A method and system for diagnosing faults of water-cooled chillers
Through the classification method based on dynamic center of mass, the problems of low accuracy and slow speed in the prior art are solved, and efficient fault detection is achieved, especially accurate diagnosis in nonlinear situations.
Patent Information
- Application Number
- CN202210539172.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-05-18
- Publication Date
- 2024-11-26
- Estimated Expiration
- 2042-05-18
AI Technical Summary
The prior art has problems of low accuracy and slow speed in the fault diagnosis of water-cooled chiller units. Especially when dealing with nonlinear problems, it is difficult to effectively detect soft faults, resulting in reduced system performance and increased energy consumption.
The classification method based on dynamic centroid is adopted, and feature selection and dimensionality reduction transformation are performed by obtaining historical data and measured data, and the data is mapped to the feature space using projection functions. Combined with Manhattan distance calculation fault categories, the accuracy and speed of fault diagnosis are improved.
It improves the accuracy and speed of fault diagnosis of water-cooled chiller units, can effectively deal with nonlinear problems, reduces the calculation cost of feature decomposition, improves model training speed, and enhances the flexibility and efficiency of diagnosis.
Smart Images

Figure CN115081513B_ABST
Abstract
Description
Technical Field
[0001] The invention relates to a water-cooled chiller fault diagnosis method and system, belonging to the technical field of water-cooled chiller fault diagnosis in air-conditioning systems. Background Art
[0002] Public buildings account for more than 40% of global energy consumption, and HVAC systems are the main energy-consuming equipment in buildings. By diagnosing HVAC system faults and optimizing the operation of air-conditioning systems, building energy consumption can be reduced by 20%-30% and air-conditioning system energy consumption by 10%-40%. Early automatic fault detection and diagnosis can reduce equipment downtime, energy consumption and service costs. Therefore, improving the reliability of HVAC system equipment is crucial to ensuring the sustainability of buildings and reducing global carbon emissions.
[0003] Water-cooled chillers are widely used cold and heat source systems in air-conditioning systems. Faults in water-cooled chillers can be divided into hard faults and soft faults. Once a hard fault occurs, it will cause serious economic losses and cause the main engine to shut down, which is easy to identify. Soft faults are also called performance faults. When the fault is relatively minor, the system or equipment performance declines less and is generally difficult to detect. Only when the fault reaches a more serious level and the system or equipment performance declines more, it is easy to detect. Therefore, soft faults are slow to progress and difficult to detect, and their damage to the system and equipment and their impact on their performance are long-term. In a sense, soft faults are more harmful. The present invention focuses on the performance fault detection and diagnosis of water-cooled chillers.
[0004] Since modern industrial processes are equipped with a large number of sensors, measuring instruments and controllers, data-driven technology using process variables has been widely studied in recent years and has become increasingly popular. Common methods currently used in the field of fault diagnosis include principal component analysis (PCA), support vector machine (SVM), linear discriminant analysis (FDA), etc. These methods have certain limitations, such as low accuracy when dealing with nonlinear problems and slow model training speed when the data scale reaches a certain level. Summary of the invention
[0005] The purpose of the present invention is to overcome the deficiencies in the prior art and provide a water-cooled chiller fault diagnosis method and system, which can improve the water-cooled chiller fault diagnosis accuracy and diagnosis speed. To achieve the above purpose, the present invention is implemented by adopting the following technical solutions:
[0006] In a first aspect, the present invention provides a method for diagnosing a fault of a water-cooled chiller, comprising:
[0007] Obtain historical and measured data from on-site water-cooled chiller systems;
[0008] Perform feature selection on the acquired historical data and measured data to obtain training data sets and data sets to be classified;
[0009] Perform dimensionality reduction transformation on the training data set to obtain the projection function of the training data set in the feature space and the feature data set;
[0010] Use the projection function to perform dimensionality reduction transformation on the data set to be classified, and obtain the coordinates of the data to be classified in the feature space in the data set to be classified;
[0011] Based on the obtained coordinates, a dynamic centroid-based classification method is used to match the fault categories in the feature data set to obtain the fault categories measured in the field water-cooled chiller system.
[0012] In combination with the first aspect, further, the acquisition of historical data and measured data of the on-site water-cooled chiller system includes: data directly collected by sensors installed in the on-site water-cooled chiller system and data calculated based on the directly collected data.
[0013] In combination with the first aspect, optionally, the sensor includes a temperature sensor, a pressure sensor, a flow sensor and a power sensor.
