Fault diagnosis method and device for evaporative cooling system and evaporative cooling system

Abstract

The invention discloses a fault diagnosis method and device for an evaporative cooling system and the evaporative cooling system, and the method comprises the steps: carrying out the decomposition through employing a fixed window dynamic mode, obtaining the modal information and the modal number in different operation states, and carrying out the fault diagnosis through employing a model with the highest classification accuracy in a plurality of machine learning models, the technical problems that in the prior art, when a traditional fault diagnosis technology is used for conducting fault diagnosis on an evaporative cooling system of a wind turbine generator, accuracy is low, adaptability is poor, and safety and reliability are low are solved. The technical effect of performing intelligent fault diagnosis on the evaporative cooling system with high self-adaptability and high diagnosis accuracy is achieved, and the reliability and safety of system operation are improved.

Description

Technical Field

[0001] The embodiments of the present invention relate to the field of detection technology, and in particular to a fault diagnosis method and device for an evaporative cooling system, and an evaporative cooling system. Background Art

[0002] Fault diagnosis technology for traditional air-cooled wind turbines is relatively mature. Mainstream methods include: time-frequency analysis, which identifies gearbox or bearing damage by decomposing the frequency domain characteristics of vibration signals; electromechanical signal monitoring, which detects generator winding short circuits or power device failures based on voltage / current waveform distortion; and data-driven methods, which uses historical SCADA (Supervisory Control and Data Acquisition) data to establish fault prediction models.

[0003] With the increasing complexity of systems and the deployment of more and more sensors, data-driven methods have gradually exposed their limitations in high-dimensional and multivariate processing. Although deep learning models have powerful feature extraction capabilities, their interpretability is poor. Especially in the fault diagnosis task of wind turbines in evaporative cooling systems, it is difficult to accurately reveal the physical meaning of the fault mode. Therefore, a new intelligent diagnosis method with high adaptability and accuracy is urgently needed. Summary of the Invention

[0004] The embodiments of the present invention provide a fault diagnosis method, device and evaporative cooling system for an evaporative cooling system, which solve the technical problems in the prior art of using traditional fault diagnosis technology to perform fault diagnosis on the evaporative cooling system of a wind turbine generator set, such as low accuracy, poor adaptability, low safety and low reliability.

[0005] An embodiment of the present invention provides a fault diagnosis method for an evaporative cooling system, the fault diagnosis method comprising:

[0006] Acquiring operating data of the evaporative cooling system within a set time period at a set sampling frequency, and preprocessing the operating data, wherein the operating data at least includes pressure data, flow data, and temperature data of the evaporative cooling system;

[0007] Based on the preprocessed operating data, fixed-window dynamic mode decomposition is used to determine modal information and modal quantities of the evaporative cooling system under different operating states, and each modal information and modal quantity is respectively formed into a modal data set, wherein one modal data set corresponds to one operating state;

[0008] Based on the modal data set, the classification accuracy of multiple preset machine learning models is evaluated, and the machine learning model with the highest classification accuracy is determined as the target learning model;

[0009] Performing Bayesian hyperparameter tuning on the target learning model to determine the optimal parameter configuration of the target learning model;

[0010] Training the target learning model using the optimal parameter configuration to obtain an optimal target learning model;

[0011] Fault diagnosis is performed on the evaporative cooling system based on the optimal target learning model.

[0012] Furthermore, the fault diagnosis method further includes:

[0013] During the process of diagnosing the fault of the evaporative cooling system, determining in real time whether the classification accuracy of the optimal target learning model during operation is lower than a preset accuracy threshold;

[0014] If it is lower, the algorithm adaptive update is started, the operating data of the evaporative cooling system is reacquired, and the classification accuracy of multiple preset machine learning models is evaluated based on the modal data set until a new optimal target learning model is determined;

[0015] If it is not lower than, continue to use the current optimal target learning model for fault diagnosis.

[0016] Furthermore, based on the pre-processed operating data, fixed-window dynamic mode decomposition is used to determine modal information and modal quantities of the evaporative cooling system under different operating states, and each modal information and modal quantity is respectively formed into a modal data set including:

[0017] Set the window size to w, divide the pre-processed running data based on the window, and for the pth window, the window data is Wherein, the operation data n is the dimension of the system state, Δt is the time step, x k is the state vector at the kth moment, k is the start time of the current window;

[0018] Based on the singular value decomposition of each window data X p , get the corresponding local modal information;

[0019] Integrating the local modal information corresponding to each of the windows to determine the modal information and modal quantity of the evaporative cooling system under different operating states;

[0020] The modal information and the modal quantities under different operating states are respectively formed into modal data sets under different operating states.

