Hydropower station governor operation fault identification method based on image deep learning

Abstract

The invention discloses a hydropower station governor operation fault identification method based on image deep learning, and belongs to the technical field of governor fault identification, and the method comprises the following steps: collecting governor signals, carrying out the preprocessing, and constructing an image data set; constructing a hydropower station governor operation fault identification model comprising two convolution layers, a composite attention module, a normalization layer, a pooling layer and a Dropout layer; inputting the image data set into the hydropower station governor operation fault identification model to obtain an identification result; through the method, the operation fault of the hydropower station governor can be effectively identified, and the accuracy is high.

Description

Technical Field

[0001] The present invention relates to the technical field of speed regulator fault identification, and in particular to a method for identifying operating faults of a speed regulator of a hydropower station based on image deep learning. Background Art

[0002] The speed governor of a hydropower station is one of the most important auxiliary control devices of a hydro-turbine generator set. The quality of its operation directly determines the safety and stable operation of the unit. The main task of the speed governor of a hydropower station is to maintain the rated speed of the hydro-turbine generator set by adjusting the output power of the hydro-turbine generator set, which includes responding to changes in the grid frequency or changes in independently set reference values, and automatically adjusting the hydraulic load of the turbine to meet the power system's requirements for frequency stability; the speed governor of a hydropower station can also participate in the automatic power generation control of the grid, and automatically distribute and adjust the power of the grid.

[0003] As the load of the power grid system changes, the frequency of the power grid system will continue to change. The turbine speed governor needs to change the guide vane opening (or needle valve stroke) accordingly to restore and maintain the unit speed to the rated value or a predetermined value. Therefore, the hydropower station speed governor can adjust the flow entering the turbine according to the increase or decrease trend of the generator load, so that the output of the turbine can adapt to the external load, and the speed can also be kept stable within the rated value range, thereby ensuring that the frequency (f = 50Hz) does not change or changes within the allowed range. Therefore, the hydropower station speed governor is inseparable from whether the hydropower station can generate electricity normally. It is very important to monitor the health of the speed governor. Once the speed governor fails, it will directly affect the safe and stable operation of the turbine unit and even the power system.

[0004] However, there are currently several problems with the diagnosis of the speed regulator operation signal:

[0005] 1) The current diagnostic methods basically start from the perspectives of time domain and frequency domain, convert the time-course signal into a characteristic value, analyze the fault from a macro perspective, and ignore the influence of time on the signal at the micro level; that is: time domain analysis extracts features by analyzing the changes in the signal waveform on the time axis, but cannot fully capture the characteristics of the signal at the micro level, especially those rapid changes or local anomalies that occur in a short period of time; frequency domain analysis converts the signal from the time domain to the frequency domain, providing the distribution of the signal's frequency components, and also ignores the time information.

[0006] 2) With the continuous increase of time series data and the deepening and expansion of analysis purposes, it is becoming increasingly difficult to improve the accuracy of time series classification problems using traditional methods. Summary of the invention

[0007] In view of the above-mentioned prior art, the present invention provides a method for identifying operating faults of a hydropower station speed governor based on image deep learning, which mainly solves the technical problems existing in the above-mentioned background technology.

[0008] To achieve the above object, the technical solution of the embodiment of the present invention is implemented as follows:

[0009] A method for identifying operating faults of a hydropower station speed regulator based on image deep learning comprises the following steps:

[0010] Collect speed governor signals, perform preprocessing, and construct image data sets;

[0011] Construct a hydropower station governor operation fault recognition model including two convolutional layers, a composite attention module, a normalization layer, a pooling layer, and a dropout layer;

[0012] The image data set is input into the hydropower station speed governor operation fault recognition model to obtain a recognition result.

[0013] Optionally, the speed regulator signal is collected, preprocessed, and an image data set is constructed, specifically:

[0014] A speed regulator signal is collected, the speed regulator signal is subjected to median filtering, the one-dimensional signal obtained after the median filtering is converted into a two-dimensional image signal, and an image data set is constructed based on the two-dimensional image signal.

