Wind field layout optimization method and device, storage medium and computer program product
Through a two-stage algorithm, combined with a meta-heuristic algorithm and a multi-agent reinforcement learning algorithm, the wind field layout is optimized, solving the problems of low accuracy and poor generalization ability in the existing technology, and achieving a more efficient and accurate wind field layout. optimization.
Patent Information
- Application Number
- CN202411452971.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-17
- Publication Date
- 2025-01-21
AI Technical Summary
When solving the problem of low accuracy and poor generalization ability when optimizing wind farm layout, the prior art has problems such as low accuracy and poor generalization ability, especially in consideration of wake effect.
The two -stage algorithm is used to optimize the wind farm layout. First, the rough search optimization is performed through the meta -inspiration algorithm, and then a refined search optimization is performed through the multi -intelligence strengthening learning algorithm to fully consider the tail flow effect.
The accuracy and efficiency of the optimization of the wind farm layout is improved, so that each fan can optimize its position independently and adaptively, and enhance the stability and convergence of the algorithm.
Smart Images
Figure CN119337726A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the technical field of wind farm layout, and in particular to a wind farm layout optimization method, device, storage medium and computer program product. Background Art
[0002] As human beings’ demand for energy increases, the shortage of fossil fuels becomes increasingly serious, forcing people to look for new environmentally friendly energy sources. Wind energy, as a renewable energy source, has the advantages of being clean and pollution-free. The application of wind power generation can effectively reduce dependence on fossil fuels and enhance the security and stability of energy supply.
[0003] In order to increase wind power production capacity and extend the service life of wind turbines, the micro-site selection and layout in the wind farm design process needs to be reasonably determined. The core challenge in the micro-site selection process of wind farms is the wake effect. The wake effect will lead to a reduction in the power generation of downstream wind turbines, which is likely to cause fatigue loads on wind turbines and shorten their service life.
[0004] When solving the problem of wind farm layout optimization, two aspects can be considered to obtain higher production capacity: improving the wake model to achieve more accurate speed loss simulation, or proposing a more advanced algorithm to obtain a layout solution with higher production capacity. Although the former can obtain a more accurate wake model, the higher simulation analysis accuracy has a very limited effect on the production capacity improvement, and a more effective optimization algorithm is often required. The current research on the optimization of wind turbine location in wind farms is still based on meta-heuristic algorithms because of their strong global search capabilities and easy implementation.
[0005] Although metaheuristic algorithms are very effective, in order to simplify the optimization process, wind farms usually need to be discretized into low-resolution grids, resulting in low accuracy of the optimal wind turbine coordinates. In addition, the hyperparameters of the metaheuristic algorithm need to be carefully determined for each wind farm to adapt to various wind conditions, resulting in poor generalization of the algorithm. In terms of layout optimization, there is still a lack of a comprehensive reinforcement learning-based framework to fully exploit the advantages of reinforcement learning methods. Summary of the invention
[0006] The technical purpose to be achieved by the embodiments of the present application is to provide a wind farm layout optimization method, device, storage medium and computer program product to solve the problems of low accuracy and poor generalization ability in the current optimization methods for solving wind farm layout optimization problems.
[0007] In order to solve the above technical problems, the embodiment of the present application provides a wind farm layout optimization method, including:
[0008] Based on the meta-heuristic algorithm and the wake model, a plurality of wind farm layout schemes are randomly generated for the target wind farm, and a preliminary optimization process is performed on the plurality of wind farm layout schemes to determine a pre-selected wind farm layout scheme;
[0009] Based on a multi-agent reinforcement learning algorithm, the pre-selected wind farm layout scheme is optimized secondary to determine the target wind farm layout.
[0010] Preferably, in the wind farm layout optimization method as described above, based on the meta-heuristic algorithm and the wake model, a plurality of wind farm layout schemes are randomly generated for the target wind farm, and a preliminary optimization process is performed on the plurality of wind farm layout schemes to determine a pre-selected wind farm layout scheme, including:
[0011] Randomly generate a plurality of wind farm layout plans according to the geographical location of the target wind farm and the technical parameters of the wind turbines;
[0012] Using a set consisting of a plurality of wind farm layout schemes as a population in a first iteration;
[0013] According to a preset wake model, a wind turbine power calculation model and a predetermined wind field rose diagram of the target wind field, obtaining a fitness value corresponding to each wind field layout scheme in the population;
[0014] If the first condition is met, the wind farm layout scheme corresponding to the maximum value in the fitness value is determined to be the pre-selected wind farm layout scheme, wherein the first condition is that the difference between the fitness values corresponding to any two wind farm layout schemes in the population is within a preset range, or the first iteration number corresponding to the current iteration reaches a first preset number;
[0015] If the first condition is not met, a genetic iteration is performed on the population to obtain a new population, the first iteration number is increased by one, and the step of returning to execute the step of obtaining the fitness value corresponding to each wind farm layout scheme in the set according to the preset wake model, the wind turbine power calculation model and the wind field rose diagram of the predetermined target wind field is performed.
