Coal mill fault diagnosis method based on multivariable time sequence classification

Abstract

The invention discloses a coal mill fault diagnosis method based on multivariable time sequence classification, which relates to the technical field of mechanical fault diagnosis and comprises the following steps: collecting time sequence data of a coal mill, preprocessing the time sequence data and storing the time sequence data into a data set; constructing a self-supervised comparative learning model, and carrying out comparative learning training on different views in the data set; and outputting features based on the self-supervised contrast learning model, and inputting the features into an enhanced linear classifier for fault classification and diagnosis. According to the method, the self-supervised contrast learning model is constructed, so that the fault diagnosis effect under different working conditions is improved, and the adaptability and the stability of the method are enhanced; according to the method, the correlation characteristics in the multivariable time sequence data are mined by reinforcing the linear classifier, comprehensive analysis is carried out, comprehensive understanding of the running state of the coal mill is improved, therefore, more accurate fault diagnosis is achieved, and better effects are achieved in the aspects of accuracy, reliability and generalization ability.

Description

Technical Field

[0001] The invention relates to the technical field of mechanical fault diagnosis, and in particular to a coal mill fault diagnosis method based on multivariate time series classification. Background Art

[0002] Coal mill is a key equipment in thermal power plant, and its operating status directly affects the safety and economy of the power plant. Due to the complex operating environment and frequent load changes of coal mill, it is prone to failure. If the fault cannot be detected and diagnosed in time and accurately, it will not only affect the normal operation of the power plant, but also may cause serious economic losses. Therefore, the coal mill fault diagnosis method has important practical significance.

[0003] Traditional fault diagnosis methods mainly rely on experience and univariate analysis, and have problems such as low diagnostic accuracy and high false alarm rate. With the development of sensing technology and information processing technology, multivariate time series analysis methods are gradually applied to the field of fault diagnosis. Multivariate time series can comprehensively reflect the operating status of equipment. Through the joint analysis of multiple monitoring parameters, the accuracy and reliability of fault diagnosis can be improved. Summary of the invention

[0004] In view of the above-mentioned problems, the present invention is proposed.

[0005] Therefore, the technical problem solved by the present invention is: the existing coal mill fault diagnosis method has univariate analysis, low diagnostic accuracy, and high false alarm rate, and how to effectively improve the accuracy and reliability of coal mill fault diagnosis through multivariate time series classification.

[0006] To solve the above technical problems, the present invention provides the following technical solutions: a coal mill fault diagnosis method based on multivariate time series classification, comprising collecting coal mill time series data, preprocessing the time series data and storing it in a data set; constructing a self-supervised comparative learning model, and performing comparative learning training on different views in the data set; based on the output features of the self-supervised comparative learning model, the features are input into a reinforced linear classifier to perform fault classification and diagnosis.

[0007] As a preferred scheme of the coal mill fault diagnosis method based on multivariate time series classification described in the present invention, wherein: the collection of coal mill time series includes collecting time series of different fault operations and normal operations of the coal mill, and the collected time series includes coal mill current, coal mill outlet temperature, coal mill inlet primary air pressure, coal mill differential pressure, coal mill primary air flow, coal mill inlet temperature and coal mill primary air temperature.

[0008] As a preferred solution of the coal mill fault diagnosis method based on multivariate time series classification described in the present invention, the storage in the data set includes length segmentation of the collected time series data to form continuous time series samples, performing Fourier transform and time series decomposition technology, converting the time series samples into original time series views, frequency views and seasonal views, and classifying and storing the preprocessed data in the data set.

[0009] As a preferred solution of the coal mill fault diagnosis method based on multivariate time series classification described in the present invention, wherein: the construction of the self-supervised contrastive learning model includes designing three neural networks with the same structure and no shared parameters as feature extractors based on the original time series view, frequency view and seasonal view in the data set, which is expressed as:

[0010] g i =G(x i ), f i =F(x i ), s i =S(x i )

[0011] Among them, G is the original time series view, F is the frequency view, S is the seasonal view, and x i is the original data of the i-th time series sample, g i is the original time series view feature sample of the i-th time series sample, f i is the frequency view feature sample of the i-th time series sample, s i is the seasonal view feature sample of the i-th time series sample; based on the data set sample, construct sample g i 、f i and i The instance contrast loss function is expressed as:

[0012]

[0013] Among them, L Inst is the instance contrast loss function, L is the indicator function, exp is the exponential function, sim is the similarity function, N is the total number of data set samples, and g i ′ is g i The enhanced sample, f i ′ is f i The enhanced sample, s i ′ is s i The enhanced samples are k, k is the sample other than i, and t is the temperature parameter.