[0021] Furthermore, based on the modal data set, the classification accuracy of multiple preset machine learning models is evaluated, and the machine learning model with the highest classification accuracy is determined as the target learning model, including:

[0022] Based on the modal dataset, multiple preset machine learning models are respectively subjected to the following formulas: Evaluate the classification accuracy, where A is the classification accuracy, TP is the number of samples correctly predicted as positive by the model, TN is the number of samples correctly predicted as negative by the model, FP is the number of samples incorrectly predicted as positive by the model, and FN is the number of samples incorrectly predicted as negative by the model;

[0023] Based on the determined classification accuracy of each of the preset machine learning models, the machine learning model with the highest classification accuracy is determined as the target learning model.

[0024] Furthermore, the multiple preset machine learning models include at least support vector machine, decision tree, k-nearest neighbor algorithm, logistic regression classification, multi-layer perceptron, and deep neural network.

[0025] Furthermore, reacquiring the operating data of the evaporative cooling system includes:

[0026] Starting from the time point when the classification accuracy fails to meet the preset accuracy threshold, the operation data of the evaporative cooling system within the set time period is acquired again at the set sampling frequency.

[0027] Furthermore, preprocessing the operating data includes:

[0028] The operation data are normalized in rows and columns based on the time series and sensor series.

[0029] An embodiment of the present invention further provides a fault diagnosis device for an evaporative cooling system, the fault diagnosis device comprising:

[0030] a data acquisition module, configured to acquire operating data of the evaporative cooling system within a set time period at a set sampling frequency, and pre-process the operating data, wherein the operating data includes at least pressure data, flow data, and temperature data of the evaporative cooling system;

[0031] a data decomposition module, configured to determine modal information and modal quantities of the evaporative cooling system under different operating states using fixed-window dynamic mode decomposition based on the preprocessed operating data, and to organize each modal information and modal quantity into a modal data set, wherein one modal data set corresponds to one operating state;

[0032] An accuracy evaluation module is used to evaluate the classification accuracy of multiple preset machine learning models based on the modal data set, and determine the machine learning model with the highest classification accuracy as the target learning model;

[0033] A parameter tuning module, configured to perform Bayesian hyperparameter tuning on the target learning model to determine the optimal parameter configuration of the target learning model;

[0034] A model training module is used to train the target learning model using the optimal parameter configuration to obtain an optimal target learning model;

[0035] A fault diagnosis module is used to perform fault diagnosis on the evaporative cooling system based on the optimal target learning model.

[0036] Furthermore, the fault diagnosis device further includes:

[0037] A model detection module is used to determine in real time whether the classification accuracy of the optimal target learning model during operation is lower than a preset accuracy threshold during the process of fault diagnosis of the evaporative cooling system;

[0038] an algorithm updating module, configured to, if the detection result of the model detection module is that the classification accuracy is lower than a preset accuracy threshold, initiate an algorithm adaptive update, reacquire the operating data of the evaporative cooling system, and evaluate the classification accuracy of multiple preset machine learning models based on the modal data set until a new optimal target learning model is determined;

[0039] If the detection result of the model detection module is that the classification accuracy is not lower than the preset accuracy threshold, the fault diagnosis module continues to use the current optimal target learning model to perform fault diagnosis.

[0040] An embodiment of the present invention further provides an evaporative cooling system, which includes the fault diagnosis device for the evaporative cooling system described in any of the above embodiments.

[0041] An embodiment of the present invention discloses a fault diagnosis method, device and evaporative cooling system for an evaporative cooling system. The method includes: obtaining operating data of the evaporative cooling system within a set time period at a set sampling frequency, and preprocessing the operating data; based on the preprocessed operating data, using fixed-window dynamic mode decomposition to determine the modal information and modal quantity of the evaporative cooling system under different operating states, and respectively forming each modal information and modal quantity into a modal data set; based on the modal data set, evaluating the classification accuracy of multiple preset machine learning models, and determining the machine learning model with the highest classification accuracy as the target learning model; performing Bayesian hyperparameter tuning on the target learning model to determine the optimal parameter configuration of the target learning model; using the optimal parameter configuration to train the target learning model to obtain the optimal target learning model; and performing fault diagnosis on the evaporative cooling system based on the optimal target learning model. The present invention obtains modal information and modal quantity under different operating states by adopting fixed-window dynamic mode decomposition, and uses the model with the highest classification accuracy among multiple machine learning models for fault diagnosis. It solves the technical problems of low accuracy, poor adaptability, low safety and reliability in the prior art when using traditional fault diagnosis technology to diagnose faults in the evaporative cooling system of wind turbines, achieves the technical effect of intelligent fault diagnosis of the evaporative cooling system with high adaptability and high diagnostic accuracy, and improves the reliability and safety of system operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 This is a flow chart of a fault diagnosis method for an evaporative cooling system provided by an embodiment of the present invention;

