Hydropower station data operation and maintenance method and system

Abstract

The invention discloses a hydropower station data operation and maintenance method, which comprises the following steps of: arranging a sensor network on hydraulic facilities and electromechanical gold junction equipment, acquiring data, preprocessing the data, registering in space and time scales, converting data signals into acoustic, optical and electric signals, and transmitting the acoustic, optical and electric signals to a server; the method comprises the following steps: performing signal fusion of a three-dimensional sonar signal and an RGB image based on an attention concentration mechanism, performing deep analysis on data, extracting a sensor data mode of hydraulic facilities and electromechanical junction equipment under normal working condition operation, establishing a fault diagnosis model, performing real-time diagnosis on the hydraulic facilities and the electromechanical junction equipment, and constructing a response decision knowledge base. Establishing a hydropower optimization scheduling model, and displaying the information of the hydropower station; the system integrates the Internet of Things, big data, artificial intelligence, cloud computing, BIM three-dimensional modeling and digital twinning technologies, realizes real-time monitoring, intelligent diagnosis and optimal scheduling of equipment and facilities of the hydropower station, and improves the overall digital intelligent operation management level of the hydropower station.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of data operation and maintenance method, and relates to a hydropower station data operation and maintenance method. BACKGROUND

[0002] Under the double carbon background, the construction of clean energy bases is accelerating, and new energy such as wind energy, solar energy and energy storage, and extra-high voltage power grids are developing rapidly. Pumped storage power stations are ushering in a new peak of development due to their flexible adjustment characteristics, and their unique advantages in peak shaving and phase modulation also bring many challenges to the traditional operation mode of hydropower stations. On the one hand, under the constraints of maximum new energy grid connection and maximum hydropower generation, hydropower dispatching requires high, on the other hand, frequent dispatching increases the operating load of hydropower stations, and it is of great importance to ensure the safe operation of facilities and equipment. Although the existing hydropower station operation platform realizes automatic control to a certain extent, there are still deficiencies in digitalization, integration and fine management, the safety risk exposure of operation and maintenance personnel is high, and it is unable to meet the management needs after the frequent intensification of hydropower dispatching after new energy grid connection. SUMMARY

[0003] The purpose of the present application is to provide a hydropower station data operation and maintenance method, which solves the problem of the inability of the prior art to meet the management needs after the frequent intensification of hydropower dispatching after new energy grid connection.

[0004] The second purpose of the present application is to provide a hydropower station data operation and maintenance system.

[0005] The first technical solution adopted by the present application is a hydropower station data operation and maintenance method, comprising the following steps: Step 1: arranging a sensor network on the hydraulic facilities and mechanical and electrical junction equipment and collecting data; Step 2: preprocessing the data, registering on the spatial and temporal scales, and converting the data signals into sound, light and electrical signals; Step 3: signal fusion of three-dimensional sonar signals and RGB images based on attention focusing mechanism; Step 4: deep analysis of the data to extract the sensor data mode of the hydraulic facilities and mechanical and electrical junction equipment under normal operating conditions; Step 5: establishing a fault diagnosis model to diagnose the hydraulic facilities and mechanical and electrical junction equipment in real time; Step 6: constructing a response decision knowledge base; Step 7: establishing a hydropower optimization dispatching model to display the information of the hydropower station.

[0006] The first technical solution of the present application is also characterized in that: The hydraulic facilities in step 1 include barrage, diversion tunnel, spillway stilling basin; the mechanical and electrical equipment includes water turbine, generator, transformer, high-voltage switch, relay protection device, excitation system, pressure steel pipe and gate.

[0007] Step 2 is specifically: the data collected by the sensor network is transmitted to the data acquisition terminal through the AES encryption transmission mode, the data is cleaned, denoised and normalized, invalid or abnormal data is removed, the data is registered in space and time scale, and the data signal is directly or converted and processed into sound, light and electric signal, and the basic state characteristics of the hydraulic facilities are obtained from the space, time, sound and light signals, and the basic state characteristics of the mechanical and electrical equipment are obtained from the space, time, light and electric signals.

[0008] Step 3 is specifically performed according to the following steps: Step 3.1, establish a sonar data set, and the sonar data set is represented as a point cloud , wherein each point p i ∈R 4 contains three-dimensional coordinates (x, y, z) and intensity value; the image data is represented as I∈R （H×W×3） , wherein H and W represent the height and width of the image respectively, and the units are mm; Step 3.2, use a hierarchical point cloud feature extraction network based on PointNet++, select key points by farthest point sampling, group point sets in a spherical neighborhood around each key point, extract local features for each point set using a multi-layer perceptron, and obtain key point features by maximum pooling aggregation; Step 3.3, repeat step 3.2 twice to obtain features of different scales, and unify the feature dimension to D s by 1D convolution, and output a feature matrix F s ∈R M ×D s , wherein M is the number of key points; Step 3.4, use a variant of MobileNetV2 to extract image features, extract spatial features through multiple convolution layers Conv2D and pooling layers, and obtain a feature map F i ∈R Hf×Wf×Di , wherein H f ×W f is the feature map size, Di is the feature dimension, the spatial dimension is flattened into L=H f ×W f positions, and F i ∈R L×Di is obtained. Step 3.5, the sonar point cloud coordinates p j =(x j ,y j ,z jProjecting onto the image plane and performing spatial alignment:

[0009] Among them, (u j ,v j () represents the coordinates of the projected image. This is the intrinsic parameter matrix. This is the extrinsic parameter matrix; sonar features F s Based on the corresponding positions assigned to the image feature map according to the projection coordinates, a sonar feature map F with the same resolution as the image feature map is obtained. s '∈R Hf×Wf×D For positions not projected, fill with zeros; Step 3.6: Align the sonar features and image features Unify the dimension to D through linear transformation: , , where Q s This is the query vector for sonar information. The query vector is for image information. This is the sonar feature projection matrix. The image signal feature projection matrix, , ; Sonar-to-image attention for:

[0010] Among them, K s V is the index vector of sonar information. s The feature representation of the retrieved sonar information, where T is the length of the time or spatial sequence; Using image features as the query and sonar features as the key and value, the output is: , Image-to-sonar attention for:

[0011] Among them, K i V is the index vector of image information. i The feature representation of the retrieved image information; Using sonar features as the query and image features as the key and value, the output is: ; Step 3.7, Gated fusion of the two attention outputs:

[0012]

[0013]

[0014]

[0015] wherein, is a sigmoid function, and Wg, bg are learnable parameters, is a sonar-to-image gating vector, is an image-to-sonar gating vector, and is the fused feature; Step 3.8, the fused feature and is spliced to obtain , and then multi-scale features are extracted through three parallel convolution paths:

[0016]

[0017]

[0018] The three features , , are spliced in the channel dimension, and are fused through 1x1 convolution:

[0019] wherein, is a multi-scale aggregated feature.

