Hydroelectric generating set group linkage fault diagnosis method and system based on neural network

Abstract

The invention discloses a hydroelectric generating set group linkage fault diagnosis method based on a neural network, and relates to the technical field of fault diagnosis, and the diagnosis steps are as follows: S1, obtaining hydroelectric generating set group linkage fault historical cases in different regions, and sorting the historical cases to form a fault case library; s2, based on a neural network, learning a non-linear relationship between the hydroelectric generating set characteristics and faults, and obtaining relevance between the hydroelectric generating set characteristics and fault causes; and S3, observing and obtaining fault features existing in the current hydroelectric generating set group linkage, and carrying out similarity calculation on the current fault features and the features in the fault cases. The invention further discloses a hydroelectric generating set group linkage fault diagnosis system based on the neural network. According to the invention, by constructing the fault case library, collecting fault cases in different regions, constructing the combinatorial optimization model and extracting key fault features by using the quantum annealing machine, the optimal fault path is found more quickly, the diagnosis reliability is improved, and the whole system can diagnose faults efficiently and accurately.

Description

Technical Field

[0001] The present invention relates to the technical field of fault diagnosis, and in particular to a method and system for diagnosing hydropower group linkage faults based on a neural network. Background Art

[0002] As electricity demand continues to grow, hydropower, as a clean and renewable energy source, plays an increasingly important role in the power supply system. The scale of hydropower groups continues to expand, and their operational stability and reliability are directly related to the safe and efficient operation of the power system. In actual operation, hydropower groups are prone to chain failures due to the influence of multiple factors such as complex operating environment, equipment aging, and improper operation. Once such failures occur, not only will the hydropower units be shut down, resulting in huge economic losses, but they may also have a serious impact on the stability of the power system and even cause large-scale power outages.

[0003] Traditional hydropower unit fault diagnosis methods, often based on simple threshold judgments or empirical rules, are unable to effectively address complex failure modes such as the coordinated failure of hydropower units. These methods cannot accurately capture the complex characteristics exhibited by hydropower units during failure, nor the nonlinear relationship between fault characteristics and fault causes. This results in poor accuracy and timeliness of fault diagnosis. For example, when faced with coordinated failures caused by multiple coupling factors, traditional methods often have difficulty accurately locating the root cause of the fault, thus delaying the timing of fault repair.

[0004] In recent years, although some advanced technologies have been introduced into the field of hydropower unit fault diagnosis, there are still many shortcomings. For example, although signal processing-based methods can analyze some fault signals, they have limited ability to mine deep fault characteristics hidden in large amounts of operating data; and in computer server operations, a large amount of computer computing resources are required. In this regard, we propose a hydropower unit group linkage fault diagnosis method and system based on neural networks. Summary of the Invention

[0005] In order to solve the above technical problems, a method and system for diagnosing hydropower group linkage faults based on neural network are provided. This technical solution solves the above problems.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a method for diagnosing hydropower group linkage faults based on a neural network, wherein the diagnostic steps are as follows:

[0007] S1. Obtain historical cases of hydropower group linkage failures in different regions, organize the historical cases, and form a failure case database;

[0008] S2, learning the nonlinear relationship between hydropower unit characteristics and faults based on neural networks, and obtaining the correlation between hydropower unit characteristics and fault causes;

[0009] S3. Observe and obtain the fault characteristics existing in the current hydropower group linkage, calculate the similarity between the current fault characteristics and the characteristics in the fault case, select cases similar to the current fault, obtain historical case diagnosis results, use the historical diagnosis results as the preliminary diagnosis results of the current hydropower group linkage, and obtain the fault range interval;

[0010] S4. Collect the operating data of the hydropower unit, extract the key fault features after pre-processing, build a combination optimization model, input the features into the quantum annealing machine, and conduct in-depth calculation and analysis;

[0011] S5. After multiple iterations of convergence, the quantum state is converged to the optimal fault path in the fault range interval, and the cause of the current hydropower group linkage fault is determined in the fault range interval.

[0012] Preferably, in step S1, the sources for obtaining fault history cases include enterprise fault record database, equipment maintenance and repair reports and academic database; fault history cases are classified to form a fault case library; the classification methods include: classification by fault type, classification by fault cause and classification by region.