[0043] Figure 2 This is a comparison chart of the classification accuracy of the six models after training provided by an embodiment of the present invention;

[0044] Figure 3 This is an iterative result diagram of Bayesian optimization of a decision tree provided by an embodiment of the present invention;

[0045] Figure 4 is a flow chart of another fault diagnosis method for an evaporative cooling system provided by an embodiment of the present invention;

[0046] Figure 5 This is a structural diagram of a fault diagnosis device for an evaporative cooling system provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0047] The present invention will be further described in detail below with reference to the accompanying drawings and examples. It will be understood that the specific embodiments described herein are intended only to illustrate the present invention and are not intended to limit the present invention. It should also be noted that, for ease of description, the accompanying drawings only illustrate portions relevant to the present invention, not all structures.

[0048] It should be noted that the terms "first," "second," and so on, in the specification, claims, and drawings of the present invention are used to distinguish different objects, and are not intended to limit a specific order. The following embodiments of the present invention can be implemented independently or in combination with each other, and the present invention does not impose specific limitations on this.

[0049] Figure 1 This is a flow chart of a fault diagnosis method for an evaporative cooling system provided by an embodiment of the present invention.

[0050] like Figure 1 As shown, the fault diagnosis method of the evaporative cooling system specifically includes the following steps:

[0051] S101 , acquiring operating data of the evaporative cooling system within a set time period at a set sampling frequency, and preprocessing the operating data, wherein the operating data at least includes pressure data, flow data, and temperature data of the evaporative cooling system.

[0052] Specifically, the sampling frequency and duration can be set as needed. In an embodiment of the present invention, operating data is preferably collected at no less than 3,000 time intervals Δt. The operating data includes at least pressure data, flow data, and temperature data of the evaporative cooling system, specifically including variables such as condenser water supply flow rate, inlet and outlet flow rates and pressures, coolant filter pressure, stator inlet flow rate and pressure, and circulating pump outlet pressure. The evaporative cooling system also has 60 stator bars connected in parallel, each with five temperature measurement points. Therefore, a total of 300 temperature data points can be collected. By taking the maximum temperature sensor corresponding to every 10 bars, a total of 34 variables are collected.

[0053] Optionally, preprocessing the operating data includes: performing row and column normalization processing on the operating data based on the time series and the sensor series.

[0054] Specifically, the collected operation data is pre-processed. It is necessary to perform row and column normalization processing on the time series and sensor series of the collected operation data, and normalize them to the range of [0, 1].

[0055] S102: Based on the preprocessed operating data, fixed-window dynamic mode decomposition is used to determine the modal information and modal quantity of the evaporative cooling system under different operating states, and each modal information and modal quantity is respectively organized into a modal data set, where one modal data set corresponds to each operating state.

[0056] Optionally, S102 specifically includes:

[0057] Set the window size to w, divide the preprocessed running data based on the window, for the pth window, the window data is Among them, the operating data n is the dimension of the system state, Δt is the time step, x k is the state vector at the kth moment, k is the start time of the current window; based on the singular value decomposition of each window data X p , and obtain the corresponding local modal information; integrate the local modal information corresponding to each window to determine the modal information and modal quantity of the evaporative cooling system under different operating states; and form the modal information and modal quantity under different operating states into modal data sets under different operating states.

[0058] Specifically, the preprocessed running data is Where n is the dimension of the system state, Δt is the time step, and the data is represented as a matrix where each column is the state vector at a moment. Select a window size w and divide the data according to the window. For the pth window, the window data is represented as For each window data X p , solve the dynamic mode decomposition, the core is to use the singular value decomposition (SVD) data matrix X p , and obtain a set of local modal information; by integrating the dynamic mode decomposition results of multiple windows, the modal information and modal quantity of the system in different operating states can be obtained. For each window, its modal information Φp and modal quantity m are recorded. p , and obtain the final number of modes and modal characteristics.