[0014] As a preferred solution of the coal mill fault diagnosis method based on multivariate time series classification described in the present invention, the comparative learning training includes adopting a joint learning method to construct a prototype clustering method, which is expressed as:

[0015]

[0016]

[0017] in, is the prototype cluster center vector of category c under the original time series view features, is the prototype cluster center vector of category c under the frequency view feature, is the prototype cluster center vector of category c under seasonal view features, is the i-th cluster center of the frequency view feature, is the i-th cluster center of the seasonal view feature, is the i-th cluster center of the original time series view feature; based on prototype joint contrast learning, construct sample g i 、f i and i The joint learning contrast loss function is expressed as:

[0018]

[0019] Among them, L Prot is the joint learning contrast loss function, is the joint cluster center of the original time series view and the seasonal view, is the joint cluster center of the seasonal view and the original time series view, is the joint cluster center of the frequency view and the seasonal view, is the joint cluster center of the original time series view and the frequency view, is the joint cluster center of the frequency view and the original time series view, is the joint clustering center of the seasonal view and the frequency view, M is the number of comparison samples of the i-th coal mill sample, M contains 2 positive samples and M-2 negative samples, and j is the sequence number of the comparison sample; based on the instance contrast loss function and the joint learning contrast loss function, the overall objective function is constructed, which is expressed as:

[0020]

[0021] Among them, L′ is the overall objective function, and λ is the parameter that balances the two types of loss functions.

[0022] As a preferred solution of the coal mill fault diagnosis method based on multivariate time series classification described in the present invention, the enhanced linear classifier includes outputting features based on a contrastive learning algorithm, inputting the features into the enhanced linear classifier, completing the classification and diagnosis of coal mill faults, constructing a linear classifier decision function based on logistic regression, and calculating the prediction label, which is expressed as:

[0023]

[0024] in, is the predicted label, w is the weight vector, b is the bias term, σ is the sigmoid function, z is the input feature vector, z contains the original time series view feature vector, frequency view vector and season view vector; construct a cross entropy loss function to train the linear classifier, expressed as:

[0025]

[0026] Among them, L″ is the cross entropy loss function, and y is the true fault type label belonging to the normal data label.

[0027] As a preferred scheme of the coal mill fault diagnosis method based on multivariate time series classification described in the present invention, the fault classification and diagnosis includes dividing the coal mill operation status into four modes: normal mode, coal mill blockage, coal mill short-circuit, and coal mill internal fault based on the output result of the enhanced linear classifier.

[0028] Another object of the present invention is to provide a coal mill fault diagnosis system based on multivariate time series classification, which can output features based on a self-supervised comparative learning model and input the features into an enhanced linear classifier to perform fault classification and diagnosis, thereby solving the problems of low diagnostic accuracy and high false alarm rate in current coal mill fault diagnosis methods.

[0029] As a preferred solution of the coal mill fault diagnosis system based on multivariate time series classification described in the present invention, it includes: a data processing module, a self-supervised contrast learning module, and a fault analysis module; the data processing module is used to collect coal mill time series data, pre-process the time series data and store them in a data set; the self-supervised contrast learning module is used to construct a self-supervised contrast learning model, and perform contrast learning training on different views in the data set; the fault analysis module is used to output features based on the self-supervised contrast learning model, input the features into a reinforced linear classifier, and perform fault classification and diagnosis.

[0030] A computer device comprises a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of a coal mill fault diagnosis method based on multivariate time series classification.

[0031] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of a coal mill fault diagnosis method based on multivariate time series classification.

