Cooling state diagnosis method and device of rotor evaporative cooling generator and medium

Abstract

The invention relates to the technical field of generator cooling, and particularly provides a cooling state diagnosis method and device for a rotor evaporative cooling generator and a medium, and the method comprises the steps: obtaining a state parameter of a to-be-diagnosed rotor evaporative cooling generator; performing feature extraction on the state parameters based on a principal component analysis method to obtain principal component features of the state parameters; inputting the diagnosis result into the trained integrated learning model to obtain a diagnosis result output by each model in the trained integrated learning model; wherein the trained integrated learning model comprises a trained convolutional neural network model, a random forest model and a gradient boosting model; obtaining a cooling state diagnosis result based on the diagnosis result output by each model; feature extraction and dimension reduction are carried out through a principal component analysis method, the data dimension is reduced, key information in the data is reserved, and the diagnosis efficiency is improved; and the data is comprehensively diagnosed through the integrated learning model, so that the accuracy of the diagnosis result is improved, and the cooling state diagnosis cost is reduced.

Description

Technical Field

[0001] The invention relates to the technical field of generator cooling, and in particular to a cooling state diagnosis method, equipment and medium for a rotor evaporative cooling generator. Background Art

[0002] The cooling methods of generators mainly include three technologies: air cooling, water cooling and evaporative cooling. Air cooling removes heat by blowing air through the stator, rotor windings and core surface. As the motor capacity increases, its cooling capacity gradually approaches its limit. Water cooling directly circulates cooling water inside the stator and rotor windings, using the high specific heat of water to absorb heat. Its structure is complex and maintenance is difficult. It is prone to pipe blockage and water leakage, and even leads to short circuit risks. Evaporative cooling transfers heat through the phase change absorption of the cooling medium. Its cooling efficiency is significantly better than air cooling and water cooling. It inherits the advantages of water cooling and overcomes the complexity and leakage risk of water cooling.

[0003] Evaporative cooling technology has been well applied in generator stator windings, but generator rotor evaporative cooling technology is still under research. During generator rotor evaporative cooling, there are usually multiple cooling states. By determining the cooling state, it can be judged whether the operating state of the generator rotor during evaporative cooling is normal. Therefore, it is necessary to accurately diagnose the cooling state of the rotor evaporative cooling generator.

[0004] At present, the cooling state diagnosis of generator rotor evaporative cooling mainly relies on manual inspection, experience-based diagnosis or diagnosis based on various state parameters during evaporative cooling; the cost of manual inspection diagnosis is high; when diagnosing the type of cooling state based on various state parameters during evaporative cooling, it is necessary to calculate and judge based on the various state parameters of the rotor evaporation obtained. Due to the closed nature of various equipment during cooling, its internal operating state parameters are difficult to obtain directly. Usually, its cooling state can only be indirectly inferred by measuring the state parameters of the coolant at the inlet and outlet or along the process, which makes the cooling state diagnosis result of the generator rotor evaporative cooling less accurate and the diagnostic efficiency low.

[0005] Accordingly, the art needs a new generator rotor evaporative cooling status diagnosis solution to solve the above problems. Summary of the invention

[0006] In order to overcome the above defects, the present invention is proposed to solve or at least partially solve the technical problems of poor accuracy, low diagnostic efficiency and high diagnostic cost of the existing generator cooling state diagnosis method with rotor evaporative cooling.

[0007] In a first aspect, a cooling state diagnosis method for a rotor evaporative cooling generator is provided, the method comprising:

[0008] Obtaining state parameters of the rotor evaporative cooling generator to be diagnosed;

[0009] Extracting features of the state parameters of the rotor evaporative cooling of the generator to be diagnosed based on the principal component analysis method to obtain the principal component features of the state parameters;

[0010] Inputting the principal component features of the state parameters into a trained ensemble learning model to obtain the diagnostic results output by each model in the trained ensemble learning model; wherein the trained ensemble learning model includes a trained convolutional neural network model, a trained random forest model, and a trained gradient boosting model;

[0011] Based on the diagnosis results output by the various models, a cooling state diagnosis result of the rotor evaporative cooling generator to be diagnosed is obtained.

[0012] In a technical solution of the cooling state diagnosis method of the rotor evaporative cooling generator, the state parameters of the rotor evaporative cooling generator to be diagnosed are extracted based on the principal component analysis method to obtain the principal component characteristics of the state parameters, including:

[0013] Based on the state parameters of the rotor evaporative cooling generator to be diagnosed, obtaining a covariance matrix between various features of the state parameters;

[0014] Based on the covariance matrix between the characteristics of the state parameters, obtaining the eigenvalues ​​and eigenvectors of the covariance matrix;

[0015] Based on the size of the eigenvalues, the eigenvalues ​​are arranged in descending order to obtain the arrangement order of the eigenvalues ​​and the explained variance ratio corresponding to each eigenvalue;

[0016] Based on the arrangement order of the eigenvalues ​​and the explained variance ratios corresponding to the eigenvalues, a cumulative explained variance ratio and a corresponding cumulative explained variance ratio sequence are obtained;

[0017] Based on the cumulative explained variance ratio sequence, all eigenvalues ​​corresponding to the cumulative explained variance ratio that reach a preset cumulative explained variance ratio threshold are selected;

[0018] The eigenvectors of all eigenvalues ​​corresponding to the cumulative explained variance ratio are used as the principal component features of the state parameter.

[0019] In a technical solution of the cooling state diagnosis method of the above-mentioned rotor evaporative cooling generator, the inputting the principal component characteristics of the state parameters into the trained ensemble learning model to obtain the diagnostic results output by each model in the trained ensemble learning model includes:

[0020] Based on the principal component characteristics of the state parameters and the trained convolutional neural network model, a first diagnosis result and a fully connected layer extraction feature are obtained, wherein the fully connected layer extraction feature is obtained based on the fully connected layer extraction of the trained convolutional neural network model;

[0021] Obtaining a second diagnosis result based on the fully connected layer extracted features and the trained random forest model;

[0022] Based on the fully connected layer extracted features and the trained gradient boosting model, a third diagnosis result is obtained.

