Hydroelectric generating set intelligent fault diagnosis method based on multi-sensor data fusion

Abstract

The invention belongs to the technical field of hydroelectric generating set fault diagnosis, and particularly discloses a hydroelectric generating set intelligent fault diagnosis method based on multi-sensor data fusion. The method comprises the following steps: firstly, constructing two branch convolutional neural networks to respectively extract time domain features and frequency domain features of sensing data, and then constructing a central convolutional neural network to perform feature extraction on fusion features obtained by fusing the shallow time domain features and the frequency domain features; fusing the fusion feature obtained by the convolution of the # imgabs0 # layer and the time domain feature and the frequency domain feature obtained by the convolution of the # imgabs1 # layer in the branch convolutional neural network, and updating the fusion feature obtained by the convolution of the # imgabs2 # layer; outputting a preliminary diagnosis result by the central convolutional neural network; and finally, performing decision fusion based on information entropy on a preliminary diagnosis result obtained by the multiple paths of sensing data to obtain a final diagnosis result. Compared with the existing diagnosis method, the diagnosis method provided by the invention has higher anti-interference capability and robustness.

Description

Technical Field

[0001] This application belongs to the technical field of hydropower unit fault diagnosis, and more specifically, relates to an intelligent fault diagnosis method for hydropower units based on multi-sensor data fusion. Background Technique

[0002] The intelligent fault diagnosis of hydropower units is of great significance for ensuring the safety of power station production and the stability of the power grid. With the continuous development of artificial intelligence and monitoring technologies, data-driven intelligent fault diagnosis methods have become a current research hotspot and have certain application conditions in the actual production environment. The data-driven intelligent fault diagnosis method analyzes the operating state of the unit in real time, can timely detect potential faults and hidden dangers of the unit, and improve the intelligent management level of the power station.

[0003] The data-driven fault diagnosis of hydropower units is mainly divided into two types: fault diagnosis based on traditional machine learning and fault diagnosis based on deep learning. Fault diagnosis based on traditional machine learning mainly includes Random Forest (RF), Support Vector Machines (SVM), etc. These models have simple principles, fast training speeds, and good real-time performance, but they rely on signal processing methods and expert experience. In the face of complex data distributions in the actual industrial environment, the effect of fault recognition still has deficiencies. Fault diagnosis based on deep learning mainly includes Convolutional Neural Network (CNN), Deep Auto-encoder (DAE), etc. These methods have powerful nonlinear capabilities and learning abilities, can accurately extract the information contained in complex data, and reduce the interference of human factors at the same time. Some deep learning models have achieved good results in the field of hydropower unit fault diagnosis, but most of these methods use single-sensor data for model training and fault recognition, which will cause the omission of some fault information. When the sensor fails or is interfered by strong noise, the diagnostic performance of the model will drop significantly, which is not conducive to the accurate diagnosis of the fault state of the equipment. Summary of the Invention

[0004] In view of the above defects or improvement requirements of the prior art, this application provides an intelligent fault diagnosis method for hydropower units based on multi-sensor data fusion, aiming to solve the technical problem that it is easy to miss fault information when using single-sensor data for fault recognition in the prior art.

[0005] To achieve the above object, in the first aspect, this application provides a training method for an intelligent fault diagnosis model of a hydropower unit, including: Collect multi-channel sensing data of the hydropower unit and preprocess the sensing data; Construct two branch convolutional neural networks to respectively extract the time-domain features and frequency-domain features of the sensing data, and perform attention fusion on the time-domain features and frequency-domain features obtained by the first-layer convolution to obtain fused features; Construct a central convolutional neural network to extract features from the fused features, and use the fused features obtained by the -th layer convolution therein, the time-domain features and frequency-domain features obtained by the -th layer convolution in the branch convolutional neural network to perform attention fusion and then update the fused features obtained by the -th layer convolution; the central convolutional neural network finally outputs a preliminary diagnosis result; Obtain the corresponding information entropy from the sensitivity of each path of sensor data to the fault, and perform decision fusion on the preliminary diagnosis results obtained from each path of sensor data based on the information entropy to obtain the final diagnosis result; use the gap between the final diagnosis result and the actual label as a loss function to train the parameters of each convolutional neural network, and finally obtain the intelligent fault diagnosis model of the hydropower unit.