[0013] Preferably, the neural network model construction step in step S2 is:

[0014] Determine the problem to be solved and the characteristics of the input and output;

[0015] Determine the neural network structure, including the number of nodes in the input layer, the number of nodes in the hidden layer, and the number of nodes in the output layer;

[0016] Input the acquired historical cases into the neural network model, train and optimize the model, initialize the weights and bias terms, perform forward propagation, calculate the loss, and backpropagate to update the parameters, repeating until convergence;

[0017] Find out the influence of each feature on the fault cause, and explore the potential correlation between the characteristics of hydropower units and the fault cause when the hydropower group linkage fault occurs.

[0018] Preferably, the association mining steps are:

[0019] Calculate the sum of the absolute values of the weights from the input layer to the hidden layer and from the hidden layer to the output layer. The weight from the i-th feature of the input layer to the j-th neuron of the l-th hidden layer is The weight from the jth neuron in the lth hidden layer to the kth neuron in the output layer is The influence degree of the i-th feature on the k-th neuron in the output layer is I ik Expressed as:

[0020]

[0021] Where L is the number of hidden layers, n l is the number of neurons in the lth hidden layer, by calculating I ik Quantify the impact of each feature on different fault cause categories, explore the potential correlation between hydropower unit features and fault causes when hydropower group linkage failure occurs, and calculate the results I ik A large value indicates that the i-th feature plays a role in determining the cause of the k-th fault.

[0022] Preferably, the similarity calculation step in step S3 is:

[0023] Based on the correlation between the hydropower unit characteristics and the fault causes, a weight value w1 is assigned to the corresponding characteristics;

[0024] The acquired features are vectorized based on discretization processing, and the vectorized data are calculated;

[0025] Assume that the current fault feature vector is X=(x1,x2,…,x n ), the eigenvector in the fault case is Y=(y1,y2,…,y n ), similarity value d E The formula for (X,Y) is:

[0026]

[0027] where d E (X, Y) is the calculated similarity value, d E The smaller the (X,Y) value, the more similar the two eigenvectors are;

[0028] Sort the calculation results from high to low, filter out the top-ranked cases, and for the filtered cases, give the diagnosis reasons based on the historical diagnosis results to preliminarily determine the current fault range.

[0029] Preferably, in step S4, data collection is performed through vibration sensors, temperature sensors, and pressure sensors. After preprocessing the collected data, features related to the fault are extracted based on professional knowledge and experience. A combinatorial optimization model is constructed in the quantum annealing machine, and the extracted features are input to search for matching fault modes.

[0030] Preferably, step S4 specifically includes:

[0031] Assume that the operating data set of the hydropower unit is D'={d'1,d'2,…,d' n}, where d' i represents the i-th data sample;

[0032] Based on professional knowledge and experience, the feature extraction function f(D') is defined to extract the fault feature vector X=(x1,x2,…,x n ), that is, X = f(D'). For vibration data, define x1 as the vibration amplitude feature;

[0033] Extract features for different types of data and build k' single feature models M1, M2, ..., M k’ , each model M j Process a specific part of the feature vector X and output the intermediate result y' j , that is, y' j =M j (X);

[0034] Constructing the combination optimization model M comb , which integrates the outputs of multiple single models, and its objective function is to minimize the uncertainty of fault path identification;

[0035] Define the fault path Z=(z1,z2,…,z m ), where zi∈{0,1} represents the presence or absence of possible fault causes;

[0036] The combinatorial optimization problem is defined as: minF(Z)=M opt (y'1,y'2,…,y' k ,Z), where F(Z) represents the unreasonable degree of fault path Z under given characteristic analysis results;

[0037] Map the combinatorial optimization problem into a quantum Hamiltonian H(s) = (1-s)H0 + sH1;

[0038] Where H0 is the initial Hamiltonian, H1 is the problem Hamiltonian, and s is the annealing parameter; the problem Hamiltonian H1 is constructed as an energy function reflecting the optimization target F(Z);

[0039] Initialize the system to the ground state |ψ0> of H0;

[0040] By gradually adjusting s from 0 to 1, the system state evolves from |ψ0> to |ψ(s)> close to the H1 ground state;

[0041] The final quantum state is measured to obtain the fault path estimate Z*, which represents the most likely combination of fault causes.

[0042] Preferably, the iterative convergence step in step S5 is:

[0043] Repeat the quantum annealing machine to find the matching pattern and iterate. During each iteration, the direction of quantum state evolution is adjusted based on measurement feedback to converge the quantum state to the state representing the optimal fault path;

[0044] Set the convergence criteria. When the convergence criteria are reached, terminate the iteration process and output the currently measured path as the optimal fault path.

