Thirdly, the present invention also provides an electronic device, the electronic device comprising: at least one processor; and a memory communicatively connected to the at least one processor; the memory storing computer program instructions executable by the at least one processor, the computer program instructions being executed by the at least one processor to enable the at least one processor to execute the industrial Internet of Things intelligent decision-making method based on a large model as described in the first aspect.

[0110] The beneficial effects of this invention lie in achieving multi-dimensional collaborative perception of equipment status through residual compensation algorithms and dynamic threshold calculations that adapt to operating conditions. Compared to traditional fixed threshold methods, this reduces the system's false alarm rate. It constructs an intelligent closed-loop system from perception to decision-making to execution and feedback, shortening the mean time to failure (MTTR) and thus reducing equipment downtime costs. Based on the closed-loop feedback mechanism, the system develops continuous evolution capabilities, with a steadily increasing proportion of autonomous decision-making. Attached Figure Description

[0111] 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, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0112] Figure 1 This is a schematic flowchart illustrating a method according to an embodiment of the present invention.

[0113] Figure 2 This is a schematic block diagram of a system according to an embodiment of the present invention. Detailed Implementation

[0114] To make the objectives, features, and advantages of this invention more apparent and understandable, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings of the specific embodiments. Obviously, the embodiments described below are only some embodiments of this invention, and not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0115] like Figure 1 As shown, this embodiment of the invention provides an intelligent decision-making method for the Industrial Internet of Things based on a large model, including:

[0116] S1. Based on real-time equipment data, operating parameters, and historical data, an adaptive dynamic threshold is generated using a residual compensation algorithm. Real-time equipment data includes current, vibration amplitude, and temperature. Operating parameters include load rate, ambient temperature, and runtime. Historical data consists of normal / abnormal time-series data from the past set time period.

[0117] S2. Based on dynamic thresholds and real-time device data, calculate an anomaly score, which is a weighted sum of the proportion of single-parameter exceedances, the correlation of multiple-parameter anomalies, and historical similarity multiplied by sensitivity.

[0118] S3. Input real-time equipment data, anomaly scores, and multimodal structured context into the large model to generate decision instructions with confidence assessment and record audit logs; the decision instructions include operation actions, operation parameters, and constraints; the structured context includes maintenance records, production plans, spare parts inventory status, and safe operating procedures.

[0119] S4. When the confidence level of the decision instruction reaches the confidence level threshold, an automated operation is triggered, and closed-loop feedback is performed based on the execution effect of the automated operation to dynamically update the confidence level, confidence level threshold, residual compensation coefficient, and anomaly score weight. When the anomaly score exceeds the set threshold but the confidence level does not reach the confidence level threshold, an early warning notification is triggered.

[0120] In this embodiment of the invention, the dynamic threshold includes an upper limit and a lower limit, wherein:

[0121]

[0122]

[0123] It should be noted that, This is the load compensation coefficient, initially set to 0.2. This is the environmental compensation coefficient, initially set to 0.05. This refers to three standard deviations from the baseline threshold. The frequency band energy weight is fitted using historical fault data and is set to 0.1.

[0124] The steps for calculating the baseline value are as follows:

[0125] Obtain the rated power of the equipment and real-time power And calculate the load rate;

[0126] Outlier values ​​are removed from the current, vibration amplitude, and temperature data collected by the equipment under normal conditions, excluding data collected during equipment maintenance or when sensors malfunction.

[0127] The data is divided into different operating condition segments based on the load rate. For the normal data within each operating condition segment, a time window is set, and the arithmetic mean of the data within the time window is calculated to obtain the segmental baseline value. , among which, the k The segment baseline value for each working condition segment is In this embodiment of the invention, the time window size is 1 hour, and the average value of the data within each working condition segment window is calculated to obtain the segmented baseline value.

[0128] The data is divided into three segments based on load rate: low load (0-30%), medium load (30-70%), and high load (70-100%); auxiliary segments are made based on ambient temperature: normal temperature conditions: temperature ≤50℃; high temperature conditions: temperature >50℃.

[0129] The global baseline value, i.e., the baseline value, is obtained by weighted summation of the baseline values ​​of different segments. ;

[0130] in, , , , For example, if the motor has a rated power of 10kW and a real-time power of 7kW, then the load rate is 70%.

[0131] In the formula, N The number of data points within the window. This refers to real-time data from normal devices within the specified time window. M The total number of operating condition segments. For the current working conditions, the first k The weight of each working condition segment r The current load rate, For the first k The average load rate of each operating condition segment This is the sharpening factor.

