A method, system, terminal and computer-readable storage medium for determining online faults of a chiller

Abstract

The present application discloses a method, system, terminal and computer-readable storage medium for determining online faults of a chiller, the method comprising: obtaining multiple feature groups of the chiller, and obtaining the working condition sub-cluster data corresponding to each of the multiple feature groups according to the multiple feature groups; constructing a first autoencoder model corresponding to each working condition sub-cluster data, and based on each first autoencoder model, obtaining the reconstruction error result corresponding to each working condition sub-cluster data according to each corresponding working condition sub-cluster data; constructing a second autoencoder model corresponding to each fault type of the chiller, inputting each reconstruction error result into the second autoencoder model, and obtaining the fault result, wherein the fault result includes faults of known fault types and unknown fault types. The present application can realize the rapid detection and fault diagnosis of the operating status of the chiller across equipment and working conditions, and has the ability to identify unknown faults, thereby improving the accuracy of the fault detection and diagnosis of the chiller.

Description

Technical Field

[0001] The present application relates to the field of chiller detection technology, and in particular to a chiller online fault determination method, system, terminal and computer-readable storage medium. Background Art

[0002] Chillers are typical energy-consuming equipment in buildings that are large consumers of energy. Their energy consumption is considerable. If they operate in an unhealthy state, it may lead to a large amount of energy waste. Therefore, it is very necessary to perform fault detection and diagnosis on chillers.

[0003] In recent years, people have paid more and more attention to data-driven fault detection and diagnosis methods, and downplayed traditional model-based methods. Data-driven methods are efficient, convenient and accurate. Their rise is inseparable from the reduction in the cost of IoT sensors and the popularization of building energy management systems, which provide a basis for obtaining a large amount of real-time operation data of central air-conditioning systems. However, model-based methods have problems such as modeling difficulties, high dependence on expert knowledge, and poor generalization performance between devices. In current data-driven methods, people often use multiple classifiers to directly establish fault detection and diagnosis models based on partial experimental or simulation data. Using a single model to simultaneously complete the detection and diagnosis tasks of the chiller may have problems such as low diagnostic accuracy, insufficient robustness and generalization performance, and inability to diagnose unknown faults.

[0004] Therefore, the prior art still needs to be improved and developed. Summary of the invention

[0005] The main purpose of the present application is to provide a method, system, terminal and computer-readable storage medium for determining online faults of a chiller, aiming to solve the problem of low accuracy of fault detection and diagnosis of chillers in the prior art.

[0006] A first aspect of an embodiment of the present application provides a method for determining an online fault of a chiller, and the method for determining an online fault of a chiller comprises the following steps: obtaining multiple feature groups of the chiller, and obtaining operating condition subcluster data corresponding to each of the multiple feature groups based on the multiple feature groups; constructing a first autoencoder model corresponding to each of the operating condition subcluster data, and based on each of the first autoencoder models, obtaining a reconstruction error result corresponding to each of the operating condition subcluster data based on the corresponding operating condition subcluster data; constructing a second autoencoder model corresponding to each fault type of the chiller, and inputting each of the reconstruction error results into the multiple second autoencoder models to obtain a fault result, wherein the fault result includes faults of known fault types and unknown fault types.

[0007] Optionally, in one embodiment of the present application, the acquiring of multiple feature groups of the chiller and, based on the multiple feature groups, obtaining operating condition sub-cluster data corresponding to each of the multiple feature groups specifically includes: acquiring online operation data of the chiller and processing the online operation data to obtain a data set; dividing the data set to obtain multiple feature groups of the chiller; and performing operating condition clustering on the multiple feature groups using a clustering method to obtain operating condition sub-cluster data corresponding to each of the multiple feature groups.

[0008] Optionally, in one embodiment of the present application, the first autoencoder model corresponding to each of the working condition subcluster data is constructed, and based on the first autoencoder model, a reconstruction error result corresponding to each of the working condition subcluster data is obtained according to each corresponding working condition subcluster data, specifically including: for each of the working condition subcluster data, a first autoencoder model corresponding to the working condition subcluster data is constructed, and a first detection threshold of the first autoencoder model is obtained; a working condition subcluster data is input into the corresponding first autoencoder model to obtain a first reconstruction error matrix; and a reconstruction error result corresponding to the working condition subcluster data is obtained according to the first reconstruction error matrix and the first detection threshold.

[0009] Optionally, in one embodiment of the present application, constructing a second autoencoder model corresponding to each fault type of the chiller specifically includes: obtaining the total number of fault categories of the chiller, and constructing a corresponding plurality of second initial autoencoder models based on the total number of fault categories; for each of the second initial autoencoder models, obtaining input features corresponding to the second initial autoencoder model, and determining target parameters of the second initial autoencoder model; training the second initial autoencoder model according to the input features and the target parameters to obtain a second autoencoder model.

[0010] Optionally, in one embodiment of the present application, the target parameters include a target number of early stopping rounds, a target learning rate and a target number of hidden layer nodes; determining the target parameters of the second initial autoencoder model specifically includes: determining the initialization parameters of the second initial autoencoder model, wherein the initialization parameters include the number of early stopping rounds, the learning rate and the number of hidden layer nodes to be optimized; optimizing the number of early stopping rounds, the learning rate and the number of hidden layer nodes based on the Bayesian optimization method to obtain the target number of early stopping rounds, the target learning rate and the target number of hidden layer nodes.

[0011] Optionally, in one embodiment of the present application, each of the reconstruction error results is input into multiple second autoencoder models to obtain a fault result, specifically including: inputting the reconstruction error results corresponding to each of the operating condition subcluster data into multiple second autoencoder models respectively to obtain corresponding multiple second reconstruction error matrices; processing multiple second reconstruction error matrices to obtain a summary matrix; and obtaining the fault result of the chiller according to the summary matrix.