[0014] In combination with the first aspect, further, the feature selection is: selecting parameters for fault diagnosis, and the feature selection principle is:
[0015] Parameters that change significantly when a fault occurs, parameters that change stably when a fault occurs, and parameters that have low acquisition or calculation costs.
[0016] In combination with the first aspect, further, the expression of the dimensionality reduction transformation is:
[0017] x→z=Θ T K(:,X) (1)
[0018] In formula (1), the training data set is an N×n matrix X, which is composed of N historical data, X = (X1, X2, ..., X N ) T ; Formula (1) reduces the dimension of the n-dimensional matrix X to the (c-1)-dimensional feature space z, where x represents the matrix of the matrix X in the n-dimensional feature space, and c represents the number of fault categories in the training data set;
[0019] Transformation matrix Θ = [α1, α2, ..., α c-1 ] is a N×(c-1) matrix;
[0020]
[0021] In combination with the first aspect, further, the projection function is expressed by the following formula:
[0022]
[0023] In formula (2), w is the projection vector, which realizes the mapping of the matrix X of the training data set to the matrix X′ of the feature data set. i is the eigenvector, and K is the kernel function.
[0024] In combination with the first aspect, further, the classification method based on dynamic centroid includes:
[0025] Based on the feature data set, the centroid of each type of fault in the feature space is calculated to determine the direction of each fault type cluster;
[0026] Based on the coordinates of the data to be classified in the feature space, the dynamic centroid of each type of fault in the feature space is determined;
[0027] Calculate the Manhattan distance between the coordinates of the data to be classified in the feature space and the dynamic centroid;
[0028] The fault category represented by the dynamic centroid closest to the Manhattan distance is the fault category of the data to be classified.
[0029] In combination with the first aspect, further, determining the dynamic centroid of each type of fault in the feature space includes:
[0030] Calculate the Manhattan distance of each data to be classified from the origin ||x||1;
[0031] Based on the calculated Manhattan distance ||x||1, the location coordinates of each fault category cluster are used as the dynamic centroid of the fault category, ||x k ||1=||x||1,||x k ||1 is the Manhattan distance between the samples in the k-th fault cluster and the origin.
[0032] In combination with the first aspect, further, the Manhattan distance between the coordinates of the data to be classified in the feature space and the dynamic centroid is calculated by the following formula:
[0033]
[0034] In formula (3), Y′ i represents the coordinate of the data to be classified i in the feature space, k represents the fault label, k∈(1,c), c represents the number of fault categories; i∈(1,N), N represents the number of data in the matrix X of the training data set;
[0035] y′ i Represents a specific coordinate of the data to be classified i in the feature space, x′ i Indicates that the feature data corresponding to the data to be classified i is relative to y′i Specific coordinates of
[0036] averX′ i k It represents the dynamic centroid of the k-th fault in the feature space and is expressed by the following formula:
[0037] averX′ i k =mean([X′ i ] n×(c-1) ) (4)
[0038] In formula (4), X′ i represents the coordinates of the feature data corresponding to the data to be classified i in the feature space, and n represents the number of the kth type of faults in the matrix X of the training data set.
[0039] In a second aspect, the present invention provides a water-cooled chiller fault diagnosis system, comprising:
[0040] Acquisition module: used to obtain historical data and measured data of the on-site water-cooled chiller system;
[0041] Feature selection module: used to select features from the acquired historical data and measured data to obtain training data sets and data sets to be classified;
[0042] The first dimensionality reduction module is used to perform dimensionality reduction transformation on the training data set to obtain the projection function of the training data set in the feature space and the feature data set;
[0043] The second dimensionality reduction module is used to perform dimensionality reduction transformation on the data set to be classified using a projection function to obtain the coordinates of the data to be classified in the feature space of the data set to be classified;
[0044] Fault diagnosis module: It is used to match the fault category from the feature data set based on the obtained coordinates and adopt the dynamic centroid-based classification method to obtain the fault category actually measured in the on-site water-cooled chiller system.
[0045] In a third aspect, the present invention provides a computer-readable storage medium storing one or more programs, wherein the one or more programs include instructions, which, when executed by a computing device, cause the computing device to perform any of the methods described in the first aspect.
