A hydroelectric generator unit fault diagnosis method, system and computer device of an SDAE network

Abstract

The present application belongs to the field of hydroelectric generating set fault diagnosis, and discloses a hydroelectric generating set fault diagnosis method and system based on a hybrid intelligent algorithm optimized SDAE network. A Sin chaotic mapping and a Levy flight strategy are introduced into a Harris hawk algorithm, and a reverse learning and Cauchy mutation strategy are fused into a sand cat swarm algorithm. Then, the improved Harris hawk algorithm and the improved sand cat swarm algorithm are fused by using a switching criterion. The fused hybrid intelligent algorithm is used to adaptively optimize a stacked denoising autoencoder (SDAE) network, and original data of the hydroelectric generating set are input into the optimized SDAE network for fault diagnosis. The algorithm has the advantages of the HHO algorithm in global search and the characteristics of the SCSO algorithm in fast convergence speed and high-precision optimization. Compared with existing hydroelectric generating set fault diagnosis methods, the present application has better diagnosis efficiency and higher fault diagnosis accuracy.

Description

Technical Field

[0001] This invention belongs to the field of fault diagnosis of hydro-generator sets, and particularly relates to a method and system for fault diagnosis of hydro-generator sets using a hybrid intelligent algorithm to optimize the SDAE network. Background Technology

[0002] Currently, in today's rapidly developing energy sector, hydro-turbine generator sets, as an important clean energy power generation device, are crucial for ensuring power supply and promoting sustainable economic development. However, due to the complex operating environment and diverse equipment structures of hydro-turbine generator sets, fault diagnosis faces numerous challenges. Against this backdrop, achieving effective and accurate fault diagnosis of hydro-turbine generator sets is particularly important. By accurately identifying and resolving potential problems, the reliability and operating efficiency of the equipment can be maximized, maintenance costs reduced, and thus, the continuous and stable meeting of power demand can be ensured. Therefore, in-depth research into fault diagnosis methods for hydro-turbine generator sets, and improving the accuracy and efficiency of diagnosis, is of great significance for promoting the development and popularization of clean energy technologies.

[0003] In the field of fault diagnosis research, the core is to select fault diagnosis algorithms with high accuracy, short diagnosis time, and strong robustness. The invention disclosed in patent announcement number CN110597240B is a fault diagnosis method for hydro-generator sets. This invention uses a one-dimensional convolutional neural network (1-DCNN) and gated recurrent units (GRU) for fault diagnosis of hydro-generator sets. This research mainly focuses on shallow machine learning models, which have shown high accuracy in diagnosing simple fault types. However, with the increasing complexity of hydro-generator set operating conditions, the limitations of shallow learning algorithms have become increasingly apparent, making it difficult to meet the needs of accurate diagnosis. To solve this problem, this invention utilizes deep learning algorithms for fault diagnosis. SDAE, as a typical network structure model in deep learning, is an unsupervised learning neural network model used to learn effective representations of data. This network is constructed through a multi-layered autoencoder network, possessing strong representation and feature extraction capabilities. It can detect and diagnose equipment or system faults by learning high-level representations of data, thereby improving the accuracy and efficiency of fault diagnosis.

[0004] However, the structure and parameters of the SDAE network directly affect its fault diagnosis performance. Traditional manual selection of SDAE network hyperparameters is time-consuming, labor-intensive, and suffers from insufficient generalization ability. The invention disclosed in patent announcement number CN112879200B presents a fault diagnosis method for large hydro-generator units. This invention uses a particle swarm optimization algorithm to optimize a stacked sparse autoencoder, and then uses the optimized stacked sparse autoencoder to perform fault diagnosis on the large hydro-generator unit. However, the particle swarm optimization algorithm is prone to getting trapped in local optima, resulting in low fault diagnosis accuracy.

[0005] Based on the above analysis, the problems and shortcomings of the existing technology are as follows:

[0006] (1) Limitations of shallow models: Early research mainly focused on shallow machine learning models, such as one-dimensional convolutional neural networks (1-DCNN) and gated recurrent units (GRU). These models perform well in simple fault diagnosis, but their diagnostic capabilities are insufficient for complex operating conditions and multiple fault types.

[0007] (2) Difficulty in tuning hyperparameters for deep models: The performance of SDAE is highly dependent on the selection of its structure and hyperparameters. Traditional manual hyperparameter tuning methods are time-consuming and laborious, and are prone to causing insufficient generalization ability of the model.

[0008] (3) Optimization algorithms are prone to getting trapped in local optima: Existing optimization algorithms themselves are prone to getting trapped in local optima, which affects the accuracy of fault diagnosis. Summary of the Invention

[0009] To address the problems existing in the prior art, this invention provides a method and system for fault diagnosis of hydro-generator sets using a hybrid intelligent algorithm to optimize the SDAE network.

[0010] This invention is implemented as follows: a method for fault diagnosis of hydro-generator units using a hybrid intelligent algorithm to optimize the SDAE network, wherein the method is as follows:

[0011] Sin chaotic mapping and Levy flight strategy are introduced into Harris Eagle algorithm, and reverse learning and Cauchy mutation strategy are integrated into Sand Cat swarm algorithm. Then, switching criterion is used to fuse the improved Harris Eagle algorithm and the improved Sand Cat swarm algorithm. The fused hybrid intelligent algorithm is used to adaptively optimize stacked noise reduction autoencoder (SDAE) network, and the original data of hydro turbine generator set is input into the optimized SDAE network for fault diagnosis.