[0012] Optionally, in one embodiment of the present application, obtaining the fault result of the chiller according to the summary matrix specifically includes: performing weighted processing on the summary matrix to obtain a weighted summary matrix; averaging the weighted summary matrix column by column to obtain an average relative total reconstruction error corresponding to each fault category; if any element in the average relative total reconstruction error is greater than a preset threshold, then the fault result is determined to be an unknown fault of an unknown fault type; if at least one element in the average relative total reconstruction error is less than or equal to the preset threshold, then the fault result is determined to be a known fault of a known fault type.

[0013] A second aspect of the embodiment of the present application further provides a chiller online fault determination system, wherein the chiller online fault determination system comprises:

[0014] A data processing module, used for acquiring a plurality of feature groups of the chiller, and obtaining, based on the plurality of feature groups, operating condition sub-cluster data corresponding to each of the plurality of feature groups;

[0015] an abnormality detection confirmation module, used to construct a first autoencoder model corresponding to each of the operating condition sub-cluster data, and based on each of the first autoencoder models, obtain a reconstruction error result corresponding to each of the operating condition sub-cluster data according to each of the corresponding operating condition sub-cluster data;

[0016] A fault type determination module is used to construct a second autoencoder model corresponding to each fault type of the chiller, input each of the reconstruction error results into multiple second autoencoder models, and obtain a fault result, wherein the fault result includes faults of known fault types and unknown fault types.

[0017] The third aspect of an embodiment of the present application also provides a terminal, wherein the terminal includes: a memory, a processor, and a chiller online fault determination program stored in the memory and executable on the processor, wherein the chiller online fault determination program implements the steps of the chiller online fault determination method as described above when executed by the processor.

[0018] The fourth aspect of the embodiment of the present application also provides a computer-readable storage medium, wherein the computer-readable storage medium stores a chiller online fault determination program, and when the chiller online fault determination program is executed by a processor, the steps of the chiller online fault determination method as described above are implemented.

[0019] Beneficial effects: The present application provides a method, system, terminal and computer-readable storage medium for determining online faults of a chiller. The method establishes an independent autoencoder model for typical faults, and obtains the fault result by comparing the output of the autoencoder model corresponding to different faults with the threshold ratio during testing. The method can realize rapid detection and fault diagnosis of the operating status of the chiller across equipment and working conditions, and has the ability to identify unknown faults, thereby improving the accuracy, efficiency and robustness of the chiller fault detection and diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0021] Figure 1 It is a flow chart of a preferred embodiment of the method for determining online faults of a chiller unit of the present application;

[0022] Figure 2 It is a technical roadmap for data-driven chiller fault detection and fault diagnosis in a preferred embodiment of the chiller online fault determination method of the present application;

[0023] Figure 3 It is a detailed flow chart of the fault diagnosis steps based on the autoencoder cluster in the preferred embodiment of the online fault determination method of the chiller unit of the present application;

[0024] Figure 4 It is a technical roadmap for parameter optimization in a preferred embodiment of the method for determining online faults of chillers of the present application;

[0025] Figure 5 This is a schematic diagram of the self-encoder cluster diagnosis decision in a preferred embodiment of the chiller online fault determination method of the present application;

[0026] Figure 6 It is a structural diagram of a preferred embodiment of the online fault determination system for chillers of the present application;

[0027] Figure 7 This is a structural diagram of a preferred embodiment of the terminal of this application.

[0028] Description of reference numerals:

[0029] 100, data processing module; 200, abnormality detection and confirmation module; 300, fault type determination module. DETAILED DESCRIPTION

[0030] In order to make the purpose, technical scheme and effect of the present application clearer and more specific, the technical scheme in the embodiments of the present application will be described clearly and completely in combination with the drawings in the embodiments of the present application. The described embodiments are only possible technical implementations of the present application, not all possible implementations. Based on the embodiments in the present application, those skilled in the art can completely combine the embodiments of the present application to obtain other embodiments without creative work, and these embodiments are also within the scope of protection of the present application.

[0031] In related technologies, unhealthy operation of chillers may lead to a large amount of hidden energy waste, so fault detection and diagnosis are necessary. Current data-driven fault detection and diagnosis methods may have problems such as insufficient diagnostic accuracy, insufficient robustness and generalization performance, and low computational efficiency. If hyperparameter-sensitive tree-based machine learning methods are used, the unavoidable hyperparameter optimization greatly reduces the computational efficiency.

[0032] First, the nouns involved in the embodiments of the present application are introduced:

[0033] LightGBM (Light Gradient Boosting Machine): Lightweight gradient boosting machine is an efficient and scalable machine learning algorithm based on gradient boosting decision tree. It optimizes the performance of traditional algorithms, has faster training speed and lower memory consumption, and is suitable for processing large-scale data and complex features.

[0034] Shapley: Chinese usually does not have a direct translation, but Shapley value has a corresponding explanation in Chinese. In machine learning, Shapley value is used to evaluate feature importance, that is, to evaluate the contribution of each feature to the model prediction results.

[0035] Nstopping: The number of early stopping rounds (or the number of stopping rounds). In the training process of machine learning models, Nstopping is a parameter used to control the early stopping mechanism. The early stopping mechanism is a method to prevent model overfitting. It monitors the performance of the model on the validation set and stops the training process when the performance on the validation set does not improve within Nstopping rounds of training. This can prevent the model from overfitting on the training set, thereby improving the generalization ability of the model.

[0036] The following describes the chiller online fault determination method, system, terminal and computer-readable storage medium of the embodiment of the present application with reference to the accompanying drawings. In view of the problem of low accuracy of chiller fault detection and diagnosis in the above-mentioned related technologies, the present application provides a chiller online fault determination method, in which a typical fault independent autoencoder model is established, and the fault result is obtained by comparing the output of the autoencoder model corresponding to different faults with the threshold ratio during testing. It can realize the rapid detection and fault diagnosis of the operating status of the chiller across equipment and working conditions, and has the ability to identify unknown faults, thereby improving the accuracy, efficiency and robustness of chiller fault detection and diagnosis. As a result, the technical problem of low accuracy of chiller fault detection and diagnosis in the related technologies is solved.