[0023] In a technical solution of the cooling state diagnosis method of the rotor evaporative cooling generator, the cooling state diagnosis result of the rotor evaporative cooling generator to be diagnosed is obtained based on the diagnosis results output by each model, including:

[0024] Based on a voting mechanism, a cooling state diagnosis result of the rotor evaporative cooling generator to be diagnosed is obtained according to the first diagnosis result, the second diagnosis result and the third diagnosis result.

[0025] In a technical solution of the cooling state diagnosis method of the above-mentioned rotor evaporative cooling generator, the fully connected layer extracts features as global feature vectors, and the trained convolutional neural network model includes a one-dimensional convolution layer, a maximum pooling layer, a flattening layer, a fully connected layer and an output layer;

[0026] The obtaining of a first diagnosis result and a fully connected layer extracted feature based on the principal component features of the state parameters and the trained convolutional neural network model comprises:

[0027] Inputting the principal component features of the state parameters into the one-dimensional convolutional layer to extract local features and obtain a local feature map;

[0028] Inputting the local feature map into the maximum pooling layer for downsampling to reduce the feature dimension, thereby obtaining a reduced-dimensional feature map;

[0029] Inputting the downsampled feature map into the flattening layer for flattening operation to obtain a flattened one-dimensional feature vector;

[0030] Inputting the flattened one-dimensional feature data into the fully connected layer for feature integration to obtain the global feature vector;

[0031] The global feature vector is input into the output layer to obtain a first diagnosis result.

[0032] In a technical solution of the cooling state diagnosis method of the above-mentioned rotor evaporative cooling generator, the second diagnosis result is obtained based on the fully connected layer extracted features and the trained random forest model, including:

[0033] Inputting the global feature vector into the trained random forest model to obtain a second diagnosis result;

[0034] The obtaining of a third diagnosis result based on the fully connected layer extracted features and the trained gradient boosting model comprises:

[0035] The global feature vector is input into the trained gradient boosting model to obtain a third diagnosis result.

[0036] In a technical solution of the cooling state diagnosis method of the above-mentioned rotor evaporative cooling generator, the method further includes training the convolutional neural network model according to the following steps:

[0037] Acquire historical state parameters of the rotor evaporative cooling generator and real labels corresponding to the historical state parameters, wherein the real labels are used to indicate the real cooling state of the rotor evaporative cooling generator;

[0038] Based on the principal component analysis method, feature extraction is performed on the historical state parameters of the rotor evaporative cooling generator to obtain the principal component features of the historical state parameters;

[0039] Inputting the principal component features of the historical state parameters into a convolutional neural network model to be trained to obtain a predictive diagnosis result and a predictive label, wherein the predictive label is used to indicate a predicted cooling state of the rotor evaporative cooling generator;

[0040] Calculating a loss function based on the true label and the predicted label;

[0041] Based on the loss function and the optimization algorithm, the gradient of the parameters of each module in the convolutional neural network model to be trained is obtained, and based on the gradient of the parameters of each module in the convolutional neural network model to be trained, the parameters of each module in the convolutional neural network model to be trained are updated.

[0042] In a technical solution of the cooling state diagnosis method of the rotor evaporative cooling generator, before extracting the characteristics of the state parameters of the rotor evaporative cooling generator to be diagnosed based on the principal component analysis method, the method further includes:

[0043] One or more pre-processing operations of centralization and standardization are performed on the state parameters of the rotor evaporative cooling generator to be diagnosed.

[0044] In a second aspect, an intelligent device is provided, comprising at least one processor; and a memory communicatively connected to the at least one processor; wherein a computer program is stored in the memory, and when the computer program is executed by the at least one processor, the method described in any one of the technical solutions of the above-mentioned method for diagnosing the cooling state of the rotor evaporative cooling generator is implemented.

[0045] In a third aspect, a computer-readable storage medium is provided, in which a plurality of program codes are stored, wherein the program codes are suitable for being loaded and run by a processor to execute the method described in any one of the technical solutions of the above-mentioned method for diagnosing the cooling state of the rotor evaporative cooling generator.

[0046] The above one or more technical solutions of the present invention have at least one or more of the following beneficial effects:

[0047] In the technical solution of the cooling state diagnosis method of the rotor evaporative cooling generator provided by the present invention, the state parameters of the rotor evaporative cooling generator to be diagnosed are obtained; based on the principal component analysis method, feature extraction is performed on the state parameters of the rotor evaporative cooling generator to be diagnosed, and the principal component characteristics of the state parameters are obtained; the principal component characteristics of the state parameters are input into a trained integrated learning model to obtain the diagnosis results output by each model in the trained integrated learning model; wherein the trained integrated learning model includes a trained convolutional neural network model, a trained random forest model and a trained gradient boosting model; based on the diagnosis results output by each model, the cooling state diagnosis result of the rotor evaporative cooling generator to be diagnosed is obtained; the principal component analysis method is used to extract features and reduce the dimension of the data, thereby reducing the data dimension and retaining the key information in the original data, thereby improving the diagnosis efficiency; the integrated learning model is used to perform comprehensive diagnosis on the data, thereby improving the accuracy of the diagnosis result and greatly reducing the cost of cooling state diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] The disclosure of the present invention will become more easily understood with reference to the accompanying drawings. It is easy for those skilled in the art to understand that these drawings are only for illustrative purposes and are not intended to limit the scope of protection of the present invention. Among them:

[0049] Figure 1 is a flow chart of main steps of a cooling state diagnosis method for a rotor evaporative cooling generator according to an embodiment of the present invention;

[0050] Figure 2 is a schematic flow chart of main steps of a method for obtaining principal component features according to an embodiment of the present invention;

[0051] Figure 3It is a flowchart diagram of the main steps of a method for obtaining the diagnostic results output by each model in a trained ensemble learning model according to an embodiment of the present invention;

[0052] Figure 4 is a schematic diagram of a trained convolutional neural network model according to an embodiment of the present invention;

[0053] Figure 5 is a visualization result diagram of the first two principal component features after being processed by the principal component analysis method according to an embodiment of the present invention;

[0054] Figure 6 It is a schematic diagram of the change of the loss function of the training set and the test set with the iteration cycle when the convolutional neural network model is trained according to an embodiment of the present invention;

[0055] Figure 7 It is a schematic diagram of the change of the accuracy of the training set and the test set with the iteration cycle when the convolutional network model is trained according to an embodiment of the present invention;

[0056] Figure 8 is a confusion matrix obtained after classifying a test set using a convolutional neural network model trained according to an embodiment of the present invention;

[0057] Fig. 9 is a confusion matrix obtained by cooling state diagnosis of an integrated learning model according to an embodiment of the present invention;

[0058] Fig.10 It is a main structural diagram of a smart device according to an embodiment of the present invention.