[0045] The convergence criteria include: when the change in the path cost value obtained from multiple consecutive measurements is less than a threshold, it is judged that the convergence criteria is met; when the measurement results converge to the same path within a proportional number of times, it is judged that the optimal fault path is converged.

[0046] Preferably, the iterative convergence condition is specifically expressed as follows: the number of iterations at convergence is Na, and the quantum bit state |ψ N >Corresponding to the optimal fault path quantum state within the preliminary screening fault range, the optimal fault path is determined by obtaining classical information from the quantum state. There exists a measurement function U(|ψ Na >), for |ψ Na >Measure and get classical information x opt , that is, x opt =U(|ψ Na >), x opt is the representation of the optimal fault path.

[0047] The neural network-based hydropower group linkage fault diagnosis system includes:

[0048] A case acquisition module configured to acquire historical case data;

[0049] a case analysis module configured to analyze the acquired case data;

[0050] The preliminary analysis module is configured to compare the current fault situation with the case data and preliminarily determine the range of the fault;

[0051] The quantum analysis module is configured to perform computational analysis based on a quantum annealing machine. After iteration and convergence of the iterative judgment module, the optimal fault path is obtained within the preliminary fault range and the fault judgment result is obtained.

[0052] Compared with the prior art, the present invention has the following beneficial effects:

[0053] The present invention builds a fault case library, collects fault cases from different regions, provides rich and real references for diagnosis, broadens diagnostic ideas, shortens diagnostic time, uses neural network analysis cases, explores deep-level fault characteristics, clarifies the complex relationship between fault characteristics and causes, improves diagnostic accuracy, avoids misdiagnosis and missed diagnosis, quickly narrows the scope of fault diagnosis, builds a combination optimization model and uses a quantum annealing machine to extract key fault characteristics, finds the optimal fault path more quickly, and improves diagnostic reliability. The overall system can diagnose faults efficiently and accurately, reduces the computing resources of the server, and quickly obtains fault judgment results. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 This is a flow chart of the diagnostic steps of the present invention;

[0055] Figure 2 This is a framework diagram of the fault diagnosis system of the present invention. DETAILED DESCRIPTION

[0056] The following description is intended to disclose the present invention so that those skilled in the art can implement the present invention. The preferred embodiments described below are merely examples, and those skilled in the art may conceive of other obvious variations.

[0057] Reference Figure 1 and Figure 2 As shown in the figure, the neural network-based fault diagnosis method for hydropower group linkage is as follows:

[0058] S1. Obtain historical cases of hydropower group linkage failures in different regions, organize the historical cases, and form a failure case database;

[0059] S2, learning the nonlinear relationship between hydropower unit characteristics and faults based on neural networks, and obtaining the correlation between hydropower unit characteristics and fault causes;

[0060] S3. Observe and obtain the fault characteristics existing in the current hydropower group linkage, calculate the similarity between the current fault characteristics and the characteristics in the fault case, select cases similar to the current fault, obtain historical case diagnosis results, use the historical diagnosis results as the preliminary diagnosis results of the current hydropower group linkage, and obtain the fault range interval;

[0061] S4. Collect the operating data of the hydropower unit, extract the key fault features after pre-processing, build a combination optimization model, input the features into the quantum annealing machine, and conduct in-depth calculation and analysis;

[0062] S5. After multiple iterations of convergence, the quantum state is converged to the optimal fault path in the fault range interval, and the cause of the current hydropower group linkage fault is determined in the fault range interval.