[0032] Beneficial effects of the present invention: The coal mill fault diagnosis method based on multivariate time series classification provided by the present invention can more accurately identify the fault type and fault location of the coal mill through comparative learning technology and multivariate time series data analysis, thereby improving the accuracy and reliability of fault diagnosis; through advanced machine learning algorithms and efficient data processing technology, it can quickly process and analyze multivariate time series data, realize real-time diagnosis of coal mill faults, and reduce fault downtime; by constructing a self-supervised comparative learning model, comprehensively analyzing and comparative learning of multivariate time series data, adaptively adjusting the diagnosis model, improving the fault diagnosis effect under different working conditions, and enhancing the adaptability and stability of the method; by strengthening the linear classifier to fully explore the correlation features in the multivariate time series data, conducting a comprehensive analysis, and improving the comprehensive understanding of the coal mill operating status, thereby realizing more accurate fault diagnosis, the present invention has achieved better results in terms of accuracy, reliability and generalization ability. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use 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 ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0034] Figure 1 An overall flow chart of a coal mill fault diagnosis method based on multivariate time series classification provided by the first embodiment of the present invention;

[0035] Figure 2 A multi-view learning method structure diagram of a coal mill fault diagnosis method based on multivariate time series classification provided by the first embodiment of the present invention;

[0036] Figure 3 A module schematic diagram of a coal mill fault diagnosis system based on multivariate time series classification provided in the third embodiment of the present invention. DETAILED DESCRIPTION

[0037] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative work should fall within the scope of protection of the present invention.

[0038] Example 1, reference Figure 1-Figure 2 , which is an embodiment of the present invention, provides a coal mill fault diagnosis method based on multivariate time series classification, comprising:

[0039] S1: Collect coal mill time series data, preprocess the time series data and store them in a data set.

[0040] Furthermore, collecting the coal mill time series includes collecting the time series of different fault operations and normal operations of the coal mill, and the collected time series include coal mill current, coal mill outlet temperature, coal mill inlet primary air pressure, coal mill differential pressure, coal mill primary air flow, coal mill inlet temperature and coal mill primary air temperature.

[0041] It should be noted that storing in a data set includes length segmentation of the collected time series data to form continuous time series samples, performing Fourier transform and time series decomposition techniques, converting the time series samples into original time series views, frequency views and seasonal views, and classifying and storing the preprocessed data in a data set.

[0042] It should also be noted that the frequency signal samples of the time series signal samples are generated by Fourier transform using the frequency. In addition to the frequency signal, the time signal is decomposed into trend, season and residual signals using the time series decomposition technology method. In the model of the present invention, the seasonal signal decomposed by the time series decomposition technology is selected as a new view together with the time signal, which allows comparative learning of the seasonal signal at the same time as the frequency signal and the time signal, thereby providing insights that are more sensitive to time series. The collected data is preprocessed by Fourier transform and time series decomposition technology, and the time series data is converted into the original time series view, frequency view and seasonal view, providing multi-perspective input for the model, thereby improving the representation and learning capabilities of the model, and can extract features related to the operating status of the coal mill, thereby improving the accuracy of fault diagnosis, and can reduce the impact of noise and outliers on fault diagnosis, thereby improving the accuracy and reliability of fault diagnosis.

[0043] S2: Build a self-supervised contrastive learning model to perform contrastive learning training on different views in the dataset.

[0044] Furthermore, constructing a self-supervised contrastive learning model includes designing three neural networks with the same structure and no shared parameters as feature extractors based on the original time series view, frequency view and seasonal view in the dataset, which are expressed as:

[0045] g i =G(x i ),f i =F(x i ),s i =S(x i )

[0046] Among them, G is the original time series view, F is the frequency view, S is the seasonal view, and x i is the original data of the i-th time series sample, g i is the original time series view feature sample of the i-th time series sample, f i is the frequency view feature sample of the i-th time series sample, s i is the seasonal view feature sample of the i-th time series sample; based on the data set sample, construct sample g i 、f i and i The instance contrast loss function is expressed as:

[0047]

[0048] Among them, L Inst is the instance contrast loss function, L is the indicator function, exp is the exponential function, sim is the similarity function, N is the total number of data set samples, and g i ′ is the enhanced sample of gi, f i ′ is f i The enhanced sample, s i ′ is s i The enhanced samples are k, k is the sample other than i, and t is the temperature parameter.