[0015] Optionally, the two convolutional layers are used to extract features from the image data in the image data set;

[0016] The composite attention module is used to capture local features and global features of the images output by the two convolutional layers;

[0017] The normalization layer is used to normalize the feature image output by the composite attention module;

[0018] The pooling layer is used to reduce the dimension of the feature image output by the normalization layer;

[0019] The Dropout layer is used to regularize the feature image output by the pooling layer to prevent overfitting.

[0020] Optionally, the composite attention module includes a first multi-head attention mechanism, a fully connected layer, a second multi-head attention mechanism, and a third multi-head attention mechanism;

[0021] The first multi-head attention mechanism is connected to the two convolutional layers to learn local features of the feature graph output by the two convolutional layers;

[0022] The fully connected layer is connected to the first multi-head attention mechanism, and is used to integrate the local features extracted by the first multi-head attention mechanism to form a more global feature representation;

[0023] The second multi-head attention mechanism is connected to the fully connected layer to provide global information;

[0024] The third multi-head attention mechanism is connected to the second multi-head attention mechanism for learning local features.

[0025] Optionally, the first multi-head attention mechanism includes a 3x3 convolution layer, a normalization layer, a residual layer, and a 1x1 convolution layer.

[0026] Optionally, the second multi-head attention mechanism includes a 1x1 convolutional layer, a normalization layer, a Transformer and a 1x1 convolutional layer.

[0027] Optionally, the Transformer includes an encoder-decoder structure, the encoder is composed of multiple first residual blocks and multiple second residual blocks, the first residual blocks include multi-head attention and normalization layers, and the second residual blocks include a bit-by-bit feedforward network and a normalization layer; the first residual blocks and the second residual blocks are connected in a residual connection manner;

[0028] The decoder includes a plurality of third residual blocks, a plurality of fourth residual blocks, a plurality of fifth residual blocks and a fully connected layer, the third residual block includes a hidden multi-head attention and a normalization layer, the fourth residual block includes a multi-head attention and a normalization layer, the fifth residual block includes a bit-by-bit feedforward network and a normalization layer, the third residual block, the fourth residual block and the fifth residual block are connected in a residual connection manner, and the fifth residual block is connected to the fully connected layer;

[0029] Optionally, the third multi-head attention mechanism includes a 3x3 convolutional layer, a normalization layer, a residual layer, and a 1x1 convolutional layer.

[0030] Optionally, the method further includes: using a cross entropy loss function to train the hydropower station governor operation fault identification model; the cross entropy loss function expression is:

[0031] L = -[ylog(p)+(1-y)log(1-p)]

[0032] Among them, y is the true value and p is the probability of correct prediction.

[0033] The beneficial effect of the present invention lies in: the present invention provides a method for identifying operating faults of a hydropower station speed governor based on image deep learning, which first converts the collected time domain signal or frequency domain signal of the speed governor into a time-course image to construct an image data set, and then the image data set can be trained through a hydropower station speed governor operating fault recognition model, specifically, the image data set is first subjected to a convolution layer to extract features, and then the image features output by the convolution layer are input into a composite attention module, the composite attention module can effectively fuse features through a multi-head attention mechanism and a fully connected layer, and realize multi-level and multi-scale capture of image features, and then the image features processed by the composite attention module are input into a normalization layer, a pooling layer and a Dropout layer, and finally a classification result is obtained to realize the identification of operating faults of a hydropower station speed governor. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 A schematic diagram of a flow chart of a method for identifying operating faults of a hydropower station governor based on image deep learning provided in an embodiment of the present invention;

[0035] Figure 2 A schematic diagram of a hydropower station governor operation fault identification model provided in an embodiment of the present invention;

[0036] Figure 3 A schematic diagram of a Transformer structure provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0037] The technical solution of the present invention is further elaborated in detail below in conjunction with the drawings and specific embodiments of the specification. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by technicians in the technical field of the present invention. The terms used in the specification of the present invention herein are only for the purpose of describing specific embodiments and are not intended to limit the present invention. In the following description, the expression "some embodiments" is related to a subset of all possible embodiments, but it should be understood that "some embodiments" can be the same subset or different subsets of all possible embodiments, and can be combined with each other without conflict.

