A method and device for optimizing a turbine blade cooling structure based on reinforcement learning

Abstract

The application provides a turbine blade cooling structure optimization method and device based on reinforcement learning, the method comprises the following steps: in response to inputting a preset design condition, obtaining an initial cooling structure of a turbine blade based on the design condition; in response to preset intelligent agent initialization being completed, training the intelligent agent based on a preset reinforcement learning algorithm in a preset reinforcement learning environment framework to obtain a trained intelligent agent; optimizing the cooling parameters of the cooling structure based on the trained intelligent agent to obtain optimized cooling parameters; and determining an optimized cooling structure based on the optimized cooling parameters. The application can be applied to more complex or complete cooling structures, realizes automatic optimization and determination of the cooling parameters of the cooling structure, and reduces the experience dependence and time cost of the designer in the cooling parameter design process.

Description

Technical Field

[0001] This application relates to the field of deep learning technology, and more specifically, to a method and apparatus for optimizing the cooling structure of turbine blades based on reinforcement learning. Background Technology

[0002] In the field of aero-engines, turbine blade cooling structures are critical, and their design is extremely complex and time-consuming, heavily reliant on the expertise of the designers. Furthermore, to design a set of parameters that meet the requirements, designers often need to adjust these parameters multiple times, repeating complex steps such as blade modeling, mesh generation, and simulation calculations. They then evaluate the suitability of these design parameters based on the obtained temperature field values. Typically, designers need to spend ten to fifteen days obtaining a set of parameters that meet the design conditions. However, these parameters are interconnected, and traditional design methods require a high level of specialized expertise to optimize the cooling structure.

[0003] Currently, the optimization of cooling structures is often achieved by using heuristic algorithms to fine-tune the cooling structure of turbine blades.

[0004] However, this method is only applicable to simple or localized cooling structures. In the implementation process, designers not only need to iteratively modify the parameters of the basic structure and cooling structure, but also need to continuously perform simulation calculations to evaluate the cooling effect. This consumes a lot of time to search for a better solution and is highly dependent on the designer's experience. Summary of the Invention

[0005] In view of this, the purpose of this application is to provide a method and device for optimizing the cooling structure of turbine blades based on reinforcement learning. By introducing reinforcement learning algorithms to train the agent during the design process of the turbine blade cooling structure, and optimizing the cooling parameters of the cooling structure based on the trained agent, more complex or complete cooling structures can be applied. This realizes the automatic optimization and determination of the cooling parameters of the cooling structure, reducing the dependence on the designer's experience and time cost in the design process of cooling parameters.

[0006] In a first aspect, embodiments of this application provide a method for optimizing the cooling structure of turbine blades based on reinforcement learning, the method comprising:

[0007] In response to preset design conditions, the initial cooling structure of the turbine blade is obtained based on the design conditions; the design conditions include blade model configuration file and operating conditions.

[0008] In response to the completion of the preset agent initialization, the agent is trained based on the preset reinforcement learning algorithm within the preset reinforcement learning environment framework to obtain the trained agent; wherein, the reinforcement learning algorithm corresponds to the policy network.

[0009] The cooling parameters of the cooling structure are optimized based on the trained agent to obtain the optimized cooling parameters, and the optimized cooling structure is determined based on the optimized cooling parameters.

[0010] In one possible implementation, the step of obtaining the initial cooling structure of the turbine blades based on preset design conditions in response to input includes:

[0011] In response to the input preset design conditions, the external flow channel simulation calculation is performed based on the design conditions to generate the initial basic structure of the turbine blades;

[0012] Based on the initial basic structure of the turbine blade, a corresponding initial cooling structure is obtained; wherein, the basic structure is used to determine the basic parameters of the cooling structure of the turbine blade; and the cooling structure is used to determine the cooling parameters of the cooling structure of the turbine blade.

[0013] In one possible implementation, before training the preset agent based on a preset policy network within a preset reinforcement learning environment framework, the agent is initialized through the following steps:

[0014] Within a preset reinforcement learning environment framework, the agent resets the environment and reads the corresponding hole parameter information and operating conditions based on the blade model configuration file, the operating conditions, and the preset hole parameter file; wherein, the hole parameter file includes the cooling parameters of the cooling structure;

[0015] The hole parameter information and operating conditions are processed based on a preset full three-dimensional evaluation algorithm to obtain the initial surface temperature of the first turbine blade after environmental reset.

[0016] In one possible implementation, training the agent to obtain a trained agent includes:

[0017] The agent inputs the surface temperature of the first turbine blade and the cooling parameters of the cooling structure read from the orifice parameter file into the policy network to generate a corresponding target action, and then submits the target action to the reinforcement learning environment framework for implementation, so that the reinforcement learning environment framework can tune the cooling parameters based on the target action;

[0018] In response to the completion of the cooling parameter tuning, the tuned cooling parameters and the operating conditions are calculated based on the full three-dimensional evaluation algorithm to obtain the tuned second turbine blade surface temperature, thereby completing one round of training for the agent.

[0019] If the termination signal generated by the preset reward function is false, the agent will continue to be trained in a new round based on the surface temperature of the second turbine blade and the adjusted cooling parameters, so as to finally obtain the trained agent.

[0020] In one possible implementation, the method further includes:

[0021] If the termination signal generated by the preset reward function is false, the agent updates the policy network based on the tuned cooling parameters and the surface temperature of the second turbine blade.

[0022] If the updated policy network fails to converge, training ends.

[0023] In one possible implementation, the step of optimizing the cooling parameters of the cooling structure based on the trained agent to obtain the optimized cooling parameters includes:

[0024] Under the reinforcement learning algorithm, based on the trained agent and the updated policy network, as well as the basic parameters corresponding to the cooling parameters and the infrastructure, the initial surface temperature of the third turbine blade is obtained in the reinforcement learning environment framework.