[0132] The baseline value update mechanism updates the baseline value every hour under normal operating mode; under special maintenance mode, the baseline value is updated every 15 minutes within 72 hours after equipment repair; the anomaly handling mechanism suspends the baseline value update and issues an alarm when an anomaly is detected.

[0133] In this embodiment of the invention, the method for calculating frequency band energy includes:

[0134] Vibration signals After applying the Hanning window, an FFT transform is performed to obtain the spectrum. ;

[0135] The frequency band energy is obtained by extracting the spectrum of the target frequency band and calculating the energy proportion of the target frequency band; where, the target frequency band ;

[0136] .

[0137] In some embodiments, the step of calculating anomaly scores based on dynamic thresholds and real-time device data includes:

[0138] For each parameter, count the number of times it exceeds the dynamic threshold within the sliding time window; use the proportion of these exceeding times to the total number of sampling points as the exceedance ratio for that parameter; calculate the arithmetic mean of the exceedance ratios for all parameters to obtain the single-parameter exceedance ratio. The parameters include temperature, vibration amplitude, and current. For example, if there are 600 temperature sampling points (100Hz × 60 seconds × 10 minutes), and 120 of them exceed the upper limit of the dynamic threshold, then the exceedance rate is 20%.

[0139] Calculate the Pearson correlation coefficient matrix among the parameters; take the mean of the absolute values ​​of the off-diagonal elements as the synchronicity score. Use 1 minus the synchronicity score as the associated outlier. ;

[0140]

[0141] ,

[0142] In the formula, n is the number of parameters. These are the elements of the correlation coefficient matrix. For parameters and covariance, The standard deviation is used; time series data within a sliding time window, with each column representing a parameter and each row representing sampled values ​​at the same time point, constructing a parameter data matrix X; for example, the correlation coefficient between temperature and vibration under normal conditions. When abnormal, it drops to If this occurs, the synchronicity score decreases and the correlation outlier increases.

[0143] Extract the multidimensional feature vector of the current device; calculate its normalized Euclidean distance with the historical fault feature vector; use 1 minus the normalized distance as the historical similarity;

[0144] Extract the current device feature vector (Including average temperature, vibration frequency energy, current fluctuation range, etc.), retrieve the most similar fault feature vector from the historical fault database. ;

[0145] Calculate the normalized Euclidean distance

[0146] Historical similarity

[0147] These are the maximum and minimum value vectors of all feature vectors in the historical fault feature database, respectively. For example, if the Euclidean distance between the current vibration spectrum and the historical bearing fault spectrum is 0.3, and the maximum historical distance is 1.5, then the historical similarity is... .

[0148] The anomaly score is obtained by weighting the proportion of single-parameter out-of-range values, the correlation of multiple parameters and historical similarity, and multiplying it by the sensitivity.

[0149] The formula for calculating the anomaly score is as follows:

[0150]

[0151] In the formula, As a weighting factor, Sensitivity is dynamically adjusted based on equipment operating time and environmental conditions. Base sensitivity. Sensitivity of new equipment (running time <100 hours) This is used to reduce false alarms in extreme environments (temperature > 50℃ or humidity > 90%). Used to improve sensitivity; otherwise, the sensitivity is the same as the specified sensitivity.

[0152] The output range of the anomaly score is 0-100, and different levels of alarms are triggered according to the score threshold: score > 90: emergency shutdown; 80 < score ≤ 90: planned maintenance; score ≤ 80: continuous monitoring.

[0153] In some embodiments, the generation of large model decision instructions in S3 includes:

[0154] The system takes structured context data as input, including maintenance records, production plans, spare parts inventory status, and safe operating procedures. It outputs decision options with confidence levels, including operating actions, confidence levels, operating parameters, and constraints. The confidence level threshold is initialized based on the equipment type and dynamically updated according to the execution results.

[0155] Structured data template:

[0156] {

[0157] "context": {

[0158] "abnormal_score": 85,

[0159] "temperature_threshold": [15, 100],

[0160] "maintenance_history": ["2025-02-20 Bearing Replacement"],

[0161] "production_phase": "peak"

[0162] }

[0163] }

[0164] Model decision output format (decision options with confidence scores):

[0165] {

[0166] "options": [{

[0167] "action": "emergency_shutdown",

[0168] "confidence": 0.91,

[0169] "params": {"delay_min": 5},

[0170] "constraints": ["require_digital_signature"]

[0171] }]

[0172] }

[0173] Confidence threshold strategy: Critical equipment: Shutdown operation threshold ≥ 0.9, early warning threshold ≥ 0.75, other cases ignored. Ordinary equipment: Shutdown operation threshold ≥ 0.85, early warning threshold ≥ 0.65, other cases ignored.