[0046] Compared with the prior art, the water-cooled chiller fault diagnosis method and system provided by the embodiment of the present invention have the following beneficial effects:
[0047] The historical data and measured data of the on-site water-cooled chiller system are obtained; the obtained historical data and measured data are respectively subjected to feature selection to obtain a training data set and a data set to be classified; the training data set is subjected to dimensionality reduction transformation to obtain a projection function and a feature data set of the training data set in the feature space; the projection function is used to perform dimensionality reduction transformation on the to-be-classified data set to obtain the coordinates of the to-be-classified data in the feature space in the to-be-classified data set; based on the obtained coordinates, a classification method based on dynamic centroid is used to match the fault category in the feature data set to obtain the fault category measured in the on-site water-cooled chiller system. The present invention uses LDA to project high-dimensional data into low-dimensional space, and the classification method based on dynamic centroid can effectively handle the situation where the data is linearly inseparable, solve nonlinear problems, and make the diagnosis accuracy higher. At the same time, it can cleverly reduce the computational cost of feature decomposition, and further improve the model training speed under the premise of ensuring classification accuracy. The present invention does not require an in-depth understanding of the physical characteristics of the chiller system, and has higher efficiency and flexibility. BRIEF DESCRIPTION OF THE DRAWINGS
[0048] Figure 1 is a flow chart of a water-cooled chiller fault diagnosis method provided by Embodiment 1 of the present invention;
[0049] Figure 2 is the distribution of 6500 sets of original training data in three-dimensional space in Example 2 of the present invention;
[0050] Figure 3 is the distribution of the centroid of the fault category in the three-dimensional space after the training data in Example 2 of the present invention is transformed by dimensionality reduction;
[0051] Figure 4 It is the distribution of the online data in the three-dimensional space after the dimensionality reduction transformation in Example 2 of the present invention. DETAILED DESCRIPTION
[0052] The present invention will be further described below in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and cannot be used to limit the protection scope of the present invention.
[0053] Embodiment 1:
[0054] This embodiment provides a water-cooled chiller fault diagnosis method, including:
[0055] Obtain historical and measured data from on-site water-cooled chiller systems;
[0056] Perform feature selection on the acquired historical data and measured data to obtain training data sets and data sets to be classified;
[0057] Perform dimensionality reduction transformation on the training data set to obtain the projection function of the training data set in the feature space and the feature data set;
[0058] Use the projection function to perform dimensionality reduction transformation on the data set to be classified, and obtain the coordinates of the data to be classified in the feature space in the data set to be classified;
[0059] Based on the obtained coordinates, a dynamic centroid-based classification method is used to match the fault categories in the feature data set to obtain the fault categories measured in the field water-cooled chiller system.
[0060] The specific steps are as follows:
[0061] Step 1: Obtain historical and measured data of the on-site water-cooled chiller system.
[0062] Data collected directly by sensors installed in the on-site water-cooled chiller system and data calculated based on the directly collected data.
[0063] Sensors include temperature sensors, pressure sensors, flow sensors and power sensors.
[0064] Step 2: Perform feature selection on the acquired historical data and measured data to obtain the training data set and the data set to be classified.
[0065] Select parameters for fault diagnosis from the data obtained in step 1. The principle of feature selection is: parameters that change significantly when a fault occurs, parameters that change stably when a fault occurs, and parameters with low acquisition or calculation costs.
[0066] Step 3: Perform dimensionality reduction transformation on the training data set to obtain the projection function of the training data set in the feature space and the feature data set.
[0067] The training data set contains N data, where N is a natural number greater than 0, c is the number of fault categories in the training set, and each data represents the operating status of the unit, which is a labeled n-dimensional vector k is X i The label of k∈(1,c), n is the number of diagnostic indicators used for fault detection after feature selection, and p is the fault diagnosis indicator. The training data set consisting of N historical data is an N×n matrix X, X=(X1,X2…X N ) T , each row vector in X represents a raw data representing the operating status of the unit.
[0068] The expression of dimensionality reduction transformation is:
[0069] x→z=Θ T K(:,X) (1)
[0070] Formula (1) reduces the dimension of the n-dimensional matrix X to the (c-1)-dimensional feature space z, where x represents the matrix of the matrix X in the n-dimensional feature space, and the transformation matrix Θ = [α1, α2, …, α c-1 ] is a N×(c-1) matrix;
[0071] The projection function is expressed as follows:
[0072]
[0073] In formula (2), w is the projection vector, which realizes the mapping of the matrix X of the training data set to the matrix X′ of the feature data set. i is the eigenvector, and K is the kernel function.
[0074] The feature data set is the original n-dimensional training data that is converted into a (c-1)-dimensional vector X′ in the feature space after dimensionality reduction. i , X′ i =(x′1,x′2,…x′ c-1 ) T The transformed data constitutes a feature data set of N×(c-1) matrix X′, X′=(X′1,X′2,…X′ N ) T .