[0037] In this application, the online operation data of the chiller is first preprocessed, and then clustered based on typical characteristic working conditions and an autoencoder model is established. The test outputs the operating status of the chiller and the reconstruction error results. For the chiller detected as abnormal, the reconstruction error is further input into the fault diagnosis module to complete the judgment of the fault type. In the training phase of the diagnosis module, the important input features corresponding to the typical faults are first screened based on the Shapley value, and then the partial hyperparameters of the autoencoder are optimized through the Bayesian optimization algorithm, and finally an independent autoencoder model of the typical fault is constructed for standby. In the test phase, the data is passed into each encoder model, and the reconstruction error is calculated and weighted integrated to realize the diagnosis of typical faults and unknown faults. While this technology realizes efficient fault detection of the chiller, it also establishes an independent autoencoder model for typical faults to form a cluster to solve the fault diagnosis problem. This diagnostic method takes the reconstruction error as input, which is an incremental form of the original data, and is expected to improve the diagnostic generalization ability. The computational efficiency of the autoencoder model is greatly improved under the GPU accelerated deep learning framework. Due to the low data pressure of the single model, the diagnostic accuracy and robustness are improved, and the diagnosis of unknown faults is also realized.

[0038] The technical solution of the present application is described in detail with specific embodiments below. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0039] The method for determining an online fault of a chiller described in the preferred embodiment of the present application is as follows: Figure 1 As shown, the chiller online fault determination method comprises the following steps:

[0040] In step S101, a plurality of feature groups of the chiller are acquired, and based on the plurality of feature groups, operating condition sub-cluster data corresponding to each of the plurality of feature groups are obtained.

[0041] It is worth noting that see Figure 2, the embodiment of the present application includes two parts: fault detection (i.e., abnormality detection) and fault diagnosis (i.e., abnormality diagnosis). The fault detection part completes the judgment of whether the chiller has a fault, and the fault diagnosis part further judges whether the data of the chiller detected by the fault detection part belongs to one of the predefined faults or an undefined new fault. The fault detection part includes three sub-modules, namely, data preprocessing, typical feature group working condition clustering, and autoencoder model establishment and detection; the fault diagnosis includes four sub-modules, namely, data preparation, feature group optimization, parameter optimization, and diagnostic technology based on autoencoder clusters, so as to determine the fault result of the chiller through fault detection and fault diagnosis, that is, which type of fault it belongs to.

[0042] In one possible implementation, in the fault detection step, online operation data of the chiller is obtained, and the online operation data is processed to obtain a data set; the data set is divided to obtain multiple feature groups of the chiller; and a clustering method is used to cluster the operating conditions of the multiple feature groups to obtain operating condition sub-cluster data corresponding to each of the multiple feature groups.

[0043] Specifically, in the anomaly detection process, data preprocessing is first performed. The original operating data of the chiller that can be captured is deduplicated, outliers are processed, missing values ​​are processed, data smoothing, feature calculation, feature extraction and other operations are performed to convert the original data into a data set that can be used by the algorithm. Furthermore, data deduplication is performed to remove the data that is read repeatedly in the original data, and only the earliest data is retained every 10 minutes, and the timestamp is standardized; outlier processing is performed to remove serious outliers using rules for each column of data. Missing value processing, a continuous segment missing retrieval and processing technology is designed for important variables including cooling water inlet temperature, cooling water outlet temperature, chilled water inlet temperature, chilled water outlet temperature and other variables. First, the missing ratio of each parameter in the pre-determined important parameters is checked one by one. When the missing ratio of important parameters exceeds 80%, the machine is marked as a "severely missing important parameter" machine and no subsequent operations are performed. When the missing ratio of no important parameters exceeds 80%, the important parameter column with the largest missing ratio is selected, and a certain length is set as the threshold. When it is greater than the threshold, the parameter is considered to be continuously missing. For example, if the threshold is set to 3, data segments with 4 or more consecutive missing items are automatically screened out as continuous missing segments, and all features of the entire segment of the continuous missing segment data are deleted. For the remaining data segments with continuous missing items less than or equal to 3, the nearest neighbor filling is performed based on the principle that the operating parameters of the cold machine are not prone to drastic changes in a short period of time. For the remaining parameters, nearest neighbor filling and reverse filling are used. If the same strategy is used for all parameters, it may lead to large data loss. Data smoothing, use weighted moving average to smooth the data, create a sliding window of a certain length, assign different weights to the data in the window, and the data close to the current moment has a greater weight. The data points in the window are multiplied by the weights and then added. Feature calculation, use the original parameter information that the chiller can obtain to calculate and derive important features such as: refrigerant suction superheat, refrigerant exhaust superheat, refrigerant subcooling, condenser heat transfer approximation, evaporator heat transfer approximation, system flow coefficient, etc. Feature extraction, in addition to the above data smoothing, which belongs to the category of feature extraction, dimensionality reduction methods can also be used to reduce the dimensionality of some dimensional data to check the data distribution characteristics.

[0044] Specifically, in the anomaly detection process, typical feature group working condition clustering is performed next. In this embodiment, based on domain knowledge, the complex operating parameters in the chiller are divided according to certain rules to form multiple low-dimensional feature groups, and then multiple feature groups are used to cluster working conditions. On the one hand, the input dimension of the clustering algorithm is reduced, making the working condition clustering of the distance-based clustering algorithm more effective and reducing the impact of the dimensionality curse; on the other hand, some selection criteria make the working condition clustering more targeted and meaningful, and improve the interpretability of the working condition clustering.