[0174] In some embodiments, the automated operation includes triggering one of the following automated operations based on the type of decision instruction:

[0175] Emergency shutdown, reduced load operation, automatic generation of maintenance work orders and push to the MES system, integration with the ERP system to reserve spare parts inventory, and sending early warning notifications to remind responsible persons.

[0176] If the anomaly score is greater than 90 and the associated device status is "normal":

[0177] Generate emergency repair orders and automatically reserve spare parts warehouse slots.

[0178] else if 80 < anomaly score <= 90 and production plan = "idle period":

[0179] Generate a planned maintenance order and push it to the maintenance queue.

[0180] else:

[0181] Record early warnings and continuously monitor.

[0182] In some embodiments, in S4, after the automated operation is triggered, the step of dynamically updating the instruction parameters based on the execution effect of the automated operation and performing closed-loop feedback includes:

[0183] Collect actual operating status data of the equipment after executing decision commands, including equipment fault repair time, downtime, maintenance costs, and operation success rate;

[0184] The confidence level of a single decision instruction in the large model is updated based on the success rate of the last N operations; whereby... In this embodiment of the invention, , .

[0185] The performance evaluation index of the closed-loop feedback mechanism is calculated based on the collected data; the performance evaluation index is a reward function.

[0186]

[0187] in,

[0188]

[0189] In the formula, For MTBF improvement rate, To achieve cost savings, This represents the number of erroneous operations. These are the weighting coefficients;

[0190] like If the score falls below the set threshold for T consecutive times, the confidence threshold, residual compensation coefficient, and anomaly score weight of the large model decision-making process will be dynamically adjusted.

[0191] The confidence threshold is adjusted in steps of ±0.05, and does not exceed the safety range for the device type;

[0192] The weights for abnormal ratings are redistributed every 24 hours, with priority given to reducing the weights for historical similarity.

[0193] For example, a certain pump frequently shuts down unexpectedly due to sudden load changes. (4 consecutive times <0.4);

[0194] Adjustment process: Increase the shutdown confidence threshold (0.9→0.93);

[0195] Reduce load compensation factor (0.2→0.18);

[0196] The historical similarity weight was reduced from 0.3 to 0.25. After the adjustment, false stops were reduced by 35%. It rose back to 0.65.

[0197] In addition, for example, there are more false alarms of vibration in high-temperature environments ( (3 consecutive times <0.5);

[0198] Adjustment process: Increase environmental compensation coefficient (0.05→0.06);

[0199] Sensitivity decreased from 1.2 to 1.0; vibration frequency range was optimized (500-1000Hz → 600-900Hz).

[0200] like Figure 2 As shown, this embodiment of the invention also provides an industrial IoT intelligent decision-making method based on a large model, including:

[0201] The dynamic threshold calculation module is used to generate adaptive dynamic thresholds based on real-time equipment data, operating parameters, and historical data using a residual compensation algorithm. The real-time equipment data includes current, vibration amplitude, and temperature; the operating parameters include load rate, ambient temperature, and runtime; and the historical data consists of normal / abnormal time-series data from the past set time period.

[0202] An anomaly detection and scoring module is used to calculate an anomaly score based on dynamic thresholds and real-time device data. The anomaly score is a weighted sum of the proportion of single-parameter exceedances, the correlation of multiple-parameter anomalies, and historical similarity, multiplied by a sensitivity.

[0203] The large model decision instruction generation module is used to receive anomaly scores and structure the multimodal data into contextual data, generate decision instructions with confidence assessment, and record audit logs; the decision instructions include operation actions, operation parameters, and constraints.

[0204] The automated execution engine module is used to trigger automated operations when the confidence level of a decision instruction reaches a preset confidence threshold or the abnormal score exceeds a set threshold. It also provides closed-loop feedback based on the execution effect of the automated operations, dynamically updating the confidence level, confidence threshold, residual compensation coefficient, and abnormal score weight.

[0205] The dynamic threshold includes an upper limit and a lower limit, wherein:

[0206]

[0207]

[0208] This is the load compensation coefficient. This is the environmental compensation coefficient. This refers to three standard deviations from the baseline threshold. Frequency band energy weight.