[0020] In step 4, data is analyzed in depth through a data mining algorithm.

[0021] In step 5, a fault diagnosis model of the hydraulic facility and the mechanical and electrical equipment is established through a machine learning algorithm. The faults of the hydraulic facility are classified into cracks, abrasion, and defects, and the faults of the mechanical and electrical equipment are classified into defects, loosening, deterioration, tip discharge, corona discharge, suspension discharge, air gap discharge, and surface discharge. The collected acoustic-optical data or photoelectric data are characterized and learned to realize the identification of the faults and propose a fault warning threshold and a warning mode.

[0022] In step 6, a response decision knowledge base under the warning mode is constructed according to the fault diagnosis type. The response decision knowledge base includes a facility and equipment abnormal feature library, a historical disposal case library, and a dynamic strategy generation engine. When a warning signal is detected, similar historical cases are matched through a graph neural network, and then a decision suggestion including a disposal priority, an operation step, and an expected impact is generated in combination with real-time equipment state data.

[0023] Step 7 is specifically performed according to the following steps: Step 7.1, considering the power generation plan under the grid connection of new energy, equipment state, power price, environmental constraints, set the objective function: Objective 1: Maximize power generation revenue For:

[0024] Wherein, is the processing power of t period, unit: MW, price t is the electricity price of t period, unit: yuan / MWh, Objective 2: Minimize ecological discharge For:

[0025]

[0026] Wherein, S t is the additional ecological discharge, unit: m 3 / s, Q t is the flow actually used for power generation, unit: m 3 / s, E t is the ecological flow requirement, unit: m 3 / s; Step 7.2, set the constraint condition, New energy grid connection constraint:

[0027] Equipment facility state constraint: Unit output limit:

[0028] is the approved output; Flow limit:

[0029] Ecological flow guarantee constraint:

[0030] Wherein, R t is the total discharge flow, is the ecological flow; Reservoir water balance:

[0031] Wherein, I t is the inflow, is, is; Reservoir capacity constraint:

[0032] Output calculation:

[0033] Wherein, η is the unit efficiency, H t The net water head, The gravity acceleration; A visual display platform is constructed to display information of the hydropower station.

[0034] The second technical solution adopted by the present application is a hydropower station data operation and maintenance system, which comprises an Internet of Things sensing layer, a data processing and sensing layer, a large model decision and control layer and a man-machine interaction and display layer connected in sequence.

[0035] The second technical solution of the present application is also characterized by: The Internet of Things sensing layer comprises a data acquisition terminal and a sensor network, the sensor network comprises a mobile laser radar acquisition dog, a mobile point cloud acquisition dog, a mobile sonar scanning dog, an infrared sensor, a temperature sensor, a vibration sensor, a current sensor, a voltage sensor, a pressure sensor and a microwave detector, the mobile laser radar acquisition dog, the mobile point cloud acquisition dog and the mobile sonar scanning dog regularly perform data acquisition and detection in a water diversion tunnel, a barrage, a water inlet and a tailrace, the infrared sensor and the temperature sensor are used for temperature monitoring of a generator rotor, a main transformer bushing and a high-voltage switch contact, the vibration sensor is used for fault monitoring of bearing wear, rotor imbalance and structural looseness of a hydraulic turbine and an excitation system, the current sensor and the voltage sensor are used for excitation current and excitation voltage waveform monitoring of an excitation system and secondary side current and voltage signal monitoring of a relay protection device, the pressure sensor is used for monitoring of water flow excitation and pipe wall stress vibration of a pressure steel pipe, and the microwave detector is used for structural deformation and displacement monitoring of a gate and a valve. The data processing and storage layer comprises a data preprocessing module, a big data storage system and a data mining and analysis module, the data preprocessing module is used for cleaning, denoising and normalization processing of the collected data, a storage layer of the big data storage system is constructed based on Hadoop HDFS and Apache Ozone, Hadoop HDFS is used for storing high-frequency acquisition time series monitoring data, Apache Ozone is used for managing video monitoring and three-dimensional model data, a computing acceleration layer adopts a combination of Spark+Alluxio, a hierarchical storage strategy is implemented by the big data storage system according to data characteristics of different types: hot data resides in memory and NVMeSSD, warm data is stored by using HDD RAID6, and cold data is migrated to a Blu-ray library, and the data mining and analysis module is used for registration of data in space and time scales, and direct or transformed processing of data signals into sound, light and electrical signals; The large model decision and control layer includes an intelligent diagnosis system, an intelligent control module and an optimized scheduling system, the intelligent diagnosis system adopts a machine learning algorithm to establish a fault diagnosis model based on the result output by the data mining and analysis module, to perform real-time diagnosis on the hydraulic facilities and mechanical and electrical equipment, and to propose a fault warning threshold and a warning mode, the intelligent control module constructs a rapid response decision knowledge base under the warning mode based on the diagnosis result, and the optimized scheduling system constructs a hydropower optimized scheduling model by comprehensively considering a power generation plan, equipment state, power price and environmental constraint factors under new energy grid connection, to provide a power generation scheduling scheme. The man-machine interaction and display layer includes a visual display platform and a mobile terminal application, the visual display platform displays the hydropower station in-station scene, equipment operation state, fault diagnosis result and scheduling scheme on the screen through three-dimensional modeling and virtual reality.

