Industrial equipment intelligent control method and system

Abstract

The invention discloses an industrial equipment intelligent control method and system, and the method comprises the following steps: S1, collecting and preprocessing equipment operation data, and constructing a state data set; s2, establishing a trend prediction model by adopting Gaussian process regression, and outputting a state prediction value; s3, constructing a graph structure representation model, and combining a graph convolutional network and a semi-supervised learning method to carry out joint training and identify a working condition category and an operation level; s4, based on the equipment operation data, calculating a fuzzy membership degree and a fuzzy entropy index, and generating an entropy distribution curve; s5, fusing the state prediction value, the working condition category, the operation grade and the fuzzy entropy index to generate a state evaluation result; s6, adjusting control strategy parameters according to a state evaluation result, and issuing a control instruction to an equipment control system; and S7, collecting feedback data, updating the state data set, and realizing closed-loop iterative optimization. The method has the advantages of accurate prediction, stable identification, flexible evaluation, fast control, self-learning and the like, and is suitable for complex and changeable industrial application scenes.

Description

Technical Field

[0001] The present invention relates to the technical field of industrial automation control, and particularly to an intelligent control method and system for industrial equipment. Background Art

[0002] In the current field of industrial manufacturing and automation, the intelligent control of industrial equipment has become a key technical link to improve production efficiency, reduce energy consumption, and ensure the stable operation of equipment. With the development of Industry 4.0 and intelligent manufacturing, the traditional equipment control methods have gradually revealed problems that cannot meet the requirements of complex working condition changes and dynamic optimization. In the prior art, industrial equipment mostly relies on preset logic rules, PID controllers, or control strategies based on empirical knowledge for operation control. Although these methods have certain effects under a single stable working condition, they show obvious limitations in the modern industrial environment with large working condition fluctuations, complex data, and frequent changes in equipment states.

[0003] On the one hand, the existing industrial equipment control systems often lack the in-depth understanding and modeling ability of the equipment operation state, and cannot achieve refined and personalized prediction and control. Especially in the context of high-frequency acquisition and strong interference, a large amount of industrial sensing data presents non-linear, strong noise, and time-series dynamic characteristics, making it difficult for traditional control methods to extract effective information from it. On the other hand, although some studies have tried to introduce data-driven algorithmic methods, such as state recognition and prediction technologies based on support vector machines, decision trees, or convolutional neural networks, most models have high requirements for data quality and lack the modeling ability to handle uncertainty and small sample conditions, resulting in frequent problems such as weak generalization ability, poor real-time performance, and insufficient control accuracy.

[0004] In terms of state modeling, the existing methods mostly adopt simple fitting or empirical regression models, lack the confidence measure of the state prediction error, lack the estimation of the future state fluctuation range, and cannot meet the requirements of high-reliability equipment regulation. At the same time, for the coupling relationship, operation dependence, state propagation, etc. among multiple devices in complex industrial systems, most of the existing methods adopt static modeling and cannot dynamically depict the structural influence and behavior linkage among devices. In addition, a large number of methods completely rely on supervised data for training, face the problems of difficult label acquisition and high annotation cost in the actual industrial environment, and lack an effective mechanism for utilizing unlabeled data.

[0005] In terms of equipment state evaluation, most of the existing methods judge the operation state based on simple rules or threshold analysis, and cannot reflect the fuzzy, uncertain, and dynamic change characteristics of the equipment operation state, resulting in the lack of flexibility and stability of the state evaluation results, and further affecting the decision-making basis of downstream control strategies. Especially in sensitive stages such as the early faults of key equipment and the critical state of working conditions, traditional methods often lack sufficient response mechanisms and cannot effectively sense abnormal trends.

[0006] In terms of control strategy generation, most existing control systems rely on fixed logic or simple feedback regulation, lacking the ability to generate dynamic strategy parameters based on the results of state evaluation, and unable to perform real-time adaptive regulation according to the current state changes of the equipment, resulting in poor regulation effects and even possible control interference or feedback oscillation. In addition, most control systems adopt an open-loop control structure, unable to achieve strategy optimization and system self-learning based on feedback results, and unable to form a complete closed-loop intelligent regulation system.

[0007] The existing industrial equipment control methods generally have the following defects: lack of high-precision prediction ability for complex time-series data, lack of dynamic state recognition mechanism for structural data, lack of fuzzy evaluation means for the uncertainty of operating states, lack of parameterized strategy generation mechanism based on evaluation results, and lack of the ability to continuously learn from control results, ultimately leading to unstable control effects, untimely responses, and low operating efficiency, seriously restricting the implementation of the intelligent manufacturing system and the improvement of operating performance.

[0008] Therefore, how to provide an intelligent control method and system for industrial equipment is an urgent problem to be solved by those skilled in the art. Summary of the Invention

[0009] An object of the present invention is to propose an intelligent control method and system for industrial equipment. The present invention integrates a Gaussian process regression modeling, a state recognition method combining a graph convolutional network and semi-supervised learning, and an operating state evaluation algorithm based on fuzzy entropy, constructs a closed-loop self-learning intelligent regulation process for industrial equipment, details the specific steps of predicting, identifying, evaluating, and regulating the operating state of the equipment under complex working conditions, and has the advantages of high prediction accuracy, strong robustness of state recognition, flexible controllability of evaluation results, and adaptive optimization of control strategies.