[0012] Furthermore, the following steps are included:

[0013] Step 1: Install accelerometers to collect vibration data of the hydro-generator unit, and divide the data into training set samples and test set samples at a ratio of 4:1. Normalize all samples.

[0014] Step 2: Initialize the Harris Eagle Algorithm (HHO) using Sin chaotic mapping, and introduce the Levy flight strategy to improve the optimization capability of the HHO algorithm and avoid getting trapped in local optima;

[0015] Step 3: Introduce reverse learning and Cauchy mutation strategies to perturb the current optimal solution of the sand cat population, and propose an improved sand cat group algorithm SCSO.

[0016] Step 4: Propose a switching criterion that integrates the improved HHO algorithm and the improved SCSO algorithm into an HHO-SCSO hybrid intelligent algorithm;

[0017] Step 5: Input the training samples from Step 1 into the stacked noise reduction autoencoder network SDAE, and use the HHO-SCSO hybrid intelligent algorithm from Step 4 to adaptively extract the optimal parameters of the SDAE network, thereby obtaining an SDAE network with the optimal structure.

[0018] Step 6: Input the test set samples from Step 1 into the SDAE network with structural parameters for fault diagnosis, and obtain the fault type of the hydro-generator unit through multiple training iterations.

[0019] Furthermore, in step one, the vibration data of the hydro-generator unit collected by the accelerometer is the vibration data of the sliding bearing, including three types: normal lubrication, boundary friction, and severe friction.

[0020] In step one, all samples are normalized, and the normalization function is:

[0021]

[0022] In the formula, This represents the normalized sample data; This represents the original sample data.

[0023] Furthermore, in step two, the Harris Eagle algorithm HHO is initialized using Sin chaotic mapping, the method being:

[0024] The mathematical formula for the Sin chaotic mapping is shown below:

[0025]

[0026] The chaotic sequence is obtained from the mathematical expression of the Sin chaotic map. By mapping to the initial search space of the HHO algorithm, the starting position of the Harris eagle population can be determined:

[0027]

[0028] In the formula, This marks the starting point of the Harris Eagle population. and These represent the lower and upper limits of the distribution range for the Harris Eagle population.

[0029] Furthermore, in step two, the method for introducing the Levi flight strategy is as follows:

[0030]

[0031] In the formula, This represents the current optimal solution for the Harris Eagle population after the Levi flight disturbance. This represents the current optimal solution for the Harris Eagle population before the Levi flight disturbance. The step size is random. Dot product; For a random search path, the following constraints must be satisfied:

[0032]

[0033] In the formula, It follows a standard normal distribution; ;

[0034]

[0035] After the Levi flight update is completed, the fitness before and after the update needs to be compared; if the fitness after the update is lower, then the new population is considered the optimal solution; if the fitness after the update is not as good as before, then the original optimal solution is maintained.

[0036] Furthermore, in step three, reverse learning and Cauchy mutation strategies are introduced to perturb the optimal solution of the population. The method is as follows:

[0037] The mathematical description of reverse learning is as follows:

[0038]

[0039] In the formula, The solution is the reverse one; These are the upper and lower limits of the population. This is the optimal solution for the current population.

[0040] The definition of Cauchy variation is as follows:

[0041]

[0042] In the formula, The solution for the sand cat population after Cauchy mutation update; It is a random number that follows a Cauchy distribution; the meanings of the other parameters are the same as described above.

[0043] After each iteration, a random number is generated. ,according to Sum of probabilities The choice between reverse learning and Cauchy mutation to perturb and update the current optimal solution position depends on the magnitude of the difference. , This represents the current iteration number; The maximum number of iterations; if Use reverse learning to perturb the current optimal solution; if Cauchy mutation is used to perturb the current optimal solution; after the update, the fitness before and after the perturbation is compared, and the position with the smaller fitness is selected as the new optimal solution.

[0044] Choose probability The probability decreases non-linearly with increasing iterations. In the initial stages of the algorithm, the selection probability... The value is relatively large, therefore The probability is relatively high, and reverse learning perturbation updates are mainly used; in the later stages of the algorithm, the probability is selected. The value is small. The probability is relatively high, and the main method used is Cauchy mutation perturbation update.

[0045] Furthermore, step four proposes a switching criterion that integrates the improved HHO algorithm and the improved SCSO algorithm into an HHO-SCSO hybrid intelligent algorithm. The specific method is as follows:

[0046] S1: Set the algorithm switching criterion as: calculate the most recent The standard deviation of the fitness values ​​in each iteration, if continuous The standard deviation of the fitness value in each iteration does not exceed the threshold. If so, the switching criteria are met;

[0047] S2: The specific implementation method is as follows: In the early stage of the hybrid intelligent algorithm, the HHO algorithm is used for global search to converge to the interval where the global optimum is located and narrow the range of the global optimum. During the iteration of the HHO algorithm, if the algorithm switching criterion is met, the algorithm is switched to the SCSO algorithm for updating. In the later stage of the hybrid intelligent algorithm, the convergence interval of the HHO algorithm is assigned to the SCSO algorithm as the initial interval, and the SCSO algorithm is used to quickly converge to the global optimum within the small interval.

[0048] Furthermore, in step five, the optimal parameters of the SDAE network are adaptively extracted using the HHO-SCSO hybrid intelligent algorithm described in step four. The SDAE network has three hidden layers, so there are seven parameters that need to be optimized: the number of nodes in the three hidden layers, the sparsity coefficients in the three hidden layers, and the zeroing ratio of the input data dropout.

