A method and system for detecting and diagnosing faults of chillers

Abstract

The present invention discloses a chiller fault detection and diagnosis method and system, including: constructing a three-dimensional modeling data set; using a tensor analysis method to extract dynamic features between batches from the three-dimensional modeling data set, and using a slow feature analysis method to extract dynamic features within the same batch, thereby constructing a tensor slow feature analysis model; extracting tensor slow features from the working condition data to be tested, constructing monitoring statistics and determining control limits to perform fault detection; extracting tensor slow features from fault working condition data and historical fault data respectively, assigning different distribution weights and misclassification costs to the tensor slow features of each type of historical fault data, iteratively training a data gravity classifier according to the distribution weights, updating the weights of the data gravity classifier's correctly classified samples and misclassified samples according to the misclassification costs, thereby constructing a cost-sensitive data gravity classifier; and obtaining the fault category according to the cost-sensitive data gravity classifier. The performance of chiller fault detection and fault identification is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of fault detection and diagnosis, and in particular to a method and system for detecting and diagnosing faults of a chiller. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] The chiller is the main energy-consuming equipment in the HVAC system, including the compressor, evaporator, condenser and expansion valve, which realizes the cooling and heating of the variable air volume air conditioning unit, and plays an important role in achieving the optimal control and energy management of the HVAC system. The failure of the chiller will not only affect the operation of the HVAC system, but also lead to an increase in operating energy consumption. Therefore, the realization of chiller fault detection and diagnosis technology is of great significance to ensure the normal operation of the HVAC system and reduce the system energy consumption.

[0004] By analyzing the actual operating status of the chiller, it is found that the chiller has the characteristics of periodic repeated operation (one operating cycle corresponds to one working day). Therefore, the chiller can be regarded as a special intermittent control process. The operating status of the chiller on a working day is defined as an operating batch. Indoor personnel usually do not make significant changes to the settings of air-conditioning temperature and humidity in consecutive days or even weeks. The outdoor climate will also remain relatively stable in a short period of time. Therefore, the chiller will show the characteristics of periodic cycles in the operation of several consecutive days.

[0005] However, there are still offsets and differences between different batches of data (i.e., significant inter-batch time-varying dynamic characteristics). At the same time, during the daily operation of the chiller, changes in ambient temperature and humidity (caused by outdoor weather changes), changes in indoor load and set parameters, and changes in the chiller operation mode under closed-loop control will all lead to significant intra-batch time-varying dynamic characteristics. In addition, the frequency of occurrence of different fault modes in actual operation of the chiller is also different; the number of occurrences of some fault modes is much greater than that of other fault modes, resulting in an imbalance in the number of fault data of different categories.

[0006] However, existing chiller fault detection and identification methods generally do not consider the above characteristics, and have the following problems: (1) There are dynamic characteristics between batches in the operation of chillers between adjacent days, but the existing chiller fault detection methods cannot handle the time-varying dynamic characteristics between batches; (2) There are time-varying dynamic characteristics within batches in the operation of chillers on each day, and the existing chiller dynamic monitoring methods cannot effectively capture the time-varying dynamic characteristics within batches. (3) In the fault identification stage of chillers, the collected fault data of different categories are imbalanced in quantity, resulting in low accuracy of fault identification of traditional fault identification methods for small amounts of data. Summary of the invention

[0007] In order to solve the above problems, the present invention proposes a chiller fault detection and diagnosis method and system, which processes the time-varying dynamic characteristics of the chiller between batches and within batches based on tensor slow feature analysis, and solves the problem of unbalanced distribution of fault data of different categories in quantity based on cost-sensitive data gravity classifier, thereby improving the performance of chiller fault detection and fault identification.

[0008] In order to achieve the above object, the present invention adopts the following technical solution:

[0009] In a first aspect, the present invention provides a chiller fault detection and diagnosis method, comprising:

[0010] Obtain the normal operating data of the chiller and use it to construct a 3D modeling data set;

[0011] The tensor analysis method is used to extract dynamic features between batches of the 3D modeling dataset, and the slow feature analysis method is used to extract dynamic features within the same batch, so as to construct a tensor slow feature analysis model.

