Double-layer fault diagnosis method and system for industrial interconnection system
By applying state-space modeling, Granger causality analysis, and alternating direction multiplier method in industrial interconnection systems, a distributed residual generator is constructed, which solves the problems of accuracy and complexity in fault diagnosis in existing technologies and achieves efficient fault source identification and location.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- UNIV OF SCI & TECH BEIJING
- Filing Date
- 2025-12-03
- Publication Date
- 2026-04-21
AI Technical Summary
Existing fault diagnosis technologies struggle to achieve efficient and accurate fault source localization in industrial interconnected systems. In particular, when facing reliability challenges involving multiple intertwined sources, existing methods suffer from decreased diagnostic sensitivity, high false alarm rates, and the risk of unplanned downtime. Furthermore, mechanistic modeling and data-driven methods fail to fully leverage their complementary advantages.
Based on the state-space model of the industrial interconnection system, a distributed residual generator is constructed through distributed subspace identification, Granger causality analysis and alternating direction multiplier method to realize fault detection and location, and combined with a centralized processing unit to identify the fault source subsystem.
It improves the accuracy of fault diagnosis and reduces computational complexity, enabling more accurate capture of system dynamic characteristics, rapid identification and location of fault sources, and reduction of unplanned downtime risks.
Smart Images
Figure CN121901602A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of fault diagnosis technology for industrial interconnection systems, and in particular to a two-layer fault diagnosis method and system for industrial interconnection systems. Background Technology
[0002] Industrial Internet systems, as key infrastructure driving the digital and intelligent transformation of manufacturing, achieve comprehensive interconnection and intelligent collaboration across the entire production process through the deep integration of information technology and physical processes. However, while achieving efficient and flexible operation, these systems also face multi-source reliability challenges due to their highly integrated and complexly coupled architecture. At the physical level, various sensors, actuators, controllers, and process equipment, during long-term continuous operation, are affected by multiple factors such as mechanical wear, material aging, electromagnetic interference, and environmental disturbances, generally exhibiting progressive or sudden fault characteristics such as accuracy degradation, response lag, and intermittent instability. These physical anomalies often have strong correlation and transmission; the performance degradation of local components can cascade through production line topology and control loops, gradually evolving into overall system performance degradation or functional failure. In particular, the nonlinear dynamic characteristics and time-varying operating conditions prevalent in physical systems further increase the complexity of state identification and fault attribution. If critical equipment anomalies are not captured and located in a timely manner, it will directly lead to process deviations, energy efficiency degradation, or even production interruptions, posing a substantial threat to the continuity and economy of industrial operations.
[0003] Against this backdrop, building an efficient and accurate fault diagnosis mechanism is of significant necessity and urgency for ensuring the stable operation of industrial interconnected systems. With the continuous increase in system integration and process complexity, the physical processes encompass numerous coupled equipment units, nested control structures, and continuous material flow. Even minor anomalies in any link, such as sensor drift, actuator jamming, or controller parameter inaccuracies, can be amplified cascaded through multi-loop interactions and dynamic responses within the system. Existing fault diagnosis methods mostly focus on monitoring and analyzing single-point signals or independent equipment, lacking the overall modeling capability for system-level fault propagation mechanisms and cross-equipment correlations, making it difficult to accurately locate the fault source. Taking the common flow control anomalies in process industries as an example, although the symptoms may appear uniform, the underlying causes may involve multiple physical factors such as valve body mechanical structure wear, reduced pump output capacity, and detection unit calibration drift. Relying solely on localized information for judgment can easily lead to one-sided and outdated diagnostic conclusions, failing to effectively guide maintenance decisions and potentially causing unplanned downtime due to improper handling, exacerbating production losses and safety risks.
[0004] Existing fault diagnosis technologies are mainly divided into two mainstream paradigms: model-based and data-driven. Model-based methods rely on precise mechanistic models of the system, achieving fault detection and isolation by analyzing the residuals between the model and measured data. They are characterized by transparent diagnostic logic and strong interpretability of results. However, industrial interconnected systems generally exhibit complex characteristics such as unmodeled dynamics, time-varying dynamics, and multi-loop coupling, making accurate modeling extremely difficult, leading to decreased diagnostic sensitivity and increased false alarm rates. Data-driven methods, on the other hand, break free from the dependence on mechanistic models, using algorithms such as machine learning and deep learning to mine fault characteristics and operational patterns from historical data, demonstrating significant advantages in identifying the state of complex systems. However, these methods are highly dependent on the quality and completeness of training data, the internal reasoning mechanism of the model is opaque, and the generalization ability is limited when facing unknown operating conditions or small-sample fault scenarios. Currently, the two methodologies are relatively independent in terms of theoretical foundation and application path, failing to fully leverage the complementary advantages of prior mechanistic knowledge and data-driven intelligence, thus restricting further improvements in the accuracy and practicality of fault diagnosis in industrial interconnected systems. Summary of the Invention
[0005] To address the problems in the prior art, this invention provides a two-layer fault diagnosis method and system for industrial interconnection systems. First, based on the state-space models of each subsystem in the industrial interconnection system, the invention derives the input-output data of each subsystem and utilizes this data to achieve distributed identification of the subspace parameter matrix, aiming to fully leverage the complementary advantages of mechanism modeling and data-driven approaches. Second, the Granger causality analysis method is used to calculate the coupling connections between subsystems, and the alternating direction multiplier method is combined to achieve distributed optimization of the subspace parameter matrix. Finally, a distributed residual generator is designed and residuals are evaluated at the subsystem level to achieve distributed fault detection in the industrial interconnection system. Simultaneously, the residual signals of each subsystem are transmitted to a centralized processing unit. When a fault is detected, the source subsystem is identified by calculating the comprehensive fault score of each subsystem. To achieve the above objectives, the technical solution is as follows:
[0006] On one hand, the present invention provides a two-layer fault diagnosis method for industrial interconnection systems, the method comprising: S1. Based on the state-space expressions of each subsystem of the industrial interconnection system, obtain the input and output data of each subsystem of the industrial interconnection system; S2. Based on the input and output data of each subsystem of the industrial interconnection system, the past input and output parameter matrix and future input parameter matrix related to the subsystem are obtained through the distributed subspace identification method. S3. Based on the input and output data of each subsystem of the industrial interconnection system, the coupling connection relationship between each subsystem is obtained using the multivariate Granger causal analysis method. S4. Based on the coupling relationship between the subsystems, the alternating direction multiplier method is used to optimize the past input-output parameter matrix and the future input parameter matrix related to the subsystem, so as to obtain the optimized past input-output parameter matrix and the optimized future input parameter matrix. S5. Based on the optimized past input-output parameter matrix and the optimized future input parameter matrix, construct a distributed residual generator to obtain the distributed fault detection status of the industrial interconnection system. S6. Based on the distributed fault detection of the industrial interconnection system, the fault source subsystem is obtained by processing the characteristic changes of each subsystem of the industrial interconnection system.