[0049] Another objective of this invention is to provide a fault diagnosis system for hydro-generator sets using a hybrid intelligent algorithm to optimize the SDAE network, comprising:

[0050] Data acquisition module: used to install accelerometers and collect vibration data of hydro-generator units, and divide the collected data into training set and test set according to preset ratio, and normalize all data;

[0051] Hybrid intelligent algorithm optimization module: The Harris Eagle algorithm is initialized by introducing Sin chaotic mapping, and the Sand Cat Group algorithm is improved by introducing reverse learning and Cauchy mutation strategy; the improved HHO algorithm and the improved SCSO algorithm are fused based on the switching criterion to form the HHO-SCSO hybrid intelligent algorithm.

[0052] SDAE Network Optimization Module: The HHO-SCSO hybrid intelligent algorithm is used to adaptively optimize the structure and parameters of the stacked noise reduction autoencoder (SDAE) network, resulting in an SDAE network with optimal structure.

[0053] Fault diagnosis module: The test set samples are input into the optimized SDAE network, and through multiple training iterations, the fault types of the hydro-generator unit are output.

[0054] Another object of the present invention is to provide a computer device, the computer device including a memory and a processor, the memory storing a computer program, the computer program being executed by the processor causing the processor to perform the steps of a method for diagnosing faults in a hydro-generator unit using a hybrid intelligent algorithm to optimize the SDAE network.

[0055] Based on the above technical solutions and the technical problems solved, the advantages and positive effects of the technical solution to be protected by this invention are as follows:

[0056] First, this invention proposes a novel HHO-SCSO hybrid intelligent algorithm, which combines the advantages of the HHO algorithm in global search with the characteristics of the SCSO algorithm in fast convergence speed and high-precision optimization. In the early stages of algorithm iteration, Sin chaotic mapping and the Lévy flight strategy are introduced to accelerate the convergence speed of the HHO algorithm and improve its global search performance. In the later stages of algorithm iteration, reverse learning and Cauchy mutation strategies are introduced to compensate for the SCSO algorithm's tendency to get trapped in local optima. Compared with existing optimization algorithms, the HHO-SCSO hybrid intelligent algorithm of this invention has higher convergence accuracy and better optimization performance.

[0057] This invention proposes to optimize the hyperparameters of the SDAE network using the HHO-SCSO hybrid intelligent algorithm, which solves the problems of time-consuming and laborious manual selection of SDAE network parameters by the traditional enumeration method and insufficient generalization ability, thereby improving the efficiency and accuracy of fault diagnosis.

[0058] Compared with existing fault diagnosis methods for hydro-generator sets, the present invention has better diagnostic efficiency and higher fault diagnosis accuracy.

[0059] Second, the technical problem solved by this invention is:

[0060] Limitations of shallow models: Existing fault diagnosis methods for hydro-generator units mainly rely on shallow machine learning models, which are not adaptable to complex fault types and operating conditions and cannot meet the requirements for high-precision diagnosis.

[0061] Deep model parameter tuning is difficult: Although deep learning models such as SDAE have shown potential in fault diagnosis, their performance is greatly affected by network structure and hyperparameters. Traditional manual parameter tuning is time-consuming and laborious, and has limited generalization ability.

[0062] The optimization algorithm is prone to getting trapped in local optima: Existing optimization algorithms are prone to getting trapped in local optima when automatically adjusting SDAE parameters, resulting in low accuracy in fault diagnosis.

[0063] The significant technological advancements achieved by this invention are as follows:

[0064] Hybrid Intelligent Algorithm Optimization of SDAE: A hybrid intelligent algorithm integrating the improved Harris Eagle Algorithm (HHO) and the improved Sand Cat Swarm Algorithm (SCSO) is proposed. By adaptively optimizing the SDAE network structure and parameters, the limitations of traditional manual parameter tuning are overcome, and the generalization ability and diagnostic accuracy of the model are improved.

[0065] Enhanced Global Search and Local Optimization Capabilities: The global search capability of the HHO algorithm is enhanced by introducing Sin chaotic mapping and Lévy flight strategy to avoid getting trapped in local optima; at the same time, the SCSO algorithm is improved by using reverse learning and Cauchy mutation strategy to improve local optimization capability and accelerate convergence speed.

[0066] Improved accuracy and efficiency in fault diagnosis: By optimizing the SDAE network through hybrid intelligent algorithms, more accurate and efficient fault diagnosis of hydro-generator units has been achieved, providing strong support for ensuring the stable operation of the power system.

[0067] Third, the expected benefits and commercial value of the technical solution of this invention after transformation are as follows: the proposed method achieves an average fault diagnosis accuracy of 98.21% for its hydro-generator units, which is an improvement of 8.19% in average diagnosis accuracy. It effectively improves the fault diagnosis level of hydro-generator units and has good expected benefits and commercial value.

[0068] This invention solves a long-standing but unresolved technical problem: the structure and parameters of the SDAE network directly impact the fault diagnosis performance of hydro-generator units. Traditional manual selection of SDAE network hyperparameters is time-consuming, labor-intensive, and suffers from insufficient generalization ability. This invention innovatively proposes a hybrid intelligent algorithm combining an improved Harris Hawks Optimization (HHO) algorithm and an improved Sand Cat Swarm Optimization (SCSO) algorithm. This hybrid intelligent algorithm adaptively determines the optimal structural parameters of the SDAE network, and finally, the optimized SDAE network is used for fault diagnosis of hydro-generator units. Compared to traditional SDAE networks, the proposed method achieves an average fault diagnosis accuracy of 98.21%, an improvement of 8.19%. This successfully solves the technical problem of low fault diagnosis accuracy in hydro-generator units using previous methods.