[0209] In some embodiments, the dynamic threshold calculation module performs the baseline value calculation steps as follows:

[0210] Obtain the rated power of the equipment and real-time power And calculate the load rate;

[0211] Outlier values ​​are removed from the current, vibration amplitude, and temperature data collected by the equipment under normal conditions, excluding data collected during equipment maintenance or when sensors malfunction.

[0212] The data is divided into different operating condition segments based on the load rate, and a time window is set for the normal data within each operating condition segment. The arithmetic mean of the data within the window is then calculated to obtain the segmented baseline value. , among which, the k The segment baseline value for each working condition segment is ;

[0213] The global baseline value, i.e., the baseline value, is obtained by weighted summation of the baseline values ​​of different segments. ;

[0214] in, , , , ;

[0215] In the formula, N The number of data points within the window. This is the real-time data of the normal devices within the window. M The total number of operating condition segments. For the current working conditions, the first k The weight of each working condition segment r The current load rate, For the first k The average load rate of each operating condition segment This is the sharpening factor.

[0216] In some embodiments, the dynamic threshold calculation module calculates frequency band energy using the following method:

[0217] Vibration signals After applying the Hanning window, an FFT transform is performed to obtain the spectrum. ;

[0218] The frequency band energy is obtained by extracting the spectrum of the target frequency band and calculating the energy proportion of the target frequency band; where, the target frequency band ;

[0219] .

[0220] In some embodiments, the anomaly detection and scoring module calculates anomaly scores based on dynamic thresholds and real-time device data, including the following steps:

[0221] For each parameter, count the number of times it exceeds the dynamic threshold within the sliding time window; use the proportion of these exceeding times to the total number of sampling points as the exceedance ratio for that parameter; calculate the arithmetic mean of the exceedance ratios for all parameters to obtain the single-parameter exceedance ratio. The parameters include temperature, vibration amplitude, and current.

[0222] Calculate the Pearson correlation coefficient matrix among the parameters; take the mean of the absolute values ​​of the off-diagonal elements as the synchronicity score. Use 1 minus the synchronicity score as the associated outlier. ;

[0223]

[0224] ,

[0225] In the formula, n is the number of monitoring parameters. These are the elements of the correlation coefficient matrix. For parameters and covariance, The standard deviation is used; the time series data within the sliding time window, with each column representing a parameter and each row representing the sampled values ​​at the same time point, constructs a parameter data matrix X.

[0226] Extract the multidimensional feature vector of the current device; calculate its normalized Euclidean distance with the historical fault feature vector; use 1 minus the normalized distance as the historical similarity;

[0227] Extract the current device feature vector (Including average temperature, vibration frequency energy, current fluctuation range, etc.), retrieve the most similar fault feature vector from the historical fault database. ;

[0228] Calculate the normalized Euclidean distance

[0229] Historical similarity

[0230] These are the maximum and minimum value vectors of all feature vectors in the historical fault feature database, respectively.

[0231] The anomaly score is obtained by weighting the proportion of single-parameter out-of-range values, the correlation of multiple parameters and historical similarity, and multiplying it by the sensitivity.

[0232] The formula for calculating the anomaly score is as follows:

[0233]

[0234] In the formula, As a weighting factor, The sensitivity is dynamically adjusted based on equipment operating time and environmental conditions.

[0235] In some embodiments, the large model decision instruction generation module is specifically used to receive input structured context data, including maintenance records, production plans, spare parts inventory status, and safety operating procedures, and output decision options with confidence levels, including operating actions, confidence levels, operating parameters, and constraints.

[0236] In some embodiments, the automated operations of the automated execution engine module include triggering one of the following automated operations based on the type of decision instruction:

[0237] Emergency shutdown, reduced load operation, automatic generation of maintenance work orders and push to the MES system, integration with the ERP system to reserve spare parts inventory, and sending early warning notifications to remind responsible persons.

[0238] In some embodiments, after an automated operation is triggered, the step of dynamically updating the instruction parameters based on the closed-loop feedback of the automated operation's execution effect includes:

[0239] Collect actual operating status data of the equipment after executing decision commands, including equipment fault repair time, downtime, maintenance costs, and operation success rate;

[0240] The confidence level of a single decision instruction in the large model is updated based on the success rate of the last N operations; whereby... In this embodiment of the invention, , .

[0241] The performance evaluation index of the closed-loop feedback mechanism is calculated based on the collected data; the performance evaluation index is a reward function.