[0059] Reference numerals:

[0060] 101: memory; 102: processor. DETAILED DESCRIPTION

[0061] Some embodiments of the present invention are described below with reference to the accompanying drawings. It should be understood by those skilled in the art that these embodiments are only used to explain the technical principles of the present invention and are not intended to limit the protection scope of the present invention.

[0062] In the description of the present invention, "module" and "processor" may include hardware, software or a combination of the two. A module may include hardware circuits, various suitable sensors, communication ports, and memories, and may also include software parts, such as program codes, or a combination of software and hardware. The processor may be a central processing unit, a microprocessor, an image processor, a digital signal processor, or any other suitable processor. The processor has data and / or signal processing functions. The processor may be implemented in software, hardware, or a combination of the two. Computer-readable storage media include any suitable medium that can store program codes, such as a magnetic disk, a hard disk, an optical disk, a flash memory, a read-only memory, a random access memory, and the like. The term "A and / or B" means all possible combinations of A and B, such as only A, only B, or A and B. The term "at least one A or B" or "at least one of A and B" has a similar meaning to "A and / or B", and may include only A, only B, or A and B. The singular terms "one" and "the" may also include plural forms.

[0063] Generator cooling methods mainly include three technologies: air cooling, water cooling and evaporative cooling;

[0064] Air cooling is one of the earliest methods used. It removes heat by blowing air through the stator, rotor windings and core surface. It has a simple structure, reliable operation and convenient maintenance. However, as the motor capacity increases, the cooling capacity of air cooling gradually approaches its limit.

[0065] In-water cooling directly circulates cooling water inside the stator and rotor windings, using the high specific heat of water to absorb heat. It has a significant cooling effect, can make the electromagnetic load of the motor higher, the winding temperature more uniform, and the material use more efficient; but its structure is complex and difficult to maintain, and it is prone to pipe blockage and water leakage, and even leads to short circuit risks. Therefore, large-capacity hydro-turbine generators are gradually reducing the use of water cooling solutions;

[0066] Evaporative cooling technology transfers heat through the phase change heat absorption of the cooling medium, and its cooling efficiency is significantly better than air cooling and water cooling; this technology is not only non-toxic, pollution-free, and non-corrosive to motor components, but also has high insulation, fireproof, and arc-extinguishing properties, and can effectively suppress the occurrence of electrical faults; its operating temperature is low and uniform, which can avoid local overheating, while eliminating the need for fans, reducing wind friction losses, and increasing the overall efficiency by 0.1% to 0.2%, making maintenance easy and the operation safe and reliable; evaporative cooling technology inherits the advantages of water cooling, overcomes the complexity and leakage risks of water cooling, and has been well applied in generator stator windings, but generator rotor evaporative cooling technology is still under research. When the generator rotor is evaporatively cooled, there are usually multiple cooling states, and the determination of the cooling state can determine whether the operating state of the generator rotor during evaporative cooling is normal. Therefore, it is necessary to accurately diagnose the cooling state of the rotor evaporatively cooled generator to improve the method of generator rotor evaporative cooling.

[0067] In order to deeply study the performance of rotor evaporative cooling technology and its stability under different working conditions, we can simulate the cooling process in actual operation through experiments, verify the reliability of the design, and optimize the parameters of generator rotor evaporative cooling; after a large number of experiments, it is confirmed that the cooling states of generator rotor evaporative cooling mainly include: normal, condenser failure, insufficient work quality, inverter failure resulting in low speed, and power failure resulting in excessive power.

[0068] At present, the cooling state diagnosis of generator rotor evaporative cooling failure mainly relies on manual inspection, experience-based diagnosis or diagnosis based on various state parameters during evaporative cooling; the cost of manual inspection diagnosis is high; when judging the type of cooling state based on various state parameters during evaporative cooling, it is necessary to calculate and judge based on the various state parameters of the rotor evaporation obtained. Due to the closed nature of various equipment during cooling, its internal operating state parameters are difficult to obtain directly. Usually, its cooling state can only be indirectly inferred by measuring the state parameters of the coolant at the inlet and outlet or along the process, which makes the cooling state diagnosis result of the generator rotor evaporative cooling less accurate and the diagnostic efficiency low.

[0069] See attached Figure 1 , Figure 1 FIG. 1 is a flow chart showing the main steps of a cooling state diagnosis method for a rotor evaporative cooling generator according to an embodiment of the present invention. Figure 1 As shown, the cooling state diagnosis method of the rotor evaporative cooling generator in the embodiment of the present invention mainly includes the following steps S101 to S104.

[0070] Step S101: obtaining state parameters of the rotor evaporative cooling generator to be diagnosed.

[0071] In this embodiment, the state parameters of the rotor evaporative cooling generator to be diagnosed are collected, and a data set is constructed according to the collected state parameters.

[0072] Optionally, the state parameters of the rotor evaporative cooling generator to be diagnosed are state parameters of the generator rotor during evaporative cooling, and specifically may be state parameters of the cooling equipment around the rotor of the generator during evaporative cooling, such as state parameters of the cooling pipe and state parameters of the condenser. It should be noted that the state parameters of the rotor evaporative cooling generator to be diagnosed here are only exemplary and do not limit the implementation methods of the present application. In actual applications, the required state parameters can be obtained according to actual conditions to perform cooling state diagnosis of the rotor evaporative cooling generator.