[0069] Third, this invention proposes a hybrid intelligent algorithm-optimized SDAE network fault diagnosis method for hydro-generator units, which mainly solves the problems of low accuracy and poor stability of traditional fault diagnosis methods. By integrating the improved Harris Eagle algorithm and Sand Cat swarm algorithm, the SDAE network is adaptively optimized, thereby improving the accuracy and efficiency of fault diagnosis.

[0070] In terms of technological innovation, this invention employs Sin chaotic mapping and the Lévy flight strategy to improve the Harris Eagle algorithm, enhancing its global search capability and avoiding the trap of local optima. Simultaneously, by introducing reverse learning and Cauchy mutation strategies, innovative improvements are made to the Sand Cat Swarm Optimization algorithm, increasing its convergence speed and optimization performance.

[0071] The mathematical model and algorithm design of this invention achieve adaptive optimization of the SDAE network, enabling the network to better learn and identify fault characteristics of hydro-generator units. By using a switching criterion, two improved intelligent algorithms are cleverly combined to form a powerful hybrid intelligent optimization method, significantly improving the reliability of fault diagnosis.

[0072] In summary, this invention, through innovative algorithm fusion and mathematical model design, not only addresses the shortcomings of traditional fault diagnosis methods but also achieves technological advancement, providing a more efficient and accurate solution for fault diagnosis of hydroelectric generator units. The implementation of this technology will help improve the safe operation level of hydropower station equipment, reduce downtime due to faults, and thus bring significant economic and social benefits.

[0073] Fourth, this invention solves several technical problems in fault diagnosis of hydro-generator sets in industrial applications. Traditional fault diagnosis methods are often limited by data processing capabilities and algorithm accuracy, resulting in inaccurate diagnostic results, and may even lead to missed or misdiagnosed faults. This invention significantly improves the accuracy and sensitivity of fault diagnosis by introducing a hybrid intelligent algorithm to optimize the SDAE network.

[0074] In industrial applications, the stable operation of hydro-generator units is crucial. However, existing fault diagnosis technologies often fail to detect potential faults in a timely manner, posing hidden dangers to industrial production. This invention, through adaptive optimization of the SDAE network, can more effectively identify abnormal conditions during unit operation and provide early warnings, thereby significantly reducing the probability of fault occurrence.

[0075] Furthermore, traditional fault diagnosis methods are inefficient when processing complex data and cannot meet real-time requirements. This invention utilizes the efficient search and optimization capabilities of hybrid intelligent algorithms to significantly improve data processing speed and diagnostic efficiency, enabling real-time monitoring and rapid response, and providing strong support for the continuity and stability of industrial production.

[0076] In summary, this invention represents a significant technological advancement in industrial applications. It not only improves the accuracy and efficiency of fault diagnosis but also provides strong technical support for the safe operation of hydroelectric generator sets. The promotion and application of this technology will greatly enhance the maintenance level of industrial equipment, reduce operating costs, and bring long-term economic benefits to industrial development. Attached Figure Description

[0077] Figure 1 This is an overall flowchart of the fault diagnosis of hydro-generator units using a hybrid intelligent algorithm to optimize the SDAE network, provided in an embodiment of the present invention.

[0078] Figure 2 This is a flowchart of the improved HHO-SCSO hybrid intelligent algorithm provided in this embodiment of the invention;

[0079] Figure 3 These are fitness curves of SDAE trained using different optimization algorithms provided in this embodiment of the invention;

[0080] Figure 4 This is the initial fault classification confusion matrix diagram of the hydro-generator unit provided in the embodiments of the present invention;

[0081] Figure 5 This is a confusion matrix diagram of the first fault classification of the optimized and unoptimized SDAE networks provided by the HHO-SCSO hybrid intelligent algorithm in this embodiment of the invention. Detailed Implementation

[0082] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0083] This invention first proposes an improved Harris Hawks Optimization (HHO) algorithm and an improved Sand Cat Swarm Optimization (SCSO) algorithm. Specifically, it introduces Sin chaotic mapping and Lévy flight strategy to accelerate the convergence speed of the HHO algorithm and improve the global search effect, and integrates back-learning and Cauchy mutation strategies to compensate for the SCSO algorithm's tendency to get trapped in local optima. Then, a switching criterion is proposed to fuse the improved HHO and SCSO algorithms into an HHO-SCSO hybrid intelligent algorithm, achieving complementary advantages and overcoming their respective shortcomings. Finally, the HHO-SCSO hybrid intelligent algorithm is used to adaptively optimize the SDAE network, and the original data of the hydro-generator unit is input into the optimized SDAE network for fault diagnosis. Compared with existing hydro-generator unit fault diagnosis methods, this invention has better diagnostic efficiency and higher fault diagnosis accuracy.

[0084] like Figure 1 The diagram shown is an overall flowchart of the fault diagnosis process for hydro-generator units using a hybrid intelligent algorithm to optimize the SDAE network, as provided in an embodiment of the present invention.

[0085] Step 1: Install accelerometers to collect vibration data of the hydro-generator unit, and divide the data into training set samples and test set samples at a ratio of 4:1. Normalize all samples.

[0086] Step 2: Initialize the Harris Eagle Algorithm (HHO) using a Sin chaotic mapping to enhance population diversity, thereby accelerating the algorithm's convergence speed and improving global search performance. Introduce the Lévy flight strategy to enhance the HHO algorithm's optimization capability and avoid getting trapped in local optima.

