AI identification decision method based on industrial Internet of Things

Abstract

The invention relates to the technical field of industrial Internet of Things and artificial intelligence, in particular to an AI identification decision method based on the industrial Internet of Things. The method comprises the steps that multi-modal data of equipment operation are collected and preprocessed, and a multi-modal embedded feature vector is obtained; based on the multi-modal embedded feature vector, introducing a fuzzy state scoring mechanism fused with multi-modal cooperative enhancement, and generating an equipment operation fuzzy state score; based on the equipment operation fuzzy state score, identifying the abrupt change trend of the equipment operation state, and obtaining a score fluctuation response value; constructing a strategy library containing candidate control action vectors, and performing comprehensive efficiency evaluation on the candidate control action vectors in combination with the equipment operation fuzzy state score and the score fluctuation response value to obtain a joint score value; and based on the joint score value, intelligent adjustment of equipment operation is realized. The technical problems that a traditional identification decision-making method lacks different modal collaborative modeling capability, identification and control separation are caused, and control action selection lacks evaluation basis are solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of industrial Internet of Things and artificial intelligence, and in particular to an AI identification and decision method based on industrial Internet of Things. BACKGROUND

[0002] With the development of Industrial Internet of Things (IIoT) technology, the breadth and depth of equipment operation state monitoring and data collection have been significantly improved, providing a data basis for realizing intelligent perception and closed-loop control of equipment. However, in actual industrial production processes, the equipment operation state is usually in a fuzzy transition area between "normal" and "abnormal", showing high nonlinearity, weak symptom and multi-source heterogeneity, which makes it difficult for traditional identification methods based on threshold judgment or rule engine to accurately perceive potential abnormalities or trend changes of equipment.

[0003] Although some current researches use neural networks, support vector machines and other methods to classify and identify equipment fault states, most of the methods still have the following problems: (1) lack of collaborative modeling ability for fuzzy states: the coupling effect between multiple modalities (such as temperature, pressure, vibration, etc.) is often involved in the equipment operation process, and existing methods mostly focus on single variable modeling, which is difficult to capture the collaborative degradation characteristics between multiple modalities; (2) identification and control are separated, and the response is lagging: most methods only focus on the identification stage, and fail to dynamically convert the identification results into executable control decision strategies, lacking a response mechanism for real-time control; (3) control action selection lacks evaluation basis: even if there is a control strategy library, the dynamic trade-off between the current state change rate of the equipment, the control cost and the expected benefit is not considered, and the control action selection lacks systematic evaluation basis, resulting in inaccurate and uncontrollable response strategies.

[0004] Therefore, it is urgent to propose a method that can integrate multiple modal information of equipment operation, real-time perceive the evolution trend of fuzzy state, and intelligently match the response strategy combined with the control cost, to realize the integration of equipment state identification and control strategy selection, and improve the stability and safety of equipment operation. SUMMARY

[0005] The present application provides an AI identification and decision method based on industrial Internet of Things to solve the technical problems of traditional identification and decision methods lacking collaborative modeling ability of different modalities, separation of identification and control, and lack of evaluation basis for control action selection.

[0006] The AI identification and decision method based on industrial Internet of Things of the present application specifically includes the following technical solutions: An AI identification and decision method based on industrial Internet of Things includes the following steps: S1. Collecting multi-modal data of the equipment operation and pre-processing to obtain multi-modal embedding feature vectors; introducing a fuzzy state score mechanism based on multi-modal collaborative enhancement based on the multi-modal embedding feature vectors to generate an equipment operation fuzzy state score; and constructing an equipment operation data based on the multi-modal data of the equipment operation, the multi-modal embedding feature vectors and the equipment operation fuzzy state score; S2. Identifying the mutation trend of the equipment operation state based on the equipment operation fuzzy state score to obtain a score fluctuation response value; constructing a strategy library containing candidate control action vectors, combining the equipment operation fuzzy state score and the score fluctuation response value to comprehensively evaluate the candidate control action vectors to obtain a joint score value; and realizing intelligent adjustment of the equipment operation based on the joint score value.