[0025] Based on the surface temperature of the third turbine blade and the cooling parameters, a corresponding second target action is generated, and the second target action is executed by the reinforcement learning environment framework to optimize the cooling parameters of the cooling structure, thereby obtaining the surface temperature of the fourth turbine blade and the optimized cooling parameters.

[0026] The changes in the surface temperature of the turbine blade are evaluated based on the surface temperature of the fourth turbine blade, and the optimized cooling parameters are determined based on the evaluation results.

[0027] In one possible implementation, the evaluation of the change in the surface temperature of the turbine blade based on the surface temperature of the fourth turbine blade, and the determination of the optimized cooling parameters based on the evaluation results, includes:

[0028] In response to determining, based on the surface temperature of the fourth turbine blade and the initial surface temperature of the third turbine blade, that the surface temperature of the turbine blade rises or the cooling parameters exceed the preset parameter boundaries during the implementation of the second target action, the agent accumulates the second target action to obtain total modification information for the cooling structure.

[0029] Based on the total modification information, the cooling parameters are modified to obtain the final optimized cooling parameters.

[0030] Secondly, embodiments of this application also provide a turbine blade cooling structure optimization device based on reinforcement learning, the turbine blade cooling structure optimization device comprising:

[0031] The acquisition module is used to obtain the initial cooling structure of the turbine blade based on the input preset design conditions; the design conditions include the blade model configuration file and operating conditions.

[0032] The training module is used to train the agent based on a preset reinforcement learning algorithm within a preset reinforcement learning environment framework after the preset agent initialization is completed, thereby obtaining the trained agent; wherein the reinforcement learning algorithm corresponds to a policy network.

[0033] The optimization module is used to optimize the cooling parameters of the cooling structure based on the trained agent, obtain the optimized cooling parameters, and determine the optimized cooling structure based on the optimized cooling parameters.

[0034] In one possible implementation, the acquisition module is specifically used for:

[0035] In response to the input preset design conditions, the external flow channel simulation calculation is performed based on the design conditions to generate the initial basic structure of the turbine blades;

[0036] Based on the initial basic structure of the turbine blade, a corresponding initial cooling structure is obtained; wherein, the basic structure is used to determine the basic parameters of the cooling structure of the turbine blade; and the cooling structure is used to determine the cooling parameters of the cooling structure of the turbine blade.

[0037] In one possible implementation, the agent initialization is accomplished through the following steps:

[0038] Before training the preset agent based on a preset policy network within the preset reinforcement learning environment framework, the agent resets the environment and reads the corresponding hole parameter information and operating conditions based on the blade model configuration file, the operating conditions, and the preset hole parameter file within the preset reinforcement learning environment framework; wherein, the hole parameter file includes the cooling parameters of the cooling structure.

[0039] The hole parameter information and operating conditions are processed based on a preset full three-dimensional evaluation algorithm to obtain the initial surface temperature of the first turbine blade after environmental reset.

[0040] In one possible implementation, the training module is specifically used for:

[0041] The agent inputs the surface temperature of the first turbine blade and the cooling parameters of the cooling structure read from the orifice parameter file into the policy network to generate a corresponding target action, and then submits the target action to the reinforcement learning environment framework for implementation, so that the reinforcement learning environment framework can tune the cooling parameters based on the target action;

[0042] In response to the completion of the cooling parameter tuning, the tuned cooling parameters and the operating conditions are calculated based on the full three-dimensional evaluation algorithm to obtain the tuned second turbine blade surface temperature, thereby completing one round of training for the agent.

[0043] If the termination signal generated by the preset reward function is false, the agent will continue to be trained in a new round based on the surface temperature of the second turbine blade and the adjusted cooling parameters, so as to finally obtain the trained agent.

[0044] In one possible implementation, the device further includes:

[0045] An update module is used to update the policy network based on the tuned cooling parameters and the surface temperature of the second turbine blade in response to a false termination signal generated by a preset reward function.

[0046] The judgment module is used to terminate training if the updated policy network fails to converge.

[0047] In one possible implementation, the tuning module is specifically used for:

[0048] Under the reinforcement learning algorithm, based on the trained agent and the updated policy network, as well as the basic parameters corresponding to the cooling parameters and the infrastructure, the initial surface temperature of the third turbine blade is obtained in the reinforcement learning environment framework.

[0049] Based on the surface temperature of the third turbine blade and the cooling parameters, a corresponding second target action is generated, and the second target action is executed by the reinforcement learning environment framework to optimize the cooling parameters of the cooling structure, thereby obtaining the surface temperature of the fourth turbine blade and the optimized cooling parameters.

[0050] The changes in the surface temperature of the turbine blade are evaluated based on the surface temperature of the fourth turbine blade, and the optimized cooling parameters are determined based on the evaluation results.

[0051] In one possible implementation, the tuning module is specifically used for:

[0052] In response to determining, based on the surface temperature of the fourth turbine blade and the initial surface temperature of the third turbine blade, that the surface temperature of the turbine blade rises or the cooling parameters exceed the preset parameter boundaries during the implementation of the second target action, the agent accumulates the second target action to obtain total modification information for the cooling structure.

[0053] Based on the total modification information, the cooling parameters are modified to obtain the final optimized cooling parameters.

[0054] Thirdly, embodiments of this application provide an electronic device, including: a processor, a storage medium, and a bus. The storage medium stores machine-readable instructions executable by the processor. When the electronic device is running, the processor communicates with the storage medium via the bus, and the processor executes the machine-readable instructions to perform the steps of the reinforcement learning-based turbine blade cooling structure optimization method as described in any of the first aspects.

[0055] Fourthly, embodiments of this application provide a computer-readable storage medium storing a computer program, which, when executed by a processor, performs the steps of the reinforcement learning-based turbine blade cooling structure optimization method described in any one of the first aspects.

