Hydroelectric generating set bolt service stress nephogram inversion optimization algorithm

Abstract

The invention discloses a hydroelectric generating set bolt service stress nephogram inversion optimization algorithm, and relates to the technical field of hydroelectric generating set bolt stress measurement, and the method comprises the following steps: S1, finite element simulation model construction: S2, carrying out data cleaning on physical field simulation data and field measurement data obtained in S1, and obtaining a physical field data set; s3, constructing a stress nephogram inversion model; and S4, searching for optimal parameters. According to the method, the defects in the aspects of actual environment modeling and structure optimization in the prior art are overcome, and a new technical approach is provided for maintenance and performance improvement of the bolts of the hydroelectric generating set. And high-precision simulation in an actual operation environment is realized, and the defects in the aspect of digital modeling in the prior art are overcome. Based on theoretical physical constraints, a reinforcement learning algorithm can be constructed to invert stress distribution of a bolt in long-term service, comprehensive information is provided for structural health monitoring, and certain limitation exists in the aspect in the prior art.

Description

Technical Field

[0001] The invention relates to the technical field of hydropower unit bolt stress calculation, and in particular to an inversion optimization algorithm for a hydropower unit bolt service stress cloud diagram. Background Art

[0002] In the fault diagnosis of hydropower equipment, the existing technologies mainly focus on structural mechanics and material engineering, as well as stress analysis of bolts. The following are the characteristics of the existing technical solutions for fault diagnosis of hydropower bolts: (1) In terms of structural mechanics and materials engineering, existing technologies mainly use the principles of structural mechanics and materials engineering to model and analyze the bolt structure of hydropower units. They focus on the static analysis of the structure and consider the basic mechanical properties of the material, such as strength and stiffness.

[0003] (2) In terms of bolt stress analysis, the stress analysis of bolts is mainly based on the principle of statics, focusing on the stress state of the bolts under stress conditions; according to the material and geometric shape of the bolts, the stress distribution under stress state is calculated.

[0004] (3) In terms of finite element simulation technology: Common technical solutions use finite element simulation technology, but there are usually simplifications in modeling, which may not accurately reflect the actual working environment. While considering the overall structure, finite element simulation has certain difficulties in fine modeling of local areas.

[0005] (4) In terms of digital twin technology: Most existing technologies do not fully utilize digital twin technology, lack accurate digital simulation of the actual operating environment, and fail to fully consider the impact of actual operating data on the structure and bolt stress. Summary of the invention

[0006] The present invention aims to provide an inversion optimization algorithm for the service stress cloud map of the bolts of a hydropower unit. By designing a finite element simulation model of the bolts of the hydropower unit, the actual operating status of the components can be accurately characterized; by cleaning the finite element simulation data and the actual measurement data, high-quality data can be obtained for the training and testing of the prediction model; an embedded physical knowledge neural network is constructed to achieve finite element input parameter optimization and rapid reconstruction of the three-dimensional flow field; the finite element evaluation index is used to construct the system's evaluator network and the participant network based on gradient descent iterative update, so as to achieve the embedded physical knowledge neural network parameter update and iterative optimization of the finite element simulation input parameters of the hydropower unit component model.

[0007] In order to achieve the above-mentioned object of the invention, the technical solution of the present invention is as follows: An inversion optimization algorithm for the service stress cloud diagram of a hydropower unit bolt comprises the following steps: S1: Finite element simulation model construction, including the following steps: S1.1: Determine the basic structure of the component: The hydropower unit is decomposed into a fluid calculation domain and a structural calculation domain, and three-dimensional modeling is performed on the fluid calculation domain of the hydropower unit and the top cover bolt structure calculation domain using three-dimensional modeling software; The fluid calculation domain includes the volute, fixed guide vanes and movable guide vanes, runner area and tailwater pipe; the structural calculation domain includes the outer top cover flange, inner top cover and bolts.

[0008] S1.2: Establishment and calculation of finite element simulation model: After meshing the three-dimensional models of the fluid calculation domain and the top cover bolt structure calculation domain in step S1.1, a finite element simulation model is established. Contact constraints of each component and boundary conditions of the fluid calculation domain are set in the finite element simulation model. Multiple groups of different operating loads are input into the finite element simulation model to establish a finite element simulation model of the hydropower unit. Flow-solid coupling simulation calculations are carried out to obtain physical field simulation data under different operating loads.

[0009] The mesh division is to use unstructured tetrahedral meshes to mesh the entire flow channel of the hydropower unit, and to encrypt the meshes of the top cover flow surface area, the impeller and the guide vane.

[0010] The boundary conditions of the fluid calculation domain include the volute inlet (using the pressure inlet boundary condition, given the total working head of the hydropower unit), the wall (using the no-slip wall boundary condition), the tailwater outlet (using the mass flow boundary condition), and the runner area (since the runner is a rotating component, this area uses the rotating reference coordinate system method, the runner speed value is given, and a frozen rotor is used at the dynamic and static interface).

[0011] The contact constraints are horizontal contact between the inner top cover flange and the outer top cover flange, horizontal contact between the bolt head and the inner top cover flange, and contact between the bolt bottom thread and the inner top cover thread hole.

[0012] Different working load conditions include: multiple groups of working load conditions with different load positions, different load intensities and different load application methods.

[0013] The fluid-solid coupling simulation calculation includes applying the gravity load of the turbine according to the gravitational acceleration, applying the preload of the bolts and locking the preload of the bolts, applying water pressure to the top cover and the water-guide bearing seat, that is, the water pressure load calculated by the external fluid is transmitted to the top cover and the water-guide bearing seat through the fluid-solid coupling surface, and solving the fluid-solid coupling equation through fluid-solid coupling simulation according to the working load conditions and boundary conditions to obtain the physical field simulation data under different working conditions.

[0014] The establishment of the finite element simulation model also includes mesh independence verification. The mesh independence verification increases the number of meshes on the basis of the existing finite element simulation model. When the change of the physical field does not exceed 5% with the increase of the number of meshes, the verification is completed. When the change of the physical field exceeds 5% with the increase of the number of meshes, the number of meshes continues to be increased for independence verification until the verification is completed.

[0015] S2: Perform data cleaning on the physical field simulation data and field measurement data obtained in step S1.2 to obtain a physical field data set; Data cleaning includes the following steps: S2.1: Handle missing values: Detect missing values ​​in physical field simulation data and field measurement data, and handle missing values ​​by filling in missing values ​​or deleting rows or columns containing missing values.

[0016] S2.2: Handling outliers: Detect outliers in physical field simulation data and field measurement data, and handle outliers using methods including substitution, deletion, normalization, and verification of data.

[0017] S2.3: Convert data types: Determine the data type required for each field based on the analysis objectives or modeling requirements, and adjust the types of physical field simulation data and field measurement data accordingly, including conversion of numerical and categorical data.

[0018] S2.4: Standardized data: Standardize the physical field simulation data and field measurement data so that each set of data has the same dimension and distribution, matches the input dimension of the reinforcement learning algorithm, and obtains the physical field data set; to facilitate the subsequent construction of the service stress cloud map inversion optimization algorithm.

