Multi-sensor fused air energy heat pump fault prediction and diagnosis system

Abstract

The invention relates to the technical field of fault prediction and diagnosis systems, in particular to a multi-sensor fusion air energy heat pump fault prediction and diagnosis system, which is characterized in that a data acquisition module is used for acquiring a multi-dimensional data set in the operation process of an air energy heat pump; the data preprocessing module generates a topological variation matrix based on the data set; the fault diagnosis module comprises a feature extraction sub-module, a feature compression sub-module, a fault classification sub-module, a mode matching sub-module and a fault backtracking sub-module which are connected in sequence, and a fault classification result, an optimal matching mode and a fault reason are obtained based on progressive processing of a topological variation matrix; the fault prediction module predicts a future fault type and occurrence time based on the fault diagnosis result; the decision-making module generates an optimization strategy according to a fault prediction result, the accuracy of the system on pattern matching is remarkably improved through combination of modulus operation and an Euler function, and meanwhile, the depth and reliability of fault reason analysis are further enhanced through application of a Bessel function in fault backtracking.

Description

Technical Field

[0001] The present invention relates to the technical field of fault prediction and diagnosis systems, and specifically to a multi-sensor fusion air energy heat pump fault prediction and diagnosis system. Background Art

[0002] At present, common fault diagnosis methods can be roughly divided into two categories: rule-based diagnosis methods and machine learning-based diagnosis methods. Rule-based diagnosis methods rely on pre-set thresholds and logical rules to judge the operating status of the equipment. This type of method is simple to implement, but it often seems powerless when faced with complex multi-dimensional fault modes. In addition, its ability to identify nonlinear fault characteristics is limited, which can easily lead to misjudgment or missed judgment.

[0003] On the other hand, diagnostic methods based on machine learning have received widespread attention in recent years. Such methods usually rely on a large amount of historical data and achieve fault mode identification and classification through model training.

[0004] However, existing machine learning methods still have the following problems:

[0005] 1. Insufficient feature extraction: When processing multidimensional data, traditional methods have difficulty in fully extracting features that reflect the nature of the fault, especially the limited ability to capture complex nonlinear relationships and dynamic changes.

[0006] 2. Low efficiency of feature screening and compression: When the data dimension is high, existing methods are prone to lose important information when screening key features, resulting in reduced accuracy of diagnostic results.

[0007] 3. Insufficient fault classification accuracy: For the classification of multiple fault modes, existing methods find it difficult to establish accurate distinction boundaries between fault modes, especially when there is overlap or noise interference between modes, the classification accuracy is significantly reduced.

[0008] 4. Insufficient ability to trace back fault causes: Existing methods are often unable to deeply trace the specific causes of faults and can only provide surface fault information, which limits their application in complex scenarios.

[0009] 5. Poor real-time performance and response speed: When processing large-scale data, traditional methods have high computational complexity, resulting in long system response time and difficulty in meeting real-time requirements.

[0010] In summary, the existing fault diagnosis methods have obvious limitations when facing multi-dimensional complex data and dynamic operating environments. These problems have greatly affected the reliability and practicality of the air energy heat pump fault diagnosis system, and a new technical solution is urgently needed to overcome these technical bottlenecks. Summary of the invention

[0011] The present invention aims to solve the above-mentioned problems and provide an air energy heat pump fault prediction and diagnosis system capable of multi-sensor fusion.

[0012] The present invention proposes a multi-sensor fusion air energy heat pump fault prediction and diagnosis system, the system includes a data acquisition module, a data preprocessing module, a fault diagnosis module, a fault prediction module and a decision module;

[0013] Wherein, the data acquisition module is used to obtain a multi-dimensional data set during the operation of the air energy heat pump;

[0014] The data preprocessing module generates a topological variation matrix based on the data set;

[0015] The fault diagnosis module includes a feature extraction submodule, a feature compression submodule, a fault classification submodule, a pattern matching submodule and a fault backtracking submodule connected in sequence, and obtains a fault classification result, an optimal matching pattern and a fault cause based on the progressive processing of the topological variation matrix;

[0016] The fault prediction module predicts the future fault type and occurrence time based on the fault diagnosis result;

[0017] The decision module generates an optimization strategy according to the fault prediction result.