[0007] Optionally, in S1, the input and output data of each subsystem of the industrial interconnection system are obtained based on the state-space expressions of each subsystem, including: S11. Based on the state-space expressions of each subsystem of the industrial interconnection system, the improved state-space model of the industrial interconnection system is obtained. S12. Based on the improved state space model of the industrial interconnection system, the input-output model of the industrial interconnection system is obtained. S13. Based on the input-output model of the industrial interconnection system, obtain the input-output dataset model of the industrial interconnection system. S14. Based on the input-output dataset model of the industrial interconnection system, the input-output data of each subsystem of the industrial interconnection system are obtained through decomposition.
[0008] Optionally, in S2, based on the input and output data of each subsystem of the industrial interconnection system, a distributed subspace identification method is used to obtain the past input and output parameter matrix and the future input parameter matrix related to the subsystem, including: S21. Based on the input and output data of each subsystem of the industrial interconnection system, the Hankel matrix is obtained by performing the least squares method. S22. Based on the Hankel matrix, perform LQ decomposition to obtain the decomposed numerical matrix; S23. Based on the numerical matrix after decomposition, obtain the past input / output parameter matrix and the future input parameter matrix related to the subsystem.
[0009] Optionally, in S3, based on the input and output data of each subsystem of the industrial interconnection system, and using a multivariate Granger causality analysis method, the coupling connection relationships between the subsystems are obtained, including: S31. Based on the input and output data of each subsystem of the industrial interconnection system, establish an unrestricted model and a restricted model; S32. Based on the unrestricted model and the restricted model, calculate the sum of squared residuals of the unrestricted model and the sum of squared residuals of the restricted model. S33. Based on the sum of squared residuals of the unrestricted model and the sum of squared residuals of the restricted model, obtain the statistic. S34. Based on this statistic, the standardized coupling strength of each subsystem is obtained; S35. Based on the standardized coupling strength of each subsystem, the coupling connection relationship between each subsystem is obtained.
[0010] Optionally, in S4, based on the coupling relationship between the subsystems, the alternating direction multiplier method is used to optimize the past input / output parameter matrix and future input parameter matrix related to the subsystem, resulting in the optimized past input / output parameter matrix and optimized future input parameter matrix, including: S41. Based on the coupling relationship between the subsystems, introduce auxiliary variables to construct the augmented Lagrange function; S42. Based on the augmented Lagrange function, the alternating direction multiplier method is used for iterative solution to obtain the auxiliary variables of past data coefficients and future input coefficients. S43. Based on the past data coefficient auxiliary variable and the future input coefficient auxiliary variable, the original residual norm and the dual residual norm are obtained through calculation. S44. Iterate and converge the original residual norm and the dual residual norm. When both the original residual norm and the dual residual norm are less than the preset threshold, the optimized past input-output parameter matrix and the optimized future input parameter matrix are obtained.
[0011] Optionally, in S5, a distributed residual generator is constructed based on the optimized past input-output parameter matrix and the optimized future input parameter matrix to obtain the distributed fault detection status of the industrial interconnection system, including: S51. Based on the optimized past input-output parameter matrix and the optimized future input parameter matrix, construct the residual generator for each subsystem; S52. Based on the residual generator of each subsystem, perform residual evaluation to obtain the residual statistics of each subsystem; S53. Based on the set false alarm rate, the alarm threshold is calculated using the chi-square distribution table. S54. Compare the alarm threshold with the residual statistics of each subsystem to obtain the distributed fault detection status of the industrial interconnection system.
[0012] Optionally, in S6, based on the distributed fault detection of the industrial interconnection system, the fault source subsystem is obtained by processing the characteristic changes of each subsystem of the industrial interconnection system, including: S61. Based on the distributed fault detection of the industrial interconnection system, obtain the residual signals of each subsystem within the time interval and the residual signals of each subsystem within the time window. S62. Based on the residual signals of each subsystem within the time interval, obtain the benchmark mean and benchmark standard deviation of the residual signals of each subsystem. S63. Based on the residual signals of each subsystem within the time window, obtain the window mean and window standard deviation of the residual signals of each subsystem. S64. Based on the baseline mean and baseline standard deviation of the residual signals of each subsystem and the window mean and window standard deviation of the residual signals of each subsystem, obtain the change in the statistical characteristics of each subsystem. S65. Based on the changes in the statistical characteristics of each subsystem, the fault source subsystem is obtained through the comprehensive fault score of each subsystem.