[0059] Optionally, decompose the data matrix X by singular value decomposition p , specifically including:

[0060] Construct the input matrix X and output matrix Y, where:

[0061] X=[x1,x2,…,x w-1 ],Y=[x2,x3,…,x w ]

[0062] Construct the singular value decomposition, specifically:

[0063]

[0064] Among them, U b and V b are the left singular vector matrix and the right singular vector matrix, Σ b Is a diagonal singular value matrix, set the threshold α to truncate X p The singular values ​​of and singular vectors and V b , reducing data complexity, the dimension after dimensionality reduction is b.

[0065] Based on the dynamic mode decomposition assumed by Koopman's theorem, a linear mapping matrix R is established, which can make the state matrix separated by Δt satisfy the approximate linear mapping relationship:

[0066] Y≈RX

[0067] Using the singular values ​​and singular vectors after dimensionality reduction, the pseudo-inverse operation yields:

[0068]

[0069] To extract modal information (decay rate, frequency, and amplitude), calculate The eigenvalues ​​and eigenvectors of :

[0070]

[0071] Among them, λ is the eigenvalue, and its real part represents the decay rate decay i =Re(ω i ), the imaginary part represents the oscillation angular frequency Frequency is v is the eigenvector, which represents the mode of the system and describes the evolution of the mode in time. The amplitude ai of each dynamic mode is calculated by the following formula:

[0072]

[0073] The dataset constructed in this way includes the modes (amplitude, decay rate and frequency) and the number of modes in each window, corresponding to the operating status under each window.

[0074] S103, based on the modal data set, evaluate the classification accuracy of multiple preset machine learning models, and determine the machine learning model with the highest classification accuracy as the target learning model.

[0075] Optionally, S103 specifically includes:

[0076] Based on the modal dataset, the formulas are used for various preset machine learning models: The classification accuracy is evaluated, where A is the classification accuracy, TP (True Positives) is the number of samples correctly predicted by the model as positive, TN (True Negatives) is the number of samples correctly predicted by the model as negative, FP (False Positives) is the number of samples incorrectly predicted by the model as negative, and FN (False Negatives) is the number of samples incorrectly predicted by the model as positive. Based on the determined classification accuracy of each preset machine learning model, the machine learning model with the highest classification accuracy is determined as the target learning model.

[0077] Optionally, the multiple preset machine learning models include at least support vector machine, decision tree, k-nearest neighbor algorithm, logistic regression classification, multi-layer perceptron, and deep neural network.

[0078] Specifically, when running various preset machine learning models based on the modal dataset, the hyperparameters that affect their performance are listed according to the selected model, and a regular search range is set for each hyperparameter. The rough parameter configurations of various preset machine learning models are as follows:

[0079] (1) Support Vector Machine (SVM). The search range is set to: penalty coefficient [1e-3, 1e3], kernel scale [1e-3, 1e2], and normalization [true, false].

[0080] Preferably, the grass parameter configuration is: kernel function: rbf; kernel scale: 1; penalty coefficient: 1; kernel scale: 1 / number of features.

[0081] (2) Decision Tree (DT). The search range is set as follows: maximum number of splits [2, 100], minimum number of leaf node samples [1, 50], minimum number of parent node samples [2, 100], splitting criteria ['gdi', 'deviance', 'twoing'], whether to prune ['on', 'off'], whether to merge leaf nodes ['on', 'off'], and proxy splitting ['on', 'off'].

[0082] Preferably, the GRASS parameter configuration is: maximum number of splits: 10; minimum number of leaf node samples: 1; minimum number of parent node samples: 100; splitting standard: GDI; whether to prune: on; whether to merge leaf nodes: on; proxy splitting: off.

[0083] (3) K-nearest neighbor algorithm (KNN). The search range is set as: number of neighbors [1, 50], distance metric ['euclidean', 'cityblock', 'chebychev', 'minkowski'], and distance weight ['equal', 'inverse', 'squaredinverse'].

[0084] Preferably, the grass parameter configuration is: number of neighbors: 5; distance metric: Euclidean distance; distance weight: reverse.

[0085] (4) Logistic regression classification (LR). The search range is set to: regularization strength [1e-5, 10], solver ['liblinear', 'lbfgs', 'newton-cg', 'sag', 'saga']. It should be noted that different solvers have different regularization methods. For example, 'sag' has regularization methods ['l1', 'l2', 'elasticnet', 'none'].