[0049] It should be noted that the contrastive learning training includes using joint learning to construct a prototype clustering method, which is expressed as:

[0050]

[0051] in, is the prototype cluster center vector of category c under the original time series view features, is the prototype cluster center vector of category c under the frequency view feature, is the prototype cluster center vector of category c under seasonal view features, is the i-th cluster center of the frequency view feature, is the i-th cluster center of the seasonal view feature, is the i-th cluster center of the original time series view feature; based on prototype joint contrast learning, construct sample g i 、f i and i The joint learning contrast loss function is expressed as:

[0052]

[0053] Among them, L Prot is the joint learning contrast loss function, is the joint cluster center of the original time series view and the seasonal view, is the joint cluster center of the seasonal view and the original time series view, is the joint cluster center of the frequency view and the seasonal view, is the joint cluster center of the original time series view and the frequency view, is the joint cluster center of the frequency view and the original time series view, is the joint clustering center of the seasonal view and the frequency view, M is the number of comparison samples of the i-th coal mill sample, M contains 2 positive samples and M-2 negative samples, and j is the sequence number of the comparison sample; based on the instance contrast loss function and the joint learning contrast loss function, the overall objective function is constructed, which is expressed as:

[0054]

[0055] Among them, L′ is the overall objective function, and λ is the parameter that balances the two types of loss functions.

[0056] It should also be noted that if Figure 2 As shown, in the present invention, three views of time signal, frequency signal and seasonal signal are used, and instance-level contrastive learning is used in a single view to obtain the hidden feature representation in the view. After obtaining the hidden feature representation of the three views, prototype-based contrastive learning is used to combine the feature representations of different views, which includes calculating the clustering index in the view, which is represented as v in the figure. g 、v f 、v s , and then calculate the indexes from the other two views, which are represented in the figure as Three neural networks with the same structure and no shared parameters are designed as feature extractors. The encoder of each individual view is composed of a 3-layer CNN pyramid network. A maximum pooling module is added after each convolution block. The enhanced samples corresponding to each individual view are obtained by dropout after each convolution block. A self-supervised contrastive learning model is constructed. Through comprehensive analysis and contrastive learning of multivariate time series data, the diagnosis model is adaptively adjusted to improve the fault diagnosis effect under different working conditions and enhance the adaptability and stability of the method. The model learns the feature representation of time series data to capture the pattern differences of the coal mill under different operating conditions. Using normal data and different fault mode data, the model automatically learns the feature differences between data through a contrastive learning strategy without explicit labels. By maximizing the feature consistency of the same type of data and minimizing the feature differences of different types of data, the model can effectively learn data patterns without labels.

[0057] S3: Based on the output features of the self-supervised contrastive learning model, the features are input into the enhanced linear classifier for fault classification and diagnosis.

[0058] Furthermore, the enhanced linear classifier includes outputting features based on the contrastive learning algorithm, inputting the features into the enhanced linear classifier, completing the classification and diagnosis of the coal mill fault, constructing the linear classifier decision function based on logistic regression, and calculating the predicted label, which is expressed as:

[0059]

[0060] in, is the predicted label, w is the weight vector, b is the bias term, σ is the sigmoid function, z is the input feature vector, z contains the original time series view feature vector, frequency view vector and season view vector; construct a cross entropy loss function to train the linear classifier, expressed as:

[0061]

[0062] Among them, L″ is the cross entropy loss function, and y is the true fault type label belonging to the normal data label.

[0063] It should be noted that fault classification and diagnosis includes classifying the coal mill operation status into four modes: normal mode, coal mill coal blockage, coal mill coal shortage, and coal mill internal fault based on the output results of the enhanced linear classifier.

[0064] It should also be noted that the labeled coal mill operation data is used to conduct supervised learning training on the linear classifier, further optimize the model, fully explore the correlation features in the multivariate time series data, conduct comprehensive analysis, and improve the comprehensive understanding of the coal mill operation status, thereby achieving more accurate fault diagnosis, so that it can accurately classify the operation status of the coal mill. The trained linear classifier can classify the input coal mill data into four modes: normal mode, coal mill blockage, coal mill shortage, and coal mill internal failure.

[0065] Example 2 is an embodiment of the present invention, which provides a coal mill fault diagnosis method based on multivariate time series classification. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through economic benefit calculation and simulation experiments.

[0066] The system is simulated and verified using MATLAB / Simulink software. The simulation parameters are shown in Table 1.

[0067] Table 1 Simulation parameters

[0068] Parameter settings Parameter Value Training data set (number) 1196 Test data set (number) 697 Feature Dimension 256 Time series length 128 Number of variables (number of channels) 7 G, F, S backbone model Pyramid three-layer convolution Pyramid convolution first layer input / output (7，32) Pyramid convolution second layer input / output (32，64) Pyramid convolution third layer input / output (64，128) Pyramid convolution kernel size 7 Negative sample clipping rate 0.6

[0069] Refer to Table 2 to record and analyze the experimental data.