[0010] An intelligent control method for industrial equipment according to an embodiment of the present invention includes the following steps:

[0011] S1. Obtain the operating data of the industrial equipment, perform preprocessing, and construct a standardized state data set;

[0012] S2. Based on the standardized state data set, use the Gaussian process regression algorithm to model the equipment operating trend, generate a state prediction model within a specified time window, and output a state prediction value;

[0013] S3. Construct a graph structure representation model with the equipment operating state as nodes and the association relationship between equipment as edges, map the standardized state data into the graph structure, and use the graph convolutional network and semi-supervised learning method to jointly train the labeled state data and unlabeled state data, and output the working condition category and operating level corresponding to the current equipment operating state;

[0014] S4. Calculate the fuzzy membership function values corresponding to each moment based on the device operation data at the current moment and historical moments, calculate the fuzzy entropy index based on the membership function values, and generate a fuzzy entropy distribution curve of the device operation;

[0015] S5. Based on the state prediction value, operating condition category, operation level, and fuzzy entropy index, comprehensively evaluate the current device operation state and generate a state evaluation result;

[0016] S6. According to the state evaluation result, adjust the corresponding dynamic control strategy parameters, send a control instruction to the device control system, and perform real-time regulation operations;

[0017] S7. Collect the feedback data after the device executes the regulation, update the state dataset, and achieve closed-loop self-learning and iterative optimization of the control strategy.

[0018] Optionally, the operation data includes temperature, current, vibration frequency, energy consumption, and operation time, and the preprocessing includes outlier removal, denoising, and standardization.

[0019] Optionally, S2 specifically includes:

[0020] S21. Extract the target variable sequence Y = {y1, y2, …, y m} for modeling from the standardized state dataset, where y k is the device operation state value at the k-th moment, and m is the length of the time series;

[0021] S22. Construct the input variable set X = {x1, x2, …, x m}, where x i is the operation feature vector corresponding to the i-th moment, and the feature dimension is d, that is

[0022] S23. Based on the input variable set X and the target variable sequence Y, establish a Gaussian process regression model, and define the covariance function between the prediction variables as the kernel function:

[0023]

[0024] where k(x i , x j ) is the kernel function, is the signal variance of the kernel function, exp(·) is the natural exponential function with base e, l is the length scale hyperparameter, ||x i - x j || is the Euclidean distance between the input vectors, and x j is the operation feature vector corresponding to the j-th moment;

[0025] S24. Calculate the mean function μ(x * ) and covariance function Σ(x * ) based on the kernel function:

[0026]

[0027] where x * is the input feature vector at the prediction moment, is the observation noise variance, I is the identity matrix, K(X, X) is the covariance matrix composed of the kernel function k(x i , x j ), and k(x * , X) is the covariance vector between the prediction point and the training points;

[0028] S25. Output the state prediction value based on the mean function and covariance function to complete the prediction of the operating state of the target device within the specified time window.

[0029] Optionally, the state prediction value is calculated through the mean function and covariance function:

[0030]

[0031] where is the state prediction value at the prediction time point, μ(x * ) is the mean function, representing the point estimate prediction of x * , Σ(x * ) is the covariance function, representing the magnitude of the prediction uncertainty, ρ is the uncertainty modulation coefficient, v is the variance estimate value of the covariance function, calculated by the internal covariance propagation mechanism of the Gaussian process, ∈ is the stability constant to prevent the denominator from being zero, m is the length of the time series, ω i represents the weight parameter used to construct the auxiliary non - linear perturbation term, tanh(·) is the hyperbolic tangent function used to introduce non - linear perturbation features, <x * , x i > is the inner product of the input feature vector x * at the prediction moment and the operating feature vector x i corresponding to the i - th moment, b i is the offset term, and δ is the global adjustment bias used to correct the overall prediction trend.

[0032] Optionally, the specific content of S3 includes:

[0033] S31. Construct a graph - structure representation model G=(V, E), where V = {v1, v2, …, v n} represents the operating - state nodes of the device at each moment, and E = {e ij} represents any two state nodes vi The associated edge with v j and the total number of nodes is n;

[0034] S32. For each node v i Allocate a feature vector where h i is obtained by the intermediate layer mapping transformation of the operating parameter feature vector x i and the feature dimension is d;

[0035] S33. Construct an adjacency matrix where A ij = 1 indicates that there is a connection edge e i between node v j and node v ij , otherwise A ij = 0;

[0036] S34. According to the node feature matrix and the adjacency matrix A, use a graph convolutional network for feature propagation and aggregation, and calculate the updated feature matrix of each layer:

[0037]

[0038] where, H (l+1) is the updated feature matrix of the (l + 1)-th layer, H (l) is the updated feature matrix of the l-th layer, is the adjacency matrix with self-loops added, is 's degree matrix, I n is the n×n identity matrix, W (l) is the weight matrix of the l-th layer, σ(·) is the activation function, and the initial feature matrix H (0) = H;

[0039] S35. Set marker values for some nodes, and the remaining nodes are in an unmarked state. Combine the marked nodes and unmarked nodes for semi-supervised training, and optimize based on the cross-entropy loss function to obtain the working condition category label c i ∈C and the operating level label r i ∈R, where C is the preset set of working condition categories and R is the set of operating levels.

[0040] Optionally, the working condition category label c i ∈C indicates the working condition type to which the operating state of the device corresponding to node v i belongs, and the operating level label r i ∈R indicates the operating level of node v iThe load level corresponding to the operating state of the device, where C is a set of preset working condition categories, satisfying C = {c1, c2, …, c q}, R is a set of operating levels, satisfying R = {r1, r2, …, r p}, q and p are the number of working condition categories and the number of operating levels respectively, satisfying q ≥ 2 and p ≥ 2.