[0006] Preferably, the attention fusion is specifically as follows: Compress the feature vectors of each feature to be fused through pooling, and form a compressed vector from all the compression values; Calculate the similarity of the compressed vector to obtain the attention weights of each feature to be fused; Based on the attention weights, perform weighted summation on each feature to be fused to obtain the fused features.

[0007] Preferably, performing decision fusion on the preliminary diagnosis results obtained from each path of sensor data based on the information entropy to obtain the final diagnosis result is specifically as follows: Obtain the weight ratio of each information entropy among all the information entropies; Normalize the weight ratio of each information entropy to obtain the fusion weights corresponding to each path of sensor data; Use the fusion weights of each path of sensing data to perform weighted summation on the preliminary diagnosis results obtained from each path of sensing data to obtain the final diagnosis result.

[0008] Preferably, the weight ratio of each information entropy is:

[0009] wherein, is the information entropy corresponding to the -th path of sensor data, is the total number of paths of sensors; the fusion weight corresponding to the -th path of sensor data

[0010] Among them, is the activation function; The final diagnosis result is:

[0011] Among them, represents the probability of belonging to the th type of fault, is the preliminary diagnosis result of each path of sensing data.

[0012] Preferably, the corresponding information entropy is obtained from the sensitivity of each path of sensor data to the fault, specifically:

[0013] Among them, is the information entropy corresponding to the th path of sensor data; represents the possible probability of the th data in the th path of sensor data; is the length of the sensor data; is the preset logarithm base.

[0014] Preferably, the sensing data is subjected to Fourier transform to convert the sensing data in the time domain into the sensing data in the frequency domain, the time-domain features are obtained from the sensing data in the time domain, and the frequency-domain features are obtained from the sensing data in the frequency domain.

[0015] Preferably, the branch convolutional neural network includes four convolutional-pooling layers; the central convolutional neural network includes three convolutional-pooling layers and a fully connected layer.

[0016] Preferably, the cross entropy is used to measure the gap between the final diagnosis result and the actual label, and the cross entropy is:

[0017] Among them, is the actual label, is the final diagnosis result, and the subscript represents the rd in the distribution, and the base of the logarithm is a preset value.

[0018] In a second aspect, the present application provides a method for intelligent fault diagnosis of a hydropower unit based on multi-sensor data fusion, including: Collect multi-path sensing data of the hydropower unit and preprocess the sensing data; Obtain the time-domain signal and frequency-domain signal of each path of the sensing data; Input the time-domain signal and frequency-domain signal of each path of the sensing data into the intelligent fault diagnosis model of the hydropower unit as described in any one of claims 1-8 respectively; The intelligent fault diagnosis model of the hydropower unit outputs the fault diagnosis result of the hydropower unit.

[0019] In a third aspect, the present application provides an electronic device, including: At least one memory for storing programs; At least one processor for executing the programs stored in the memory. When the programs stored in the memory are executed, the processor is used to execute the method described in any one of the first aspects, or is used to execute the method described in the second aspect.

[0020] Generally speaking, compared with the prior art through the above technical solutions conceived by the present application, the following beneficial effects are obtained: (1) The present application proposes a fault diagnosis method based on multi-path sensing data and multi-dimensional data fusion, which has stronger anti-interference ability and robustness compared with the prior art that relies on single-sensor data for fault diagnosis.

[0021] (2) In the process of data feature extraction and fusion, the present application's fault diagnosis method further fuses the data features of different convolutional depths of the model, and the obtained features are more representative and easier to mine fault information from the feature data.

[0022] (3) The present application's fault diagnosis method performs decision fusion based on information entropy on the diagnosis results obtained from multi-path data features, thereby fully considering the association between different sensor differences and the reliability of diagnosis results, and effectively improving the diagnosis accuracy. Description of the Drawings

[0023] Figure 1 is a schematic flowchart of the method for training the intelligent fault diagnosis model of the hydropower unit provided by the embodiment of the present application.

[0024] Figure 2 is a schematic structural diagram of the intelligent fault diagnosis model of the hydropower unit provided by the embodiment of the present application.