[0063] This application obtains historical cases of hydropower group linkage failures in different regions and builds a case library, which provides a rich and real reference basis for subsequent fault diagnosis. The case library can comprehensively cover fault situations under various complex working conditions, so that the diagnosis process no longer relies on single or limited experience, which greatly broadens the diagnostic thinking. When facing new faults, similar situations can be quickly searched from the case library, shortening the diagnosis time and improving the diagnosis efficiency. The neural network is used to learn the nonlinear relationship between the characteristics of the hydropower unit and the fault, breaking through the limitation of traditional methods that are difficult to capture complex characteristics and nonlinear correlations. The powerful learning ability of the neural network can accurately mine the hidden features in massive data. According to the deep fault characteristics in the data, the complex relationship between the fault characteristics and the fault causes is clarified, which makes the diagnosis more accurate, can deeply analyze the root cause of the fault, provide strong support for targeted maintenance, and effectively avoid maintenance delays and resource waste caused by misdiagnosis or missed diagnosis; calculate the similarity between the current fault characteristics and the characteristics in the case library, and select cases with high similarity as the preliminary diagnosis results, which can quickly narrow the scope of fault diagnosis. This method greatly improves the pertinence of diagnosis, reduces unnecessary troubleshooting work, and saves a lot of manpower, material resources and time costs. When facing complex linkage faults caused by multi-factor coupling, traditional methods often require comprehensive Various possible factors are checked, and this method can quickly lock the key areas through similarity screening, greatly improving the efficiency of fault location; after collecting the operating data of the hydropower unit and preprocessing it, a combined optimization model is constructed and the variables are represented by the quantum state of superconducting quantum bits with the help of a quantum annealing machine. The solution space is explored based on the quantum tunneling effect. This process can efficiently extract key fault features from complex data. The powerful computing power and unique quantum tunneling mechanism of the quantum annealing machine can search for the optimal solution faster than traditional computing methods, that is, the optimal fault path in the fault range interval that has been initially screened. The optimal fault path obtained after multiple iterations of convergence can It can accurately locate fault problems, improve the accuracy and reliability of fault diagnosis, and further ensure the stable operation of hydropower groups; the comprehensive application of this method and system can realize efficient and accurate diagnosis of hydropower group linkage faults, and by quickly locating the cause and scope of the fault, take effective maintenance measures in time, reduce the downtime of hydropower units, and reduce the economic losses caused by the fault. At the same time, it effectively guarantees the safe and stable operation of the power system and avoids large-scale power outages caused by hydropower group faults. It is of great significance to maintain the reliability and stability of power supply, meet the growing development needs of the hydropower industry, and promote the sustainable development of the hydropower industry.

[0064] The purpose of obtaining the correlation in step S2 of this application is to clarify the correlation between the characteristics and the cause of the fault, monitor the relevant characteristic parameters in real time during the operation of the hydropower unit, and once abnormal changes in certain characteristic parameters are found, the impending fault can be predicted in advance based on the correlation, so that preventive measures can be taken, such as arranging maintenance and replacing parts in advance, to avoid the occurrence and further expansion of the fault, and reduce downtime and maintenance costs;

[0065] After understanding the correlation, a more scientific and reasonable maintenance plan is formulated based on the actual operating characteristics and failure risks of the unit. Key characteristic parameters with strong correlation with failures are monitored and regularly inspected. For different types of failures, maintenance tools, spare parts and other resources are prepared in a targeted manner to improve the efficiency and pertinence of maintenance work and avoid over-maintenance or under-maintenance.

[0066] It helps to deeply understand the operating characteristics of hydropower units. Based on the relationship between the characteristics and the causes of failures, the operating parameters of the units can be optimized to make the units operate in an efficient and safe state, thereby improving the overall performance and reliability of the units and extending the service life of the units.

[0067] The results of the neural network can provide a clear direction for fault diagnosis. When abnormal characteristics of the unit are monitored, the components or systems that may have faults can be quickly identified based on the learned correlations, narrowing the scope of fault investigation and improving the accuracy and efficiency of fault location.

[0068] In step S1, the method for obtaining historical fault cases is to obtain them from the enterprise fault record database, equipment maintenance and repair reports, and academic databases; the historical case organization method is to classify the cases through classification methods, including classification by fault type, classification by fault cause, and classification by region.

[0069] In this application, the method for obtaining historical fault cases is to obtain them from the enterprise fault record database, the equipment maintenance and repair report and the academic database; the historical case collation method classifies the cases through classification, and the classification methods include classification by fault type, classification by fault cause and classification by region; the fault case analysis method constructs a neural network model, determines the structure of the neural network, including the number of input layer nodes, the number of hidden layer nodes and the number of output layer nodes, inputs the obtained historical cases into the constructed neural network model, trains and optimizes the model, analyzes the weights and biases of the trained neural network model, finds the degree of influence of each feature on the fault cause, and explores the potential association rules between the characteristics of the hydropower units and the fault causes when the hydropower group fails.