[0045] Furthermore, in the process of selecting feature groups, there are many operating parameters for chillers. From the perspective of physical types of parameters, they can be divided into electrical parameters, temperature parameters, pressure parameters, flow parameters and frequency parameters. They can also be divided from the perspective of medium types, namely water-related features, refrigerant-related features and chiller operation features. For example, pressure parameters include refrigerant saturated evaporation pressure, refrigerant saturated condensation pressure, intermediate pressure, cooling water inlet pressure, chilled water inlet pressure, compressor lubricating oil pressure, etc. By selecting the first several features with greater correlation (within 10 to reduce the impact of dimensionality disaster) for certain faults or certain components, several feature groups are formulated. In the process of operating condition clustering, the operating condition clustering is completed for the selected feature groups using clustering algorithms, and several operating condition sub-clusters corresponding to each feature group are formed. Under the premise that different feature groups provide information, the operating condition clustering results may be quite different, which makes some data may not belong to the same clustering cluster, realizing the "softening" of the operating condition boundary division, and minimizing the negative impact of the "hard structure" of the data forced by the operating condition clustering on the subsequent establishment of the autoencoder model application.

[0046] It can be understood that data preprocessing is to convert the original operating data into a data set that can be used by the algorithm. The purpose of typical feature group working condition clustering is to reduce the input dimension of the clustering algorithm and improve the effectiveness of working condition clustering. First, the feature group is selected based on domain knowledge, and then the clustering algorithm is used to cluster the feature group working conditions.

[0047] In step S102, a first autoencoder model corresponding to each of the operating condition sub-cluster data is constructed, and based on each of the first autoencoder models, a reconstruction error result corresponding to each of the operating condition sub-cluster data is obtained according to each of the corresponding operating condition sub-cluster data.

[0048] In one possible implementation, in the fault detection step, for each of the operating condition subcluster data, a first autoencoder model corresponding to the operating condition subcluster data is constructed, and a first detection threshold of the first autoencoder model is obtained; the operating condition subcluster data is input into the corresponding first autoencoder model to obtain a first reconstruction error matrix; and based on the first reconstruction error matrix and the first detection threshold, a reconstruction error result corresponding to the operating condition subcluster data is obtained.

[0049] Specifically, in the anomaly detection process, the established autoencoder model (ie, the first autoencoder model) is finally used for detection.

[0050] It should be noted that the autoencoder model in fault detection is the first autoencoder model, and the autoencoder model in fault diagnosis is the second autoencoder model. In other words, both are autoencoder models, and their functions are to learn the characteristics of data, but the functions they ultimately implement are different. The autoencoders all output reconstruction error results. The autoencoder for detection distinguishes whether it is a fault or normal, and the autoencoder for diagnosis distinguishes whether it is a certain type of fault. The autoencoder for fault diagnosis is equivalent to each outputting a similarity information (relative total reconstruction error) and then summing them up to take the most similar fault. If they are all very dissimilar, it is an unknown fault. The reconstruction error result output by fault detection is the input of diagnostic detection.

[0051] Furthermore, in the first autoencoder model training phase, the sub-cluster data of each working condition of the chiller is taken out, and an independent autoencoder model is established for each working condition sub-cluster data. The training data is re-transmitted into the autoencoder model to obtain the reconstructed training data, and the reconstructed training data is subtracted from the original training data to obtain the reconstruction error matrix. The square sum of each row in the reconstruction error matrix is ​​calculated to obtain the total reconstruction error sequence, and the 95% quantile of the total reconstruction error sequence is taken as the detection threshold (the first detection threshold), and the model object is stored for standby. In the first autoencoder model testing phase, the data of the test machine divided into each working condition sub-cluster is taken out; the data is transmitted to the corresponding autoencoder model trained in the training phase for testing, and integrated to form a reconstruction error matrix; the total reconstruction error is calculated, and the 95% quantile of the total reconstruction error sequence is taken as the test score, which is compared with the training threshold. If it is greater than the threshold, the test machine is abnormal; if it is less than or equal to the threshold, the machine is in normal state.

[0052] It should be noted that the present application can summarize a large number of parameters of the chiller according to certain standards, realize the enhanced characterization of a certain component or a certain fault type of the chiller, reduce the input feature dimension of the clustering algorithm and the autoencoder model, and reduce the impact of the dimensionality curse, thereby improving the clustering performance of the clustering algorithm working conditions and the data information extraction performance of the autoencoder model, and thus improving the accuracy of fault detection.

[0053] In step S103, a second autoencoder model corresponding to each fault type of the chiller is constructed, and each of the reconstructed error results is input into multiple second autoencoder models to obtain a fault result, wherein the fault result includes faults of known fault types and unknown fault types.

[0054] It can be understood that in this application, in the preparation stage, the autoencoder model is initialized and relevant parameters are set; the model is trained, and various types of fault data are used to train the autoencoder model; the error benchmark is calculated, and the error benchmark value of each fault type is calculated through the training data; the test data is passed into the model to calculate the reconstruction error; the reconstruction information is processed to form a total reconstruction error sequence and a summary matrix; the weight is optimized, and the weight is set according to the degree of data abnormality and imbalance; the relative error is summarized, and the average relative total reconstruction error of each fault category is calculated; the result is output, and the fault type or unknown fault is judged according to the average relative total reconstruction error. Through the above steps, accurate diagnosis of machine faults can be achieved, and unknown faults can be identified.

[0055] In one possible implementation, in the fault diagnosis step, the total number of fault categories of the chiller is obtained, and a corresponding plurality of second initial autoencoder models are constructed according to the total number of fault categories; for each of the second initial autoencoder models, the input features corresponding to the second initial autoencoder model are obtained, and the target parameters of the second initial autoencoder model are determined; the second initial autoencoder model is trained according to the input features and the target parameters to obtain a second autoencoder model.

[0056] It should be noted that in the abnormal diagnosis step, data preparation is first performed. The reconstructed error results output by the abnormal detection step are divided into abnormal segment data with reference to the information provided by the daily abnormal detection results. If it is a training stage, the data segments are labeled after the data is divided and enter the training process. If it is a test data, the reconstructed error data is not labeled after the data is divided and input into the test step.