[0086] Preferably, the grass parameter configuration is: regularization strength: 0.5; regularization method: L2; solver: lbfgs.

[0087] (5) Multilayer Perceptron (MLP). The search range is set as follows: hidden layer structure [(64, 32, 7), (128, 64, 7), (32, 16, 7)], maximum number of iterations [50, 300], mini-batch size [16, 128], initial learning rate [1e-5, 1e-1], activation function ['sigmoid', 'tanh', 'relu'].

[0088] Preferably, the grass parameter configuration is: fully connected layer: 64-32-7; maximum number of iterations: 50; mini-batch size: 32; initial learning rate: 0.001; activation function: sigmoid.

[0089] (6) Deep neural network (DNN). The search range is set as follows: hidden layer structure [(300, 500, 300), (512, 256, 128), (256, 256, 128)], maximum number of iterations [50, 300], mini-batch size [64, 256], initial learning rate [1e-5, 1e-1], activation function ['relu', 'tanh', 'leakyrelu'].

[0090] Preferably, the grass parameter configuration is: fully connected layers: 300-500-300; maximum number of iterations: 120; mini-batch size: 128; initial learning rate: 0.001; activation function: ReLU.

[0091] For example, Figure 2 As shown, Figure 2 A comparison chart of the classification accuracy of the six models after training is given. Among them, through classification accuracy screening, the decision tree (DT) is used as the target learning model, and its accuracy is 98.63%.

[0092] S104: Perform Bayesian hyperparameter tuning on the target learning model to determine the optimal parameter configuration of the target learning model.

[0093] Specifically, after determining the target learning model, the objective function of Bayesian optimization is selected. That is, the goal is to maximize the accuracy of the model and minimize the loss of the model. The objective function is as follows:

[0094] maximize Accuracy(w,hyperparameter);

[0095] Subject to accuracy ≥ 99%;

[0096] 1≤w≤100;

[0097] Hyperparameters∈Conventional search range;

[0098] After setting the search space and objective function, the Bayesian optimization algorithm is initialized. The optimizer configuration includes a maximum number of iterations of 300, an expected improvement acquisition function, and a Gaussian process surrogate model constructed with the Matérn5 / 2 kernel function. Bayesian optimization continuously updates the probability model by collecting new data, thereby optimizing hyperparameters.

[0099] The optimization process is as follows Figure 3 As shown in the figure, when the number of iterations is 0-50, the objective function value (accuracy) drops rapidly, from the initial value of -0.8 to -0.99431, indicating that Bayesian optimization efficiently explores the potential optimal area in the parameter space through the Gaussian process surrogate model and the expected improvement acquisition function. At the same time, the optimization stop condition has been met so far. The optimization process takes only 4.39s, proving that the fault diagnosis method has significant advantages in time cost for dynamic adjustment operations. In the convergence stage (iterations 50-300), the objective function value tends to be stable, ranging from -0.99431 to -0. The optimization process is close to the global optimal solution. The minimum target value and the estimated optimal target value gradually coincide with each other, verifying the accurate fitting of the surrogate model to the distribution of the objective function. When the number of iterations is 55, the objective function reaches the global optimal value of -0.99695, the corresponding window size is 15, and the total time consumption is 11.69s. The optimization process reaches the optimal value within 55 iterations, accounting for 18.3% of the total number of iterations. This optimization process provides the optimal hyperparameter combination for the decision tree model in the dynamic pattern decomposition feature space, thereby improving the accuracy and generalization ability of the machine learning model.

[0100] S105: Train the target learning model using the optimal parameter configuration to obtain the optimal target learning model.

[0101] Specifically, after the optimization process is completed, the Bayesian optimization algorithm will provide a set of optimal hyperparameter configurations, namely the above-mentioned optimal parameter configurations. This set of configuration parameters can enable the target learning model to exhibit the best performance within a given search space; using the optimal parameter configuration, the target learning model is trained on the complete dataset to obtain the optimal target learning model.

[0102] S106: Perform fault diagnosis on the evaporative cooling system based on the optimal target learning model.

[0103] Specifically, the optimized optimal target learning model is finally used to perform fault diagnosis on the evaporative cooling system, which can realize intelligent fault diagnosis of the evaporative cooling system and improve the reliability and safety of system operation.

