Hydropower station drainage system adaptive diagnosis method based on artificial intelligence

Abstract

The invention discloses a hydropower station drainage system adaptive diagnosis method based on artificial intelligence, and relates to the technical field of hydropower stations, and the method comprises the steps: deploying a plurality of sensors in a drainage system, including a flow sensor, a pressure sensor and a temperature sensor, for collecting the operation data of the drainage system in real time, and constructing a drainage system operation data set; and the drainage system operation data set is transmitted to a central data processing unit through a sensor network. The method has the beneficial effects that fault prediction and diagnosis are performed by using the deep learning model combining the graph neural network and the Transform, and the operation parameters are optimized in real time in combination with the adaptive control algorithm, so that intelligent monitoring and adaptive control of the drainage system are realized, and the overall performance and reliability of the system are improved.

Description

Technical Field

[0001] The invention relates to the technical field of hydropower stations, and in particular to an artificial intelligence-based adaptive diagnosis method for a drainage system of a hydropower station. Background Art

[0002] As an important form of renewable energy generation, the operational stability and efficiency of hydropower stations are directly related to the reliability of power supply. The drainage system is one of the key auxiliary systems of hydropower stations, responsible for handling excess water generated during the operation of the hydropower station and preventing water accumulation from damaging the equipment. Traditional hydropower station drainage systems usually rely on manual inspections and basic automated monitoring methods, which have significant limitations and defects.

[0003] The monitoring of traditional drainage systems mainly relies on preset fixed parameters and simple alarm mechanisms. This method often lags behind in dealing with sudden failures and anomalies, making it difficult to respond in a timely manner. Specifically, the traditional method has the following shortcomings:

[0004] Traditional drainage systems mainly rely on regular manual inspections and simple automated monitoring, which makes it difficult to achieve real-time monitoring and early warning. When a system failure occurs, it is often impossible to detect and handle it in time, which can easily lead to drainage system blockage, leakage and other problems, which may further cause equipment damage and downtime.

[0005] Traditional drainage system monitoring is mainly based on simple data processing methods, which cannot effectively process and analyze large amounts of real-time data. Due to the lack of advanced data processing and analysis capabilities, the system cannot fully utilize operating data for fault prediction and diagnosis, resulting in low accuracy and reliability of fault detection.

[0006] Traditional drainage system control methods usually use fixed parameters or simple control strategies, which make it difficult to adaptively adjust parameters according to actual operating conditions. The operating environment of the drainage system is complex and changeable, and different parameter settings are required under different operating conditions to ensure the best performance of the system. Fixed parameter control methods often lead to low system operation efficiency and energy waste.

[0007] Traditional fault detection and diagnosis models for drainage systems are usually simple and fail to fully consider the multi-dimensional data and complex relationships during the operation of drainage systems. These models often fail to capture the dynamic characteristics and potential failure modes of the system, resulting in poor accuracy and timeliness of fault detection.

[0008] Therefore, how to provide an adaptive diagnosis method for the drainage system of a hydropower station based on artificial intelligence is an urgent problem to be solved by those skilled in the art. Summary of the invention

[0009] In view of the above problems or problems existing in the prior art, the present invention is proposed.

[0010] Therefore, the purpose of the present invention is to provide an artificial intelligence-based adaptive diagnosis method for a hydropower station drainage system, which can solve the problems mentioned in the background technology.

[0011] In order to solve the above technical problems, the present invention provides the following technical solutions: an artificial intelligence-based adaptive diagnosis method for a hydropower station drainage system, comprising S1, deploying a plurality of sensors in the drainage system, including a flow sensor, a pressure sensor and a temperature sensor, for real-time collection of operation data of the drainage system, and constructing a drainage system operation data set;

[0012] S2, transmitting the drainage system operation data set to the central data processing unit through the sensor network;

[0013] S3, preprocessing the collected drainage system operation data set, including data cleaning, noise reduction, missing value filling and feature extraction;

[0014] S4. Use the deep learning model combining graph neural network and Transformer to train historical key feature parameters and real-time key feature parameters to establish a drainage system health status model;

[0015] S5. Through the self-attention mechanism, combined with the prediction results of the drainage system health status model, the drainage system health status model is continuously trained;

[0016] S6. Based on the trained drainage system health status model, analyze real-time data, predict potential failures and abnormalities of the drainage system, and generate early warning signals;

[0017] S7. According to the prediction results of the drainage system health status model, the operation parameters of the drainage system are adaptively adjusted, and an adaptive control algorithm based on reinforcement learning is adopted to achieve dynamic optimization of the operation parameters of the drainage system;

[0018] S8. When a drainage system fault or abnormality is detected, a fault diagnosis report is automatically generated, which includes the fault type, cause analysis and treatment suggestions;

[0019] S9. Regularly perform self-inspections on the entire diagnostic system, evaluate the system's performance and accuracy, and optimize and update the system as needed.

[0020] As a preferred solution of the artificial intelligence-based adaptive diagnosis method for a hydropower station drainage system of the present invention, the S1 specifically includes:

[0021] S11. Deploy flow sensors, pressure sensors and temperature sensors at different locations of the drainage system to collect real-time operation data of the drainage system;

[0022] S12, the flow sensor is used to measure the flow rate Q(t) of the liquid in the drainage system, where Q represents the flow rate and t represents the time;

[0023] S13, the pressure sensor is used to measure the pressure P(t) in the drainage system, where P represents pressure and t represents time;

[0024] S14, the temperature sensor is used to measure the temperature T(t) in the drainage system, where T represents temperature and t represents time;

[0025] S15. Construct the collected flow, pressure and temperature data into a drainage system operation data set:

[0026] {(Q(t),,P(t),,T(t)),∣,t,∈,[0,,T max ]};

[0027] Among them, T max Indicates the maximum time range for data collection.