[0041] Optionally, the S4 specifically includes:

[0042] S41. Based on the target variable sequence Y, obtain the sequence of device operating state values {y t-T+1 , y t-T+2 , …, y t} at the current moment and the previous T moments, where y k is the device operating state value at the kth moment, t is the current moment, and T is the set time window length;

[0043] S42. According to the numerical range of y k at each moment, construct a fuzzy membership function:

[0044]

[0045] where μ k (y k ) is the membership degree of the operating state value y k at the kth moment, a k , b k , c k are the control points of the fuzzy membership function corresponding to the kth moment, satisfying a k < b k < c k ;

[0046] S43. Based on the membership degree sequence calculate the fuzzy entropy value within the time window:

[0047]

[0048] where FE is the fuzzy entropy value within the time window [t - T + 1, t], measuring the fuzzy uncertainty of the device operating state, and lnμ k (y k ) is the natural logarithm of the membership degree;

[0049] S44. Arrange the fuzzy entropy values calculated within different time windows in chronological order to form a fuzzy entropy distribution curve F = {FE t-T+1 , FE t-T+2 , …, FE t}, where FE tis the fuzzy entropy value corresponding to the j-th moment, which is used to describe the fuzzy change trend of the equipment operation state in the time dimension.

[0050] Optionally, the S5 specifically includes:

[0051] S51. Construct a comprehensive evaluation vector Among them, is the state prediction value at the prediction time point, c t is the working condition category label value at the current moment, which is normalized to a numerical level, r t is the operation level value at the current moment, which is normalized to a numerical level, FE t is the fuzzy entropy value corresponding to the current moment;

[0052] S52. Construct a non-linear comprehensive evaluation function with regularization constraints, in the form of a linear weighted combination:

[0053]

[0054] Among them, S t is the comprehensive evaluation score of the state at the current moment, w1 and w2 are the weight factors of the state prediction value and the working condition category, α, β, and γ are the combined modulation coefficients of each sub-function, λ is the entropy modulation factor, which is used to punish the state with high fuzzy uncertainty, Z is the normalization constant, which is obtained by accumulating α, β, and γ, and is used to perform scaling normalization processing on the evaluation result;

[0055] S53. Based on the comprehensive evaluation score of the state, divide the current operation state of the equipment into multiple fixed threshold intervals, and generate a state evaluation result label e t ∈E, where E is the preset set of operation state labels, and the state evaluation result is used to drive the generation of subsequent control strategies.

[0056] Optionally, the S53 specifically includes:

[0057] S531. Set the state evaluation threshold set Θ = {θ1, θ2, …, θ k}, where satisfies θ1 < θ2 < … < θ k , and the set of operation state labels is E = {e1, e2, …, e k};

[0058] S532. Based on the comprehensive evaluation score S t , construct a mapping function g(S t , Θ), which is defined as:

[0059]

[0060] Among them, e tAs the state evaluation result label, satisfying e t ∈E, e k is the evaluation result label at the k-th moment, e i is the evaluation result label at the i-th moment, is the indicator function, which returns 1 if the condition holds, otherwise returns 0, indicates that the current score exceeds the previous i - 1 thresholds. If S t >θ k , then directly assign it to e k ;

[0061] S533. Use the state label e t as the final state evaluation result to drive the generation of subsequent control strategies.

[0062] An intelligent control system for industrial equipment according to an embodiment of the present invention includes:

[0063] A data processing module, configured to obtain the operation data of industrial equipment, and perform missing value filling, outlier removal, denoising, time synchronization, and standardization processing on the obtained data to construct a standardized state data set;

[0064] A state prediction module, configured to model the equipment operation trend based on the standardized state data set by using a Gaussian process regression algorithm, generate a state prediction model within a specified time window, and output a state prediction value;

[0065] A graph structure modeling module, configured to construct a graph structure representation model with the equipment operation state as nodes and the association relationship between equipment as edges, map the standardized state data into the graph structure, and perform joint training on the labeled state data and unlabeled state data through a graph convolutional network and a semi-supervised learning method, and output the working condition category and operation level corresponding to the current equipment operation state;

[0066] A fuzzy entropy calculation module, configured to calculate the fuzzy membership function value corresponding to each moment based on the equipment operation data at the current moment and historical moments, and calculate the fuzzy entropy index based on the membership function value to generate a fuzzy entropy distribution curve of equipment operation;

[0067] A state evaluation module, configured to comprehensively evaluate the current equipment operation state based on the state prediction value, working condition category, operation level, and fuzzy entropy index, and generate a state evaluation result;

[0068] A strategy control module, configured to adjust the corresponding dynamic control strategy parameters according to the state evaluation result, send a control instruction to the equipment control system, and perform real-time regulation operations;

[0069] A feedback update module, which is used to collect the feedback data after the device executes regulation, update the status data set, and achieve closed-loop self-learning and iterative optimization of the control strategy.

[0070] The beneficial effects of the present invention are as follows:

[0071] First of all, by introducing the Gaussian process regression model, the present invention effectively improves the prediction accuracy of the operating status of industrial equipment. Compared with traditional linear regression or neural network methods, Gaussian process regression can not only perform high-confidence modeling under small sample conditions, but also provide variance estimation for the prediction results, enabling the control system to have stronger stability and forward-looking in the face of working condition fluctuations, thereby avoiding sudden abnormalities and response lags during the operation of the equipment.

[0072] Secondly, the present invention combines the graph convolutional network and the semi-supervised learning method to establish a graph model based on the structural relationship of device status, realizing the joint learning and recognition of the marked and unmarked device status. This method overcomes the common problem of insufficient status labels in the industrial field, makes full use of the structured information and temporal evolution law between devices, and realizes the highly robust recognition and refined classification of the operating status of complex device clusters, providing a clear and interpretable working condition basis for subsequent control decisions.