[0018] Preferably, the data preprocessing module is used to construct a topological variation matrix based on the multi-dimensional data set, and the topological variation matrix is ​​obtained by the following formula:

[0019]

[0020] Among them, M t (i,j) is the topological variation relationship between node i and node j at time t, D t (i) and D t (j) are the operating data of node i and node j at time t, σ is the standard deviation, k is the adjustment parameter, and t is time.

[0021] Preferably, the feature extraction submodule obtains a feature matrix based on the topological variation matrix through the following steps:

[0022] F t =Eig(M t )

[0023] Among them, F t is the characteristic matrix, Eig(M t ) is the topological variation matrix M t The eigenvalue decomposition results of .

[0024] Preferably, the feature compression submodule further performs feature screening on the feature matrix based on the symmetry group in group theory, and the screening matrix is ​​obtained by the following formula:

[0025] S t ={g∈G|φ(g,F t )>θ}

[0026] Among them, S t is the feature screening matrix, G is the symmetry group, φ(g,F t ) is the feature importance function, and θ is the screening threshold. The feature vector after screening is:

[0027] F ′ t =S t ·F t

[0028] Among them, F ′ t is the compressed feature matrix.

[0029] Preferably, the fault classification submodule uses Legendre polynomials to perform fault classification based on the feature matrix output by the feature compression submodule, and the classification result is obtained by the following formula:

[0030]

[0031] Among them, C t is the classification result, P n (x) is the Legendre polynomial, α n is the classification weight, and N is the expansion order.

[0032] Preferably, the pattern matching submodule matches the optimal fault mode through modulus operation and Euler function based on the fault classification result, and the matching function is obtained by the following formula:

[0033]

[0034] Among them, d(P i ,C t ) is the failure mode P i And the classification result C t The matching degree, φ(P i ) is the Euler function.

[0035] Preferably, the fault backtracking submodule is based on the optimal fault mode The Bessel function is used to trace back the cause of the fault. The traceback function is obtained by the following formula:

[0036]

[0037] Among them, R(x) is the fault cause weight function, J n (x) is the Bessel function.

[0038] Preferably, the fault prediction module predicts possible future fault types and occurrence times based on the fault classification results and backtracking results in combination with historical data.

[0039] Preferably, the decision module generates an optimization strategy based on the fault prediction result, including adjusting operating parameters, issuing alarms and suggesting shutdown maintenance.

[0040] Preferably, the system forms a complete workflow from data acquisition and processing to fault diagnosis, prediction and decision-making through progressive algorithm modules, wherein the output of each algorithm module is directly used as the input of the subsequent module to ensure the collaborative relationship and logical consistency between the modules of the system.

[0041] The beneficial technical effects brought by the technical solution of the present invention are:

[0042] Through the close collaboration of five modules, from macro-architecture to micro-algorithm, the present invention forms an efficient, accurate and intelligent fault diagnosis system. In terms of the overall architecture, the data flow and information interaction between modules have a high degree of logical consistency and synergy. Among them, the data acquisition module provides high-quality data input for subsequent algorithms; the data preprocessing module uses the topological matrix mutation algorithm to construct a dynamic association model, laying the foundation for feature extraction; the fault diagnosis module uses a progressive algorithm link, from multi-dimensional feature extraction to fault cause tracing, layer by layer, to achieve efficient transformation from data to information and then to knowledge.