[0028] As a preferred solution of the artificial intelligence-based adaptive diagnosis method for the drainage system of a hydropower station of the present invention, the S3 specifically includes:

[0029] S31, cleaning the collected drainage system operation data set to remove duplicate data and invalid data;

[0030] S32, using Kalman filtering to perform noise reduction processing on the drainage system operation data set after data cleaning;

[0031] S33, estimating and supplementing the missing values ​​of the drainage system operation data set after the noise reduction process to generate a complete drainage system operation data set;

[0032] S34. Extract features from the complete drainage system operation data set, perform feature conversion on the flow data, pressure data and temperature data, and extract key feature parameters, including the mean value of the flow rate. The average flow rate Indicates that in the time range T max Average flow rate within:

[0033]

[0034] in, is the mean flow rate, T max The maximum time range for data collection;

[0035] Variance of pressure Variance of pressure Indicates that in the time range T max Variance of internal pressure:

[0036]

[0037] in, is the variance of pressure, is the mean value of pressure;

[0038] Frequency distribution of temperature f(T):

[0039]

[0040] Where f(T) is the frequency distribution of temperature and δ represents the Dirac function.

[0041] As a preferred solution of the artificial intelligence-based adaptive diagnosis method for a hydropower station drainage system of the present invention, the S4 specifically includes:

[0042] S41, dividing the key characteristic parameters in the preprocessed drainage system operation data set into historical key characteristic parameters and real-time key characteristic parameters;

[0043] S42. Construct a deep learning model based on a combination of multi-scale adaptive graph neural network and spatiotemporal Transformer to capture the relationship and spatiotemporal dependency between the characteristics of the hydropower station drainage system;

[0044] S43. In the multi-scale adaptive graph neural network part, the characteristic relationship of the drainage system at different scales is captured through multi-scale convolution operations. The multi-dimensional relationship between the flow, pressure and temperature data at different locations is captured through multi-scale convolution to reflect the dynamic characteristics of the drainage system:

[0045]

[0046] Among them, H (l+1) represents the node representation of the l+1th layer, σ represents the activation function, represents the set of neighbor nodes of node i at the kth scale, is the normalization constant at the kth scale, is the element in the adaptive adjacency matrix at the kth scale, W (l,k) and b (k) ) represent the weight matrix and bias vector of the k-th scale of the l-th layer respectively;

[0047] S44. In the spatiotemporal Transformer part, define the spatiotemporal self-attention mechanism to capture the long-term and short-term dependencies of the drainage system time series data and capture the changing rules of the drainage system data in the time dimension:

[0048] Z=LayerNorm(X+MultiHead(Q(t),P(t),T(t)));

[0049] MultiHead(Q(t),P(t),T(t))=Concat(h ead1,…,h ead h )W O ;

[0050]

[0051] Among them, Z represents the output of the spatiotemporal Transformer layer, LayerNorm represents layer normalization, and MultiHead represents the multi-head self-attention mechanism. Represent the weight matrices of output, query, key and value respectively, and Attention represents the self-attention mechanism;

[0052] S45. Combining multi-scale adaptive graph neural network and spatiotemporal Transformer, the MAGNN-STT model is formed, which comprehensively considers the multi-dimensional dependency relationship and time series characteristics between drainage system characteristics and provides optimized diagnostic results:

[0053] MAGNN-STT(X,A,Q(t),P(t),T(t))=α·MAGNN(X,A)+β·STT(Q(t),P(t),T(t));

[0054] Among them, α and β are adaptive weight parameters used to balance the output of the multi-scale adaptive graph neural network and the spatiotemporal Transformer;

[0055] S46. Use historical key feature parameters to train the MAGNN-STT model. Through training, optimize the MAGNN-STT model parameters so that the MAGNN-STT model can predict the health status and potential failures of the drainage system:

[0056]

[0057] Where N represents the number of training samples, y i Indicates the actual value, represents the predicted value, λ is the regularization parameter, θ and φ represent the model parameters of MAGNN and STT respectively;

[0058] S47. Based on the trained MAGNN-STT model, the real-time key characteristic parameters are analyzed and a drainage system health status model is established to predict potential failures and anomalies of the drainage system.

[0059] As a preferred solution of the artificial intelligence-based adaptive diagnosis method for a hydropower station drainage system of the present invention, the S5 specifically includes:

[0060] S51. Define the self-attention mechanism for the drainage system of hydropower stations, design a multi-level and multi-dimensional feature capture mechanism, and diagnose the drainage system under complex hydrological environment;

[0061] S52. Construct the input matrix of the self-attention mechanism, and use the historical key characteristic parameters and real-time key characteristic parameters of the drainage system as the rows and columns of the input matrix, including flow, pressure and temperature:

[0062]

[0063] S53. Design an adaptive weight matrix, combine environmental characteristics and seasonal changes, perform weighted processing on data from different time periods, and train the sensitivity of the drainage system health status model to seasonal fluctuations:

[0064] W = diag(ω1,ω2,ω3);

[0065] Among them, ω1, ω2, ω3 represent the adaptive weights of flow, pressure and temperature characteristics respectively;