[0016] Specifically, in the wind farm layout optimization method as described above, performing a genetic iteration on the population to obtain a new population includes:
[0017] According to the fitness values corresponding to the wind farm layout schemes in the population, the parent individual in the current iteration is determined by a roulette wheel selection method, wherein the parent individual is the code corresponding to the wind farm layout scheme in the set for genetic iteration;
[0018] The parent individuals are subjected to crossover and mutation operations to obtain a plurality of offspring individuals, and the plurality of offspring individuals constitute a new population, and the first iteration number is increased by one, and the offspring individuals are codes corresponding to the wind farm layout scheme in the new population.
[0019] Preferably, the wind farm layout optimization method as described above performs secondary optimization on the pre-selected wind farm layout scheme based on a multi-agent reinforcement learning algorithm to determine the target wind farm layout, including:
[0020] Get the number of time steps, strategy network and value network of the preset target trajectory;
[0021] According to the strategy network and the value network, the state, action reward and advantage function corresponding to each time step of the pre-selected wind farm layout scheme in the target preset trajectory are obtained;
[0022] The state, action reward, advantage function corresponding to each time step in the preset trajectory of the target and the state corresponding to the next time step are stored as experience data in a preset experience pool;
[0023] Sampling a batch of the experience data from the preset experience pool, and iteratively updating the strategy network and the value network based on the experience data;
[0024] If the second number of iterations corresponding to the iterative update reaches a second preset number, performing a second optimization on the preselected wind farm layout scheme according to the updated strategy network to determine the target wind farm layout;
[0025] If the second number of iterations is less than the second preset number, return to the step of obtaining the state, action reward and advantage function corresponding to each time step of the pre-selected wind farm layout scheme in the target preset trajectory according to the strategy network and the value network.
[0026] Specifically, in the wind farm layout optimization method as described above, the state, action reward and advantage function corresponding to each time step of the pre-selected wind farm layout scheme in the target preset trajectory are obtained according to the strategy network and the value network, including:
[0027] According to the preselected wind farm layout scheme, determining a first state of each of the wind turbines at a first time step of the target preset trajectory;
[0028] Determining, according to each of the wind turbines, a corresponding action at the first time step according to a corresponding strategy network, wherein the action represents a position change of the wind turbine;
[0029] According to the action, obtaining a second state of each wind turbine and the action reward at a second time step of the target preset trajectory, wherein the second time step is a time step next to the first time step, and the action reward is a difference between the total capacity of the wind farm at the second state and the total capacity of the wind farm at the first state;
[0030] If the number of steps corresponding to the second time step is the number of time steps, obtaining the advantage function corresponding to each time step in the target preset trajectory according to the value network;
[0031] If the number of steps corresponding to the second time step is less than the number of time steps, the second time step is updated to the first time step, and the number of steps corresponding to the second time step is increased by one, and the step of returning to execute each of the fans according to the corresponding strategy network to determine the corresponding action at the first time step, wherein the action represents the change in position of the fan.
[0032] Preferably, in the wind farm layout optimization method as described above, the pre-selected wind farm layout scheme is optimized twice based on the multi-agent reinforcement learning algorithm, and before determining the target wind farm layout, the method further includes:
[0033] The pre-selected wind farm layout scheme is de-gridded to obtain the de-gridded pre-selected wind farm layout scheme.
[0034] Another embodiment of the present application provides a device for optimizing wind farm layout, including:
[0035] A first processing module is used to randomly generate multiple wind farm layout schemes for a target wind farm based on a meta-heuristic algorithm and a wake model, and to perform preliminary optimization processing on the multiple wind farm layout schemes to determine a pre-selected wind farm layout scheme;
[0036] The second processing module is used to perform secondary optimization on the pre-selected wind farm layout scheme based on a multi-agent reinforcement learning algorithm to determine the target wind farm layout.
[0037] Yet another embodiment of the present application provides an electronic device, comprising a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the computer program implements the steps of the wind farm layout optimization method as described above when executed by the processor.
[0038] Yet another embodiment of the present application provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the wind farm layout optimization method described above are implemented.
[0039] Another embodiment of the present application provides a computer program product, including computer instructions, which, when executed by a processor, implement the steps of the wind farm layout optimization method as described above.
