Intelligent optimization method and device for water turbine, electronic equipment and storage medium

Abstract

The invention provides an intelligent optimization method and device for a water turbine, electronic equipment and a storage medium, and effectively solves various problems existing in optimization of the water turbine by an existing intelligent optimization technology for the water turbine. The method comprises the following steps: scanning a water turbine in a hydropower station to obtain a point cloud set corresponding to a plurality of parts of the water turbine, and traversing point cloud data in the point cloud set to obtain a virtual model corresponding to the water turbine; collecting and inputting various data to a water turbine quality prediction model, and processing the various data based on a virtual model corresponding to the water turbine to output a quality prediction result; generating an optimization strategy for the water turbine based on the quality prediction result, and controlling the water turbine to execute the optimization strategy to obtain an execution result; and evaluating the execution result to obtain an evaluation result, and controlling the water turbine to execute the optimization strategy based on the evaluation result so as to complete intelligent optimization of the water turbine.

Description

Technical Field

[0001] This application relates to the technical field of intelligentization of hydroelectric power generation equipment. Specifically, it relates to a method and device for intelligent optimization of a water turbine, an electronic device, and a storage medium. Background Art

[0002] The existing intelligent optimization technologies for water turbines have the following technical bottlenecks:

[0003] 1. There are limitations in the mechanical adjustment of control strategies

[0004] Patent documents CN206770098U (adding a magnet structure between the built-in water turbine and the fan blade) and CN109931203A (using a variable-frequency motor to compensate for the output of the water turbine) show that the existing intelligent optimization technologies for water turbines rely on fixed compensation strategies (such as industrial-frequency / variable-frequency motors) and cannot dynamically match the changes in the operating conditions of the water turbine. For example, the three-phase asynchronous motor adopted in CN109931203A has an efficiency of less than 30% under low load, and the control logic is only based on preset parameters, unable to optimize the matching of the guide vane opening and the speed in real time, resulting in an efficiency loss of up to 12%.

[0005] 2. Data islands and insufficient model generalization ability

[0006] The patent "Federated Learning Based on Pipeline Mode" (November 28, 2024) applied by a certain power investment group and the aerospace science and industry federated learning framework (October 24, 2024) point out that currently, cross-station data cannot be shared due to privacy barriers, and the accuracy of the centralized training model decreases by 18% - 25% in heterogeneous data scenarios. For example, the literature CN202310075826A (bearing temperature prediction method) only uses data from a single power station to train the model, and the prediction error increases by 32% when deployed across power stations.

[0007] 3. Defects in model staticization and real-time performance

[0008] The literature CN202411051075A (water turbine temperature data preprocessing method) reveals that the existing preprocessing methods rely on offline statistical rules (such as fixed threshold filtering) and cannot dynamically adapt to the changes in the unit status. For example, its preprocessing process requires manual annotation of the stable state duration threshold, resulting in the need to repeatedly adjust parameters when deploying new models, and the model update is delayed by more than 48 hours. In addition, the BP neural network model adopted in CN202310075826A has an update cycle as long as 7 days and cannot capture sudden vibration anomalies.

[0009] 4. Problem of the breakage of the decision-control chain

[0010] The patents of Haqi robots (January 31, 2025) and CN109931203A9 show that there is a lack of closed-loop control between the existing maintenance decision-making system and the actuators. For example, the variable-frequency compensation strategy of CN109931203A is not linked to the health prediction, resulting in the compensation action lagging behind the actual fault occurrence (with an average delay of 15 minutes), and the compensation amount cannot be dynamically optimized through reinforcement learning.

[0011] 5. Insufficient coupling between digital twin and physical entity

[0012] The patent of Zhitiangong Technology's spiral water turbine (December 17, 2024) points out that the traditional modeling method has insufficient accuracy (error > 1mm) and does not integrate multi-physical field simulation data. For example, the empirical formula method used in the Yajiang water bucket design patent (January 7, 2025) 10 cannot reflect the dynamic stress distribution of the runner, resulting in a deviation of more than 40% between the simulation result and the actual wear. Summary of the Invention

[0013] In view of this, the purpose of this application is to provide a water turbine intelligent optimization method, device, electronic device and storage medium. The water turbine intelligent optimization method, device, electronic device and storage medium effectively solve various problems existing in the existing water turbine intelligent optimization technology when optimizing water turbines.

[0014] The embodiment of this application provides a water turbine intelligent optimization method, and the method includes:

[0015] Scan the water turbine in the hydropower station to obtain a point cloud set corresponding to multiple parts of the water turbine, and traverse the point cloud data in the point cloud set to obtain a virtual model corresponding to the water turbine;

[0016] Collect various data of the water turbine during operation, and input the various data into a pre-trained water turbine quality prediction model, so that the water turbine quality prediction model processes the various data based on the virtual model corresponding to the water turbine and outputs a quality prediction result; the water turbine quality prediction model is established across hydropower stations; the quality prediction result is displayed based on the virtual model;

[0017] Generate an optimization strategy for the water turbine based on the quality prediction result, and control the virtual model corresponding to the water turbine to execute the optimization strategy to obtain an execution result;

[0018] Evaluate the execution result based on multiple evaluation dimensions to obtain an evaluation result, and control the water turbine to execute the optimization strategy based on the evaluation result to complete the intelligent optimization of the water turbine.