[0242]

[0243] in,

[0244]

[0245] In the formula, For MTBF improvement rate, To achieve cost savings, This represents the number of erroneous operations. These are the weighting coefficients;

[0246] like If the score falls below the set threshold for T consecutive times, the confidence threshold, residual compensation coefficient, and anomaly score weight of the large model decision-making process will be dynamically adjusted.

[0247] This invention also provides an electronic device, comprising: a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other via the communication bus. The communication bus can be used for information transmission between the electronic device and sensors. The processor can call logical instructions in memory to execute the following methods: S1. Based on real-time equipment data, operating parameters, and historical data, an adaptive dynamic threshold is generated using a residual compensation algorithm; wherein, real-time equipment data includes current, vibration amplitude, and temperature; operating parameters include load rate, ambient temperature, and runtime; historical data are normal / abnormal time-series data from the past set time; S2. Based on the dynamic threshold and real-time equipment data, an anomaly score is calculated, wherein the anomaly score is a weighted sum of the proportion of single parameter exceeding the standard, the correlation of multiple parameter anomalies, and historical similarity, multiplied by sensitivity; S3. The anomaly score and the structured context of multimodal data are input into a large model to generate a decision instruction with confidence assessment and record an audit log; the decision instruction includes operation actions, operation parameters, and constraints; S4. When the confidence of the decision instruction reaches a preset confidence threshold or the anomaly score exceeds a set threshold, an automated operation is triggered, and closed-loop feedback is performed based on the execution effect of the automated operation to dynamically update the confidence, confidence threshold, residual compensation coefficient, and anomaly score weight.

[0248] Furthermore, the logical instructions in the aforementioned memory can be implemented as software functional units and sold or used as independent products, and can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0249] Although the present invention has been described in detail with reference to the accompanying drawings and preferred embodiments, the present invention is not limited thereto. Various equivalent modifications or substitutions can be made to the embodiments of the present invention by those skilled in the art without departing from the spirit and essence of the invention, and such modifications or substitutions should all be within the scope of the present invention. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should also be covered within the protection scope of the present invention.

Claims

1. An intelligent decision-making method for the Industrial Internet of Things based on a large model, characterized in that, include: S1. Based on real-time equipment data, operating parameters, and historical data, an adaptive dynamic threshold is generated using a residual compensation algorithm. Real-time equipment data includes current, vibration amplitude, and temperature. Operating parameters include load rate, ambient temperature, and runtime. Historical data consists of normal / abnormal time-series data from the past set time period. S2. Based on dynamic thresholds and real-time device data, calculate an anomaly score, which is a weighted sum of the proportion of single-parameter out-of-range values, multi-parameter correlated outliers, and historical similarity multiplied by sensitivity; specific steps include: For each parameter, count the number of times it exceeds the dynamic threshold within the time window; use the proportion of these exceeding times to the total number of sampling points as the exceedance ratio for that parameter; calculate the arithmetic mean of the exceedance ratios for all parameters to obtain the single-parameter exceedance ratio. The parameters include temperature, vibration amplitude, and current. Calculate the Pearson correlation coefficient matrix among the parameters; take the mean of the absolute values ​​of the off-diagonal elements as the synchronicity score. Use 1 minus the synchronicity score as the associated outlier. ; Extract the multidimensional feature vector of the current device; calculate its normalized Euclidean distance with the historical fault feature vectors; subtract the normalized distance from 1 as the historical similarity. ; The anomaly score is obtained by weighting the proportion of single-parameter out-of-range values, the correlation of multiple parameters and historical similarity, and multiplying it by the sensitivity. The formula for calculating the anomaly score is as follows: In the formula, As a weighting factor, Sensitivity is set based on device operating time and environmental conditions; S3. Input real-time equipment data, anomaly scores, and multimodal structured context into the large model to generate decision instructions with confidence assessment and record audit logs; the decision instructions include operation actions, operation parameters, and constraints; the structured context includes maintenance records, production plans, spare parts inventory status, and safe operating procedures. S4. When the confidence level of the decision instruction reaches the confidence threshold, an automated operation is triggered, and closed-loop feedback is performed based on the execution effect of the automated operation to dynamically update the confidence level, confidence threshold, residual compensation coefficient, and anomaly score weight; when the anomaly score exceeds the set threshold but the confidence level does not reach the confidence threshold, an early warning notification is triggered; the steps of dynamically updating the calculation parameters based on the execution effect of the automated operation after it is triggered include: Collect actual operating status data of the equipment after executing decision commands, including equipment fault repair time, downtime, maintenance costs, and operation success rate; The confidence level of a single decision instruction in the large model is updated based on the success rate of the last N operations; whereby... ; All are constants; The performance evaluation index of the closed-loop feedback mechanism is calculated based on the collected data; the performance evaluation index is a reward function. in, In the formula, For MTBF improvement rate, For cost savings, This represents the number of erroneous operations. These are the weighting coefficients; like If the score falls below the set threshold for T consecutive times, the confidence threshold, residual compensation coefficient, and anomaly score weight of the large model decision-making process will be dynamically adjusted.