[0012] The tensor slow feature analysis model is used to extract the tensor slow features from the working condition data to be tested, and the monitoring statistics are constructed based on the tensor slow features and the control limits are determined to detect the faults of the chiller.

[0013] Tensor slow features are extracted from fault condition data and historical fault data respectively, and different distribution weights and misclassification costs are assigned to the tensor slow features of each type of historical fault data. The data gravity classifier is iteratively trained according to the distribution weights, and the weights of the correctly classified samples and misclassified samples of the data gravity classifier are updated according to the misclassification costs, so as to construct a cost-sensitive data gravity classifier.

[0014] The fault category of the tensor slow feature of the fault condition data is obtained according to the cost-sensitive data gravity classifier.

[0015] As an optional implementation, the three-dimensional modeling data set is preprocessed, including: dividing the three-dimensional modeling data set into K time slice matrices according to the sampling time; arranging the time slice matrices into a two-dimensional matrix in the batch direction; at each sampling time, normalizing the two-dimensional matrix to zero mean and unit variance along each variable, dividing the processed two-dimensional data matrix into K time slice matrices again, and rearranging the time slice matrix into a three-dimensional matrix.

[0016] As an optional implementation, the objective function of the tensor slow feature analysis model is:

[0017]

[0018] Among them, U and V are load matrices, X i is the data in the i-th batch, and I is the number of batches in the 3D modeling dataset;

[0019] The load matrices U and V are obtained by performing tensor decomposition on the three-dimensional modeling data set, and the tensor slow features within the same batch of chillers are extracted based on the load matrices U and V.

[0020] As an alternative implementation, the tensor slow feature Y i Expressed as: Y i =U T X i V.

[0021] As an optional implementation, the misclassification cost c(y i )for:

[0022]

[0023]

[0024] Among them, F i ,F j is the partial F value of the i-th and j-th tensor slow features; K represents the number of sampling points in a batch.

[0025] As an optional implementation, when iteratively training the data gravity classifier, at the tth iteration, the data gravity classifier h t The weight α t for:

[0026]

[0027] Among them, c(y i ) is the sample y i The misclassification cost, D t (i) is the tth iteration sample y i The weight of .

[0028] As an optional implementation, after updating the weights according to the misclassification cost, at the tth iteration, the sample y i The weight D(i) is updated as:

[0029] Among them, γ i is the category label, Z t It is the normalized coefficient of the updated weight D(i) after the tth iteration.

[0030] In a second aspect, the present invention provides a chiller fault detection and diagnosis system, comprising:

[0031] A data set construction module is configured to obtain normal operating data of the chiller and construct a three-dimensional modeling data set based on the normal operating data of the chiller;

[0032] A model building module is configured to use a tensor analysis method to extract dynamic features between batches and a slow feature analysis method to extract dynamic features within the same batch for the three-dimensional modeling data set, thereby building a tensor slow feature analysis model;

[0033] A fault detection module is configured to extract tensor slow features from the working condition data to be tested using a tensor slow feature analysis model, construct monitoring statistics based on the tensor slow features and determine control limits to perform fault detection on the chiller;

[0034] The classifier construction module is configured to extract tensor slow features from fault condition data and historical fault data respectively, assign different distribution weights and misclassification costs to the tensor slow features of each type of historical fault data, iteratively train the data gravity classifier according to the distribution weights, and update the weights of the correctly classified samples and misclassified samples of the data gravity classifier according to the misclassification costs, so as to construct a cost-sensitive data gravity classifier;

[0035] The fault identification module is configured to obtain the fault category of the tensor slow feature of the fault condition data according to the cost-sensitive data gravity classifier.

[0036] In a third aspect, the present invention provides an electronic device comprising a memory and a processor, and computer instructions stored in the memory and executed on the processor, wherein when the computer instructions are executed by the processor, the method described in the first aspect is performed.