[0019] In combination with the first aspect, the embodiments of the present application provide a first possible implementation manner of the first aspect, wherein generating an optimization strategy for the water turbine based on the quality prediction result includes:

[0020] Determine multiple strategy generation dimensions based on the quality prediction result, and generate strategy dimension results corresponding to the strategy generation dimensions based on the multiple strategy generation dimensions and the quality prediction result;

[0021] Fuse the strategy dimension results corresponding to the multiple strategy generation dimensions to obtain the optimization strategy for the water turbine.

[0022] In combination with the first aspect, the embodiments of the present application provide a second possible implementation manner of the first aspect, wherein determining multiple strategy generation dimensions based on the quality prediction result includes:

[0023] Determine the spatial dimension mapped by each sub-result based on the state attribute corresponding to each sub-result in the quality prediction result;

[0024] Obtain the corresponding strategy generation dimension based on the correspondence between the spatial dimension and the strategy generation dimension.

[0025] In combination with the first aspect, the embodiments of the present application provide a third possible implementation manner of the first aspect, wherein the water turbine quality prediction model processes the multiple data based on the virtual model corresponding to the water turbine and outputs a quality prediction result, including:

[0026] Extract the features of the multiple data at multiple scales to obtain multiple data features, and call a preset fusion method based on the multiple data features;

[0027] Process the multiple data features through the preset fusion method and the corresponding weight coefficients to output a quality prediction result.

[0028] In combination with the first aspect, the embodiments of the present application provide a fourth possible implementation manner of the first aspect, wherein extracting the features of the multiple data at multiple scales to obtain multiple data features includes:

[0029] Pre-establish a local feature sub-model and a heterogeneous model, and set different feature extraction methods for the local feature sub-model and the heterogeneous model;

[0030] Input the multiple data into the local feature sub-model and the heterogeneous model, so that the local feature sub-model and the heterogeneous model perform multi-scale feature extraction on the multiple data.

[0031] In combination with the first aspect, the embodiments of the present application provide a fifth possible implementation manner of the first aspect. Specifically, the steps for establishing a turbine quality prediction model across hydropower stations include:

[0032] Based on the parameter combinations of the local quality prediction models in each hydropower station received and aggregated by the central server, a global model is obtained; the global model includes the prediction errors and communication costs of the local quality prediction models; the local quality prediction models are obtained by learning the virtual models of the turbines in each hydropower station;

[0033] Search for the optimal hyperparameter combination in the parameter combinations of each local quality prediction model, and fuse the global model and the optimal hyperparameter combination to obtain the turbine quality prediction model.

[0034] In combination with the first aspect, the embodiments of the present application provide a sixth possible implementation manner of the first aspect. After traversing the point cloud data in the point cloud set to obtain the virtual model corresponding to the turbine, the following steps are included:

[0035] Calculate the stress of the part corresponding to the virtual model and determine whether it meets the preset high-risk conditions;

[0036] If so, determine the position information corresponding to the stress and label the virtual model based on the position information.

[0037] In a second aspect, the embodiments of the present application provide a turbine intelligent optimization device, which includes:

[0038] A scanning module for scanning the turbine in the hydropower station to obtain a point cloud set corresponding to multiple parts of the turbine, and traversing the point cloud data in the point cloud set to obtain the virtual model corresponding to the turbine;

[0039] An acquisition module for acquiring various data during the operation of the turbine and inputting the various data into a pre-trained turbine quality prediction model, so that the turbine quality prediction model processes the various data based on the virtual model corresponding to the turbine and outputs a quality prediction result; the turbine quality prediction model is established across hydropower stations; the quality prediction result is displayed based on the virtual model;

[0040] An execution module for generating an optimization strategy for the turbine based on the quality prediction result, and controlling the virtual model corresponding to the turbine to execute the optimization strategy to obtain an execution result;

[0041] An evaluation module for evaluating the execution result based on multiple evaluation dimensions to obtain an evaluation result, and controlling the turbine to execute the optimization strategy based on the evaluation result to complete the intelligent optimization of the turbine.

[0042] In a third aspect, an embodiment of the present application provides an electronic device, comprising: a processor, a memory and a bus, wherein the memory stores machine-readable instructions executable by the processor, and when the electronic device is running, the processor and the memory communicate via the bus, and when the machine-readable instructions are executed by the processor, the steps of any one of the methods for intelligent optimization of a hydro-turbine are performed.

[0043] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the computer program executes any one of the steps of a method for intelligent optimization of a water turbine.