2. The industrial IoT intelligent decision-making method based on a large model according to claim 1, characterized in that, The dynamic threshold includes an upper limit and a lower limit, wherein: In the formula, This is the load compensation coefficient. This is the environmental compensation coefficient. This refers to three standard deviations from the baseline threshold. Frequency band energy weight.

3. The industrial IoT intelligent decision-making method based on a large model according to claim 2, characterized in that, The steps for calculating the baseline value are as follows: Obtain the rated power and real-time power of the equipment, and calculate the load rate; Outlier values ​​are removed from the current, vibration amplitude, and temperature data collected by the equipment under normal conditions, excluding data collected during equipment maintenance or when sensors malfunction. The data is divided into different operating conditions based on the load rate. For the normal data in each operating condition segment, a time window is set to calculate the arithmetic mean of the data within the time window to obtain the segment baseline value. The global baseline value, or baseline value, is obtained by weighted summation of the baseline values ​​of different segments.

4. The industrial IoT intelligent decision-making method based on a large model according to claim 3, characterized in that, Methods for calculating frequency band energy include: Vibration signals After applying the Hanning window, an FFT transform is performed to obtain the spectrum. ; The frequency band energy is obtained by extracting the spectrum of the target frequency band and calculating the energy proportion of the target frequency band; where, the target frequency band ; 。 5. The industrial IoT intelligent decision-making method based on a large model according to claim 4, characterized in that, The generation of large model decision instructions in S3 includes: It receives anomaly scores, real-time equipment data, maintenance records, production plans, and spare parts inventory status, and outputs decision options with confidence levels, including operation actions, confidence levels, operation parameters, and constraints. The confidence level threshold is initialized based on the equipment type and dynamically updated according to the execution effect of automated operations.

6. The industrial IoT intelligent decision-making method based on a large model according to claim 5, characterized in that, The automated operations include triggering one of the following automated operations based on the type of decision instruction: Emergency shutdown, reduced load operation, automatic generation of maintenance work orders and push to the MES system, integration with the ERP system to reserve spare parts inventory, and sending early warning notifications to remind responsible persons.

7. An industrial Internet of Things (IoT) intelligent decision-making system based on a large model, implementing the method of any one of claims 1-6, characterized in that, include: The dynamic threshold calculation module is used to generate adaptive dynamic thresholds based on real-time equipment data, operating parameters, and historical data using a residual compensation algorithm. The real-time equipment data includes current, vibration amplitude, and temperature; the operating parameters include load rate, ambient temperature, and runtime; and the historical data consists of normal / abnormal time-series data from the past set time period. An anomaly detection and scoring module is used to calculate an anomaly score based on dynamic thresholds and real-time device data. The anomaly score is a weighted sum of the proportion of single-parameter exceedances, the correlation of multiple-parameter anomalies, and historical similarity, multiplied by sensitivity. The large model decision instruction generation module is used to input real-time equipment data, anomaly scores, and multimodal structured context into the large model, generate decision instructions with confidence assessment, and record audit logs; the decision instructions include operation actions, operation parameters, and constraints; the structured context includes maintenance records, production plans, spare parts inventory status, and safety operation procedures; The automated execution engine module is used to trigger automated operations when the confidence level of a decision instruction reaches the confidence level threshold, and to provide closed-loop feedback based on the execution effect of the automated operations, dynamically updating the confidence level, confidence level threshold, residual compensation coefficient, and anomaly score weight; when the anomaly score exceeds the set threshold but the confidence level does not reach the confidence level threshold, an early warning notification is triggered.

8. An electronic device, characterized in that, The electronic device includes: at least one processor; and a memory communicatively connected to the at least one processor; the memory stores computer program instructions executable by the at least one processor, the computer program instructions being executed by the at least one processor to enable the at least one processor to execute the industrial Internet of Things intelligent decision-making method based on a large model as described in any one of claims 1 to 6.

Images