[0038] In the following description, a large number of specific details are provided to provide a more thorough understanding of the present invention. However, it is apparent to those skilled in the art that the present invention can be implemented without one or more of these details. In other examples, in order to avoid confusion with the present invention, some technical features well known in the art are not described.

[0039] It should be understood that the present invention can be implemented in different forms and should not be interpreted as being limited to the embodiments presented here. On the contrary, providing these embodiments will make the disclosure thorough and complete, and the scope of the present invention will be fully conveyed to those skilled in the art. And the purpose of the terms used herein is only to describe specific embodiments and is not intended to be a limitation of the present invention. When used herein, the singular forms of "one", "one" and "said / the" are also intended to include plural forms, unless the context clearly indicates another way. It should also be understood that the terms "compose" and / or "include" when used in this specification determine the presence of the features, integers, steps, operations, elements and / or parts, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, parts and / or groups. When used herein, the term "and / or" includes any and all combinations of the relevant listed items.

[0040] It should also be noted that when an element is referred to as being "fixed to" another element, it may be directly on the other element or there may be a central element. When an element is considered to be "connected to" another element, it may be directly connected to the other element or there may be a central element at the same time. The terms "vertical", "horizontal", "inside", "outside", "left", "right" and similar expressions used herein are for illustrative purposes only and are not intended to be the only implementation method.

[0041] In order to fully understand the present invention, a detailed structure will be proposed in the following description to illustrate the technical solution proposed by the present invention. The optional embodiments of the present invention are described in detail as follows, but in addition to these detailed descriptions, the present invention may also have other implementations.

[0042] Example

[0043] Please refer to the attached Figure 1 and attached Figure 2 The present application provides a method for identifying operating faults of a hydropower station speed regulator based on image deep learning, comprising the following steps:

[0044] Collect speed governor signals, perform preprocessing, and construct image data sets;

[0045] Construct a hydropower station governor operation fault recognition model including two convolutional layers, a composite attention module, a normalization layer, a pooling layer, and a dropout layer;

[0046] The image data set is input into the hydropower station speed governor operation fault recognition model to obtain a recognition result.

[0047] As an optional implementation, the speed regulator signal is collected, preprocessed, and an image data set is constructed, specifically:

[0048] A speed regulator signal is collected, the speed regulator signal is subjected to median filtering, the one-dimensional signal obtained after the median filtering is converted into a two-dimensional image signal, and an image data set is constructed based on the two-dimensional image signal.

[0049] Specifically, after collecting the speed regulator signal, the time domain signal or frequency domain signal in the speed regulator signal is extracted, the time domain signal or frequency domain signal is subjected to median filtering to remove the noise in the signal, and then converted into an image signal, thereby obtaining an image data set; by converting the time domain signal or frequency domain signal into an image signal, the characteristics of the signal can be analyzed at multiple resolution levels, which is conducive to processing signals with rich local characteristics, and at the same time, the influence of time on the signal in the microscopic level can be emphasized;

[0050] The formula for median filtering is:

[0051] y[t]=1 / 2*x[t]+1 / 4*(x[t-1]+x[t+1])

[0052] Where t is a certain moment, x[t] is the input value at moment t, and y[t] is the output value at the corresponding moment t.