[0044] An embodiment of the present application provides a method for intelligent optimization of a turbine. The method first scans a turbine in a hydropower station to obtain a point cloud set corresponding to multiple parts of the turbine, and traverses the point cloud data in the point cloud set to obtain a virtual model corresponding to the turbine; secondly, a variety of data of the turbine during operation are collected, and the various data are input into a pre-trained turbine quality prediction model, so that the turbine quality prediction model processes the various data based on the virtual model corresponding to the turbine to output a quality prediction result; the turbine quality prediction model is established across hydropower stations; the quality prediction result is displayed based on the virtual model; then, an optimization strategy for the turbine is generated based on the quality prediction result, and the virtual model corresponding to the turbine is controlled to execute the optimization strategy to obtain an execution result; finally, the execution result is evaluated based on multiple evaluation dimensions to obtain an evaluation result. Based on the evaluation result, the turbine is controlled to execute the optimization strategy to complete the intelligent optimization of the turbine, thereby solving the problems of the existing intelligent optimization technology of turbines when optimizing the turbine, such as the limitations of mechanical adjustment of the control strategy, insufficient data islands and model generalization capabilities, model staticization and real-time defects, decision-control chain breaks, and insufficient coupling between digital twins and physical entities. The optimization effect when optimizing the turbine is ensured, the data barriers and model generalization bottlenecks are broken, high-precision dynamic mapping and simulation are achieved, real-time response and decision-making capabilities are enhanced, operating efficiency and equipment life cycle management are optimized, and innovative applications of technology integration are promoted. The "digital twin + federated learning + reinforcement learning" collaborative architecture is pioneered to open up the full-link closed loop of data perception, intelligent decision-making and execution control, and provide a systematic solution for the intelligent upgrade of the hydropower industry. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] To more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the accompanying drawings required for the embodiments. It should be understood that the following drawings only show some embodiments of the present application, and thus should not be regarded as limiting the scope. For those of ordinary skill in the art, without creative efforts, other related drawings can also be obtained based on these drawings.

[0046] Figure 1 It shows a schematic flow chart of the first hydroturbine intelligent optimization method provided by the embodiments of the present application;

[0047] Figure 2 It shows a schematic flow chart of obtaining the quality prediction result provided by the embodiments of the present application;

[0048] Figure 3 It shows a schematic flow chart of obtaining the optimization strategy provided by the embodiments of the present application;

[0049] Figure 4 It shows a structural block diagram of the first hydroturbine intelligent optimization device provided by the embodiments of the present application;

[0050] Figure 5 It shows a schematic structural diagram of the first electronic device provided by the embodiments of the present application. Detailed implementation manners

[0051] To make the objectives, technical solutions, and advantages of the embodiments of the present application clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present application in conjunction with the accompanying drawings in the embodiments of the present application. It should be understood that the accompanying drawings in the present application only serve the purposes of illustration and description, and are not used to limit the protection scope of the present application. Additionally, it should be understood that the schematic drawings are not drawn to actual scale. The flowcharts used in the present application show the operations implemented according to some embodiments of the present application. It should be understood that the operations in the flowchart may not be implemented in sequence, and steps without logical context relationships may be reversed or implemented simultaneously. Moreover, those skilled in the art can add one or more other operations to the flowchart or remove one or more operations from the flowchart under the guidance of the content of the present application.

[0052] In addition, the described embodiments are only some embodiments of the present application, rather than all of the embodiments. The components of the embodiments of the present application usually described and illustrated in the accompanying drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present application provided in the accompanying drawings is not intended to limit the scope of the present application to be protected, but only represents the selected embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without creative efforts belong to the scope of protection of the present application.

[0053] It should be noted that the term "including" will be used in the embodiments of the present application to indicate the existence of the features stated thereafter, but does not exclude the addition of other features.

[0054] Currently, there are various problems in the existing hydroturbine intelligent optimization technology when optimizing hydroturbines, such as limitations in the mechanical adjustment of control strategies, data islands and insufficient model generalization ability, defects in model staticization and real-time performance, breakage of the decision-control chain, and insufficient coupling between digital twins and physical entities.

[0055] Based on this, the embodiments of the present application provide a hydroturbine intelligent optimization method, device, electronic device, and storage medium, which will be described below through embodiments.

[0056] Embodiment 1

[0057] To facilitate the understanding of this embodiment, first, a hydroturbine intelligent optimization method disclosed in the embodiments of the present application will be introduced in detail. As Figure 1 shown in the flowchart of a hydroturbine intelligent optimization method, a hydroturbine intelligent optimization method provided by the present application includes:

[0058] S101. Scan the hydroturbine in the hydropower station to obtain a point cloud set corresponding to multiple parts of the hydroturbine, and traverse the point cloud data in the point cloud set to obtain a virtual model corresponding to the hydroturbine;

[0059] S102. Collect various data of the hydroturbine during operation and input the various data into a pre-trained hydroturbine quality prediction model, so that the hydroturbine quality prediction model processes the various data based on the virtual model corresponding to the hydroturbine and outputs a quality prediction result; the hydroturbine quality prediction model is established across hydropower stations; the quality prediction result is displayed based on the virtual model;

[0060] S103. Generate an optimization strategy for the hydroturbine based on the quality prediction result, and control the virtual model corresponding to the hydroturbine to execute the optimization strategy to obtain an execution result;

[0061] S104. Evaluate the execution result based on multiple evaluation dimensions to obtain an evaluation result, and control the hydroturbine to execute the optimization strategy based on the evaluation result to complete the intelligent optimization of the hydroturbine.