[0013] On the other hand, the present invention provides a two-layer fault diagnosis system for industrial interconnection systems, which is applied to a two-layer fault diagnosis method for industrial interconnection systems, the system comprising: The first acquisition module is used to obtain the input and output data of each subsystem of the industrial interconnection system based on the state space expression of each subsystem of the industrial interconnection system. The parameter matrix module is used to obtain the past input / output parameter matrix and future input parameter matrix related to the subsystems based on the input / output data of each subsystem of the industrial interconnection system through a distributed subspace identification method. The coupling connection module is used to obtain the coupling connection relationship between the subsystems based on the input and output data of each subsystem of the industrial interconnection system and the multivariate Granger causal analysis method. The optimization module is used to optimize the past input-output parameter matrix and future input parameter matrix associated with each subsystem using the alternating direction multiplier method, based on the coupling connection relationship between the subsystems, to obtain the optimized past input-output parameter matrix and the optimized future input parameter matrix. The fault monitoring module is used to construct a distributed residual generator based on the optimized past input-output parameter matrix and the optimized future input parameter matrix to obtain the distributed fault detection status of the industrial interconnection system. The fault location module is used to identify the fault source subsystem by processing the characteristic changes of each subsystem of the industrial interconnection system based on the distributed fault detection of the industrial interconnection system.
[0014] Compared with the prior art, the technical solution of the present invention has at least the following beneficial effects: The above-mentioned scheme has several advantages. First, based on the state-space model of the industrial internet system, it derives a distributed subspace identification method for the industrial internet system, leveraging the complementary advantages of mechanism modeling and data-driven approaches. Second, it uses Granger causality analysis to calculate the causal coupling strength between subsystems, comprehensively reflecting the coupling relationships between them. Third, it uses the alternating direction multiplier method to perform distributed optimization of the subsystem subspace parameter matrix, which can more accurately capture the dynamic characteristics of the system. Fourth, it combines the distributed fault detection method on the subsystem side with the fault location method of the centralized processing unit, reducing the computational complexity of the fault diagnosis method for the industrial internet system while ensuring the accuracy of fault detection / location. Attached Figure Description
[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0016] Figure 1 This is a flowchart of an embodiment of the two-layer fault diagnosis method for industrial interconnection systems of the present invention; Figure 2 This is a flowchart illustrating the process of obtaining input and output data of each subsystem of an industrial interconnection system in an embodiment of the two-layer fault diagnosis method for industrial interconnection systems of the present invention. Figure 3 This is a flowchart illustrating the process of obtaining the past input / output parameter matrix and the future input parameter matrix related to the subsystem in an embodiment of the two-layer fault diagnosis method for industrial interconnection systems of the present invention. Figure 4 This is a flowchart illustrating the coupling connection relationship between subsystems in an embodiment of the two-layer fault diagnosis method for industrial interconnection systems of the present invention. Figure 5 This is a flowchart illustrating the process of obtaining the optimized past input / output parameter matrix and the optimized future input parameter matrix in an embodiment of the two-layer fault diagnosis method for industrial interconnection systems of the present invention. Figure 6 This is a flowchart illustrating the distributed fault detection status of an industrial interconnection system in an embodiment of the two-layer fault diagnosis method for industrial interconnection systems of the present invention. Figure 7 This is a flowchart illustrating the process of obtaining the fault source subsystem in an embodiment of the two-layer fault diagnosis method for industrial interconnection systems of the present invention. Figure 8 This is a system block diagram of an embodiment of the two-layer fault diagnosis system for industrial interconnection systems of the present invention. Detailed Implementation
[0017] The technical solution of the present invention will now be described with reference to the accompanying drawings.
[0018] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.
[0019] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.
[0020] like Figure 1 The flowchart shown is an embodiment of the two-layer fault diagnosis method for industrial interconnection systems according to the present invention. The present invention provides a two-layer fault diagnosis method for industrial interconnection systems, which is implemented by a two-layer fault diagnosis system for industrial interconnection systems. The method includes: S1. Based on the state-space expressions of each subsystem of the industrial interconnection system, obtain the input and output data of each subsystem of the industrial interconnection system; Specifically, such as Figure 2 The flowchart shown in this embodiment of the two-layer fault diagnosis method for industrial interconnection systems of the present invention illustrates the process of obtaining input and output data of each subsystem of the industrial interconnection system. In step S1, the input and output data of each subsystem of the industrial interconnection system are obtained based on the state-space expressions of each subsystem, including: S11. Based on the state-space expressions of each subsystem of the industrial interconnection system, the improved state-space model of the industrial interconnection system is obtained. Furthermore, each subsystem can be written in the following linear time-invariant discrete-time state-space representation: (1) In the formula, Indicates the first i The system state of each subsystem; Indicates the first i System inputs for each subsystem; Indicates the first i The system output of each subsystem; Indicates the first i Process noise of each subsystem; Indicates the first i Measurement noise of each subsystem; Indicates the first j The system state of each subsystem; Indicates the first k At the [time]th moment i The system state of each subsystem; Indicates the first k At the [time]th moment i System inputs for each subsystem; Indicates the first k At the [time]th moment i The system output of each subsystem; Indicates the first k At the [time]th moment i Process noise of each subsystem; Indicates the first k At the [time]th moment i Measurement noise of each subsystem; Indicates the first k+1 At the [time]th moment i The system state of each subsystem; Indicates the relationship with the first i A set of interconnected subsystems; Indicates the first i The local state matrix of each subsystem; Indicates the first i The input matrix of each subsystem; Indicates the first i The output matrix of each subsystem; Indicates the first i The direct transfer matrix of each subsystem; Indicates the first i Subsystem and the first j The connection matrix between the subsystems.