[0057] For details, see Figure 3 In the diagnosis step based on the autoencoder cluster, the training phase: autoencoder model initialization: according to the total number of fault categories, initialize the same number of autoencoder models, use the optimized features in as the input features of the autoencoder model here, and initialize the number of input layer nodes at the same time; use the optimal number of early stopping rounds, learning rate, and number of hidden layer nodes confirmed in as the initialization parameters of each encoder model; autoencoder model training, each type of fault data is passed into the corresponding autoencoder model to complete model training; error benchmark calculation, all training data are re-transmitted into the corresponding autoencoder model, assuming that there are N types of faults, N reconstruction matrices with the same shape as the training data will be obtained, and the reconstruction matrix is ​​subtracted from the original matrix to obtain the reconstruction error matrix, and the square sum of each row of data in the reconstruction matrix is ​​obtained to obtain the total reconstruction error of each data, and the total reconstruction error is averaged, which is the error benchmark value corresponding to the fault type. .

[0058] Furthermore, in the process of feature group optimization, a complete data set with labels for each fault type is constructed; a LightGBM binary classification model is defined for each fault type; each fault binary classification model is trained with fault samples as the positive class and normal samples as the negative class; Shapley value calculation, the Shapley value is calculated separately under each fault type binary classification model to obtain a Shapley value matrix with the same length and dimension as the training data, the absolute value of the matrix as a whole is taken to obtain the contribution of each dimensional feature of the sample point to the model prediction of the positive class, and the average of each column of the matrix is ​​further calculated to obtain the average contribution of each dimensional feature to the model prediction of the positive class under the support of the complete training sample, which can characterize the importance of the feature; feature importance sorting, the average Shapley value calculated for each model, that is, the average contribution, is arranged in descending order, and according to the M value entered by the user (M is an arbitrary number representing the number of features the user wants to select), the first M features in descending order are taken as the input features of each model (the second autoencoder model).

[0059] In a possible implementation, the target parameters include a target number of early stopping rounds, a target learning rate, and a target number of hidden layer nodes. In the step of determining the target parameters, the initialization parameters of the two initial autoencoder models are determined, wherein the initialization parameters include the number of early stopping rounds, the learning rate, and the number of hidden layer nodes to be optimized; the number of early stopping rounds, the learning rate, and the number of hidden layer nodes are optimized based on the Bayesian optimization method to obtain the target number of early stopping rounds, the target learning rate, and the target number of hidden layer nodes.

[0060] For details, see Figure 4 In the process of parameter optimization, the parameters to be optimized are defined. After evaluation, the parameters that have a greater impact on the autoencoder are the learning rate, the number of hidden layer nodes, and the number of early stopping rounds Nstopping in the early stopping mechanism set to minimize the negative impact of overfitting on the generalization performance of the model (the training of the autoencoder model is stopped if the error on the validation set does not decrease for consecutive Nstopping rounds); parameter optimization, based on the Bayesian optimization algorithm, the three parameters of the autoencoder model hidden layer node number, early stopping round number and learning rate are optimized. The default number of optimization rounds is 100 rounds. The optimization aims to minimize the average relative total reconstruction error of K-fold cross validation. The indicators are defined as follows:

[0061] Taking the training of a fault autoencoder model as an example, the total fault data set is not randomly shuffled and divided into K parts according to the time series. K-1 parts are taken as the training set and the validation set, and the other part is taken as the test set. The ratio of the training set to the validation set is 4:1. After the current model is trained using the training set and the validation set, the training set and the validation set are re-transmitted for testing to obtain the reconstructed data. The training set and the reconstructed data are subtracted to obtain the reconstruction error matrix. The square sum of the data row by row in the reconstruction error matrix is ​​calculated to obtain the total reconstruction error sequence. The mean of the total reconstruction error sequence is used as the error metric on the training set. Repeat the above error metric calculation operation on the test set to obtain the error metric on the test set ,by is a single-fold validation indicator After completing K-fold cross validation, Sum and get the total The K-fold cross validation average relative total error is then divided by K. This application hopes that this value is as small as possible, indicating that the model has learned the characteristics of the fault data well, has no serious overfitting, and has good generalization ability. Record the optimal hyperparameters, and record the optimal number of hidden layer nodes, number of early stopping rounds, and learning rate found in the dictionary for model establishment.

[0062] In one possible implementation, in the fault diagnosis step, the reconstruction error results corresponding to each of the operating condition subcluster data are respectively input into multiple second autoencoder models to obtain corresponding multiple second reconstruction error matrices; the multiple second reconstruction error matrices are processed to obtain a summary matrix; and based on the summary matrix, the fault result of the chiller is obtained.

[0063] Specifically, in data testing, the machine data to be tested is passed into each diagnostic autoencoder model to obtain a reconstruction matrix; the reconstruction error matrix between the original matrix and the reconstructed matrix is ​​calculated. Reconstruction information processing, the square sum of the samples in each reconstruction error matrix is ​​calculated to form a total reconstruction error sequence; the elements in the sequence represent the total reconstruction error size of the test sample under the corresponding autoencoder. Reconstruction information integration, the total reconstruction error sequence of each fault type is divided by the corresponding error reference value to form a relative value; splicing to form a new summary matrix, which represents the similarity between each data of the machine to be tested and each fault type. Diagnosis result output, weight optimization, set weights according to the degree of data abnormality, give higher weights to data with a large degree of abnormality, set weights according to the imbalance of fault data, and balance the summary diagnostic performance; relative total reconstruction error summary, average the summary matrix column by column, and obtain the average relative total reconstruction error under each fault category; judgment, if all elements are greater than the unknown fault threshold, it is considered to be an unknown fault, otherwise, the fault type corresponding to the minimum relative total reconstruction error is selected as the diagnosis result.

[0064] In one possible implementation, in the step of determining the fault result, the summary matrix is ​​weighted to obtain a weighted summary matrix; the weighted summary matrix is ​​averaged column by column to obtain an average relative total reconstruction error corresponding to each fault category; if any element in the average relative total reconstruction error is greater than a preset threshold, the fault result is determined to be an unknown fault of an unknown fault type; if at least one element in the average relative total reconstruction error is less than or equal to the preset threshold, the fault result is determined to be a known fault of a known fault type.