[0053] As an optional implementation, the two convolutional layers are used to extract features from the image data in the image data set;

[0054] The composite attention module is used to capture local features and global features of the images output by the two convolutional layers;

[0055] The normalization layer is used to normalize the feature image output by the composite attention module;

[0056] The pooling layer is used to reduce the dimension of the feature image output by the normalization layer;

[0057] The Dropout layer is used to regularize the feature image output by the pooling layer to prevent overfitting;

[0058] Specifically, the image dataset is divided into a training set and a dataset in a ratio of 8:2; the image data of the training set is input into two 3*3 convolutional layers in sequence to extract features of the image. The first 3*3 convolutional layer can extract low-level features of the image, and the second 3*3 convolutional layer extracts more complex features based on the first 3*3 convolutional layer. The features output by the second 3*3 convolutional layer are then sent to the composite attention module. The composite attention module can enhance the understanding of the image context by the hydropower station governor operation fault recognition model, combine global information and local information, emphasize important features and suppress unimportant features, and then The processed image features are input into the normalization layer. The normalization layer reduces the variance of each layer input through normalization processing, improves the generalization ability of the hydropower station speed governor operation fault recognition model, and then passes through the pooling layer to select the most significant features. Finally, the neurons are randomly discarded through the Dropout layer to prevent overfitting. The consistency of the hydropower station speed governor operation fault recognition model is maintained by continuously adjusting the proportion of neuron outputs. Finally, through forward propagation, the output layer will output a probability vector, in which each element represents the probability of belonging to a certain category. The category with the largest probability is selected as the prediction result, thereby obtaining the hydropower station speed governor operation fault recognition result.

[0059] As an optional implementation, the composite attention module includes a first multi-head attention mechanism, a fully connected layer, a second multi-head attention mechanism, and a third multi-head attention mechanism;

[0060] The first multi-head attention mechanism is connected to the two convolutional layers to learn local features of the feature graph output by the two convolutional layers;

[0061] The fully connected layer is connected to the first multi-head attention mechanism, and is used to integrate the local features extracted by the first multi-head attention mechanism to form a more global feature representation;

[0062] The second multi-head attention mechanism is connected to the fully connected layer to provide global information;

[0063] The third multi-head attention mechanism is connected to the second multi-head attention mechanism for learning local features.

[0064] Specifically, the first multi-head attention mechanism includes a 3x3 convolution layer, a normalization layer, a residual layer, and a 1x1 convolution layer.

[0065] The second multi-head attention mechanism includes a 1x1 convolution layer, a normalization layer, a Transformer and a 1x1 convolution layer.

[0066] Please refer to the attached Figure 3, the Transformer includes an encoder-decoder structure, the encoder is composed of multiple first residual blocks and multiple second residual blocks, the first residual block includes a multi-head attention and a normalization layer, and the second residual block includes a bit-by-bit feedforward network and a normalization layer; the first residual block and the second residual block are connected in a residual connection manner;

[0067] The decoder includes a plurality of third residual blocks, a plurality of fourth residual blocks, a plurality of fifth residual blocks and a fully connected layer, the third residual block includes a hidden multi-head attention and a normalization layer, the fourth residual block includes a multi-head attention and a normalization layer, the fifth residual block includes a bit-by-bit feedforward network and a normalization layer, the third residual block, the fourth residual block and the fifth residual block are connected in a residual connection manner, and the fifth residual block is connected to the fully connected layer;

[0068] The third multi-head attention mechanism includes a 3x3 convolution layer, a normalization layer, a residual layer, and a 1x1 convolution layer.

[0069] As an optional implementation, the method further includes: using a cross entropy loss function to train the hydropower station governor operation fault identification model; the cross entropy loss function is expressed as:

[0070] L = -[ylog(p)+(1-y)log(1-p)]

[0071] Among them, y is the true value and p is the probability of correct prediction.

[0072] Specifically, the cross entropy loss function optimizes the hydropower station speed governor operation fault identification model by calculating the difference between the probability distribution predicted by the hydropower station speed governor operation fault identification model and the probability distribution of the true label, helping the hydropower station speed governor operation fault identification model to better learn fault characteristics and improve the accuracy and efficiency of fault identification.