[0073] In one embodiment, in order to reduce the impact of different dimensions on the data, it is usually necessary to perform one or more preprocessing operations such as centering (i.e., subtracting the mean) and standardization (i.e., dividing by the standard deviation) on the data to ensure that each variable contributes equally to the analysis; if the preprocessing operation is not performed correctly, the calculated eigenvalue may be abnormal. It should be noted that performing one or more preprocessing operations such as centering and standardization on the data is only an exemplary description. In actual applications, the preprocessing operation can be selected as needed.

[0074] Step S102: extracting features of the state parameters of the rotor evaporative cooling generator to be diagnosed based on the principal component analysis method to obtain principal component features of the state parameters.

[0075] In this embodiment, principal component analysis (PCA) is used to extract features from the data set, reduce the dimension, and retain the principal component features whose cumulative explained variance ratio reaches a preset cumulative explained variance ratio threshold.

[0076] PCA is a popular data dimensionality reduction technique. Its core idea is to project the original data into a new coordinate system through linear transformation. The basis vectors of this new coordinate system are the principal component features, which are sorted according to the degree to which they can explain the variance of the original data. The principal component features can capture most of the important information of the data, and usually fewer principal component features can retain most of the variability in the original data. In this way, although we reduce the dimension of the data, we still retain the core characteristics of the data, thereby achieving the purpose of reducing the dimension while retaining key information.

[0077] In this embodiment, see the attached Figure 2 , Figure 2 FIG. 1 is a flow chart showing the main steps of a method for obtaining principal component features according to an embodiment of the present invention. Figure 2 As shown, the method for obtaining the principal component features in the embodiment of the present invention mainly includes the following steps.

[0078] Step S201: based on the state parameters of the rotor evaporative cooling generator to be diagnosed, obtaining a covariance matrix between various characteristics of the state parameters;

[0079] Specifically, the state parameters of the rotor evaporative cooling generator to be diagnosed include m samples, each sample has n features, and the matrix X corresponding to the n features is:

[0080]

[0081] Calculate the mean of each feature (i.e. column vector) for:

[0082]

[0083] Subtract each value in the matrix X from the average value of the column vector calculated above to obtain a new matrix X′:

[0084]

[0085] Calculate the transposed matrix X' of the matrix X' T And the covariance matrix S:

[0086]

[0087] Step S202: based on the covariance matrix between the various features of the state parameters, obtaining the eigenvalues ​​and eigenvectors of the covariance matrix;

[0088] Specifically, the eigenvalue λ and eigenvector α of the covariance matrix S are calculated, that is, the following determinant is solved:

[0089] |S-λE|=0

[0090] Where E is an n-dimensional unit vector

[0091] Step S203: based on the size of the eigenvalues, the eigenvalues ​​are arranged in descending order to obtain the arrangement order of the eigenvalues ​​and the explained variance ratio corresponding to each eigenvalue;

[0092] Specifically, according to the size of the eigenvalue, the eigenvalues ​​are sorted from high to low, and the explained variance ratio corresponding to each eigenvalue is obtained.

[0093] For example, suppose there is a 3D data set, and the PCA method is executed to obtain the following three eigenvalues: λ1=3.0, λ2=2.0, λ3=1.0. The order of the above eigenvalues ​​from high to low is the order of the eigenvalues. The proportion of each eigenvalue to the total of all eigenvalues ​​is calculated to obtain the explained variance ratio corresponding to each eigenvalue. It should be noted that the three-dimensional data set assumed here is only an example of an application scenario and is not a limitation of this embodiment. In actual applications of the embodiments of the present invention, the number and value of the actual eigenvalues ​​are obtained based on actual acquisition.

[0094] The calculation formula for the explained variance ratio p corresponding to the i-th eigenvalue is:

[0095]

[0096] When λ1=3.0, the explained variance ratio is

[0097] When λ2=2.0, the explained variance ratio is

[0098] When λ3=1.0, the explained variance ratio is

[0099] Step S204: based on the arrangement order of the eigenvalues ​​and the explained variance ratios corresponding to the eigenvalues, obtaining the cumulative explained variance ratios and the corresponding cumulative explained variance ratio sequences;

[0100] For example, when accumulated to the first eigenvalue λ1, the cumulative explained variance ratio is 0.5;

[0101] When accumulated to the second eigenvalue λ2, the cumulative explained variance ratio is 0.5+0.333=0.833;

[0102] When the third eigenvalue λ3 is accumulated, the cumulative explained variance ratio is 0.5+0.333+0.167=1;

[0103] Among them, the order of the corresponding relationship between the above cumulative eigenvalues ​​and the cumulative explained variance ratios is the cumulative explained variance ratio sequence.

[0104] Step S205: based on the cumulative explained variance ratio sequence, selecting all eigenvalues ​​corresponding to the cumulative explained variance ratio that reaches a preset cumulative explained variance ratio threshold;

[0105] In practical applications, the number of principal components to be retained may be determined based on the cumulative explained variance ratio. For example, if you want to retain at least 80% of the information, then the preset cumulative explained variance ratio threshold is 80%. According to the cumulative explained variance ratio sequence, all eigenvalues ​​corresponding to the cumulative explained variance ratio exceeding 83.33% are determined as the required eigenvalues. In this example, the first two eigenvalues ​​λ1 and λ2 are selected to be retained (the cumulative explained variance ratio of the first two eigenvalues ​​reaches 83.3%, reaching the preset cumulative explained variance ratio threshold). It should be noted that retaining the principal component characteristics with a cumulative explained variance ratio of 80% here is only an exemplary explanation. In practical applications, the value of the cumulative explained variance ratio threshold can be selected as needed.

[0106] Step S206: taking the eigenvectors of all eigenvalues ​​corresponding to the cumulative explained variance ratio as the principal component features of the state parameters.

[0107] Specifically, the eigenvectors corresponding to the selected eigenvalues ​​are formed into an eigenvector matrix, and the data of the matrix X′ is projected into the eigenvector matrix to obtain multiple principal component features, each of which carries a portion of the information in the original data set.

[0108] Step S103: Input the principal component features of the state parameters into the trained ensemble learning model to obtain the diagnostic results output by each model in the trained ensemble learning model; wherein the trained ensemble learning model includes a trained convolutional neural network model, a trained random forest model and a trained gradient boosting model.