[0043] The complementarity between modules is reflected in the following aspects: the topological matrix mutation algorithm makes up for the shortcomings of traditional feature extraction in dynamic association recognition; group theory screening effectively solves the contradiction between high-dimensional feature screening and compression; Legendre polynomial classification establishes precise boundaries between complex patterns. The superposition effect of each module algorithm ultimately enables the system to maintain efficient fault diagnosis performance in complex environments.

[0044] The present invention also has a significant synergistic effect. The five modules work together to not only solve the core problems in the prior art, but also significantly improve the overall robustness, accuracy and real-time performance of the system. For example, through the combination of modular operations and Euler functions, the accuracy of the system in pattern matching has been significantly improved. At the same time, the application of Bessel functions in fault backtracking further enhances the depth and reliability of fault cause analysis.

[0045] Through the above design, the present invention provides a systematic solution to the problems existing in the fault diagnosis of air-source heat pumps. It not only achieves technical breakthroughs in fault detection, classification and cause tracing, but also significantly outperforms the existing technologies in real-time, accuracy and diagnostic depth. It has broad application prospects and important practical value. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 It is the overall system logic block diagram of the present invention.

[0047] Figure 2 It is a logic box of the fault diagnosis module of the present invention. DETAILED DESCRIPTION

[0048] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the specific implementation methods, structures, features and effects thereof are described in detail below in conjunction with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics in one or more embodiments may be combined in any suitable form.

[0049] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0050] See also Figure 1-2 The present invention provides a multi-sensor fusion air energy heat pump fault prediction and diagnosis system, which includes a data acquisition module 1, a data preprocessing module 2, a fault diagnosis module 3, a fault prediction module 4 and a decision module 5. The data acquisition module 1 is used to obtain a multi-dimensional data set during the operation of the air energy heat pump, which includes but is not limited to the solenoid valve switch state, compressor current, fan current, water valve opening and pressure state. The data acquisition module 1 collects the above parameters in real time through multiple high-precision sensors and transmits them to the data preprocessing module 2 for processing.

[0051] The data preprocessing module 2 is responsible for cleaning and normalizing the collected raw data. This module further uses a specific algorithm to convert the processed data into a topological variation matrix to provide input data support for the fault diagnosis module 3. The construction of the topological variation matrix can capture the dynamic correlation between data, especially in the expression of potential fault characteristics between different component operating states.

[0052] Fault diagnosis module 3 includes five progressive sub-modules, namely feature extraction sub-module, feature compression sub-module, fault classification sub-module, pattern matching sub-module and fault backtracking sub-module. The output of each sub-module is directly used as the input of the next sub-module, thereby realizing full-link diagnosis from fault detection to cause analysis.

[0053] The fault prediction module 4 further predicts the possible fault types and their occurrence time in the future based on the diagnosis results of the fault diagnosis module 3. Finally, the decision module 5 generates corresponding optimization strategies based on the results provided by the fault prediction module 4. These strategies include but are not limited to adjusting operating parameters, issuing alarms or suggesting shutdown maintenance to ensure the safe and efficient operation of the air energy heat pump.

[0054] In a preferred embodiment of the present invention, the data preprocessing module 2 is used to construct a topological variation matrix. Specifically, the module generates the topological variation matrix according to the following formula:

[0055]

[0056] Among them, M t (i,j) is the topological variation relationship between node i and node j at time t, D t (i) and D t (j) are the operation data of node i and node j at time t, σ is the standard deviation, k is the adjustment parameter, and t is time. The topological variation matrix is ​​generated using sensor data to dynamically capture the nonlinear associations between nodes.

[0057] For example, when the compressor current gradually deviates from the normal range, the topological variation matrix can capture the abnormal trend through the dynamic change of weights. This matrix construction method lays a solid foundation for the data analysis of subsequent modules.