[0025] Figure 3 is a schematic diagram of feature extraction of the time-domain feature and frequency-domain feature of the sensing data provided by the embodiment of the present application.

[0026] Figure 4 is a schematic diagram of multi-dimensional feature fusion based on the attention mechanism provided by the embodiment of the present application.

[0027] Figure 5 is a schematic diagram of decision fusion based on information entropy provided by the embodiment of the present application.

[0028] Figure 6 It is a schematic structural diagram of an electronic device provided by an embodiment of the present application. Specific embodiments

[0029] In order to make the objectives, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0030] The terms "first" and "second" in the description and claims of this article are used to distinguish different objects, rather than to describe a specific order of the objects. For example, the first convolutional neural network and the second convolutional neural network are used to distinguish different convolutional neural networks, rather than to describe the specific order of the convolutional neural networks.

[0031] In the embodiments of the present application, words such as "exemplary" or "for example" are used to represent examples, illustrations or explanations. Any embodiment or design solution described as "exemplary" or "for example" in the embodiments of the present application should not be construed as being more preferred or having more advantages than other embodiments or design solutions. Exactly speaking, the use of words such as "exemplary" or "for example" is intended to present related concepts in a specific manner.

[0032] In the description of the embodiments of the present application, unless otherwise specified, the meaning of "a plurality of" refers to two or more. For example, a plurality of convolutional neural networks refers to two or more convolutional neural networks, etc.; a plurality of sensors refers to two or more sensors, etc.

[0033] Now, the technical terms involved in the embodiments of the present application will be introduced: One-Hot is a method for converting discrete categorical features into binary vectors.

[0034] Adam (Adaptive Moment Estimation) is a widely used adaptive learning rate optimization algorithm.

[0035] CNN (Convolutional Neural Network) is a deep learning model specifically used to process grid-structured data.

[0036] AMDC_CNN (Adaptive Multi-scale Dilated Convolutional Neural Network) is an improved convolutional neural network architecture, and its main feature is adaptive multi-scale dilated convolution.

[0037] E_CNN (Encoder CNN) is a network model that gradually compresses the spatial dimensions of the input data through convolutional layers and pooling layers to extract high-level semantic features.

[0038] D_CNN (Decoder CNN) is a network model that gradually restores the spatial resolution through transposed convolution or interpolation to generate an output that matches the input size.

[0039] AMMFN (Adaptive Multi-scale Multi-level Fusion Network) is a multi-scale and multi-level adaptive fusion network, mainly used for tasks such as image segmentation, object detection, and medical image analysis.

[0040] MsfHGNN (Multi-Scale Fusion Heterogeneous Graph Neural Network) is a multi-scale fusion heterogeneous graph neural network, mainly used for processing complex relational data.

[0041] MRSFN (Multi-Resolution and Scale-Aware Feature Fusion Network) is a multi-resolution and multi-scale feature fusion network, mainly used for high-precision image segmentation and object detection tasks.

[0042] DRCNN (Dilated Residual Convolutional Neural Network) is an improved CNN architecture that combines dilated convolution and residual connections, mainly used for dense prediction tasks.

[0043] The embodiments of the present application will now be described with reference to the accompanying drawings.

[0044] As Figure 1 shown, the method for training the intelligent fault diagnosis model of the hydropower unit provided by the embodiments of the present application includes the following steps: Step 1, data preprocessing: In this embodiment, the vibration signals at multiple measuring points of the hydropower unit are selected as the input of the model. First, the vibration monitoring data of the hydropower unit is sliced to construct a data sample set, and the size of each sample is manually adjusted according to the training effect of the model. Secondly, a fault label is added to each sample, and the typical fault categories of the hydropower unit are compiled using the One-Hot encoding method. Then, the sample data is normalized to improve the stability of the model and accelerate the convergence of the model. Finally, the processed sample data is divided into a training set and a test set.

[0045] Step 2, Construction of an intelligent fault diagnosis model for a hydroelectric generating unit: As Figure 2 shown, the intelligent fault diagnosis model for a hydroelectric generating unit includes multiple groups of convolutional neural networks, and each group of convolutional neural networks performs feature extraction and feature fusion on multiple dimensions of a path of sensing data.