[0021] The state-space model for industrial interconnection systems is as follows: (2) in, , , , .
[0022] In the formula, Indicates the system status of the industrial internet system; This represents the system input of the industrial internet system; This represents the system output of the industrial interconnection system; This represents the process noise of an industrial interconnected system; This indicates the measurement noise of an industrial interconnected system; Indicates the first k The system status of the industrial internet system at any given moment; Indicates the first k The system input of the industrial internet system at any given moment; Indicates the first k The system output of the industrial interconnection system at any given moment; Indicates the first k Process noise of an industrial interconnection system at any given moment; Indicates the first k Measurement noise of industrial interconnection systems at any given moment; Indicates the first k+1 The system status of the industrial internet system at any given moment; Represents the state matrix of an industrial interconnection system; Represents the input matrix of an industrial interconnection system; This represents the output matrix of an industrial interconnection system; This represents the direct transfer matrix of an industrial interconnection system.
[0023] S12. Based on the improved state space model of the industrial interconnection system, the input-output model of the industrial interconnection system is obtained. Furthermore, the input-output model of an industrial internet system is as follows: (3) in, , , , , , , .
[0024] In the formula, s The dimension of the data vector; Indicates the first k At any given moment and system output y Related data vectors; Indicates the first k At any moment and system input Related data vectors; Indicates the first k Time-of-flight and process noise Related data vectors; Indicates the first k Time and Measurement Noise Related data vectors; Indicates the first ks The system status of the industrial internet system at any given moment; Indicates the first ks The system output of the industrial interconnection system at any given moment; Indicates the first ks The system input of the industrial internet system at any given moment; and Indicates the first ksProcess noise of an industrial interconnection system at any given moment; Indicates the first ks Measurement noise of industrial interconnection systems at any given moment; Represents the extended observability matrix of an industrial interconnected system; and Represents the system input respectively and process noise The relevant lower triangular parameter matrix.
[0025] S13. Based on the input-output model of the industrial interconnection system, obtain the input-output dataset model of the industrial interconnection system. Furthermore, the input / output dataset model of the industrial internet system is as follows: (4) in, , , , , .
[0026] In the formula, , , , , These represent the Hankel matrices corresponding to system state, system output, system input, process noise, and measurement noise, respectively; M represents the number of columns in the Hankel matrix. , , , and They represent the first ks-M+1 The system status, system output, system input, process noise, and measurement noise of the industrial interconnection system at any given time. , , , They represent the first k-M+1 The system status, system output, system input, process noise, and measurement noise of the industrial interconnection system at any given time.
[0027] To estimate the unmeasurable variables in formula (4) We introduce the following state observer based on a Kalman filter: (5) In the formula, Indicates the first k+1 System state estimation at time 10:00; This represents the system state estimate at time k; Indicates the first k System output estimation at time 1; Indicates the first k The system residual at any given time; Let represent the gain of the Kalman filter. Based on equation (5), we can further obtain:
[0028] (6) In the formula, and They represent the first k-1 The system output and system input of the industrial interconnection system at a given moment; for a sufficiently large parameter s, Substituting formula (6) into formula (4), we get:
[0029] (7) in, , .
[0030] In the formula, For parameterized matrices; The input and output Hankel matrix represents the past. The past input Hankel matrix; This is the past output Hankel matrix.
[0031] S14. Based on the input-output dataset model of the industrial interconnection system, the input-output data of each subsystem of the industrial interconnection system are obtained through decomposition. Furthermore, by decomposing formula (7), we can obtain the distributed input-output dataset model of each subsystem. (8) in, , , , .
[0032] In the formula, , , They represent the first i The Hankel matrices corresponding to the system output, process noise, and measurement noise of each subsystem; Indicates the first j The Hankel matrix corresponding to the system input of each subsystem; , , Representing the subsystem iThe relevant past input / output parameter matrix, future input parameter matrix, and process noise parameter matrix; , , and They represent the first ks-M+1 At the [time]th moment i The system output, system input, process noise, and measurement noise of each subsystem. , , They represent the first k-M+1 At the [time]th moment i The system output, system input, process noise, and measurement noise of each subsystem; , , They represent the first ks At the [time]th moment i The system output, system input, process noise, and measurement noise of each subsystem.
[0033] S2. Based on the input and output data of each subsystem of the industrial interconnection system, the past input and output parameter matrix and future input parameter matrix related to the subsystem are obtained through the distributed subspace identification method. Specifically, such as Figure 3 The flowchart shown in this embodiment of the two-layer fault diagnosis method for industrial interconnection systems of the present invention obtains the past input-output parameter matrix and the future input parameter matrix related to the subsystem. In step S2, based on the input-output data of each subsystem of the industrial interconnection system, the distributed subspace identification method is used to obtain the past input-output parameter matrix and the future input parameter matrix related to the subsystem, including: S21. Based on the input and output data of each subsystem of the industrial interconnection system, the Hankel matrix is obtained by performing the least squares method. S22. Based on the Hankel matrix, perform LQ decomposition to obtain the decomposed numerical matrix; S23. Based on the numerical matrix after decomposition, obtain the past input / output parameter matrix and the future input parameter matrix related to the subsystem.