[0058] In a preferred embodiment of the present invention, the feature extraction submodule obtains a feature matrix by performing eigenvalue decomposition on the topological variation matrix. Specifically, the feature extraction process is as follows:

[0059] F t =Eig(M t )

[0060] Among them, F t is the characteristic matrix at time t, Eig(M t ) represents the matrix M t Perform eigenvalue decomposition. This feature extraction method can effectively extract the core features in complex fault modes and avoid the computational complexity caused by high-dimensional data in traditional algorithms.

[0061] Preferably, in this embodiment, if the dimension of the topological variation matrix is ​​100×100, the dimension of the feature matrix after feature extraction can be reduced to 10×10. This dimensionality reduction process can significantly reduce the computational burden while maintaining the main information.

[0062] In one embodiment of the present invention, the feature compression submodule further compresses and screens the feature matrix based on the symmetry group in group theory. The module performs feature screening using the following formula:

[0063] S t ={g∈G|φ(g,F t )>θ}

[0064] Among them, S t is the feature screening matrix, G is the feature matrix F t The corresponding symmetric group set, φ(g,F t ) is the feature importance function, θ is the screening threshold, preferably between 0.5 and 0.7. t , using the subgroup structure of the symmetry group to screen out the most important features.

[0065] For example, for a compressor current feature, its feature value may be 0.8 during normal operation, but may drop to 0.3 during abnormal operation. At this time, selecting θ=0.6 can effectively eliminate features of low importance and improve diagnostic accuracy.

[0066] The compressed feature vector is expressed as follows:

[0067] F ′ t =S t ·F t

[0068] Among them, F ′ t It is the final compressed feature matrix, which is used for the next step of fault classification.

[0069] In one embodiment of the present invention, the fault classification submodule uses Legendre polynomials to perform nonlinear fault classification based on the output result of the feature compression submodule. The specific formula is as follows:

[0070]

[0071] Among them, C t is the fault classification result, P n (x) is the Legendre polynomial, defined as:

[0072]

[0073] α nis a weight coefficient. Preferably, N is 4 to 8, which is used to control the complexity of the classification model. In practical applications, if the characteristic value of the water valve opening fluctuates abnormally, the classification result C t It may be output as a level 3 fault. At this time, the system will take different response strategies according to the fault level. Use the special function Legendre polynomial for classification. ′ t Perform nonlinear projection to enhance classification capabilities.

[0074] The present invention realizes comprehensive diagnosis and classification of air energy heat pump faults through five progressive modules, and greatly improves the accuracy and efficiency of diagnosis by combining topological variation matrix, group theory and special functions. The input and output of each module have a close logical relationship, which ensures the efficiency and robustness of the system.

[0075] In a preferred embodiment of the present invention, the pattern matching submodule is based on the classification result C output by the fault classification submodule. t ,The method combining modular operation and Euler function is used to match fault patterns. Specifically, the matching degree between the classification result and the known fault pattern set is calculated by the following formula:

[0076]

[0077] Among them, d(P i ,C t ) is the failure mode P i And the classification result C t The matching degree, φ(P i ) is the Euler function, defined as:

[0078]

[0079] Among them, P represents the prime number set, q is P i By calculating the matching degree, the mode with the smallest matching degree is selected as the optimal fault mode. In actual scenarios, for example, if the matching result of a certain fault mode P1 is 5, and the matching result of another mode P2 is 3, the system will select the mode P2 with a higher matching degree as the optimal mode. The Euler function is used in this algorithm to improve the resolution of pattern matching, especially in the case of multiple similar fault modes, and the fault mode is matched using modular operations and Euler functions.

[0080] The role of the pattern matching submodule is to further refine the fault classification results, thereby providing more accurate input for the fault backtracking submodule. By introducing modular operations and Euler functions, not only the computational complexity is effectively reduced, but also the algorithm's ability to distinguish subtle differences between patterns is significantly enhanced.

[0081] In one embodiment of the present invention, the fault backtracking submodule uses the optimal fault mode The cause of the failure is traced back and the Bessel function is used to assign weights to the failure modes. The specific traceback function is defined as follows:

[0082]

[0083] Among them, R(x) is the fault cause weight function, J n (x) is the Bessel function of the first kind, defined as:

[0084]

[0085] Wherein, Γ(m+n+1) is the gamma function.