[0046] Each group of convolutional neural networks includes two branch convolutional neural networks and a central convolutional neural network: Among them, the two branch convolutional neural networks respectively extract the time-domain features and frequency-domain features of the sensing data, and perform attention fusion on the shallow time-domain features and frequency-domain features obtained by the first-layer convolution to obtain fusion features; The central convolutional neural network performs feature extraction on the fusion features, and performs attention fusion on the fusion features obtained by the -th layer convolution, the time-domain features and frequency-domain features obtained by the -th layer convolution in the branch convolutional neural network, and then updates the fusion features obtained by the -th layer convolution; the central convolutional neural network finally outputs a preliminary diagnosis result; The corresponding information entropy is obtained from the sensitivity of each path of sensor data to faults, and the preliminary diagnosis results obtained from each path of sensor data are fused by decision-making based on the information entropy to obtain a final diagnosis result.

[0047] Step 3: Training of the intelligent fault diagnosis model for a hydroelectric generating unit: Input the training set into the constructed intelligent fault diagnosis model for a hydroelectric generating unit to obtain the final diagnosis result of the training set, calculate the cross entropy between the final diagnosis result and the corresponding sample One-Hot encoded label. And use the Adam optimization algorithm to update the matrix parameters of the model. Adjust the structural parameters of the model through multiple trainings, and finally obtain the intelligent fault diagnosis model for the hydroelectric generating unit.

[0048] Step 4: Testing of the intelligent fault diagnosis model for a hydroelectric generating unit: After the intelligent fault diagnosis model for a hydroelectric generating unit is trained, fix the parameters of the model, input the test set into the model, and verify the prediction accuracy rate of the model.

[0049] In this embodiment, in order to fully extract the effective information in each path of sensing signals, a multi-dimensional feature extraction method is adopted. First, the sensing signals are transformed from the time domain to the frequency domain by Fourier transform, and then convolutional neural networks are used to extract the fault features of the signals from two dimensions of the time-domain signal and the frequency-domain signal respectively, as Figure 3 shown: Convolutional neural networks can be well applied to the time series analysis of sensor data, and can effectively capture local features in the signal without being affected by their positions. During the feature extraction process, after the sensing signal is transformed from the time domain to the frequency domain through the FFT transformation, the length of the data is reduced by half. In order to achieve the effective fusion of features of different dimensions, it is necessary to perform a pooling operation with a pooling kernel size of 2 on the time domain signal additionally to keep the length of the signal space consistent.

[0050] In order to effectively fuse deep features of different dimensions, a multi-channel feature fusion method based on the attention mechanism is disclosed in this embodiment. The specific fusion process is as Figure 4 shown: First, the shallow time domain features and frequency domain features of the sensing signal are fused through attention to obtain fused features, and then the deep features of the fused features are further extracted, and the fused features with different deep convolution depths are further fused and feature extracted. Finally, a preliminary diagnosis result is obtained. During the fusion process, the attention mechanism is used to weight the importance of features in different channels. Specifically: The feature vectors of each feature to be fused are compressed through a pooling operation into a single numerical value, and all the compressed values form a compressed vector; average pooling is used for compression in this embodiment.

[0051] The attention weights of each feature to be fused are calculated through the similarity calculation of the compressed vector; Based on the attention weights, each feature to be fused is weighted and summed to obtain the fused features.

[0052] Among them, fusing the shallow time domain features and frequency domain features of the sensing signal through attention to obtain the fused features is specifically:

[0053] In the formula: represents the first-layer fused feature obtained from the sensing data; , respectively represent the attention weights of the time domain feature and the frequency domain feature; the subscript represents the th path of sensing data.

[0054] Among them, fusing the fused feature obtained from the th layer of convolution, the time domain feature and the frequency domain feature obtained from the th layer of convolution in the branch convolutional neural network through attention is specifically:

[0055] In the formula, is the attention weight of the fused feature; represents the fused feature after fusion in the current convolutional layer.