[0109] In this embodiment, see the attached Figure 3 , Figure 3 The present invention is a flowchart of the main steps of a method for obtaining the diagnostic results output by each model in a trained ensemble learning model according to an embodiment of the present invention. The principal component features of the operating data are input into the trained ensemble learning model to obtain the diagnostic results output by each model in the trained ensemble learning model, including the following steps:

[0110] S301: Based on the principal component features of the state parameters and the trained convolutional neural network model, a first diagnosis result and a fully connected layer extraction feature are obtained, where the fully connected layer extraction feature is obtained based on the fully connected layer extraction of the trained convolutional neural network model;

[0111] Specifically, the fully connected layer extracts features as global feature vectors, see Appendix Figure 4 , Figure 4 Schematic diagram of a trained convolutional neural network model according to an embodiment of the present invention. Figure 4As shown, the trained convolutional neural network model includes a one-dimensional convolution layer, a maximum pooling layer, a flattening layer, a fully connected layer, and an output layer;

[0112] The method for obtaining the first diagnosis result and the fully connected layer extracted features includes:

[0113] 1. The principal component features of the state parameters are input into the one-dimensional convolution layer for local feature extraction to obtain a local feature map. In the one-dimensional convolution layer, 32 filters and a convolution kernel of size 2 are used, and the ReLU activation function is used for processing to introduce nonlinear features.

[0114] 2. Input the local feature map into the maximum pooling layer for downsampling to reduce the feature dimension and obtain a reduced-dimensional feature map; wherein the maximum pooling layer includes a convolution kernel of size 2.

[0115] 3. Input the downsampled feature map into the flattening layer for flattening to obtain the flattened one-dimensional feature vector;

[0116] 4. Input the flattened one-dimensional feature data into the fully connected layer for feature integration to obtain the global feature vector;

[0117] Among them, in the fully connected layer, a hidden layer containing 64 units is used, and the ReLU activation function is applied to further learn complex features;

[0118] 5. Input the global feature vector into the output layer to obtain the first diagnosis result;

[0119] Among them, the output layer consists of five units, and the softmax activation function is used to generate the probability distribution of 5 categories to achieve multi-classification tasks.

[0120] In this embodiment, the method further includes training the convolutional neural network model according to the following steps:

[0121] (1) Obtaining historical state parameters of the rotor evaporative cooling generator and real labels corresponding to the historical state parameters, wherein the real labels are used to indicate the real cooling state of the rotor evaporative cooling generator; the cooling state includes multiple cooling states such as normal, condenser failure, insufficient work quality, etc., corresponding to labels marked as 0, 1, 2, ..., n respectively;

[0122] (2) Based on the principal component analysis method, feature extraction is performed on the historical state parameters of the rotor evaporative cooling generator to obtain the principal component features of the historical state parameters; the principal component features of the historical state parameters are divided into a training set and a test set in a ratio of 8:2;

[0123] (3) inputting the training set of the principal component features of the historical state parameters into the convolutional neural network model to be trained to obtain a predictive diagnosis result and a predictive label, wherein the predictive label is used to indicate the predicted cooling state of the rotor evaporative cooling generator;

[0124] (4) Calculate the loss function based on the true label and the predicted label;

[0125] (5) Based on the loss function and the optimization algorithm, the gradient of the parameters of each module in the convolutional neural network model to be trained is obtained, and based on the gradient of the parameters of each module in the convolutional neural network model to be trained, the parameters of each module in the convolutional neural network model to be trained are updated.

[0126] Optionally, select the adam optimization algorithm, which is a commonly used adaptive optimization algorithm that can efficiently adjust the learning rate to accelerate convergence; the loss function uses the categorical_crossentropy loss function, which is suitable for multi-classification tasks and can calculate the error between the probability distribution of the model output and the true label; the evaluation indicator is set to accuracy to monitor the accuracy of the classification task; under this configuration, the model will use the adam optimization algorithm to continuously adjust the weights to minimize the classification error and evaluate the performance of the model by accuracy; the iteration cycle is 100.

[0127] Use the trained convolutional neural network model to classify the test set data, and use the classification accuracy to evaluate the classification performance of the model. As the iteration cycle changes, the accuracy and loss value of the model tend to be stable and will no longer fluctuate greatly. The trained convolutional neural network model is output.

[0128] It should be noted that the selection of the adam optimization algorithm, categorical_crossentropy loss function and accuracy evaluation index here is only for illustrative purposes and is not a limitation to this embodiment. In practical applications of the embodiments of the present invention, the actual optimization algorithm, loss function and type of evaluation index can be selected according to actual needs.

[0129] S302: Obtain a second diagnosis result based on the fully connected layer extracted features and the trained random forest model;

[0130] Specifically, the fully connected layer extracts features as a global feature vector, and the global feature vector is input into the trained random forest model to obtain the second diagnosis result.

[0131] Exemplarily, the trained random forest model completes the classification task by constructing 100 decision trees (set by the parameter n_estimators = 100) and combining it with a majority voting mechanism; during the training process of the random forest model, each tree is constructed by performing random sampling (bootstrap) with replacement on the training data and randomly selecting some features; its parameter random_state = 42 ensures that the random results of the model are repeatable. In the end, each tree learns independently, and the results are predicted by combining the majority voting method. The model performs robustly and has the ability to resist overfitting. It should be noted that the number of decision trees in the random forest model here is only an exemplary description and is not a limitation on the embodiments of the present invention.

[0132] S303: Obtain a third diagnosis result based on the fully connected layer extracted features and the trained gradient boosting model;

[0133] Specifically, the fully connected layer extracts features as a global feature vector, and the global feature vector is input into the trained gradient boosting model to obtain the third diagnosis result.

[0134] Exemplarily, the trained gradient boosting model trains 100 weak classifiers (set by the parameter n_estimators = 100) by gradually optimizing the residuals. These weak classifiers are usually shallow decision trees. When training the gradient boosting model, in each iteration, the model minimizes the loss function by fitting the residuals predicted in the previous step, gradually reducing the error. The parameter random_state = 42 also ensures that the random results are repeatable. Gradient boosting uses the superposition effect of multiple weak classifiers to generate a powerful classification model that can capture complex patterns of data and is suitable for tasks that require high precision. It should be noted that the number of weak classifiers in the gradient boosting model here is only an exemplary description and is not a limitation on the embodiments of the present invention.