[0034] Furthermore, the past input / output parameter matrix and future input parameter matrix related to subsystem i can be obtained by solving the following least squares problem: (9) In the formula, The Frobenius norm of the matrix is represented by . The row space is orthogonally projected onto the line space by and In Zhang Cheng's row space, the solution can be expressed as:
[0035] (10) In the formula, Represents the pseudo-inverse of a matrix; , These represent the parameter matrices between the output of subsystem i and the inputs of subsystem 1 and subsystem N, respectively.
[0036] The Hankel matrix in formula (10) is decomposed using the following LQ decomposition: (11) We can obtain: (12) in, , .
[0037] In the formula, , , , , , , , , This represents the numerical matrix obtained by performing LQ decomposition on the Hankel matrix.
[0038] Based on this, we can obtain the subsystem i Related past input / output parameter matrices and future input parameter matrices: (13) S3. Based on the input and output data of each subsystem of the industrial interconnection system, the coupling connection relationship between each subsystem is obtained using the multivariate Granger causal analysis method. Specifically, such as Figure 4 The flowchart shown in this embodiment of the two-layer fault diagnosis method for industrial interconnection systems illustrates the process of obtaining the coupling connection relationships between subsystems. In step S3, based on the input and output data of each subsystem of the industrial interconnection system and using a multivariate Granger causality analysis method, the coupling connection relationships between the subsystems are obtained, including: S31. Based on the input and output data of each subsystem of the industrial interconnection system, establish an unrestricted model and a restricted model; Furthermore, the unrestricted model is established as follows: (14) The constrained model is: (15) In the formula, Indicates the first k Subsystem at a given moment j The system output; Indicates the first kp Subsystem at a given moment j The system output; Indicates the first kp Subsystem at a given moment i The system output; Indicates the first kp Subsystem at a given moment i System input; This represents the error term; P represents the maximum lag order. , , These represent the first output of the target subsystem, the output of the source subsystem, and the input of the source subsystem, respectively. p The coefficient of the first lag term.
[0039] S32. Based on the unrestricted model and the restricted model, calculate the sum of squared residuals of the unrestricted model and the sum of squared residuals of the restricted model. Furthermore, the calculation method is as follows: , (16) In the formula, , Let represent the sum of squared residuals for the unrestricted model and the restricted model, respectively; , These represent the predicted values of the unrestricted model and the restricted model, respectively; T represents the length of the time series.
[0040] S33. Based on the sum of squared residuals of the unrestricted model and the sum of squared residuals of the restricted model, obtain the statistic. Furthermore, statistics F The calculation method is as follows: (17) S34. Based on this statistic, the standardized coupling strength of each subsystem is obtained; Furthermore, at the significance level If: (18) Then the null hypothesis is rejected, and the subsystem is considered to be... i Granger-induced subsystem j .
[0041] For the detected significant causal relationships, calculate the normalized coupling strength of each subsystem. : (19) S35. Based on the standardized coupling strength of each subsystem, the coupling connection relationship between each subsystem is obtained.
[0042] Furthermore, based on the standardized coupling strength of the subsystems, the coupling connection relationships between the subsystems are constructed: (20) In the formula, For subsystem j Subsystem i The coupling strength.
[0043] S4. Based on the coupling relationship between the subsystems, the alternating direction multiplier method is used to optimize the past input-output parameter matrix and the future input parameter matrix related to the subsystem, so as to obtain the optimized past input-output parameter matrix and the optimized future input parameter matrix. Specifically, such as Figure 5 The flowchart shown in this embodiment of the two-layer fault diagnosis method for industrial interconnection systems of the present invention obtains the optimized past input-output parameter matrix and the optimized future input parameter matrix. In step S4, based on the coupling connection relationship between the subsystems, the alternating direction multiplier method is used to optimize the past input-output parameter matrix and the future input parameter matrix associated with the subsystem, resulting in the optimized past input-output parameter matrix and the optimized future input parameter matrix, including: S41. Based on the coupling relationship between the subsystems, introduce auxiliary variables to construct the augmented Lagrange function; Furthermore, for each subsystem, the following optimization problem is constructed; (twenty one) In the formula, This represents the regularization parameter, used to control the strength of regularization.
[0044] By introducing auxiliary variables, an unconstrained optimization problem can be transformed into a constrained optimization problem: (twenty two) In the formula, Auxiliary variables representing past data coefficients of subsystem i; This represents the auxiliary variable representing the future input coefficients of subsystem i.
[0045] Based on this, the following augmented Lagrangian function can be constructed: (twenty three) In the formula, , Let represent the Lagrange multipliers of the past data coefficients and future input coefficients of subsystem i, respectively; This represents the penalty coefficient, used to control the intensity of the penalty for constraint violation.
[0046] S42. Based on the augmented Lagrange function, the alternating direction multiplier method is used for iterative solution to obtain the auxiliary variables of past data coefficients and future input coefficients. Furthermore, the iterative solution is performed using the alternating direction multiplier method, which includes the following three sub-steps: (i) Fix the auxiliary variables and Lagrange multipliers, and update the coefficient matrix of past data: (twenty four) And the future input coefficient matrix: (25) In the formula, , , , , , Let represent the past input / output parameter matrix, future input parameter matrix, past data coefficient auxiliary variable, future input coefficient auxiliary variable, past data coefficient Lagrange multiplier matrix, and future input coefficient Lagrange multiplier matrix, respectively, related to subsystem i at time k. , Let represent the past input / output parameter matrix and the future input parameter matrix related to subsystem i at time k+1, respectively.