[0073] Finally, by constructing a fuzzy entropy calculation mechanism based on the fuzzy membership function, the present invention quantitatively evaluates the uncertainty of the device operating status, and fuses the state prediction results, working condition categories, operating levels, and fuzzy entropy indicators to form a comprehensive state evaluation system, driving the dynamic control strategy generation module to realize real-time parameter update. At the same time, the system constructs a complete feedback learning path, realizes the iterative optimization of the model by the device regulation behavior, establishes an intelligent control mechanism integrating closed-loop control and self-learning, and significantly improves the adaptive ability, stability, and intelligence level of the control system. Description of the Drawings

[0074] The drawings are used to provide a further understanding of the present invention, and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention, and do not constitute a limitation to the present invention. In the drawings:

[0075] Figure 1 is a flow chart of an intelligent control method for industrial equipment proposed by the present invention;

[0076] Figure 2 is a flow chart of graph structure construction and graph convolutional joint training of an intelligent control method for industrial equipment proposed by the present invention;

[0077] Figure 3 is a module structure diagram of an intelligent control system for industrial equipment proposed by the present invention. Detailed Embodiments

[0078] Now, the present invention will be further described in detail with reference to the accompanying drawings. These drawings are all simplified schematic diagrams, only schematically showing the basic structure of the present invention, so they only show the components related to the present invention.

[0079] Referring to Figure 1 and Figure 2 , an intelligent control method for industrial equipment includes the following steps:

[0080] S1. Obtain the operation data of the industrial equipment, perform preprocessing, and construct a standardized state data set;

[0081] S2. Based on the standardized state data set, use the Gaussian process regression algorithm to model the equipment operation trend, generate a state prediction model within a specified time window, and output state prediction values;

[0082] S3. Construct a graph structure representation model with the equipment operation state as nodes and the association relationship between equipment as edges, map the standardized state data into the graph structure, and use the graph convolutional network and semi-supervised learning method to jointly train the labeled state data and unlabeled state data, and output the working condition category and operation level corresponding to the current equipment operation state;

[0083] S4. Based on the equipment operation data at the current moment and historical moments, calculate the fuzzy membership function values corresponding to each moment, calculate the fuzzy entropy index based on the membership function values, and generate a fuzzy entropy distribution curve of the equipment operation;

[0084] S5. Based on the state prediction values, working condition categories, operation levels, and fuzzy entropy indexes, comprehensively evaluate the current equipment operation state, and generate a state evaluation result;

[0085] S6. According to the state evaluation result, adjust the corresponding dynamic control strategy parameters, send control instructions to the equipment control system, and perform real-time regulation operations;

[0086] S7. Collect the feedback data after the equipment executes the regulation, update the state data set, and realize closed-loop self-learning and iterative optimization of the control strategy.

[0087] The intelligent control method for industrial equipment proposed by the present invention constructs a complete process covering state perception, modeling prediction, intelligent identification, comprehensive evaluation, and dynamic control by introducing Gaussian process regression, graph convolutional network, fuzzy entropy evaluation, and closed-loop self-learning mechanism. This method can achieve high-precision prediction of equipment operation state, accurate identification of complex working conditions and uncertainty evaluation, and optimize the strategy through real-time control feedback, thereby greatly improving the operation efficiency, stability, and intelligent level of industrial equipment.

[0088] In this embodiment, the operating data includes temperature, current, vibration frequency, energy consumption, and operating time, and the preprocessing includes outlier removal, denoising, and standardization.

[0089] By defining the composition of the operating data and the preprocessing method, the present invention ensures the quality and consistency of the input data, effectively removes noise, outliers, and redundant information, improves the robustness and accuracy of subsequent modeling algorithms, and provides a reliable basis for equipment status perception and modeling.

[0090] In this embodiment, S2 specifically includes:

[0091] S21. Extract the target variable sequence Y = {y1, y2, …, y m} for modeling from the standardized state dataset, where y k is the equipment operating state value at the k-th moment, and m is the length of the time series;

[0092] S22. Construct the input variable set X = {x1, x2, …, x m}, where x i is the operating feature vector corresponding to the i-th moment, and the feature dimension is d, that is

[0093] S23. Based on the input variable set X and the target variable sequence Y, establish a Gaussian process regression model, and define the covariance function between the prediction variables as the kernel function:

[0094]

[0095] Among them, k(x i , x j ) is the kernel function, is the signal variance of the kernel function, exp(·) is the natural exponential function with e as the base, l is the length-scale hyperparameter, ||x i - x j || is the Euclidean distance between the input vectors, and x j is the operating feature vector corresponding to the j-th moment;

[0096] S24. Based on the kernel function, calculate the mean function μ(x * ) and the covariance function Σ(x * ) of the Gaussian process prediction distribution:

[0097]

[0098] Among them, x * is the input feature vector at the prediction moment, is the observation noise variance, I is the identity matrix, and K(X, X) is composed of the kernel function k(x i, x j ) The covariance matrix formed by, k(x * , X) is the covariance vector between the prediction point and the training point;

[0099] S25. Based on the mean function and the covariance function, output the state prediction value to complete the operation state prediction of the target device within the specified time window.

[0100] In the state prediction link of the present invention, a Gaussian process regression model is introduced, and combined with mathematical expression forms such as covariance functions and mean functions, high-confidence modeling and prediction of the future operation state of the device are carried out, which has the advantages of strong non-linear modeling ability, quantifiable uncertainty, and good generalization ability, and significantly improves the stability and practicality of the prediction.

[0101] In this embodiment, the state prediction value is calculated through the mean function and the covariance function:

[0102]

[0103] Among them, is the state prediction value at the prediction time point, μ(x * ) is the mean function, indicating the point estimate prediction of x * , Σ(x * ) is the covariance function, indicating the amplitude of the prediction uncertainty, ρ is the uncertainty modulation coefficient, v is the variance estimate value of the covariance function, calculated by the internal covariance propagation mechanism of the Gaussian process, ∈ is the stability constant to prevent the denominator from being zero, m is the length of the time series, ω i represents the weight parameter used to construct the auxiliary non-linear perturbation term, tanh(·) is the hyperbolic tangent function, used to introduce non-linear perturbation characteristics, <x * , x i > is the inner product of the input feature vector x * at the prediction moment and the operation feature vector x i corresponding to the i-th moment, b i is the offset term, and δ is the global adjustment bias, used to correct the overall prediction trend.