[0040] Compared with the prior art, the wind farm layout optimization method, device, storage medium and computer program product provided in the embodiments of the present application have at least the following beneficial effects:
[0041] The wind farm layout optimization method proposed in this application adopts a two-stage algorithm to optimize the wind farm layout, in which a coarse search optimization is first performed through a meta-heuristic algorithm, and then a fine search optimization is performed through a multi-agent reinforcement learning algorithm, so that each wind turbine can independently and adaptively optimize its position, and fully consider the wake effect, thereby improving the accuracy and efficiency of wind farm layout optimization. BRIEF DESCRIPTION OF THE DRAWINGS
[0042] Figure 1 One of the flow diagrams of the wind farm layout optimization method in this application;
[0043] Figure 2 This is the second flow chart of the wind farm layout optimization method in this application;
[0044] Figure 3 This is the third flow chart of the wind farm layout optimization method in this application;
[0045] Figure 4 This is the fourth flow chart of the wind farm layout optimization method in this application;
[0046] Figure 5 This is the fifth flow chart of the wind farm layout optimization method in this application;
[0047] Figure 6 A schematic diagram of the structure of the device for optimizing the wind farm layout in this application. DETAILED DESCRIPTION
[0048] In order to make the technical problems, technical solutions and advantages to be solved by the application clearer, the following will be described in detail in conjunction with the accompanying drawings and specific embodiments. In the following description, specific details such as specific configurations and components are provided only to help fully understand the embodiments of the application. Therefore, it should be clear to those skilled in the art that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of the application. In addition, for clarity and brevity, the description of known functions and structures has been omitted.
[0049] It should be understood that the references to "one embodiment" or "an embodiment" throughout the specification mean that the specific features, structures, or characteristics associated with the embodiment are included in at least one embodiment of the present application. Therefore, the references to "in one embodiment" or "in an embodiment" appearing throughout the specification do not necessarily refer to the same embodiment. In addition, these specific features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.
[0050] In the various embodiments of the present application, it should be understood that the size of the serial numbers of the following processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.
[0051] It should be understood that the term "and / or" in this article is only a description of the association relationship of the associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / " in this article generally indicates that the associated objects before and after are in an "or" relationship.
[0052] In the embodiments provided in the present application, it should be understood that "B corresponding to A" means that B is associated with A, and B can be determined according to A. However, it should also be understood that determining B according to A does not mean determining B only according to A, and B can also be determined according to A and / or other information.
[0053] See also Figure 1 An embodiment of the present application provides a wind farm layout optimization method, comprising:
[0054] Step S101, based on the meta-heuristic algorithm and the wake model, randomly generate multiple wind farm layout plans for the target wind farm, and perform preliminary optimization processing on the multiple wind farm layout plans to determine the pre-selected wind farm layout plan; Step S102, based on the multi-agent reinforcement learning algorithm, perform secondary optimization on the pre-selected wind farm layout plan to determine the target wind farm layout.
[0055] In this embodiment, based on the meta-heuristic algorithm and the wake model, multiple wind farm layout plans are randomly generated, and multiple wind farm layout plans are preliminarily optimized to determine the pre-selected wind farm layout plan, thereby improving the energy output and layout rationality of the wind farm layout plan, and on the basis of narrowing the search space, it is also conducive to accelerating the convergence of the reinforcement learning algorithm, providing a good starting point for subsequent multi-agent reinforcement learning, and is conducive to improving the overall optimization efficiency. At the same time, the wake effect is fully considered to facilitate the accurate generation of the optimal layout.
[0056] After determining the pre-selected wind farm layout plan, in order to conduct a more fine-grained search, each wind turbine can be regarded as an independent intelligent agent. Since the wind turbines in the layout optimization problem have a cooperative relationship and all aim to maximize the total energy output of the wind farm, the pre-selected wind farm layout plan is secondary optimized based on the multi-agent reinforcement learning algorithm to determine the final target wind farm layout; at the same time, in order to ensure the stability and convergence of the algorithm, the multi-agent proximal policy optimization (MAPPO) is preferably used in this application for secondary optimization to achieve a satisfactory decision-making effect. Among them, the use of multi-agent reinforcement learning algorithms, especially MAPPO for optimization, can conceptualize the wind farm in the wind farm as a maze, so that the wind turbines can learn to autonomously find the best position, ensuring a high degree of generalization ability, accuracy and interpretability.
[0057] In summary, the wind farm layout optimization method proposed in this application adopts a two-stage algorithm to optimize the wind farm layout, in which a coarse search optimization is first performed through a metaheuristic algorithm, and then a fine search optimization is performed through a multi-agent reinforcement learning algorithm, so that each wind turbine can independently and adaptively optimize its position, and fully consider the wake effect, thereby improving the accuracy and efficiency of wind farm layout optimization.
