Hydroelectric generating set excitation power unit fault diagnosis method based on Attention-CNN and fault knowledge base

Abstract

The invention discloses a hydroelectric generating set excitation power unit fault diagnosis method based on Attention-CNN and a fault knowledge base, and the method comprises the steps: combining normal data and fault simulation data, associating with a corresponding type code, and constructing a single-bridge excitation power unit fault knowledge base; dividing waveform data in the knowledge base into a training set and a test set according to a preset proportion, and inputting the waveform of the excitation voltage and the waveform of the AC side voltage as two-dimensional vectors into an Attention-CNN model for training and testing; and excitation voltage and alternating current side voltage waveforms during the operation period of the excitation system are collected in real time, the excitation voltage and alternating current side voltage waveforms are converted into two-dimensional vectors, the two-dimensional vectors are input into the trained Attention-CNN model, and type codes are output. According to the method, the fault diagnosis efficiency and accuracy of the single-bridge excitation system of the small-sized hydroelectric generating set are improved by using a deep learning technology, automatic identification of a complex fault mode is realized, and the operation stability and the fault processing capability of the small-sized hydroelectric generating set are improved.

Description

Technical Field

[0001] The invention relates to a fault diagnosis method for an excitation power unit of a hydropower unit based on Attention-CNN and a fault knowledge base, belonging to the technical field of fault diagnosis of a single-bridge excitation power unit of a small hydropower unit. Background Art

[0002] Small hydropower units are an important part of modern hydropower plants, and the stable operation of their generators depends on the accurate control provided by the excitation system. The main function of the excitation system is to provide the required excitation current for the rotor of the small hydropower generator to maintain the voltage stability of the generator and the reasonable distribution of reactive power. In modern power systems, the reliable operation of the excitation system is crucial to the safety and stability of small hydropower units and the entire power grid.

[0003] The single-bridge excitation power unit is an important part of the excitation system. It converts AC power into DC power through a three-phase bridge rectifier circuit to provide a stable DC excitation current for the generator rotor. However, with the long-term operation of the equipment, affected by the external environment and the aging of components, key components in the single-bridge power unit, such as thyristors, are prone to failures, such as open circuit failures and short circuit failures. If the fault is not diagnosed and handled in time, it may lead to unstable unit operation and even cause more serious system accidents, affecting the reliability of the entire power generation system.

[0004] The traditional single-bridge excitation system fault diagnosis method determines the fault type by manually analyzing the fault waveform or using frequency domain and time domain analysis methods. When dealing with various faults caused by long-term operation of equipment, the traditional fault diagnosis method often relies on limited fault data or manual experience, resulting in the accuracy of fault identification and response speed being difficult to meet actual needs.

[0005] In view of this, the present invention is proposed. Summary of the invention

[0006] The present invention provides a fault diagnosis method for the excitation power unit of a hydropower unit based on Attention-CNN and a fault knowledge base. The method constructs a knowledge base covering a variety of fault modes and their waveform characteristics. Based on the Attention-CNN model, the method effectively solves the problems caused by incomplete construction of the fault knowledge base, missing fault data, low diagnostic accuracy, poor real-time performance, and inability to effectively deal with complex fault modes.

[0007] The technical solution of the present invention is:

[0008] According to a first aspect of the present invention, a method for diagnosing faults in an excitation power unit of a hydropower unit based on Attention-CNN and a fault knowledge base is provided, comprising the following steps: constructing a single-bridge excitation power unit fault knowledge base; dividing waveform data in the knowledge base into a training set and a test set according to a preset ratio, and inputting the excitation voltage and AC side voltage waveforms as two-dimensional vectors into an Attention-CNN model for training and testing; collecting the excitation voltage and AC side voltage waveforms during the operation of the excitation system in real time, converting them into two-dimensional vectors, inputting them into a trained Attention-CNN model, and outputting a type code.