[0037] In a fourth aspect, the present invention provides a computer-readable storage medium for storing computer instructions, wherein when the computer instructions are executed by a processor, the method described in the first aspect is performed.

[0038] Compared with the prior art, the present invention has the following beneficial effects:

[0039] The present invention proposes a chiller fault detection and diagnosis method and system. According to the periodic repetitive operation characteristics of the chiller, normal operating condition data of several adjacent days are used to construct a three-dimensional modeling data set, and the significant time-varying dynamic characteristics between different batches of data of the chiller are fully considered.

[0040] The present invention proposes a chiller fault detection and diagnosis method and system, constructs a fault detection algorithm based on tensor slow feature analysis, and directly establishes a monitoring model on a normalized three-dimensional normal operating condition data set, which can efficiently process the time-varying dynamic characteristics of the chiller between different batches of data and within the same batch of data in actual operation.

[0041] The present invention proposes a method and system for detecting and diagnosing faults of chillers, constructs a fault identification method based on a cost-sensitive data gravity classifier, effectively solves the problem of imbalanced distribution of fault data of different categories in chillers in terms of quantity, and significantly improves the type identification accuracy of a small number of fault data to be identified.

[0042] Advantages of additional aspects of the present invention will be given in part in the following description, and in part will become obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0044] Figure 1 A schematic flow chart of a chiller fault detection and diagnosis method provided in Example 1 of the present invention;

[0045] Figure 2 A schematic diagram of a 3D modeling dataset preprocessing process provided in Example 1 of the present invention;

[0046] Figure 3 A processing flow chart of a tensor slow feature analysis model provided in Example 1 of the present invention;

[0047] Figure 4 A flowchart of fault identification of a cost-sensitive data gravity classifier provided in Example 1 of the present invention;

[0048] Figure 5 This is a flow chart of the cost-sensitive data gravity classifier algorithm provided in Example 1 of the present invention. DETAILED DESCRIPTION

[0049] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0050] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0051] It should be noted that the terms used herein are only for describing specific embodiments, and are not intended to limit exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that the terms "include" and "have" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0052] In the absence of conflict, the embodiments of the present invention and the features of the embodiments may be combined with each other.

[0053] Example 1

[0054] like Figure 1 As shown, this embodiment provides a chiller fault detection and diagnosis method, which is based on the combination of tensor slow feature analysis and cost-sensitive gravity classifier. In the fault detection stage, the fault detection based on tensor slow feature analysis effectively handles the time-varying dynamic characteristics of the chiller between batches and within batches. In the fault identification stage, the fault identification based on the cost-sensitive data gravity classifier solves the problem of imbalanced distribution of fault data of different categories in quantity, and improves the performance of chiller fault detection and fault identification. Specifically including:

[0055] Obtain the normal operating data of the chiller for several consecutive days and use it to construct a 3D modeling data set;

[0056] The tensor analysis method is used to extract dynamic features between batches of the 3D modeling dataset, and the slow feature analysis method is used to extract dynamic features within the same batch, so as to construct a tensor slow feature analysis model.

[0057] The tensor slow feature analysis model is used to extract the tensor slow features from the working condition data to be tested, and the monitoring statistics are constructed based on the tensor slow features and the control limits are determined to detect the faults of the chiller.

[0058] Tensor slow features are extracted from fault condition data and historical fault data respectively, and different distribution weights and misclassification costs are assigned to the tensor slow features of each type of historical fault data. The data gravity classifier is iteratively trained according to the distribution weights, and the weights of the correctly classified samples and misclassified samples of the data gravity classifier are updated according to the misclassification costs, so as to construct a cost-sensitive data gravity classifier.

[0059] The fault category of the tensor slow feature of the fault condition data is obtained according to the cost-sensitive data gravity classifier.