[0058] See also Figure 2 Preferably, in the wind farm layout optimization method as described above, the method randomly generates multiple wind farm layout schemes for the target wind farm based on the meta-heuristic algorithm and the wake model, and performs preliminary optimization processing on the multiple wind farm layout schemes to determine the pre-selected wind farm layout scheme, including:
[0059] Step S201, randomly generating a plurality of wind farm layout plans according to the geographical location of the target wind farm and the technical parameters of the wind turbines;
[0060] Step S202, taking a set consisting of a plurality of wind farm layout schemes as a population in a first iteration;
[0061] Step S203, obtaining the fitness value corresponding to each of the wind farm layout schemes in the population according to a preset wake model, a wind turbine power calculation model and a predetermined wind farm rose diagram of the target wind farm;
[0062] Step S204: if the first condition is met, determining that the wind farm layout scheme corresponding to the maximum value in the fitness value is the pre-selected wind farm layout scheme, wherein the first condition is that the difference between the fitness values corresponding to any two wind farm layout schemes in the population is within a preset range, or that the first iteration number corresponding to the current iteration reaches a first preset number;
[0063] Step S205, if the first condition is not met, a genetic iteration is performed on the population to obtain a new population, and the first iteration number is increased by one, and the step of returning to execute the step of obtaining the fitness value corresponding to each wind farm layout scheme in the set according to the preset wake model, the wind turbine power calculation model and the predetermined wind field rose diagram of the target wind field is obtained.
[0064] In this embodiment, the steps of preliminary optimization are illustrated by taking the adaptive genetic algorithm in the meta-heuristic algorithm combined with the Monte-Carlo Tree Search as an example. In the preliminary optimization of the model, firstly, based on the geographical location of the target wind farm where the wind turbines need to be arranged and the technical parameters of the wind turbines to be arranged, a plurality of wind farm layout schemes are randomly generated, each of which includes the arrangement position of each wind turbine, and the number of wind farm layout schemes generated can be set according to demand.
[0065] Then, the initial population of inheritance is determined. In this application, the set of the aforementioned randomly generated multiple wind farm layout schemes is used as the population in the first iteration, i.e., the initial population. If necessary, the initial population may also be part of multiple wind farm layout schemes.
[0066] To facilitate subsequent genetic operations, after determining the initial population, the fitness value of each individual in the population, i.e., the wind farm layout scheme, will be determined. In this embodiment, the total energy output corresponding to each individual is preferably used as its fitness value, so that the higher the fitness value, the better the wind farm layout scheme corresponding to the individual. The fitness value is calculated based on the preset wake model, the wind turbine power calculation model, and the wind field rose diagram of the predetermined target wind field, wherein the wind field rose diagram is obtained by classifying the wind direction according to the wind direction and wind speed data of the target wind field over a period of time, wherein the wind field rose diagram includes 8 or 16 wind direction sectors, and the radius of each wind direction sector represents the frequency of occurrence of the directional wind of the wind direction.
[0067] After obtaining the fitness value, it will be judged whether the first condition is met, that is, whether the current population converges and whether the first iteration number corresponding to the current iteration reaches the first preset number. When one of them is met, it can be determined that the first condition is met. At this time, the preliminary wind farm layout plan can be determined from the current population.
[0068] If it is determined through judgment that the first condition is not met at present, it means that genetic iteration needs to continue, so a genetic iteration is performed on the population to obtain a new population, and the number of first iterations is increased by one, and then the above steps of obtaining the fitness value are returned to perform repeated iterations until the first condition is met.
[0069] It should be noted that, since the fitness value obtained for the first time is highly random, it generally does not meet the first condition, so the judgment can be ignored and genetic iteration can be performed directly.
[0070] To facilitate understanding by those skilled in the art, the wake model and the wind turbine power calculation model used in this embodiment are exemplified below.
[0071] Taking the Jensen single wake model as an example, assuming that the momentum in the wake region is conserved, then the wake velocity v at a distance x from the wind turbine in the wake region is x It can be calculated by the first algorithm, where the first algorithm is expressed as:
[0072]
[0073] Where v0 is the initial wind speed, D is the diameter of the wind turbine, and parameters a and k w is an intermediate parameter, and k w =0.5 / ln(h / z0), C T is the wind turbine thrust coefficient, h is the wind turbine hub height, and z0 is the ground roughness constant.
[0074] If the downstream fan is in the wake area of the upstream fan, its position generally has two situations: completely in the wake area and partially in the wake area. If the downstream fan is partially in the wake area, the overlapping area of the two fans can be calculated according to the second algorithm, which is expressed as:
[0075]
[0076] Among them, the parameters parameter R ij For fan T j In fan T i The radius of the projection surface at, r is the radius of the wind turbine, d ij For fan T j In fan T i The projection surface at the fan T i The distance between the centers of circles.