[0135] S104: Based on the diagnosis results output by each model, a cooling state diagnosis result of the rotor evaporative cooling generator to be diagnosed is obtained.

[0136] In this embodiment, based on a voting mechanism, a cooling state diagnosis result of the rotor evaporative cooling generator to be diagnosed is obtained according to the first diagnosis result, the second diagnosis result and the third diagnosis result.

[0137] For example, when the diagnostic results output by the trained convolutional neural network model and the trained random forest model are both condenser failure, and the diagnostic result output by the trained gradient boosting model is insufficient work quality, then based on the voting mechanism, it can be seen that the voting result is 2:1, and the cooling state diagnosis result of the rotor evaporative cooling generator to be diagnosed is condenser failure. It should be noted that the diagnostic results and voting results output by each model here are only exemplary descriptions and are not limitations on the embodiments of the present invention.

[0138] In an application scenario according to an embodiment of the present invention, a generator rotor evaporative cooling test bench can be built for testing. By obtaining the state parameters of the test bench and determining the cooling state corresponding to the state parameters, the cooling state of the rotor evaporative cooling generator can be diagnosed. It should be noted that the construction of the generator rotor evaporative cooling test bench and obtaining the state parameters of the test bench to diagnose the cooling state of the rotor evaporative cooling generator is only an exemplary description and does not limit the implementation method of the present application. In actual applications, the state parameters can also be obtained during the evaporative cooling of the generator rotor to diagnose the cooling failure of the rotor evaporative cooling generator as needed.

[0139] In this embodiment, the generator rotor evaporative cooling test bench includes a cooling pipe, a condenser, etc.; the method for obtaining specific state parameters is: arranging pressure sensors at the inlet and outlet positions of the cooling pipe of the generator rotor evaporative cooling test bench and at the inlet and outlet positions of the condenser working fluid; arranging flow sensors at the inlet and outlet positions of the cooling pipe; arranging temperature sensors at the inlet and outlet positions of the cooling pipe, at the inlet and outlet positions of the condenser working fluid, at the inlet and outlet positions of the condenser cooling water, and along the cooling pipe.

[0140] The method for testing the evaporative cooling of a generator rotor and diagnosing the cooling state on a generator rotor evaporative cooling test bench comprises the following steps:

[0141] Step 1: Obtain the state parameters of the generator rotor evaporative cooling test bench to be diagnosed.

[0142] In this embodiment, after setting various sensors, data is collected on the generator rotor evaporative cooling test bench. There are five cooling states in the generator rotor evaporative cooling test bench: normal, condenser failure, insufficient work quality, inverter failure resulting in too low speed, and power failure resulting in too high power. The five cooling states are marked with five types of labels 0-4 respectively;

[0143] Collecting a plurality of sample data in each of the five cooling states of the generator rotor evaporative cooling as state parameters of the generator rotor evaporative cooling test bench to be diagnosed;

[0144] The collected state parameters of the cooling pipe inlet working fluid pressure (Pa), cooling pipe outlet working fluid pressure (Pa), condenser inlet working fluid pressure (Pa), condenser outlet working fluid pressure (Pa), cooling pipe inlet flow (mL / min), cooling pipe outlet flow (mL / min), cooling pipe inlet working fluid temperature (℃), cooling pipe outlet working fluid temperature (℃), condenser inlet working fluid temperature (℃), condenser outlet working fluid temperature (℃), cooling water inlet water temperature (℃), cooling water outlet water temperature (℃), and average temperature along the cooling pipe (℃) are used as 13 features to construct a data set. It should be noted that the types of state parameters obtained here and the positions at which the state parameters are obtained are only exemplary and do not limit the implementation methods of this application. In actual applications, the types of state parameters and the acquisition positions can be set as needed.

[0145] Step 2: Based on the principal component analysis method, the state parameters of the generator rotor evaporative cooling test bench to be diagnosed are extracted to obtain the principal component characteristics of the state parameters.

[0146] In this embodiment, principal component analysis (PCA) is used to extract features from the data set (13 features), reduce the dimension, and retain the principal component features whose cumulative explained variance ratio reaches a preset cumulative explained variance ratio threshold.

[0147] In this embodiment, see the attached Figure 5 , Figure 5 is a visualization result diagram of the first two principal component features after being processed by the principal component analysis method according to an embodiment of the present invention. Figure 5 It can be seen that after the data is reduced in dimension by PCA, the data features of different categories are quite different, indicating that its feature extraction ability is strong.

[0148] Step 3: Input the principal component features of the state parameters into the trained ensemble learning model to obtain the diagnostic results output by each model in the trained ensemble learning model; wherein the trained ensemble learning model includes a trained convolutional neural network model, a trained random forest model and a trained gradient boosting model.

[0149] In this embodiment, the principal component features of the state parameters are input into a trained convolutional neural network model to obtain a first diagnostic result and fully connected layer extracted features, where the fully connected layer extracted features are extracted based on the fully connected layer of the convolutional neural network; the fully connected layer extracted features are input into a trained random forest model to obtain a second diagnostic result; the fully connected layer extracted features are input into a trained gradient boosting model to obtain a third diagnostic result.

[0150] In this embodiment, the method further includes training the convolutional neural network model according to the following steps:

[0151] (1) obtaining historical state parameters of the generator rotor evaporative cooling test bench and real labels corresponding to the historical state parameters, wherein the real labels are used to indicate the real cooling state of the generator rotor evaporative cooling test bench;

[0152] (2) Based on the principal component analysis method, the historical state parameters of the generator rotor evaporative cooling test bench are feature extracted to obtain the principal component characteristics of the historical state parameters; the principal component characteristics of the historical state parameters are divided into a training set and a test set in a ratio of 8:2;

[0153] (3) inputting the training set of the principal component features of the historical state parameters into the convolutional neural network model to be trained to obtain a predictive diagnosis result and a predictive label, wherein the predictive label is used to indicate the predicted cooling state of the generator rotor evaporative cooling test bench;

[0154] (4) Calculate the loss function based on the true label and the predicted label;

[0155] (5) Based on the loss function and the optimization algorithm, the gradient of the parameters of each module in the convolutional neural network model to be trained is obtained, and based on the gradient of the parameters of each module in the convolutional neural network model to be trained, the parameters of each module in the convolutional neural network model to be trained are updated.