[0019] S3: using the physical field data set obtained in step S2, establishing a reinforcement learning stress cloud map inversion model for the bolts of the hydropower unit, and iteratively updating the reinforcement learning stress cloud map inversion model for the bolts of the hydropower unit, and verifying the effect of the finite element stress cloud map inversion by comparing the running results of the actual working conditions with the finite element simulation and the inversion results; The method of establishing a reinforcement learning stress cloud map inversion model for bolts of a hydropower unit comprises: constructing a Markov decision chain model with the physical field data set obtained in step S2, establishing a reinforcement learning model, applying the reinforcement learning model to solve the Markov decision chain model, and coupling the reinforcement learning model with the Markov decision chain model to obtain a reinforcement learning stress cloud map inversion model for bolts of a hydropower unit; The finite element input parameters of the physical field data set are used as the input of the reinforcement learning stress cloud map inversion model of the hydropower unit bolts to invert the stress cloud map of the hydropower unit bolts, wherein the stress cloud map of the hydropower unit bolts is the stress value of each grid point of the finite element simulation model after grid division, wherein the stress value of the boundary of the stress cloud map is the boundary stress value; The Markov decision chain model consists of the states s t 、Action a t , Markov decision chain model feedback r ( s t ); The status s t The stress inversion result of the initialization decision of the Markov decision chain model is t The stress inversion result is s t+1 , action a t The step size for changing the stress magnitude each time is used to change the state of each iterative cycle. s t ; Through the Navier-Stokes equation, using the solid wall boundary condition, that is, the condition that the velocity of the fluid at the solid boundary is zero, the stress boundary value of the theoretical physical field is calculated through the fluid physical parameters in the hydropower unit; the general form of the Navier-Stokes equation is: Among them, ρ is the density of the fluid, v is the velocity field, p is the pressure field, μ is the viscosity coefficient of the fluid, and f is the external force, such as gravity; For the bolts of the hydropower unit, set the solid wall boundary condition, set the velocity at the boundary to zero, solve the equation, and get the current state s t The stress boundary value of Markov decision chain model feedback is the current state t The mean square error of the stress value and the stress value of the physical field data set, plus the difference between the calculated theoretical stress cloud map boundary value and the boundary stress value inverted by the stress cloud map inversion model, is used to measure the current predicted stress, that is, the state s t The difference between the real data set; the iterative formula of the Markov decision chain model state in each cycle is as follows: .

[0020] The reinforcement learning model includes a strategy network I (participants) and a value network I (evaluators); The strategy network I is responsible for making decisions, that is, making decisions on the direction of change (increase or decrease) of the finite element input parameters; the value network I is responsible for evaluating the quality of the decision output by the strategy network I, that is, calculating the error based on the predicted value of the finite element simulation result and the true value of the simulation result in the data set. The smaller the error, the better the decision; The strategy network I consists of a radial basis function neural network, including an input layer, two hidden layers and an output layer, which generates the corresponding control strategy by gradually accumulating system experience; The value network I part includes an input layer, a hidden layer and an output layer. It evaluates the performance of the current finite element parameter decision by comparing the difference between the physical field simulation data of the finite element simulation results and the actual test data. It generates rewards (error becomes smaller) or penalties (error becomes larger) as feedback values ​​for adaptive learning according to the trend of error changes, and introduces a long-term cost function as an evaluation criterion. The input and output dimensions (data format) of the strategy network I and the value network I are determined by the number of input parameters of the finite element simulation and the number of nodes in the extracted finite element simulation physical field.

[0021] The iterative update includes: updating the strategy network I through a stochastic gradient descent algorithm, and correcting the state value of the Markov decision chain model through the Bellman iterative equation, and performing simulation application and verification to achieve inversion prediction of the stress cloud map; The updating of the policy network I by the stochastic gradient descent algorithm includes using the gradient method in the embedded physical knowledge policy network I to update the policy network I by stochastic gradient descent (how to update is described in the next paragraph), and ensuring that the finite element parameters generated by the policy network I can maximize the Markov decision chain model evaluation index through the evaluation of the value network I; The stochastic gradient descent algorithm is used to train the objective function of the strategy network I, measure the performance of the current strategy, and guide the update of network parameters; the stochastic gradient descent algorithm is composed of the sum of two parts: the first part is a loss function composed of the difference between the finite element simulation result corresponding to the sensor and the actual measurement value, and the second part is the difference between the field boundary condition predicted by the strategy network I and the ideal boundary condition; the predicted boundary condition is the velocity and pressure on the fluid boundary in the simulation result, and the ideal boundary condition is the velocity and pressure on the fluid boundary that meets the previous fluid-solid coupling basic equation; The updating of the policy network I and the value network I by the stochastic gradient descent algorithm includes: first, the target Markov decision chain model randomly generates the initial state of the Markov decision chain model or the current state s after iterative update t , based on the strategy π generated by the initialization of the strategy network I, according to the current Markov decision chain model state s t To decide to perform action a t , and the action a t Interact with the Markov decision chain model and transfer to the new state s according to the iterative formula of the Markov decision chain model state t+1 , get the Markov decision chain model feedback ; Value Network I estimates the value of an action through feedback from a Markov decision chain model , The expression is: Among them, S is the state of the Markov decision chain model, G t Feedback for Markov decision chain model The mathematical expectation of Each state-action group (s t , a t )'s advantage function A π : The gradient of the strategy network I is calculated based on the advantage function, and the strategy network I is updated according to the stochastic gradient descent algorithm, thereby correcting the value judgment and strategy formulation process of the strategy network I on the state of the Markov decision chain model; the iteration of the value network I needs to correct the state value of the Markov decision chain model based on the Bellman iteration equation, and calculate the parameter gradients before and after the correction, and update the value network I according to the stochastic gradient descent algorithm; Among them, the gradient calculation of policy network I is as follows: To define the network parameters of the strategy network I, In a given state s t Next, action a t The log probability of the corresponding strategy; is the advantage function.

[0022] The iterative formula of the value function is as follows: in, is the attenuation coefficient, Feedback for the Markov decision chain model.

[0023] The simulation application and verification include: the reinforcement learning algorithm interacts with the Markov decision chain model in each iteration cycle, and the strategy network I is based on the state of the Markov decision chain model s t Generate decision action a t , the decision action a t To make a decision on the direction of change of the finite element input parameters; the Markov decision chain model is based on the decision action a t Feedback is generated, and the Markov decision chain model state changes from s tTransform to s t+1 The value network I is responsible for evaluating the quality of the decision output by the strategy network I. The evaluation is based on the feedback of the Markov decision chain model. , that is, the difference between the predicted value of the finite element simulation result and the calculated true value of the simulation result in the data set plus the difference between the theoretical stress cloud map boundary value calculated based on the physical formula and the inverted stress boundary value. The sum of the two differences is the error. The smaller the error, the better the decision. When the error is greater than 5%, the stress cloud map inversion optimization algorithm continues to iterate; when the error is less than 5%, the stress cloud map inversion optimization algorithm iteration is completed.

[0024] S4: optimal parameter search, including using the hydropower unit bolt reinforcement learning stress cloud map inversion model obtained in step S3 to establish a hydropower unit bolt reinforcement learning working condition parameter search model, perform optimal working condition parameter search, and verify and visualize the search results to obtain a hydropower unit bolt service stress cloud map inversion optimization algorithm; The S4 step includes the following steps: S4.1: Using the hydropower unit bolt reinforcement learning stress cloud map inversion model, a hydropower unit operating parameter search model is constructed, including: Obtain the operating environment of the hydropower unit under different working conditions through historical data or measurements, select the appropriate boundary type, and determine the value of each boundary condition based on the measured data or standard specifications of the working condition; select the objective function in the optimization task based on performance indicators or energy efficiency, and map the stress indicators in the simulation results to the objective function to ensure that the objective function can truly reflect the stress state of the equipment components and is closely related to the actual operating conditions; set the range of each action variable in the environment and the constraints of the water pressure distribution based on the actual situation, and construct a parameter search model for the working condition of the hydropower unit; Boundary types include fixed boundary conditions and symmetric boundary conditions.