[0066] S54. Use the self-attention mechanism to calculate the interdependence between drainage system features, and calculate the correlation between features through weighted dot product:

[0067]

[0068] Where Q, K and V represent the query matrix, key matrix and value matrix respectively, W is the adaptive weight matrix, d k Represents the dimension of the key;

[0069] S55. Combined with the multi-level attention mechanism, we design global and local feature capture modules. The global module captures the overall trend, and the local module captures mutations and anomalies:

[0070] Z global =GlobalAttention(X hist ,X real );

[0071] Z local =LocalAttention(X hist ,X real );

[0072] Among them, Z global and Z local Respectively represent the results of global and local feature capture;

[0073] S56. Through comprehensive analysis of global and local characteristics, the parameters of the drainage system health status model are dynamically updated so that the drainage system health status model can reflect the operating status and potential risks of the drainage system:

[0074]

[0075] Among them, θ represents the model parameters, η represents the learning rate, L represents the loss function, and λ is the regularization parameter.

[0076] As a preferred solution of the artificial intelligence-based adaptive diagnosis method for a hydropower station drainage system of the present invention, the S6 specifically includes:

[0077] S61. Use the trained drainage system health status model to input real-time key characteristic parameters, including flow pressure and temperature f(T) real ;

[0078] S62. Analyze the real-time data through the health status model and calculate the health score of the drainage system:

[0079]

[0080] Where HS(t) represents the health score at time t, α, β, and γ are the weight parameters obtained in model training;

[0081] S63, comparing the health score HS(t) with a preset threshold, when the health score is lower than the threshold, it is determined as a potential fault or abnormality, and an early warning signal is generated;

[0082] S64. Classify the generated early warning signals to identify the type and severity of potential failures:

[0083]

[0084] Among them, θ1, θ2 and θ3 are preset thresholds used to classify the risk level of the warning signal.

[0085] As a preferred solution of the artificial intelligence-based adaptive diagnosis method for a hydropower station drainage system of the present invention, the S7 specifically includes:

[0086] S71. Based on the prediction results of the drainage system health status model, identify key operating parameters that need to be optimized, including drainage rate, pump pressure, and water temperature adjustment parameters;

[0087] S72, construct an adaptive control algorithm and define the state space S, including the real-time state characteristics of the drainage system, such as the current drainage rate R(t), pump pressure P b (t) and water temperature T w (t) value and its changing trend:

[0088]

[0089] S73, define action space A, including multi-dimensional adjustment operations on drainage rate, pump pressure and water temperature parameters:

[0090] A={a R ,a P b,a T w|a R ∈[R min ,R max ],a P b∈[Pb min ,Pb max ],a T w∈[Tw min ,Tw max ]};

[0091] S74. Design an adaptive reward function R, taking into account the health score HS(t) of the drainage system, environmental factors and energy consumption, to evaluate the effect of the adjustment operation:

[0092] R(s,a)=λ1·HS(t)+λ2·η(t)-λ3·E(t)-λ4·C(t);

[0093] Among them, s represents the current state, a represents the current action, HS(t) is the health score, η(t) represents the operating efficiency of the drainage system, E(t) represents the energy consumption, C(t) represents the cost of the adjustment operation, and λ1, λ2, λ3, and λ4 are weight parameters;

[0094] S75. A reinforcement learning algorithm based on evolutionary strategy is used to optimize the adaptive control strategy and iteratively update the strategy parameter θ:

[0095]

[0096] Among them, α is the learning rate, π θ (a i |s i ) indicates that the strategy is in state s i Next select action a i The probability of represents the gradient of the policy parameters, and n is the number of samples;

[0097] S76. Through continuous iteration and environmental interaction, the evolutionary strategy enables the drainage system to adaptively adjust operating parameters to achieve optimal performance under different operating conditions;

[0098] S77. Based on the optimized adaptive control strategy, the operating parameters of the drainage system are monitored and adjusted in real time, and environmental feedback is used for continuous optimization to train the system's intelligence level and adaptive capabilities.

[0099] Beneficial effects of the present invention: The present invention utilizes a deep learning model combining a graph neural network and a Transformer to perform fault prediction and diagnosis, and combines an adaptive control algorithm to optimize operating parameters in real time, thereby realizing intelligent monitoring and adaptive control of the drainage system and improving the overall performance and reliability of the system.

[0100] A variety of sensors are used to collect the operation data of the drainage system in real time, and a complete drainage system operation data set is constructed to ensure the real-time and accuracy of the data, effectively improving the monitoring capability of the system. By using a deep learning model that combines graph neural networks and Transformer, historical data and real-time data are analyzed to establish a drainage system health status model. The drainage system health status model can be continuously trained through a self-attention mechanism to improve the accuracy and robustness of fault prediction, enabling the system to detect potential faults and anomalies at an early stage and generate early warning signals, greatly reducing the probability of failures, maintenance costs and system downtime.

[0101] The present invention utilizes a deep learning model that combines a multi-scale adaptive graph neural network and a spatiotemporal Transformer to capture the relationships and spatiotemporal dependencies between the characteristics of a hydropower station’s drainage system. Through a multi-level and multi-dimensional feature capture mechanism, it effectively processes large-scale real-time data, ensuring the accuracy and reliability of the diagnostic results. The application of intelligent models enhances the adaptability and flexibility of the system under different operating conditions.