[0104] By introducing the uncertainty modulation term, the covariance propagation mechanism and the non-linear perturbation expression, the present invention effectively improves the sensitivity to abnormal trends and mutation risks during the state prediction process, and enhances the interpretability of the state prediction value and the decision-making security of the control system.

[0105] In this embodiment, the specific content of S3 includes:

[0106] S31. Construct a graph structure representation model G=(V, E), where V={v1, v2,..., v n}(indicating the operating state nodes of the device at each moment, E = {e ij}(indicating any two state nodes v i and v j The associated edge between them, and the total number of nodes is n;

[0107] S32. Assign a feature vector to each node v i where h is obtained by the intermediate layer mapping transformation of the operating parameter feature vector x i with a feature dimension of d; i

[0108] S33. Construct the adjacency matrixwhere A ij = 1 indicates that there is a connection edge e between node v i and node v j otherwise A ij ij = 0; ij

[0109] S34. According to the node feature matrixand the adjacency matrix A, use the graph convolutional network for feature propagation and aggregation to calculate the updated feature matrix of each layer:

[0110]

[0111] where H (l+1) is the updated feature matrix of the l+1 layer, H (l) is the updated feature matrix of the l layer, is the adjacency matrix with self-loops added, is 's degree matrix, I n is the n×n identity matrix, W (l) is the weight matrix of the l layer, σ(·) is the activation function, and the initial feature matrix H (0) = H;

[0112] S35. Set the marked values for some nodes, and the rest of the nodes are in the unmarked state. Jointly train the marked nodes and unmarked nodes for semi-supervised training, and optimize based on the cross-entropy loss function to obtain the working condition category label c i ∈C and the operating level label r i ∈R, where C is the preset set of working condition categories and R is the set of operating levels.

[0113] The present invention constructs a graph structure model with device states as nodes and associations between devices as edges, and combines a graph convolutional network for feature propagation and aggregation, which can deeply mine the structural relationships between device operating states, and improve the recognition accuracy through semi-supervised learning in the case of insufficient data labels, realizing a more stable and efficient device condition recognition ability.

[0114] In this embodiment, the condition category label c i ∈C represents the condition type to which the device operating state corresponding to node v i belongs, and the operation level label r i ∈R represents the load level of the device operating state corresponding to node v i . Among them, C is a preset condition category set, satisfying C={c1, c2, …, c q}, R is an operation level set, satisfying R={r1, r2, …, r p}, q and p are the numbers of condition categories and operation levels respectively, satisfying q≥2, p≥2.

[0115] By setting the combination mechanism of the condition category label and the operation level label, the present invention can finely divide the device operating state, not only identify the current operating environment of the device, but also quantify its load level, which helps to realize the configuration of differential control strategies and the fine management of devices.

[0116] In this embodiment, S4 specifically includes:

[0117] S41. Based on the target variable sequence Y, obtain the device operating state value sequence {y t-T+1 , y t-T+2 , …, y t} at the current moment and the previous T moments, where y k is the device operating state value at the kth moment, t is the current moment, and T is the set time window length;

[0118] S42. According to the numerical range of y k at each moment, construct a fuzzy membership function:

[0119]

[0120] where μ k (y k ) is the membership of the operating state value y k at the kth moment, a k , b k , c k are the control points of the fuzzy membership function corresponding to the kth moment, satisfying a k <b k <c k ;

[0121] S43. Calculate the fuzzy entropy value within the time window based on the membership degree sequence: Calculate the fuzzy entropy value within the time window:

[0122]

[0123] where FE is the fuzzy entropy value within the time window [t - T + 1, t], which measures the fuzzy uncertainty of the equipment operation state, and lnμ k (y k ) is the natural logarithm of the membership degree;

[0124] S44. Arrange the fuzzy entropy values calculated within different time windows in chronological order to form a fuzzy entropy distribution curve F = {FE t-T+1 , FE t-T+2 , …, FE t}, where FE t is the fuzzy entropy value corresponding to the j-th moment, which is used to describe the fuzzy change trend of the equipment operation state in the time dimension.

[0125] By adopting the fuzzy membership function and the fuzzy entropy calculation method, the present invention can effectively model the fuzziness and uncertainty of the operation state, overcome the defects of the traditional threshold classification method with strong rigidity and poor sensitivity, and improve the adaptability of the state assessment in a multi-source complex environment.

[0126] In this embodiment, the specific steps of S5 are as follows:

[0127] S51. Construct a comprehensive evaluation vector where, is the state prediction value at the prediction time point, c t is the working condition category label value at the current moment, which is converted into a numerical level after normalization, r t is the operation level value at the current moment, which is converted into a numerical level after normalization, and FE t is the fuzzy entropy value corresponding to the current moment;

[0128] S52. Construct a non-linear comprehensive evaluation function with regularization constraints, in the form of a linear weighted combination:

[0129]

[0130] where S t is the comprehensive evaluation score of the state at the current moment, w1 and w2 are the weight factors of the state prediction value and the working condition category, α, β, and γ are the combined modulation coefficients of each sub-function, λ is the entropy modulation factor, which is used to penalize the state with high fuzzy uncertainty, Z is the normalization constant, which is obtained by adding α, β, and γ, and is used to perform scaling normalization processing on the evaluation result;

[0131] S53. Based on the comprehensive state evaluation score, divide the current operating state of the device into multiple fixed threshold intervals to generate a state evaluation result label e t ∈E, where E is a preset set of operating state labels, and the state evaluation result is used to drive the generation of subsequent control strategies.