[0087] Step 3: Introduce reverse learning and Cauchy mutation strategies to perturb the current optimal solution of the sand cat population, overcoming the shortcomings of the traditional sand cat group algorithm (SCSO) in that it is easy to get trapped in local optima during the iteration process;

[0088] Step 4: Propose a switching criterion that integrates the improved HHO algorithm and the improved SCSO algorithm into an HHO-SCSO hybrid intelligent algorithm to achieve complementary advantages of the two algorithms and make up for their respective shortcomings.

[0089] Step 5: Input the training samples from Step 1 into the Stacked Denoising Autoencoder Network (SDAE), and use the HHO-SCSO hybrid intelligent algorithm from Step 4 to adaptively extract the optimal parameters of the SDAE network, thereby obtaining an SDAE network with the optimal structure;

[0090] Step 6: Input the test set samples from Step 1 into the SDAE network with structural parameters for fault diagnosis, and obtain the fault type of the hydro-generator unit through multiple training iterations.

[0091] In step one, the vibration data of the hydro-generator set collected by the accelerometer is the vibration data of the sliding bearing, including three types: normal lubrication, boundary friction and severe friction.

[0092] In step one, all samples are normalized, and the normalization function is:

[0093] (1)

[0094] In the formula, This represents the normalized sample data; This represents the original sample data.

[0095] In step two, the Harris Eagle Algorithm (HHO) is initialized using the Sin chaotic mapping. The specific method is as follows:

[0096] The mathematical formula for the Sin chaotic mapping is shown below:

[0097] (2)

[0098] The chaotic sequence can be obtained from the mathematical expression of the Sin chaotic mapping (2). By mapping to the initial search space of the HHO algorithm, the starting position of the Harris Eagle population can be determined, as shown in equation (3):

[0099] (3)

[0100] In the formula, This marks the starting point of the Harris Eagle population. and These represent the lower and upper limits of the distribution range for the Harris Eagle population.

[0101] In step two, the Lévy flight strategy is introduced to improve the optimization capability of the HHO algorithm and avoid getting trapped in local optima. The specific method is as follows:

[0102] The definition of Levy's flight strategy is as follows:

[0103] (4)

[0104] In the formula, This represents the current optimal solution for the Harris Eagle population after the Levi flight disturbance. This represents the current optimal solution for the Harris Eagle population before the Levi flight disturbance. The step size is random. Dot product; For a random search path, the following constraints must be satisfied:

[0105] (5)

[0106] In the formula, It follows a standard normal distribution; .

[0107] (6)

[0108] After the Levy flight update is completed, the fitness before and after the update needs to be compared. If the fitness after the update is lower, then the new population is considered the optimal solution; if the fitness after the update is worse than before, then the original optimal solution is maintained.

[0109] Step three introduces reverse learning and Cauchy mutation strategies to perturb the optimal solution of the population. The specific method is as follows:

[0110] The mathematical description of reverse learning is as follows:

[0111] (7)

[0112] In the formula, The solution is the reverse one; These are the upper and lower limits of the population. This is the optimal solution for the current population.

[0113] The definition of Cauchy variation is as follows:

[0114] (8)

[0115] In the formula, The solution for the sand cat population after Cauchy mutation update; It is a random number that follows a Cauchy distribution; the meanings of the other parameters are the same as described above.

[0116] After each iteration, a random number is generated. ,according to Sum of probabilities The choice between reverse learning and Cauchy mutation to perturb and update the current optimal solution position depends on the magnitude of the difference. The definition is shown in equation (9). If Use reverse learning to perturb the current optimal solution; if Cauchy mutation is used to perturb the current optimal solution. After the update, the fitness before and after the perturbation is compared, and the position with the smaller fitness is selected as the new optimal solution.

[0117] (9)

[0118] In the formula, This represents the current iteration number; This represents the maximum number of iterations.

[0119] From equation (9), we can see that the selection probability The probability decreases non-linearly with increasing iterations. In the initial stages of the algorithm, the selection probability... The value is relatively large, therefore The probability is relatively high, and reverse learning perturbation updates are mainly used; in the later stages of the algorithm, the probability is selected. The value is small. The probability is relatively high, and the main method used is Cauchy mutation perturbation update.

[0120] Step four proposes a switching criterion that integrates the improved HHO algorithm and the improved SCSO algorithm into an HHO-SCSO hybrid intelligent algorithm. The specific method is as follows:

[0121] S1: Set the algorithm switching criterion as: calculate the most recent The standard deviation of the fitness values ​​in each iteration, if continuous The standard deviation of the fitness value in each iteration does not exceed the threshold. If so, the switching criteria are met.

[0122] S2: The specific implementation method is as follows: In the early stage of the hybrid intelligent algorithm, the HHO algorithm is used for global search to converge to the interval where the global optimum is located and narrow the range of the global optimum. During the iteration of the HHO algorithm, if the algorithm switching criterion is met, the algorithm is switched to the SCSO algorithm for updating. In the later stage of the hybrid intelligent algorithm, the convergence interval of the HHO algorithm is assigned to the SCSO algorithm as the initial interval, and the SCSO algorithm is used to quickly converge to the global optimum within the small interval.

[0123] In step five, the optimal parameters of the SDAE network are adaptively extracted using the HHO-SCSO hybrid intelligent algorithm described in step four. The specific steps are as follows:

[0124] S1: The SDAE network has 3 hidden layers, therefore 7 parameters need to be optimized. These are the number of nodes in the 3 hidden layers, the sparsity coefficients in the 3 hidden layers, and the proportion of input data dropout to zero.