[0102] The present invention uses an adaptive control algorithm based on reinforcement learning to adjust the operating parameters of the drainage system in real time, thereby realizing dynamic optimization of the system operating parameters. It adopts evolutionary strategies and deep Q networks for training and optimization, and optimizes the adaptive control strategy by iteratively updating the strategy parameters. The adaptive control algorithm designs an adaptive reward function, which comprehensively considers the health score, operating efficiency, energy consumption and cost of adjustment operations of the drainage system, so that the system can achieve optimal performance under different operating conditions. Through continuous iteration and environmental interaction, the system can autonomously learn and optimize, thereby improving the level of intelligence and adaptive capabilities. BRIEF DESCRIPTION OF THE DRAWINGS

[0103] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0104] Figure 1 It is a flow chart of the present invention.

[0105] Figure 2 It is a structural block diagram of the MAGNN-STT model of the present invention. DETAILED DESCRIPTION

[0106] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.

[0107] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0108] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor is it a separate or selective embodiment that is mutually exclusive with other embodiments.

[0109] Example 1

[0110] Reference Figure 1-2 , which is the first embodiment of the present invention, and which provides an artificial intelligence-based adaptive diagnosis method for a hydropower station drainage system, comprising S1, deploying a plurality of sensors in the drainage system, including a flow sensor, a pressure sensor, and a temperature sensor, for real-time collection of operation data of the drainage system, and constructing a drainage system operation data set;

[0111] S2, transmitting the drainage system operation data set to the central data processing unit through the sensor network;

[0112] S3, preprocessing the collected drainage system operation data set, including data cleaning, noise reduction, missing value filling and feature extraction;

[0113] S4. Use the deep learning model combining graph neural network and Transformer to train historical key feature parameters and real-time key feature parameters to establish a drainage system health status model;

[0114] S5. Through the self-attention mechanism, combined with the prediction results of the drainage system health status model, the drainage system health status model is continuously trained;

[0115] S6. Based on the trained drainage system health status model, analyze real-time data, predict potential failures and abnormalities of the drainage system, and generate early warning signals;

[0116] S7. According to the prediction results of the drainage system health status model, the operation parameters of the drainage system are adaptively adjusted, and an adaptive control algorithm based on reinforcement learning is adopted to achieve dynamic optimization of the operation parameters of the drainage system;

[0117] S8. When a drainage system fault or abnormality is detected, a fault diagnosis report is automatically generated, which includes the fault type, cause analysis and treatment suggestions;

[0118] S9. Regularly perform self-inspections on the entire diagnostic system, evaluate the system's performance and accuracy, and optimize and update the system as needed.

[0119] In this implementation, S1 specifically includes:

[0120] S11. Deploy flow sensors, pressure sensors and temperature sensors at different locations of the drainage system to collect real-time operation data of the drainage system;

[0121] S12, the flow sensor is used to measure the flow rate Q(t) of the liquid in the drainage system, where Q represents the flow rate and t represents the time;

[0122] S13, the pressure sensor is used to measure the pressure P(t) in the drainage system, where P represents pressure and t represents time;

[0123] S14, the temperature sensor is used to measure the temperature T(t) in the drainage system, where T represents temperature and t represents time;

[0124] S15. Construct the collected flow, pressure and temperature data into a drainage system operation data set:

[0125] {(Q(t),,P(t),,T(t)),∣,t,∈,[0,,T max ]};

[0126] Among them, T max Indicates the maximum time range for data collection.

[0127] In this implementation, S3 specifically includes:

[0128] S31, cleaning the collected drainage system operation data set to remove duplicate data and invalid data;

[0129] S32, using Kalman filtering to perform noise reduction processing on the drainage system operation data set after data cleaning;

[0130] S33, estimating and supplementing the missing values ​​of the drainage system operation data set after the noise reduction process to generate a complete drainage system operation data set;

[0131] S34. Extract features from the complete drainage system operation data set, perform feature conversion on the flow data, pressure data and temperature data, and extract key feature parameters, including the mean value of the flow rate. The average flow Indicates that in the time range T max Average flow rate within:

[0132]

[0133] in, is the mean flow rate, T max The maximum time range for data collection;

[0134] Variance of pressure Variance of pressure Indicates that in the time range T max Variance of internal pressure:

[0135]

[0136] in, is the variance of pressure, is the mean value of pressure;

[0137] Frequency distribution of temperature f(T):

[0138]

[0139] Where f(T) is the frequency distribution of temperature and δ represents the Dirac function.

[0140] In this implementation, S4 specifically includes:

[0141] S41, dividing the key characteristic parameters in the preprocessed drainage system operation data set into historical key characteristic parameters and real-time key characteristic parameters;

[0142] S42. Construct a deep learning model based on a combination of multi-scale adaptive graph neural network and spatiotemporal Transformer to capture the relationship and spatiotemporal dependency between the characteristics of the hydropower station drainage system;

[0143] S43. In the multi-scale adaptive graph neural network part, the characteristic relationship of the drainage system at different scales is captured through multi-scale convolution operations. The multi-dimensional relationship between the flow, pressure and temperature data at different locations is captured through multi-scale convolution to reflect the dynamic characteristics of the drainage system:

[0144]

[0145] Among them, H (l+1) represents the node representation of the l+1th layer, σ represents the activation function, represents the set of neighbor nodes of node i at the kth scale, is the normalization constant at the kth scale, is the element in the adaptive adjacency matrix at the kth scale, W (l,k) and b (k) ) represent the weight matrix and bias vector of the k-th scale of the l-th layer respectively;