[0070] Table 2 Experimental data record table

[0071] Comparison Models Accuracy Accuracy Recall F1 score AUROC AUPRC KNN 0.7479 0.8437 0.8171 0.8334 0.8307 0.8428 TS2vec 0.9121 0.8228 0.8518 0.8215 0.9389 0.8278 TS-TCC 0.9548 0.9344 0.9094 0.9344 0.9767 0.9363 Our invention 0.9656 0.9698 0.9703 0.9536 0.9789 0.9367

[0072] By comparing the diagnostic accuracy, precision, recall, F1 score, AUROC and AUPRC of different fault types, we can see the beneficial effects of our invention. For example, the diagnostic accuracy of our invented model for different fault types such as coal blockage in coal mill, coal breakage in coal mill, internal fault in coal mill, etc. is over 95%, which is significantly improved compared with other comparison models (KNN: about 70%, TS2vec: about 90%, TS-TCC: about 95%). This shows that our invented model can more accurately identify coal mill faults, and has higher precision, recall and F1 score, which can effectively reduce the risk of misdiagnosis and missed diagnosis, and improve the reliability of coal mill fault diagnosis. Through this comprehensive evaluation, the potential fault types of coal mill can be accurately identified, and targeted repairs and maintenance can be carried out, thereby Improve the operating efficiency and safety of the coal mill. Our invention is superior to the traditional coal mill fault diagnosis method because our invention not only takes into account the operating status and fault characteristics of the coal mill, but also introduces multivariate time series analysis and self-supervised comparative learning technology, which can more comprehensively and accurately identify the potential faults of the coal mill, and accurately classify and diagnose faults by strengthening the linear classifier. Compared with the existing technology, our invention provides a more advanced and reliable coal mill fault diagnosis method. By integrating multivariate time series analysis and self-supervised comparative learning technology, it can more accurately identify the potential faults of the coal mill, and accurately classify and diagnose faults by strengthening the linear classifier. This not only improves the efficiency of coal mill fault diagnosis, but also helps to improve the operating efficiency and safety of the coal mill. Therefore, our invention is creative.

[0073] Example 3, reference Figure 3 , which is an embodiment of the present invention, provides a coal mill fault diagnosis system based on multivariate time series classification, including a data processing module, a self-supervised comparative learning module, and a fault analysis module.

[0074] The data processing module is used to collect the time series data of the coal mill, pre-process the time series data and store it in the data set; the self-supervised contrastive learning module is used to build a self-supervised contrastive learning model and perform contrastive learning training on different views in the data set; the fault analysis module is used to output features based on the self-supervised contrastive learning model, input the features into the enhanced linear classifier, and perform fault classification and diagnosis

[0075] If the function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods of each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0076] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in conjunction with such instruction execution systems, devices or apparatuses. For the purposes of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in conjunction with such instruction execution systems, devices or apparatuses.

[0077] More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk case (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be a paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering, or processing in another suitable manner as necessary, and then stored in a computer memory.

[0078] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc. It should be noted that the above embodiments are only used to illustrate the technical solution of the present invention and are not limited. Although the present invention is described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that the technical solution of the present invention can be modified or replaced by equivalents without departing from the spirit and scope of the technical solution of the present invention, which should be included in the scope of the claims of the present invention.

[0079] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A coal mill fault diagnosis method based on multivariate time series classification, characterized in that: include: Collect coal mill time series data, pre-process the time series data and store them in a data set; Build a self-supervised contrastive learning model to perform contrastive learning training on different views in the dataset; Based on the output features of the self-supervised contrastive learning model, the features are input into the enhanced linear classifier for fault classification and diagnosis.

2. The coal mill fault diagnosis method based on multivariate time series classification according to claim 1, characterized in that: The collected coal mill time series includes collecting time series of different fault operations and normal operation of the coal mill, and the collected time series includes coal mill current, coal mill outlet temperature, coal mill inlet primary air pressure, coal mill differential pressure, coal mill primary air flow, coal mill inlet temperature and coal mill primary air temperature.