[0132] By constructing a non-linear, multi-factor weighted comprehensive state evaluation function, this invention fully integrates multiple dimensions such as state prediction values, working condition categories, operating levels, and fuzzy entropy indicators to achieve a comprehensive quantitative evaluation of the device's operating state, providing an accurate and flexible decision-making basis for dynamic control strategies.

[0133] In this embodiment, the S53 specifically includes:

[0134] S531. Set a set of state evaluation thresholds Θ = {θ1, θ2, …, θ k}, where satisfies θ1 < θ2 < … < θ k , and the set of operating state labels is E = {e1, e2, …, e k};

[0135] S532. Based on the comprehensive evaluation score S t , construct a mapping function g(S t , Θ), defined as:

[0136]

[0137] where e t is the state evaluation result label, satisfying e t ∈E, e k is the evaluation result label at the k-th moment, e i is the evaluation result label at the i-th moment, is an indicator function that returns 1 if the condition holds, otherwise returns 0, indicates that the current score exceeds the previous i - 1 thresholds. If S t > θ k , then directly assign it to e k ;

[0138] S533. Use the state label e t as the final state evaluation result to drive the generation of subsequent control strategies.

[0139] By designing a complex state label mapping function and a multi-threshold segmentation mechanism, this invention can intelligently divide the device state level according to the comprehensive evaluation score, ensuring that the control system has a fine perception ability of operation risks and efficiency differences, and further enhancing the dynamic adaptability of the control strategy.

[0140] ReferenceFigure 3 , an intelligent control system for industrial equipment, comprising:

[0141] A data processing module, configured to obtain the operation data of the industrial equipment, and perform missing value filling, outlier removal, denoising, time synchronization and normalization processing on the obtained data to construct a normalized status data set;

[0142] A status prediction module, configured to model the equipment operation trend based on the normalized status data set by using the Gaussian process regression algorithm, generate a status prediction model within a specified time window, and output status prediction values;

[0143] A graph structure modeling module, configured to construct a graph structure representation model with the equipment operation status as nodes and the association relationship between equipment as edges, map the normalized status data into the graph structure, and perform joint training on the labeled status data and unlabeled status data through a graph convolutional network and a semi-supervised learning method, and output the working condition category and operation level corresponding to the current equipment operation status;

[0144] A fuzzy entropy calculation module, configured to calculate the fuzzy membership function value corresponding to each moment based on the equipment operation data at the current moment and historical moments, and calculate the fuzzy entropy index based on the membership function value to generate a fuzzy entropy distribution curve of the equipment operation;

[0145] A status evaluation module, configured to comprehensively evaluate the current equipment operation status based on the status prediction value, working condition category, operation level and fuzzy entropy index, and generate a status evaluation result;

[0146] A strategy control module, configured to adjust the corresponding dynamic control strategy parameters according to the status evaluation result, send control instructions to the equipment control system, and perform real-time regulation operations;

[0147] A feedback update module, configured to collect the feedback data after the equipment executes the regulation, update the status data set, and realize closed-loop self-learning and iterative optimization of the control strategy.

[0148] The intelligent control system for industrial equipment provided by the present invention has a complete structure and clear modules, and can realize the whole-process closed-loop management from data acquisition, status modeling, intelligent recognition, status evaluation to strategy regulation and self-learning feedback. The system design is applicable to a variety of industrial scenarios, and can significantly improve the intelligent level, response speed and operation stability of equipment control.

[0149] Example 1:

[0150] To verify the feasibility of the present invention in implementation, the present invention is applied to the high-load CNC machining workshop of a heavy manufacturing factory. This workshop is equipped with multiple large CNC machine tools, undertaking the processing tasks of high-precision metal parts. The daily operation time of the equipment reaches more than 18 hours, and extremely high requirements are placed on the equipment stability and control accuracy. However, in the actual production process, this workshop frequently encounters problems such as sudden equipment shutdowns, lag in regulation and control responses, and abnormal fluctuations in energy consumption. Through research on this production environment, it is found that the means of equipment status monitoring lag, and the control strategy lacks intelligence, unable to identify changes in working conditions and potential faults in real time, which greatly restricts production efficiency.

[0151] In this embodiment, three CNC machining devices of model MGC-800 are selected as the test objects, and the intelligent control method and system for industrial equipment described in the present invention are deployed to collect and analyze their operation data in real time. During the deployment process, five types of sensors, namely temperature, current, voltage, vibration frequency, and machining time, are installed on each device, and data cleaning, anomaly elimination, and normalization and standardization processing are completed through the edge computing module to construct a unified status data set. The system automatically models the collaborative relationship between multiple devices, and takes the interaction of the operation status nodes of each device with other devices in task scheduling, energy distribution, etc. as edges to construct a graph structure, which is input into the graph convolutional network for status recognition.

[0152] To improve the prediction ability, the system uses Gaussian process regression modeling for the status data collected during the operation of each device, and predicts the operation trend for the next 30 minutes before each round of task processing. The system automatically outputs the change trends of the predicted temperature, current, and vibration data every 5 minutes, and combines the operation level and working condition category labels identified by the graph convolutional network to perform fuzzy entropy evaluation on the status. The uncertainty distribution at each time point is extracted through the fuzzy membership function, and then the overall fuzzy degree is quantified using the fuzzy entropy function to realize the evaluation of the stability of the current operation state.

[0153] This system has been running on-site for 15 consecutive days, 18 hours a day, recording rich operation data, and has conducted a comparative analysis with the traditional control mode without using this system. The following is the summary table of the core indicators recorded during the implementation process:

[0154] Table 1 Comparison table of the application effects of the present invention in CNC equipment control

[0155]

[0156]

[0157] The results show that after the deployment of the present invention, the average downtime rate of the equipment has been reduced from the original 7.3% to 2.1%, the processing cycle of per unit product has been shortened from an average of 26.5 minutes to 24.2 minutes, and the volatility of equipment energy consumption has been reduced by 31.6%. At the same time, through the state prediction and fuzzy entropy early warning mechanism, 4 slight abnormal trends of the equipment have been successfully captured and the control strategy adjustment has been triggered in advance, avoiding potential interruption risks.