[0056] This application provides a method and apparatus for optimizing turbine blade cooling structure based on reinforcement learning. Responding to preset design conditions, an initial cooling structure for the turbine blade is obtained based on these conditions. After the preset agent initialization is completed, the agent is trained using a preset reinforcement learning algorithm within a preset reinforcement learning environment framework, resulting in a trained agent. The cooling parameters of the cooling structure are then optimized based on the trained agent, yielding optimized cooling parameters. Finally, the optimized cooling structure is determined based on these optimized cooling parameters. This application, by introducing a reinforcement learning algorithm to train the agent during the turbine blade cooling structure design process and optimizing the cooling parameters based on the trained agent, allows for the application of more complex or complete cooling structures. It achieves automatic optimization and determination of cooling parameters, reducing the reliance on designer experience and time costs during the cooling parameter design process.

[0057] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0058] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0059] Figure 1 This is a flowchart of a reinforcement learning-based turbine blade cooling structure optimization method according to an embodiment of this application;

[0060] Figure 2 This is a schematic diagram of the cooling structure optimization process;

[0061] Figure 3 This is a schematic diagram of the iterative interaction process between the intelligent agent and the environment;

[0062] Figure 4 This is a flowchart of a reinforcement learning-based turbine blade cooling structure optimization method according to another embodiment of this application;

[0063] Figure 5 A flowchart of a reinforcement learning-based turbine blade cooling structure optimization method according to another embodiment of this application;

[0064] Figure 6 This is a schematic diagram of the agent training process;

[0065] Figure 7 This is a flowchart of a reinforcement learning-based turbine blade cooling structure optimization method according to another embodiment of this application;

[0066] Figure 8 This is a schematic diagram of a turbine blade cooling structure optimization device based on reinforcement learning according to an embodiment of this application.

[0067] Figure 9 This is a schematic diagram of the structure of an electronic device according to an embodiment of this application. Detailed Implementation

[0068] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. It should be understood that the accompanying drawings in this application are for illustrative and descriptive purposes only and are not intended to limit the scope of protection of this application. Furthermore, it should be understood that the schematic drawings are not drawn to scale. The flowcharts used in this application illustrate operations implemented according to some embodiments of this application. It should be understood that the operations in the flowcharts may not be implemented in sequence, and steps without logical contextual relationships may be reversed or implemented simultaneously. In addition, those skilled in the art, guided by the content of this application, may add one or more other operations to the flowcharts, or remove one or more operations from the flowcharts.

[0069] Furthermore, the described embodiments are merely some, not all, of the embodiments of this application. The components of the embodiments of this application described and illustrated herein can typically be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of the application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0070] It should be noted that the term "comprising" will be used in the embodiments of this application to indicate the presence of the features declared thereafter, but does not exclude the addition of other features.

[0071] In the field of aero-engines, turbine blade cooling structures are critical, and their design is extremely complex and time-consuming, heavily reliant on the expertise of the designers. Furthermore, to obtain a set of parameters that meet the requirements, designers often need to adjust these parameters multiple times, repeating complex steps such as blade modeling, mesh generation, and simulation calculations. They then evaluate the suitability of these design parameters based on the obtained temperature field values. Typically, designers need to spend over ten days to obtain a set of parameters that meet the design conditions. However, these parameters are interconnected, and traditional design methods require a high level of specialized expertise to optimize the cooling structure.

[0072] Currently, the optimization of cooling structures is often achieved by using heuristic algorithms to fine-tune the cooling structure of turbine blades.

[0073] However, this method is only applicable to simple or localized cooling structures. In the implementation process, designers not only need to iteratively modify the parameters of the basic structure and cooling structure, but also need to continuously perform simulation calculations to evaluate the cooling effect. This consumes a lot of time to search for a better solution and is highly dependent on the designer's experience.

[0074] To address this issue, this application provides a method and apparatus for optimizing turbine blade cooling structures based on reinforcement learning. By introducing reinforcement learning algorithms to train an agent during the turbine blade cooling structure design process, and optimizing the cooling parameters of the cooling structure based on the trained agent, more complex or complete cooling structures can be applied. This achieves automatic optimization and determination of the cooling parameters of the cooling structure, reducing the reliance on the designer's experience and time costs during the cooling parameter design process.

[0075] Figure 1 This is a flowchart of a reinforcement learning-based turbine blade cooling structure optimization method according to an embodiment of this application, as follows: Figure 1 As shown in the embodiments of this application, the turbine blade cooling structure optimization method based on reinforcement learning may specifically include:

[0076] S101, responding to the input preset design conditions, obtains the initial cooling structure of the turbine blades based on the design conditions.

[0077] S102, in response to the completion of the preset agent initialization, the agent is trained based on the preset reinforcement learning algorithm within the preset reinforcement learning environment framework to obtain the trained agent.

[0078] S103, the cooling parameters of the cooling structure are optimized based on the trained agent to obtain the optimized cooling parameters, and the optimized cooling structure is determined based on the optimized cooling parameters.

[0079] In the above-mentioned reinforcement learning-based turbine blade cooling structure optimization method, reinforcement learning algorithms are introduced into the turbine blade cooling structure design process to train the agent, and the cooling parameters of the cooling structure are optimized based on the trained agent. More complex or complete cooling structures can be applied, realizing the automatic optimization and determination of the cooling parameters of the cooling structure, reducing the dependence on the designer's experience and time cost in the cooling parameter design process.

[0080] The exemplary steps described above in the embodiments of this application are illustrated below with specific examples:

[0081] S101, responding to the input preset design conditions, obtains the initial cooling structure of the turbine blades based on the design conditions.

[0082] It should be noted that this application uses reinforcement learning (RL) algorithms to optimize the turbine blade cooling structure.