3. The coal mill fault diagnosis method based on multivariate time series classification according to claim 2 is characterized in that: The storing in the data set includes length segmenting the collected time series data to form continuous time series samples, performing Fourier transform and time series decomposition technology, converting the time series samples into original time series views, frequency views and seasonal views, and classifying and storing the pre-processed data in the data set.

4. The coal mill fault diagnosis method based on multivariate time series classification according to claim 3 is characterized in that: The self-supervised contrastive learning model is constructed by designing three neural networks with the same structure and no shared parameters as feature extractors based on the original time series view, frequency view and seasonal view in the data set, which are expressed as: g i =G(x i ),f i =F(x i ),s i =S(x i ) Among them, G is the original time series view, F is the frequency view, S is the seasonal view, and x i is the original data of the i-th time series sample, g i is the original time series view feature sample of the i-th time series sample, f i is the frequency view feature sample of the i-th time series sample, s i is the seasonal view feature sample of the i-th time series sample; Based on the data set sample, construct sample g i 、f i and i The instance contrast loss function is expressed as: Among them, L Inst is the instance contrast loss function, L is the indicator function, exp is the exponential function, sim is the similarity function, N is the total number of data set samples, and g i ′ is g i The enhanced sample, f i ′ is f i The enhanced sample, S i ′ is S i The enhanced samples are k, k is the sample other than i, and t is the temperature parameter.

5. The coal mill fault diagnosis method based on multivariate time series classification according to claim 4 is characterized in that: The contrastive learning training includes using joint learning to construct a prototype clustering method, which is expressed as: in, is the prototype cluster center vector of category c under the original time series view features, is the prototype cluster center vector of category c under the frequency view feature, is the prototype cluster center vector of category c under seasonal view features, is the i-th cluster center of the frequency view feature, is the i-th cluster center of the seasonal view feature, is the i-th cluster center of the original time series view feature; Based on prototype-based joint contrastive learning, construct sample g i 、f i and i The joint learning contrast loss function is expressed as: Among them, L Prot is the joint learning contrast loss function, is the joint cluster center of the original time series view and the seasonal view, is the joint cluster center of the seasonal view and the original time series view, is the joint cluster center of the frequency view and the seasonal view, is the joint cluster center of the original time series view and the frequency view, is the joint cluster center of the frequency view and the original time series view, is the joint cluster center of the seasonal view and the frequency view, M is the number of comparison samples of the i-th coal mill sample, M contains 2 positive samples and M-2 negative samples, and j is the comparison sample number; Based on the instance contrast loss function and the joint learning contrast loss function, the overall objective function is constructed, which is expressed as: Among them, L′ is the overall objective function, and λ is the parameter that balances the two types of loss functions.

6. The coal mill fault diagnosis method based on multivariate time series classification according to claim 5, characterized in that: The enhanced linear classifier includes outputting features based on a contrastive learning algorithm, inputting the features into the enhanced linear classifier, completing the classification and diagnosis of coal mill faults, constructing a linear classifier decision function based on logistic regression, and calculating the predicted label, which is expressed as: in, is the predicted label, w is the weight vector, b is the bias term, σ is the sigmoid function, z is the input feature vector, and z contains the original time series view feature vector, frequency view vector, and season view vector; Construct a cross entropy loss function to train a linear classifier, expressed as: Among them, L ″ is the cross entropy loss function, y is the true fault type label belonging to the normal data label.

7. The coal mill fault diagnosis method based on multivariate time series classification according to claim 6, characterized in that: The fault classification and diagnosis includes classifying the coal mill operation status into four modes: normal mode, coal mill coal blockage, coal mill coal shortage, and coal mill internal fault based on the output result of the enhanced linear classifier.

8. A system using the coal mill fault diagnosis method based on multivariate time series classification as claimed in any one of claims 1 to 7, characterized in that: Including data processing module, self-supervised contrast learning module, and fault analysis module; The data processing module is used to collect coal mill time series data, pre-process the time series data and store them in a data set; The self-supervised contrastive learning module is used to construct a self-supervised contrastive learning model to perform contrastive learning training on different views in the data set; The fault analysis module is used to output features based on the self-supervised contrastive learning model, and input the features into the enhanced linear classifier to perform fault classification and diagnosis.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the coal mill fault diagnosis method based on multivariate time series classification described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the coal mill fault diagnosis method based on multivariate time series classification described in any one of claims 1 to 7 are implemented.

Images