[0077] But usually, the wind turbine may be located in the common area of the wake caused by multiple wind turbines, resulting in superimposed speed loss. i The actual speed can be calculated by the third algorithm, where the third algorithm is expressed as:
[0078]
[0079] Among them, v i Indicates fan Ti Received wind speed;
[0080] v0 represents the initial wind speed;
[0081] A ij Indicates fan T i With fan T j Overlap of wake regions;
[0082] A r Indicates fan T i The rotor area;
[0083] v ij Indicates fan T i In fan T j Speed under the influence of wake turbulence.
[0084] See also Figure 3 Specifically, in the wind farm layout optimization method as described above, performing a genetic iteration on the population to obtain a new population includes:
[0085] Step S301, determining the parent individual in the current iteration by a roulette wheel selection method according to the fitness value corresponding to each wind farm layout scheme in the population, wherein the parent individual is the code corresponding to the wind farm layout scheme in the set for genetic iteration;
[0086] Step S302, crossover and mutation operations are performed on the parent individuals to obtain multiple offspring individuals, and the multiple offspring individuals constitute a new population, and the first iteration number is increased by one, and the offspring individuals are the codes corresponding to the wind farm layout scheme in the new population.
[0087] In this embodiment, when performing genetic iteration, a roulette wheel selection method is used to select better individuals from the population as parent individuals in genetic iteration, ensuring that individuals with high fitness values have a greater probability of being selected. Then, by performing crossover and mutation operations on the parent individuals, multiple offspring individuals can be obtained, and the multiple offspring individuals can constitute a new population, wherein the crossover operation includes but is not limited to at least one of single-point crossover and multi-point crossover, and the mutation operation randomly changes part of its genes with a certain probability, increases the diversity of the population, and avoids falling into a local optimum. Preferably, the mutation operation is achieved by randomly selecting a gene position in the parent individual and performing binary code flipping.
[0088] See also Figure 4 Preferably, the wind farm layout optimization method as described above performs secondary optimization on the pre-selected wind farm layout scheme based on a multi-agent reinforcement learning algorithm to determine the target wind farm layout, including:
[0089] Step S401, obtaining the number of time steps, strategy network and value network of the preset target trajectory;
[0090] Step S402, obtaining the state, action reward and advantage function corresponding to each time step of the pre-selected wind farm layout scheme in the target preset trajectory according to the strategy network and the value network;
[0091] Step S403, storing the state, action reward, advantage function corresponding to each time step in the preset trajectory of the target and the state corresponding to the next time step as experience data in a preset experience pool;
[0092] Step S404, sampling a batch of the experience data from the preset experience pool, and iteratively updating the strategy network and the value network based on the experience data;
[0093] Step S405, if the second iteration number corresponding to the iterative update reaches a second preset number, performing secondary optimization on the pre-selected wind farm layout scheme according to the updated strategy network to determine the target wind farm layout;
[0094] Step S406, if the second iteration number is less than the second preset number, return to the step of obtaining the state, action reward and advantage function corresponding to each time step of the pre-selected wind farm layout scheme in the target preset trajectory according to the strategy network and the value network.
[0095] In this embodiment, when the pre-selected wind farm layout scheme is optimized for secondary optimization based on the centralized training with decentralized execution (CTDE) in the multi-agent reinforcement learning method, MAPPO is preferably used for optimization. The following is a specific example of optimization based on MAPPO, wherein the number of time steps, the policy network and the value network of the preset target trajectory are first obtained, wherein the number of time steps, the policy network and the value network are preferably pre-configured. Then, according to the policy network and the value network, the state, action reward, advantage function and the state corresponding to the next time step of the pre-selected wind farm layout scheme in each time step in the target preset trajectory are obtained, and the state is stored in the preset experience pool as historical data for easy training; then a batch of experience data is sampled from the preset experience pool, and the policy network and the value network are iteratively updated based on the experience data, wherein the policy network is preferably iteratively updated based on the batch gradient ascent method, and the value network is iteratively updated based on the batch gradient descent method. Since the above method is relatively conventional, the specific update process will not be repeated here. After the update is completed, it is determined whether the second iteration number corresponding to the iterative update reaches the second preset number. If it reaches the second preset number, the pre-selected wind farm layout scheme can be optimized again according to the updated strategy network to determine the target wind farm layout; otherwise, return to the above steps of obtaining the state, action reward and advantage function corresponding to each time step of the pre-selected wind farm layout scheme in the target preset trajectory according to the strategy network and the value network, so as to perform multiple iterations. This allows each wind turbine to optimize its position independently and adaptively, improving the accuracy and efficiency of wind farm layout optimization.