[0025] S4.2: In the hydropower unit operating parameter search model obtained in step S4.1, a reinforcement learning environment for operating design parameters of the operating parameter search model is constructed by using the maximum entropy reinforcement learning algorithm, including: S4.21: Construct a maximum entropy reinforcement learning algorithm and introduce an adjustable entropy term into the maximum entropy reinforcement learning algorithm ; The maximum entropy reinforcement learning algorithm includes value network II and policy network II. In each iteration cycle, policy network II is based on the state of the reinforcement learning environment. s t Take action on it a t , and obtain the reward function from the reinforcement learning environment, the state s tTransition to a new state s t+1 ; Value Network II evaluates the sum of reward function and policy entropy To understand the current state of the reinforcement learning environment s t Perform value assessment; based on advantage function To calculate the gradient of the policy network II, and then update the policy network II according to the stochastic gradient descent algorithm, and then correct the value judgment and strategy formulation process of the policy network II on the system state; the iteration of the value network II needs to correct the value of the system state based on the Bellman iteration equation, and calculate the parameter gradients before and after the correction, and update the value network II according to the stochastic gradient descent algorithm; To define the network parameters of strategy network II, is in a given state s t Next, action The log probability of the corresponding strategy; is the advantage function; S4.22: In the reinforcement learning working condition parameter search model of the hydropower unit bolt, the control variables in the finite element simulation environment are analyzed, and by mapping these control variables to actions in reinforcement learning, the action space of reinforcement learning is determined, so that the strategy network II can adjust these variables for optimization under different working conditions; S4.23: Determine the state space in the reinforcement learning environment. Considering the state changes caused by the actions, the strategy characteristics can be defined as the stress distribution of the bolts. The inherent characteristics that do not change in the environment are the dimensions of the constructed three-dimensional hydropower unit model and the fluid-structure coupling calculation form. S4.24: Determine environmental constraints and design a reinforcement learning reward function based on the objective function; the reward function provides positive feedback for actions close to the optimal parameters, guides the strategy network II to learn and select the optimal working condition parameters, and ensures that the reward function reflects the optimization direction of the objective function; S4.3: Obtaining the optimal operating parameters: The operating parameter search model uses the maximum entropy reinforcement learning algorithm in the reinforcement learning environment to complete the online search and optimization of the operating parameters, iterates the reinforcement learning optimal strategy based on the maximum entropy reinforcement learning, and searches for the optimal operating parameters based on the optimal strategy; including: Based on the reinforcement learning environment established in S4.2, the reinforcement learning environment is solved by the maximum entropy reinforcement learning algorithm; By cumulative reward and policy entropy The two items constitute the evaluation function of the value function, and the optimal strategy is calculated through the evaluation function of the value function. The optimal strategy is imported into the strategy network II and output to obtain the optimal working condition parameters; The evaluation function of the value function is: in, For state actions The probability distribution of the group is The expected reward value when s t and a t are the state and action at time t respectively; For the status s t Next action a t The reward value obtained; Status s t The entropy value of strategy π when The temperature parameter determines its ability to explore different strategies; π For control strategies; π * is the optimal control strategy.

[0026] S4.4: Verify the optimal operating parameters obtained from the search, including: The finite element simulation is performed using the optimal operating parameters obtained through the search. The differences between the physical field simulation data in the finite element simulation results and the physical field simulation data output by the operating parameter search model are compared to verify the feasibility and effectiveness of the search algorithm. A sampling algorithm is used to generate multiple new parameter points near the optimal search parameters. The optimal working condition parameters obtained by the search and the sampling points near them and the corresponding finite element simulation data are expanded into a new training set for online iterative update of the reinforcement learning stress cloud map inversion model. The optimal operating parameters are verified on the actual equipment, the actual working performance is observed and measured, and the finite element simulation model, stress cloud map inversion model and operating parameter search model are integrated. The overall model is a digital twin model. The operating parameters before and after optimization are applied to the actual equipment and the finite element simulation model respectively, and the comparison results before and after optimization are observed to confirm the effectiveness of the optimal operating parameters obtained in the digital twin model.

[0027] S4.5: Display the parameter search results, including: The parameter search and iteration process is visualized in the form of icons, including the time and accuracy of each iteration. When the accuracy of coincidence between the optimal parameters obtained by the search and the physical field under actual working conditions reaches more than 90%, the iterative training stops and the inversion optimization algorithm of the service stress cloud map of the bolts of the hydropower unit is obtained.

[0028] The service stress cloud map inversion optimization algorithm also includes verification and error analysis of the physical field simulation results in step S1.2, and the verification and error analysis of the physical field simulation results include: verifying the finite element simulation model in step S1.2, and comparing the error between the physical field simulation data predicted by the finite element simulation model and the actual measurement value; when the error is greater than 10%, temporarily retaining the group of finite element simulation data, and using the finite element simulation data that meets the error condition to establish an inversion optimization algorithm model, and optimizing the working condition parameter load in the current finite element simulation through the inversion optimization algorithm, and updating the finite element simulation until the error is less than 10%, and the error condition is met; adding the group of finite element simulation data to the training set of the inversion optimization algorithm model; The field measurement data includes stress in the actual environment. The stress measurement in the actual environment adopts a hollow cylindrical force sensor, which includes a built-in elastic element, a resistance strain gauge, a temperature compensation circuit and a conversion circuit. The sensor measures the actual average stress on the bolt. A single sensor is installed between the top cover bolt and the inner top cover. The field test condition is the entire start-stop process of the unit under high water head from shutdown to startup, changing the opening and finally shutting down. The bolt stress distribution measured under actual conditions is compared and analyzed with the finite element simulation results. The actual measured value is the average stress of the actual bolt measured by the hollow cylindrical force sensor; the field test condition is the entire start-stop process of the unit under high water head from shutdown to startup, changing the opening degree to the final shutdown.

[0029] Beneficial effects of the present invention: 1. The inversion optimization algorithm for the service stress cloud map of the bolts of the hydropower unit in the present invention constructs a stress cloud map inversion model based on the finite element stress simulation results for different working conditions of the hydropower unit, and searches for the optimal working condition parameters of the bolts of the hydropower unit based on the constructed stress cloud map inversion model. Compared with other working condition parameter search algorithms, this algorithm uses the finite element simulation database to establish a stress cloud map inversion model. During the search process, the search algorithm does not need to interact with long-term finite element simulation, but interacts with the efficient stress cloud map inversion model, which greatly improves the search efficiency and saves hardware resources such as computing power and time resources. In addition, during the model training process, the accuracy of the model is strictly controlled, so the working condition parameter search results of this algorithm are accurate and reliable.

[0030] 1. The finite element simulation model of the bolts of the hydropower unit in the present invention establishes an accurate prediction model of the normal state of the power generation equipment components for different working conditions and their corresponding physical field simulation data, so as to measure the physical field simulation data under various working conditions. The difference between the model prediction value and the actual measurement value is used to evaluate the accuracy of the prediction model, accurately characterize the operating state of the power generation equipment components, realize the equipment state simulation, and provide simulation data for the cloud map inversion optimization.

[0031] 2. The efficient data cleaning and preprocessing method in the present invention develops high-quality and effective data cleaning methods for local finite element simulation test data and actual measurement data of power generation equipment components to ensure high-quality data output and establish highly robust joint data for use as a basis for training and testing core algorithm links.

[0032] 3. The present invention adopts a reinforcement learning algorithm based on the policy gradient optimization algorithm and the Bellman iterative equation, and guides the convergence of reinforcement learning based on embedded physical knowledge, making the autonomous decision-making of Markov decision chain model parameters more intelligent and accurate. The reinforcement learning algorithm enhances the learning and adaptability of the Markov decision chain model, and has a positive effect on improving the performance of the Markov decision chain model.

[0033] 4. The optimal operating condition parameter search algorithm in the present invention: For the prediction model of physical field simulation data under different working conditions, a deep reinforcement learning framework based on intelligent technology is used to develop an automatic search algorithm for optimal operating condition parameters. A reasonable objective function for updating the optimal strategy is constructed to achieve reasonable exploration of the optimal parameters of power generation equipment components and enhance the algorithm's exploration ability for the prediction model. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 The present invention is a schematic diagram of the process flow of the inversion optimization algorithm for the service stress cloud diagram of the bolts of the hydropower unit. DETAILED DESCRIPTION

[0035] The present invention is further described in detail below in conjunction with examples, but the embodiments of the present invention are not limited thereto.