[0070] Find the influence of each feature on the cause of the fault by calculating the sum of the absolute values of the weights from the input layer to the hidden layer and from the hidden layer to the output layer. The weight of the i-th feature of the input layer to the j-th neuron of the l-th hidden layer is The weight from the jth neuron in the lth hidden layer to the kth neuron in the output layer is The influence degree of the i-th feature on the k-th neuron in the output layer is I ik Expressed as:

[0071]

[0072] Where L is the number of hidden layers, n l is the number of neurons in the lth hidden layer, by calculating I ik Quantify the impact of each feature on different fault cause categories, and then mine the potential association rules between hydropower unit features and fault causes when hydropower group linkage failure occurs. ik If the value is large, it means that the i-th feature plays a role in judging the cause of the k-th fault.

[0073] By quantifying the impact of each feature on different fault cause categories, this application can deeply explore the potential association rules between the characteristics of hydropower units and the fault causes when hydropower cluster linkage faults occur. For example, if a specific electrical parameter feature is found to have a great impact on the fault category caused by design defects in multiple faults, then in the subsequent equipment design improvement, operation monitoring and maintenance, we can focus on this electrical parameter to prevent such faults from occurring in advance, effectively reduce the fault incidence rate, and ensure the stable operation of the hydropower cluster.

[0074] The similarity calculation steps in step S3 are as follows: based on the obtained correlation degree between the hydropower unit characteristics and the fault causes, a weight value w1 is assigned to the corresponding characteristics;

[0075] The acquired features are vectorized based on discretization processing, and the vectorized data are calculated;

[0076] Assume that the current fault feature vector is X=(x1,x2,…,x n ), the eigenvector in the fault case is Y=(y1,y2,…,y n ), similarity value d E The formula for (X,Y) is:

[0077]

[0078] where d E (X, Y) is the calculated similarity value, d E The smaller the (X,Y) value, the more similar the two eigenvectors are;

[0079] The smaller the calculated Euclidean distance, the more similar the two feature vectors are. After the calculation is completed, the similarity values are sorted from high to low, and the cases with the highest similarity rankings are selected. For these selected cases, the diagnosis causes are given and the current fault range is preliminarily determined.

[0080] The weight value allocation method is the hierarchical analysis method, which is an existing technology and will not be described in detail here.

[0081] The Euclidean distance of this application is rigorous and intuitive in mathematical principles, and can accurately reflect the true distance between two points in vector space. In the analysis of hydropower group fault characteristics, it can quantify complex fault characteristics into specific numerical differences. When processing multi-dimensional fault characteristics such as vibration frequency and voltage fluctuation, the Euclidean distance can comprehensively consider the differences in each dimension and accurately derive the similarity of two feature vectors, freeing fault feature matching from subjective and fuzzy judgments and greatly improving diagnostic accuracy.

[0082] In step S4, the operating data of the hydropower unit is collected based on vibration sensors, temperature sensors and pressure sensors, and the data is preprocessed, including removing noise, filling missing values and cleaning and correcting outliers. Based on professional knowledge and experience, the features related to the fault are extracted to build a combined optimization model. After combining multiple single feature models, the extracted features are input into the quantum annealing machine, and the variables are represented by the quantum state of superconducting quantum bits. Based on the quantum tunneling effect, matching fault modes are searched.

[0083] In the data collection process of this application, the vibration sensor adopts the principle of piezoelectric effect and is tightly installed in the key mechanical components, bearings, and shaft positions of the hydropower unit. When the component vibrates, the piezoelectric material inside the sensor generates a charge signal proportional to the vibration amplitude, thereby accurately capturing the vibration of the mechanical components of the unit. Vibration changes as subtle as the micron level can be sensed. These vibration data can intuitively reflect the hidden dangers of mechanical wear and imbalance faults; temperature sensors mostly use high-precision thermocouples or thermal resistor sensors, which are deployed at key hot spots in the electrical windings and mechanical friction parts of the equipment;

[0084] The collected data often have problems with noise, missing values and outliers, so careful data preprocessing is required. In terms of noise removal, the wavelet transform algorithm is used. This algorithm can decompose the data into different frequency bands, effectively filter out high-frequency noise signals caused by external electromagnetic interference and sensor errors, so that the data can truly reflect the operating status of the unit. For filling missing values, a method based on the K-nearest neighbor algorithm is used to find the K neighbor samples most similar to the missing value samples, and use the data features of the neighbor samples for weighted averaging to fill the missing values, ensuring data integrity and avoiding the impact of missing key data on subsequent analysis. For cleaning and correcting outliers, the isolation forest algorithm is used. This algorithm can quickly identify outliers in the data and regard them as outliers. Then, they are corrected through statistical methods or model-based predictions to prevent extreme erroneous data from misleading the diagnosis direction, providing a high-quality data foundation for subsequent feature extraction.