[0009] Furthermore, the construction of the single-bridge excitation power unit fault knowledge base is specifically as follows: simulating various fault modes in the single-bridge excitation power unit through a simulation platform, generating two-dimensional vector forms of excitation voltage and AC side voltage waveform data under different fault modes as fault simulation data; collecting data of the single-bridge excitation power unit in a normal state from the electric field site, and converting it into a two-dimensional vector form of excitation voltage and AC side voltage waveform data as normal data; combining normal data and fault simulation data, and associating them with corresponding type codes to construct a single-bridge excitation power unit fault knowledge base.

[0010] Furthermore, the type code includes a normal operating condition code and four types of failure mode codes, specifically: a first type code, a second type code, a third type code, a fourth type code, and a fifth type code.

[0011] Furthermore, the failure modes include single thyristor open circuit failure, two thyristor open circuit failures in the same phase, two thyristor open circuit failures on different polarity sides, and two thyristor open circuit failures on different polarity sides in different phases.

[0012] Furthermore, the Attention-CNN model is constructed by introducing Attention on the basis of the CNN model.

[0013] Furthermore, the Attention-CNN model specifically includes: a convolution layer, which is used to extract local and global features of waveform data through multiple convolution kernels; a self-attention layer, which is used to weight features; a pooling layer, which is used for dimensionality reduction; and a fully connected layer, which is used to associate features with type codes in a knowledge base and output classification results.

[0014] Furthermore, there are two convolution layers, the convolution kernel size of each convolution layer is 3×3, and the activation function of the convolution layer is the ReLU activation function.

[0015] Furthermore, the pooling layer uses a 2×2 maximum pooling operation.

[0016] Furthermore, the learning rate is set to 0.001; the optimizer adopts the Adam optimizer, and the cross entropy loss function is used for model training.

[0017] According to a second aspect of the present invention, a hydropower unit excitation power unit fault diagnosis system based on Attention-CNN and a fault knowledge base is provided, comprising:

[0018] A building module for building a single-bridge excitation power unit fault knowledge base;

[0019] A training module is used to divide the waveform data in the knowledge base into a training set and a test set according to a preset ratio, and the excitation voltage and AC side voltage waveforms are input into the Attention-CNN model as two-dimensional vectors for training and testing;

[0020] The diagnostic module is used to collect the excitation voltage and AC side voltage waveforms during the operation of the excitation system in real time, convert them into two-dimensional vectors to input the trained Attention-CNN model, and output type coding.

[0021] Furthermore, the fault diagnosis system is integrated with a SCADA data acquisition system; wherein, during the operation of the generator set, the SCADA data acquisition system collects excitation voltage and AC side voltage waveform data in real time and transmits them to the fault diagnosis system.

[0022] The beneficial effects of the present invention are as follows: in the field of single-bridge excitation system fault diagnosis of small hydropower units, the present invention constructs a knowledge base covering a variety of fault modes and their waveform characteristics based on simulated fault waveforms and historical normal data, and stores them as two-dimensional vector data that can be used for deep learning model training and recognition; the knowledge base not only contains the voltage and current waveforms of small hydropower units under normal operating conditions, but also covers waveform data under various typical fault modes; by combining the fault knowledge base generated by simulation and the feature extraction capability of the Attention-CNN model, automatic fault identification and classification are achieved, which significantly improves the efficiency and accuracy of fault diagnosis and helps to ensure the safe and stable operation of the power system. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 is a flow chart of the method of the present invention;

[0024] Figure 2 It is the schematic diagram of single bridge excitation power unit;

[0025] Figure 3 This is the waveform diagram of T1 fault excitation voltage and AC measurement voltage;

[0026] Figure 4 This is the waveform diagram of T4 fault excitation voltage and AC measurement voltage;

[0027] Figure 5 It is a joint monitoring process of SCADA system and fault diagnosis system. DETAILED DESCRIPTION

[0028] In order to make the purpose, technical scheme and advantages of the embodiments of the present invention clearer, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention. It should be noted that the embodiments in this application and the features in the embodiments can be combined with each other arbitrarily without conflict.