[0086] In this embodiment, the fault backtracking submodule firstly calculates the fault status of the optimal mode. Calculate the weight function R(x) of each type of fault and sort the fault causes. For example, if the fault weight of abnormal water valve opening is 0.7, and the fault weight of abnormal compressor current is 0.3, the system can prompt the user to check the water valve opening first. Use Bessel function to trace the cause of the fault and calculate the impact weight of each mode on the fault.

[0087] The introduction of Bessel function in this algorithm has multiple advantages. It can nonlinearly distribute the weights of different fault modes, effectively avoiding misdiagnosis caused by similar fault modes, and at the same time improve the accuracy and stability of fault cause tracing through integral calculation.

[0088] In one embodiment of the present invention, the fault prediction module is based on the fault classification result C t The module uses the fault backtracking result R(x) and historical data to predict the possible fault types and their occurrence time in the future. This module models the fault trend through the time series analysis algorithm. Specifically, it uses the sliding window technology and the autoregressive integrated moving average model (ARIMA) for prediction.

[0089] The fault prediction model first constructs a time series {C t -1,C t-2 ,…,C t-n}, and then make predictions using the following formula:

[0090]

[0091] in, is the predicted future fault classification result, μ is a constant term, φ i and θ j are the autoregressive coefficient and moving average coefficient in the ARIMA model, ∈t-j is the prediction error.

[0092] In a preferred embodiment, if the current classification result shows that the system is in a level 2 fault state, and historical data shows that a similar pattern will evolve into a level 3 fault state after 24 hours of operation, the prediction module will issue an alarm in advance based on historical trends.

[0093] The core function of the fault prediction module is to detect potential problems in advance and reduce the risk of more serious consequences due to failure to handle faults in a timely manner.

[0094] In one embodiment of the present invention, the decision module generates the optimal response strategy based on the future fault type and occurrence time output by the fault prediction module and the results of the fault diagnosis module. These strategies may include adjusting operating parameters, issuing alarm prompts, or recommending shutdown maintenance.

[0095] Decision module 5 generates a response plan through the following logic:

[0096] If the predicted fault level is level 1, the system recommends mitigating the fault trend by adjusting operating parameters (such as reducing the compressor speed).

[0097] If the predicted fault level is level 2, the system will sound an alarm to prompt maintenance personnel to pay attention and monitor the equipment status in real time.

[0098] If the predicted fault level is level 3 or above, the system recommends immediate shutdown for maintenance to prevent equipment damage.

[0099] For example, when the wind turbine current is abnormal and the prediction model shows that it will reach a level 3 fault state within the next two hours, the decision module will automatically adjust the wind turbine's operating frequency and prompt the user to check the equipment.

[0100] The innovation of the decision-making module lies in its close collaboration with other modules, which can dynamically adjust response strategies based on real-time data to improve the overall stability and security of the system.

[0101] The present invention forms a complete workflow from data acquisition to diagnosis, prediction and decision-making. The modules achieve collaborative optimization through the connection of input and output, and each step provides accurate and reliable data support for the next step.

[0102] Specifically, the data acquisition module transmits the multidimensional data to the data preprocessing module to generate the topological variation matrix M t Subsequently, the fault diagnosis module performs feature extraction, feature compression, classification, pattern matching and backtracking analysis in sequence, gradually narrowing the scope of the fault and locating the specific cause. Finally, the prediction and decision-making module comprehensively analyzes the results and gives the optimal response strategy.

[0103] The complete data flow and workflow design of the present invention ensures the robustness and efficiency of the system in complex scenarios, can greatly improve the fault management capability of the air-to-energy heat pump, reduce equipment operation risks and extend equipment service life.