[0104] The present invention obtains modal information and modal quantity under different operating states by adopting fixed-window dynamic mode decomposition, and uses the model with the highest classification accuracy among multiple machine learning models for fault diagnosis. It solves the technical problems of low accuracy, poor adaptability, low safety and reliability in the prior art when using traditional fault diagnosis technology to diagnose faults in the evaporative cooling system of wind turbines, achieves the technical effect of intelligent fault diagnosis of the evaporative cooling system with high adaptability and high diagnostic accuracy, and improves the reliability and safety of system operation.

[0105] On the basis of the above technical solutions, Figure 4 This is a flow chart of another method for diagnosing a fault in an evaporative cooling system provided by an embodiment of the invention. Figure 4 As shown, the fault diagnosis method of the evaporative cooling system further includes:

[0106] S401, during the process of diagnosing the fault of the evaporative cooling system, determining in real time whether the classification accuracy of the optimal target learning model during operation is lower than a preset accuracy threshold.

[0107] S402: If it is lower than , then start the algorithm adaptive update, reacquire the operating data of the evaporative cooling system, and perform the action of evaluating the classification accuracy of multiple preset machine learning models based on the modal data set until a new optimal target learning model is determined;

[0108] S403: If it is not lower than, continue to use the current optimal target learning model to perform fault diagnosis.

[0109] Specifically, in order to ensure the accuracy of fault diagnosis, the preset accuracy threshold is usually set at 99%. That is, when the classification accuracy of the optimal target learning model during actual operation is lower than 99%, the algorithm adaptive update is started, and when it is higher than 99%, it continues to run. This ensures that the preset machine learning model will not be updated frequently, nor will it fail to receive timely response due to its degraded fault diagnosis performance, resulting in delayed judgment.

[0110] Optionally, reacquiring the operating data of the evaporative cooling system in S402 includes: starting from the time point when the classification accuracy fails to meet a preset accuracy threshold, reacquiring the operating data of the evaporative cooling system within a set time period at a set sampling frequency.

[0111] Specifically, when reacquiring the operating data of the evaporative cooling system, it is necessary to collect the latest operating data, which is the time node from which the current classification accuracy fails to meet the preset accuracy threshold, and collect the operating data of the evaporative cooling system during no less than 3,000 time intervals Δt. After obtaining the new operating data, steps S101 to S105 are repeated to determine a new optimal target learning model that meets the preset accuracy threshold.

[0112] The embodiment of the present invention also provides a fault diagnosis device for an evaporative cooling system, such as Figure 5 As shown, the fault diagnosis device of the evaporative cooling system specifically includes:

[0113] The data acquisition module 51 is used to acquire the operating data of the evaporative cooling system within a set time period at a set sampling frequency and pre-process the operating data, wherein the operating data at least includes pressure data, flow data, and temperature data of the evaporative cooling system;

[0114] A data decomposition module 52 is configured to determine modal information and modal quantities of the evaporative cooling system under different operating states using fixed-window dynamic mode decomposition based on the preprocessed operating data, and to organize each modal information and modal quantity into modal data sets, wherein one modal data set corresponds to each operating state;

[0115] An accuracy evaluation module 53 is configured to evaluate the classification accuracy of multiple preset machine learning models based on the modal data set, and determine the machine learning model with the highest classification accuracy as the target learning model;

[0116] A parameter tuning module 54 is used to perform Bayesian hyperparameter tuning on the target learning model to determine the optimal parameter configuration of the target learning model;

[0117] The model training module 55 is used to train the target learning model using the optimal parameter configuration to obtain the optimal target learning model;

[0118] The fault diagnosis module 56 is used to perform fault diagnosis on the evaporative cooling system based on the optimal target learning model.

[0119] Optionally, the fault diagnosis device further includes:

[0120] A model detection module is used to determine in real time whether the classification accuracy of the optimal target learning model during operation is lower than a preset accuracy threshold during the process of fault diagnosis of the evaporative cooling system;

[0121] an algorithm update module, configured to initiate an adaptive algorithm update if the classification accuracy of the model detection module is lower than a preset accuracy threshold, reacquire the operating data of the evaporative cooling system, and evaluate the classification accuracy of multiple preset machine learning models based on the modal data set until a new optimal target learning model is determined;

[0122] If the detection result of the model detection module is that the classification accuracy is not lower than the preset accuracy threshold, the fault diagnosis module continues to use the current optimal target learning model for fault diagnosis.