[0156] In this embodiment, see the attached Figure 6 -Attached Figure 8 . Figure 6 It is a schematic diagram of the change of the loss function of the training set and the test set with the iteration cycle when the convolutional neural network model is trained according to an embodiment of the present invention. The test set samples are classified based on the trained convolutional neural network model, and the classification accuracy is 96.50%; Figure 7 FIG. 1 is a schematic diagram showing how the accuracy of the training set and the test set varies with the iteration period when the convolutional network model is trained according to an embodiment of the present invention. Figure 6 and Figure 7 It can be seen that the final loss function and accuracy tend to be stable, indicating that the model is not obviously overfitting. Figure 8 is the confusion matrix obtained after classifying the test set by the convolutional neural network model trained according to an embodiment of the present invention, Figure 8 It can be seen that the trained convolutional neural network model has a relatively high accuracy, indicating that it can effectively improve the accuracy of cooling failure.

[0157] In this embodiment, the trained random forest model completes the classification task by constructing 100 decision trees (set by the parameter n_estimators = 100) and combining it with the majority voting mechanism; during the training process of the random forest model, each tree is constructed by random sampling (bootstrap) with replacement of the training data and randomly selecting some features; its parameter random_state = 42 ensures that the random results of the model are repeatable. In the end, each tree learns independently, and the results are predicted by combining the majority voting method. The model is robust and has the ability to resist overfitting. It should be noted that the number of decision trees in the random forest model here is only an exemplary description and is not a limitation on the implementation methods of the present invention.

[0158] In this embodiment, the trained gradient boosting model trains 100 weak classifiers (set by the parameter n_estimators = 100) by gradually optimizing the residuals. These weak classifiers are usually shallow decision trees. When training the gradient boosting model, in each round of iteration, the model minimizes the loss function by fitting the residuals predicted in the previous step, gradually reducing the error. The parameter random_state = 42 also ensures that the random results are repeatable; gradient boosting uses the superposition effect of multiple weak classifiers to generate a powerful classification model that can capture complex patterns of data and is suitable for tasks that require high precision. It should be noted that the number of weak classifiers in the gradient boosting model here is only an exemplary description and is not a limitation on the embodiments of the present invention.

[0159] Step 4: Based on the diagnostic results output by each model, a cooling state diagnostic result of the rotor evaporative cooling generator to be diagnosed is obtained;

[0160] In this embodiment, based on a voting mechanism, a cooling state diagnosis result of the generator rotor evaporative cooling test bench to be diagnosed is obtained according to the first diagnosis result, the second diagnosis result and the third diagnosis result.

[0161] See attached Fig. 9 , Fig. 9 is a confusion matrix obtained by diagnosing the cooling state of the integrated learning model according to an embodiment of the present invention. Fig. 9 As shown in the figure, the classification accuracy of the integrated learning model is 97%, which is better than the single convolutional neural network model.

[0162] It should be pointed out that although the various steps in the above embodiments are described in a specific order, those skilled in the art can understand that in order to achieve the effect of the present invention, different steps do not have to be executed in such an order, they can be executed simultaneously (in parallel) or in other orders. These adjusted schemes are equivalent to the technical schemes described in the present invention, and therefore will also fall within the scope of protection of the present invention.

[0163] It is understood by those skilled in the art that the present invention implements all or part of the processes in the method of the above embodiment, and can also be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned various method embodiments when executed by the processor. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable storage medium may include: any entity or device, medium, U disk, mobile hard disk, disk, optical disk, computer memory, read-only memory, random access memory, electric carrier signal, telecommunication signal and software distribution medium, etc. that can carry the computer program code.

[0164] Another aspect of the present invention provides a computer-readable storage medium.

[0165] In an embodiment of a computer-readable storage medium according to the present invention, the computer-readable storage medium may be configured to store a program for executing the cooling state diagnosis method of the rotor evaporative cooling generator of the above method embodiment, and the program may be loaded and run by a processor to implement the cooling state diagnosis method of the rotor evaporative cooling generator. For ease of explanation, only the parts related to the embodiment of the present invention are shown. For specific technical details not disclosed, please refer to the method part of the embodiment of the present invention. The computer-readable storage medium may be a storage device formed by various electronic devices. Optionally, the computer-readable storage medium in the embodiment of the present invention is a non-temporary computer-readable storage medium.

[0166] Another aspect of the present invention provides a smart device.

[0167] In an embodiment of an intelligent device according to the present invention, the intelligent device may include at least one processor; and a memory connected to the at least one processor in communication; wherein a computer program is stored in the memory, and when the computer program is executed by the at least one processor, the method described in any of the above embodiments is implemented. Fig.10 , Fig.10 Schematically shows that the memory 101 and the processor 102 are communicatively connected via a bus.

[0168] The smart device described in the present invention may be, but is not limited to, a mobile phone, a tablet computer, a desktop, a laptop, a handheld computer, a notebook computer, a vehicle-mounted device, an ultra-mobile personal computer (UMPC), a netbook, a personal digital assistant (PDA), an augmented reality (AR) or a virtual reality (VR) device, etc., and the embodiments of the present invention are not limited to this.

[0169] So far, the technical solution of the present invention has been described in conjunction with an embodiment shown in the accompanying drawings, but it is easy for those skilled in the art to understand that the protection scope of the present invention is obviously not limited to these specific embodiments. Without departing from the principle of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will fall within the protection scope of the present invention.