[0065] For details, see Figure 5 In the diagnostic step based on the autoencoder cluster, the test phase: data testing, the machine data to be tested (i.e., the reconstruction error result outputted from step 102) is inputted into each diagnostic autoencoder model (the second autoencoder model). Assuming that there are N types of faults, the machine data to be tested There are m rows and n columns in total, so we get N reconstruction matrices with m rows and n columns. ( Represents the reconstruction matrix of the first fault type, represents the reconstruction matrix of the Nth fault type), using the original matrix Subtract the reconstruction matrix Then we get the reconstruction error matrix ( represents the reconstruction error matrix of the first fault type until represents the reconstruction error matrix of the Nth fault type). Reconstruction information processing, converting the N reconstruction error matrices The square sum of each sample in is calculated to form a total reconstruction error sequence with the same length as the number of reconstruction error rows ( represents the total reconstruction error sequence for the first fault type, represents the total reconstruction error sequence for the second fault type, Represents the total reconstruction error sequence of the Nth fault type). An element in this sequence represents the total reconstruction error size of a test sample under an autoencoder test. The larger the value, the less similar it is to the fault type. Reconstruction information integration, the N total reconstruction error sequences corresponding to N faults First divide the whole by the error benchmark ( represents the error benchmark for the first fault type, Represents the error benchmark of the Nth fault type), forming the relative value of the total reconstruction error. The purpose of forming the relative value is to make the errors of the autoencoder model under different faults comparable, and then splicing to form a new summary matrix (The superscript T indicates the transposition operation, and the summary matrix is ​​formed by transposing the N total reconstruction error sequences). This matrix represents the similarity between each data of the machine under test and each fault type. The larger the value, the less similar the data is to the corresponding fault type. Finally, the diagnosis result is output, and the summary matrix In the method, the relative values ​​of the total reconstruction errors under the autoencoder model of all faults of all data are stored. Although this method can take the category with the smallest relative value of the total reconstruction error as the diagnostic prediction result of the sample, form a diagnostic label for the sample, and then output the machine diagnosis result based on the sample label quantity information, after comparison and verification, this method has brought about a large information loss after the secondary conversion of information. At the same time, since the number of samples is an integer, the decision boundary is too hard, and the decision is relatively unreliable. Therefore, the diagnostic result output of the embodiment of the present application uses the following relative total reconstruction error summary method:

[0066] Specifically, first, weight optimization, first consider the weight setting of the data abnormality degree, calculate a weight according to the total reconstruction error in the fault detection result, and the weight range is between [1, 2], and the summary matrix Each piece of data is weighted, and a higher weight is given to data with a greater degree of abnormality, so that the summary result is more reliable; secondly, the weight setting of the imbalanced fault data is considered. The data set of different fault types of the chiller can be demonstrated to be an unbalanced data set from the three aspects of fault probability, fault type and fault duration. The autoencoder corresponding to the fault type with a large amount of data can capture more information about the corresponding fault, and the model performance is relatively good. This feature may cause the fault type with a small number of data samples to have a greater possibility of generating a high total reconstruction error when detecting a new machine; in order to balance the summary diagnosis performance, this application calculates a set of weights based on the different fault data sample sizes, so that the total reconstruction error of the autoencoder corresponding to the fault type that frequently fails is relatively increased, thereby reducing the overall decision bias. Second, the relative total reconstruction error is summarized, and the summary matrix Calculate the average value column by column to obtain the N average relative total reconstruction errors corresponding to N fault categories ( represents the average relative total reconstruction error of the first fault type, represents the average relative total reconstruction error of the second fault type, represents the average relative total reconstruction error of the Nth fault type). This result integrates all the fault operation data information of the test machine and is more meaningful for decision making. Is every element in greater than the unknown fault threshold? If it is true, it means that the test machine is very dissimilar to all existing fault types and is an unknown fault. If it is false, it is considered that the test machine is similar to at least one defined fault type, and the fault type corresponding to the minimum relative total reconstruction error can be selected as the machine diagnosis result.

[0067] It should be noted that the embodiment of the present application is based on the chiller fault diagnosis process of the parameter pre-optimized autoencoder cluster. By establishing an independent autoencoder model of a typical fault, fault diagnosis is achieved by comparing the total reconstruction error and threshold ratio output by the data in the autoencoder models corresponding to different faults during testing. This can improve the diagnostic accuracy and enhance the robustness, and can realize the diagnosis of unknown fault types.

[0068] Specifically, the diagnostic accuracy is high and the robustness is strong: the fully connected autoencoder model can capture a large amount of potential nonlinear information in the data, and the model has strong capabilities. To establish an independent autoencoder for a typical fault type, it only needs to consider the data of a single fault type, with a small total amount of information and low data pressure. Compared with directly using a multi-classification model to complete fault diagnosis and integrating multiple binary classification models to complete multi-classification tasks, not only is the pressure of data imbalance small, but the total amount of data information that a unit model has to deal with is also reduced. The model capability plus the low data pressure can achieve higher diagnostic accuracy and robustness. It can realize the diagnosis of unknown fault types: It can not only realize the fault diagnosis within the scope of known typical faults, but also realize the diagnosis of unknown faults. Taking One-vs-Rest (OvR, a strategy that converts multi-classification problems into multiple binary classification sub-problems, in this strategy, multi-classification problems are decomposed into several binary classification problems) technology as a comparative case, this technology is based on multiple binary classification models, and integrates the output results of binary classification models to complete multi-classification tasks. Assuming that there are N types of typical fault types, OvR will establish N binary classification models. The positive class of each binary classification model is a typical fault, and the negative class is N types of faults except For all faults except the positive typical faults, during the test, if all the binary classification models output a probability close to 0, it cannot be concluded that this is an unknown fault type outside the defined typical faults. Since the negative classes of the binary classification models are all known fault types, it cannot be ruled out that whether the combination of the remaining fault types forms a new data feature, making the new machine biased towards the negative class; however, this is not the case when the autoencoder cluster is used in this application. When the total reconstruction error ratio of the autoencoder models of all typical fault types is seriously large, it can be concluded that the fault test chiller is not similar to any known typical fault type and is a new unknown fault type.