[0036] The present application has the following advantages: The present application can effectively reduce equipment downtime, improve water energy utilization rate and overall operation efficiency of the power station, and improve power generation efficiency by formulating a power generation plan through a multi-target constraint hydropower scheduling model under the conditions of ensuring unit safety and ecological base flow, and by reducing water abandonment rate. The present application can reduce the risk of exposure of workers through digital means, and can provide strong support for scientific decision-making through big data analysis and visual display function, so as to improve the fine management level of the hydropower station. The present application can reduce the workload of manual inspection and equipment maintenance, reduce labor cost, improve management efficiency and quality, realize real-time monitoring, diagnosis and optimized scheduling of equipment and facilities of the hydropower station through integration of Internet of Things, big data, artificial intelligence, cloud computing, BIM three-dimensional modeling and digital twin technology, ensure safe operation of key equipment of the hydropower station, reduce the safety risk exposure of the operation and maintenance personnel, improve power generation efficiency, and improve the overall digital operation and management level of the hydropower station. BRIEF DESCRIPTION OF DRAWINGS

[0037] Figure 1 is a system architecture diagram of the present application; Figure 2 is an AES encryption diagram in the present application; Figure 3 is a cross-modal fusion and feature signal extraction diagram in the present application; Figure 4 is a scheduling control diagram in the present application. DETAILED DESCRIPTION

[0038] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments.

[0039] The data operation and maintenance method for hydropower stations includes the following steps: Step 1: Deploy a sensor network and collect data on hydraulic facilities and electromechanical equipment. Hydraulic facilities include dams, water diversion tunnels, and flood discharge stilling basins. Electromechanical equipment includes turbines, generators, transformers, high-voltage switches, relay protection devices, excitation systems, pressure steel pipes, and gates. Step 2: The data collected by the sensor network is transmitted to the data acquisition terminal through AES encryption. The data is cleaned, denoised, and normalized to remove invalid or abnormal data. The data is registered on spatial and temporal scales. The data signals are directly or after conversion and processing into sound, light, and electrical signals. The basic performance characteristics of hydraulic facilities are obtained from spatial, temporal, sound, and light signals, and the basic performance characteristics of electromechanical and metal structure equipment are obtained from spatial, temporal, light, and electrical signals. Step 3: Create a sonar dataset, which is represented as a point cloud. , where each point p i ∈R 4 It contains three-dimensional coordinates (x, y, z) and intensity values; the image data is represented as I∈R （H×W×3） , where H and W represent the height and width of the image, respectively, both in mm; A hierarchical point cloud feature extraction network based on PointNet++ is used to select key points by sampling the farthest point, and group points into sets within a spherical neighborhood around each key point. For each set of points, a multilayer perceptron is used to extract local features, and the key point features are obtained by aggregation through max pooling. Max pooling is repeated twice to obtain features at different scales. Then, 1D convolution is used to unify the feature dimension to D. s The output is the feature matrix F. s ∈R M ×D s Where M is the number of key points; Image features are extracted using a variant of MobileNetV2, with spatial features extracted through multiple convolutional layers (Conv2D) and pooling layers, resulting in feature map F. i ∈R Hf×Wf×Di H f ×W f Let Di be the feature map size, and Di be the feature dimension. Flattening the spatial dimensions gives L=H. f ×W f At each position, we obtain F. i ∈R L×Di ; The coordinates of the sonar point cloud are p j =(xj , y j , z j ) are projected to image plane and spatially aligned:

[0040] where (u j , v j ) are projected image coordinates, is intrinsic matrix, is extrinsic matrix (rotation matrix R and translation vector t ); The sonar features F s are assigned to corresponding positions of image feature map according to projection coordinates, and sonar feature map F s '∈R Hf×Wf×D is obtained with the same resolution as the image feature map, and zero is filled for positions without projection; The aligned sonar features and image features are unified in dimension to D through linear transformation: , where Q s is the query vector of sonar information, is the query vector of image information, is the sonar feature projection matrix, is the image signal feature projection matrix, , ; The attention from sonar to image is:

[0041] where K s is the index vector of sonar information, V s is the feature representation of the queried sonar information, and T is the time or space sequence length; With image features as queries, sonar features as keys and values, the output is , The attention from image to sonar is:

[0042] where K i is the index vector of image information, V i is the feature representation of the queried image information; With sonar features as queries, image features as keys and values, the output is ; The two attention outputs are gated fused:

[0043]

[0044]

[0045]

[0046] where, is a sigmoid function, and Wg, bg are learnable parameters, is the gating vector from sonar to image, is the gating vector from image to sonar, and is the fused feature; The fused feature is concatenated with and to obtain , and then multi-scale features are extracted through three parallel convolution paths:

[0047]

[0048]

[0049] The three features , , are concatenated in the channel dimension and fused through 1x1 convolution:

[0050] where, is the multi-scale aggregated feature; for real-time, dynamic, and periodic mass data collection, different facilities and equipment need to establish their basic feature information library, that is, through statistical analysis and deep learning means to extract basic features under different operating conditions, and the fusion of multi-source signals can complement each other to more comprehensively and truly obtain the current information of the hydraulic facility, effectively solving the problems of chaotic monitoring data, difficulty in utilization, difficulty in data format fusion, and difficulty in effective information extraction; Step 4: Deeply analyze the data through data mining algorithms (cluster analysis, association rule mining, time series analysis) to extract sensor data patterns of hydraulic facilities and mechanical and electrical equipment under normal operating conditions; Step 5: Establish a fault diagnosis model for water facilities and mechanical and electrical equipment through machine learning algorithms (support vector machines, neural networks), conduct real-time diagnosis of water facilities and mechanical and electrical equipment, and divide the faults of water facilities into cracks, abrasion, and defects. The faults of mechanical and electrical equipment are divided into defects, loosening, degradation, tip discharge, corona discharge, suspension discharge, air gap discharge, and surface discharge. Feature labeling and learning are performed on the collected acoustic-optical data or photoelectric data to identify faults and propose fault warning thresholds and warning modes. Step 6: Build a response decision knowledge base under the warning mode according to the fault diagnosis type, which includes facility equipment abnormal feature library, historical disposal case library, and dynamic strategy generation engine. When a warning signal is detected, first match similar historical cases through graph neural network, and then generate decision suggestions including disposal priority, operation steps, and expected impact by combining real-time equipment state data. Step 7: Establish a water and electricity optimization scheduling model, considering the generation plan, equipment state, electricity price, and environmental constraints under new energy grid connection, and set the objective function: Objective 1: Maximize power generation revenue For:

[0051] Where, P(t) is the processing power at time t, unit: MW, price t P(t) is the electricity price at time t, unit: yuan / MWh, Objective 2: Minimize ecological discharge For:

[0052]

[0053] Where, S t E is the additional ecological discharge, unit: m 3 / s, Q t is the actual flow for power generation, unit: m 3 / s, E t is the ecological flow requirement, unit: m 3 / s; Set the constraint condition, New energy grid connection constraint:

[0054] Equipment and facility state constraint: Unit output limit:

[0055] is the approved output; Flow limit:

[0056] Ecological flow guarantee constraint:

[0057] Wherein, R t is the total discharge flow, is the ecological flow; Reservoir water balance:

[0058] Wherein, I t is the inflow, is, is; Reservoir capacity constraint:

[0059] Output calculation:

[0060] Wherein, η is the unit efficiency, H t is the net head, is the acceleration of gravity; A visual display platform is constructed to display information of the hydropower station.

[0061] The hydropower station data operation and maintenance system refers to Figure 1 , comprising an Internet of Things sensing layer, a data processing and sensing layer, a large model decision and control layer, and a man-machine interaction and display layer connected in sequence. The Internet of Things sensing layer collects the operation state data of the hydraulic facilities and mechanical and electrical equipment through a sensor network and transmits the data to a data acquisition terminal. The data processing and storage layer pre-processes, stores, and analyzes the collected data. The large model decision and control layer diagnoses and optimizes the scheduling of the operation state according to the data mining results and remotely controls the equipment through an intelligent control module. The man-machine interaction and display layer provides real-time monitoring and operation interfaces for the operating personnel through a visual display platform and a mobile terminal application.

[0062] The Internet of Things perception layer includes a data acquisition terminal and a sensor network. The sensor network includes a mobile laser radar acquisition dog, a mobile point cloud acquisition dog, a mobile sonar scanning dog, an infrared sensor, a temperature sensor, a vibration sensor, a current sensor, a voltage sensor, a pressure sensor, and a microwave detector. The mobile laser radar acquisition dog, the mobile point cloud acquisition dog, and the mobile sonar scanning dog periodically perform data acquisition and detection at a diversion tunnel, a barrage, an intake, and a tailrace. The infrared sensor and the temperature sensor are used for temperature monitoring of a generator rotor, a main transformer bushing, and a high-voltage switch contact. The vibration sensor is used for fault monitoring of bearing wear, rotor imbalance, and structural looseness of a hydraulic turbine and an excitation system. The current sensor and the voltage sensor are used for excitation current and excitation voltage waveform monitoring of an excitation system and secondary side current and voltage signal monitoring of a relay protection device. The pressure sensor is used for monitoring of water flow excitation and pipe wall stress vibration of a penstock. The microwave detector is used for structural deformation and displacement monitoring of a gate and a valve. The data collected by the sensors is transmitted to the data acquisition terminal through wireless communication. The data transmission mode is AES encryption transmission. Referring to Figure 2 , AES data encryption is implemented through byte substitution, row displacement, column confusion, and multiple round operations. The data processing and storage layer includes a data preprocessing module, a big data storage system, and a data mining and analysis module. The data preprocessing module is used for cleaning, denoising, and normalizing the collected data, removing invalid or abnormal data, and improving data quality. The storage layer of the big data storage system is based on Hadoop HDFS and Apache Ozone. Hadoop HDFS is used to store high-frequency acquisition time series monitoring data, and Apache Ozone is used to manage video monitoring and three-dimensional model data. The computing acceleration layer uses a combination of Spark and Alluxio to achieve millisecond-level response of real-time data through memory caching, supporting million-level data point stream processing capability per second. According to the characteristics of different types of data, the big data storage system implements a hierarchical storage strategy: setting three types of hot data, warm data, and cold data, hot data residing in memory and NVMe SSD, warm data using HDD RAID6 storage, and cold data migrating to a Blu-ray library. It can support 10 TB of data per day, providing high-concurrency and low-latency data services for power generation dispatching, fully meeting the data management needs of hydropower station production operation, ecological dispatching, and decision analysis, and supporting fast reading, writing, and querying of massive data. The data mining and analysis module is used for spatial and temporal scale registration of data and processing of data signals into sound, light, and electrical signals directly or after conversion. Through deep analysis of stored data, equipment operation rules, fault modes, and energy consumption characteristic information are mined, and the information is classified, and different problems and faults are sent to different operation and maintenance personnel. The large model decision and control layer includes an intelligent diagnosis system, an intelligent control module, and an optimization scheduling system. The intelligent diagnosis system is based on the results output by the data mining and analysis module, adopts a machine learning algorithm to establish a fault diagnosis model, performs real-time diagnosis on the hydraulic facilities and mechanical and electrical equipment, and proposes a fault warning threshold and a warning mode to provide early warning of potential faults, formulate optimal decision schemes and risk assessments for operation and maintenance personnel, and form a decision database. The intelligent control module is based on the diagnosis results and constructs a rapid response decision knowledge base under the warning mode to support user work under the warning mode. The knowledge base adopts a modular architecture design, includes a facility and equipment abnormal feature library, a historical disposal case library, and a dynamic strategy generation engine. When the intelligent diagnosis system detects a warning signal, it first matches similar historical cases through a graph neural network, and then combines real-time equipment state data to quickly generate a decision suggestion including disposal priority, operation steps, and expected impact. The intelligent diagnosis system supports multi-modal interaction, and in addition to traditional interface display, it also provides AR glasses visual guidance to overlay equipment parameters and operation guidance to the real scene through spatial positioning technology, enabling field personnel to quickly locate problem points. The optimization scheduling system considers the power generation plan, equipment state, power price, and environmental constraint factors under new energy grid connection, constructs a hydroelectric optimization scheduling model, optimizes the comprehensive operation efficiency of the hydropower station, and considers the requirements of the power grid on hydropower after new energy grid connection. After the new constraints of equipment state after frequent start and stop of hydropower facilities and equipment after new energy grid connection, the constraints faced by new developments can be effectively solved. By setting the environmental constraint as the highest priority, and arranging the power generation plan, equipment state, power price, and environmental constraint factors in turn, an optimal power generation scheduling scheme is formulated by using an optimization algorithm (such as genetic algorithm, particle swarm optimization algorithm), to realize economic operation and energy saving and emission reduction of the hydropower station, and to formulate the best power generation scheduling scheme. The man-machine interaction and display layer includes a visual display platform and a mobile terminal application. The visual display platform displays the hydropower station scene, equipment operation state, fault diagnosis result, and scheduling scheme on the screen through three-dimensional modeling and virtual reality, facilitating real-time monitoring and operation by operation and maintenance personnel. The development of the mobile terminal application is based on mobile applications for mobile phones and tablet computers, enabling operation personnel to view hydropower station operation information, receive fault alarms, perform remote operations, and record operation ports and operation and maintenance personnel at any time and anywhere, improving the flexibility and convenience of operation and management.