[0062] In step S101, if there are multiple turbines in the hydropower station, then multiple parts of each turbine are scanned using a depth camera to obtain the corresponding point cloud sets of the multiple parts. Since in practice, faults are likely to occur in the turbine, and the three parts that need to be optimized are the spiral case, guide vane, and runner, while other parts are not likely to have faults or damages, the obtained point cloud sets also correspond to these three parts, namely the spiral case, guide vane, and runner. There are multiple point clouds in the point cloud set, and each point cloud is an array with three columns in one row. The NURBS surface reconstruction technology is used to process the multiple point clouds in the point cloud set, that is, the point cloud data in the point cloud set is traversed to obtain the virtual models corresponding to the three parts, which is also to obtain the virtual model corresponding to the turbine, that is, the multiple point clouds in the point cloud set corresponding to the part are connected according to the actual structure to form a surface corresponding to the actual structure. The surface is a non-uniform rational β-spline curve, that is, a virtual model corresponding to the actual part of the turbine is constructed. The connection method based on the point clouds in the point cloud set ensures that the accuracy error between the virtual model and the actual corresponding part is less than 0.1 mm, that is, the accuracy of the obtained virtual model of the turbine is guaranteed.

[0063] In a specific implementation process of step S101, there is an embodiment: after traversing the point cloud data in the point cloud set to obtain the virtual model corresponding to the turbine, it includes:

[0064] S1011. Calculate the stress of the part corresponding to the virtual model and judge whether it meets the preset high-risk condition;

[0065] S1012. If so, determine the position information corresponding to the stress and label the virtual model based on the position information.

[0066] In steps S1011 - S1012, after obtaining the virtual models corresponding to the spiral case, guide vane, and runner of the water turbine respectively, the virtual model of this water turbine is established. For the virtual model of the water turbine, based on the material mechanics simulation unit, calculate the stress distribution of the virtual model of the water turbine. First, specify the correct material properties for each component of the water turbine, such as density, elastic modulus, Poisson's ratio, etc. These properties will directly affect the calculation results of the stress distribution. Then, according to the actual working conditions of the water turbine, set reasonable boundary conditions. For example, fix certain surfaces of certain components to simulate their restraint conditions in the real environment. Finally, apply loads that conform to the actual situation, including the pressure, gravity, centrifugal force, etc. brought by the water flow. These loads should be reasonably estimated according to the operating conditions of the water turbine and the results of fluid dynamics analysis to obtain the corresponding stress, and judge whether it meets the preset high - risk conditions according to the specific stress value. The preset high - risk condition is whether the stress value is higher than the preset high - risk threshold. If it meets the preset high - risk condition, determine the position information of the high - risk area according to the corresponding point cloud, and perform color marking on the virtual model of the water turbine according to this position information. Those that do not meet this high - risk condition are marked with other colors.

[0067] In step S102, after establishing the virtual model of the water turbine in the hydropower station, collect various data of the water turbine during operation. The various data include water turbine operation state data, environmental data, and image data. The water turbine state data includes vibration data detected by vibration sensors and pressure data detected by pressure sensors. The environmental data refers to temperature data detected by document sensors. The image data refers to data such as the flow velocity and flow rate of the water inlet obtained by the depth camera shooting the water inlet of the water turbine from above based on the reflection of water. And pre - process the collected various data. The pre - processing specifically includes wavelet denoising and time - frequency domain feature extraction, and output the pre - processed various data and the virtual model of the water turbine to a pre - trained water turbine quality prediction model, so that the water turbine quality prediction model outputs a quality prediction result based on the various data; the quality prediction result is displayed based on the virtual model. For example, if the quality prediction result is that the quality of the guide vane in the water turbine is poor, it will be displayed on the virtual model, and the specific position can be displayed through the marked high - risk area. To avoid the existence of data islands in the future, the water turbine quality prediction model is established across hydropower stations; that is, the local models established by each hydropower station are trained and learned on the central server, and then the learned models are sent back to each hydropower station. When each hydropower station trains the local model that has been learned on the central server, optimize the neural network structure based on the adaptive differential evolution algorithm to ensure the accuracy of the quality prediction result even if there are data islands in the local models of each hydropower station.

[0068] In the specific implementation process of step S102, there is an embodiment as follows: Figure 2 As shown, the turbine quality prediction model processes the multiple data based on the virtual model corresponding to the turbine and outputs a quality prediction result, including:

[0069] S10211. Extract the features of the multiple data at multiple scales to obtain multiple data features, and call a preset fusion method based on the multiple data features;

[0070] S10212. Process the multiple data features through the preset fusion method and the corresponding weight coefficients to output a quality prediction result.

[0071] In steps S10211 - S10212, the turbine quality prediction model sets different data feature extraction methods for the multiple data. The data feature extraction methods are all set in the base model layer. The base model layer includes multiple models for data feature extraction. After the base model layer extracts data features with different extraction dimensions from the multiple data, it calls a preset fusion method to fuse the data features extracted with different extraction dimensions together. The preset fusion method is the Attention mechanism. When determining that the Attention mechanism is the preset fusion method, the weight coefficients of the Attention mechanism are also determined based on actual needs. The specific calculation method of the weight coefficients can be calculated based on statistical methods (such as the entropy weight method), machine learning algorithms (such as the weight adjustment in gradient boosting trees), or expert experience, etc. The weight coefficients can also be adjusted according to the performance of the quality prediction result and real-time working condition data. Methods such as cross-validation and grid search can be used to optimize the weight coefficients. According to the selected preset fusion method and weight coefficients, the data features are fused and processed. The result of the fusion process will be used as the input of the prediction model, so as to perform quality prediction based on the turbine quality prediction model to obtain a quality prediction result. The prediction model can be linear regression, logistic regression, support vector machine, neural network, etc. The quality prediction result also includes the result corresponding to the position information of the standard high risk. This application also constructs a turbine digital twin with millimeter-level accuracy based on parametric modeling and real-time data synchronization technology, deeply fuses multi-physical field simulation data, and provides a dynamic and visual decision support basis for equipment status analysis.