[0085] Step S4 specifically includes:

[0086] Assume that the operating data set of the hydropower unit is D'={d'1,d'2,…,d' n}, where d' i represents the i-th data sample;

[0087] Based on professional knowledge and experience, the feature extraction function f(D') is defined to extract the fault feature vector X=(x1,x2,…,x n ), that is, X = f(D'). For vibration data, define x1 as the vibration amplitude feature;

[0088] Extract features for different types of data and build k' single feature models M1, M2, ..., M k’ , each model M j Process a specific part of the feature vector X and output the intermediate result y' j , that is, y' j =M j (X);

[0089] Constructing the combination optimization model M comb , which integrates the outputs of multiple single models, and its objective function is to minimize the uncertainty of fault path identification;

[0090] Define the fault path Z=(z1,z2,…,z m ), where zi∈{0,1} represents the presence or absence of possible fault causes;

[0091] The combinatorial optimization problem is defined as: minF(Z)=M opt (y'1,y'2,…,y' k,Z), where F(Z) represents the unreasonable degree of fault path Z under given characteristic analysis results;

[0092] Map the combinatorial optimization problem into a quantum Hamiltonian H(s) = (1-s)H0 + sH1;

[0093] Where H0 is the initial Hamiltonian, H1 is the problem Hamiltonian, and s is the annealing parameter; the problem Hamiltonian H1 is constructed as an energy function reflecting the optimization target F(Z);

[0094] Initialize the system to the ground state |ψ0> of H0;

[0095] By gradually adjusting s from 0 to 1, the system state evolves from |ψ0> to |ψ(s)> close to the H1 ground state;

[0096] The final quantum state is measured to obtain the fault path estimate Z*, which represents the most likely combination of fault causes.

[0097] During the data collection stage, this application uses vibration, temperature, and pressure sensors to collect operating data of hydropower units from different dimensions. Multi-source data fusion comprehensively presents the operating status of the unit, laying the foundation for diagnosis. In the data preprocessing stage, noise is removed, missing values are filled, and outliers are cleaned and corrected to purify the data, ensure data quality, and improve the accuracy of subsequent feature extraction and the reliability of the diagnostic process. In feature extraction, based on professional knowledge and experience, key features are extracted from vibration and temperature data, focusing on the core factors of the fault, supporting the construction of efficient diagnostic models, and constructing a combined optimization model to integrate the advantages of a single feature model to cover various fault scenarios; using a quantum annealing machine, superconducting quantum bit quantum states are used to represent variables, and with the help of the quantum tunneling effect, fault modes are efficiently searched and matched in the huge solution space, which greatly shortens the diagnosis time, improves positioning accuracy, and enhances the ability of hydropower groups to respond to sudden faults.

[0098] The iterative convergence steps in step S5 are:

[0099] Repeat the quantum annealing machine to find the matching pattern and iterate. During each iteration, the direction of quantum state evolution is adjusted based on measurement feedback to converge the quantum state to the state representing the optimal fault path;

[0100] Set the convergence criteria. When the convergence criteria are reached, terminate the iteration process and output the currently measured path as the optimal fault path.

[0101] The convergence criteria include: when the change in the path cost value obtained from multiple consecutive measurements is less than a threshold, it is judged that the convergence criteria is met; when the measurement results converge to the same path within a proportional number of times, it is judged that the optimal fault path is converged.

[0102] At the beginning of each iteration of this application, the quantum bit crosses the energy barrier based on the quantum tunneling probability. The quantum tunneling probability is determined by many characteristics of the quantum system, including the initial state of the quantum bit, the energy environment it is in, and the Hamiltonian of the system. This probability is not set randomly, but follows the basic principles of quantum mechanics. Under a specific quantum bit state, the distribution of the energy barrier around it will be mapped to the tunneling probability through a complex mathematical model. When the energy barrier is low, the probability of the quantum bit crossing is relatively high; conversely, when the barrier is high, the probability of crossing is low.