[0075] Step 4: Use the projection function to reduce the dimension of the data set to be classified and obtain the coordinate Y′ of the data to be classified in the feature space. i , Y′ i =(y′1,y′2,…y′ c-1 ) T .
[0076] Step 5: Based on the obtained coordinates, a dynamic centroid-based classification method is used to match the fault categories in the feature data set to obtain the fault categories measured in the on-site water-cooled chiller system.
[0077] Classification methods based on dynamic centroid include:
[0078] Step 5.1: Based on the feature data set, calculate the centroid of each type of fault in the feature space and determine the direction of each fault category cluster.
[0079] The centroid of each category is expressed as averk = mean ([X′ i ] n×(c-1) ), where k is X i The label of k∈(1,c), n is the number of faults of the kth type in the training set. The centroid determines the direction of the cluster in the feature space.
[0080] Step 5.2: Based on the coordinates of the data to be classified in the feature space, determine the dynamic centroid averX′ of each type of fault in the feature space i k , indicating that the k-th fault is related to Y′ i The corresponding centroid. When the sample to be classified changes, the dynamic centroid of the category changes accordingly.
[0081] Step 5.2.1: Calculate the Manhattan distance ||x||1 between each data to be classified and the origin.
[0082] Step 5.2.2: Based on the calculated Manhattan distance ||x||1, the location coordinates of each fault category cluster are used as the dynamic centroid of the fault category, ||x k ||1=||x||1,||x k ||1 is the Manhattan distance between the samples in the k-th fault cluster and the origin.
[0083] Step 5.3: Calculate the Manhattan distance between the coordinates of the data to be classified in the feature space and the dynamic centroid.
[0084] averX′ i k It represents the dynamic centroid of the k-th fault in the feature space and is expressed by the following formula:
[0085] averX′ i k =mean([X′ i ] n×(c-1) ) (3)
[0086] In formula (3), X′ i represents the coordinates of the feature data corresponding to the data to be classified i in the feature space, and n represents the number of the kth type of faults in the matrix X of the training data set.
[0087] Calculate the Manhattan distance between the coordinates of the data to be classified in the feature space and the dynamic centroid using the following formula:
[0088]
[0089] In formula (4), Y′ i represents the coordinate of the data to be classified i in the feature space, k represents the fault label, k∈(1,c), c represents the number of fault categories; i∈(1,N), N represents the number of data in the matrix X of the training data set; y′ i Represents a specific coordinate of the data to be classified i in the feature space, x′ i Indicates that the feature data corresponding to the data to be classified i is relative to y′ i The specific coordinates of .
[0090] Step 5.4: The fault category represented by the dynamic centroid closest to the Manhattan distance is the fault category of the data to be classified.
[0091] The present invention uses LDA to project high-dimensional data into low-dimensional space. The classification method based on dynamic centroid can effectively handle the situation where data is linearly inseparable, solve nonlinear problems, and make the diagnosis more accurate. At the same time, it can cleverly reduce the computational cost of feature decomposition, and further improve the model training speed while ensuring classification accuracy. The present invention does not require an in-depth understanding of the physical characteristics of the chiller system, and has higher efficiency and flexibility.
[0092] Embodiment 2:
[0093] This embodiment uses the ASHRAERP-1043 data set to verify the method described in the first embodiment.
[0094] ASHRAERP-1043 contains normal and fault data for a 90 ton (316kW) centrifugal chiller. Each fault mode is monitored at four severity levels, 10%, 20%, 30%, and 40%. The test data includes steady-state operating conditions and transient data. A total of approximately 864 minutes of testing was performed for each test under 27 operating conditions by changing the supplied cooling water outlet temperature, cooling water inlet temperature, and unit cooling load. Each operating condition lasted approximately 30 minutes to maintain a steady state.
[0095] Step 1: In this embodiment, six common chiller faults are selected from ASHARERP-1043, including condenser scaling, non-condensable gas in refrigerant, condenser water flow reduction, evaporator water flow reduction, refrigerant leakage, and refrigerant overcharge.
[0096] Step 2: In each test, 64 measured / calculated values were recorded. Appropriate feature selection is very effective in removing irrelevant variables and improving model performance. By analyzing the physical characteristics of each fault, features closely related to the fault can be obtained. Considering the stability and significance of parameter changes when each fault occurs, 11 features were finally selected to detect and diagnose 6 common faults, including direct measurements and calculations, as shown in Table 1. Figure 2 The figure shows the distribution of 6500 sets of original training data in three-dimensional space. It can be seen that the faults of each category are completely overlapped and mixed.