[0073] The above are only specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. The protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A method for identifying operating faults of a hydropower station governor based on image deep learning, characterized in that: The following steps are involved: Collect speed governor signals, perform preprocessing, and construct image data sets; Construct a hydropower station governor operation fault recognition model including two convolutional layers, a composite attention module, a normalization layer, a pooling layer, and a dropout layer; The image data set is input into the hydropower station speed governor operation fault recognition model to obtain a recognition result.

2. According to the method for identifying operating faults of a hydropower station speed regulator based on image deep learning in claim 1, it is characterized in that: Collect the speed regulator signal, perform preprocessing, and construct the image data set, specifically: A speed regulator signal is collected, the speed regulator signal is subjected to median filtering, the one-dimensional signal obtained after the median filtering is converted into a two-dimensional image signal, and an image data set is constructed based on the two-dimensional image signal.

3. According to the method for identifying operating faults of a hydropower station speed regulator based on image deep learning in claim 1, it is characterized in that: The two convolutional layers are used to extract features from the image data in the image data set; The composite attention module is used to capture local features and global features of the images output by the two convolutional layers; The normalization layer is used to normalize the feature image output by the composite attention module; The pooling layer is used to reduce the dimension of the feature image output by the normalization layer; The Dropout layer is used to regularize the feature image output by the pooling layer to prevent overfitting.

4. According to the method of identifying operating faults of a hydropower station speed governor based on image deep learning in claim 3, it is characterized in that: The composite attention module includes a first multi-head attention mechanism, a fully connected layer, a second multi-head attention mechanism, and a third multi-head attention mechanism; The first multi-head attention mechanism is connected to the two convolutional layers to learn local features of the feature graph output by the two convolutional layers; The fully connected layer is connected to the first multi-head attention mechanism, and is used to integrate the local features extracted by the first multi-head attention mechanism to form a more global feature representation; The second multi-head attention mechanism is connected to the fully connected layer to provide global information; The third multi-head attention mechanism is connected to the second multi-head attention mechanism for learning local features.

5. According to the method for identifying operating faults of a hydropower station speed governor based on image deep learning in claim 4, it is characterized in that: The first multi-head attention mechanism includes a 3x3 convolution layer, a normalization layer, a residual layer, and a 1x1 convolution layer.

6. According to the method of identifying operating faults of a hydropower station speed governor based on image deep learning in claim 4, it is characterized in that: The second multi-head attention mechanism includes a 1x1 convolution layer, a normalization layer, a Transformer and a 1x1 convolution layer.

7. The method for identifying operating faults of a hydropower station speed regulator based on image deep learning according to claim 6 is characterized in that: The Transformer includes an encoder-decoder structure, the encoder is composed of a plurality of first residual blocks and a plurality of second residual blocks, the first residual blocks include a multi-head attention and a normalization layer, and the second residual blocks include a bit-by-bit feedforward network and a normalization layer; the first residual blocks and the second residual blocks are connected in a residual connection manner; The decoder includes multiple third residual blocks, multiple fourth residual blocks, multiple fifth residual blocks and a fully connected layer, the third residual block includes hidden multi-head attention and a normalization layer, the fourth residual block includes multi-head attention and a normalization layer, the fifth residual block includes a bit-by-bit feedforward network and a normalization layer, the third residual block, the fourth residual block and the fifth residual block are connected by residual connection, and the fifth residual block is connected to the fully connected layer.

8. The method for identifying operating faults of a hydropower station speed governor based on image deep learning according to claim 4 is characterized in that: The third multi-head attention mechanism includes a 3x3 convolution layer, a normalization layer, a residual layer, and a 1x1 convolution layer.

9. The method for identifying operating faults of a hydropower station speed regulator based on image deep learning according to claim 1 is characterized in that: Also includes: The cross entropy loss function is used to train the hydropower station governor operation fault identification model; the cross entropy loss function expression is: L = -[ylog(p)+(1-y)log(1-p)] Among them, y is the true value and p is the probability of correct prediction.

Images