[0060] In this embodiment, in order to take into account the time-varying dynamic characteristics between batches of chillers, this embodiment obtains the normal operating data of the chillers for several consecutive days to construct a three-dimensional modeling data set; then, the three-dimensional modeling data set is preprocessed to extract the average motion trajectory of the process variables in multiple batch data sets, thereby highlighting the time-varying dynamic characteristics between different batches; after preprocessing, a normalized three-dimensional modeling data set is constructed again, and the expanded two-dimensional matrix is ​​arranged into a normalized three-dimensional training data set to construct a tensor slow feature analysis model. Figure 2 As shown, the specific process includes the following:

[0061] (1) Use the normal operating data of the chiller for several consecutive days to build a 3D modeling dataset Where I represents the number of batches in the data set, J represents the number of process variables in a batch, and K represents the number of sampling points in a batch;

[0062] 3D modeling dataset Divide into K time slice matrices according to the sampling time k=1,2,…,K; time slice matrix k=1,2,…,K are arranged into a two-dimensional matrix according to the batch direction

[0063] (2) At each sampling moment, along each variable, a two-dimensional matrix Perform zero mean and unit variance normalization; for the two-dimensional matrix The above standardization process is performed to extract the average operation trajectory of the process variables under normal operation of multiple intermittent processes, highlighting the time-varying dynamic characteristics between different operation batches. Its mathematical expression is as follows:

[0064]

[0065] (3) The standardized two-dimensional data matrix X(I×KJ) is again divided into K time slice matrices, and these time slice matrices are rearranged into a three-dimensional matrix.

[0066] In this embodiment, a tensor slow feature analysis model is established based on the preprocessed three-dimensional modeling data set to process the time-varying dynamic characteristics of the chiller between batches and within batches, and implement fault detection;

[0067] Firstly, the tensor analysis method is used to extract the dynamic characteristics of batches of chillers on the preprocessed 3D modeling data set. The tensor analysis method regards each batch of data in the 3D modeling data set as a second-order tensor. There is no need to expand the 3D modeling data into a 2D matrix form, which will not destroy the original spatial structure of each batch of data. The local neighborhood structure of the data is maintained, and the time-varying dynamic characteristics between batches are effectively processed.

[0068] Then, the slow feature method is used to extract the slowly changing tensor slow feature information from each batch of data to characterize the potential driving force that causes the change of the chiller operating conditions and more fully handle the time-varying dynamic characteristics within the batch;

[0069] Finally, monitoring statistics are constructed based on the extracted tensor slow feature information and control limits are determined to detect faults of the chiller and monitor the operating status of the chiller.

[0070] like Figure 3 As shown, the specific process includes the following: In order to avoid the problem of destroying the original data structure in the process of expanding the three-dimensional modeling data into two-dimensional data in the traditional method, this embodiment uses the tensor space Given a data set X = {X i},i=1,2,…,I, tensor slow feature algorithm finds two transformation matrices and The matrix Projection (K 1 ≤K,J 1 ≤J), so that Y i Can represent X i , where Y i =U T X i V, directly perform tensor decomposition on the 3D modeling data to obtain the load matrices U and V.

[0071] The optimization objective of the tensor slow feature analysis model is expressed as:

[0072]

[0073] Since ||A|| 2 =tr(AA T ),get:

[0074]

[0075] The objective function of the tensor slow feature analysis model is:

[0076]

[0077] In order to construct a pseudo time series to provide time series information, j=1,2,…,l represents the i-th two-dimensional matrix X in the three-dimensional matrix X(I×K×J) i The two-dimensional matrix X i The pseudo time series can be constructed based on the neighbor information of the data points as follows:

[0078]

[0079] The pseudo time series matrix of the three-dimensional matrix X(I×K×J) is constructed as follows:

[0080] T(2Il×K×J)=[t 1 ,t 2 ,t 3 ,…,t I ](6)

[0081] For 1≤i≤Il, the (2i-1)th and 2i-th two-dimensional matrices in the pseudo time series matrix T are represented as τ(2i-1) and τ(2i), respectively. Therefore, it can be considered that the time variation information exists between the two-dimensional matrices τ(2i-1) and τ(2i). The objective function of the tensor slow feature can be rewritten as:

[0082]

[0083] If you The objective function of the tensor slow feature can be obtained:

[0084]

[0085] By solving the generalized eigenvalue decomposition problem, the optimal transformation matrix U is obtained:

[0086] M V u=λN V u(9)

[0087] Similarly, since ||A|| 2 =tr(A T A), we can get:

[0088]

[0089] The objective function of tensor slow feature analysis can also be expressed as:

[0090]

[0091] Pseudo time series matrix based on the constructed three-dimensional matrix X(I×K×J):

[0092] T(2Il×K×J)=[t 1 ,t 2 ,t 3,…,t I ](12)

[0093] The objective function of tensor slow feature analysis is rewritten as:

[0094]

[0095] If you Then we can get:

[0096]

[0097] Similarly, by solving the generalized eigenvalue decomposition problem, the optimal transformation matrix V can be obtained:

[0098] M U v=λN U v(15)

[0099] It should be noted that the two generalized eigenvalue problems in equations (9) and (15) are interdependent and cannot be solved independently. Therefore, firstly, the matrix U is set to the unit matrix, and the matrix V is obtained by solving equation (16):

[0100] M U v=λN U v(16)

[0101] After solving the matrix V, continue to solve equation (9) to update the value of the matrix U:

[0102] M V u=λN V u (17)

[0103] According to equation (16) and equation (17), we can solve the former J 1 and K 1 The smallest generalized eigenvalues ​​are obtained and the load matrices U and V are constructed: and

[0104] Based on the load matrices U and V, the data X of each batch of chillers is extracted according to formula (18): i The tensor slow feature Y i , more fully extract the time-varying dynamic characteristics of the batch of chillers;

[0105] Y i =U T X i V(18)

[0106] Finally, based on the extracted tensor slow feature Y i Construct monitoring statistics T 2and SPE, and estimates the control limits corresponding to the monitoring statistics based on the three-dimensional modeling data of the normal operating condition data to implement fault detection on the chiller.

[0107] In this embodiment, after a chiller failure is detected, first, a tensor slow feature analysis model is used to extract tensor slow features from fault condition data and historical fault data;

[0108] Then, in order to solve the problem of imbalanced distribution of different types of fault data of chillers in quantity, a cost-sensitive data gravity classifier is constructed to ensure the classification accuracy of the few-class fault data of chillers. Specifically, different distribution weights and misclassification costs are assigned to the tensor slow features of each type of historical fault data, and the data gravity classifier is iteratively trained according to the distribution weights. The weights of the correctly classified samples and misclassified samples of the data gravity classifier are updated according to the misclassification costs, the weights of the correctly classified samples are reduced, and the weights of the misclassified samples are increased, so that the next round of data gravity classifier focuses on the few-class samples with large weights.

[0109] Finally, the fault categories of the tensor slow features of the fault condition data are identified by weighted voting, which improves the recognition accuracy of the few-category fault data of the chiller.

[0110] like Figure 4 As shown, the specific process includes the following:

[0111] Based on the tensor slow feature analysis model, the load matrices U and V are used to extract the tensor slow feature information of the three-dimensional normal operating data and each type of historical fault data. and Y Fi , i = 1, 2, ..., C; use the weighted method based on the partial F value to use the normal working condition data to find the weighted vector that can characterize each type of fault data to better explain The contribution of each feature to the fault; calculate the partial F value of the jth feature:

[0112]

[0113] For the i-th type of fault data Define the weight vector for:

[0114]

[0115] in,

[0116] The weight vectors of each type C fault data can be obtained

[0117] Set the tensor slow feature information Y of each type of historical fault data Fi,i=1,2,…,the misclassification cost of the samples in C, and set different weights for different types of training samples according to the misclassification cost; according to the weight distribution of the samples, the data gravity classifier is trained by an iterative method, and the weights of the different types of samples that are currently correctly classified are reduced according to the misclassification cost, and the weights of the misclassified samples are increased, so that the next round focuses on those samples with large weights, and finally the tensor slow feature information y of the fault data is treated by weighted voting. i The classification results are used to make the final decision.