[0146] S44. In the spatiotemporal Transformer part, define the spatiotemporal self-attention mechanism to capture the long-term and short-term dependencies of the drainage system time series data and capture the changing rules of the drainage system data in the time dimension:

[0147] Z=LayerNorm(X+MultiHead(Q(t),P(t),T(t)));

[0148] MultiHead(Q(t),P(t),T(t))=Concat(h ead1,…,h ead h )W O ;

[0149]

[0150] Among them, Z represents the output of the spatiotemporal Transformer layer, LayerNorm represents layer normalization, and MultiHead represents the multi-head self-attention mechanism. Represent the weight matrices of output, query, key and value respectively, and Attention represents the self-attention mechanism;

[0151] S45. Combining multi-scale adaptive graph neural network and spatiotemporal Transformer, the MAGNN-STT model is formed, which comprehensively considers the multi-dimensional dependency relationship and time series characteristics between drainage system characteristics and provides optimized diagnostic results:

[0152] MAGNN-STT(X,A,Q(t),P(t),T(t))=α·MAGNN(X,A)+β·STT(Q(t),P(t),T(t));

[0153] Among them, α and β are adaptive weight parameters used to balance the output of the multi-scale adaptive graph neural network and the spatiotemporal Transformer;

[0154] S46. Use historical key feature parameters to train the MAGNN-STT model. Through training, optimize the MAGNN-STT model parameters so that the MAGNN-STT model can predict the health status and potential failures of the drainage system:

[0155]

[0156] Where N represents the number of training samples, y i Indicates the actual value, represents the predicted value, λ is the regularization parameter, θ and φ represent the model parameters of MAGNN and STT respectively;

[0157] S47. Based on the trained MAGNN-STT model, the real-time key characteristic parameters are analyzed and a drainage system health status model is established to predict potential failures and anomalies of the drainage system.

[0158] In this implementation, S5 specifically includes:

[0159] S51. Define the self-attention mechanism for the drainage system of hydropower stations, design a multi-level and multi-dimensional feature capture mechanism, and diagnose the drainage system under complex hydrological environment;

[0160] S52. Construct the input matrix of the self-attention mechanism, and use the historical key characteristic parameters and real-time key characteristic parameters of the drainage system as the rows and columns of the input matrix, including flow, pressure and temperature:

[0161]

[0162] S53. Design an adaptive weight matrix, combine environmental characteristics and seasonal changes, perform weighted processing on data from different time periods, and train the sensitivity of the drainage system health status model to seasonal fluctuations:

[0163] W = diag(ω1,ω2,ω3);

[0164] Among them, ω1, ω2, ω3 represent the adaptive weights of flow, pressure and temperature characteristics respectively;

[0165] S54. Use the self-attention mechanism to calculate the interdependence between drainage system features, and calculate the correlation between features through weighted dot product:

[0166]

[0167] Where Q, K and V represent the query matrix, key matrix and value matrix respectively, W is the adaptive weight matrix, d k Represents the dimension of the key;

[0168] S55. Combined with the multi-level attention mechanism, we design global and local feature capture modules. The global module captures the overall trend, and the local module captures mutations and anomalies:

[0169] Z global =GlobalAttention(X hist ,X real );

[0170] Z local =LocalAttention(X hist ,X real );

[0171] Among them, Z global and Z local Respectively represent the results of global and local feature capture;

[0172] S56. Through comprehensive analysis of global and local characteristics, the parameters of the drainage system health status model are dynamically updated so that the drainage system health status model can reflect the operating status and potential risks of the drainage system:

[0173]

[0174] Among them, θ represents the model parameters, η represents the learning rate, L represents the loss function, and λ is the regularization parameter.

[0175] In this implementation, S6 specifically includes:

[0176] S61. Use the trained drainage system health status model to input real-time key characteristic parameters, including flow pressure and temperature f(T) real ;

[0177] S62. Analyze the real-time data through the health status model and calculate the health score of the drainage system:

[0178]

[0179] Where HS(t) represents the health score at time t, α, β, and γ are the weight parameters obtained in model training;

[0180] S63, comparing the health score HS(t) with a preset threshold, when the health score is lower than the threshold, it is determined as a potential fault or abnormality, and an early warning signal is generated;

[0181] S64. Classify the generated early warning signals to identify the type and severity of potential failures:

[0182]

[0183] Among them, θ1, θ2 and θ3 are preset thresholds used to classify the risk level of the warning signal.

[0184] In this implementation, S7 specifically includes:

[0185] S71. Based on the prediction results of the drainage system health status model, identify key operating parameters that need to be optimized, including drainage rate, pump pressure, and water temperature adjustment parameters;

[0186] S72, construct an adaptive control algorithm and define the state space S, including the real-time state characteristics of the drainage system, such as the current drainage rate R(t), pump pressure P b (t) and water temperature T w (t) value and its changing trend:

[0187]

[0188] S73, define action space A, including multi-dimensional adjustment operations on drainage rate, pump pressure and water temperature parameters:

[0189] A={a R ,a P b,a T w|a R ∈[R min ,R max ],a P b∈[Pb min ,Pb max ],a T w∈

[0190] [Tw min ,Tw max ]};

[0191] S74. Design an adaptive reward function R, taking into account the health score HS(t) of the drainage system, environmental factors and energy consumption, to evaluate the effect of the adjustment operation:

[0192] R(s,a)=λ1·HS(t)+λ2·η(t)-λ3·E(t)-λ4·C(t);

[0193] Among them, s represents the current state, a represents the current action, HS(t) is the health score, η(t) represents the operating efficiency of the drainage system, E(t) represents the energy consumption, C(t) represents the cost of the adjustment operation, and λ1, λ2, λ3, and λ4 are weight parameters;

[0194] S75. A reinforcement learning algorithm based on evolutionary strategy is used to optimize the adaptive control strategy and iteratively update the strategy parameter θ:

[0195]

[0196] Among them, α is the learning rate, π θ (a i |s i ) indicates that the strategy is in state s i Next select action a i The probability of represents the gradient of the policy parameters, and n is the number of samples;

[0197] S76. Through continuous iteration and environmental interaction, the evolutionary strategy enables the drainage system to adaptively adjust operating parameters to achieve optimal performance under different operating conditions;

[0198] S77. Based on the optimized adaptive control strategy, the operating parameters of the drainage system are monitored and adjusted in real time, and environmental feedback is used for continuous optimization to train the system's intelligence level and adaptive capabilities.

[0199] In large hydropower stations, traditional drainage systems mainly rely on regular manual inspections and simple automated monitoring, which makes it difficult to achieve real-time monitoring and fault prediction, resulting in low system efficiency, serious energy waste, and inability to handle sudden faults in a timely manner. This embodiment applies an artificial intelligence-based adaptive diagnosis method for a hydropower station drainage system to the drainage system of the hydropower station, and achieves fault prediction and system optimization through real-time monitoring and adaptive control, thereby improving the operating efficiency and reliability of the system.

[0200] In the drainage system of the hydropower station, multiple sensors, including flow sensors, pressure sensors and temperature sensors, are deployed to collect the operation data of the drainage system in real time. These data are transmitted to the central data processing unit through the sensor network to form a complete drainage system operation data set. In the data preprocessing, the collected drainage system operation data are cleaned, denoised, missing values ​​are filled and features are extracted. The Kalman filter algorithm is used to denoise the data, and the interpolation algorithm is used to fill the missing data. The key feature parameters extracted include the mean of the flow, the variance of the pressure and the frequency distribution of the temperature. The deep learning model combining the graph neural network and the Transformer is used to train the historical data and real-time data to establish the drainage system health status model. Through the self-attention mechanism, combined with the prediction results of the drainage system health status model, the model is continuously trained to improve the accuracy and robustness of fault prediction.

[0201] According to the prediction results of the model, the operating parameters of the drainage system are adaptively adjusted. Adopting the adaptive control algorithm based on reinforcement learning, the operating parameters of the drainage system are optimized in real time to ensure the best performance of the system under different operating conditions.

[0202] In this embodiment, the drainage system of a hydropower station was selected as the test object, and a six-month real-time monitoring and fault prediction test was conducted. The data collection frequency was once per minute, and more than 260,000 data records were collected. The specific data comparison is shown in Table 1 below:

[0203] Table 1 Comparison data of adaptive diagnosis methods for drainage systems in hydropower stations

[0204]

[0205]

[0206] As can be seen from Table 1, the traditional method only detected two faults in six months, while the method of the present invention detected five faults, and each fault was warned and handled at an early stage. Through the adaptive control algorithm, the response time of the method of the present invention when handling faults is significantly reduced, and the fault handling time is reduced from an average of 1.5 hours of the traditional method to about 20 minutes, and the downtime caused by the fault is also greatly reduced.

[0207] During the model training process, 5,000 historical data were selected for model training, and continuous training was carried out in combination with real-time data. The prediction accuracy of the model reached 95.21%, which is much higher than the 70% of the traditional method. The specific data comparison is shown in Table 2 below:

[0208] Table 2 Training sample comparison data

[0209]

[0210] By comparison, it can be seen that the method of the present invention has significant advantages in fault detection and processing, realizes intelligent monitoring and adaptive control of the drainage system, and greatly improves the operating efficiency and reliability of the system. It effectively solves the problems of insufficient real-time detection capability, limited data processing capability, lack of adaptive control, etc. in traditional methods, and provides a strong guarantee for the stable operation of the drainage system of the hydropower station.

[0211] The present invention uses a variety of sensors to collect the operation data of the drainage system in real time, builds a complete drainage system operation data set, ensures the real-time and accuracy of the data, and effectively improves the monitoring capability of the system. By using a deep learning model that combines graph neural networks and Transformers, historical data and real-time data are analyzed to establish a drainage system health status model. The drainage system health status model can be continuously trained through a self-attention mechanism to improve the accuracy and robustness of fault prediction, so that the system can detect potential faults and anomalies at an early stage, generate early warning signals, greatly reduce the probability of faults, and reduce maintenance costs and system downtime.

[0212] The present invention utilizes a deep learning model that combines a multi-scale adaptive graph neural network and a spatiotemporal Transformer to capture the relationships and spatiotemporal dependencies between the characteristics of a hydropower station’s drainage system. Through a multi-level and multi-dimensional feature capture mechanism, it effectively processes large-scale real-time data, ensuring the accuracy and reliability of the diagnostic results. The application of intelligent models enhances the adaptability and flexibility of the system under different operating conditions.