[0072] In the specific implementation process of step S10211, there is an embodiment as follows: The multi-scale extraction of the features of the multiple data to obtain multiple data features includes:

[0073] S102111. Establish a local feature sub-model and a heterogeneous model in advance, and set different feature extraction methods for the local feature sub-model and the heterogeneous model;

[0074] S102112. Input the multiple types of data into the local feature sub-model and the heterogeneous model, so that the local feature sub-model and the heterogeneous model perform multi-scale feature extraction on the multiple types of data.

[0075] In steps S102111 - S102112, the base model layer includes multiple models for data feature extraction, including a pre-established local feature sub-model and a heterogeneous model. Specifically, they can be an LSTM time series prediction model, a CNN image recognition model, and an XGBoost feature analysis model, which process the operating state data, environmental data, and image data of the water turbine respectively. The specific establishment and optimization methods are mature technologies in the prior art and will not be elaborated here. The local feature sub-model and the heterogeneous model have different feature extraction methods. The local feature sub-model is screened by the T-test screening method; the heterogeneous model is constructed by dividing historical data using the moving window technique, and it is ensured that the multiple types of data are input into the local feature sub-model and the heterogeneous model in the correct format and order for multi-scale feature extraction. The local feature sub-model and the heterogeneous model perform feature extraction in different extraction dimensions on the input multiple types of data. For example, if the local feature sub-model extracts features from the vibration dimension of the multiple types of data, the heterogeneous model extracts features from the temperature dimension, ensuring that the features extracted from the local feature sub-model and the heterogeneous model can reflect different aspects and levels of the water turbine represented by the multiple types of data.

[0076] In the specific implementation process of step S102, there is another embodiment: The specific steps for establishing the water turbine quality prediction model across hydropower stations include:

[0077] S10221. Obtain a global model based on the parameter combinations of the local quality prediction models in each hydropower station aggregated by the central server; the global model includes the prediction errors and communication costs of the local quality prediction models; the local quality prediction models are obtained by learning the virtual models of the water turbines in each hydropower station.

[0078] S10222. Search for the optimal hyperparameter combination in the parameter combinations of each local quality prediction model, and fuse the global model and the optimal hyperparameter combination to obtain the water turbine quality prediction model.

[0079] In steps S10221 - S10222, each hydropower station is based on the established local quality prediction model of the local water turbine. The local quality prediction model is established and trained according to the data of the water turbine in the corresponding hydropower station and the virtual model. The specific establishment and training methods are mature technologies in the prior art and will not be elaborated here. After the local quality prediction model is established and trained, the parameter combination of the local quality prediction model is sent to the central server. The central server aggregates and learns the parameter combinations of the local quality prediction models in each received hydropower station to obtain a global model, and the local quality prediction models between each hydropower station also learn from each other. The model prediction errors generated by the local quality prediction models between each hydropower station and the communication cost are calculated by dynamic weighted sum using a fitness function. The central server also searches for the optimal hyperparameter combination in the parameter combinations of each local quality prediction model in the mutation - crossover - selection iteration based on the differential evolution algorithm, and fuses the global model and the optimal hyperparameter combination to obtain a water turbine quality prediction model, and distributes the water turbine quality prediction model to each hydropower station. Then each hydropower station optimizes the parameters (such as learning rate, network depth) of the received water turbine quality prediction model by the differential evolution algorithm. Through the federated transfer learning framework, this application realizes cross - domain knowledge sharing and model collaborative optimization while protecting the data privacy of each hydropower station, significantly improving the adaptability and prediction accuracy of the machine learning model in different hydropower station environments, and effectively solving the long - standing data island problem in the industry.

[0080] Moreover, due to the possible heterogeneity of data between different hydropower stations (such as differences in data format, dimension, distribution, etc.), this will pose challenges to the training and aggregation of the global model. To solve this problem, this application introduces domain adaptation technology. Domain adaptation technology aims to enable the model to adapt to the data distribution of different domains (or data sources). In the hydropower station scenario, this means that the model needs to be able to adapt to the data heterogeneity between different hydropower stations. Through domain adaptation technology, the data of different hydropower stations can be transformed into a common feature space, enabling the model to effectively learn and predict in this space. Specifically, difference - based, adversarial - based, or reconstruction - based domain adaptation methods can be used to achieve this goal. These methods align the feature distributions of the source domain and the target domain through different mechanisms to solve the data heterogeneity problem.