[0056] Using multi-channel sensor signals for collaborative fault diagnosis can obtain more fault characteristics of the unit. However, the sensitivity of each sensor to faults is different under different fault conditions, and the importance of each sensor signal should be considered in collaborative fault diagnosis. This embodiment discloses a decision fusion method based on the information entropy of sensor samples. By calculating the information entropy of the input samples of the model, the weight distribution of the diagnosis results of each sensor is carried out, which improves the reliability of multi-sensor collaborative fault diagnosis. The specific fusion process is as follows: Figure 5 As shown in the figure, based on the information entropy, the preliminary diagnosis results obtained from the data of each sensor are fused to obtain the final diagnosis result. Specifically: Obtain the weight ratio of each information entropy among all information entropies; The weight ratio of each information entropy is:

[0057] where, is the information entropy corresponding to the data of the -th channel sensor, is the total number of sensor channels; Normalize the weight ratio of each information entropy to obtain the fusion weight corresponding to the data of each sensor. The fusion weight corresponding to the data of the -th channel sensor

[0058] where, is the activation function; Use the fusion weights of each channel of sensor data to perform weighted summation on the preliminary diagnosis results obtained from each channel of sensor data to obtain the final diagnosis result. The final diagnosis result is:

[0059] where, represents the probability of belonging to the -th type of fault, is the preliminary diagnosis result of each channel of sensor data.

[0060] Experimental verification: The dataset comes from the data generated during the daily operation of a hydropower station in China. As a typical rotating machine, the hydro-generator unit (HU) has a wide range of applications in the energy industry. The structure of the HU is complex and the operating environment has strong noise interference. Therefore, this dataset is selected to verify the superiority and robustness of the proposed method in a real industrial environment. The experimental verification is mainly carried out using the vibration data in the X direction of the water guide bearing of the unit shafting and the Y direction of the water guide bearing. These data are sampled at equal intervals. A total of four different fault data are used, namely normal condition, excessive vibration of the stator core, flow passage blockage fault, and mass imbalance fault. Each sample contains 1024 data points, and there are 50 samples for each fault type.

[0061] The experimental model has the same structure as the model proposed in the embodiment of the present application. Each channel consists of a central CNN and two branch CNNs, and the structures of the two branch CNNs are the same. The Adam optimizer is used for model training. During the training process, the initial learning rate is set to 0.001 and the number of training epochs is set to 100. For the two datasets, 20% of the samples are used as the test set and 80% of the samples are used as the training set.

[0062] Verification of the advantages of time-frequency domain feature fusion: To verify the correctness and effectiveness of the proposed time-frequency domain feature fusion method, experiments are carried out using the HU dataset. The time-domain signal, frequency-domain signal, and time-frequency domain signal of the dataset are respectively used as the input of the recognition model, and the model used is the single-channel form of the proposed model.

[0063] In the experimental results: The diagnostic accuracy of using the time-domain signal or the frequency-domain signal alone is higher than 83%, which indicates that both the time-domain signal and the frequency-domain signal contain information reflecting the equipment state and can diagnose the equipment state to a certain extent.

[0064] When the time-domain signal and the frequency-domain signal are jointly used as the input of the diagnostic model, the model accuracy reaches 96%, which is 12.25% and 10% higher than the diagnostic accuracies of the time-domain signal and the frequency-domain signal respectively.

[0065] It can be found that the diagnostic effects of the time-domain signal and the frequency-domain signal are different, but the best diagnostic results are obtained when the time-domain and frequency-domain signals are fused. This is because: when the time-domain and frequency-domain signals are fused and used as the input of the diagnostic model, the implicit information in the monitoring signal is more fully utilized, and the characteristics of both the time-domain signal and the frequency-domain signal are considered, effectively improving the accuracy of the diagnostic model.

[0066] Verification of the advantages of multi-sensor data feature fusion: In this experiment, the advantages of using multi-sensor data compared to single-sensor data were verified. In the HU dataset, experiments were conducted using the water-guide X-directional swing data, the water-guide Y-directional swing data, and the fused data of both. The experimental model has the same structure as the model proposed in the embodiments of this application. When using the fused data, the model uses the decision fusion strategy proposed in the embodiments of this application. To minimize the impact of randomness on the final results, ten repeated experiments were conducted for each method. In the experimental results, the diagnostic accuracy of the fused data increased by 1.75% and 4.75% respectively compared to the water-guide X-directional and Y-directional swing data. In summary, fusing different channel signals can effectively improve the accuracy of the fault diagnosis model. This is because the differences in sensors result in different reflections of fault characteristics by the signals of each channel. Fusing the characteristics of multiple channel signals can avoid the loss of fault characteristics and effectively improve the accuracy of the diagnosis model.