[0083] In this embodiment, the design conditions refer to the pre-set design conditions for the turbine blade cooling structure. These conditions include a blade model configuration file and operating conditions. The blade model configuration file includes airfoil parameters, at least including a two-dimensional airfoil path, blade height, maximum and minimum airfoil radius, blade circumference, airfoil radius, maximum XY coordinates of the airfoil, x-point of the airfoil, y-point of the airfoil, X-coordinate of the airfoil midpoint, and the number of blades. The operating conditions include at least the total pressure of the gas inlet, the static pressure of the gas outlet, the total temperature of the gas inlet, the total temperature of the cooling gas inlet, the total pressure of the cooling gas inlet, the cooling gas efficiency, and the upper limit of the cooling gas flow rate. When the user inputs the preset design conditions, the initial cooling structure of the turbine blade is obtained based on these conditions for subsequent processing. That is, as... Figure 2 As shown, you need to input the design conditions first, including the input and analysis of the two-dimensional blade profile and the input of the operating conditions, namely the blade model configuration file and operating conditions.

[0084] S102, in response to the completion of the preset agent initialization, the agent is trained based on the preset reinforcement learning algorithm within the preset reinforcement learning environment framework to obtain the trained agent.

[0085] In this embodiment, the reinforcement learning environment framework is a pre-set reinforcement learning environment framework built based on reinforcement learning algorithms, such as the gym environment framework, which is described in this application as an example; the policy network corresponding to the reinforcement learning algorithm is described in this application using PPO and SAC reinforcement learning algorithms as examples; the policy network is a pre-set policy network corresponding to the reinforcement learning algorithm used to train the agent. When training the agent, the agent needs to be initialized first. After the agent is initialized, it is trained based on the preset policy network to obtain the trained agent for subsequent processing.

[0086] As those skilled in the art will understand, in reinforcement learning, agents make sequential decisions through interaction with their environment. Therefore, the environment in which the agent interacts plays a crucial role in the training of reinforcement learning algorithms. To design a training environment for reinforcement learning agents, the open-source Python environment simulation library gym is commonly used. gym is an open-source Python environment simulation library developed and maintained by the non-profit company OpenAI for developing and comparing the performance of reinforcement learning algorithms. It enables reinforcement learning algorithms and their environment to interact and communicate by providing a standard API.

[0087] It should be noted that the agent initialization is completed through the following steps: Under the preset reinforcement learning environment framework, the agent resets the environment and reads the corresponding hole parameter information and operating conditions based on the blade model configuration file, operating conditions, and preset hole parameter file; the hole parameter information and operating conditions are processed based on the preset full 3D evaluation algorithm to obtain the initial turbine blade surface temperature after the environment reset. The hole parameter file includes the cooling parameters of the cooling structure, which include the diameter of the turbulence column, the diameter of the film cooling hole, the inclination angle of the film cooling hole, the diameter of the impact hole, the hole row spacing, and the single-row hole spacing. The full 3D evaluation algorithm is the preset full 3D evaluation algorithm.

[0088] For example, using the pre-written gym environment framework, the agent first resets the environment by copying the necessary files to the training directory, namely the blade model configuration file, operating conditions, and preset hole parameter files. It reads these files to obtain the hole parameter information and operating conditions, and processes the hole parameter information and operating conditions based on the full three-dimensional evaluation algorithm to obtain the initial state after environmental processing, which is the initial surface temperature of the first turbine blade, for subsequent processing.

[0089] As those skilled in the art will understand, reinforcement learning is a major class of machine learning methods that aims to learn optimal behavioral strategies through interaction with the environment. Simply put, machine interaction with the environment means that the machine makes decisions and generates a set of actions in a given state of the environment, applies these actions to the environment, and the environment then changes, providing the machine with a new state and an immediate reward. In reinforcement learning training, the machine continuously interacts with the environment, with the ultimate goal of maximizing the expected cumulative reward obtained during the interaction process by following the learned strategy. Unlike other machine learning methods, in reinforcement learning, the machine making decisions is called an agent. Unlike models in supervised learning, agents learn behavioral strategies and output actions selected under the guidance of those strategies, emphasizing their ability to alter the environment.

[0090] For example, such as Figure 3 As shown, the interaction between the agent and the environment is an iterative process. The agent first perceives the current state of the environment and selects an appropriate action based on its own policy. Once the action is applied to the environment, the environment will provide a reward signal and trigger a state transition. The agent adjusts its policy based on the state before and after the action, the applied action, and the reward, and selects a new action based on the new state in the next round of interaction. Through this interaction process, the agent gradually learns the optimal decision-making policy in the current environment.

[0091] S103, the cooling parameters of the cooling structure are optimized based on the trained agent to obtain the optimized cooling parameters, and the optimized cooling structure is determined based on the optimized cooling parameters.

[0092] In this embodiment of the application, the cooling parameters of the cooling structure are optimized based on the trained agent obtained in step S102 to obtain the optimized cooling parameters, and the optimized cooling structure is determined based on the optimized cooling parameters.

[0093] The turbine blade cooling structure optimization method based on reinforcement learning provided in this application responds to preset design conditions, obtains an initial cooling structure for the turbine blade based on these conditions, and, after the preset agent initialization is completed, trains the agent using a preset reinforcement learning algorithm within a preset reinforcement learning environment framework to obtain a trained agent. The trained agent then optimizes the cooling parameters of the cooling structure to obtain optimized cooling parameters, and finally determines the optimized cooling structure based on these optimized parameters. This reinforcement learning-based turbine blade cooling structure optimization method, by introducing a reinforcement learning algorithm to train the agent during the turbine blade cooling structure design process and optimizing the cooling parameters based on the trained agent, can apply more complex or complete cooling structures. It achieves automatic optimization and determination of the cooling parameters, reducing the reliance on designer experience and time costs during the cooling parameter design process.