[0007] Preferably, the S1 specifically comprises: In the implementation process of the fuzzy state score mechanism based on multi-modal collaborative enhancement, the deviation degree of the feature distance between the modal pairs of embedding feature vectors relative to the average feature distance of the modal pairs under the normal state of the equipment operation is calculated based on the multi-modal embedding feature vectors, and the collaborative weight coefficient and the collaborative correlation strength between the modes are introduced to calculate the equipment operation fuzzy state score.

[0008] Preferably, the S2 specifically comprises: The risk response adjustment coefficient is introduced by constructing a nonlinear score fluctuation response function through calculating the change amount of the equipment operation fuzzy state score and combining the historical average fuzzy state score value to obtain the score fluctuation response value.

[0009] Preferably, the S2 specifically comprises: When the score fluctuation response value exceeds the preset response threshold, the control action decision stage is entered, and a strategy library containing candidate control action vectors is constructed; the elements in the candidate control action vectors represent the adjustment amplitude of the candidate control action vectors to the controllable parameters.

[0010] Preferably, the S2 specifically comprises: Based on the candidate control action vectors, the current equipment operation fuzzy state criticality, the score fluctuation trend and the control action economic performance ratio are quantified by combining the equipment operation fuzzy state score and the score fluctuation response value to construct a response optimality score function to obtain a joint score value.

[0011] Preferably, the S2 specifically comprises: In the process of constructing the response goodness scoring function, based on the current equipment operation fuzzy state score and combined with a preset risk threshold, the criticality of the current equipment operation fuzzy state score is quantified; based on the score fluctuation response value and combined with the score fluctuation response value in historical equipment operation data, a penalty structure is constructed; combined with the equipment operation fuzzy state score at effective moments in historical equipment operation data, a control action normalized cost term is constructed to quantify the score improvement rate under unit control cost of candidate control action vectors; the effective moment refers to the moment corresponding to the historical multimodal embedding feature vector with the smallest feature distance to the current multimodal embedding feature vector.

[0012] Preferably, S2 specifically includes: Based on the equipment operation fuzzy state score of the effective time in the historical equipment operation data, the decrease value of the equipment operation fuzzy state score after executing the candidate control action vector is calculated, and the control cost of the candidate control action vector is combined to construct the control action normalized cost term.

[0013] Preferably, S2 specifically includes: The candidate control action vector that maximizes the joint score is taken as the optimal candidate control action vector. The optimal candidate control action vector is then converted into an operation command that the industrial equipment can recognize and sent to the control system for execution, thereby realizing closed-loop regulation of equipment operation.

[0014] The beneficial effects of the technical solution of the present invention are: 1. This invention proposes a fuzzy state scoring mechanism that integrates multimodal collaborative enhancement. By introducing collaborative weight coefficients and collaborative correlation strengths between different modal data, the scores of each modality are weighted and fused. This can more effectively identify subtle state changes of equipment in the "normal-abnormal" transition range, significantly improve the ability to identify early degradation trends and fuzzy fault symptoms, and provide a more reliable state input basis for control decisions.

[0015] 2. The response excellence scoring function constructed in this invention takes the fuzzy state score of the equipment operation as a continuous variable input, rather than a binary judgment, and constructs a triple correlation function of score-trend-effect in the control action decision stage, avoiding the separation of identification and control, and has the advantages of early trend response and multi-dimensional control action evaluation.

[0016] 3. In the process of optimizing control actions, this invention introduces a normalized cost term for control actions and constructs a response quality scoring function in combination with the score improvement effect. This effectively balances the score recovery capability with the actual execution cost, realizes the dynamic selection of the optimal control action, avoids problems such as over-control and ineffective control, and improves the economy and feasibility of the control strategy. Attached Figure Description

[0017] Figure 1 This is a flowchart of an AI recognition and decision-making method based on the Industrial Internet of Things as described in this invention. Detailed Implementation

[0018] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments. 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.