[0047] (ii) Fix the original variables and Lagrange multipliers, and update the auxiliary variables of past data coefficients: (26) And future input coefficient auxiliary variables: (27) In the formula, , Let represent the auxiliary variables of past data coefficients and future input coefficients related to subsystem i at time k+1, respectively.
[0048] (iii) Update the Lagrange multiplier matrix of past data coefficients and future input coefficients: , (28) In the formula, , Let represent the Lagrange multiplier matrix of past data coefficients and the Lagrange multiplier matrix of future input coefficients related to subsystem i at time k+1, respectively.
[0049] S43. Based on the past data coefficient auxiliary variable and the future input coefficient auxiliary variable, the original residual norm and the dual residual norm are obtained through calculation. Furthermore, the original residual norm is calculated: (29) And dual residual norm: (30) S44. Iterate and converge the original residual norm and the dual residual norm. When both the original residual norm and the dual residual norm are less than the preset threshold, the optimized past input-output parameter matrix and the optimized future input parameter matrix are obtained.
[0050] Furthermore, iteration stops when both the original residual norm and the dual residual norm are simultaneously less than a preset threshold: (31) In the formula, , These represent the original residual convergence threshold and the dual residual convergence threshold, respectively.
[0051] S5. Based on the optimized past input-output parameter matrix and the optimized future input parameter matrix, construct a distributed residual generator to obtain the distributed fault detection status of the industrial interconnection system. Specifically, such as Figure 6 The flowchart shown in this embodiment of the two-layer fault diagnosis method for industrial interconnection systems illustrates the process of obtaining distributed fault detection information for the industrial interconnection system. In step S5, a distributed residual generator is constructed based on the optimized past input / output parameter matrix and the optimized future input parameter matrix to obtain the distributed fault detection information for the industrial interconnection system, including: S51. Based on the optimized past input-output parameter matrix and the optimized future input parameter matrix, construct the residual generator for each subsystem; Furthermore, the residual generators for each subsystem can be constructed in the following form: (32) in, .
[0052] In the formula, Indicates the first i The Hankel matrix corresponding to the residual signals of each subsystem; Indicates the first ks-M+1 At the [time]th moment i The residual signal of each subsystem Indicates the first k-M+1 At the [time]th moment iThe residual signal of each subsystem; Indicates the first ks At the [time]th moment i The residual signal of each subsystem; Indicates the first k At the [time]th moment i The residual signal of each subsystem.
[0053] Based on this, the residual generators of each subsystem can be further expressed as: (33) in, , , , , .
[0054] In the formula, , They represent the first KS-1 The system output and system input of the industrial interconnection system at any given moment; , They represent the first k-2s-1 The system output and system input of the industrial interconnection system at any given moment; , They represent the first k Time and Subsystem i The residual signal and the data vector related to the system output; Indicates the first k Time and Subsystem j The system input-related data vector; , They represent the first KS-1 A data vector related to the system output and system input of the industrial interconnection system at each moment.
[0055] S52. Based on the residual generator of each subsystem, perform residual evaluation to obtain the residual statistics of each subsystem; Furthermore, (34) in, .
[0056] In the formula, Let be the residual statistic of the i-th subsystem. This represents the covariance of the residual signal of the i-th subsystem; This represents a chi-square distribution with a certain number of degrees of freedom.
[0057] S53. Based on the set false alarm rate, the alarm threshold is calculated using the chi-square distribution table. Furthermore, Given a false alarm rate The threshold can be calculated using the chi-square distribution table. : (35) S54. Compare the alarm threshold with the residual statistics of each subsystem to obtain the distributed fault detection status of the industrial interconnection system.
[0058] Furthermore, by comparing formula (34) with formula (35) and running the following decision logic, distributed fault detection in industrial interconnection systems can be achieved: (36) S6. Based on the distributed fault detection of the industrial interconnection system, the fault source subsystem is obtained by processing the characteristic changes of each subsystem of the industrial interconnection system.
[0059] Specifically, such as Figure 7 The flowchart shown in this embodiment of the two-layer fault diagnosis method for industrial interconnection systems illustrates the process of obtaining the fault source subsystem. In step S6, based on the distributed fault detection of the industrial interconnection system, the fault source subsystem is obtained by processing the characteristic changes of each subsystem of the industrial interconnection system, including: S61. Based on the distributed fault detection of the industrial interconnection system, obtain the residual signals of each subsystem within the time interval and the residual signals of each subsystem within the time window. Furthermore, within the time interval Internal collection of residual signals of each subsystem within the time interval ; For time Define a sliding window The residual signals of each subsystem within the collection time window are the residual signals. .
[0060] S62. Based on the residual signals of each subsystem within the time interval, obtain the benchmark mean and benchmark standard deviation of the residual signals of each subsystem. Furthermore, the baseline mean of the residual signal: (37) Benchmark standard deviation: (38) In the formula, For subsystem i The baseline mean of the residual signal; The baseline window length; The starting point of the reference window; For subsystem i The baseline standard deviation of the residual signal.
[0061] S63. Based on the residual signals of each subsystem within the time window, obtain the window mean and window standard deviation of the residual signals of each subsystem. Furthermore, the window mean: (39) Window standard deviation: (40) In the formula, Representation Subsystem i The window mean of the residual signal at time k; To detect the length of the window; Representation Subsystem i The residual signal is k The standard deviation of the time window.