[0063] Embodiment 1 The method for data operation and maintenance of a hydropower station includes the following steps: Step 1: Arrange a sensor network on the hydraulic facilities and mechanical and electrical equipment and collect data. The hydraulic facilities include river dams, water diversion tunnels, and flood discharge stilling basins. The mechanical and electrical equipment includes water turbines, generators, transformers, high-voltage switches, relay protection devices, excitation systems, pressure steel pipes, and gates. Step 2: Preprocess the data, align it in space and time scales, and convert the data signals into acoustic, optical, and electrical signals; Step 3: Perform signal fusion of three-dimensional sonar signals and RGB images based on an attention focusing mechanism; Step 4: Perform deep analysis of the data and extract sensor data patterns of the hydraulic facilities and mechanical and electrical equipment under normal operating conditions; Step 5: Establish a fault diagnosis model for real-time diagnosis of the hydraulic facilities and mechanical and electrical equipment; Step 6: Construct a response decision knowledge base; Step 7: Establish a hydropower optimization scheduling model to display information of the hydropower station.

[0064] Embodiment 2: The method for data operation and maintenance of a hydropower station includes the following steps: Step 1: Arrange a sensor network on the hydraulic facilities and mechanical and electrical equipment and collect data; Step 2: Transmit the data collected by the sensor network to a data collection terminal through AES encryption transmission, clean, denoise, and normalize the data, remove invalid or abnormal data, align the data in space and time scales, and directly or after conversion, process the data signals into acoustic, optical, and electrical signals. The basic behavior characteristics of the hydraulic facilities are obtained from the space, time, acoustic, and optical signals, and the basic behavior characteristics of the mechanical and electrical equipment are obtained from the space, time, optical, and electrical signals; Step 3: Perform signal fusion of three-dimensional sonar signals and RGB images based on an attention focusing mechanism; Step 4: Perform deep analysis of the data and extract sensor data patterns of the hydraulic facilities and mechanical and electrical equipment under normal operating conditions; Step 5: Establish a fault diagnosis model for real-time diagnosis of the hydraulic facilities and mechanical and electrical equipment; Step 6: Construct a response decision knowledge base; Step 7: Establish a hydropower optimization scheduling model to display information of the hydropower station.

[0065] Embodiment 3: The method for data operation and maintenance of a hydropower station includes the following steps: Step 1: Arrange a sensor network on the hydraulic facilities and mechanical and electrical equipment and collect data; Step 2: Preprocess the data, align it in space and time scales, and convert the data signals into acoustic, optical, and electrical signals; Step 3: Establish a sonar data set, which is represented as a point cloud where each point p i ∈R 4, containing three-dimensional coordinates (x, y, z) and intensity values; the image data is represented as I ∈ R （H×W×3） , where H and W represent the height and width of the image, respectively, both in mm; Using a hierarchical point cloud feature extraction network based on PointNet++, key points are selected by farthest point sampling, and point sets are grouped in a spherical neighborhood around each key point. Local features are extracted for each point set using a multi-layer perceptron, and key point features are obtained by maximum pooling aggregation. The maximum pooling aggregation is repeated twice to obtain features of different scales, and the feature dimensions are unified to D s , the output is a feature matrix F s ∈R M ×D s , where M is the number of key points; Using a variant of MobileNetV2 to extract image features, spatial features are extracted through multiple convolutional layers Conv2D and pooling layers to obtain a feature map F i ∈R Hf×Wf×Di , where H f ×W f is the feature map size, and Di is the feature dimension. The spatial dimension is flattened into L = H f ×W f positions to obtain F i ∈R L×Di ; Project the sonar point cloud coordinates p j = (x j , y j , z j ) to the image plane for spatial alignment:

[0066] where (u j , v j ) are the projected image coordinates, is the intrinsic matrix, is the extrinsic matrix; Assign the sonar features F s to the corresponding positions of the image feature map according to the projection coordinates to obtain a sonar feature map F s ’ ∈ R Hf×Wf×D with the same resolution as the image feature map. For positions that are not projected, use zero padding; Align the sonar features and image features by linear transformation to unify the dimensions to D: , , where Q s is the query vector of sonar information, The query vector is for image information. This is the sonar feature projection matrix. The image signal feature projection matrix, , ; Sonar-to-image attention for:

[0067] Among them, K s V is the index vector of sonar information. s The feature representation of the retrieved sonar information, where T is the length of the time or spatial sequence; Using image features as the query and sonar features as the key and value, the output is: , Image-to-sonar attention for:

[0068] Among them, K i V is the index vector of image information. i The feature representation of the retrieved image information; Using sonar features as the query and image features as the key and value, the output is: ; Gated fusion of two attention outputs:

[0069]

[0070]

[0071]

[0072] in, For the sigmoid function, Wg and bg are both learnable parameters. This is the gate vector from the sonar to the image. This is the gating vector from the image to the sonar. and Features after fusion; The fused features and spliced ​​together Then, multi-scale features are extracted through three parallel convolutional paths:

[0073]

[0074]

[0075] Then three features 、 、 In the channel dimension splicing, and through 1x1 convolution fusion:

[0076] Among them, Multi-scale aggregation features; Step 4: Deep analysis of data, extraction of sensor data mode of hydraulic facilities and mechanical and electrical equipment under normal operating conditions; Step 5: Establish a fault diagnosis model to diagnose the hydraulic facilities and mechanical and electrical equipment in real time; Step 6: Build a response decision knowledge base; Step 7: Establish a hydropower optimization scheduling model to display the information of the hydropower station.