[0118] The algorithm flow of the cost-sensitive data gravity classifier is as follows Figure 5 As shown; given training data {y F (i),γ i},i=1,2,…,n F , where γ i =1,2,…,C is the category label of the data, and different initial distribution weights are set for each type of historical fault data

[0119]

[0120] In order to improve the classification accuracy of minority class samples, a larger misclassification cost is imposed on minority class samples and a smaller misclassification cost is imposed on majority class samples; The misclassification cost c(y i ) is defined as:

[0121]

[0122] in:

[0123] In the data gravity classifier, the distribution weight of each sample is used as its quality, rather than simply considering the quality of the sample as 1; dataset Y Fi Y Fj The data gravity of the rth sample point in the tth iteration is defined as:

[0124]

[0125] Calculate the C-type fault data for the test data y Fj Gravity After that, the test data y Fj The category of is consistent with the category of the fault dataset that exerts the greatest attraction on it:

[0126]

[0127] At the tth iteration, the data gravity classifier h t The weight αt Calculated as:

[0128]

[0129] In the tth iteration, sample y i The weight D(i) is updated as:

[0130]

[0131] Among them, c(y i ) is the sample y i The misclassification cost, γ i Represents sample y i The category label, Z t Represents the normalized coefficient of the updated weight D(i) after the tth iteration.

[0132] Cost-sensitive data gravity classifier for test sample y new When classifying, set y new The mass m new =1; first calculate T weak classifiers for y new Then, the classification results of each weak classifier are weighted, and the T weak classifiers are determined according to the voting principle. new The final integrated classification result is:

[0133]

[0134] In this embodiment, in the fault detection and fault identification of the chiller, multiple sensors are used to collect the actual operation data of the chiller, monitor whether the chiller has a fault, and identify the type of fault data detected. First, the tensor slow feature information of the data is extracted based on the tensor slow feature analysis model, and the corresponding monitoring statistics are calculated to monitor the operating status of the chiller; after the fault is detected, the cost-sensitive data gravity classifier is used to identify the fault type of the chiller based on the tensor slow feature information of the fault data and historical fault data.

[0135] Example 2

[0136] This embodiment provides a chiller fault detection and diagnosis system, including:

[0137] A data set construction module is configured to obtain normal operating data of the chiller and construct a three-dimensional modeling data set based on the normal operating data of the chiller;

[0138] A model building module is configured to use a tensor analysis method to extract dynamic features between batches and a slow feature analysis method to extract dynamic features within the same batch for the three-dimensional modeling data set, thereby building a tensor slow feature analysis model;

[0139] A fault detection module is configured to extract tensor slow features from the working condition data to be tested using a tensor slow feature analysis model, construct monitoring statistics based on the tensor slow features and determine control limits to perform fault detection on the chiller;

[0140] The classifier construction module is configured to extract tensor slow features from fault condition data and historical fault data respectively, assign different distribution weights and misclassification costs to the tensor slow features of each type of historical fault data, iteratively train the data gravity classifier according to the distribution weights, and update the weights of the correctly classified samples and misclassified samples of the data gravity classifier according to the misclassification costs, so as to construct a cost-sensitive data gravity classifier;

[0141] The fault identification module is configured to obtain the fault category of the tensor slow feature of the fault condition data according to the cost-sensitive data gravity classifier.

[0142] It should be noted that the above modules correspond to the steps described in Example 1, and the examples and application scenarios implemented by the above modules and the corresponding steps are the same, but are not limited to the contents disclosed in the above Example 1. It should be noted that the above modules, as part of the system, can be executed in a computer system such as a set of computer executable instructions.

[0143] In further embodiments, there is also provided:

[0144] An electronic device includes a memory and a processor, and computer instructions stored in the memory and executed on the processor, wherein when the computer instructions are executed by the processor, the method described in Embodiment 1 is performed. For the sake of brevity, it will not be described in detail here.

[0145] It should be understood that in this embodiment, the processor may be a central processing unit CPU, and the processor may also be other general-purpose processors, digital signal processors DSP, application-specific integrated circuits ASIC, off-the-shelf programmable gate arrays FPGA or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0146] The memory may include a read-only memory and a random access memory, and provide instructions and data to the processor. A portion of the memory may also include a non-volatile random access memory. For example, the memory may also store information about the device type.