[0125] S2: Initialize HHO algorithm parameters, and initialize the population using Sin chaotic mapping;

[0126] S3: Use the classification error rate as the fitness function to evaluate each group and select the group with the minimum fitness value as the local optimum.

[0127] S4: Update the population position using the initial update mechanism of the HHO algorithm;

[0128] S5: Re-evaluate the fitness value and update the positions of the current optimal solution and the global optimal solution based on the results;

[0129] S6: Based on the definition of Levi flight, use Levi flight to adjust the population's position;

[0130] S7: Based on the change in fitness value, determine whether to adopt the updated population as the new optimal solution. If the fitness value decreases, then replace it; otherwise, keep the original optimal solution unchanged.

[0131] S8: Evaluate whether the algorithm conversion conditions are met. If they are met, convert to the SCSO algorithm; otherwise, return to S4.

[0132] S9: Use the population limit of the HHO algorithm to set the initial population range of the SCSO algorithm, and use the Sin chaotic mapping for population initialization.

[0133] S10: Evaluate the fitness of the population, and then update the population position according to the update rules of the SCSO algorithm;

[0134] S11: Re-evaluate fitness values ​​to determine the global optimal position and calculate the selection probability. ;

[0135] S12: If The current population position is perturbed according to the definition of reverse learning; if the condition is not met, the current population position is perturbed according to the definition of Cauchy mutation. After the perturbation, the changes in fitness values ​​before and after the perturbation are compared. If the fitness value decreases, the perturbed population position is taken as the current optimal solution; otherwise, the current optimal solution remains unchanged.

[0136] S13: If the current iteration number is... Greater than the limit value If the condition is met, the iteration stops, and the seven optimal parameters of the SDAE network are obtained; otherwise, return to S10 until the condition is met.

[0137] Preferred implementation schemes such as Figures 2 to 5 As shown, a method for fault diagnosis of hydro-generator units using a hybrid intelligent algorithm to optimize the SDAE network is described, with the following specific steps:

[0138] S1: A split S-195 diesel engine main bearing was used, and a 30 N·m weight was applied to simulate the bearing load. To simulate friction failure of the sliding bearing, the lubrication oil circuit was cut off, and a 6010 rolling bearing was used to support both ends of the bearing. For power transmission, an electric motor, a continuously variable transmission (CVT), and a flexible coupling were used. To monitor the bearing's operating status, an RNT-200 sensor was placed on the upper surface of the bearing cover, and signals were acquired at a frequency of 500 kHz.

[0139] S2: During the experiment, once the spindle speed stabilized, the oil supply to the bearing bush was immediately cut off, and data was recorded at a frequency of acquiring signals every 2 minutes until a sharp increase in bearing temperature or a decrease in speed was observed, at which point the experiment was stopped. A total of 300 samples were selected in the experiment, each containing 4096 data points. These included 100 normal lubrication samples (62 training samples and 38 test samples), 100 boundary friction samples (63 training samples and 37 test samples), and 100 severe friction samples (63 training samples and 37 test samples). The One-Hot encoding of the three types of bearing vibration signals is shown in Table 1.

[0140] Table 1 Bearing Failure Types and Codes

[0141]

[0142] This experiment successfully simulated the operating state and friction fault conditions of the sliding bearing in a hydro-generator unit, and the sample selection covered various degrees of friction faults, demonstrating representativeness. Furthermore, the selection of training and testing samples provided a reliable foundation for subsequent data analysis and algorithm validation.

[0143] S3: Each sample in the original data has 4096 data points, meaning the SDAE network has 4096 input layer nodes. However, research indicates that excessively high input layer dimensionality in neural networks not only affects diagnostic speed but also reduces fault diagnosis accuracy. Therefore, this invention uses principal component analysis to reduce the dimensionality of the original data from 4096 to 100, resulting in 100 input layer nodes for the SDAE network. The bearing data in this invention has three types (normal lubrication, boundary friction, and severe friction), meaning the SDAE network has 3 output layer nodes.

[0144] S4: The training samples from the experimental data are input into the hydro-generator fault diagnosis system using the hybrid intelligent algorithm-optimized SDAE network for training. The feature vectors of the training samples are input into the constructed SDAE network, and the stacked DAE network is trained layer by layer. The improved HHO-SCSO hybrid intelligent algorithm is used to adaptively extract the hyperparameters of the SDAE network. The population size of the improved HHO-SCSO hybrid intelligent algorithm is set to 50, the maximum number of iterations is set to 100, and the switching criteria of the improved HHO-SCSO hybrid intelligent algorithm are specified. threshold The improved HHO-SCSO hybrid intelligent algorithm optimizes the hyperparameters of the SDAE network as follows: Figure 2 As shown;

[0145] S5: Improved HHO-SCSO hybrid intelligent algorithm, HHO optimization algorithm, SCSO optimization algorithm, PSO optimization algorithm. See the iterative process for optimizing SDAE network parameters. Figure 3 The results of optimizing the SDAE network parameters using the improved HHO-SCSO hybrid intelligent algorithm are shown in Table 2.