[0103] The convergence condition of the iteration is: let the number of iterations at convergence be Na, at which time the quantum bit state |ψ N >Corresponding to the optimal fault path quantum state within the initial screening fault range, classical information is obtained from this quantum state to determine the optimal fault path. There exists a measurement function U(|ψ Na >), by |ψ Na > Make a measurement and get the classical information x opt , that is, x opt =U(|ψ Na >), x opt is the representation of the optimal fault path.

[0104] The convergence condition of the iteration of this application is: suppose the number of iterations at convergence is , at this time the quantum bit state corresponds to the optimal fault path quantum state within the preliminary screening fault range interval, and classical information is obtained from this quantum state to determine the optimal fault path. There is a measurement function, and by measuring , the classical information is obtained, which is the representation of the optimal fault path.

[0105] The neural network-based hydropower group linkage fault diagnosis system includes:

[0106] A case acquisition module configured to acquire historical case data;

[0107] a case analysis module configured to analyze the acquired case data;

[0108] The preliminary analysis module is configured to compare the current fault situation with the case data and preliminarily determine the range of the fault;

[0109] The quantum analysis module is configured to perform computational analysis based on a quantum annealing machine. After iteration and convergence of the iterative judgment module, the optimal fault path is obtained within the preliminary fault range and the fault judgment result is obtained.

[0110] The basic principles, main features, and advantages of the present invention are shown and described above. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions only describe the principles of the present invention. Various changes and improvements are possible without departing from the spirit and scope of the present invention, and such changes and improvements fall within the scope of the invention as claimed.

Claims

1. A method for fault diagnosis of hydropower generator groups based on neural network, characterized in that: The diagnostic steps are: S1. Obtain historical cases of hydropower group linkage failures in different regions, organize the historical cases, and form a failure case database; S2, learning the nonlinear relationship between hydropower unit characteristics and faults based on neural networks, and obtaining the correlation between hydropower unit characteristics and fault causes; S3. Observe and obtain the fault characteristics existing in the current hydropower group linkage, calculate the similarity between the current fault characteristics and the characteristics in the fault case, select cases similar to the current fault, obtain historical case diagnosis results, use the historical diagnosis results as the preliminary diagnosis results of the current hydropower group linkage, and obtain the fault range interval; S4. Collect the operating data of the hydropower unit, extract the key fault features after pre-processing, build a combination optimization model, input the features into the quantum annealing machine, and conduct in-depth calculation and analysis; S5. After multiple iterations of convergence, the quantum state is converged to the optimal fault path in the fault range interval, and the cause of the current hydropower group linkage fault is determined in the fault range interval.

2. The neural network-based hydropower group linkage fault diagnosis method according to claim 1 is characterized in that: The sources of fault history cases in step S1 include enterprise fault record database, equipment maintenance and repair reports, and academic databases; Classify historical fault cases to form a fault case library; Classification methods include: classification by fault type, classification by fault cause and classification by region.

3. The neural network-based hydropower group linkage fault diagnosis method according to claim 1 is characterized in that: The steps for building the neural network model in step S2 are: Determine the problem to be solved and the characteristics of the input and output; Determine the neural network structure, including the number of nodes in the input layer, the number of nodes in the hidden layer, and the number of nodes in the output layer; Input the acquired historical cases into the neural network model, train and optimize the model, initialize the weights and bias terms, perform forward propagation, calculate the loss, and backpropagate to update the parameters, repeating until convergence; Find out the influence of each feature on the fault cause, and explore the potential correlation between the characteristics of hydropower units and the fault cause when the hydropower group linkage fault occurs.

4. The neural network-based hydropower group linkage fault diagnosis method according to claim 3 is characterized in that: The steps of association mining are: Calculate the sum of the absolute values of the weights from the input layer to the hidden layer and from the hidden layer to the output layer. The weight from the i-th feature of the input layer to the j-th neuron of the l-th hidden layer is The weight from the jth neuron in the lth hidden layer to the kth neuron in the output layer is The influence degree of the i-th feature on the k-th neuron in the output layer is I ik Expressed as: Where L is the number of hidden layers, n l is the number of neurons in the lth hidden layer, by calculating I ik Quantify the impact of each feature on different fault cause categories, explore the potential correlation between hydropower unit features and fault causes when hydropower group linkage failure occurs, and calculate the results I ik A large value indicates that the i-th feature plays a role in determining the cause of the k-th fault.