[0094] Furthermore, such as Figure 4 As shown, step S101 in the above embodiment, "responding to the input preset design conditions, obtaining the initial cooling structure of the turbine blades based on the design conditions," may specifically include the following steps:

[0095] S401, in response to the input preset design conditions, perform external flow channel simulation calculations based on the design conditions to generate the initial basic structure of the turbine blades.

[0096] In this embodiment of the application, when preset design conditions are input, external flow channel simulation calculations are performed based on these conditions to generate the initial basic structure of the turbine blades for subsequent processing, for example... Figure 2 As shown.

[0097] It should be noted that the basic structure is obtained through one of the following steps: selecting the basic structure based on a preset scheme database; recommending a set of blade cooling structures as the basic structure based on a preset intelligent recommendation algorithm; and determining the basic structure in response to the input turbine blade structural parameters. In other words, the determination or design of the basic structure includes three methods: the first is to select a set of existing design schemes from the scheme database (i.e., the database storing basic structure schemes); the second is to use an intelligent recommendation algorithm to recommend a set of blade cooling structure designs, and a process check is required before running the intelligent recommendation algorithm to ensure the integrity of the algorithm's input; the third is to determine the basic structure by manually inputting the turbine blade structural parameters through the user interface.

[0098] S402, based on the initial basic structure of the turbine blades, yields the corresponding initial cooling structure.

[0099] In this embodiment of the application, the initial cooling structure is obtained based on the initial basic structure of the turbine blade obtained in step S401.

[0100] It should be noted that the basic structure is used to determine the basic parameters of the turbine blade cooling structure. The basic parameters include at least the blade wall structure, the distance between the inner and outer walls and the wall thickness, the number of inner baffles, the position of the inner baffles and their thickness; the cooling structure is used to determine the cooling parameters of the turbine blade cooling structure. The cooling parameters include at least the diameter of the turbulence column, the diameter of the impact hole, the diameter of the film cooling hole, the angle of the film cooling hole, the spacing between the holes in the row, and the spacing between holes in a single row.

[0101] Furthermore, such as Figure 5 As shown, step S102 in the above embodiment, "training the agent to obtain the trained agent," may specifically include the following steps:

[0102] S501, the agent inputs the surface temperature of the first turbine blade and the cooling parameters of the cooling structure read from the orifice parameter file into the policy network to generate the corresponding target action, and then hands the target action over to the reinforcement learning environment framework for implementation, so that the reinforcement learning environment framework can adjust the cooling parameters based on the target action.

[0103] In this embodiment, the reinforcement learning environment framework includes an action space, a state space, and a reward function. The target action is the set of all actions included in the action space, representing the range and possibilities of the agent's possible behaviors. During agent training, the agent inputs the surface temperature of the first turbine blade and the cooling parameters of the cooling structure read from the orifice parameter file into the policy network to generate the corresponding target action. The target action is then implemented by the reinforcement learning environment framework, allowing the framework to tune the cooling parameters based on the target action, resulting in tuned cooling parameters. For example, after agent initialization, the agent processes the turbine blade surface temperature and the cooling parameters of the cooling structure read from the orifice file into the state required by the policy network and inputs it into the policy network to generate the corresponding target action. After the agent decides on the action, it submits the target action to the gym, which then parses the action and operates the corresponding orifice parameter file according to the action of the corresponding dimension to simulate the parameter tuning process of the designer, thereby achieving parameter tuning of the cooling parameters.

[0104] S502, in response to the completion of the cooling parameter tuning, calculates the tuned cooling parameters and operating conditions based on the full three-dimensional evaluation algorithm to obtain the surface temperature of the second turbine blade after tuning, so as to complete one round of training of the agent.

[0105] In this embodiment of the application, after the cooling parameters are adjusted in a training round, the adjusted cooling parameters and operating conditions are calculated based on the full three-dimensional evaluation algorithm to obtain the surface temperature of the second turbine blade after adjustment, so as to complete a training round of the agent. At this time, it is indicated that a training round has been completed.

[0106] S503, in response to receiving a false termination signal generated by the preset reward function, the agent continues to be trained in a new round based on the surface temperature of the second turbine blade and the adjusted cooling parameters, so as to finally obtain the trained agent.

[0107] In this embodiment, the reward function is a pre-encapsulated reward function. During a training round, the agent generates a corresponding termination signal based on the encapsulated reward function. The termination signal is judged. If the termination signal is false, the agent continues to be trained in a new round based on the surface temperature of the second turbine blade and the adjusted cooling parameters until the trained agent is finally obtained.

[0108] Furthermore, in response to a false termination signal generated by the preset reward function, the agent updates the policy network based on the tuned cooling parameters and the surface temperature of the second turbine blade; if the updated policy network fails to converge, training ends. Specifically, when the termination signal generated by the preset reward function is false, the policy network is updated based on the tuned cooling parameters and the surface temperature of the second turbine blade to obtain the updated policy network. After the update of the entire stage is completed, the agent checks whether the policy network of this stage has converged; if it has not converged, training ends.

[0109] It should be noted that, for example, such as Figure 6 As shown, after initialization, the agent generates actions using the policy network and submits them to the gym environment framework for implementation. Based on the encapsulated reward function, it generates a tuple of state, action, reward, and termination signal, which is then fed back to the agent. If the termination signal is false, the agent uses the new state (second turbine blade surface temperature) to generate a new target action and repeats the above process. If the termination signal is true, the agent ends the current trajectory acquisition round, updates the policy network based on the obtained data (including the second turbine blade surface temperature and the adjusted cooling parameters), and resets the environment to begin a new round of interaction. After the entire phase update is complete, the agent checks whether the policy network has converged. If converged, it modifies the target temperature of the environment, saves the latest cooling parameters, and continues to the next phase of training; otherwise, it terminates training.