[0077] Example 4: The hydropower station data operation and maintenance method comprises the following steps: Step 1: Arrange a sensor network on the hydraulic facilities and mechanical and electrical equipment and collect data; Step 2: Preprocess, space and time scale registration of data, and convert data signals into sound, light and electrical signals; Step 3: Signal fusion of three-dimensional sonar signals and RGB images based on attention focusing mechanism; Step 4: Deep analysis of data through data mining algorithm, extraction of sensor data mode of hydraulic facilities and mechanical and electrical equipment under normal operating conditions; Step 5: Establish a fault diagnosis model of hydraulic facilities and mechanical and electrical equipment through machine learning algorithm, diagnose the hydraulic facilities and mechanical and electrical equipment in real time, the faults of the hydraulic facilities are classified as cracks, abrasion and defects, the faults of the mechanical and electrical equipment are classified as defects, loosening, deterioration, tip discharge, corona discharge, suspension discharge, air gap discharge and surface discharge, feature labeling and learning are performed on the collected sound and light data or photoelectric data, the faults are recognized, and a fault warning threshold and a warning mode are proposed; Step 6: According to the fault diagnosis type, build a response decision knowledge base under the warning mode, the response decision knowledge base includes an abnormal feature library of facilities and equipment, a historical disposal case library and a dynamic strategy generation engine, when a warning signal is detected, first match similar historical cases through a graph neural network, then combine real-time equipment state data to generate a decision suggestion including disposal priority, operation steps and expected impact; Step 7: Establish a hydropower optimization scheduling model, comprehensively consider the power generation plan, equipment state, power price and environmental constraints under new energy grid connection, set the objective function: Objective 1: Maximize power generation revenue For:

[0078] Where, P is the processing power in MW, price t Q is the electricity price in Yuan / MWh, t is the time period, Objective 2: Minimize ecological flow For:

[0079]

[0080] Where, S t Q is the additional ecological flow in m 3 / s, Q t is the flow actually used for power generation in m 3 / s, E t is the ecological flow requirement in m 3 / s; Set constraints, New energy grid connection constraints:

[0081] Device facility state constraints: Unit output limit:

[0082] is the approved output; Flow limit:

[0083] Ecological flow guarantee constraints:

[0084] Where, R t is the total outflow, is the ecological flow; Reservoir water balance:

[0085] Where, I t is the inflow, is, is; Reservoir capacity constraints:

[0086] Output calculation:

[0087] wherein η is the unit efficiency, H t is the net water head, is the acceleration due to gravity; A visual display platform is constructed to display information of the hydropower station.

[0088] Example 5: The operation and maintenance method of the high-voltage switch cabinet is expanded. The breaker moving contact area, bus connection, and cable connection point inside the high-voltage switch cabinet are the layout positions of the temperature sensor. These positions are prone to temperature abnormalities due to current concentration heating. By installing temperature sensors in these areas, the temperature changes of the key parts can be monitored in real time. Platinum resistance temperature sensors are used, which have high precision and good stability, and are suitable for temperature monitoring inside the high-voltage switch cabinet.

[0089] The data acquisition terminal is responsible for converting the analog signals collected by the temperature sensor into digital signals. It is equipped with high-precision ADC converters and temperature compensation circuits to improve data accuracy. Data is transmitted to the data processing and storage layer through an encrypted communication interface (TLS / SSL encrypted Ethernet). In abnormal conditions, the data acquisition terminal can automatically switch to a high-frequency sampling mode to obtain more detailed temperature change data. After receiving the data, the data processing and storage layer will immediately encrypt the data. Advanced encryption algorithms (AES-256) are used to encrypt temperature data to ensure data security during transmission and storage.

[0090] The data processing and storage layer integrates high-performance microprocessors, digital filtering algorithms, and temperature trend prediction models to analyze temperature data in real time, determine whether the temperature exceeds the preset safety threshold, and trigger an alarm mechanism in abnormal conditions. The real-time data pool uses ring buffer technology to support high-frequency data throughput, ensuring real-time and consistency of temperature information.

[0091] Through analysis of historical data, a fault prediction model can be established to provide early warning of potential fault risks. When the temperature continues to rise or the current fluctuates frequently, the system can issue a warning to prompt maintenance personnel to check and maintain.

[0092] The intelligent diagnosis system includes an intelligent diagnosis engine based on machine learning algorithms that can dynamically optimize temperature warning thresholds, enabling an upgrade from passive monitoring to active warning. The storage and analysis of historical data also use encryption protocols to ensure data integrity and security.

[0093] A three-dimensional dynamic digital twin of the high-voltage switch cabinet is built based on the Unity3D engine for subsequent visualization and analysis. The temperature data collected by the sensors is mapped in real time to the three-dimensional model, and the running state of the switch cabinet is displayed through the visualization interface. Encryption protocols are used in data transmission and storage to ensure the security of data transmission to the three-dimensional model display platform. The three-dimensional dynamic digital twin not only displays the current temperature state, but also performs trend analysis combined with historical data to help operation and maintenance personnel better understand the running status of the equipment.

[0094] The visualization platform supports remote monitoring function, operation and maintenance personnel can access the platform through the network, real-time view the temperature state of the switch cabinet, when the temperature exceeds the preset threshold, the system will automatically trigger the alarm mechanism, through the sound and light alarm, SMS notification and other ways to remind the operation and maintenance personnel.