[0062] S64. Based on the baseline mean and baseline standard deviation of the residual signals of each subsystem and the window mean and window standard deviation of the residual signals of each subsystem, obtain the change in the statistical characteristics of each subsystem. Furthermore, the changes in the statistical characteristics of each subsystem are calculated using formulas (37) to (40): (41) In the formula, Representation Subsystem i The residual signal is k The change in the mean over time; Representation Subsystem i The residual signal is k Change in standard deviation over time; S65. Based on the changes in the statistical characteristics of each subsystem, the fault source subsystem is obtained through the comprehensive fault score of each subsystem.
[0063] Furthermore, the normalized rate of change: (42) In the formula, Representation Subsystem i The residual signal is k The rate of change of the mean at time t; Representation Subsystem i The residual signal is k Rate of change of standard deviation at time t; It represents a very small positive number, preventing division by zero.
[0064] The overall fault score for each subsystem is calculated using formula (42): (43) By combining the overall fault scores of each subsystem, the source subsystem of the fault can be identified: (44) In the formula, represents the fault source subsystem number at time k.
[0065] like Figure 8 The diagram shown is a system block diagram of an embodiment of the two-layer fault diagnosis system for industrial interconnection systems according to the present invention. The present invention provides a two-layer fault diagnosis system for industrial interconnection systems, which is applied to a two-layer fault diagnosis method for industrial interconnection systems. The system includes: a first acquisition module, a parameter matrix module, a coupling connection module, an optimization module, a fault monitoring module, and a fault location module. Specifically, The first acquisition module is used to obtain the input and output data of each subsystem of the industrial interconnection system based on the state space expression of each subsystem of the industrial interconnection system. The parameter matrix module is used to obtain the past input / output parameter matrix and future input parameter matrix related to the subsystems based on the input / output data of each subsystem of the industrial interconnection system through a distributed subspace identification method. The coupling connection module is used to obtain the coupling connection relationship between the subsystems based on the input and output data of each subsystem of the industrial interconnection system and the multivariate Granger causal analysis method. The optimization module is used to optimize the past input-output parameter matrix and future input parameter matrix associated with each subsystem using the alternating direction multiplier method, based on the coupling connection relationship between the subsystems, to obtain the optimized past input-output parameter matrix and the optimized future input parameter matrix. The fault monitoring module is used to construct a distributed residual generator based on the optimized past input-output parameter matrix and the optimized future input parameter matrix to obtain the distributed fault detection status of the industrial interconnection system. The fault location module is used to identify the fault source subsystem by processing the characteristic changes of each subsystem of the industrial interconnection system based on the distributed fault detection of the industrial interconnection system.
[0066] This invention provides a two-layer fault diagnosis method and system for industrial interconnection systems. First, based on the state-space models of each subsystem within the industrial interconnection system, the invention derives the input-output data of each subsystem and utilizes this data to achieve distributed identification of the subspace parameter matrix, aiming to fully leverage the complementary advantages of mechanistic modeling and data-driven approaches. Second, the invention uses Granger causality analysis to calculate the coupling connections between subsystems and combines this with the alternating direction multiplier method to achieve distributed optimization of the subspace parameter matrix. Finally, a distributed residual generator is designed and residuals are evaluated at the subsystem level to achieve distributed fault detection in the industrial interconnection system. Simultaneously, the residual signals from each subsystem are transmitted to a centralized processing unit. When a fault is detected, the source subsystem is identified by calculating the comprehensive fault score of each subsystem.
[0067] It is understood that the present invention has been described through the above embodiments and should not be construed as limiting the implementation and scope of the present invention. Those skilled in the art will recognize that various changes or equivalent substitutions can be made to these features and embodiments without departing from the spirit and scope of the present invention. Furthermore, under the teachings of the present invention, these features and embodiments can be modified to adapt to specific situations and materials without departing from the spirit and scope of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed herein, and all embodiments falling within the scope of the claims of this application are within the protection scope of the present invention.
Claims
1. A two-layer fault diagnosis method for industrial interconnection systems, characterized in that, The method includes: S1. Based on the state-space expressions of each subsystem of the industrial interconnection system, obtain the input and output data of each subsystem of the industrial interconnection system; S2. Based on the input and output data of each subsystem of the industrial interconnection system, the past input and output parameter matrix and the future input parameter matrix related to the subsystem are obtained through the distributed subspace identification method. S3. Based on the input and output data of each subsystem of the industrial interconnection system, the coupling connection relationship between each subsystem is obtained using the multivariate Granger causal analysis method. S4. Based on the coupling connection relationship between the subsystems, the past input-output parameter matrix and future input parameter matrix related to the subsystem are optimized using the alternating direction multiplier method to obtain the optimized past input-output parameter matrix and the optimized future input parameter matrix. S5. Based on the optimized past input-output parameter matrix and the optimized future input parameter matrix, construct a distributed residual generator to obtain the distributed fault detection status of the industrial interconnection system. S6. Based on the distributed fault detection of the industrial interconnection system, the fault source subsystem is obtained by processing the characteristic changes of each subsystem of the industrial interconnection system.
2. The two-layer fault diagnosis method for industrial interconnection systems according to claim 1, characterized in that, In step S1, the input and output data of each subsystem of the industrial interconnection system are obtained based on the state-space expressions of each subsystem, including: S11. Based on the state-space expressions of each subsystem of the industrial interconnection system, the improved state-space model of the industrial interconnection system is obtained. S12. Based on the improved state space model of the industrial interconnection system, obtain the input-output model of the industrial interconnection system; S13. Based on the input-output model of the industrial interconnection system, obtain the input-output dataset model of the industrial interconnection system. S14. Based on the input-output dataset model of the industrial interconnection system, the input-output data of each subsystem of the industrial interconnection system are obtained by decomposition.