[0081] In step S103, the present application combines an online learning mechanism and uses the Stacking ensemble learning method to fuse the LSTM time series prediction model, the CNN image recognition model, and the XGBoost feature analysis model to obtain a reinforcement learning controller. The double-delayed deep deterministic policy gradient (TD3) algorithm is used to train and control the reinforcement learning controller. A reward function is designed for the reinforcement learning controller to balance power generation efficiency and equipment loss coefficient. The reinforcement learning controller generates an optimization strategy for the water turbine based on the quality prediction result. The optimization strategy can be that if the blade vibration is abnormal, then according to the energy efficiency and vibration data, adjust the angle or shape of the blade to optimize the impact force of the water flow and reduce vibration. The optimization strategy can also be that if the quality prediction result shows that the temperature does not meet the requirements, adjust the parameters of the cooling system of the water turbine, such as the flow rate and temperature of the cooling medium, to ensure that the water turbine operates at the optimal working temperature. The optimization strategy also includes the optimization strategy corresponding to the high-risk parts, that is, the optimization strategy is global for the water turbine. After obtaining the optimization strategy, control the virtual model corresponding to the water turbine to execute the optimization strategy to obtain an execution result, that is, execute it on the actual water turbine according to the execution result, thereby ensuring the optimization effect of the water turbine. The present application adopts a dynamic ensemble learning mechanism and an online update strategy, enabling the model to quickly adapt to the non-stationary working condition changes of the water turbine and realizing a full-process real-time closed-loop response from anomaly detection, fault warning to maintenance decision-making.

[0082] In a specific implementation process of step S103, there is an embodiment as follows: Figure 3 As shown, generating an optimization strategy for the water turbine based on the quality prediction result includes:

[0083] S1031. Determine multiple policy generation dimensions based on the quality prediction result, and generate policy dimension results corresponding to the policy generation dimensions based on the multiple policy generation dimensions and the quality prediction result;

[0084] S1032. Fuse the policy dimension results corresponding to the multiple policy generation dimensions to obtain the optimization strategy of the water turbine.

[0085] In steps S1031 - S1032, the quality prediction result obtained by this application based on the water turbine quality prediction model is obtained by processing various data collected from the water turbine. Therefore, the policy generation dimension is also determined based on various data of the water turbine, including two policy generation dimensions: the state space and the action space dimension. And based on the multiple policy generation dimensions and the quality prediction result, the policy dimension results corresponding to the policy generation dimensions are generated. That is, the reinforcement learning controller generates the corresponding policy dimension results based on the quality prediction result in each policy generation dimension. That is, the water turbine in each dimension is optimized according to the policy dimension result, and the policy dimension results corresponding to the multiple policy generation dimensions are fused according to the power generation efficiency and the equipment loss coefficient to obtain the optimization policy of the water turbine. That is, the optimization policy of the water turbine should be reflected in the power generation efficiency and the equipment loss coefficient, and can be specifically determined according to actual needs. The optimization policy also needs to be consistent with the overall operation goal and long-term plan of the hydropower station, considering the order and priority of the implementation of the optimization policy to ensure the effective utilization of resources and maximize the benefits, and coordinating the conflicts and overlaps between different policies to avoid unnecessary duplicate work or mutual interference.

[0086] In the specific implementation process of step S1031, there is an embodiment: determining multiple policy generation dimensions based on the quality prediction result, including:

[0087] S10311. Based on the state attributes corresponding to each sub-result in the quality prediction result, determine the space dimension mapped by each sub-result;

[0088] S10312. Based on the corresponding relationship between the space dimension and the policy generation dimension, obtain the corresponding policy generation dimension.

[0089] In steps S10311 - S10312, this application does not simply predict whether the quality of the water turbine in the hydropower station is good or bad based on the quality prediction result, but conducts a comprehensive quality prediction of the water turbine based on each sub-result in the quality prediction result. Each sub-result has a corresponding state attribute. The quality prediction result is generated based on the pressure, vibration, guide vane angle, etc. of the water turbine. Then the corresponding state attributes also include the pressure pulsation value, efficiency deviation, vibration amplitude, and guide vane angle adjustment amount. And based on the preset mapping relationship, determine the space dimension mapped by each sub-result. That is, the space dimension is the state space and the action space. Then, according to the preset mapping relationship, the pressure pulsation value, efficiency deviation, and vibration amplitude are mapped to the state space, and the guide vane angle adjustment amount is mapped to the action space. The policy generation dimension corresponding to the state space is the state space, and the policy generation dimension corresponding to the action space is the action space dimension. Thus, the optimization policy of the corresponding water turbine is obtained based on the quality prediction result.

[0090] S104. After the optimization strategy is executed on the virtual model of the water turbine, the execution result is evaluated based on multiple evaluation dimensions. The evaluation result includes two dimensions: power generation efficiency and equipment loss coefficient, which can also be specifically set according to the actual situation. The specific values of power generation efficiency and equipment loss coefficient are specifically adjusted according to the actual working conditions. Then, based on the evaluation results corresponding to the two evaluation dimensions of power generation efficiency and equipment loss coefficient on the virtual model, and focusing on observing whether there is an obvious improvement in the marked places on the virtual model of the water turbine, it is determined whether to apply the optimization strategy to the actual water turbine of the hydropower station. If the evaluation result is excellent, the water turbine is controlled to execute the optimization strategy, adjust the guide vane opening and speed, and realize the full-condition adaptive control of the water turbine to complete the intelligent optimization of the water turbine. If the evaluation result is average, the optimization strategy needs to be adjusted in order to obtain an optimization strategy with an excellent evaluation result to complete the intelligent optimization of the water turbine, realizing an adaptive control strategy driven by reinforcement learning, dynamically balancing equipment loss while ensuring power generation efficiency, extending the service life of key components, and significantly reducing the operation and maintenance cost.