[0019] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0020] The following description, in conjunction with the accompanying drawings, details a specific scheme for an AI-based recognition and decision-making method for industrial IoT provided by this invention.

[0021] See attached document Figure 1 The diagram illustrates a flowchart of an AI-based identification and decision-making method for the Industrial Internet of Things (IIoT) according to an embodiment of the present invention. The method includes the following steps: S1. Collect multimodal data of equipment operation and preprocess it to obtain multimodal embedding feature vectors; based on the multimodal embedding feature vectors, introduce a fuzzy state scoring mechanism that integrates multimodal collaborative enhancement to generate a fuzzy state score of equipment operation; construct equipment operation data based on the multimodal data of equipment operation, multimodal embedding feature vectors and fuzzy state score of equipment operation. Multimodal data such as sound, temperature, and vibration from equipment operation are collected through microphones, temperature sensors, and triaxial accelerometers in the Industrial Internet of Things (IIoT). This multimodal data is then preprocessed, including time alignment, standardization, feature extraction, and dimensionality unification, to obtain multimodal embedding feature vectors. ,in, Indicates the first The modality at time... The embedded feature vector, , indicating modal index, The number of modes is indicated; the techniques used in the above preprocessing are all well-known to those skilled in the art and will not be elaborated here. To more effectively identify subtle state changes in equipment operating in the "normal-abnormal" transition range during industrial field operations, a fuzzy state scoring mechanism integrating multimodal collaborative enhancement is introduced. Based on multimodal embedded feature vectors, the deviation of the feature distance between the embedded feature vectors of any modality pair from the average feature distance of the modality pair under normal equipment operation is calculated, and enhancement is achieved through the Softplus function. Collaborative weight coefficients and collaborative correlation coefficients between modalities are introduced to calculate the fuzzy state score of equipment operation. Equipment operation data is constructed based on the multimodal data of equipment operation, the multimodal embedded feature vectors, and the fuzzy state score. The specific calculation formula for the fuzzy state score of equipment operation is as follows: , in, Indicates time The equipment operation fuzzy status score indicates that the higher the value, the more ambiguous the equipment operation status and the further it deviates from the normal state; the normal state refers to the fuzzy status score of the equipment operation. Below the risk threshold The equipment operating status at that time; the risk threshold , representing the risk threshold of the fuzzy state score, is used to determine whether the equipment is operating in a normal state. It is obtained by fitting the score distribution of historical equipment operation data using a Gaussian distribution percentile method with a confidence level of 97%. The calculation method of the risk threshold is a well-known technique in the art and will not be elaborated here. The number of modes; , All are modal indexes. ; Represents the modal collaborative weight coefficient, and represents the first modal collaborative weight coefficient. The modality and the first The degree of synergistic influence of each modality in the fuzzy state scoring of equipment operation satisfies the normalization condition. By analyzing the abnormal samples of historical equipment operation data, each mode pair The proportion of samples in abnormal states is obtained by normalization, and the abnormal samples are the fuzzy state scores of equipment operation. Higher than or equal to the risk threshold Equipment operating data at that time; Indicates the first The modality and the first The co-correlation coefficients among the modalities can be obtained by using the Pearson correlation coefficient method on historical equipment operating data. The Pearson correlation coefficient method is a well-known technique in the field and will not be elaborated upon here. The higher the value, the higher the fuzzy state score of the device operation; if the mode pair is an undirected mode pair, then... , ; This is a Softplus function used to enhance the responsiveness of device fuzzy state scoring in abnormal conditions. , representing the abnormal increment of the modal response to cooperative bias, if A 0 indicates that the equipment is operating in an abnormal state; conversely, a 0 indicates that the equipment is operating normally. Indicates the first The modality at time... Embedded feature vectors; Indicates the first The modality at time... Embedded feature vectors; Indicates the first The modality and the first The modality at time... The feature distance is used to measure whether the collaborative relationship between the two has deviated; Representing mode pairs The average feature distance under normal operating conditions of historical equipment can be obtained by statistically analyzing normal samples in historical equipment operating data. These normal samples are the fuzzy state scores of equipment operation. Below the risk threshold Equipment operating data at that time; The above formula introduces a fuzzy state scoring mechanism that integrates multimodal collaborative enhancement, which significantly enhances the industrial IoT's ability to respond to "collaborative abnormal" states of equipment operation, and is particularly suitable for the identification of multimodal coupled faults in complex industrial scenarios.