[0104] In order to verify the superiority of the present invention in air energy heat pump fault diagnosis, an experimental comparison was conducted. Through actual operation data, the fault diagnosis system based on topological matrix variation and special function group theory described in the present invention was compared with the traditional fault diagnosis method based on simple linear model.

[0105] During the test, several air-to-energy heat pump devices under actual working conditions were selected for operation data collection, and compared based on the following main performance indicators:

[0106] 1. Fault Detection Accuracy: It is used to measure the diagnostic system’s ability to identify faults.

[0107] 2. Fault classification accuracy: used to evaluate the correctness of fault classification.

[0108] 3. Root Cause Analysis Accuracy: refers to the system's ability to trace the specific cause of a fault.

[0109] 4. Prediction Lead Time: The time in which the system predicts a fault will occur, in hours.

[0110] 5. System Response Time: The time from data input to fault result output, measured in seconds.

[0111] The testing standards and methods are as follows:

[0112] 1. Fault detection accuracy: Compare the known fault points in actual operation with the fault points detected by the system, and calculate the accuracy by comparing the fault labels. The following formula is used:

[0113]

[0114] 2. Fault classification accuracy: Under various fault modes, the classification accuracy is calculated through the confusion matrix.

[0115] 3. Fault cause backtracking capability: Compare the actual fault cause with the cause traced by the system, and calculate the accuracy based on the matching rate.

[0116] 4. Fault prediction lead time: The lead time is calculated based on the difference between the actual fault occurrence time and the system prediction time.

[0117] 5. System response time: Use the system log to record the response time of each diagnostic run and take the average value.

[0118] Embodiment: A system based on the present invention;

[0119] Equipment operating conditions: 500 hours of operation, collecting 1 million data points, including compressor current, fan current, water valve opening and pressure status.

[0120] Algorithm: The five progressive algorithm modules of the present invention are adopted, including topological variation matrix, group theory screening, Legendre polynomial classification, Euler function matching and Bessel function backtracking.

[0121] Comparative example: traditional linear model;

[0122] Equipment operating conditions: Same as above.

[0123] Algorithm: Use the traditional fault diagnosis method based on linear regression.

[0124] The test results are shown in the following table:

[0125] index Example Comparative Example Improvement Fault detection accuracy (%) 98.7 85.3 +13.4 Fault classification accuracy (%) 96.5 78.2 +18.3 Fault cause tracing capability (%) 94.8 72.5 +22.3 Failure prediction lead time (hours) 3.5 1.2 +2.3 System response time (seconds) 2.1 4.8 -2.7

[0126] The result analysis and superiority verification are as follows:

[0127] From the above test results, it can be clearly seen that the embodiments of the present invention are superior to the traditional linear model in all indicators. The specific analysis is as follows:

[0128] 1. Improvement of fault detection accuracy: This invention uses topological variation matrix and group theory screening algorithm to greatly improve the recognition ability of complex nonlinear features. The detection accuracy is increased from 85.3% of the traditional model to 98.7%, reducing the missed detection rate.

[0129] 2. Improvement of fault classification accuracy: The introduction of Legendre polynomials enables the classification model to better fit the fault feature boundaries, thereby improving the classification accuracy by 18.3 percentage points.

[0130] 3. Improved fault cause tracing capability: The accuracy of cause tracing using Bessel function is significantly improved. Its complex weight distribution mechanism effectively avoids misjudgment caused by pattern overlap in traditional linear methods.

[0131] 4. Improvement in fault prediction lead time: The present invention can predict faults 2.3 hours earlier than traditional models, providing users with more adequate response time and helping to take preventive measures in a timely manner.

[0132] 5. Optimization of system response time: Although the present invention adopts multiple progressive algorithms, the optimized algorithm structure shortens the system response time to 2.1 seconds, which is 2.7 seconds less than the comparison ratio.