[0036] Example 1 like Figure 1 As shown, this embodiment provides an inversion optimization algorithm for the service stress cloud diagram of the bolts of a hydropower unit, comprising the following steps: S1: Finite element simulation model construction, including the following steps: S1.1: Determine the basic structure of the components: decompose the hydropower unit into the fluid calculation domain and the structural calculation domain, and use 3D modeling software to perform 3D modeling on the fluid calculation domain of the hydropower unit and the top cover bolt structure calculation domain; S1.2: Establishment and calculation of finite element simulation model: After meshing the three-dimensional model of the fluid calculation domain and the top cover bolt structure calculation domain in step S1.1, a finite element simulation model is established. Contact constraints of various components and boundary conditions of the fluid calculation domain are set in the finite element simulation model. Multiple groups of different working condition loads are input into the finite element simulation model to establish a finite element simulation model of the hydropower unit. Fluid-solid coupling simulation calculations are performed to obtain physical field simulation data under different working condition loads. S2: Perform data cleaning on the physical field simulation data and field measurement data obtained in step S1.2 to obtain a physical field data set; S3: Construction of stress cloud map inversion model: Using the physical field data set obtained in step S2, a stress cloud map inversion model of the hydropower unit bolt reinforcement learning is established, and the stress cloud map inversion model of the hydropower unit bolt reinforcement learning is iteratively updated. The effect of the finite element stress cloud map inversion is verified by comparing the running results of the actual working conditions with the finite element simulation and the inversion results; S4: optimal parameter search, including using the hydropower unit bolt reinforcement learning stress cloud map inversion model obtained in step S3 to establish a hydropower unit bolt reinforcement learning operating parameter search model, perform optimal operating parameter search, and verify and visualize the search results to obtain the optimal operating parameters of the hydropower unit bolts.

[0037] The inversion optimization algorithm for the service stress cloud map of the bolts of the hydropower unit in this embodiment constructs a stress cloud map inversion model based on the finite element stress simulation results for different working conditions of the hydropower unit, and searches for the optimal working condition parameters of the bolts of the hydropower unit based on the constructed stress cloud map inversion model. Compared with other working condition parameter search algorithms, this algorithm uses the finite element simulation database to establish a stress cloud map inversion model. During the search process, the search algorithm does not have to interact with long-term finite element simulation, but interacts with the efficient stress cloud map inversion model, which greatly improves the search efficiency and saves hardware resources such as computing power and time resources. In addition, during the model training process, the accuracy of the model is strictly controlled, so the working condition parameter search results of this algorithm are accurate and reliable.

[0038] Example 2 The difference between this embodiment and embodiment 1 is that, in this embodiment, step S1 includes the following steps: S1.1: Determine the basic structure of the component: The hydropower unit is decomposed into a fluid calculation domain and a structural calculation domain, and three-dimensional modeling is performed on the fluid calculation domain of the hydropower unit and the top cover bolt structure calculation domain using three-dimensional modeling software; The fluid calculation domain includes the volute, fixed guide vanes and movable guide vanes, runner area and tailwater pipe; the structural calculation domain includes the outer top cover flange, inner top cover and bolts.

[0039] S1.2: Establishment and calculation of finite element simulation model: After meshing the three-dimensional models of the fluid calculation domain and the top cover bolt structure calculation domain in step S1.1, a finite element simulation model is established. Contact constraints of each component and boundary conditions of the fluid calculation domain are set in the finite element simulation model. Multiple groups of different operating loads are input into the finite element simulation model to establish a finite element simulation model of the hydropower unit. Flow-solid coupling simulation calculations are carried out to obtain physical field simulation data under different operating loads.

[0040] The mesh division is to use unstructured tetrahedral meshes to mesh the entire flow channel of the hydropower unit, and to encrypt the meshes of the top cover flow surface area, the impeller and the guide vane.

[0041] The boundary conditions of the fluid calculation domain include the volute inlet (using the pressure inlet boundary condition, given the total working head of the hydropower unit), the wall (using the no-slip wall boundary condition), the tailwater outlet (using the mass flow boundary condition), and the runner area (since the runner is a rotating component, this area uses the rotating reference coordinate system method, the runner speed value is given, and a frozen rotor is used at the dynamic and static interface).

[0042] The contact constraints are horizontal contact between the inner top cover flange and the outer top cover flange, horizontal contact between the bolt head and the inner top cover flange, and contact between the bolt bottom thread and the inner top cover thread hole.

[0043] Different working load conditions include: multiple groups of working load conditions with different load positions, different load intensities and different load application methods.

[0044] The fluid-solid coupling simulation calculation includes applying the gravity load of the turbine according to the gravitational acceleration, applying the preload of the bolts and locking the preload of the bolts, applying water pressure to the top cover and the water-guide bearing seat, that is, the water pressure load calculated by the external fluid is transmitted to the top cover and the water-guide bearing seat through the fluid-solid coupling surface, and solving the fluid-solid coupling equation through fluid-solid coupling simulation according to the working load conditions and boundary conditions to obtain the physical field simulation data under different working conditions.

[0045] The establishment of the finite element simulation model also includes a grid independence verification, which increases the number of grids on the basis of the existing finite element simulation model; when the change of the physical field does not exceed 5% as the number of grids increases, the verification is completed; when the change of the physical field exceeds 5% as the number of grids increases, the number of grids is continued to be increased for independence verification until the verification is completed. The remaining steps are the same as in Example 1.

[0046] The finite element simulation model of the bolts of the hydropower unit in this embodiment establishes an accurate prediction model of the normal state of the power generation equipment components for different working conditions and their corresponding physical field simulation data, so as to measure the physical field simulation data under various working conditions. The difference between the model prediction value and the actual measurement value is used to evaluate the accuracy of the prediction model, accurately characterize the operating state of the power generation equipment components, realize the equipment state simulation, and provide simulation data for the cloud map inversion optimization.

[0047] Example 3 The difference between this embodiment and embodiment 1 is that, in this embodiment, the data cleaning in step S2 includes the following steps: S2.1: Handle missing values: Detect missing values ​​in physical field simulation data and field measurement data, and handle missing values ​​by filling in missing values ​​or deleting rows or columns containing missing values.

[0048] S2.2: Handling outliers: Detect outliers in physical field simulation data and field measurement data, and handle outliers using methods including substitution, deletion, normalization, and verification of data.

[0049] S2.3: Convert data types: Determine the data type required for each field based on the analysis objectives or modeling requirements, and adjust the types of physical field simulation data and field measurement data accordingly, including conversion of numerical and categorical data.

[0050] S2.4: Standardizing data: Standardize the physical field simulation data and field measurement data so that each set of data has the same dimension and distribution, matches the input dimension of the reinforcement learning algorithm, and obtains a physical field data set; so as to facilitate the subsequent construction of the service stress cloud map inversion optimization algorithm. The remaining steps are the same as in Example 1.

[0051] The highly efficient data cleaning and preprocessing method in this embodiment develops a high-quality and effective data cleaning method for local finite element simulation test data and actual measurement data of power generation equipment components to ensure high-quality data output and establish highly robust joint data for use as a basis for training and testing core algorithm links.

[0052] Example 4 The difference between this embodiment and embodiment 1 is that, in this embodiment, step S3 includes the following steps: The method of establishing a reinforcement learning stress cloud map inversion model for bolts of a hydropower unit comprises: constructing a Markov decision chain model with the physical field data set obtained in step S2, establishing a reinforcement learning model, applying the reinforcement learning model to solve the Markov decision chain model, and coupling the reinforcement learning model with the Markov decision chain model to obtain a reinforcement learning stress cloud map inversion model for bolts of a hydropower unit; The finite element input parameters of the physical field data set are used as the input of the reinforcement learning stress cloud map inversion model of the hydropower unit bolts to invert the stress cloud map of the hydropower unit bolts, wherein the stress cloud map of the hydropower unit bolts is the stress value of each grid point of the finite element simulation model after grid division, wherein the stress value of the boundary of the stress cloud map is the boundary stress value; The Markov decision chain model consists of the states s t 、Action a t , Markov decision chain model feedback r ( s t ); The statuss t The stress inversion result of the initialization decision of the Markov decision chain model is t The stress inversion result is s t+1 , action a t The step size for changing the stress magnitude each time is used to change the state of each iterative cycle. s t； Through the Navier-Stokes equation, using the solid wall boundary condition, that is, the condition that the velocity of the fluid at the solid boundary is zero, the stress boundary value of the theoretical physical field is calculated through the fluid physical parameters in the hydropower unit; the general form of the Navier-Stokes equation is: Among them, ρ is the density of the fluid, v is the velocity field, p is the pressure field, μ is the viscosity coefficient of the fluid, and f is the external force, such as gravity; For the bolts of the hydropower unit, set the solid wall boundary condition, set the velocity at the boundary to zero, solve the equation, and get the current state s t The stress boundary value of Markov decision chain model feedback is the current state t The mean square error of the stress value and the stress value of the physical field data set, plus the difference between the calculated theoretical stress cloud map boundary value and the boundary stress value inverted by the stress cloud map inversion model, is used to measure the current predicted stress, that is, the state s t The difference between the real data set; the iterative formula of the Markov decision chain model state in each cycle is as follows: .