[0158] As can be seen from the above data, the deployment of the present invention in the actual industrial environment has significantly improved the equipment operation efficiency, energy-saving effect and system stability, especially showing excellent performance in abnormal trend prediction, state recognition accuracy and real-time regulation. Through the control mechanism driven by graph structure learning and fuzzy entropy, the system not only improves the adaptability to complex working conditions, but also realizes the self-optimization of the control strategy in the feedback learning, and has good industrial application prospects and promotion value.

[0159] The above is only the preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention, according to the technical solution of the present invention and its inventive concept, makes equivalent replacement or change, and should be covered by the protection scope of the present invention.

Claims

1. An intelligent control method for industrial equipment, characterized in that, It includes the following steps: S1. Obtain the operation data of industrial equipment, perform preprocessing, and construct a standardized status data set; S2. Based on the standardized status data set, use the Gaussian process regression algorithm to model the equipment operation trend, generate a status prediction model within a specified time window, and output status prediction values; S3. Construct a graph structure representation model with the equipment operation status as nodes and the association relationship between equipment as edges, map the standardized status data into the graph structure, and use the graph convolutional network and semi-supervised learning method to jointly train the labeled status data and unlabeled status data, and output the working condition category and operation level corresponding to the current equipment operation status; S4. Based on the equipment operation data at the current moment and historical moments, calculate the fuzzy membership function values corresponding to each moment, calculate the fuzzy entropy index based on the membership function values, and generate a fuzzy entropy distribution curve of equipment operation; S5. Based on the status prediction values, working condition category, operation level, and fuzzy entropy index, comprehensively evaluate the current equipment operation status and generate a status evaluation result; S6. According to the status evaluation result, adjust the corresponding dynamic control strategy parameters, send control instructions to the equipment control system, and perform real-time regulation operations; S7. Collect the feedback data after the equipment executes the regulation, update the status data set, and realize closed-loop self-learning and iterative optimization of the control strategy.

2. The intelligent control method for an industrial device according to claim 1, characterized in that, The operation data includes temperature, current, vibration frequency, energy consumption, and operation time, and the preprocessing includes outlier removal, denoising, and standardization.

3. The intelligent control method for an industrial device according to claim 1, wherein, The specific content of S2 includes: S21. Extract the target variable sequence Y = {y1, y2, …, y m} for modeling from the standardized state dataset, where y k is the device operation status value at the k-th moment, and m is the length of the time series; S22. Construct the input variable set \(X = \{x_1, x_2, \ldots, x\) m \}, where \(x\) i is the operation feature vector corresponding to the \(i\)-th moment, and the feature dimension is \(d\), that is S23. Based on the input variable set X and the target variable sequence Y, establish a Gaussian process regression model, and define the covariance function between prediction variables as the kernel function: where \(k(x\) i , x\) j ) is the kernel function, is the signal variance of the kernel function, exp(·) is the natural exponential function with base e, l is the length-scale hyperparameter, \(\|x\) i - x\) j \(\|\) is the Euclidean distance between the input vectors, and \(x\) j is the operating feature vector corresponding to the \(j\)-th moment; S24. Calculate the mean function μ(x * ) and covariance function Σ(x * ) based on the kernel function: where x * is the input feature vector at the prediction time, is the observation noise variance, I is the identity matrix, K(X,X) is the covariance matrix formed by the kernel function k(x i , x j ), k(x * , X) is the covariance vector between the prediction point and the training points; S25. Based on the mean function and covariance function, output the status prediction values, and complete the operation status prediction of the target equipment within the specified time window.

4. An intelligent control method for industrial equipment according to claim 3, characterized in that, The status prediction values are calculated through the mean function and covariance function: Among them, is the state prediction value at the predicted time point, μ(x * ) is the mean function, representing the point estimate prediction of x * , Σ(x * ) is the covariance function, representing the amplitude of the prediction uncertainty, ρ is the uncertainty modulation coefficient, v is the variance estimate of the covariance function, calculated by the internal covariance propagation mechanism of the Gaussian process, ∈ is the stability constant to prevent the denominator from being zero, m is the length of the time series, ω i represents the weight parameter used to construct the auxiliary non-linear perturbation term, tanh(·) is the hyperbolic tangent function, used to introduce non-linear perturbation characteristics, <x * , x i > is the inner product of the input feature vector x * at the prediction moment and the operating feature vector x i corresponding to the i-th moment, b i is the offset term, and δ is the global adjustment bias, used to correct the overall prediction trend.

5. An intelligent control method for industrial equipment according to claim 1, characterized in that, The specific content of S3 includes: S31. Construct a graph structure representation model G=(V, E), where V = {v1, v2, …, v n} represents the operation state nodes of the device at each moment, and E = {e ij} represents the associated edges between any two state nodes v i and v j , and the total number of nodes is n; S32. For each node v i allocate a feature vector where h i is obtained by mapping transformation of the operating parameter feature vector x i through the middle layer, and the feature dimension is d; S33. Construct the adjacency matrix where A ij = 1 indicates that there is a connecting edge e i between node v j and node v ij , otherwise A ij = 0; S34. According to the node feature matrix and the adjacency matrix A, use a graph convolutional network for feature propagation and aggregation to calculate the updated feature matrix for each layer: Among them, H (l+1) is the updated feature matrix of the (l + 1)-th layer, and H (l) is the updated feature matrix of the l-th layer. is the adjacency matrix with self-loops added. is 's degree matrix, and I n is the n×n identity matrix. W (l) is the weight matrix of the l-th layer, σ(·) is the activation function, and the initial feature matrix H (0) = H; S35. Set marker values for some nodes, and the remaining nodes are in an unmarked state. Combine the marked nodes and unmarked nodes for semi-supervised training, optimize based on the cross-entropy loss function, and obtain the operating condition class label c i ∈ C and the operating level label r i ∈ R, where C is the preset operating condition class set and R is the operating level set.