[0096] See also Figure 5 Specifically, the wind farm layout optimization method as described above, obtaining the state, action reward and advantage function corresponding to each time step of the pre-selected wind farm layout scheme in the target preset trajectory according to the strategy network and the value network, includes:
[0097] Step S501, determining a first state of each wind turbine at a first time step of the target preset trajectory according to the preselected wind farm layout scheme;
[0098] Step S502, determining a corresponding action at the first time step according to the corresponding strategy network of each wind turbine, wherein the action represents a position change of the wind turbine;
[0099] Step S503, according to the action, obtaining a second state of each wind turbine and the action reward at a second time step of the target preset trajectory, wherein the second time step is a time step next to the first time step, and the action reward is a difference between the total capacity of the wind farm at the second state and the total capacity of the wind farm at the first state;
[0100] Step S504, if the number of steps corresponding to the second time step is the number of time steps, obtaining the advantage function corresponding to each time step in the target preset trajectory according to the value network;
[0101] Step S505: if the number of steps corresponding to the second time step is less than the number of time steps, the second time step is updated to the first time step, and the number of steps corresponding to the second time step is increased by one, and the step of returning to execute each of the fans according to the corresponding strategy network to determine the corresponding action at the first time step, wherein the action represents the change in position of the fan.
[0102] In this embodiment, an example is given of how to obtain the state, action reward and advantage function corresponding to each time step in the target preset trajectory. First, the first state of each wind turbine in the first time step of the target preset trajectory is determined according to the preselected wind farm layout plan. The first state is the position of each wind turbine in the preselected wind farm layout plan; then based on the position and the corresponding strategy network, the action to be taken in the first time step can be determined, so that the wind turbine changes position according to the position change represented by the action, and the second state of each wind turbine in the second time step of the target preset trajectory is obtained, and further based on the total energy output corresponding to the wind farm in the second state and the total energy output corresponding to the wind farm in the first state, the difference is made, and the difference is used as the action reward corresponding to the action.
[0103] Furthermore, it will be determined whether the number of steps corresponding to the current second time step reaches the number of time steps in the target preset trajectory. If it reaches, it is determined that the target preset trajectory is completed. At this time, the corresponding advantage function can be obtained according to the value network, where the advantage function is the difference between the corresponding action value function and the state value function in the value network, where the action value function represents the expected total return of performing the preset action in the corresponding state, and the state value function represents the expected total return in the corresponding state. The advantage function can be understood as the difference between the actual performance of the action and the average performance. If it is not reached, it is determined that the target preset trajectory is not completed. At this time, the second time step is updated to the first time step, and the next iteration is performed.
[0104] See also Figure 6 Another embodiment of the present application provides a device for optimizing wind farm layout, comprising:
[0105] The first processing module 601 is used to perform preliminary optimization processing on the plurality of wind farm layout schemes based on a meta-heuristic algorithm and a wake model to determine a pre-selected wind farm layout scheme;
[0106] The second processing module 602 is used to perform secondary optimization on the pre-selected wind farm layout scheme based on a multi-agent reinforcement learning algorithm to determine a target wind farm layout.
[0107] Preferably, in the above device, the first processing module comprises:
[0108] A first processing submodule is used to randomly generate a plurality of wind farm layout schemes according to the geographical location of the target wind farm and the technical parameters of the wind turbines;
[0109] A second processing submodule is used to use a set consisting of a plurality of wind farm layout schemes as a population in a first iteration;
[0110] A third processing submodule is used to obtain the fitness value corresponding to each wind farm layout scheme in the population according to a preset wake model, a wind turbine power calculation model and a predetermined wind farm rose diagram of the target wind farm;
[0111] a fourth processing submodule, configured to determine that the wind farm layout scheme corresponding to the maximum value in the fitness value is the preselected wind farm layout scheme if a first condition is met, wherein the first condition is that the difference between the fitness values corresponding to any two wind farm layout schemes in the population is within a preset range, or that the first iteration number corresponding to the current iteration reaches a first preset number;
[0112] The fifth processing submodule is used to perform a genetic iteration on the population to obtain a new population if the first condition is not met, increase the first iteration number by one, and return to execute the step of obtaining the fitness value corresponding to each wind farm layout scheme in the set according to the preset wake model, the wind turbine power calculation model and the wind field rose diagram of the predetermined target wind field.
[0113] Specifically, in the above-mentioned device, the fourth processing submodule includes:
[0114] A first processing unit is used to determine the parent individual in the current iteration by a roulette wheel selection method according to the fitness value corresponding to each wind farm layout scheme in the population, wherein the parent individual is a code corresponding to the wind farm layout scheme in the set for genetic iteration;
[0115] The second processing unit is used to perform crossover and mutation operations on the parent individuals to obtain multiple offspring individuals, and the multiple offspring individuals constitute a new population, and the offspring individuals are codes corresponding to the wind farm layout scheme in the new population.