[0069] Next, the chiller online fault determination system proposed according to the embodiment of the present application is described with reference to the accompanying drawings.

[0070] Figure 6 It is a structural diagram of the chiller online fault determination system of an embodiment of the present application.

[0071] like Figure 6 As shown, the chiller online fault determination system includes: a data processing module 100 , an abnormality detection and confirmation module 200 and a fault type determination module 300 .

[0072] Specifically, the data processing module 100 is used to obtain multiple feature groups of the chiller, and obtain the operating condition sub-cluster data corresponding to each of the multiple feature groups according to the multiple feature groups;

[0073] The abnormality detection confirmation module 200 is used to construct a first autoencoder model corresponding to each of the operating condition sub-cluster data, and obtain a reconstruction error result corresponding to each of the operating condition sub-cluster data based on each of the first autoencoder models and according to each of the corresponding operating condition sub-cluster data;

[0074] The fault type determination module 300 is used to construct a second autoencoder model corresponding to each fault type of the chiller, input each of the reconstruction error results into multiple second autoencoder models, and obtain a fault result, wherein the fault result includes faults of known fault types and unknown fault types.

[0075] Figure 7 This is a structural diagram of a terminal provided in an embodiment of the present application. The terminal may include:

[0076] A memory 501 , a processor 502 , and a computer program stored in the memory 501 and executable on the processor 502 .

[0077] When the processor 502 executes the program, the method for determining an online fault of a chiller provided in the above embodiment is implemented.

[0078] Furthermore, the terminal further includes:

[0079] The communication interface 503 is used for communication between the memory 501 and the processor 502 .

[0080] The memory 501 is used to store computer programs that can be executed on the processor 502 .

[0081] The memory 501 may include a high-speed RAM memory, and may also include a non-volatile memory (non-volatile memory), such as at least one disk memory.

[0082] If the memory 501, the processor 502 and the communication interface 503 are implemented independently, the communication interface 503, the memory 501 and the processor 502 can be connected to each other through a bus and communicate with each other. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component (PCI) bus or an Extended Industry Standard Architecture (EIS) bus. The bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 7 Only one thick line is used in the diagram, but this does not mean that there is only one bus or only one type of bus.

[0083] Optionally, in a specific implementation, if the memory 501, the processor 502 and the communication interface 503 are integrated on a chip, the memory 501, the processor 502 and the communication interface 503 can communicate with each other through an internal interface.

[0084] The processor 502 may be a central processing unit (CPU), or an application specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of the present application.

[0085] This embodiment also provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the above-mentioned method for determining an online fault of a chiller is implemented.

[0086] One embodiment of the present application provides a computer program product, including a computer program, which, when executed by a processor, implements the present application Figure 1 A method for determining an online fault of a chiller provided in any of the corresponding embodiments.

[0087] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or N embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, without contradiction.

[0088] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include at least one of the features. In the description of this application, "N" means at least two, such as two, three, etc., unless otherwise clearly and specifically defined.

[0089] Any process or method description in a flowchart or otherwise described herein may be understood to represent a module, fragment or portion of code comprising one or N executable instructions for implementing the steps of a custom logical function or process, and the scope of the preferred embodiments of the present application includes alternative implementations in which functions may not be performed in the order shown or discussed, including performing functions in a substantially simultaneous manner or in reverse order depending on the functions involved, which should be understood by technicians in the technical field to which the embodiments of the present application belong.

[0090] 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 storage 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 combination with these instruction execution systems, devices or apparatuses. For the purposes of this specification, a "computer-readable storage 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 combination with these instruction execution systems, devices or apparatuses. More specific examples (non-exhaustive list) of computer-readable storage media include the following: an electrical connection with one or N wirings (electronic device), a portable computer disk box (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 storage medium may even be paper or other suitable medium on which the program is printed, since the program can be obtained electronically by optically scanning the paper or other medium and then editing, interpreting or processing in other suitable ways as necessary, and then storing it in a computer memory.

[0091] It should be understood that the various parts of the present application can be implemented by hardware, software, firmware or a combination thereof. In the above embodiment, N 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.

[0092] A person skilled in the art may understand that all or part of the steps in the method for implementing the above-mentioned embodiment may be completed by instructing related hardware through a program, and the program may be stored in a computer-readable storage medium, which, when executed, includes one or a combination of the steps of the method embodiment.

[0093] In addition, each functional unit in each embodiment of the present application may be integrated into a processing module, or each unit may exist physically separately, or two or more units may be integrated into one module. The above-mentioned integrated module may be implemented in the form of hardware or in the form of a software functional module. If the integrated module is implemented in the form of a software functional module and sold or used as an independent product, it may also be stored in a computer-readable storage medium.

[0094] The storage medium mentioned above may be a read-only memory, a magnetic disk or an optical disk, etc. Although the embodiments of the present application have been shown and described above, it can be understood that the above embodiments are exemplary and cannot be understood as limiting the present application. A person of ordinary skill in the art may change, modify, replace and modify the above embodiments within the scope of the present application.

[0095] It should be understood that the application of the present application is not limited to the above examples. For ordinary technicians in this field, improvements or changes can be made based on the above description. All these improvements and changes should fall within the scope of protection of the claims attached to this application.