[0147] A computer-readable storage medium is used to store computer instructions, and when the computer instructions are executed by a processor, the method described in Example 1 is completed.

[0148] The method in Example 1 can be directly embodied as a hardware processor, or a combination of hardware and software modules in the processor. The software module can be located in a mature storage medium in the field such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory, or an electrically erasable programmable memory, a register, etc. The storage medium is located in the memory, and the processor reads the information in the memory and completes the steps of the above method in combination with its hardware. To avoid repetition, it is not described in detail here.

[0149] Those skilled in the art will appreciate that the units, i.e., algorithm steps, of the various examples described in conjunction with this embodiment can be implemented in electronic hardware or in a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0150] 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. A chiller fault detection and diagnosis method, characterized in that: include: Obtain the normal operating data of the chiller and use it to construct a 3D modeling data set; The tensor analysis method is used to extract dynamic features between batches of the 3D modeling dataset, and the slow feature analysis method is used to extract dynamic features within the same batch, so as to construct a tensor slow feature analysis model. The tensor slow feature analysis model is used to extract the tensor slow features from the working condition data to be tested, and the monitoring statistics are constructed based on the tensor slow features and the control limits are determined to detect the faults of the chiller. Tensor slow features are extracted from fault condition data and historical fault data respectively, and different distribution weights and misclassification costs are assigned to the tensor slow features of each type of historical fault data. The data gravity classifier is iteratively trained according to the distribution weights, and the weights of the correctly classified samples and misclassified samples of the data gravity classifier are updated according to the misclassification costs, so as to construct a cost-sensitive data gravity classifier. Set the tensor slow feature information Y of each type of historical fault data Fi ,i=1,2,…,the misclassification cost of the samples in C, and set different weights for different types of training samples according to the misclassification cost; according to the weight distribution of the samples, the data gravity classifier is trained by an iterative method, and the weights of the different types of samples that are currently correctly classified are reduced according to the misclassification cost, and the weights of the misclassified samples are increased, so that the next round focuses on those samples with large weights, and finally the tensor slow feature information y of the fault data is treated by weighted voting. i The classification results are used to make the final decision; The specific algorithm of the cost-sensitive data gravity classifier is as follows: given the training data {y F (i),γ i },i=1,2,…,n F , where γ i =1,2,…,C is the category label of the data, and different initial distribution weights are set for each type of historical fault data In order to improve the classification accuracy of minority class samples, a larger misclassification cost is imposed on minority class samples and a smaller misclassification cost is imposed on majority class samples; Y Fi Sample y i The misclassification cost c(y i ) is defined as: in: Among them, F i ,F j is the partial F value of the slow feature of the i-th and j-th tensors; K represents the number of sampling points in a batch; In the data gravity classifier, the distribution weight of each sample is used as its quality, rather than simply considering the quality of the sample as 1; dataset Y Fi Y Fj The data gravity of the rth sample point in the tth iteration is defined as: Calculate the C-type fault data for the test data y Fj Gravity After that, the test data y Fj The category of is consistent with the category of the fault dataset that exerts the greatest attraction on it: At the tth iteration, the data gravity classifier h t The weight α t Calculated as: Among them, c(y i ) is the sample y i The misclassification cost, D t (i) is the tth iteration sample y i The weight of In the tth iteration, sample y i The weight D(i) is updated as: Among them, γ i Represents sample y i The category label, Z t Represents the normalized coefficient of the weight D(i) update after the tth iteration; The fault category of the tensor slow feature of the fault condition data is obtained according to the cost-sensitive data gravity classifier.

2. A chiller fault detection and diagnosis method as claimed in claim 1, characterized in that: The three-dimensional modeling data set is preprocessed, including: dividing the three-dimensional modeling data set into K time slice matrices according to the sampling time; arranging the time slice matrix into a two-dimensional matrix according to the batch direction; at each sampling time, the two-dimensional matrix is ​​normalized to zero mean and unit variance along each variable, the processed two-dimensional data matrix is ​​divided into K time slice matrices again, and the time slice matrix is ​​rearranged into a three-dimensional matrix.