[0116] Preferably, in the above-mentioned device, the second processing module comprises:
[0117] A sixth sub-processing module is used to obtain the number of time steps, the strategy network and the value network of the preset target trajectory;
[0118] A seventh sub-processing module, configured to obtain, according to the strategy network and the value network, the state, action reward and advantage function corresponding to each time step of the pre-selected wind farm layout scheme in the target preset trajectory;
[0119] An eighth sub-processing module is used to store the state, action reward, advantage function corresponding to each time step in the preset trajectory of the target and the state corresponding to the next time step as experience data in a preset experience pool;
[0120] A ninth sub-processing module, configured to sample a batch of the experience data from the preset experience pool, and iteratively update the strategy network and the value network based on the experience data;
[0121] a tenth sub-processing module, configured to perform secondary optimization on the pre-selected wind farm layout scheme according to the updated strategy network to determine the target wind farm layout if the second iteration number corresponding to the iterative update reaches a second preset number;
[0122] An eleventh sub-processing module is used for returning to the step of obtaining the state, action reward and advantage function corresponding to each time step of the pre-selected wind farm layout scheme in the target preset trajectory according to the strategy network and the value network if the second iteration number is less than the second preset number.
[0123] Specifically, in the above-mentioned device, the seventh sub-processing module includes:
[0124] A third processing unit is used to determine a first state of each wind turbine at a first time step of the target preset trajectory according to the preselected wind farm layout scheme;
[0125] A fourth processing unit, configured to determine a corresponding action at the first time step according to a corresponding strategy network of each wind turbine, wherein the action represents a position change of the wind turbine;
[0126] a fifth processing unit, configured to obtain, according to the action, a second state of each of the wind turbines at a second time step of the target preset trajectory and the action reward, wherein the second time step is a time step next to the first time step, and the action reward is a difference between a total capacity of the wind farm at the second state and a total capacity of the wind farm at the first state;
[0127] a sixth processing unit, configured to obtain, according to the value network, the advantage function corresponding to each time step in the target preset trajectory if the number of steps corresponding to the second time step is the number of time steps;
[0128] The seventh processing unit is used to update the second time step to the first time step if the number of steps corresponding to the second time step is less than the number of time steps, and increase the number of steps corresponding to the second time step by one, and return to execute the step of each of the fans determining the corresponding action at the first time step according to the corresponding strategy network, wherein the action represents the change in position of the fan.
[0129] The device embodiment of the present application is a device corresponding to the embodiment of the above-mentioned wind farm layout optimization method. All the implementation means in the above-mentioned method embodiment are applicable to the embodiment of the device and can achieve the same technical effect. The above-mentioned device provided in the embodiment of the present application can implement all the method steps implemented in the embodiment of the above-mentioned wind farm layout optimization method and can achieve the same technical effect. The parts and beneficial effects of this embodiment that are the same as those in the method embodiment will not be described in detail here.
[0130] Another embodiment of the present application provides an electronic device, including a processor, a memory, and a computer program stored in the memory and executable on the processor. When the computer program is executed by the processor, the steps of the wind farm layout optimization method as described above are implemented, and the same technical effect can be achieved. To avoid repetition, it will not be described here.
[0131] Another embodiment of the present application provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the wind farm layout optimization method as described above are implemented, and the same technical effect can be achieved. To avoid repetition, it will not be repeated here.
[0132] Another embodiment of the present application provides a computer program product, including computer instructions, which, when executed by a processor, implement the steps of the wind farm layout optimization method as described above and can achieve the same technical effect. To avoid repetition, they are not described here.
[0133] In addition, the present application may repeat reference numerals and / or letters in different examples. This repetition is for the purpose of simplicity and clarity, and does not in itself indicate the relationship between the various embodiments and / or settings discussed.
[0134] It should also be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusions.
[0135] The above is a preferred embodiment of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles described in the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.
Claims
1. A wind farm layout optimization method, characterized in that: include: Based on the meta-heuristic algorithm and the wake model, a plurality of wind farm layout schemes are randomly generated for the target wind farm, and a preliminary optimization process is performed on the plurality of wind farm layout schemes to determine a pre-selected wind farm layout scheme; Based on a multi-agent reinforcement learning algorithm, the pre-selected wind farm layout scheme is optimized secondary to determine the target wind farm layout.