[0123] Optionally, the data decomposition module 52 is specifically configured to:

[0124] Set the window size to w, divide the preprocessed running data based on the window, for the pth window, the window data is Among them, the operating data n is the dimension of the system state, Δt is the time step, x k is the state vector at the kth moment, k is the start time of the current window;

[0125] Based on the singular value decomposition of each window data X p , get the corresponding local modal information;

[0126] The local modal information corresponding to each window is integrated to determine the modal information and modal quantity of the evaporative cooling system under different operating conditions;

[0127] The modal information and modal quantities under different operating states are respectively composed into modal data sets under different operating states.

[0128] Optionally, the accuracy evaluation module 53 is specifically configured to:

[0129] Based on the modal dataset, the formulas are used for various preset machine learning models: The classification accuracy is evaluated, where A is the classification accuracy, TP is the number of samples correctly predicted as positive by the model, TN is the number of samples correctly predicted as negative by the model, FP is the number of samples incorrectly predicted as positive by the model, and FN is the number of samples incorrectly predicted as negative by the model;

[0130] Based on the determined classification accuracy of each preset machine learning model, the machine learning model with the highest classification accuracy is determined as the target learning model.

[0131] Optionally, the algorithm updating module is specifically configured to:

[0132] Starting from the time point when the classification accuracy fails to meet the preset accuracy threshold, the operation data of the evaporative cooling system within the set time period is obtained again at the set sampling frequency.

[0133] Optionally, the data acquisition module 51 is specifically configured to:

[0134] For the operation data, row and column normalization are performed based on time series and sensor series respectively.

[0135] The fault diagnosis device for an evaporative cooling system provided in an embodiment of the present invention has the same technical features as the fault diagnosis method for an evaporative cooling system provided in the above embodiment, and therefore can also solve the same technical problems and achieve the same technical effects.

[0136] An embodiment of the present invention further provides an evaporative cooling system, which includes the fault diagnosis device for the evaporative cooling system in any of the above embodiments.

[0137] The evaporative cooling system provided by the embodiment of the present invention includes the fault diagnosis device of the evaporative cooling system in the above embodiment. Therefore, the evaporative cooling system provided by the embodiment of the present invention also has the beneficial effects described in the above embodiment, which will not be repeated here.

[0138] In the description of the embodiments of the present invention, unless otherwise specified or limited, the terms "mounted," "connected," and "connected" should be understood in a broad sense. For example, they may refer to fixed, detachable, or integral connections; mechanical or electrical connections; direct or indirect connections through an intermediate medium; and internal communication between two components. Those skilled in the art will understand the specific meanings of the above terms in the present invention based on specific circumstances.

[0139] Finally, it should be noted that the above are only preferred embodiments of the present invention and the technical principles employed. Those skilled in the art will understand that the present invention is not limited to the specific embodiments described herein, and that various obvious changes, readjustments, and substitutions can be made by those skilled in the art without departing from the scope of protection of the present invention. Therefore, although the present invention has been described in detail through the above embodiments, the present invention is not limited to the above embodiments and may include many other equivalent embodiments without departing from the scope of the present invention. The scope of the present invention is determined by the scope of the appended claims.

Claims

1. A fault diagnosis method for an evaporative cooling system, characterized in that: The fault diagnosis method comprises: Acquiring operating data of the evaporative cooling system within a set time period at a set sampling frequency, and preprocessing the operating data, wherein the operating data at least includes pressure data, flow data, and temperature data of the evaporative cooling system; Based on the preprocessed operating data, fixed-window dynamic mode decomposition is used to determine modal information and modal quantities of the evaporative cooling system under different operating states, and each modal information and modal quantity is respectively formed into a modal data set, wherein one modal data set corresponds to one operating state; Based on the modal data set, the classification accuracy of multiple preset machine learning models is evaluated, and the machine learning model with the highest classification accuracy is determined as the target learning model; Performing Bayesian hyperparameter tuning on the target learning model to determine the optimal parameter configuration of the target learning model; Training the target learning model using the optimal parameter configuration to obtain an optimal target learning model; Fault diagnosis is performed on the evaporative cooling system based on the optimal target learning model.

2. The fault diagnosis method of the evaporative cooling system according to claim 1, characterized in that: The fault diagnosis method further includes: During the process of diagnosing the fault of the evaporative cooling system, determining in real time whether the classification accuracy of the optimal target learning model during operation is lower than a preset accuracy threshold; If it is lower, the algorithm adaptive update is started, the operating data of the evaporative cooling system is reacquired, and the classification accuracy of multiple preset machine learning models is evaluated based on the modal data set until a new optimal target learning model is determined; If it is not lower than, continue to use the current optimal target learning model for fault diagnosis.