[0029] Example 1: Figure 1-4 As shown, a fault diagnosis method for the excitation power unit of a hydropower unit based on Attention-CNN and a fault knowledge base includes the following steps:

[0030] S1. Build a single-bridge excitation power unit fault knowledge base;

[0031] Specifically: various fault modes in a single-bridge excitation power unit are simulated through the Matlab simulation platform, and the two-dimensional vector form of the excitation voltage and AC side voltage waveform data under different fault modes is generated as fault simulation data; the data of the single-bridge excitation power unit in normal state is collected from the electric field site and converted into the two-dimensional vector form of the excitation voltage and AC side voltage waveform data as normal data; the normal data and the fault simulation data are combined, associated with the corresponding type code, and a single-bridge excitation power unit fault knowledge base is constructed.

[0032] The failure modes include single thyristor open circuit failure, two thyristor open circuit failures in the same phase, two thyristor open circuit failures in different polarity sides, and two thyristor open circuit failures in different phases and different polarity sides. The failure modes included in step 1 are shown in Table 1 below:

[0033] Table 1 Single bridge excitation power unit fault knowledge base

[0034]

[0035] Specifically:

[0036] Single thyristor open circuit fault, the fault mode includes 6 different situations, which are the failure of each thyristor (T1, T2, T3, T4, T5, T6) in the three-phase bridge rectifier circuit. Figure 3 The waveform data of T1 fault excitation voltage and AC voltage measurement obtained through Matlab simulation platform are shown; Figure 4 The T4 fault excitation voltage and AC measured voltage waveform data obtained through the Matlab simulation platform are displayed.

[0037] Two thyristors of the same phase have open circuit faults, and the fault modes include open circuit faults of three adjacent thyristor combinations of T1&T4, T2&T5, and T3&T6.

[0038] The two thyristors on the same pole side have open circuit faults, and the fault modes include 6 different thyristor combinations such as T1&T3, such as combined open circuit faults of T1 and T3, T3 and T5, T1 and T5, T2 and T6, T4 and T6, and T4 and T2.

[0039] Two thyristors at different phases and different polarity sides have open circuit faults. The fault modes include open circuit faults of six adjacent thyristor combinations, such as T1 & T2, such as combined open circuit faults of T1 and T2, T2 and T3, T3 and T4, T4 and T5, T5 and T6, and T1 and T6.

[0040] S2. The waveform data in the knowledge base is divided into a training set and a test set according to a preset ratio, and the excitation voltage and AC side voltage waveforms are input into the Attention-CNN model as two-dimensional vectors for training and testing; illustratively, the waveform data in the knowledge base is divided into a training set and a test set in a ratio of 7:3.

[0041] Furthermore, the Attention-CNN model includes: a convolution layer for extracting local and global features of waveform data through multiple convolution kernels; a self-attention layer for weighting features; a pooling layer for reducing dimensionality and reducing the amount of calculation; and a fully connected layer for associating features with type codes in a knowledge base and outputting classification results.

[0042] The convolution layer consists of two layers, the convolution kernel size of each convolution layer is 3×3, and the activation function of the convolution layer is the ReLU activation function; the pooling layer uses a 2×2 maximum pooling operation; the learning rate is set to 0.001; the optimizer uses the Adam optimizer, and the cross entropy loss function is used for model training. After training, the model has online learning and self-adaptation capabilities, and can automatically update the weight parameters by receiving new fault data during the model operation process, ensuring the accuracy and stability of the model in long-term operation.

[0043] Specifically, the convolution layer extracts features from the input waveform through the convolution kernel.

[0044] The input waveform is: Convolution Kernel Perform a convolution operation: Among them, Y(i,j) represents the convolution result, m×n is the size of the convolution kernel, and (i,j) is the current position of the convolution operation; the output of the convolution layer is the feature map, which is expressed as: Among them, σ(·) is the activation function, b kis the bias term, K ij is the weight of the kth convolution kernel. The convolution layer extracts features of different directions and scales through multiple convolution kernels, which reflect the local characteristics of the fault waveform.

[0045] After the output feature map of the convolutional layer, the attention mechanism is introduced to extract the features related to the fault by calculating the attention weight of each feature position. The calculation formula of the feature map after attention weighting is as follows:

[0046] First, calculate the relevance score for each feature position: ij =tanh(W a F ij +b a ).