[0096] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit it. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for determining an online fault of a chiller, characterized in that: The chiller online fault determination method comprises: Acquire multiple feature groups of the chiller, and obtain operating condition sub-cluster data corresponding to each of the multiple feature groups according to the multiple feature groups; Constructing first autoencoder models corresponding to each of the operating condition sub-cluster data, and obtaining reconstruction error results corresponding to each of the operating condition sub-cluster data based on the first autoencoder models and according to the corresponding operating condition sub-cluster data; Constructing a second autoencoder model corresponding to each fault type of the chiller, inputting each of the reconstruction error results into a plurality of the second autoencoder models to obtain a fault result, wherein the fault result includes faults of known fault types and unknown fault types; The constructing of the first autoencoder model corresponding to each of the operating condition sub-cluster data, and obtaining the reconstruction error result corresponding to each of the operating condition sub-cluster data based on the first autoencoder model and according to the corresponding operating condition sub-cluster data, specifically includes: For each of the operating condition sub-cluster data, construct a first autoencoder model corresponding to the operating condition sub-cluster data, and obtain a first detection threshold of the first autoencoder model; Inputting each of the operating condition sub-cluster data into the corresponding first autoencoder model to obtain a first reconstruction error matrix; Obtaining a reconstruction error result corresponding to each of the operating condition sub-cluster data according to the first reconstruction error matrix and the first detection threshold; The step of constructing a second autoencoder model corresponding to each fault type of the chiller specifically includes: Obtaining the total number of fault categories of the chiller, and constructing a corresponding plurality of second initial autoencoder models according to the total number of fault categories; For each of the second initial autoencoder models, obtaining input features corresponding to the second initial autoencoder model, and determining target parameters of the second initial autoencoder model; The second initial autoencoder model is trained according to the input features and the target parameters to obtain a second autoencoder model.

2. The method for determining an online fault of a chiller according to claim 1, characterized in that: The acquiring of the plurality of feature groups of the chiller and obtaining, according to the plurality of feature groups, the operating condition sub-cluster data corresponding to each of the plurality of feature groups specifically includes: Acquiring online operation data of the chiller and processing the online operation data to obtain a data set; Dividing the data set to obtain multiple feature groups of the chiller; A clustering method is used to cluster the operating conditions of the plurality of feature groups to obtain operating condition sub-cluster data corresponding to each of the plurality of feature groups.

3. The method for determining an online fault of a chiller according to claim 1, characterized in that: The target parameters include the target number of early stopping rounds, the target learning rate and the target number of hidden layer nodes; The determining of the target parameters of the second initial autoencoder model specifically includes: Determine initialization parameters of the two initial autoencoder models, wherein the initialization parameters include the number of early stopping rounds to be optimized, the learning rate, and the number of hidden layer nodes; The number of early stopping rounds, the learning rate and the number of hidden layer nodes are optimized based on the Bayesian optimization method to obtain the target number of early stopping rounds, the target learning rate and the target number of hidden layer nodes.

4. The method for determining an online fault of a chiller according to claim 1, characterized in that: Inputting each of the reconstruction error results into a plurality of the second autoencoder models to obtain a fault result specifically includes: Inputting the reconstruction error results corresponding to each of the operating condition sub-cluster data into a plurality of the second autoencoder models respectively to obtain a plurality of corresponding second reconstruction error matrices; Processing a plurality of the second reconstruction error matrices to obtain a summary matrix; According to the summary matrix, a fault result of the chiller is obtained.

5. The method for determining an online fault of a chiller according to claim 4, characterized in that: The step of obtaining the fault result of the chiller according to the summary matrix specifically includes: Performing weighted processing on the summary matrix to obtain a weighted summary matrix; The weighted summary matrix is ​​averaged column by column to obtain the average relative total reconstruction error corresponding to each fault category; If any element of the average relative total reconstruction error is greater than a preset threshold, determining that the fault result is an unknown fault of an unknown fault type; If at least one element of the average relative total reconstruction error is less than or equal to a preset threshold, it is determined that the fault result is a known fault among known fault types.

6. A chiller online fault determination system, characterized in that: The chiller online fault determination system comprises: A data processing module, used for acquiring a plurality of feature groups of the chiller, and obtaining, based on the plurality of feature groups, operating condition sub-cluster data corresponding to each of the plurality of feature groups; an abnormality detection confirmation module, used to construct a first autoencoder model corresponding to each of the operating condition sub-cluster data, and based on each of the first autoencoder models, obtain a reconstruction error result corresponding to each of the operating condition sub-cluster data according to each of the corresponding operating condition sub-cluster data; The constructing of the first autoencoder model corresponding to each of the operating condition sub-cluster data, and obtaining the reconstruction error result corresponding to each of the operating condition sub-cluster data based on the first autoencoder model and according to the corresponding operating condition sub-cluster data, specifically includes: For each of the operating condition sub-cluster data, construct a first autoencoder model corresponding to the operating condition sub-cluster data, and obtain a first detection threshold of the first autoencoder model; Inputting each of the operating condition sub-cluster data into the corresponding first autoencoder model to obtain a first reconstruction error matrix; Obtaining a reconstruction error result corresponding to each of the operating condition sub-cluster data according to the first reconstruction error matrix and the first detection threshold; A fault type determination module is used to construct a second autoencoder model corresponding to each fault type of the chiller, input each of the reconstruction error results into a plurality of the second autoencoder models, and obtain a fault result, wherein the fault result includes faults of known fault types and unknown fault types; The step of constructing a second autoencoder model corresponding to each fault type of the chiller specifically includes: Obtaining the total number of fault categories of the chiller, and constructing a corresponding plurality of second initial autoencoder models according to the total number of fault categories; For each of the second initial autoencoder models, obtaining input features corresponding to the second initial autoencoder model, and determining target parameters of the second initial autoencoder model; The second initial autoencoder model is trained according to the input features and the target parameters to obtain a second autoencoder model.

7. A terminal, characterized in that: The terminal includes: a memory, a processor, and a chiller online fault determination program stored in the memory and executable on the processor. When the chiller online fault determination program is executed by the processor, the steps of the chiller online fault determination method as described in any one of claims 1 to 5 are implemented.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a chiller online fault determination program, and when the chiller online fault determination program is executed by the processor, the steps of the chiller online fault determination method according to any one of claims 1 to 5 are implemented.

Images