[0146] Table 2. Improved HHO-SCSO Hybrid Intelligent Algorithm for Optimizing SDAE Network Parameters

[0147]

[0148] S6: Input the test sample data into the optimized SDAE network for fault diagnosis. Based on the Python programming output, obtain the initial fault classification confusion matrix of the hydro-generator unit as follows: Figure 4 As shown, the final result is the average of 10 fault diagnosis results. The final fault diagnosis results are shown in Table 3:

[0149] Table 3. Fault diagnosis results of hydro-generator units using SDAE network optimized by hybrid intelligent algorithm

[0150]

[0151] As shown in Table 3, the hydro-generator fault diagnosis system with hybrid intelligent algorithm-optimized SDAE network exhibits very high fault diagnosis accuracy and fast fault diagnosis time. Under normal lubrication conditions, the accuracy reaches 97.36%, under boundary friction conditions it reaches 100%, and under severe friction conditions it reaches 97.29%. Moreover, in all cases, the fault diagnosis time is less than 1 second.

[0152] Overall, the hydro-generator fault diagnosis system with hybrid intelligent algorithm-optimized SDAE network can accurately and efficiently identify fault types and has excellent fault diagnosis performance.

[0153] S7: To demonstrate that the fault diagnosis method for hydro-generator units using the hybrid intelligent algorithm-optimized SDAE network is superior to the fault diagnosis method using the original unoptimized SDAE network, experimental data were input into both the hybrid intelligent algorithm-optimized SDAE network fault diagnosis method and the original unoptimized SDAE network for fault diagnosis. The initial fault classification confusion matrix is ​​as follows: Figure 5 As shown in the figure, the left figure is the fault diagnosis result of the hydro-generator unit with the SDAE network optimized by the hybrid intelligent algorithm, and the right figure is the fault diagnosis result of the original unoptimized SDAE network. The final result is the average of 10 fault diagnosis results, as shown in Table 4.

[0154] Table 4. Comparison of fault diagnosis results between hydro-generator units using SDAE networks optimized with those using the original unoptimized SDAE networks.

[0155]

[0156] As shown in Table 4, the fault diagnosis method for hydro-generator sets using the hybrid intelligent algorithm-optimized SDAE network achieves 8.19% higher accuracy, 0.91% lower loss rate, and 1.88 seconds shorter diagnosis time compared to the original, unoptimized SDAE network fault diagnosis method. Therefore, this invention demonstrates that the fault diagnosis method for hydro-generator sets using the hybrid intelligent algorithm-optimized SDAE network proposed in this invention has higher accuracy and shorter fault diagnosis time than the original, unoptimized SDAE network fault diagnosis method.

[0157] It should be noted that embodiments of the present invention can be implemented in hardware, software, or a combination of both. The hardware portion can be implemented using dedicated logic; the software portion can be stored in memory and executed by a suitable instruction execution system, such as a microprocessor or dedicated-design hardware. Those skilled in the art will understand that the above-described devices and methods can be implemented using computer-executable instructions and / or included in processor control code, for example, such code provided on a carrier medium such as a disk, CD, or DVD-ROM, a programmable memory such as read-only memory (firmware), or a data carrier such as an optical or electronic signal carrier. The devices and modules of the present invention can be implemented by hardware circuitry such as very large-scale integrated circuits or gate arrays, semiconductors such as logic chips, transistors, or programmable hardware devices such as field-programmable gate arrays, programmable logic devices, etc., or by software executed by various types of processors, or by a combination of the above-described hardware circuitry and software, such as firmware.

[0158] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications, equivalent substitutions, and improvements made by those skilled in the art within the scope of the technology disclosed in the present invention, and within the spirit and principles of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for fault diagnosis of hydro-generator units using a hybrid intelligent algorithm to optimize the SDAE network, characterized in that, The method is as follows: Sin chaotic mapping and Levy flight strategy are introduced into Harris Eagle algorithm, and reverse learning and Cauchy mutation strategy are integrated into Sand Cat swarm algorithm. Then, the improved Harris Eagle algorithm and improved Sand Cat swarm algorithm are fused using switching criterion. The fused hybrid intelligent algorithm is used to adaptively optimize stacked noise reduction autoencoder (SDAE) network, and the original data of hydro turbine generator set is input into the optimized SDAE network for fault diagnosis. The steps include the following: Step 1: Install accelerometers to collect vibration data of the hydro-generator unit, and divide the data into training set samples and test set samples at a ratio of 4:

1. Normalize all samples. Step 2: Initialize the Harris Eagle Algorithm (HHO) using Sin chaotic mapping, and introduce the Levy flight strategy to improve the optimization capability of the HHO algorithm and avoid getting trapped in local optima; Step 3: Introduce reverse learning and Cauchy mutation strategies to perturb the current optimal solution of the sand cat population, and propose an improved sand cat group algorithm SCSO. Step 4: Propose a switching criterion to integrate the improved HHO algorithm and the improved SCSO algorithm into an HHO-SCSO hybrid intelligent algorithm; Step 5: Input the training samples from Step 1 into the stacked noise reduction autoencoder network SDAE, and use the HHO-SCSO hybrid intelligent algorithm from Step 4 to adaptively extract the optimal parameters of the SDAE network, thereby obtaining an SDAE network with the optimal structure. Step 6: Input the test set samples from Step 1 into the SDAE network with structural parameters for fault diagnosis, and obtain the fault type of the hydro-generator unit through multiple training iterations; Step four proposes a switching criterion that merges the improved HHO algorithm and the improved SCSO algorithm into an HHO-SCSO hybrid intelligent algorithm. The specific method is as follows: S1: Set the algorithm switching criterion as: calculate the most recent The standard deviation of the fitness values ​​in each iteration, if continuous The standard deviation of the fitness value in each iteration does not exceed the threshold. If so, the switching criteria are met; S2: The specific implementation method is as follows: In the early stage of the hybrid intelligent algorithm, the HHO algorithm is used to perform a global search to converge to the interval where the global optimum is located and narrow the range of the global optimum; during the iteration of the HHO algorithm, if the algorithm switching criterion is met, switch to the SCSO algorithm for updating; in the later stage of the hybrid intelligent algorithm, the convergence interval of the HHO algorithm is assigned to the SCSO algorithm as the initial interval, and the SCSO algorithm is used to quickly converge to the global optimum within the small interval.