2. The wind farm layout optimization method according to claim 1, characterized in that: The method randomly generates multiple wind farm layout schemes for the target wind farm based on the meta-heuristic algorithm and the wake model, and performs preliminary optimization processing on the multiple wind farm layout schemes to determine the pre-selected wind farm layout scheme, including: randomly generating a plurality of wind farm layout plans according to the geographical location of the target wind farm and the technical parameters of the wind turbines; Using a set consisting of a plurality of wind farm layout schemes as a population in a first iteration; According to a preset wake model, a wind turbine power calculation model and a predetermined wind field rose diagram of the target wind field, obtaining a fitness value corresponding to each wind field layout scheme in the population; If the first condition is met, the wind farm layout scheme corresponding to the maximum value in the fitness value is determined to be the pre-selected wind farm layout scheme, wherein the first condition is that the difference between the fitness values corresponding to any two wind farm layout schemes in the population is within a preset range, or the first iteration number corresponding to the current iteration reaches a first preset number; If the first condition is not met, a genetic iteration is performed on the population to obtain a new population, the first iteration number is increased by one, and the step of returning to execute the step of obtaining the fitness value corresponding to each wind farm layout scheme in the set according to the preset wake model, the wind turbine power calculation model and the wind field rose diagram of the predetermined target wind field is performed.
3. The wind farm layout optimization method according to claim 2, characterized in that: The performing a genetic iteration on the population to obtain a new population comprises: According to the fitness values corresponding to the wind farm layout schemes in the population, the parent individual in the current iteration is determined by a roulette wheel selection method, wherein the parent individual is the code corresponding to the wind farm layout scheme in the set for genetic iteration; The parent individuals are subjected to crossover and mutation operations to obtain a plurality of offspring individuals, and the plurality of offspring individuals constitute a new population, and the first iteration number is increased by one, and the offspring individuals are codes corresponding to the wind farm layout scheme in the new population.
4. The wind farm layout optimization method according to claim 1, characterized in that: Based on a multi-agent reinforcement learning algorithm, the pre-selected wind farm layout scheme is optimized twice to determine the target wind farm layout, including: Get the number of time steps, strategy network and value network of the preset target trajectory; According to the strategy network and the value network, the state, action reward and advantage function corresponding to each time step of the pre-selected wind farm layout scheme in the target preset trajectory are obtained; The state, action reward, advantage function corresponding to each time step in the preset trajectory of the target and the state corresponding to the next time step are stored as experience data in a preset experience pool; Sampling a batch of the experience data from the preset experience pool, and iteratively updating the strategy network and the value network based on the experience data; If the second number of iterations corresponding to the iterative update reaches a second preset number, performing a second optimization on the preselected wind farm layout scheme according to the updated strategy network to determine the target wind farm layout; If the second number of iterations is less than the second preset number, return to the step of obtaining the state, action reward and advantage function corresponding to each time step of the pre-selected wind farm layout scheme in the target preset trajectory according to the strategy network and the value network.
5. The wind farm layout optimization method according to claim 4, characterized in that: The state, action reward and advantage function corresponding to each time step of the pre-selected wind farm layout scheme in the target preset trajectory are obtained according to the strategy network and the value network, including: According to the preselected wind farm layout scheme, determining a first state of each wind turbine at a first time step of the target preset trajectory; Determining, according to each of the wind turbines, a corresponding action at the first time step according to a corresponding strategy network, wherein the action represents a position change of the wind turbine; According to the action, obtaining a second state of each wind turbine and the action reward at a second time step of the target preset trajectory, wherein the second time step is a time step next to the first time step, and the action reward is a difference between the total capacity of the wind farm at the second state and the total capacity of the wind farm at the first state; If the number of steps corresponding to the second time step is the number of time steps, obtaining the advantage function corresponding to each time step in the target preset trajectory according to the value network; If the number of steps corresponding to the second time step is less than the number of time steps, the second time step is updated to the first time step, and the number of steps corresponding to the second time step is increased by one, and the step of returning to execute each of the fans according to the corresponding strategy network to determine the corresponding action at the first time step, wherein the action represents the change in position of the fan.
6. A device for optimizing wind farm layout, characterized in that: include: A first processing module is used to randomly generate a plurality of wind farm layout schemes for a target wind farm based on a meta-heuristic algorithm and a wake model, and to perform preliminary optimization processing on the plurality of wind farm layout schemes to determine a pre-selected wind farm layout scheme; The second processing module is used to perform secondary optimization on the pre-selected wind farm layout scheme based on a multi-agent reinforcement learning algorithm to determine the target wind farm layout.
7. An electronic device, characterized in that: The invention comprises a processor, a memory and a computer program stored in the memory and executable on the processor, wherein the computer program, when executed by the processor, implements the steps of the wind farm layout optimization method according to any one of claims 1 to 5.
8. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the wind farm layout optimization method according to any one of claims 1 to 5 are implemented.
9. A computer program product, characterized in that The method comprises computer instructions, which, when executed by a processor, implement the steps of the wind farm layout optimization method according to any one of claims 1 to 5.
Citation Information
Cited By
Offshore wind plant micro-siting optimization method considering regular layout
CN121168738A