[0022] S2. Based on the fuzzy state score of equipment operation, identify the sudden change trend of equipment operation status and obtain the score fluctuation response value; construct a strategy library containing candidate control action vectors, combine the fuzzy state score of equipment operation and the score fluctuation response value, and conduct a comprehensive performance evaluation of the candidate control action vectors to obtain a joint score value; based on the joint score value, realize intelligent adjustment of equipment operation.

[0023] To achieve early detection of abrupt changes in equipment operating status, a nonlinear scoring fluctuation response function is proposed. This function dynamically quantifies the growth rate and risk level of the current equipment operating fuzzy status score, yielding the scoring fluctuation response value. The specific formula is as follows: , in, Indicates time The rating fluctuation response value is used to identify abrupt trends in the operating status of the equipment; , is the risk response adjustment coefficient, used to make system-level adaptive adjustments to the score fluctuation response value under different equipment types and operating conditions. It is obtained by backtracking the correlation between the change in the fuzzy state score of equipment operation and the score fluctuation risk in historical equipment operation data, and fitting it with the minimum mean square error method. The minimum mean square error method is a technical means well known to those skilled in the art, and will not be described in detail here. It represents the change in the fuzzy state score of equipment operation, and is used to measure the trend of the change in the fuzzy state score of equipment operation at the most recent moment. It can be used as a sensitive indicator of early abnormal behavior. Indicates time Fuzzy status scoring of equipment operation; Indicates time Fuzzy status scoring of equipment operation; This indicates the degree of deviation of the current equipment operating fuzzy state score from the historical average fuzzy state score. It is used to measure the degree of risk of the current equipment operation. The larger the value, the more the equipment operation deviates from the normal state. This represents the average fuzzy state score of the device within its historical normal operating range, i.e., the historical average fuzzy state score. The historical normal operating range is defined as the period from the current moment... During the backtracking process, the first interval found that satisfies the following conditions The above conditions are: interval The fuzzy state score of the equipment operation at any given time is less than the risk threshold. At the same time The previous moment and The equipment operation fuzzy state score at the next moment is greater than or equal to the risk critical threshold. The length of the historical normal operating interval is , An index representing a relative time step within a historical normal operating range. Indicates the first in the historical normal operating range Fuzzy status score of device operation at any given time; , representing the scoring amplification factor, can be obtained through expert experience; It is a non-linear amplification term that reflects the degree of risk deviation in the current equipment operating state, improving the early identification capability of "slowly changing faults". The introduction of 1 is to prevent fuzzy scoring of equipment operating state. When approaching 0, The value is set to 0 to maintain the basic response, which is used to reflect the degree of risk deviation of the current equipment operating status. The above formula not only describes the sudden increase of the equipment operating fuzzy status score itself, but also combines the degree of deviation of the equipment operating fuzzy status score from the historical average fuzzy status score value, effectively improving the early detection capability of sudden changes in equipment operating status and improving the foresight of the response. When the rating fluctuates, the response value Exceeding the response threshold At that time, the control action decision-making stage begins; the response threshold... The scoring fluctuation response value can be calculated by analyzing the abnormal state periods in historical equipment