[0110] This completes the training process for the intelligent agent.

[0111] Furthermore, such as Figure 7As shown, step S103 in the above embodiment, "optimizing the cooling parameters of the cooling structure based on the trained agent to obtain the optimized cooling parameters," may specifically include the following steps:

[0112] S701, under the reinforcement learning algorithm, obtains the initial surface temperature of the third turbine blade based on the trained agent and the updated policy network, as well as the basic parameters corresponding to the cooling parameters and the basic structure, within the reinforcement learning environment framework.

[0113] In this embodiment of the application, under the selected reinforcement learning algorithm, the initial surface temperature of the third turbine blade is obtained in the reinforcement learning environment framework based on the trained agent, the updated policy network, and the basic parameters corresponding to the cooling parameters and the basic structure obtained in the above embodiment, for subsequent processing.

[0114] S702 generates a corresponding second target action based on the surface temperature and cooling parameters of the third turbine blade, and then entrusts the second target action to the reinforcement learning environment framework for implementation in order to optimize the cooling parameters of the cooling structure, thereby obtaining the surface temperature of the fourth turbine blade and the optimized cooling parameters.

[0115] In this embodiment, a second target action is generated based on the surface temperature and cooling parameters of the third turbine blade and executed by a reinforcement learning environment framework to optimize the cooling parameters of the cooling structure, thereby obtaining the surface temperature of the fourth turbine blade and the optimized cooling parameters. It should be noted that the optimization process is similar to the training process. The AI ​​loads the corresponding parameters according to the selected reinforcement learning algorithm, uses the gym environment framework to obtain the initial state (turbine blade surface temperature), decides the action, and delivers it to the environment for implementation. For details, please refer to the training process in the above embodiment.

[0116] S703 evaluates the changes in turbine blade surface temperature based on the surface temperature of the fourth turbine blade, and determines the optimized cooling parameters based on the evaluation results.

[0117] In this embodiment, the change in the surface temperature of the turbine blade is the rise or fall of the surface temperature of the turbine blade. The change in the surface temperature of the turbine blade is evaluated based on the surface temperature of the fourth turbine blade, and the optimized cooling parameters are determined based on the evaluation results.

[0118] As one possible implementation, in response to the determination, based on the surface temperature of the fourth turbine blade and the initial surface temperature of the third turbine blade, that the turbine blade surface temperature rises or the cooling parameters exceed preset parameter boundaries during the implementation of the second target action, the agent accumulates the second target action to obtain total modification information for the cooling structure; based on the total modification information, the cooling parameters are modified to obtain the final optimized cooling parameters. Here, the parameter boundaries are the pre-set upper and lower boundaries of the cooling parameters, and the total modification information is the total modification information or total change information between the initial cooling parameters and the final determined cooling parameters of the cooling structure.

[0119] Specifically, when the optimization process is repeated, if the agent detects that the surface temperature of the turbine blade rises instead of falling after a certain second target action, that is, the surface temperature of the fourth turbine blade is higher than that of the third turbine blade, or in other words, the cooling parameters are out of bounds, the agent will accumulate the second target actions of each previous round to obtain the total modification information of the cooling structure. Based on the total modification information, the cooling parameters will be modified to obtain the final optimized cooling parameters.

[0120] This completed the process of optimizing the cooling parameters.

[0121] It should be noted that this application uses PPO and SAC reinforcement learning algorithms as examples for description, but other reinforcement learning algorithms can also be used, and no further limitations are imposed on them.

[0122] Figure 8 This is a flowchart of a reinforcement learning-based turbine blade cooling structure optimization device according to an embodiment of this application, as shown below. Figure 8 As shown in the figure, the turbine blade cooling structure optimization device 800 based on reinforcement learning in this application embodiment may specifically include:

[0123] The acquisition module 801 is used to obtain the initial cooling structure of the turbine blade based on the input preset design conditions; the design conditions include the blade model configuration file and operating conditions.

[0124] The training module 802 is used to train the agent based on a preset reinforcement learning algorithm in a preset reinforcement learning environment framework after the preset agent initialization is completed, so as to obtain the trained agent; wherein, the reinforcement learning algorithm corresponds to the policy network.

[0125] The tuning module 803 is used to tune the cooling parameters of the cooling structure based on the trained agent, obtain the tuned cooling parameters, and determine the tuned cooling structure based on the tuned cooling parameters.

[0126] In one possible implementation, the acquisition module is specifically used for:

[0127] In response to the input preset design conditions, the external flow channel simulation calculation is performed based on the design conditions to generate the initial basic structure of the turbine blades;

[0128] The initial cooling structure is derived from the initial basic structure of the turbine blade; wherein, the basic structure is used to determine the basic parameters of the cooling structure of the turbine blade; and the cooling structure is used to determine the cooling parameters of the cooling structure of the turbine blade.

[0129] In one possible implementation, agent initialization is accomplished through the following steps:

[0130] Before training the preset agent based on the preset policy network within the preset reinforcement learning environment framework, the agent resets the environment and reads the corresponding hole parameter information and operating conditions based on the blade model configuration file, operating conditions, and preset hole parameter file; wherein, the hole parameter file includes the cooling parameters of the cooling structure.

[0131] Based on a preset full three-dimensional evaluation algorithm, the bore parameter information and operating conditions are processed to obtain the initial surface temperature of the first turbine blade after environmental reset.