[0097] Table 1 Feature selection results
[0098]
[0099]
[0100] Step 3: After preprocessing the training data, the method described in Example 1 (SRKDA method) is used to perform dimensionality reduction processing on 6500 groups of samples. The specific process of dimensionality reduction can be described by the following three sub-steps.
[0101] Step 3.1: Generate orthogonal vectors:
[0102]
[0103] where y0 = [1, 1, ..., 1] T is a matrix with all elements equal to 1. Let y0 be the first eigenvector of W, and use the Gram-Schmidt method to orthogonalize the other eigenvectors. After removing the vector y0, the remaining c-1 eigenvectors of W can be expressed as:
[0104]
[0105] Step 3.2: Add an element of 1 to the end of each input data to get c-1 vectors in is the solution to the following regularized least squares problem:
[0106]
[0107] in is the eigenvector The i-th element of is the RKHS (Reproducing kernel Hilbert space) associated with the Mercer kernel, |||| K is the corresponding norm in RKHS.
[0108] Step 3.3: Order It is a (M+1)×(c-1) matrix. The dimensionality reduction of sample features is achieved by mapping it to a (c-1) dimensional subspace, namely:
[0109]
[0110] Step 4: Calculate the centroid of the training data set in the feature space after dimensionality reduction. Figure 3 The distribution of the six fault categories in the feature space after the training data is transformed by dimensionality reduction (only three dimensions are shown).
[0111] Step 5: Use 2700 test samples as online data. After dimensionality reduction, they are distributed in the feature space as follows: Figure 4 As shown in Figure 2, each type of fault is a cluster starting from the origin, and the centroid of the fault category calculated in the offline stage represents the direction of the category cluster.
[0112] Step 6: This embodiment compares the proposed method with FDA and SVM, and compares the diagnostic accuracy and model training time of the three methods for different numbers of offline samples.
[0113] 2000, 3500, 5000, and 6500 sets of dynamic data were selected from the brief data set of ASHARE-RP1043 as training sets. 0.7*0.3 samples of the training set were extracted from the remaining data as test sets. The results are shown in Table 2.
[0114] Table 2 Effect of the number of training samples on the performance of fault diagnosis model
[0115]
[0116] The performance of the model was evaluated from two aspects: classification accuracy and training time. As the number of training samples increased, both the classification accuracy and training time increased. FDA had the shortest training time and the lowest accuracy. The method described in Example 1 (SRKDA method) achieved better performance than FDA, which shows the effectiveness of the kernel method. The method described in Example 1 (SRKDA method) is better than SVM. On each training set, the difference in their misclassification rates was less than 5%. The disadvantage of the traditional KDA method is that the computational cost is very high when the number of samples is very large, and our experiments show that the method described in Example 1 (SRKDA method) and SVM have the same training speed and are better than SVM.
[0117] Embodiment three:
[0118] An embodiment of the present invention provides a water-cooled chiller fault diagnosis system, comprising:
[0119] Acquisition module: used to obtain historical data and measured data of the on-site water-cooled chiller system;
[0120] Feature selection module: used to select features from the acquired historical data and measured data to obtain training data sets and data sets to be classified;
[0121] The first dimensionality reduction module is used to perform dimensionality reduction transformation on the training data set to obtain the projection function of the training data set in the feature space and the feature data set;
[0122] The second dimensionality reduction module is used to perform dimensionality reduction transformation on the data set to be classified using a projection function to obtain the coordinates of the data to be classified in the feature space of the data set to be classified;
[0123] Fault diagnosis module: It is used to match the fault category from the feature data set based on the obtained coordinates and adopt the dynamic centroid-based classification method to obtain the fault category actually measured in the on-site water-cooled chiller system.
[0124] Embodiment 4:
[0125] An embodiment of the present invention provides a computer-readable storage medium storing one or more programs, characterized in that the one or more programs include instructions, which, when executed by a computing device, enable the computing device to execute a method in the method described in embodiment one.
[0126] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.
[0127] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.
[0128] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.
[0129] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.
[0130] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the technical principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.