[0213] The present invention uses an adaptive control algorithm based on reinforcement learning to adjust the operating parameters of the drainage system in real time, thereby realizing dynamic optimization of the system operating parameters. It adopts evolutionary strategies and deep Q networks for training and optimization, and optimizes the adaptive control strategy by iteratively updating the strategy parameters. The adaptive control algorithm designs an adaptive reward function, which comprehensively considers the health score, operating efficiency, energy consumption and cost of adjustment operations of the drainage system, so that the system can achieve optimal performance under different operating conditions. Through continuous iteration and environmental interaction, the system can autonomously learn and optimize, thereby improving the level of intelligence and adaptive capabilities.

[0214] It is important to note that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. An adaptive diagnostic method for a hydropower station drainage system based on artificial intelligence, characterized in that: include, S1. Deploy multiple sensors in the drainage system, including flow sensors, pressure sensors, and temperature sensors, to collect the operation data of the drainage system in real time and build a drainage system operation data set; S2, transmitting the drainage system operation data set to the central data processing unit through the sensor network; S3, preprocessing the collected drainage system operation data set, including data cleaning, noise reduction, missing value filling and feature extraction; S4. Use the deep learning model combining graph neural network and Transformer to train historical key feature parameters and real-time key feature parameters to establish a drainage system health status model; S5. Through the self-attention mechanism, combined with the prediction results of the drainage system health status model, the drainage system health status model is continuously trained; S6. Based on the trained drainage system health status model, analyze real-time data, predict potential failures and abnormalities of the drainage system, and generate early warning signals; S7. According to the prediction results of the drainage system health status model, the operation parameters of the drainage system are adaptively adjusted, and an adaptive control algorithm based on reinforcement learning is adopted to achieve dynamic optimization of the operation parameters of the drainage system; S8. When a drainage system fault or abnormality is detected, a fault diagnosis report is automatically generated, which includes the fault type, cause analysis and treatment suggestions; S9. Regularly perform self-inspections on the entire diagnostic system, evaluate the system's performance and accuracy, and optimize and update the system as needed.

2. The artificial intelligence-based adaptive diagnosis method for a hydropower station drainage system according to claim 1, characterized in that: The S1 specifically includes: S11. Deploy flow sensors, pressure sensors and temperature sensors at different locations of the drainage system to collect real-time operation data of the drainage system; S12, the flow sensor is used to measure the flow rate Q(t) of the liquid in the drainage system, where Q represents the flow rate and t represents the time; S13, the pressure sensor is used to measure the pressure P(t) in the drainage system, where P represents pressure and t represents time; S14, the temperature sensor is used to measure the temperature T(t) in the drainage system, where T represents temperature and t represents time; S15. Construct the collected flow, pressure and temperature data into a drainage system operation data set: {(Q(t),、P(t),、T(t))、∣、t、∈、[0,、T max ]}; Among them, T max Indicates the maximum time range for data collection.

3. The artificial intelligence-based adaptive diagnosis method for a hydropower station drainage system according to claim 2, characterized in that: The S3 specifically includes: S31, cleaning the collected drainage system operation data set to remove duplicate data and invalid data; S32, using Kalman filtering to perform noise reduction processing on the drainage system operation data set after data cleaning; S33, estimating and supplementing the missing values ​​of the drainage system operation data set after the noise reduction process to generate a complete drainage system operation data set; S34. Extract features from the complete drainage system operation data set, perform feature conversion on the flow data, pressure data and temperature data, and extract key feature parameters, including the mean value of the flow rate. The average flow rate Indicates that in the time range T max Average flow rate within: in, is the mean flow rate, T max The maximum time range for data collection; Variance of pressure Variance of pressure Indicates that in the time range T max Variance of internal pressure: in, is the variance of pressure, is the mean value of pressure; Frequency distribution of temperature f(T): Where f(T) is the frequency distribution of temperature and δ represents the Dirac function.

4. The artificial intelligence-based adaptive diagnosis method for a hydropower station drainage system according to claim 3, characterized in that: The S4 specifically includes: S41, dividing the key characteristic parameters in the preprocessed drainage system operation data set into historical key characteristic parameters and real-time key characteristic parameters; S42. Construct a deep learning model based on a combination of multi-scale adaptive graph neural network and spatiotemporal Transformer to capture the relationship and spatiotemporal dependency between the characteristics of the hydropower station drainage system; S43. In the multi-scale adaptive graph neural network part, the characteristic relationship of the drainage system at different scales is captured through multi-scale convolution operations. The multi-dimensional relationship between the flow, pressure and temperature data at different locations is captured through multi-scale convolution to reflect the dynamic characteristics of the drainage system: Among them, H (l+1) represents the node representation of the l+1th layer, σ represents the activation function, represents the set of neighbor nodes of node i at the kth scale, is the normalization constant at the kth scale, is the element in the adaptive adjacency matrix at the kth scale, W (l,k) and b (k) ) represent the weight matrix and bias vector of the k-th scale of the l-th layer respectively; S44. In the spatiotemporal Transformer part, define the spatiotemporal self-attention mechanism to capture the long-term and short-term dependencies of the drainage system time series data and capture the changing rules of the drainage system data in the time dimension: Z=LayerNorm(X+MultiHead(Q(t),P(t),T(t))); MultiHead(Q(t),P(t),T(t))=Concat(head1,…,head h )W O ; Among them, Z represents the output of the spatiotemporal Transformer layer, LayerNorm represents layer normalization, MultiHead represents the multi-head self-attention mechanism, and W O , Represent the weight matrices of output, query, key and value respectively, and Attention represents the self-attention mechanism; S45. Combining multi-scale adaptive graph neural network and spatiotemporal Transformer, the MAGNN-STT model is formed, which comprehensively considers the multi-dimensional dependency relationship and time series characteristics between drainage system characteristics and provides optimized diagnostic results: MAGNN-STT(X,A,Q(t),P(t),T(t))=α·MAGNN(X,A)+β·STT(Q(t),P(t),T(t)); Among them, α and β are adaptive weight parameters used to balance the output of the multi-scale adaptive graph neural network and the spatiotemporal Transformer; S46. Use historical key feature parameters to train the MAGNN-STT model. Through training, optimize the MAGNN-STT model parameters so that the MAGNN-STT model can predict the health status and potential failures of the drainage system: Where N represents the number of training samples, y i Indicates the actual value, represents the predicted value, λ is the regularization parameter, θ and φ represent the model parameters of MAGNN and STT respectively; S47. Based on the trained MAGNN-STT model, the real-time key characteristic parameters are analyzed and a drainage system health status model is established to predict potential failures and anomalies of the drainage system.