[0132] In one possible implementation, the training module is specifically used for:

[0133] The agent inputs the surface temperature of the first turbine blade and the cooling parameters of the cooling structure read from the orifice parameter file into the policy network to generate the corresponding target action, and then hands the target action over to the reinforcement learning environment framework for implementation, so that the reinforcement learning environment framework can tune the cooling parameters based on the target action;

[0134] In response to the completion of the cooling parameter tuning, the tuned cooling parameters and operating conditions are calculated based on the full three-dimensional evaluation algorithm to obtain the surface temperature of the second turbine blade after tuning, so as to complete one round of training of the agent.

[0135] If the termination signal generated by the preset reward function is false, the agent will continue to be trained in a new round based on the surface temperature of the second turbine blade and the adjusted cooling parameters, so as to finally obtain the trained agent.

[0136] In one possible implementation, the reinforcement learning-based turbine blade cooling structure optimization device further includes:

[0137] The update module is used to update the policy network based on the tuned cooling parameters and the surface temperature of the second turbine blades in response to the received termination signal generated by the preset reward function being false.

[0138] The decision module is used to terminate training if the updated policy network fails to converge.

[0139] In one possible implementation, the tuning module is specifically used for:

[0140] Under the reinforcement learning algorithm, the initial surface temperature of the third turbine blade is obtained based on the trained agent and the updated policy network, as well as the basic parameters corresponding to the cooling parameters and the basic structure, within the reinforcement learning environment framework.

[0141] The second objective action is generated based on the surface temperature and cooling parameters of the third turbine blade, and then the second objective action is implemented by the reinforcement learning environment framework to optimize the cooling parameters of the cooling structure, thereby obtaining the surface temperature of the fourth turbine blade and the optimized cooling parameters.

[0142] The changes in the surface temperature of the turbine blades were evaluated based on the surface temperature of the fourth turbine blade, and the optimized cooling parameters were determined based on the evaluation results.

[0143] In one possible implementation, the tuning module is specifically used for:

[0144] In response to the determination based on the surface temperature of the fourth turbine blade and the initial surface temperature of the third turbine blade that the surface temperature of the turbine blade rises or the cooling parameters exceed the preset parameter boundaries during the implementation of the second target action, the agent accumulates the second target action to obtain the total modification information for the cooling structure.

[0145] Based on the overall modification information, the cooling parameters are modified to obtain the final optimized cooling parameters.

[0146] The turbine blade cooling structure optimization device based on reinforcement learning provided in this application responds to preset design conditions, obtains an initial cooling structure for the turbine blade based on the design conditions, and, in response to the initialization of a preset agent, trains the agent based on a preset reinforcement learning algorithm within a preset reinforcement learning environment framework to obtain a trained agent. Based on the trained agent, the cooling parameters of the cooling structure are optimized to obtain optimized cooling parameters, and the optimized cooling structure is determined based on the optimized cooling parameters. This reinforcement learning-based turbine blade cooling structure optimization device, by introducing a reinforcement learning algorithm to train the agent during the turbine blade cooling structure design process and optimizing the cooling parameters of the cooling structure based on the trained agent, can apply more complex or complete cooling structures, achieving automatic optimization and determination of the cooling parameters of the cooling structure, reducing the reliance on the designer's experience and the time cost in the cooling parameter design process.

[0147] like Figure 9As shown in the embodiment of this application, an electronic device 900 includes a processor 901, a memory 902, and a bus. The memory 902 stores machine-readable instructions that can be executed by the processor 901. When the electronic device is running, the processor 901 communicates with the memory 902 via the bus. The processor 901 executes the machine-readable instructions to perform the steps of the turbine blade cooling structure optimization method based on reinforcement learning described above.

[0148] Specifically, the memory 902 and processor 901 mentioned above can be general-purpose memory and processor, without any specific limitations. When the processor 901 runs the computer program stored in the memory 902, it can execute the above-mentioned turbine blade cooling structure optimization method based on reinforcement learning.

[0149] Corresponding to the above-described reinforcement learning-based turbine blade cooling structure optimization method, this application embodiment also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, performs the steps of the above-described reinforcement learning-based turbine blade cooling structure optimization method.

[0150] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems and devices described above can be referred to the corresponding processes in the method embodiments, and will not be repeated here. In the several embodiments provided in this 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 modules is only a logical functional division, and in actual implementation, there may be other division methods. Furthermore, 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 displayed or discussed mutual coupling or direct coupling or communication connection can be through some communication interfaces; the indirect coupling or communication connection of devices or modules can be electrical, mechanical, or other forms.

[0151] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0152] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0153] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a processor-executable, non-volatile, computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the 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 to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the deployment methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, ROM, RAM, magnetic disks, or optical disks.

[0154] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for optimizing the cooling structure of turbine blades based on reinforcement learning, characterized in that, The method for optimizing the turbine blade cooling structure includes: In response to input preset design conditions, an initial cooling structure for the turbine blade is obtained based on the design conditions. The design conditions include a blade model configuration file and operating conditions. Obtaining the initial cooling structure for the turbine blade in response to input preset design conditions includes: performing external flow channel simulation calculations based on the design conditions to generate the initial basic structure of the turbine blade; obtaining the corresponding initial cooling structure based on the initial basic structure of the turbine blade; the basic structure is used to determine the basic parameters of the turbine blade's cooling structure; the cooling structure is used to determine the cooling parameters of the turbine blade's cooling structure; the basic parameters include at least the blade wall structure, the distance and thickness between the inner and outer walls, the number of inner baffles, the position and thickness of the inner baffles; the cooling parameters include at least the diameter of the turbulence column, the diameter of the impact hole, the diameter of the film cooling hole, the angle of the film cooling hole, the hole row spacing, and the spacing of a single row of holes. In response to the completion of the preset agent initialization, the agent is trained based on the preset reinforcement learning algorithm within the preset reinforcement learning environment framework to obtain the trained agent; wherein, the reinforcement learning algorithm corresponds to the policy network. The cooling parameters of the cooling structure are optimized based on the trained agent to obtain optimized cooling parameters, and an optimized cooling structure is determined based on the optimized cooling parameters. Specifically, optimizing the cooling parameters of the cooling structure based on the trained agent to obtain optimized cooling parameters includes: under the reinforcement learning algorithm, obtaining the initial surface temperature of the third turbine blade within the reinforcement learning environment framework based on the trained agent, the updated policy network, and the basic parameters corresponding to the cooling parameters and the basic structure; generating a corresponding second target action based on the third turbine blade surface temperature and the cooling parameters, and entrusting the second target action to the reinforcement learning environment framework for implementation, in order to optimize the cooling parameters of the cooling structure and obtain the fourth turbine... The process involves evaluating the change in the turbine blade surface temperature based on the fourth turbine blade surface temperature and determining the optimized cooling parameters based on the evaluation results. This evaluation includes: in response to a determination based on the fourth turbine blade surface temperature and the initial third turbine blade surface temperature that the turbine blade surface temperature rises or the cooling parameters exceed a preset parameter boundary during the implementation of the second target action, the agent accumulates the second target action to obtain total modification information for the cooling structure; and modifies the cooling parameters based on the total modification information to obtain the final optimized cooling parameters.