Claims

1. A method for diagnosing the cooling state of a rotor evaporative cooling generator, characterized in that: include: Obtaining state parameters of the rotor evaporative cooling generator to be diagnosed; Extracting features of the state parameters of the rotor evaporative cooling generator to be diagnosed based on the principal component analysis method to obtain principal component features of the state parameters; Inputting the principal component features of the state parameters into a trained ensemble learning model to obtain the diagnostic results output by each model in the trained ensemble learning model; wherein the trained ensemble learning model includes a trained convolutional neural network model, a trained random forest model, and a trained gradient boosting model; Based on the diagnosis results output by the various models, a cooling state diagnosis result of the rotor evaporative cooling generator to be diagnosed is obtained.

2. The cooling state diagnosis method of the rotor evaporative cooling generator according to claim 1, characterized in that: The method of extracting the characteristics of the state parameters of the rotor evaporative cooling generator to be diagnosed based on the principal component analysis method to obtain the principal component characteristics of the state parameters includes: Based on the state parameters of the rotor evaporative cooling generator to be diagnosed, obtaining a covariance matrix between various features of the state parameters; Based on the covariance matrix between the characteristics of the state parameters, obtaining the eigenvalues ​​and eigenvectors of the covariance matrix; Based on the size of the eigenvalues, the eigenvalues ​​are arranged in descending order to obtain the arrangement order of the eigenvalues ​​and the explained variance ratio corresponding to each eigenvalue; Based on the arrangement order of the eigenvalues ​​and the explained variance ratios corresponding to the eigenvalues, a cumulative explained variance ratio and a corresponding cumulative explained variance ratio sequence are obtained; Based on the cumulative explained variance ratio sequence, all eigenvalues ​​corresponding to the cumulative explained variance ratio that reach a preset cumulative explained variance ratio threshold are selected; The eigenvectors of all eigenvalues ​​corresponding to the cumulative explained variance ratio are used as the principal component features of the state parameter.

3. The cooling state diagnosis method of the rotor evaporative cooling generator according to claim 1, characterized in that: The step of inputting the principal component features of the state parameters into a trained integrated learning model to obtain the diagnostic results output by each model in the trained integrated learning model comprises: Based on the principal component characteristics of the state parameters and the trained convolutional neural network model, a first diagnosis result and a fully connected layer extraction feature are obtained, wherein the fully connected layer extraction feature is obtained based on the fully connected layer extraction of the trained convolutional neural network model; Obtaining a second diagnosis result based on the fully connected layer extracted features and the trained random forest model; Based on the fully connected layer extracted features and the trained gradient boosting model, a third diagnosis result is obtained.

4. The cooling state diagnosis method of the rotor evaporative cooling generator according to claim 3, characterized in that: The step of obtaining the cooling state diagnosis result of the rotor evaporative cooling generator to be diagnosed based on the diagnosis results output by each model includes: Based on a voting mechanism, a cooling state diagnosis result of the rotor evaporative cooling generator to be diagnosed is obtained according to the first diagnosis result, the second diagnosis result and the third diagnosis result.

5. The cooling state diagnosis method of the rotor evaporative cooling generator according to claim 3, characterized in that: The fully connected layer extracts features as global feature vectors, and the trained convolutional neural network model includes a one-dimensional convolutional layer, a maximum pooling layer, a flattening layer, a fully connected layer and an output layer; The obtaining of a first diagnosis result and a fully connected layer extracted feature based on the principal component features of the state parameters and the trained convolutional neural network model comprises: Inputting the principal component features of the state parameters into the one-dimensional convolutional layer to extract local features and obtain a local feature map; Inputting the local feature map into the maximum pooling layer for downsampling to reduce the feature dimension, thereby obtaining a reduced-dimensional feature map; Inputting the downsampled feature map into the flattening layer for flattening operation to obtain a flattened one-dimensional feature vector; Inputting the flattened one-dimensional feature data into the fully connected layer for feature integration to obtain the global feature vector; The global feature vector is input into the output layer to obtain a first diagnosis result.

6. The cooling state diagnosis method of the rotor evaporative cooling generator according to claim 5, characterized in that: The step of extracting features based on the fully connected layer and the trained random forest model to obtain a second diagnosis result includes: Inputting the global feature vector into the trained random forest model to obtain a second diagnosis result; The obtaining of a third diagnosis result based on the fully connected layer extracted features and the trained gradient boosting model comprises: The global feature vector is input into the trained gradient boosting model to obtain a third diagnosis result.

7. The cooling state diagnosis method of a rotor evaporative cooling generator according to claim 1, characterized in that: The method also includes training the convolutional neural network model according to the following steps: Acquire historical state parameters of the rotor evaporative cooling generator and real labels corresponding to the historical state parameters, wherein the real labels are used to indicate the real cooling state of the rotor evaporative cooling generator; Based on the principal component analysis method, feature extraction is performed on the historical state parameters of the rotor evaporative cooling generator to obtain the principal component features of the historical state parameters; Inputting the principal component features of the historical state parameters into a convolutional neural network model to be trained to obtain a predictive diagnosis result and a predictive label, wherein the predictive label is used to indicate a predicted cooling state of the rotor evaporative cooling generator; Calculating a loss function based on the true label and the predicted label; Based on the loss function and the optimization algorithm, the gradient of the parameters of each module in the convolutional neural network model to be trained is obtained, and based on the gradient of the parameters of each module in the convolutional neural network model to be trained, the parameters of each module in the convolutional neural network model to be trained are updated.

8. The cooling state diagnosis method for a rotor evaporative cooling generator according to any one of claims 1 to 7, before extracting features of the state parameters of the rotor evaporative cooling generator to be diagnosed based on the principal component analysis method, the method further comprises: One or more pre-processing operations of centralization and standardization are performed on the state parameters of the rotor evaporative cooling generator to be diagnosed.

9. A smart device, characterized in that: include: at least one processor; and, a memory communicatively coupled to the at least one processor; Wherein, a computer program is stored in the memory, and when the computer program is executed by the at least one processor, the cooling state diagnosis method of the rotor evaporative cooling generator according to any one of claims 1 to 8 is implemented.

10. A computer-readable storage medium storing a plurality of program codes, characterized in that: The program code is suitable for being loaded and run by a processor to execute the cooling state diagnosis method for a rotor evaporative cooling generator according to any one of claims 1 to 8.

Images