[0133] Through the above experiments, it can be seen that the present invention has significant advantages in fault diagnosis, classification, backtracking and prediction capabilities, and the system response is faster. This superior performance can provide a strong guarantee for the operation safety and maintenance efficiency of air energy heat pumps.

[0134] Although the above describes the specific implementation mode of the present invention in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art on the basis of the technical solution of the present invention without creative work are still within the scope of protection of the present invention.

Claims

1. The multi-sensor fusion air energy heat pump fault prediction and diagnosis system is characterized by: The system includes a data acquisition module, a data preprocessing module, a fault diagnosis module, a fault prediction module and a decision-making module; Wherein, the data acquisition module is used to obtain a multi-dimensional data set during the operation of the air energy heat pump; The data preprocessing module generates a topological variation matrix based on the data set; The fault diagnosis module includes a feature extraction submodule, a feature compression submodule, a fault classification submodule, a pattern matching submodule and a fault backtracking submodule connected in sequence, and obtains a fault classification result, an optimal matching pattern and a fault cause based on the progressive processing of the topological variation matrix; The fault prediction module predicts the future fault type and occurrence time based on the fault diagnosis result; The decision module generates an optimization strategy according to the fault prediction result.

2. The air energy heat pump fault diagnosis system based on topological matrix variation and special function group theory according to claim 1 is characterized in that: The data preprocessing module is used to construct a topological variation matrix based on a multi-dimensional data set. The topological variation matrix is ​​obtained by the following formula: Among them, M t (i, j) is the topological variation relationship between node i and node j at time t, D t (i) and D t (j) are the operating data of node i and node j at time t, σ is the standard deviation, k is the adjustment parameter, and t is time.

3. The system according to claim 2, characterized in that The feature extraction submodule obtains a feature matrix based on the topological variation matrix through the following steps: F t =Eig(M t ) Among them, F t is the characteristic matrix, Eig(M t ) is the topological variation matrix M t The eigenvalue decomposition results of .

4. The system according to claim 3, characterized in that The feature compression submodule further performs feature screening on the feature matrix based on the symmetry group in group theory. The screening matrix is ​​obtained by the following formula: S t ={g∈G∣φ(g,F t )>θ} Among them, S t is the feature screening matrix, G is the symmetry group, φ(g,F t ) is the feature importance function, and θ is the screening threshold. The feature vector after screening is: F ′ t =S t ·F t Among them, F ′ t is the compressed feature matrix.

5. The system according to claim 4, characterized in that The fault classification submodule uses Legendre polynomials to classify faults based on the feature matrix output by the feature compression submodule. The classification result is obtained by the following formula: Among them, C t is the classification result, P n (x) is the Legendre polynomial, α n is the classification weight, and N is the expansion order.

6. The system according to claim 5, characterized in that The pattern matching submodule matches the optimal fault mode through modulus operation and Euler function based on the fault classification result. The matching function is obtained by the following formula: Among them, d(P i ,C t ) is the failure mode P i And the classification result C t The matching degree, φ(P i ) is the Euler function.

7. The system according to claim 6, characterized in that The fault backtracking submodule is based on the optimal fault mode The Bessel function is used to trace back the cause of the fault. The traceback function is obtained by the following formula: Among them, R(x) is the fault cause weight function, J n (x) is the Bessel function.

8. The system according to claim 1, characterized in that The fault prediction module predicts possible future fault types and occurrence times based on the fault classification results and backtracking results in combination with historical data.

9. The system according to claim 1, characterized in that The decision module generates an optimization strategy based on the fault prediction results, including adjusting operating parameters, issuing alarms, and suggesting shutdown maintenance.

10. The system according to any one of claims 1 to 9, characterized in that: The system forms a complete workflow from data acquisition and processing to fault diagnosis, prediction and decision-making through progressive algorithm modules, wherein the output of each algorithm module directly serves as the input of the subsequent module to ensure the collaborative association and logical consistency between the modules of the system.

Images