[0091] Embodiment 2

[0092] The present application also provides a water turbine intelligent optimization device, as Figure 4 shown in the block diagram of a water turbine intelligent optimization device. The functions implemented by this water turbine intelligent optimization device correspond to the steps of executing a water turbine intelligent optimization method on a terminal device. This device can be understood as a component of a server including a processor. The water turbine intelligent optimization device described in the present application includes:

[0093] A scanning module 401, configured to scan the water turbine in the hydropower station to obtain a point cloud set corresponding to multiple parts of the water turbine, and traverse the point cloud data in the point cloud set to obtain a virtual model corresponding to the water turbine;

[0094] An acquisition module 402, configured to acquire various data of the water turbine during operation, and input the various data into a pre-trained water turbine quality prediction model, so that the water turbine quality prediction model processes the various data based on the virtual model corresponding to the water turbine and outputs a quality prediction result; the water turbine quality prediction model is established across hydropower stations; the quality prediction result is displayed based on the virtual model;

[0095] An execution module 403, configured to generate an optimization strategy for the water turbine based on the quality prediction result, and control the virtual model corresponding to the water turbine to execute the optimization strategy to obtain an execution result;

[0096] An evaluation module 404, configured to evaluate the execution result based on multiple evaluation dimensions to obtain an evaluation result, and control the water turbine to execute the optimization strategy based on the evaluation result, so as to complete the intelligent optimization of the water turbine.

[0097] In a feasible implementation manner, the execution module includes:

[0098] A generation module, configured to determine multiple policy generation dimensions based on the quality prediction result, and generate a policy dimension result corresponding to the policy generation dimension based on the multiple policy generation dimensions and the quality prediction result;

[0099] A fusion module, configured to fuse the policy dimension results corresponding to the multiple policy generation dimensions to obtain the optimization strategy of the water turbine.

[0100] In a feasible implementation manner, the execution module further includes:

[0101] A first correspondence module, configured to determine the spatial dimension mapped by each sub-result based on the state attribute corresponding to each sub-result in the quality prediction result;

[0102] A second correspondence module, configured to obtain the corresponding policy generation dimension based on the correspondence between the spatial dimension and the policy generation dimension.

[0103] In a feasible implementation manner, the acquisition module includes:

[0104] An extraction module, configured to extract the features of the multiple types of data at multiple scales to obtain multiple data features, and call a preset fusion method based on the multiple data features;

[0105] A processing module, configured to process the multiple data features through the preset fusion method and the corresponding weight coefficients to output a quality prediction result.

[0106] In a feasible implementation manner, the acquisition module further includes:

[0107] A establishment module, configured to pre-establish a local feature sub-model and a heterogeneous model, and set different feature extraction methods for the local feature sub-model and the heterogeneous model;

[0108] An input module, configured to input the multiple types of data into the local feature sub-model and the heterogeneous model, so that the local feature sub-model and the heterogeneous model perform multi-scale feature extraction on the multiple types of data.

[0109] In a feasible implementation manner, the acquisition module also includes:

[0110] An aggregation module, configured to obtain a global model based on parameter combinations of local quality prediction models in each hydropower station received by a central server; the global model includes the prediction error and communication cost of the local quality prediction model; the local quality prediction model is obtained by learning a virtual model of a water turbine in each hydropower station;

[0111] A searching module, configured to search for an optimal hyperparameter combination in the parameter combinations of each local quality prediction model, and fuse the global model and the optimal hyperparameter combination to obtain a water turbine quality prediction model.

[0112] In a feasible implementation manner, the scanning module includes:

[0113] A calculation module, configured to calculate the stress of the part corresponding to the virtual model, and determine whether it meets a preset high-risk condition;

[0114] A marking module, configured to, if so, determine the position information corresponding to the stress, and mark the virtual model based on the position information.

[0115] Embodiment 3

[0116] The present application further provides an electronic device, as Figure 5 shown, including: a processor 501, a memory 502, and a bus 503. The memory 502 stores machine-readable instructions executable by the processor 501. When the electronic device runs, the processor 501 communicates with the memory 502 through the bus 603. When the machine-readable instructions are executed by the processor 501, the steps of any one of the water turbine intelligent optimization methods are executed.

[0117] Embodiment 4

[0118] The present application further provides a computer-readable storage medium, on which a computer program is stored. When the computer program is run by a processor, the steps of any one of the water turbine intelligent optimization methods are executed.

[0119] Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working processes of the systems and devices described above can refer to the corresponding processes in the method embodiments, and will not be elaborated herein. In the several embodiments provided in the present application, it should be understood that the disclosed systems, devices, and methods can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of the modules is only a logical function division, and there can be other division methods in actual implementation. For another example, multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the couplings, direct couplings, or communication connections shown or discussed with each other can be through some communication interfaces. The indirect couplings or communication connections of the devices or modules can be in electrical, mechanical, or other forms.

[0120] The modules described as separate components may or may not be physically separated, and the components shown as modules may or may not be physical units, that is, they can be located in one place or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0121] In addition, in each embodiment of the present application, the functional units can be integrated into one processing unit, or each unit can exist physically alone, or two or more units can be integrated into one unit.