3. A chiller fault detection and diagnosis method as claimed in claim 1, characterized in that: The objective function of the tensor slow feature analysis model is: Among them, U and V are load matrices, X i is the data in the i-th batch, and I is the number of batches in the 3D modeling dataset; The load matrices U and V are obtained by performing tensor decomposition on the three-dimensional modeling data set, and the tensor slow features within the same batch of chillers are extracted based on the load matrices U and V.

4. A chiller fault detection and diagnosis method as claimed in claim 3, characterized in that: Tensor slow feature Y i Expressed as: Y i =U T X i V.

5. A chiller fault detection and diagnosis system, characterized in that: include: A data set construction module is configured to obtain normal operating data of the chiller and construct a three-dimensional modeling data set based on the normal operating data of the chiller; A model building module is configured to use a tensor analysis method to extract dynamic features between batches and a slow feature analysis method to extract dynamic features within the same batch for the three-dimensional modeling data set, thereby building a tensor slow feature analysis model; A fault detection module is configured to extract tensor slow features from the working condition data to be tested using a tensor slow feature analysis model, construct monitoring statistics based on the tensor slow features and determine control limits to perform fault detection on the chiller; The classifier construction module is configured to extract tensor slow features from fault condition data and historical fault data respectively, assign different distribution weights and misclassification costs to the tensor slow features of each type of historical fault data, iteratively train the data gravity classifier according to the distribution weights, and update the weights of the correctly classified samples and misclassified samples of the data gravity classifier according to the misclassification costs, so as to construct a cost-sensitive data gravity classifier; Set the tensor slow feature information Y of each type of historical fault data Fi ,i=1,2,…,the misclassification cost of the samples in C, and set different weights for different types of training samples according to the misclassification cost; according to the weight distribution of the samples, the data gravity classifier is trained by an iterative method, and the weights of the different types of samples that are currently correctly classified are reduced according to the misclassification cost, and the weights of the misclassified samples are increased, so that the next round focuses on those samples with large weights, and finally the tensor slow feature information y of the fault data is treated by weighted voting. i The classification results are used to make the final decision; The specific algorithm of the cost-sensitive data gravity classifier is as follows: given the training data {y F (i),γ i },i=1,2,…,n F , where γ i =1,2,…,C is the category label of the data, and different initial distribution weights are set for each type of historical fault data In order to improve the classification accuracy of minority class samples, a larger misclassification cost is imposed on minority class samples and a smaller misclassification cost is imposed on majority class samples; The misclassification cost c(y i ) is defined as: in: Among them, F i ,F j is the partial F value of the slow feature of the i-th and j-th tensors; K represents the number of sampling points in a batch; In the data gravity classifier, the distribution weight of each sample is used as its quality, rather than simply considering the quality of the sample as 1; dataset Y Fi Y Fj The data gravity of the rth sample point in the tth iteration is defined as: Calculate the C-type fault data for the test data y Fj Gravity After that, the test data y Fj The category of is consistent with the category of the fault dataset that exerts the greatest attraction on it: At the tth iteration, the data gravity classifier h t The weight α t Calculated as: Among them, c(y i ) is the sample y i The misclassification cost, D t (i) is the tth iteration sample y i The weight of In the tth iteration, sample y i The weight D(i) is updated as: Among them, γ i Represents sample y i The category label, Z t Represents the normalized coefficient of the weight D(i) update after the tth iteration; The fault identification module is configured to obtain the fault category of the tensor slow feature of the fault condition data according to the cost-sensitive data gravity classifier.

6. An electronic device, characterized in that: The method comprises a memory and a processor and computer instructions stored in the memory and executed on the processor, wherein when the computer instructions are executed by the processor, the method according to any one of claims 1 to 4 is completed.

7. A computer-readable storage medium, characterized in that: Used to store computer instructions, which, when executed by a processor, complete the method described in any one of claims 1 to 4.

Images