[0053] The reinforcement learning model includes a strategy network I (participants) and a value network I (evaluators); The strategy network I is responsible for making decisions, that is, making decisions on the direction of change (increase or decrease) of the finite element input parameters; the value network I is responsible for evaluating the quality of the decision output by the strategy network I, that is, calculating the error based on the predicted value of the finite element simulation result and the true value of the simulation result in the data set. The smaller the error, the better the decision; The strategy network I consists of a radial basis function neural network, including an input layer, two hidden layers and an output layer, which generates the corresponding control strategy by gradually accumulating system experience; The value network I part includes an input layer, a hidden layer and an output layer. It evaluates the performance of the current finite element parameter decision by comparing the difference between the physical field simulation data of the finite element simulation results and the actual test data. It generates rewards (error becomes smaller) or penalties (error becomes larger) as feedback values ​​for adaptive learning according to the trend of error changes, and introduces a long-term cost function as an evaluation criterion. The input and output dimensions (data format) of the strategy network I and the value network I are determined by the number of input parameters of the finite element simulation and the number of nodes in the extracted finite element simulation physical field.

[0054] The iterative update includes: updating the strategy network I through a stochastic gradient descent algorithm, and correcting the state value of the Markov decision chain model through the Bellman iterative equation, and performing simulation application and verification to achieve inversion prediction of the stress cloud map; The updating of the policy network I by the stochastic gradient descent algorithm includes using the gradient method in the embedded physical knowledge policy network I to update the policy network I by stochastic gradient descent (how to update is described in the next paragraph), and ensuring that the finite element parameters generated by the policy network I can maximize the Markov decision chain model evaluation index through the evaluation of the value network I; The stochastic gradient descent algorithm is used to train the objective function of the strategy network I, measure the performance of the current strategy, and guide the update of network parameters; the stochastic gradient descent algorithm is composed of the sum of two parts: the first part is a loss function composed of the difference between the finite element simulation result corresponding to the sensor and the actual measurement value, and the second part is the difference between the field boundary condition predicted by the strategy network I and the ideal boundary condition; the predicted boundary condition is the velocity and pressure on the fluid boundary in the simulation result, and the ideal boundary condition is the velocity and pressure on the fluid boundary that meets the previous fluid-solid coupling basic equation; The updating of the policy network I and the value network I by the stochastic gradient descent algorithm includes: first, the target Markov decision chain model randomly generates the initial state of the Markov decision chain model or the current state s after iterative update t , based on the strategy π generated by the initialization of the strategy network I, according to the current Markov decision chain model state s t To decide to perform action a t , and the action a t Interact with the Markov decision chain model and transfer to the new state s according to the iterative formula of the Markov decision chain model state t+1 , get the Markov decision chain model feedback ; Value Network I estimates the value of an action through feedback from a Markov decision chain model , The expression is: Among them, S is the state of the Markov decision chain model, G t Feedback for Markov decision chain model The mathematical expectation of Each state-action group (s t , a t )'s advantage function A π : The gradient of the strategy network I is calculated based on the advantage function, and the strategy network I is updated according to the stochastic gradient descent algorithm, thereby correcting the value judgment and strategy formulation process of the strategy network I on the state of the Markov decision chain model; the iteration of the value network I needs to correct the state value of the Markov decision chain model based on the Bellman iteration equation, and calculate the parameter gradients before and after the correction, and update the value network I according to the stochastic gradient descent algorithm; Among them, the gradient calculation of policy network I is as follows: To define the network parameters of the strategy network I, In a given state s t Next, action a t The log probability of the corresponding strategy; is the advantage function.

[0055] The iterative formula of the value function is as follows: in, is the attenuation coefficient, Feedback for the Markov decision chain model.

[0056] The simulation application and verification include: the reinforcement learning algorithm interacts with the Markov decision chain model in each iteration cycle, and the strategy network I is based on the state of the Markov decision chain model s t Generate decision action a t , the decision action a t To make a decision on the direction of change of the finite element input parameters; the Markov decision chain model is based on the decision action a t Feedback is generated, and the Markov decision chain model state changes from s t Transform to s t+1 ; The value network I is responsible for evaluating the quality of the decision output by the strategy network I. The evaluation is based on the feedback of the Markov decision chain model. , that is, the difference between the predicted value of the finite element simulation result and the actual value of the simulation result in the data set plus the difference between the theoretical stress cloud map boundary value calculated based on the physical formula and the inverted stress boundary value, the sum of the two differences is the error, the smaller the error, the better the decision; when the error is greater than 5%, continue to iterate the stress cloud map inversion optimization algorithm; when the error is less than 5%, the stress cloud map inversion optimization algorithm is iterated. The remaining steps are the same as in Example 1.

[0057] In this embodiment, a reinforcement learning algorithm based on a policy gradient optimization algorithm and Bellman iterative equation is used, and the convergence of reinforcement learning is guided based on embedded physical knowledge, making the system parameter autonomous decision more intelligent and accurate. The reinforcement learning algorithm enhances the learning and adaptability of the system and has a positive effect on improving system performance.

[0058] Example 5 The difference between this embodiment and embodiment 1 is that, in this embodiment, step S4 includes the following steps: S4.1: Using the hydropower unit bolt reinforcement learning stress cloud map inversion model, a hydropower unit operating parameter search model is constructed, including: Obtain the operating environment of the hydropower unit under different working conditions through historical data or measurements, select the appropriate boundary type, and determine the value of each boundary condition based on the measured data or standard specifications of the working condition; select the objective function in the optimization task based on performance indicators or energy efficiency, and map the stress indicators in the simulation results to the objective function to ensure that the objective function can truly reflect the stress state of the equipment components and is closely related to the actual operating conditions; set the range of each action variable in the environment and the constraints of the water pressure distribution based on the actual situation, and construct a parameter search model for the working condition of the hydropower unit; Boundary types include fixed boundary conditions and symmetric boundary conditions.

[0059] S4.2: In the hydropower unit operating parameter search model obtained in step S4.1, a reinforcement learning environment for operating design parameters of the operating parameter search model is constructed by using the maximum entropy reinforcement learning algorithm, including: S4.21: Construct a maximum entropy reinforcement learning algorithm and introduce an adjustable entropy term into the maximum entropy reinforcement learning algorithm ; The maximum entropy reinforcement learning algorithm includes value network II and policy network II. In each iteration cycle, policy network II is based on the state of the reinforcement learning environment. s t Take action on it a t , and obtain the reward function from the reinforcement learning environment, the state s t Transition to a new state st+1 ; Value Network II evaluates the sum of reward function and policy entropy To understand the current state of the reinforcement learning environment s t Perform value assessment; based on advantage function To calculate the gradient of the policy network II, and then update the policy network II according to the stochastic gradient descent algorithm, and then correct the value judgment and strategy formulation process of the policy network II on the system state; the iteration of the value network II needs to correct the value of the system state based on the Bellman iteration equation, and calculate the parameter gradients before and after the correction, and update the value network II according to the stochastic gradient descent algorithm; To define the network parameters of strategy network II, is in a given state s t Next, action The log probability of the corresponding strategy; is the advantage function; S4.22: In the reinforcement learning working condition parameter search model of the hydropower unit bolt, the control variables in the finite element simulation environment are analyzed, and by mapping these control variables to actions in reinforcement learning, the action space of reinforcement learning is determined, so that the strategy network II can adjust these variables for optimization under different working conditions; S4.23: Determine the state space in the reinforcement learning environment. Considering the state changes caused by the actions, the strategy characteristics can be defined as the stress distribution of the bolts. The inherent characteristics that do not change in the environment are the dimensions of the constructed three-dimensional hydropower unit model and the fluid-structure coupling calculation form. S4.24: Determine the environmental constraints and design a reinforcement learning reward function based on the objective function. The reward function guides the strategy network II to learn and select the optimal operating parameters by providing positive feedback for actions close to the optimal parameters, ensuring that the reward function reflects the optimization direction of the objective function.