2. The method according to claim 1, characterized in that, Before training the agent based on a preset policy network within the preset reinforcement learning environment framework, the agent is initialized through the following steps: Within a preset reinforcement learning environment framework, the agent resets the environment and reads the corresponding hole parameter information and operating conditions based on the blade model configuration file, the operating conditions, and the preset hole parameter file; wherein, the hole parameter file includes the cooling parameters of the cooling structure; The hole parameter information and operating conditions are processed based on a preset full three-dimensional evaluation algorithm to obtain the initial surface temperature of the first turbine blade after environmental reset.

3. The method according to claim 2, characterized in that, The process of training the agent to obtain the trained agent includes: The agent inputs the surface temperature of the first turbine blade and the cooling parameters of the cooling structure read from the orifice parameter file into the policy network to generate a corresponding target action, and then submits the target action to the reinforcement learning environment framework for implementation, so that the reinforcement learning environment framework can tune the cooling parameters based on the target action; In response to the completion of the cooling parameter tuning, the tuned cooling parameters and the operating conditions are calculated based on the full three-dimensional evaluation algorithm to obtain the tuned second turbine blade surface temperature, thereby completing one round of training for the agent. If the termination signal generated by the preset reward function is false, the agent will continue to be trained in a new round based on the surface temperature of the second turbine blade and the adjusted cooling parameters, so as to finally obtain the trained agent.

4. The method according to claim 3, characterized in that, The method further includes: If the termination signal generated by the preset reward function is false, the agent updates the policy network based on the tuned cooling parameters and the surface temperature of the second turbine blade. If the updated policy network fails to converge, training ends.

5. A turbine blade cooling structure optimization device based on reinforcement learning, characterized in that, The device includes: An acquisition module is used to obtain the initial cooling structure of the turbine blade based on the input preset design conditions. The design conditions include a blade model configuration file and operating conditions. Specifically, the acquisition module is used to: perform external flow channel simulation calculations based on the input preset design conditions to generate the initial basic structure of the turbine blade; and obtain the corresponding initial cooling structure based on the initial basic structure of the turbine blade. The basic structure is used to determine the basic parameters of the turbine blade's cooling structure. The cooling structure is used to determine the cooling parameters of the turbine blade's cooling structure. The basic parameters include at least the blade wall structure, the distance and thickness between the inner and outer walls, the number of inner baffles, the position and thickness of the inner baffles, and the cooling parameters include at least the diameter of the turbulence column, the diameter of the impact hole, the diameter of the film cooling hole, the angle of the film cooling hole, the spacing between the hole rows, and the spacing between the holes in a single row. The training module is used to train the agent based on a preset reinforcement learning algorithm within a preset reinforcement learning environment framework after the preset agent initialization is completed, thereby obtaining the trained agent; wherein the reinforcement learning algorithm corresponds to a policy network. The optimization module is used to optimize the cooling parameters of the cooling structure based on the trained agent, obtain optimized cooling parameters, and determine the optimized cooling structure based on the optimized cooling parameters. Specifically, the optimization module is used under the reinforcement learning algorithm, based on the trained agent and the updated policy network, and based on the cooling parameters and the basic parameters corresponding to the basic structure, to obtain the initial surface temperature of the third turbine blade within the reinforcement learning environment framework; to generate a corresponding second target action based on the third turbine blade surface temperature and the cooling parameters, and to have the second target action implemented by the reinforcement learning environment framework to optimize the cooling parameters of the cooling structure. The process involves optimizing the cooling system to obtain the surface temperature of the fourth turbine blade and the optimized cooling parameters. The system then evaluates the changes in the surface temperature of the turbine blade based on the fourth turbine blade surface temperature and determines the optimized cooling parameters based on the evaluation results. Specifically, the optimization module is further configured to, in response to a determination based on the fourth turbine blade surface temperature and the initial third turbine blade surface temperature that the turbine blade surface temperature rises or the cooling parameters exceed preset parameter boundaries during the implementation of the second target action, accumulate the second target action to obtain total modification information for the cooling structure. Based on this total modification information, the cooling parameters are modified to obtain the final optimized cooling parameters.

6. An electronic device, characterized in that, include: The device includes a processor, a memory, and a bus. The memory stores machine-readable instructions executable by the processor. When the electronic device is running, the processor communicates with the memory via the bus. When the machine-readable instructions are executed by the processor, the steps of the reinforcement learning-based turbine blade cooling structure optimization method as described in any one of claims 1 to 4 are performed.

7. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, performs the steps of the reinforcement learning-based turbine blade cooling structure optimization method as described in any one of claims 1 to 4.

Images