5. The neural network-based fault diagnosis method for hydropower group according to claim 1 is characterized in that: The similarity calculation steps in step S3 are: Based on the correlation between the hydropower unit characteristics and the fault causes, a weight value w1 is assigned to the corresponding characteristics; The acquired features are vectorized based on discretization processing, and the vectorized data are calculated; Assume that the current fault feature vector is X=(x1,x2,…,x n ), the eigenvector in the fault case is Y=(y1,y2,…,y n ), similarity value d E The formula for (X,Y) is: where d E (X, Y) is the calculated similarity value, d E The smaller the (X,Y) value, the more similar the two eigenvectors are; Sort the calculation results from high to low, filter out the top-ranked cases, and for the filtered cases, give the diagnosis reasons based on the historical diagnosis results to preliminarily determine the current fault range.

6. The neural network-based hydropower group linkage fault diagnosis method according to claim 1 is characterized in that: In step S4, data collection is performed through vibration sensors, temperature sensors, and pressure sensors. After preprocessing the collected data, fault-related features are extracted based on professional knowledge and experience. A combinatorial optimization model is constructed in the quantum annealing machine, and the extracted features are input to search for matching fault patterns.

7. The neural network-based hydropower group linkage fault diagnosis method according to claim 6 is characterized in that: Step S4 specifically includes: Assume that the operating data set of the hydropower unit is D'={d'1,d'2,…,d' n }, where d' i represents the i-th data sample; Based on professional knowledge and experience, the feature extraction function f(D') is defined to extract the fault feature vector X=(x1,x2,…,x n ), that is, X = f(D'). For vibration data, define x1 as the vibration amplitude feature; Extract features for different types of data and build k' single feature models M1, M2, ..., M k’ , each model M j Process a specific part of the feature vector X and output the intermediate result y' j , that is, y' j =M j (X); Constructing the combination optimization model M comb , which integrates the outputs of multiple single models, and its objective function is to minimize the uncertainty of fault path identification; Define the fault path Z=(z1,z2,…,z m ), where z i ∈{0,1} indicates the presence or absence of a possible fault cause; The combinatorial optimization problem is defined as: minF(Z)=M opt (y'1,y'2,…,y' k ,Z), where F(Z) represents the unreasonable degree of fault path Z under given characteristic analysis results; Map the combinatorial optimization problem into a quantum Hamiltonian H(s) = (1-s)H0 + sH1; Where H0 is the initial Hamiltonian, H1 is the problem Hamiltonian, and s is the annealing parameter; the problem Hamiltonian H1 is constructed as an energy function reflecting the optimization target F(Z); Initialize the system to the ground state |ψ0> of H0; By gradually adjusting s from 0 to 1, the system state evolves from |ψ0> to |ψ(s)> close to the H1 ground state; The final quantum state is measured to obtain the fault path estimate Z*, which represents the most likely combination of fault causes.

8. The neural network-based hydropower group linkage fault diagnosis method according to claim 1 is characterized in that: The iterative convergence steps in step S5 are: Repeat the quantum annealing machine to find the matching pattern and iterate. During each iteration, the direction of quantum state evolution is adjusted based on measurement feedback to converge the quantum state to the state representing the optimal fault path; Set the convergence criteria. When the convergence criteria are reached, terminate the iteration process and output the currently measured path as the optimal fault path. The convergence criteria include: when the change in the path cost value obtained from multiple consecutive measurements is less than a threshold, it is judged that the convergence criteria is met; when the measurement results converge to the same path within a proportional number of times, it is judged that the optimal fault path is converged.

9. The neural network-based hydropower group linkage fault diagnosis method according to claim 8 is characterized in that: The specific convergence condition of the iteration is: the number of iterations at convergence is Na, and the quantum bit state |ψ N >Corresponding to the optimal fault path quantum state within the preliminary screening fault range, the optimal fault path is determined by obtaining classical information from the quantum state. There exists a measurement function U(|ψ Na >), for |ψ Na >Measure and get classical information x opt , that is, x opt =U(|ψ Na >), x opt is the representation of the optimal fault path.

10. The neural network-based hydropower group linkage fault diagnosis system is characterized by: include: A case acquisition module configured to acquire historical case data; a case analysis module configured to analyze the acquired case data; The preliminary analysis module is configured to compare the current fault situation with the case data and preliminarily determine the range of the fault; The quantum analysis module is configured to perform computational analysis based on a quantum annealing machine. After iteration and convergence of the iterative judgment module, the optimal fault path is obtained within the preliminary fault range and the fault judgment result is obtained.

Images