[0060] S4.3: Obtaining the optimal operating parameters: The operating parameter search model uses the maximum entropy reinforcement learning algorithm in the reinforcement learning environment to complete the online search and optimization of the operating parameters, iterates the reinforcement learning optimal strategy based on the maximum entropy reinforcement learning, and searches for the optimal operating parameters based on the optimal strategy; including: Based on the reinforcement learning environment established in S4.2, the reinforcement learning environment is solved by the maximum entropy reinforcement learning algorithm; By cumulative reward and policy entropy The two items constitute the evaluation function of the value function, and the optimal strategy is calculated through the evaluation function of the value function. The optimal strategy is imported into the strategy network II and output to obtain the optimal working condition parameters; The evaluation function of the value function is: in, For state actions The probability distribution of the group is The expected reward value when s t and a t are the state and action at time t respectively; For the status s t Next action a t The reward value obtained; Status s t The entropy value of strategy π when α The temperature parameter determines its ability to explore different strategies; π For control strategies; π * is the optimal control strategy.

[0061] S4.4. Verify the optimal operating parameters obtained through the search, including: The finite element simulation is performed using the optimal operating parameters obtained through the search. The differences between the physical field simulation data in the finite element simulation results and the physical field simulation data output by the operating parameter search model are compared to verify the feasibility and effectiveness of the search algorithm. A sampling algorithm is used to generate multiple new parameter points near the optimal search parameters. The optimal working condition parameters obtained by the search and the sampling points near them and the corresponding finite element simulation data are expanded into a new training set for online iterative update of the reinforcement learning stress cloud map inversion model. The optimal operating parameters are verified on the actual equipment, the actual working performance is observed and measured, and the finite element simulation model, stress cloud map inversion model and operating parameter search model are integrated. The overall model is a digital twin model. The operating parameters before and after optimization are applied to the actual equipment and the finite element simulation model respectively, and the comparison results before and after optimization are observed to confirm the effectiveness of the optimal operating parameters obtained in the digital twin model.

[0062] S4.5. Display the parameter search results, including: The parameter search and iteration process is visualized in the form of icons, including the time and accuracy of each iteration. When the optimal parameters obtained by the search have an accuracy of more than 90% coincidence with the physical field under the actual working conditions, the iterative training stops and the inversion optimization algorithm of the service stress cloud map of the bolts of the hydropower unit is obtained. The remaining steps are the same as those in Example 1.

[0063] The optimal operating condition parameter search algorithm in this embodiment: for the prediction model of physical field simulation data under different working conditions, a deep reinforcement learning framework based on intelligent technology is used to develop an automatic search algorithm for optimal operating condition parameters. A reasonable objective function for updating the optimal strategy is constructed to achieve reasonable exploration of the optimal parameters of power generation equipment components and enhance the algorithm's ability to explore the prediction model.

[0064] Example 6 Compared with Example 1, the present embodiment is different in that, in the present embodiment, the service stress cloud map inversion optimization algorithm also includes verification and error analysis of the physical field simulation results in step S1.2, and the verification and error analysis of the physical field simulation results include: verifying the finite element simulation model in step S1.2, and comparing the error between the physical field simulation data predicted by the finite element simulation model and the actual measurement value; when the error is greater than 10%, temporarily retaining the group of finite element simulation data, and using the finite element simulation data that meets the error condition to establish an inversion optimization algorithm model, and optimizing the working condition parameter load in the current finite element simulation through the inversion optimization algorithm, and updating the finite element simulation until the error is less than 10%, and the error condition is met; adding the group of finite element simulation data to the training set of the inversion optimization algorithm model; The field measurement data includes stress in the actual environment. The stress measurement in the actual environment adopts a hollow cylindrical force sensor, which includes a built-in elastic element, a resistance strain gauge, a temperature compensation circuit and a conversion circuit. The sensor measures the actual average stress on the bolt. A single sensor is installed between the top cover bolt and the inner top cover. The field test condition is the entire start-stop process of the unit under high water head from shutdown to startup, changing the opening and finally shutting down. The bolt stress distribution measured under actual conditions is compared and analyzed with the finite element simulation results. The actual measured value is the average stress of the actual bolt measured by the hollow cylindrical force sensor; the field test condition is the entire start-stop process of the unit under high water head from shutdown to startup, changing the opening degree to the final shutdown.

[0065] It is to be understood that the present invention is described by some embodiments, and it is known to those skilled in the art that various changes or equivalent substitutions may be made to these features and embodiments without departing from the spirit and scope of the present invention. In addition, under the teachings of the present invention, these features and embodiments may be modified to adapt to specific circumstances and materials without departing from the spirit and scope of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed herein, and all embodiments falling within the scope of the claims of this application are within the scope of protection of the present invention.

Claims

1. An inversion optimization algorithm for the service stress cloud diagram of bolts of hydropower units, characterized by: The steps include: S1: Finite element simulation model construction, including the following steps: S1.1: Determine the basic structure of the components: decompose the hydropower unit into the fluid calculation domain and the structural calculation domain, and use 3D modeling software to perform 3D modeling on the fluid calculation domain of the hydropower unit and the top cover bolt structure calculation domain; S1.2: Establishment and calculation of finite element simulation model: After meshing the three-dimensional models of the fluid calculation domain and the structure calculation domain in step S1.1, a finite element simulation model is established. Contact constraints of various components and boundary conditions of the fluid calculation domain are set in the finite element simulation model. Multiple groups of different working condition loads are input into the finite element simulation model to establish a finite element simulation model of the hydropower unit. Fluid-solid coupling simulation calculations are performed to obtain physical field simulation data under different working condition loads. S2: Perform data cleaning on the physical field simulation data and field measurement data obtained in step S1.2 to obtain a physical field data set; S3: using the physical field data set obtained in step S2, establishing a reinforcement learning stress cloud map inversion model for the bolts of the hydropower unit, and iteratively updating the reinforcement learning stress cloud map inversion model for the bolts of the hydropower unit, and verifying the effect of the finite element stress cloud map inversion by comparing the running results of the actual working conditions with the finite element simulation and the inversion results; S4: optimal parameter search, including using the hydropower unit bolt reinforcement learning stress cloud map inversion model obtained in step S3 to establish a hydropower unit bolt reinforcement learning operating parameter search model, perform optimal operating parameter search, and verify and visualize the search results to obtain the optimal operating parameters of the hydropower unit bolts.

2. According to claim 1, a hydropower unit bolt service stress cloud map inversion optimization algorithm is characterized by: The service stress cloud map inversion optimization algorithm also includes verification and error analysis of the physical field simulation results in step S1.2, and the verification and error analysis of the physical field simulation results include: verifying the finite element simulation model in step S1.2, and comparing the error between the physical field simulation data predicted by the finite element simulation model and the actual measurement value; when the error is greater than 10%, temporarily retaining the group of finite element simulation data, and using the finite element simulation data that meets the error condition to establish an inversion optimization algorithm model, and optimizing the operating parameter loads in the current finite element simulation through the inversion optimization algorithm, and updating the finite element simulation until the error is less than 10%, and the error condition is met; adding the group of finite element simulation data to the training set of the inversion optimization algorithm model.

3. According to claim 1, a hydropower unit bolt service stress cloud map inversion optimization algorithm is characterized by: In step S1.1, the fluid calculation domain includes the volute, the fixed guide vane and the movable guide vane area, the runner area and the tailwater pipe; the structure calculation domain includes the outer top cover flange, the inner top cover and the bolts.