[0122] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a non-volatile computer-readable storage medium executable by a processor. Based on such an understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, a platform server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present application. The aforementioned storage medium includes: various media such as USB flash drives, mobile hard disks, ROM, RAM, magnetic disks, or optical discs that can store program codes.

[0123] The above is only the specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed in the present application, and all of them should be covered by the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A method for intelligent optimization of a hydraulic turbine, characterized in that: The method comprises: Scanning a turbine in a hydropower station to obtain a point cloud set corresponding to multiple parts of the turbine, and traversing the point cloud data in the point cloud set to obtain a virtual model corresponding to the turbine; Collecting various data of the turbine during operation, and inputting the various data into a pre-trained turbine quality prediction model, so that the turbine quality prediction model processes the various data based on a virtual model corresponding to the turbine to output a quality prediction result; the turbine quality prediction model is established across hydropower stations; and the quality prediction result is displayed based on the virtual model; generating an optimization strategy for the water turbine based on the quality prediction result, and controlling a virtual model corresponding to the water turbine to execute the optimization strategy to obtain an execution result; The execution result is evaluated based on multiple evaluation dimensions to obtain an evaluation result, and the water turbine is controlled to execute the optimization strategy based on the evaluation result to complete the intelligent optimization of the water turbine.

2. The method according to claim 1, characterized in that The generating an optimization strategy for the water turbine based on the quality prediction result comprises: Determine a plurality of policy generation dimensions based on the quality prediction result, and generate policy dimension results corresponding to the policy generation dimensions based on the plurality of policy generation dimensions and the quality prediction result; The strategy dimension results corresponding to the multiple strategy generation dimensions are integrated to obtain the optimization strategy of the turbine.

3. The method according to claim 2, characterized in that The determining of a plurality of strategy generation dimensions based on the quality prediction result includes: Determining the spatial dimension of each sub-result mapping based on the state attribute corresponding to each sub-result in the quality prediction result; Based on the corresponding relationship between the spatial dimension and the strategy generation dimension, the corresponding strategy generation dimension is obtained.

4. The method according to claim 1, characterized in that The turbine quality prediction model processes the multiple data and outputs quality prediction results based on the virtual model corresponding to the turbine, including: Extracting features of the multiple data at multiple scales to obtain multiple data features, and calling a preset fusion method based on the multiple data features; The various data features are processed by a preset fusion method and corresponding weight coefficients to output quality prediction results.

5. The method according to claim 4, characterized in that The multi-scale extraction of the features of the multiple data obtains multiple data features, including: Pre-establishing a local feature sub-model and a heterogeneous model, and setting different feature extraction methods for the local feature sub-model and the heterogeneous model; The multiple data are input into the local feature sub-model and the heterogeneous model, so that the local feature sub-model and the heterogeneous model perform multi-scale feature extraction on the multiple data.

6. The method according to claim 1, characterized in that The turbine quality prediction model is a specific step to establish across hydropower stations, including: A global model is obtained by combining the parameters of the local quality prediction models in each hydropower station aggregated and received by the central server; the global model includes the prediction error and communication cost of the local quality prediction model; the local quality prediction model is obtained by learning the virtual model of the turbine in each hydropower station; The optimal hyperparameter combination in the parameter combination of each local quality prediction model is found, and the global model and the optimal hyperparameter combination are integrated to obtain the turbine quality prediction model.

7. The method according to claim 1, characterized in that After traversing the point cloud data in the point cloud set to obtain the virtual model corresponding to the turbine, the method includes: Calculating the stress of the part corresponding to the virtual model and determining whether it meets the preset high-risk condition; If so, the position information corresponding to the stress is determined, and the virtual model is annotated based on the position information.

8. A hydraulic turbine intelligent optimization device, characterized in that: The device comprises: A scanning module, used for scanning a turbine in a hydropower station, obtaining a point cloud set corresponding to multiple parts of the turbine, and traversing the point cloud data in the point cloud set to obtain a virtual model corresponding to the turbine; A collection module, used for collecting various data of the turbine during operation, and inputting the various data into a pre-trained turbine quality prediction model, so that the turbine quality prediction model processes the various data based on a virtual model corresponding to the turbine to output a quality prediction result; the turbine quality prediction model is established across hydropower stations; and the quality prediction result is displayed based on the virtual model; An execution module, used for generating an optimization strategy for the water turbine based on the quality prediction result, and controlling the virtual model corresponding to the water turbine to execute the optimization strategy to obtain an execution result; An evaluation module is used to evaluate the execution result based on multiple evaluation dimensions to obtain an evaluation result, and control the turbine to execute the optimization strategy based on the evaluation result to complete the intelligent optimization of the turbine.

9. An electronic device, characterized in that: include: A processor, a memory and a bus, wherein the memory stores machine-readable instructions executable by the processor, and when the electronic device is running, the processor and the memory communicate via the bus, and when the machine-readable instructions are executed by the processor, the steps of a method for intelligent optimization of a water turbine as described in any one of claims 1 to 7 are performed.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of a water turbine intelligent optimization method as claimed in any one of claims 1 to 7 are executed.

Images