6. An intelligent control method for industrial equipment according to claim 5, characterized in that, The working condition category label c i ∈C indicates that the node v i corresponds to the working condition type to which the device operating state belongs. The operating level label r i ∈R indicates that the node v i corresponds to the load level of the device operating state, where C is a preset set of working condition categories, satisfying C = {c1, c2, …, c q}, R is a set of operating levels, satisfying R = {r1, r2, …, r p}, q and p are the numbers of working condition categories and operating levels respectively, satisfying q≥2, p≥2.

7. An intelligent control method for industrial equipment according to claim 1, characterized in that, The specific content of S4 includes: S41. Based on the target variable sequence Y, obtain the device operation status value sequence {y t-T+1 , y t-T+2 , …, y t} at the current moment and the previous T moments, where y k is the device operation status value at the k-th moment, t is the current moment, and T is the set time window length; S42. Construct a fuzzy membership function according to the numerical range at each moment y k : where, μ k (y k ) is the membership degree of the operating state value y k at the k-th moment, and a k , b k , c k are the control points of the fuzzy membership function corresponding to the k-th moment, satisfying a k < b k < c k ; S43. Based on the membership degree sequence Calculate the fuzzy entropy value within the time window: Among them, FE is the fuzzy entropy value within the time window [t - T + 1, t], which measures the fuzzy uncertainty of the device operation state, and lnμ k (y k ) is the natural logarithm of the membership degree; S44. Arrange the fuzzy entropy values calculated within different time windows in chronological order to form a fuzzy entropy distribution curve F = {FE t-T+1 , FE t-T+2 , …, FE t}, where FE t is the fuzzy entropy value corresponding to the j-th moment, which is used to describe the fuzzy change trend of the device operation state in the time dimension.

8. An intelligent control method for industrial equipment according to claim 1, characterized in that The specific content of S5 includes: S51. Construct a comprehensive evaluation vector Among them, is the state prediction value at the prediction time point, c t is the working condition category label value at the current moment, which is a numerical level after normalization operation, r t is the operation level value at the current moment, which is a numerical level after normalization operation, FE t is the fuzzy entropy value corresponding to the current moment; S52. Construct a non-linear comprehensive evaluation function with regularization constraints, in the form of a linear weighted combination: Among them, S t is the comprehensive evaluation score of the state at the current moment, w1 and w2 are the weight factors of the state prediction value and the working condition category, α, β, and γ are the combined modulation coefficients of each sub-function, λ is the entropy modulation factor, which is used to punish the state with high fuzzy uncertainty, Z is the normalization constant, which is obtained by adding α, β, and γ, and is used to perform scaling normalization processing on the evaluation result; S53. Based on the comprehensive state evaluation score, divide the current operating state of the device into multiple fixed threshold intervals, and generate a state evaluation result label e t ∈E, where E is a preset set of operating state labels, and the state evaluation result is used to drive the generation of subsequent control strategies.

9. The intelligent control method for an industrial device according to claim 8, characterized in that The specific content of S53 includes: S531. Set the set of state evaluation thresholds Θ = {θ1, θ2, …, θ k}, where θ1 < θ2 < … < θ k is satisfied, and the set of operating state labels is E = {e1, e2, …, e k}; S532. Based on the comprehensive evaluation score S t , construct the mapping function g(S t , Θ), which is defined as: where e t is the status evaluation result label, satisfying e t ∈E, e k is the evaluation result label at the k-th moment, e i is the evaluation result label at the i-th moment, is the indicator function, which returns 1 if the condition holds and 0 otherwise, indicates that the current score exceeds the previous i - 1 thresholds. If S t > θ k , then directly assign it to e k ; S533. Use the status label e t as the final status evaluation result to drive the generation of subsequent control strategies.

10. An intelligent control system for industrial equipment, which executes an intelligent control method for industrial equipment according to any one of claims 1 to 9, characterized in that, It includes: A data processing module, which is used to obtain the operation data of industrial equipment, fill in missing values, remove outliers, denoise, synchronize time, and standardize the obtained data, and construct a standardized status data set; A status prediction module, which is used to, based on the standardized status data set, use the Gaussian process regression algorithm to model the equipment operation trend, generate a status prediction model within a specified time window, and output status prediction values; A graph structure modeling module, which is used to construct a graph structure representation model with the equipment operation status as nodes and the association relationship between equipment as edges, map the standardized status data into the graph structure, and jointly train the labeled status data and unlabeled status data through the graph convolutional network and semi-supervised learning method, and output the working condition category and operation level corresponding to the current equipment operation status; The fuzzy entropy calculation module is used to calculate the fuzzy membership function values corresponding to each moment based on the device operation data at the current moment and historical moments, and calculate the fuzzy entropy index based on the membership function values to generate the device operation fuzzy entropy distribution curve; The state evaluation module is used to comprehensively evaluate the current device operation state based on the state prediction value, working condition category, operation level, and fuzzy entropy index to generate a state evaluation result; The strategy control module is used to adjust the corresponding dynamic control strategy parameters according to the state evaluation result, send control instructions to the device control system, and perform real-time regulation operations; The feedback update module is used to collect the feedback data after the device executes the regulation, update the state data set, and realize closed-loop self-learning and iterative optimization of the control strategy.

Images