operation data. And obtain the minimum value; in the control action decision stage, construct a system containing The policy library for the nth candidate control action vector, where the nth... Each candidate control action vector is defined as... ,in Indicates the first Candidate control action vectors For the The adjustment range of each controllable parameter This indicates the number of controllable parameters, which include main pump speed (unit: revolutions per minute), valve opening degree (unit: %), cooling time (unit: seconds), etc., for example, ,at this time The first, second, and third controllable parameters are the main pump speed, valve opening, and cooling time, respectively. Indicates the first The candidate control action vectors include three control actions: main pump speed decreases by 300 rpm, valve opening increases by 5%, and cooling time increases by 3 seconds. The strategy library is obtained by training a deep reinforcement learning algorithm based on historical equipment operation data, which is a technical means well known to those skilled in the art and will not be described in detail here. Furthermore, based on multi-objective decision-making theory, and combined with the current fuzzy state scoring of equipment operation... With rating fluctuation response value Taking into account the criticality of the current equipment's fuzzy state, the trend of score fluctuations, and the cost-effectiveness of control actions, a response goodness scoring function is constructed. The formula is as follows: , in, Indicates the scoring of the device's fuzzy operating state. With rating fluctuation response value Given the given information, for candidate control action vectors The combined score obtained from a comprehensive performance evaluation; , representing the risk threshold; Used to characterize the fuzzy state score of the current equipment operation and the risk threshold. The closer the value is to the abnormal state, the smaller the value indicates that the equipment is closer to an abnormal state. The maximum value of the rating fluctuation response value in the historical equipment operation data is used for normalization; This value measures the rate and direction of change in the current rating, reflecting the dynamic deterioration trend of the equipment's operating status. A higher value indicates an accelerating deterioration in the equipment's operating status. Constructing a penalty structure helps in selecting more robust candidate control action vectors to cope with score fluctuations; For the weighting coefficients, satisfying Setting it through expert experience, such as , , ; It is the normalized cost item for control actions. Indicates the valid time in historical equipment operation data The device operation fuzzy state score, the effective time It is a candidate control action vector The actual execution time during the operation of the historical equipment represents the historical time corresponding to the historical multimodal embedding feature vector with the smallest feature distance to the current multimodal embedding feature vector. The feature distance is calculated using the existing Euclidean distance method. The smaller the feature distance, the more similar the two times. Candidate control action vectors from historical equipment operation data After execution Fuzzy status score of device operation at any given time; Indicates the execution of candidate control action vectors The subsequent fuzzy state score decrease value of the device operation is used to measure the candidate control action vector. ; Represents candidate control action vectors The resulting control costs For the first The cost weights of each controllable parameter are obtained by fitting existing least squares methods based on historical control cost and adjustment effect data of different controllable parameters. The historical control cost and adjustment effect data can be obtained through the equipment operation database. Indicates the first Candidate control action vectors For the The adjustment range of each controllable parameter; Indicates the first Candidate control action vectors For the The maximum adjustment range of each controllable parameter is used for normalization and is obtained through the equipment operation database. To prevent extremely small constants from being divided by zero, a value of [value] can be taken. .