5. The artificial intelligence-based adaptive diagnosis method for a hydropower station drainage system according to claim 4, characterized in that: The S5 specifically includes: S51. Define the self-attention mechanism for the drainage system of hydropower stations, design a multi-level and multi-dimensional feature capture mechanism, and diagnose the drainage system under complex hydrological environment; S52. Construct the input matrix of the self-attention mechanism, and use the historical key characteristic parameters and real-time key characteristic parameters of the drainage system as the rows and columns of the input matrix, including flow, pressure and temperature: S53. Design an adaptive weight matrix, combine environmental characteristics and seasonal changes, perform weighted processing on data from different time periods, and train the sensitivity of the drainage system health status model to seasonal fluctuations: W = diag(ω1,ω2,ω3); Among them, ω1, ω2, ω3 represent the adaptive weights of flow, pressure and temperature characteristics respectively; S54. Use the self-attention mechanism to calculate the interdependence between drainage system features, and calculate the correlation between features through weighted dot product: Where Q, K and V represent the query matrix, key matrix and value matrix respectively, W is the adaptive weight matrix, d k Represents the dimension of the key; S55. Combined with the multi-level attention mechanism, we design global and local feature capture modules. The global module captures the overall trend, and the local module captures mutations and anomalies: Z global =GlobalAttention(X hist ,X real ); Z local =LocalAttention(X hist ,X real ); Among them, Z global and Z local Respectively represent the results of global and local feature capture; S56. Through comprehensive analysis of global and local characteristics, the parameters of the drainage system health status model are dynamically updated so that the drainage system health status model can reflect the operating status and potential risks of the drainage system: Among them, θ represents the model parameters, η represents the learning rate, L represents the loss function, and λ is the regularization parameter.

6. The artificial intelligence-based adaptive diagnosis method for a hydropower station drainage system according to claim 5, characterized in that: The S6 specifically includes: S61. Use the trained drainage system health status model to input real-time key characteristic parameters, including flow pressure and temperature f(T) real ; S62. Analyze the real-time data through the health status model and calculate the health score of the drainage system: Where HS(t) represents the health score at time t, α, β, and γ are the weight parameters obtained in model training; S63, comparing the health score HS(t) with a preset threshold, when the health score is lower than the threshold, it is determined as a potential fault or abnormality, and an early warning signal is generated; S64. Classify the generated early warning signals to identify the type and severity of potential failures: Among them, θ1, θ2 and θ3 are preset thresholds used to classify the risk level of the warning signal.

7. The artificial intelligence-based adaptive diagnosis method for a hydropower station drainage system according to claim 6, characterized in that: The S7 specifically includes: S71. Based on the prediction results of the drainage system health status model, identify key operating parameters that need to be optimized, including drainage rate, pump pressure, and water temperature adjustment parameters; S72, construct an adaptive control algorithm and define the state space S, including the real-time state characteristics of the drainage system, such as the current drainage rate R(t), pump pressure P b (t) and water temperature T w (t) value and its changing trend: S73, define action space A, including multi-dimensional adjustment operations on drainage rate, pump pressure and water temperature parameters: A={a R ,a P b,a T w∣a R ∈[R min ,R max ],a P b∈[Pb min ,Pb max ],a T w∈[Tw min ,Tw max ]}; S74. Design an adaptive reward function R, taking into account the health score HS(t) of the drainage system, environmental factors and energy consumption, to evaluate the effect of the adjustment operation: R(s,a)=λ1·HS(t)+λ2·η(t)-λ3·E(t)-λ4·C(t); Among them, s represents the current state, a represents the current action, HS(t) is the health score, η(t) represents the operating efficiency of the drainage system, E(t) represents the energy consumption, C(t) represents the cost of the adjustment operation, and λ1, λ2, λ3, and λ4 are weight parameters; S75. A reinforcement learning algorithm based on evolutionary strategy is used to optimize the adaptive control strategy and iteratively update the strategy parameter θ: Among them, α is the learning rate, π θ (a i |s i ) indicates that the strategy is in state s i Next select action a i The probability of represents the gradient of the policy parameters, n is the number of samples; S76. Through continuous iteration and environmental interaction, the evolutionary strategy enables the drainage system to adaptively adjust operating parameters to achieve optimal performance under different operating conditions; S77. Based on the optimized adaptive control strategy, the operating parameters of the drainage system are monitored and adjusted in real time, and environmental feedback is used for continuous optimization to train the system's intelligence level and adaptive capabilities.

Images