2. The method for fault diagnosis of hydro-generator units using a hybrid intelligent algorithm to optimize the SDAE network according to claim 1, characterized in that, In step one, the vibration data of the hydro-generator set collected by the accelerometer is the vibration data of the sliding bearing, including three types: normal lubrication, boundary friction and severe friction. In step one, all samples are normalized, and the normalization function is: ； In the formula, This represents the normalized sample data; This represents the original sample data.

3. The method for fault diagnosis of hydro-generator units using a hybrid intelligent algorithm to optimize the SDAE network according to claim 1, characterized in that, In step two, the Harris Eagle algorithm HHO is initialized using Sin chaotic mapping. The method is as follows: The mathematical formula for the Sin chaotic mapping is shown below: ； The chaotic sequence is obtained from the mathematical expression of the Sin chaotic map. By mapping to the initial search space of the HHO algorithm, the starting position of the Harris eagle population can be determined: ； In the formula, This marks the starting point of the Harris Eagle population. and These represent the lower and upper limits of the distribution range for the Harris Eagle population.

4. The method for fault diagnosis of hydro-generator units using a hybrid intelligent algorithm to optimize the SDAE network according to claim 1, characterized in that, In step two, the method for introducing the Levi flight strategy is as follows: ； In the formula, This represents the current optimal solution for the Harris Eagle population after the Levi flight disturbance. This represents the current optimal solution for the Harris Eagle population before the Levi flight disturbance. The step size is random. Dot product; For a random search path, the following constraints must be satisfied: ； In the formula, It follows a standard normal distribution; ; ； After the Levi flight update is completed, the fitness before and after the update needs to be compared; if the fitness after the update is lower, then the new population is considered the optimal solution; if the fitness after the update is not as good as before, then the original optimal solution is maintained.

5. The method for fault diagnosis of hydro-generator units using a hybrid intelligent algorithm to optimize the SDAE network according to claim 1, characterized in that, Step three introduces reverse learning and Cauchy mutation strategies to perturb the optimal solution of the population. The method is as follows: The mathematical description of reverse learning is as follows: ； In the formula, The solution is the reverse one; These are the upper and lower limits of the population. This is the optimal solution for the current population. The definition of Cauchy variation is as follows: ； In the formula, The solution for the sand cat population after Cauchy mutation update; It is a random number that follows a Cauchy distribution; the meanings of the other parameters are the same as described above. After each iteration, generate a random number. ,according to Sum of probabilities The choice between reverse learning and Cauchy mutation to perturb and update the current optimal solution position depends on the magnitude of the difference. , This represents the current iteration number; The maximum number of iterations; if Use reverse learning to perturb the current optimal solution; if Cauchy mutation is used to perturb the current optimal solution; after the update, the fitness before and after the perturbation is compared, and the position with the smaller fitness is selected as the new optimal solution. Choose probability The probability decreases non-linearly with increasing iteration count; in the early stages of the algorithm, the selection probability... The value is relatively large, therefore The probability is relatively high, so reverse learning perturbation is used for updating; in the later stages of the algorithm, the probability is selected. The value is small. The probability is relatively high, so Cauchy mutation perturbation is used for updating.

6. The method for fault diagnosis of hydro-generator units using a hybrid intelligent algorithm to optimize the SDAE network according to claim 1, characterized in that, In step five, the optimal parameters of the SDAE network are adaptively extracted using the HHO-SCSO hybrid intelligent algorithm described in step four. The SDAE network has three hidden layers, so there are seven parameters that need to be optimized: the number of nodes in the three hidden layers, the sparsity coefficients in the three hidden layers, and the proportion of zeroing the input data dropout.

7. A hydro-generator fault diagnosis system using a hybrid intelligent algorithm to optimize the SDAE network as described in claim 1, characterized in that, include: Data acquisition module: used to install accelerometers and collect vibration data of hydro-generator units, and divide the collected data into training set and test set according to preset ratio, and normalize all data; Hybrid intelligent algorithm optimization module: introduces Sin chaotic mapping to initialize Harris Eagle algorithm, and introduces reverse learning and Cauchy mutation strategy to improve Sand Cat swarm algorithm; Based on the switching criteria, the improved HHO algorithm and the improved SCSO algorithm are integrated to form the HHO-SCSO hybrid intelligent algorithm. SDAE Network Optimization Module: The HHO-SCSO hybrid intelligent algorithm is used to adaptively optimize the structure and parameters of the stacked noise reduction autoencoder (SDAE) network, resulting in an SDAE network with optimal structure. Fault diagnosis module: The test set samples are input into the optimized SDAE network, and through multiple training iterations, the fault types of the hydro-generator unit are output.

8. A computer device, characterized in that, The computer device includes a memory and a processor. The memory stores a computer program. When the computer program is executed by the processor, the processor performs the steps of the method for diagnosing faults in a hydro-generator unit that optimizes the SDAE network using a hybrid intelligent algorithm as described in claim 1.

Images