[0047] The relevance score e ij Normalize and get the attention weight:

[0048] Calculate the weighted feature map: F′ ij =α ij ·F ij .

[0049] The pooling layer is used to reduce the size of the feature map, reduce the amount of calculation and improve the model's anti-noise ability. Its formula is as follows:

[0050] P ij =maxF′ i+m,j+n .

[0051] The pooled feature map is flattened into a vector x and input into the fully connected layer for classification. The output of the fully connected layer is the probability distribution of each fault category. Among them, z j is the activation value of the jth node, w ij is the weight of the fully connected layer, x i is the flattened feature input, b j is the bias term. The activation function is used to convert these activation values ​​into probabilities. Among them, P(y=j|x) represents the probability that the input sample belongs to fault category j, and K is the total number of fault categories. By making the maximum probability prediction for the fault category, the model finally outputs the fault category and location number.

[0052] S3, collect the excitation voltage and AC side voltage waveforms during the operation of the excitation system in real time, convert them into two-dimensional vectors and input them into the trained Attention-CNN model, output type coding, and realize real-time positioning and rapid diagnosis of faults.

[0053] Embodiment 2: A fault diagnosis system for the excitation power unit of a hydropower unit based on Attention-CNN and a fault knowledge base, comprising: a construction module for constructing a single-bridge excitation power unit fault knowledge base; a training module for dividing the waveform data in the knowledge base into a training set and a test set according to a preset ratio, and inputting the excitation voltage and AC side voltage waveforms as two-dimensional vectors into the Attention-CNN model for training and testing; a diagnosis module for real-time acquisition of the excitation voltage and AC side voltage waveforms during the operation of the excitation system, converting them into two-dimensional vectors and inputting them into the trained Attention-CNN model, and outputting a type code. The term "module" as used above may be a combination of software and / or hardware that implements a predetermined function. Although the system described in the above embodiment is preferably implemented in software, the implementation of hardware, or a combination of software and hardware, is also possible and conceivable. For parts that are not described in detail in each module, please refer to the relevant description of other embodiments.

[0054] Furthermore, the fault diagnosis system is integrated with the SCADA data acquisition system; Figure 5 As shown, the integrated solution includes linking the fault diagnosis system with the SCADA data acquisition system to achieve a combination of real-time monitoring and fault diagnosis functions.

[0055] Data collection and feedback are performed through the SCADA data acquisition system. Specifically, during the operation of the generator set, the SCADA data acquisition system collects the excitation voltage and AC side voltage waveform data in real time and transmits them to the fault diagnosis system. In the architecture of the SCADA data acquisition system, data is collected by sensors or controllers on site, transmitted to the central control unit through the SCADA communication network, and then displayed and monitored in real time by the host computer. The fault diagnosis system analyzes the received voltage waveform in real time, extracts local and global features through the Attention-CNN model, and outputs the fault type in a timely manner.

[0056] In addition, during operation, the fault diagnosis system can continuously expand the simulation knowledge base by receiving new operation data and fault data. When new fault data is collected into the system, these data will be incorporated into the existing knowledge base, and the model can be retrained and the weight parameters of the model can be updated.

[0057] By applying the above technical solution, it can be seen that the present invention uses deep learning technology to improve the fault diagnosis efficiency and accuracy of the single-bridge excitation system of small hydropower units, realizes the automatic identification of complex fault modes, and improves the operating stability and fault handling capabilities of small hydropower units.

[0058] The specific implementation modes of the present invention are described in detail above in conjunction with the accompanying drawings, but the present invention is not limited to the above implementation modes, and various changes can be made within the knowledge scope of ordinary technicians in this field without departing from the purpose of the present invention.