4. According to claim 1, a hydropower unit bolt service stress cloud map inversion optimization algorithm is characterized by: In the step S1.2, the meshing is to mesh the entire flow channel of the hydropower unit using unstructured tetrahedral meshes, and to encrypt the meshes of the top cover flow surface area, the impeller and the guide vane.

5. According to claim 1, a hydropower unit bolt service stress cloud map inversion optimization algorithm is characterized by: In step S1.2, the boundary conditions of the fluid calculation domain include the volute inlet, wall, tailwater pipe outlet and runner area, and the contact constraints are the horizontal contact between the inner top cover flange and the outer top cover flange, the horizontal contact between the bolt head and the inner top cover flange, and the contact between the bottom thread of the bolt and the threaded hole of the inner top cover.

6. According to claim 1, a hydropower unit bolt service stress cloud map inversion optimization algorithm is characterized by: In step S1.2, the different working condition loads include: multiple groups of working condition loads with different load positions, different load intensities and different load application methods.

7. According to claim 1, a hydropower unit bolt service stress cloud map inversion optimization algorithm is characterized by: In step S1.2, the fluid-solid coupling simulation calculation includes applying the gravity load of the turbine according to the gravitational acceleration, applying the preload of the bolts and locking the preload of the bolts, applying water pressure to the top cover and the water-guide bearing seat, that is, the water pressure load calculated by the external fluid is transmitted to the top cover and the water-guide bearing seat through the fluid-solid coupling surface, and according to the working load conditions and boundary conditions, the fluid-solid coupling equation is solved through fluid-solid coupling simulation to obtain physical field simulation data under different working conditions.

8. According to claim 1, a hydropower unit bolt service stress cloud map inversion optimization algorithm is characterized by: In the step S1.2, the model establishment also includes mesh independence verification. The mesh independence verification increases the number of meshes on the basis of the existing finite element simulation model. When the change of the physical field does not exceed 5% with the increase of the number of meshes, the verification is completed. When the change of the physical field exceeds 5% with the increase of the number of meshes, the number of meshes continues to be increased for independence verification until the verification is completed.

9. According to claim 1, a hydroelectric generator bolt service stress cloud map inversion optimization algorithm is characterized by: The data cleaning in step S2 includes the following steps: S2.1: Handling missing values: Detect missing values ​​in physical field simulation data and field measurement data, and handle missing values ​​by filling missing values ​​or deleting rows or columns containing missing values; S2.2: Handling outliers: Detect outliers in physical field simulation data and field measurement data, and handle outliers using methods including substitution, deletion, normalization, and verification of data; S2.3: Convert data types: Determine the data type required for each field based on the analysis objectives or modeling requirements, and adjust the data types in the physical field simulation data and field measurement data accordingly; S2.4: Standardized data: Standardize the physical field simulation data and field measurement data so that each set of data has the same dimension and distribution, and obtain the physical field data set.

10. According to claim 1, a hydroelectric generator bolt service stress cloud map inversion optimization algorithm is characterized by: The step S3 of establishing a reinforcement learning stress cloud map inversion model for the bolts of the hydropower generator unit comprises: constructing a Markov decision chain model with the physical field data set obtained in step S2, establishing a reinforcement learning model, applying the reinforcement learning model to solve the Markov decision chain model, coupling the reinforcement learning model with the Markov decision chain model to obtain a reinforcement learning stress cloud map inversion model for the bolts of the hydropower generator unit; using the finite element input parameters of the physical field data set as the input of the reinforcement learning stress cloud map inversion model for the bolts of the hydropower generator unit, and inverting a stress cloud map of the bolts of the hydropower generator unit, wherein the stress cloud map of the bolts of the hydropower generator unit is the stress value of each grid point of the finite element simulation model after grid division, wherein the stress value of the boundary of the stress cloud map is the boundary stress value; the Markov decision chain model comprises a state s t 、Action a t , Markov decision chain model feedback r ( s t ), where state s t The stress inversion result of the initialization decision of the Markov decision chain model is t The stress inversion result after s t+1 , action a t is the step size for changing the stress magnitude each time, used to change the state s of each iterative cycle t ; Through the Navier-Stokes equation, using the solid wall boundary condition, the stress boundary value of the theoretical physical field is calculated through the fluid physical parameters in the hydropower unit; Markov decision chain model feedback r ( s t ) is the current state s t The mean square error of the stress value and the stress value of the physical field data set, plus the difference between the calculated theoretical stress cloud map boundary value and the boundary stress value inverted by the stress cloud map inversion model, is used to measure the current predicted stress, that is, the state s t The difference between the real data set; the iterative formula of the Markov decision chain model state in each cycle is as follows: 。 11. The inversion optimization algorithm for the service stress cloud diagram of bolts of a hydropower unit according to claim 10 is characterized by: In step S3, the reinforcement learning model includes a strategy network I and a value network I.

12. The inversion optimization algorithm for the service stress cloud diagram of the bolts of a hydropower unit according to claim 11 is characterized in that: The strategy network I is responsible for generating a decision, wherein the decision is to determine the direction of change of the finite element input parameters; The value network I is responsible for evaluating the quality of the decision output by the strategy network I. The evaluation is based on the calculation error between the predicted value of the finite element simulation result and the actual value of the simulation result in the data set. The smaller the error, the better the decision.

13. The inversion optimization algorithm for the service stress cloud diagram of bolts of a hydropower unit according to claim 11 is characterized in that: The strategy network I is composed of a radial basis function neural network, including an input layer, two hidden layers and an output layer, which generates the corresponding control strategy by gradually accumulating system experience.

14. The inversion optimization algorithm for the service stress cloud diagram of bolts of a hydropower unit according to claim 11 is characterized in that: The value network I part includes an input layer, a hidden layer and an output layer. It evaluates the performance of the current finite element parameter decision by comparing the differences between the physical field simulation data of the finite element simulation results and the actual test data. It generates rewards or penalties as feedback values ​​for adaptive learning based on the changing trend of the error, and introduces the long-term cost function as an evaluation criterion.

15. The inversion optimization algorithm for the service stress cloud diagram of bolts of a hydropower unit according to claim 11 is characterized in that: The iterative update in step S3 includes: updating the strategy network I through the stochastic gradient descent algorithm, and correcting the state value of the Markov decision chain model through the Bellman iterative equation, and performing simulation application and verification to achieve inversion prediction of the stress cloud map.

16. The inversion optimization algorithm for the service stress cloud diagram of bolts of a hydropower unit according to claim 15 is characterized in that: The updating of the policy network I by the stochastic gradient descent algorithm includes using the gradient method in the embedded physical knowledge policy network I to update the policy network I by stochastic gradient descent, and ensuring through the evaluation of the value network I that the finite element parameters generated by the policy network I can maximize the Markov decision chain model evaluation index.

17. The inversion optimization algorithm for the service stress cloud diagram of bolts of a hydropower unit according to claim 16 is characterized in that: The stochastic gradient descent algorithm is used to train the objective function of the policy network I, measure the performance of the current policy, and guide the update of network parameters; The stochastic gradient descent algorithm consists of the sum of two parts: the first part is the loss function composed of the difference between the finite element simulation results corresponding to the sensor and the actual measurement value, and the second part is the difference between the field boundary conditions predicted by the strategy network I and the ideal boundary conditions; the predicted boundary conditions are the velocity and pressure on the fluid boundary in the simulation results, and the ideal boundary conditions are the velocity and pressure on the fluid boundary that conform to the previous fluid-solid coupling basic equations.