[0024] Compared to traditional threshold-triggered response logic, the above formula uses the fuzzy state score of the equipment operation as a continuous variable input instead of a binary judgment, and constructs a triple correlation function of score-trend-effect in the control action decision stage. It has the advantages of early trend response, multi-dimensional control action evaluation, and selection of optimal response behavior. While ensuring the stability of equipment operation, it effectively improves the rationality of control actions. Furthermore, through Select the one that makes the joint score The index of the candidate control action vector that reaches the maximum value is taken as the optimal candidate control action vector; the optimal candidate control action vector is... The commands are converted into operation instructions that can be recognized by industrial equipment and sent to the PLC / DCS control system for execution via industrial IoT communication protocols such as Modbus / MQTT, realizing closed-loop intelligent adjustment from fuzzy state scoring of equipment operation to adaptive response control.

[0025] In summary, an AI-based identification and decision-making method based on the Industrial Internet of Things has been developed.

[0026] The order of the embodiments is for illustrative purposes only and does not represent the superiority or inferiority of the embodiments. The processes depicted in the drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0027] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

[0028] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. An AI-based recognition and decision-making method based on the Industrial Internet of Things, characterized in that, Includes the following steps: S1. Collect multimodal data from the device operation and preprocess it to obtain multimodal embedding feature vectors; Based on multimodal embedding feature vectors, a fuzzy state scoring mechanism that integrates multimodal collaborative enhancement is introduced to generate a fuzzy state score for equipment operation; based on the multimodal data of equipment operation, multimodal embedding feature vectors, and the fuzzy state score for equipment operation, equipment operation data is constructed. S2. Based on the fuzzy state score of equipment operation, identify the sudden change trend of equipment operation status and obtain the score fluctuation response value; construct a strategy library containing candidate control action vectors, combine the fuzzy state score of equipment operation and the score fluctuation response value, and conduct a comprehensive performance evaluation of the candidate control action vectors to obtain a joint score value; based on the joint score value, realize intelligent adjustment of equipment operation.

2. The AI-based identification and decision-making method based on the Industrial Internet of Things according to claim 1, characterized in that, S1 specifically includes: In the implementation of the fuzzy state scoring mechanism that integrates multimodal collaborative enhancement, based on the multimodal embedded feature vector, the deviation of the feature distance between the embedded feature vectors of the modal pairs from the average feature distance of the modal pairs under normal operating conditions is calculated, and the collaborative weight coefficient and collaborative correlation strength between the modal pairs are introduced to calculate the fuzzy state score of the equipment operation.

3. The AI-based identification and decision-making method based on the Industrial Internet of Things according to claim 1, characterized in that, S2 specifically includes: By calculating the change in the fuzzy state score of the equipment operation and combining it with the historical average fuzzy state score, a risk response adjustment coefficient is introduced to construct a nonlinear score fluctuation response function, and the score fluctuation response value is obtained.

4. The AI-based identification and decision-making method based on the Industrial Internet of Things according to claim 3, characterized in that, S2 specifically includes: When the rating fluctuation response value exceeds the preset response threshold, the system enters the control action decision stage and constructs a strategy library containing candidate control action vectors; the elements in the candidate control action vectors represent the adjustment range of the candidate control action vectors on the controllable parameters.

5. The AI-based identification and decision-making method based on the Industrial Internet of Things according to claim 4, characterized in that, S2 specifically includes: Based on candidate control action vectors, and combining the equipment operation fuzzy state score and score fluctuation response value, the critical degree of the current equipment operation fuzzy state, the score fluctuation trend and the cost-effectiveness of control actions are quantified, and a response goodness scoring function is constructed to obtain a joint score value.

6. The AI-based identification and decision-making method based on the Industrial Internet of Things according to claim 5, characterized in that, S2 specifically includes: In the process of constructing the response goodness scoring function, based on the current equipment operation fuzzy state score and combined with a preset risk threshold, the criticality of the current equipment operation fuzzy state score is quantified; based on the score fluctuation response value and combined with the score fluctuation response value in historical equipment operation data, a penalty structure is constructed; combined with the equipment operation fuzzy state score at effective moments in historical equipment operation data, a control action normalized cost term is constructed to quantify the score improvement rate under unit control cost of candidate control action vectors; the effective moment refers to the moment corresponding to the historical multimodal embedding feature vector with the smallest feature distance to the current multimodal embedding feature vector.

7. The AI-based identification and decision-making method based on the Industrial Internet of Things according to claim 6, characterized in that, S2 specifically includes: Based on the equipment operation fuzzy state score of the effective time in the historical equipment operation data, the decrease value of the equipment operation fuzzy state score after executing the candidate control action vector is calculated, and the control cost of the candidate control action vector is combined to construct the control action normalized cost term.

8. The AI-based identification and decision-making method based on the Industrial Internet of Things according to claim 5, characterized in that, S2 specifically includes: The candidate control action vector that maximizes the joint score is taken as the optimal candidate control action vector. The optimal candidate control action vector is then converted into an operation command that the industrial equipment can recognize and sent to the control system for execution, thereby realizing closed-loop regulation of equipment operation.

Images