[0095] Embodiment 6 The hydropower station data operation and maintenance system comprises an Internet of Things sensing layer, a data processing and sensing layer, a large model decision and control layer, and a human-computer interaction and display layer connected in sequence. The Internet of Things sensing layer comprises a data acquisition terminal and a sensor network. The sensor network comprises a mobile laser radar acquisition dog, a mobile point cloud acquisition dog, a mobile sonar scanning dog, an infrared sensor, a temperature sensor, a vibration sensor, a current sensor, a voltage sensor, a pressure sensor, and a microwave detector. The mobile laser radar acquisition dog, the mobile point cloud acquisition dog, and the mobile sonar scanning dog regularly perform data acquisition and detection in the diversion tunnel, the barrage, the water inlet, and the tail water channel. The infrared sensor and the temperature sensor are used for temperature monitoring of the generator rotor, the main transformer bushing, and the high-voltage switch contact. The vibration sensor is used for fault monitoring of bearing wear, rotor imbalance, and structural looseness of the water turbine and the excitation system. The current sensor and the voltage sensor are used for excitation current and excitation voltage waveform monitoring of the excitation system and secondary side current and voltage signal monitoring of the relay protection device. The pressure sensor is used for water flow excitation and pipe wall stress vibration monitoring of the pressure steel pipe. The microwave detector is used for structural deformation and displacement monitoring of the gate and the valve. The data processing and storage layer includes a data preprocessing module, a big data storage system, and a data mining and analysis module. The data preprocessing module is used for cleaning, denoising, and normalizing the collected data. The storage layer of the big data storage system is based on Hadoop HDFS and Apache Ozone. Hadoop HDFS is used to store high-frequency collected time series monitoring data, and Apache Ozone is used to manage video monitoring and three-dimensional model data. The computing acceleration layer uses a combination of Spark+Alluxio. According to the characteristics of different types of data, the big data storage system implements a hierarchical storage strategy: hot data resides in memory and NVMeSSD, warm data is stored using HDD RAID6, and cold data is migrated to a Blu-ray library. The data mining and analysis module is used to register data on spatial and temporal scales, and to process data signals directly or after conversion into sound, light, and electrical signals. The large model decision and control layer includes an intelligent diagnosis system, an intelligent control module, and an optimization scheduling system. The intelligent diagnosis system is based on the results output by the data mining and analysis module, and uses machine learning algorithms to establish a fault diagnosis model for real-time diagnosis of water conservancy facilities and mechanical and electrical equipment, and to propose fault warning thresholds and warning modes. The intelligent control module is based on the results of the diagnosis, and constructs a rapid response decision knowledge base under the warning mode. The optimization scheduling system considers factors such as power generation plans, equipment status, power prices, and environmental constraints under new energy grid connection, constructs a hydroelectric optimization scheduling model, and provides power generation scheduling schemes. The human-computer interaction and display layer includes a visualization display platform and a mobile terminal application. The visualization display platform displays the scene inside the hydropower station, equipment operation status, fault diagnosis results, and scheduling schemes on the screen through three-dimensional modeling and virtual reality.

Claims

1. A hydropower station data operation and maintenance method, characterized in that, The method comprises the following steps: Step 1: arranging a sensor network on the water conservancy facilities and mechanical and electrical equipment and collecting data; Step 2: pre-processing the data, registering the data in space and time scales, and converting the data signals into sound, light and electrical signals; Step 3: signal fusion of three-dimensional sonar signals and RGB images based on an attention focusing mechanism; Step 4: deep analysis of the data to extract the sensor data mode of the water conservancy facilities and mechanical and electrical equipment under normal operating conditions; Step 5: establishing a fault diagnosis model to diagnose the water conservancy facilities and mechanical and electrical equipment in real time; Step 6: constructing a response decision knowledge base; Step 7: establishing a hydropower optimization scheduling model to display information of the hydropower station.

2. The hydropower station data operation and maintenance method of claim 1, wherein, The water conservancy facilities in step 1 include river dams, water diversion tunnels and flood discharge stilling basins; the mechanical and electrical equipment includes water turbines, generators, transformers, high-voltage switches, relay protection devices, excitation systems, pressure steel pipes and gates.

3. The hydropower station data operation and maintenance method of claim 1, wherein, In step 2, the data collected by the sensor network is transmitted to the data collection terminal through AES encryption transmission, and the data is cleaned, denoised and normalized to remove invalid or abnormal data. The data is registered in space and time scales, and the data signals are directly or converted and processed into sound, light and electrical signals. The basic behavior characteristics of the water conservancy facilities are obtained from the space, time, sound and light signals, and the basic behavior characteristics of the mechanical and electrical equipment are obtained from the space, time, light and electrical signals.

4. The hydropower station data operation and maintenance method of claim 1, wherein, Step 3 is specifically performed as follows: Step 3.1, establishing a sonar dataset, the sonar dataset is represented as a point cloud where each point p i ∈R 4 contains a three-dimensional coordinate (x, y, z) and an intensity value; the image data is represented as I∈R （H×W×3） where H and W represent the height and width of the image, respectively, both in mm; Step 3.2, a hierarchical point cloud feature extraction network based on PointNet++ is used to select key points by farthest point sampling, group point sets in a spherical neighborhood around each key point, extract local features for each point set using a multi-layer perceptron, and obtain key point features by maximum pooling aggregation; Step 3.

3. Repeat step 3.2 twice to get different scale-specific features, and unify the feature dimension to D by 1D convolution s , the output is a feature matrix F s ∈R M ×D s , where M is the number of key points; Step 3.

4. Extract image features using a variant of MobileNetV2, extract spatial features through multiple convolutional layers Conv2D and pooling layers to obtain feature map F i ∈R Hf×Wf×Di where H f ×W f is the feature map size, Di is the feature dimension, and the spatial dimension is flattened into L = H f ×W f positions to obtain F i ∈R L×Di ; Step 3.