18. The inversion optimization algorithm for the service stress cloud diagram of bolts of a hydropower unit according to claim 17 is characterized in that: The updating of the policy network I and the value network I by the stochastic gradient descent algorithm includes: first, the target Markov decision chain model randomly generates the initial state of the Markov decision chain model or the current state after iterative update s t , based on the strategy π generated by the strategy network I, according to the current state of the Markov decision chain model s t To decide the action to be performed a t , the action a t Interact with the Markov decision chain model and transfer to a new state according to the iterative formula of the Markov decision chain model state , get the Markov decision chain model feedback ; Value Network I estimates the value of an action through feedback from a Markov decision chain model , The expression is: Among them, S is the state of the Markov decision chain model, G t Feedback for Markov decision chain model The mathematical expectation of Each state-action group (s t , a t )'s advantage function A π : The gradient of the strategy network I is calculated based on the advantage function, and the strategy network I is updated according to the stochastic gradient descent algorithm, thereby correcting the value judgment and strategy formulation process of the strategy network I on the state of the Markov decision chain model; the iteration of the value network I needs to correct the state value of the Markov decision chain model based on the Bellman iteration equation, and calculate the parameter gradients before and after the correction, and update the value network I according to the stochastic gradient descent algorithm; Among them, the gradient calculation of policy network I is as follows: To define the network parameters of the strategy network I, In a given state s t Next, action a t The log probability of the corresponding strategy; is the advantage function; The iterative formula of the value function is as follows: in, is the attenuation coefficient, Feedback for the Markov decision chain model.

19. The inversion optimization algorithm for the service stress cloud diagram of bolts of a hydropower unit according to claim 15 is characterized by: The simulation application and verification include: the reinforcement learning algorithm interacts with the Markov decision chain model in each iteration cycle, and the strategy network I is based on the state of the Markov decision chain model s t Generate decision action a t , the decision action a t To make a decision on the direction of change of the finite element input parameters; the Markov decision chain model is based on the decision action a t Feedback is generated, and the Markov decision chain model state changes from s t Transform to s t+1 ; The value network I is responsible for evaluating the quality of the decision output by the strategy network I. The evaluation is based on the feedback of the Markov decision chain model. , that is, the difference between the predicted value of the finite element simulation result and the calculated true value of the simulation result in the data set plus the difference between the theoretical stress cloud map boundary value calculated based on the physical formula and the inverted stress boundary value. The sum of the two differences is the error. The smaller the error, the better the decision. When the error is greater than 5%, the stress cloud map inversion optimization algorithm continues to iterate; when the error is less than 5%, the stress cloud map inversion optimization algorithm iteration is completed.

20. According to claim 1, a hydroelectric generator bolt service stress cloud map inversion optimization algorithm is characterized by: The step S4 specifically includes the following steps: S4.1: Using the reinforcement learning stress cloud map inversion model of the hydropower unit bolts obtained in step S3, a hydropower unit operating condition parameter search model is constructed; S4.2: constructing a reinforcement learning environment for the operating condition design parameters of the operating condition parameter search model in the hydropower unit operating condition parameter search model obtained in step S4.1 by using the maximum entropy reinforcement learning algorithm; S4.3: Obtaining the optimal operating parameters: The operating parameter search model uses the maximum entropy reinforcement learning algorithm in the reinforcement learning environment to complete the online search and optimization of the operating parameters, iterates the reinforcement learning optimal strategy based on the maximum entropy reinforcement learning, and searches for the optimal operating parameters based on the optimal strategy; S4.4: Verify the optimal operating parameters searched in step S4.3; S4.5: Display the parameter search results, including: the parameter search and iteration process are visualized in the form of icons, including the time and accuracy of each iteration. When the optimal parameters obtained by the search have an accuracy of more than 90% coincidence with the physical field under the actual working conditions, the iterative training stops, and an inversion optimization algorithm for the service stress cloud map of the hydropower unit bolts is obtained.

21. The inversion optimization algorithm for the service stress cloud diagram of bolts of a hydropower unit according to claim 20, characterized in that: The step S4.1 includes: obtaining the operating environment of the hydropower unit under different working conditions through historical data or measurements, selecting a suitable boundary type, the boundary type includes fixed boundary conditions and symmetric boundary conditions, and determining the value of each boundary condition according to the measured data or standard specifications of the working conditions; based on performance indicators or energy efficiency, selecting the objective function in the optimization task, and mapping the stress indicators in the simulation results to the objective function to ensure that the objective function can truly reflect the stress state of the equipment components and is closely related to the actual operating conditions; based on the actual situation, setting the range of each action variable in the environment and the constraints of the water pressure distribution, and constructing a hydropower unit operating condition parameter search model.

22. The inversion optimization algorithm for the service stress cloud diagram of bolts of a hydropower unit according to claim 20 is characterized in that: The step S4.2 includes: S4.21: Construct a maximum entropy reinforcement learning algorithm and introduce an adjustable entropy term into the maximum entropy reinforcement learning algorithm ; The maximum entropy reinforcement learning algorithm includes value network II and policy network II. In each iteration cycle, policy network II is based on the state of the reinforcement learning environment. s t Take action on it a t , and obtain the reward function from the reinforcement learning environment, the state s t Transition to a new state s t+1 ; Value Network II evaluates the sum of reward function and policy entropy To understand the current state of the reinforcement learning environment s t Perform value assessment; based on advantage function To calculate the gradient of the policy network II, and then update the policy network II according to the stochastic gradient descent algorithm, and then correct the value judgment and strategy formulation process of the policy network II on the system state; the iteration of the value network II needs to correct the value of the system state based on the Bellman iteration equation, and calculate the parameter gradients before and after the correction, and update the value network II according to the stochastic gradient descent algorithm; To define the network parameters of strategy network II, is in a given state s t Next, action The log probability of the corresponding strategy; is the advantage function; S4.22: In the reinforcement learning working condition parameter search model of the hydropower unit bolt, the control variables in the finite element simulation environment are analyzed, and by mapping these control variables to actions in reinforcement learning, the action space of reinforcement learning is determined, so that the strategy network II can adjust these variables for optimization under different working conditions; S4.23: Determine the state space in the reinforcement learning environment. Considering the state changes caused by the actions, the strategy characteristics can be defined as the stress distribution of the bolts. The inherent characteristics that do not change in the environment are the dimensions of the constructed three-dimensional hydropower unit model and the fluid-structure coupling calculation form. S4.24: Determine the environmental constraints and design a reinforcement learning reward function based on the objective function. The reward function guides the strategy network II to learn and select the optimal operating parameters by providing positive feedback for actions close to the optimal parameters, ensuring that the reward function reflects the optimization direction of the objective function.

23. The inversion optimization algorithm for the service stress cloud diagram of bolts of a hydropower unit according to claim 20 is characterized in that: The step S4.3 includes: based on the reinforcement learning environment established in S4.2, solving the reinforcement learning environment by using a maximum entropy reinforcement learning algorithm; By cumulative reward and policy entropy The two items constitute the evaluation function of the value function, and the optimal strategy is calculated through the evaluation function of the value function. The optimal strategy is imported into the strategy network II and output to obtain the optimal working condition parameters; The evaluation function of the value function is: in, For state actions The probability distribution of the group is The expected reward value when s t and a t are the state and action at time t respectively; For the status s t Next action a t The reward value obtained; Status s t The entropy value of strategy π when α The temperature parameter determines its ability to explore different strategies; π For control strategies; π * is the optimal control strategy.

24. The inversion optimization algorithm for the service stress cloud diagram of bolts of a hydropower unit according to claim 20 is characterized in that: The step S4.4 includes: The finite element simulation is performed using the optimal operating parameters obtained through the search. The differences between the physical field simulation data in the finite element simulation results and the physical field simulation data output by the operating parameter search model are compared to verify the feasibility and effectiveness of the search algorithm. A sampling algorithm is used to generate multiple new parameter points near the optimal search parameters. The optimal working condition parameters obtained by the search and the sampling points near them and the corresponding finite element simulation data are expanded into a new training set for online iterative update of the reinforcement learning stress cloud map inversion model. The optimal operating parameters are verified on the actual equipment, the actual working performance is observed and measured, and the finite element simulation model, stress cloud map inversion model and operating parameter search model are integrated. The overall model is a digital twin model. The operating parameters before and after optimization are applied to the actual equipment and the finite element simulation model respectively, and the comparison results before and after optimization are observed to confirm the effectiveness of the optimal operating parameters obtained in the digital twin model.

Images