Claims
1. A method for diagnosing faults of a water-cooled chiller. It is characterized in that include: Obtain historical and measured data from on-site water-cooled chiller systems; Perform feature selection on the acquired historical data and measured data to obtain training data sets and data sets to be classified; Perform dimensionality reduction transformation on the training data set to obtain the projection function of the training data set in the feature space and the feature data set; Use the projection function to perform dimensionality reduction transformation on the data set to be classified, and obtain the coordinates of the data to be classified in the feature space in the data set to be classified; Among them, the expression of the dimensionality reduction transformation is: x→z=Θ T K(:,X) (1) In formula (1), the training data set is an N×n matrix X, which is composed of N historical data. 1 ,X 2 ,…,X N ) T ; Formula (1) reduces the dimension of the n-dimensional matrix X to the (c-1)-dimensional feature space z, where x represents the matrix of the matrix X in the n-dimensional feature space, and c represents the number of fault categories in the training data set; Transformation matrix Θ = [α 1 ,α 2 ,…,α c-1 ] is a N×(c-1) matrix; Wherein, the projection function is expressed by the following formula: In formula (2), w is the projection vector, which realizes the mapping of the matrix X of the training data set to the matrix X′ of the feature data set. i is the eigenvector, K is the kernel function; Based on the obtained coordinates, a dynamic centroid-based classification method is used to match the fault categories in the feature data set to obtain the fault categories measured in the field water-cooled chiller system; The classification method based on dynamic centroid includes: Based on the feature data set, the centroid of each type of fault in the feature space is calculated to determine the direction of each fault type cluster; Based on the coordinates of the data to be classified in the feature space, the dynamic centroid of each type of fault in the feature space is determined; wherein the determining the dynamic centroid of each type of fault in the feature space includes: Calculate the Manhattan distance ‖x‖ of each data to be classified from the origin 1 ; Based on the calculated Manhattan distance ‖x‖ 1 , the position coordinates of each fault category cluster are taken as the dynamic centroid of the fault category, ‖x k ‖ 1 =‖x‖ 1 , ‖x k ‖ 1 is the Manhattan distance between the sample in the k-th fault cluster and the origin; Calculate the Manhattan distance between the coordinates of the data to be classified in the feature space and the dynamic centroid; wherein the Manhattan distance between the coordinates of the data to be classified in the feature space and the dynamic centroid is calculated by the following formula: In formula (3), Y′ i represents the coordinate of the data to be classified i in the feature space, k represents the fault label, k∈(1,c), c represents the number of fault categories; i∈(1,N), N represents the number of data in the matrix X of the training data set; y′ i Represents a specific coordinate of the data to be classified i in the feature space, x′ i Indicates that the feature data corresponding to the data to be classified i is relative to y′ i Specific coordinates of It represents the dynamic centroid of the k-th fault in the feature space and is expressed by the following formula: In formula (4), X′ i represents the coordinates of the feature data corresponding to the data to be classified i in the feature space, and n represents the number of the k-th type of faults in the matrix X of the training data set; The fault category represented by the dynamic centroid closest to the Manhattan distance is the fault category of the data to be classified.
2. The water-cooled chiller fault diagnosis method according to claim 1, It is characterized in that The historical data and measured data of the on-site water-cooled chiller system are obtained, including: data directly collected by sensors installed in the on-site water-cooled chiller system and data calculated based on the directly collected data.
3. The water-cooled chiller fault diagnosis method according to claim 1, It is characterized in that The feature selection is to select parameters for fault diagnosis. The feature selection principle is: Parameters that change significantly when a fault occurs, parameters that change stably when a fault occurs, and parameters that have low acquisition or calculation costs.
4. A water-cooled chiller fault diagnosis system based on the water-cooled chiller fault diagnosis method according to claim 1, It is characterized in that include: Acquisition module: used to obtain historical data and measured data of the on-site water-cooled chiller system; Feature selection module: used to select features from the acquired historical data and measured data to obtain training data sets and data sets to be classified; The first dimension reduction module is used to perform dimension reduction transformation on the training data set to obtain the projection function of the training data set in the feature space and the feature data set; The second dimensionality reduction module is used to perform dimensionality reduction transformation on the data set to be classified using a projection function to obtain the coordinates of the data to be classified in the feature space of the data set to be classified; Fault diagnosis module: It is used to match the fault category from the feature data set based on the obtained coordinates and adopt the dynamic centroid-based classification method to obtain the fault category actually measured in the on-site water-cooled chiller system.
5. A computer-readable storage medium storing one or more programs, It is characterized in that The one or more programs include instructions which, when executed by a computing device, cause the computing device to perform any one of the methods according to claims 1 to 3.