3. The two-layer fault diagnosis method for industrial interconnection systems according to claim 1, characterized in that, In step S2, based on the input and output data of each subsystem of the industrial interconnection system, a distributed subspace identification method is used to obtain the past input and output parameter matrix and the future input parameter matrix related to the subsystem, including: S21. Based on the input and output data of each subsystem of the industrial interconnection system, the Hankel matrix is obtained by performing the least squares method; S22. Based on the Hankel matrix, perform LQ decomposition to obtain the decomposed numerical matrix; S23. Based on the decomposed numerical matrix, obtain the past input / output parameter matrix and the future input parameter matrix related to the subsystem.
4. The two-layer fault diagnosis method for industrial interconnection systems according to claim 1, characterized in that, In step S3, based on the input and output data of each subsystem of the industrial interconnection system, and using multivariate Granger causality analysis, the coupling connections between the subsystems are obtained, including: S31. Based on the input and output data of each subsystem of the industrial interconnection system, establish an unrestricted model and a restricted model; S32. Based on the unrestricted model and the restricted model, calculate the sum of squared residuals of the unrestricted model and the sum of squared residuals of the restricted model. S33. Obtain the statistic based on the sum of squared residuals of the unrestricted model and the sum of squared residuals of the restricted model; S34. Based on the statistics, obtain the standardized coupling strength of each subsystem; S35. Based on the standardized coupling strength of each subsystem, the coupling connection relationship between each subsystem is obtained.
5. The two-layer fault diagnosis method for industrial interconnection systems according to claim 1, characterized in that, In step S4, based on the coupling relationship between the subsystems, the alternating direction multiplier method is used to optimize the past input / output parameter matrix and the future input parameter matrix related to the subsystem, resulting in the optimized past input / output parameter matrix and the optimized future input parameter matrix, including: S41. Based on the coupling connection relationship between the subsystems, introduce auxiliary variables to construct an augmented Lagrange function; S42. Based on the augmented Lagrange function, the alternating direction multiplier method is used for iterative solution to obtain the auxiliary variables of past data coefficients and future input coefficients; S43. Based on the past data coefficient auxiliary variables and the future input coefficient auxiliary variables, the original residual norm and the dual residual norm are obtained by calculation. S44. Iteratively converge the original residual norm and the dual residual norm. When both the original residual norm and the dual residual norm are less than a preset threshold, the optimized past input-output parameter matrix and the optimized future input parameter matrix are obtained.
6. The two-layer fault diagnosis method for industrial interconnection systems according to claim 1, characterized in that, In step S5, a distributed residual generator is constructed based on the optimized past input-output parameter matrix and the optimized future input parameter matrix to obtain the distributed fault detection status of the industrial interconnection system, including: S51. Based on the optimized past input-output parameter matrix and the optimized future input parameter matrix, construct the residual generator for each subsystem; S52. Based on the residual generator of each subsystem, perform residual evaluation to obtain the residual statistics of each subsystem; S53. Based on the set false alarm rate, the alarm threshold is calculated using the chi-square distribution table. S54. Compare the alarm threshold with the residual statistics of each subsystem to obtain the distributed fault detection status of the industrial interconnection system.
7. The two-layer fault diagnosis method for industrial interconnection systems according to claim 1, characterized in that, In step S6, based on the distributed fault detection of the industrial interconnection system, the fault source subsystem is obtained by processing the characteristic changes of each subsystem of the industrial interconnection system, including: S61. Based on the distributed fault detection of the industrial interconnection system, obtain the residual signals of each subsystem within the time interval and the residual signals of each subsystem within the time window. S62. Based on the residual signals of each subsystem within the time interval, obtain the benchmark mean and benchmark standard deviation of the residual signals of each subsystem; S63. Based on the residual signals of each subsystem within the time window, obtain the window mean and window standard deviation of the residual signals of each subsystem. S64. Based on the baseline mean and baseline standard deviation of the residual signals of each subsystem and the window mean and window standard deviation of the residual signals of each subsystem, obtain the change in the statistical characteristics of each subsystem. S65. Based on the changes in the statistical characteristics of each subsystem, the fault source subsystem is obtained through the comprehensive fault score of each subsystem.
8. A two-layer fault diagnosis system for industrial interconnection systems, used to implement the two-layer fault diagnosis method for industrial interconnection systems as described in any one of claims 1-7, characterized in that, The system includes: The first acquisition module is used to obtain the input and output data of each subsystem of the industrial interconnection system based on the state space expression of each subsystem of the industrial interconnection system. The parameter matrix module is used to obtain the past input / output parameter matrix and future input parameter matrix related to the subsystems based on the input / output data of each subsystem of the industrial interconnection system through a distributed subspace identification method. The coupling connection module is used to obtain the coupling connection relationship between the subsystems based on the input and output data of each subsystem of the industrial interconnection system and the multivariate Granger causal analysis method. The optimization module is used to optimize the past input-output parameter matrix and future input parameter matrix related to the subsystems using the alternating direction multiplier method based on the coupling connection relationship between the subsystems, so as to obtain the optimized past input-output parameter matrix and the optimized future input parameter matrix. The fault monitoring module is used to construct a distributed residual generator based on the optimized past input-output parameter matrix and the optimized future input parameter matrix to obtain the distributed fault detection status of the industrial interconnection system. The fault location module is used to obtain the fault source subsystem by processing the characteristic changes of each subsystem of the industrial interconnection system based on the distributed fault detection of the industrial interconnection system.