Claims

1. A fault diagnosis method for the excitation power unit of a hydropower unit based on Attention-CNN and fault knowledge base, characterized in that: The following steps are involved: Construct a single-bridge excitation power unit fault knowledge base; The waveform data in the knowledge base is divided into a training set and a test set according to a preset ratio. The excitation voltage and AC side voltage waveforms are input as two-dimensional vectors into the Attention-CNN model for training and testing. The excitation voltage and AC side voltage waveforms during the operation of the excitation system are collected in real time, converted into two-dimensional vectors and input into the trained Attention-CNN model, and the type code is output.

2. The fault diagnosis method for the excitation power unit of a hydropower unit based on Attention-CNN and fault knowledge base according to claim 1 is characterized in that: The construction of the single-bridge excitation power unit fault knowledge base is specifically as follows: simulating multiple fault modes in the single-bridge excitation power unit through a simulation platform, generating two-dimensional vector forms of excitation voltage and AC side voltage waveform data under different fault modes as fault simulation data; By collecting data of the single-bridge excitation power unit in normal state from the electric field site, and converting it into a two-dimensional vector form of excitation voltage and AC side voltage waveform data as normal data; Normal data and fault simulation data are combined and associated with corresponding type codes to build a single-bridge excitation power unit fault knowledge base.

3. The fault diagnosis method for the excitation power unit of a hydropower unit based on Attention-CNN and fault knowledge base according to claim 1 is characterized in that: The type code includes a normal operating condition code and four types of failure mode codes, specifically: a first type code, a second type code, a third type code, a fourth type code, and a fifth type code.

4. The fault diagnosis method for the excitation power unit of a hydropower unit based on Attention-CNN and fault knowledge base according to claim 1 is characterized in that: The failure modes include single thyristor open circuit failure, two thyristor open circuit failures in the same phase, two thyristor open circuit failures in different polarity sides, and two thyristor open circuit failures in different phases and different polarity sides.

5. The fault diagnosis method for the excitation power unit of a hydropower unit based on Attention-CNN and fault knowledge base according to claim 1 is characterized in that: The Attention-CNN model is constructed by introducing Attention on the basis of the CNN model.

6. The fault diagnosis method for the excitation power unit of a hydropower unit based on Attention-CNN and fault knowledge base according to claim 5 is characterized in that: The Attention-CNN model specifically It includes: a convolution layer, which is used to extract local and global features of waveform data through multiple convolution kernels; Self-attention layer, used to weight features; Pooling layer, used for dimensionality reduction; The fully connected layer is used to associate features with type codes in the knowledge base and output classification results.

7. The method for diagnosing faults of the excitation power unit of a hydropower unit based on Attention-CNN and fault knowledge base according to claim 6 is characterized in that: There are two convolution layers, the convolution kernel size of each convolution layer is 3×3, and the activation function of the convolution layer is the ReLU activation function.

8. The method for diagnosing faults of the excitation power unit of a hydropower unit based on Attention-CNN and fault knowledge base according to claim 6 is characterized in that: The pooling layer uses a 2×2 maximum pooling operation.

9. The method for diagnosing faults of the excitation power unit of a hydropower unit based on Attention-CNN and fault knowledge base according to claim 6 is characterized in that: The learning rate is set to 0.001; the optimizer uses the Adam optimizer and the cross entropy loss function for model training.

10. A fault diagnosis system for the excitation power unit of a hydropower unit based on Attention-CNN and fault knowledge base, characterized in that: include: A building module for building a single-bridge excitation power unit fault knowledge base; A training module is used to divide the waveform data in the knowledge base into a training set and a test set according to a preset ratio, and the excitation voltage and AC side voltage waveforms are input into the Attention-CNN model as two-dimensional vectors for training and testing; The diagnostic module is used to collect the excitation voltage and AC side voltage waveforms during the operation of the excitation system in real time, convert them into two-dimensional vectors to input the trained Attention-CNN model, and output type coding.

11. The hydropower unit excitation power unit fault diagnosis system based on Attention-CNN and fault knowledge base according to claim 10 is characterized in that: The fault diagnosis system is integrated with the SCADA data acquisition system; wherein, during the operation of the generator set, the SCADA data acquisition system collects the excitation voltage and AC side voltage waveform data in real time and transmits them to the fault diagnosis system.

Images