5. Project the sonar point cloud coordinates p j = (x j ,y j ,z j ) to the image plane for spatial alignment: wherein (u j ,v j ) are the projected image coordinates, is the intrinsic matrix, is the extrinsic matrix; The sonar features F s According to the assignment of the projection coordinates to the corresponding positions of the image feature map, a sonar feature map F of the same resolution as the image feature map is obtained s ’∈R Hf×Wf×D For positions without a projection, zero is filled. Step 3.6, aligning the sonar features and image features by a linear transformation to unify the dimensions to D: , , wherein Q s is an interrogation vector of sonar information, is an interrogation vector of image information, is a sonar feature projection matrix, is an image signal feature projection matrix, , ; Sonar to image attention is: where K s is the index vector of sonar information, V s is the feature representation of the queried sonar information, T is the time or spatial sequence length; With image features as queries, sonar features as keys and values, and output as , Attention from image to sonar is: where K i is an index vector of image information, V i is a feature representation of the queried image information; With sonar features as queries, image features as keys and values, and output as ; Step 3.7, gate fusion is performed on the two attention outputs: wherein, is a sigmoid function, Wg, bg are learnable parameters, is a gating vector from sonar to image, is a gating vector from image to sonar, and is the fused feature; Step 3.8, the fused features and are concatenated to obtain , and then multi-scale features are extracted through three parallel convolution paths: The three features are combined again , , In the channel dimension, and fused by 1x1 convolution: wherein, is a multiscale aggregation feature.

5. The hydropower station data operation and maintenance method of claim 1, wherein, In step 4, the data is deeply analyzed by a data mining algorithm.

6. The hydropower station data operation and maintenance method of claim 1, wherein, In step 5, a fault diagnosis model of the water conservancy facilities and mechanical and electrical equipment is established by a machine learning algorithm. The faults of the water conservancy facilities are classified as cracks, abrasion and defects, and the faults of the mechanical and electrical equipment are classified as defects, loosening, deterioration, tip discharge, corona discharge, suspension discharge, air gap discharge and surface discharge. The collected sound and light data or light and electrical data are feature labeled and learned to identify faults and propose fault warning thresholds and warning modes.

7. The hydropower station data operation and maintenance method of claim 6, wherein, In step 6, a response decision knowledge base under the warning mode is constructed according to the fault diagnosis type. The response decision knowledge base includes an abnormal feature library of facilities and equipment, a historical disposal case library and a dynamic strategy generation engine. When a warning signal is detected, similar historical cases are matched through a graph neural network, and a decision suggestion including disposal priority, operation steps and expected impact is generated by combining real-time equipment state data.

8. The hydropower station data operation and maintenance method of claim 1, wherein, Step 7 is specifically performed as follows: Step 7.1, considering the generation plan, equipment state, power price and environmental constraints under new energy grid connection, a target function is set: Objective 1: Maximize power generation revenue For: wherein, P is the processing power for the time period t, in MW, price t P is the processing power for the time period t, in MW, price Objective 2: Minimize ecological flow release For: where S t is the additional ecological discharge, in m 3 / s, Q t is the flow actually used for power generation, in m 3 / s, E t is the ecological flow requirement, in m 3 / s; Step 7.2, constraint conditions are set, New energy grid connection constraint: Equipment and facility state constraint: Unit power output limit: To approve an output; Flow restriction: Ecological flow guarantee constraint: wherein R t is the total outflow, is the ecological flow; Reservoir water balance: where I t is the in-flow rate, is, is; Storage capacity constraints: Output calculation: wherein η is the unit efficiency, H t is the net water head, is the gravitational acceleration; A visual display platform is constructed to display information of the hydropower station.

9. A hydropower plant data operation and maintenance system, characterized in that, The method for realizing the data operation and maintenance of the hydropower station according to any one of claims 1-8 comprises an Internet of Things sensing layer, a data processing and sensing layer, a large model decision and control layer, and a man-machine interaction and display layer connected in sequence.

10. The hydropower station data operation and maintenance system of claim 9, wherein, The Internet of Things sensing layer comprises a data acquisition terminal and a sensor network, the sensor network comprises a mobile laser radar acquisition dog, a mobile point cloud acquisition dog, a mobile sonar scanning dog, an infrared sensor, a temperature sensor, a vibration sensor, a current sensor, a voltage sensor, a pressure sensor, and a microwave detector, the mobile laser radar acquisition dog, the mobile point cloud acquisition dog, and the mobile sonar scanning dog periodically perform data acquisition and detection in a diversion tunnel, a barrage, a water inlet, and a tailrace, the infrared sensor and the temperature sensor are used for temperature monitoring of a generator rotor, a main transformer bushing, and a high-voltage switch contact, the vibration sensor is used for fault monitoring of bearing wear, rotor imbalance, and structural looseness of a hydraulic turbine and an excitation system, the current sensor and the voltage sensor are used for excitation current and excitation voltage waveform monitoring of an excitation system and secondary side current and voltage signal monitoring of a relay protection device, the pressure sensor is used for monitoring of water flow excitation and pipe wall stress vibration of a penstock, and the microwave detector is used for structural deformation and displacement monitoring of a gate and a valve. The data processing and storage layer comprises a data preprocessing module, a big data storage system, and a data mining and analysis module, the data preprocessing module is used for cleaning, denoising, and normalization processing of the collected data, a storage layer of the big data storage system is constructed based on Hadoop HDFS and Apache Ozone, Hadoop HDFS is used for storing high-frequency acquisition time series monitoring data, Apache Ozone is used for managing video monitoring and three-dimensional model data, a calculation acceleration layer adopts a combination of Spark+Alluxio, a hierarchical storage strategy is implemented by the big data storage system according to data characteristics of different types: hot data resides in memory and NVMe SSD, warm data is stored by using HDD RAID6, and cold data is migrated to a Blu-ray library, and the data mining and analysis module is used for registration of data in space and time scales and direct or converted processing of data signals into sound, light, and electrical signals. The large model decision and control layer comprises an intelligent diagnosis system, an intelligent control module, and an optimization scheduling system, the intelligent diagnosis system establishes a fault diagnosis model by using a machine learning algorithm based on an output result of the data mining and analysis module, performs real-time diagnosis on hydraulic facilities and mechanical and electrical equipment, and proposes a fault warning threshold and a warning mode, the intelligent control module constructs a rapid response decision knowledge base under the warning mode based on a diagnosis result, and the optimization scheduling system constructs a hydropower optimization scheduling model by comprehensively considering a power generation plan, a device state, a power price, and environmental constraint factors under new energy grid connection, and provides a power generation scheduling scheme. The human-computer interaction and display layer comprises a visual display platform and a mobile terminal application, and the visual display platform displays the scene in the hydropower station, the equipment operation state, the fault diagnosis result and the dispatching scheme on the screen through three-dimensional modeling and virtual reality.

Images