[0067] Verification of the advantages of the decision fusion method based on information entropy: Compare with several common decision fusion strategies respectively: Strategy 1, a fusion strategy of averaging and weighting all outputs: Strategy 2, a fusion strategy of concatenating all outputs and using a fully connected network: Conduct experiments on each of the above strategies separately, and record the accuracy of each strategy. The results are shown in Table 1:

[0068] Table 1 It can be clearly seen from the results that when different decision fusion strategies are adopted, the classification accuracy of the model decreases to varying degrees compared with the proposed strategy. The classification accuracy of using the decision fusion strategy of multi-channel averaging and weighting is slightly higher than that of using the decision fusion strategy of the fully connected layer, but the classification accuracy of the model under both strategies is relatively low. This is because neither of the two strategies considers the importance of channel signals to the diagnostic results, resulting in poor decision fusion results. The decision fusion strategy proposed in this paper fully considers the association between different faults and each channel sensor, judges the importance of the channel output results through the complexity of the input samples, makes the final fusion decision more reasonable, and improves the classification accuracy of the model.

[0069] Verification based on the hydropower unit fault dataset: Experiments were conducted using the hydropower unit fault dataset to verify the applicability and reliability of the model in a real industrial system. The final results are shown in Table 2:

[0070] Table 2 The experimental results show that the model proposed in this application has obvious advantages in the state recognition of hydropower units. It can extract and fuse the deep features of multiple sensors and multiple dimensions from the monitoring data of rotating machinery in engineering practice, and has a higher diagnostic accuracy than other models.

[0071] Based on the method in the above embodiment, an electronic device is provided in an embodiment of this application. As Figure 6 shown, the electronic device includes: a processor, a communications interface, a memory, and a communication bus. Among them, the processor, the communications interface, and the memory complete mutual communication through the communication bus. The processor can call the logical instructions in the memory to execute the method in the above embodiment.

[0072] In addition, when the logical instructions in the above memory are implemented in the form of a software functional unit and sold or used as an independent product, they can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in each embodiment of this application.

[0073] Based on the method in the above embodiment, a computer-readable storage medium is provided in an embodiment of this application. The computer-readable storage medium stores a computer program. When the computer program runs on the processor, the processor is enabled to execute the method in the above embodiment.

[0074] Based on the method in the above embodiment, a computer program product is provided in an embodiment of this application. When the computer program product runs on the processor, the processor is enabled to execute the method in the above embodiment.

[0075] It can be understood that the processor in the embodiment of this application can be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application specific integrated circuits (ASICs), field programmable gate arrays (FPGAs), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. The general-purpose processor can be a microprocessor or any conventional processor.

[0076] The method steps in the embodiments of the present application can be implemented in a hardware manner or by a processor executing software instructions. The software instructions can be composed of corresponding software modules, and the software modules can be stored in a random access memory (RAM), flash memory, read-only memory (ROM), programmable ROM (PROM), erasable PROM (EPROM), electrically erasable PROM (EEPROM), registers, hard disk, removable hard disk, CD-ROM, or any other form of storage medium well-known in the art. An exemplary storage medium is coupled to the processor, enabling the processor to read information from the storage medium and write information to the storage medium. Of course, the storage medium can also be a component of the processor. The processor and the storage medium can be located in an ASIC.

[0077] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the processes or functions described in the embodiments of the present application are generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium or transmitted through the computer-readable storage medium. The computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center in a wired manner (such as coaxial cable, optical fiber, digital subscriber line (DSL)) or a wireless manner (such as infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that the computer can access or a data storage device such as a server, data center, etc. that includes one or more integrated available media. The available medium can be a magnetic medium (such as a floppy disk, hard disk, magnetic tape), an optical medium (such as a DVD), or a semiconductor medium (such as a solid state disk (SSD)), etc.

[0078] It can be understood that the various numerical numbers involved in the embodiments of the present application are only for the convenience of description and are not used to limit the scope of the embodiments of the present application.

[0079] Those skilled in the art can easily understand that the above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent replacements, and improvements made within the spirit and principle of the present application shall be included within the protection scope of the present application.