3. The fault diagnosis method of the evaporative cooling system according to claim 1, characterized in that: Based on the pre-processed operating data, fixed-window dynamic mode decomposition is used to determine modal information and modal quantities of the evaporative cooling system under different operating states, and each modal information and modal quantity is respectively formed into a modal data set including: Set the window size to w, divide the pre-processed running data based on the window, and for the pth window, the window data is Wherein, the operation data n is the dimension of the system state, Δt is the time step, x k is the state vector at the kth moment, k is the start time of the current window; Based on the singular value decomposition of each window data X p , get the corresponding local modal information; Integrating the local modal information corresponding to each of the windows to determine the modal information and modal quantity of the evaporative cooling system under different operating states; The modal information and the modal quantities under different operating states are respectively formed into modal data sets under different operating states.

4. The fault diagnosis method of the evaporative cooling system according to claim 1, characterized in that: Based on the modal dataset, the classification accuracy of multiple preset machine learning models is evaluated, and the machine learning model with the highest classification accuracy is determined as the target learning model, including: Based on the modal dataset, multiple preset machine learning models are respectively subjected to the following formulas: Evaluate the classification accuracy, where A is the classification accuracy, TP is the number of samples correctly predicted as positive by the model, TN is the number of samples correctly predicted as negative by the model, FP is the number of samples incorrectly predicted as positive by the model, and FN is the number of samples incorrectly predicted as negative by the model; Based on the determined classification accuracy of each of the preset machine learning models, the machine learning model with the highest classification accuracy is determined as the target learning model.

5. The fault diagnosis method of the evaporative cooling system according to claim 4, characterized in that: The various preset machine learning models include at least support vector machine, decision tree, k-nearest neighbor algorithm, logistic regression classification, multi-layer perceptron, and deep neural network.

6. The fault diagnosis method of the evaporative cooling system according to claim 2, characterized in that: Retrieving the operating data of the evaporative cooling system includes: Starting from the time point when the classification accuracy fails to meet the preset accuracy threshold, the operation data of the evaporative cooling system within the set time period is acquired again at the set sampling frequency.

7. The fault diagnosis method of the evaporative cooling system according to claim 1, characterized in that: Preprocessing the operating data includes: The operation data are normalized in rows and columns based on the time series and sensor series.

8. A fault diagnosis device for an evaporative cooling system, characterized in that: The fault diagnosis device comprises: a data acquisition module, configured to acquire operating data of the evaporative cooling system within a set time period at a set sampling frequency, and pre-process the operating data, wherein the operating data includes at least pressure data, flow data, and temperature data of the evaporative cooling system; a data decomposition module, configured to determine modal information and modal quantities of the evaporative cooling system under different operating states using fixed-window dynamic mode decomposition based on the preprocessed operating data, and to organize each modal information and modal quantity into a modal data set, wherein one modal data set corresponds to one operating state; An accuracy evaluation module is used to evaluate the classification accuracy of multiple preset machine learning models based on the modal data set, and determine the machine learning model with the highest classification accuracy as the target learning model; A parameter tuning module, configured to perform Bayesian hyperparameter tuning on the target learning model to determine the optimal parameter configuration of the target learning model; A model training module is used to train the target learning model using the optimal parameter configuration to obtain an optimal target learning model; A fault diagnosis module is used to perform fault diagnosis on the evaporative cooling system based on the optimal target learning model.

9. The fault diagnosis device for an evaporative cooling system according to claim 8, characterized in that: The fault diagnosis device further comprises: A model detection module is used to determine in real time whether the classification accuracy of the optimal target learning model during operation is lower than a preset accuracy threshold during the process of fault diagnosis of the evaporative cooling system; an algorithm updating module, configured to, if the detection result of the model detection module is that the classification accuracy is lower than a preset accuracy threshold, initiate an algorithm adaptive update, reacquire the operating data of the evaporative cooling system, and evaluate the classification accuracy of multiple preset machine learning models based on the modal data set until a new optimal target learning model is determined; If the detection result of the model detection module is that the classification accuracy is not lower than the preset accuracy threshold, the fault diagnosis module continues to use the current optimal target learning model to perform fault diagnosis.

10. An evaporative cooling system, characterized in that: The evaporative cooling system includes the fault diagnosis device of the evaporative cooling system according to any one of claims 8 to 9.

Images