Claims

1. A training method for an intelligent fault diagnosis model of a hydropower unit, characterized in that Including: Collecting multi-channel sensing data of a hydropower unit and preprocessing the sensing data; Constructing two branch convolutional neural networks to extract the time-domain features and frequency-domain features of the sensing data respectively, and performing attention fusion on the time-domain features and frequency-domain features obtained by the first-layer convolution to obtain fused features; Construct a central convolutional neural network to extract the fusion features, and The fusion features obtained by layer convolution and the first branch convolutional neural network The time domain features and frequency domain features obtained by the layer convolution are updated after attention fusion. The fusion features obtained by layer convolution; the central convolutional neural network finally outputs the preliminary diagnosis results; Obtaining the corresponding information entropy from the sensitivity of each sensor data to the fault, performing decision fusion on the preliminary diagnosis results obtained from each sensor data based on the information entropy to obtain the final diagnosis result; using the gap between the final diagnosis result and the actual label as the loss function to train the parameters of each convolutional neural network, and finally obtaining the intelligent fault diagnosis model of the hydropower unit.

2. The training method according to claim 1, wherein Performing attention fusion specifically as follows: Compressing the feature vectors of each feature to be fused through pooling, and forming a compressed vector from all the compression values; Calculating the similarity of the compressed vector to obtain the attention weights of each feature to be fused; Performing weighted summation on each feature to be fused based on the attention weights to obtain the fused features.

3. The training method according to claim 1, wherein Performing decision fusion on the preliminary diagnosis results obtained from each sensor data based on the information entropy to obtain the final diagnosis result, specifically as follows: Obtaining the weight ratio of each information entropy among all the information entropies; Normalizing the weight ratio of each information entropy to obtain the fusion weights corresponding to each sensor data; Performing weighted summation on the preliminary diagnosis results obtained from each sensor data using the fusion weights of each sensor data to obtain the final diagnosis result.

4. The training method according to claim 3, wherein Weight ratio of each information entropy is as follows: Among them, is the information entropy corresponding to the road sensor data, is the total number of roads of the sensor; The fusion weight corresponding to the road sensor data is as follows: Among them, is the activation function; Final diagnosis result is: Among them, represents the probability of belonging to the category of faults, which is the preliminary diagnosis result of each path of sensing data.

5. The training method according to claim 1, characterized in that, Obtaining the corresponding information entropy from the sensitivity of each sensor data to the fault, specifically as follows: Among them, is the information entropy corresponding to the -th road sensor data; represents the possible probability of the -th data in the -th road sensor data; is the length of the sensor data; is the preset logarithm base.

6. The training method according to claim 1, wherein Performing Fourier transform on the sensing data to convert the time-domain sensing data into frequency-domain sensing data, obtaining time-domain features from the time-domain sensing data, and obtaining frequency-domain features from the frequency-domain sensing data.

7. The training method according to claim 1, characterized in that The branch convolutional neural network includes four convolutional-pooling layers; the central convolutional neural network includes three convolutional-pooling layers and one fully connected layer.

8. The training method according to claim 1, wherein Measuring the gap between the final diagnosis result and the actual label through cross-entropy, and the cross-entropy is: Among them, is the actual label, is the final diagnosis result, and the subscript represents the th in the distribution, and the base of the logarithm is a preset value.

9. An intelligent fault diagnosis method for hydropower units based on multi-sensor data fusion, characterized in that, Including: Collecting multi-channel sensing data of a hydropower unit and preprocessing the sensing data; Obtaining the time-domain signal and frequency-domain signal of each sensing data; Inputting the time-domain signal and frequency-domain signal of each sensing data into the intelligent fault diagnosis model of the hydropower unit according to any one of claims 1-8 respectively; The intelligent fault diagnosis model of the hydropower unit outputs the fault diagnosis result of the hydropower unit.

10. An electronic device, characterized in that, Including: At least one memory for storing a computer program; At least one processor for executing the program stored in the memory. When the program stored in the memory is executed, the processor is used to execute the training method according to any one of claims 1-8, or the processor is used to execute an intelligent fault diagnosis method for a hydropower unit based on multi-sensor data fusion according to claim 9.

Images