Distributed dynamic control method and system for maximum power output of wind power plant

Abstract

The invention discloses a distributed dynamic control method and system for maximum power output of a wind power plant. The distributed dynamic control method comprises the following steps: constructing a maximum optimization problem of total power of the wind power plant and solving a global optimal axial induction factor of each fan; aiming at each fan, constructing a power maximum optimization problem of a downstream fan cluster influenced by wake flow of each fan, and solving an axial induction factor of the downstream fan cluster; generating a training data set of the fan by taking the wind speed, the wind direction and the axial induction factor of the fan as input, and subtracting the maximum value of the total power of the downstream fan cluster from the total power when the axial induction factor of the fan is 0 as output; and training a power loss estimation data driving model of a downstream fan cluster influenced by wake flow based on a deep learning model so as to construct a local distributed optimal control problem and solve an axial induction factor of the fan. The invention aims to overcome the defects of low calculation speed, heavy communication burden, low fault tolerance and the like when a traditional centralized control strategy is used for dynamically controlling the large-scale wind power plant.

Description

Technical Field

[0001] The present invention belongs to the technical field of wind power generation, and particularly relates to a decentralized dynamic control method and system for maximum power output of a wind farm. Background Art

[0002] In actual operation, a wind farm often faces the wake effect caused by the wind turbines capturing wind energy, which reduces the inflow wind speed of the downstream wind turbines, decreases their effective output power, and thus significantly affects the overall efficiency and power generation of the wind farm. Traditional wind farm power optimization methods mostly adopt centralized control, which requires a centralized wind farm controller, and its solution speed is greatly extended as the scale of the wind farm increases and the number of wind turbines increases, making it difficult to adapt to time-varying wind conditions. Distributed control and data-driven methods play an important role in significantly improving the calculation efficiency and reducing the communication complexity, but their performance highly depends on a reliable communication link. In the case of communication failures between wind turbines, the control performance may significantly decline. In contrast, decentralized control eliminates the need for a central or distributed controller, enabling each wind turbine to operate independently without communication. This not only further improves the calculation efficiency but also alleviates the concern about communication failures. However, due to the wake effect, there are complex aerodynamic couplings between wind turbines, making it extremely challenging to consider these interactions in a fully decentralized framework. Summary of the Invention

[0003] The technical problem to be solved by the present invention: Aiming at the above problems of the prior art, the present invention provides a decentralized dynamic control method and system for maximum power output of a wind farm. The present invention aims to make up for the deficiencies of traditional centralized control strategies, such as slow calculation speed, heavy communication burden, and low fault tolerance in dynamic control of large-scale wind farms, improve the calculation speed of optimal control of large-scale wind farms, eliminate complex centralized communication, and improve fault tolerance.

[0004] To solve the above technical problems, the technical solution adopted by the present invention is as follows: A decentralized dynamic control method for maximum power output of a wind farm, comprising the following steps: Under the conditions of different inflow wind speeds and directions of the wind farm, construct an optimization problem for maximizing the total power of the wind farm, and use an optimization algorithm to solve the global optimal axial induction factor of each wind turbine under different wind conditions; For each wind turbine, construct an optimization problem for maximizing the power of the downstream wind turbine cluster affected by its wake under the scenario of using different axial induction factors, and use an optimization algorithm to solve the axial induction factor of the downstream wind turbine cluster; For each wind turbine, the wind speed, wind direction and the axial induction factor of the wind turbine are used as inputs, and the total power of the downstream wind turbine cluster is subtracted from the total power when the axial induction factor of the wind turbine is the preset minimum value as the output to generate a training data set for the wind turbine; For each wind turbine, a data-driven model for estimating power loss of the downstream wind turbine cluster is constructed, and the training data set of the wind turbine is used to train the data-driven model for estimating power loss of the downstream wind turbine cluster affected by the wind turbine; For each wind turbine, a local decentralized optimal control problem is constructed by combining a data-driven model for estimating the power loss of the downstream wind turbine cluster affected by the wake of the wind turbine, and a mechanical power model and a data-driven model based on a single wind turbine. The optimal axial induction factor of the wind turbine is then solved for controlling the wind turbine.

[0005] Optionally, constructing a maximum total power optimization problem of a wind farm under conditions of different inflow wind speeds and wind directions, and using an optimization algorithm to solve the global optimal axial induction factor of each wind turbine under different wind conditions includes: S101, calculate the number of wind turbines in the wind farm according to the following formula i Mechanical power captured : , in, ρ is the air density, S = πR ² is the swept area of ​​the fan impeller, R is the radius of the fan impeller, a i , v i Fan i The axial induction factor and inflow velocity, For fans i The power coefficient, ; S102, use Jensen wake model to calculate the wake wind speed: , , in, For fans i The inflow wind speed, is the inflow wind speed at the most upstream of the wind farm, Indicates the fan j The collection of wind turbines that produce wake effects, For fans i For fans j The wake wind speed, is the wake expansion coefficient of the Jensen wake model, For wind turbines along the wind directioni The j geographical distance between wind turbines i is the wake coverage area on the plane of the wind turbine j impeller caused by the wind turbines; S103. Write the j inflow wind speed of the wind turbine as a function of the inflow wind speed at the most upstream of the wind farm and the axial induction factor of the upstream wind turbine that has a wake effect on it : , Write the j captured mechanical power of the wind turbine in the functional form of the axial induction factor as shown in the following formula :

[0006] wherein, is the axial induction factor of the wind turbine j ; Write the total power of the entire wind farm as the sum of the powers of all wind turbines as a function of the most upstream inflow wind speed and the axial induction factors of all wind turbines : , wherein, represents the set of axial induction factors of all wind turbines, n is the number of wind turbines in the wind farm; S104. Construct the objective function for the optimization problem of maximizing the total power of the wind farm: , wherein, max represents taking the maximum value of the function ; S105. Use the simulated annealing particle swarm optimization algorithm to solve the global optimal axial induction factors of each wind turbine under different wind conditions.

[0007] Optionally, the steps of solving using the simulated annealing particle swarm optimization algorithm include: S201. Initialize the parameters of the particle swarm and simulated annealing. The particle swarm is composed of the results of solving the problem; S202. Update the velocity and position according to the iterative process of the particle swarm optimization algorithm: , wherein, and respectively represent the velocity and position of the particle k+ at the i first iteration, and respectively represent the velocity and position of the particle at the k -th iteration, i wherein, w represents the inertia weight, c 1 and c 2 are non-negative acceleration coefficients, r 1 and r 2 are random numbers within the interval (0, 1) used to enhance the randomness of the search process, and respectively represent the individual optimal position of the particle and the global optimal position of the particle swarm after the k -th iteration; i S203, calculate the power output of the wind farm using the objective function of the problem to obtain the fitness of the particle for the current iteration; S204, update the individual optimal position and value according to the fitness difference and the annealing criterion:

[0008] wherein, p k+1 is the acceptance probability of the system transferring from the state k at the x k -th iteration to the state k at the x k+1 +1-th iteration, E k+1 and E k respectively represent the energy levels of the system at the k +1-th iteration and the k -th iteration, T k represents the temperature of the simulated annealing mechanism; determine if the current particle is better than the global optimal solution, and if so, update the global optimal position and value; S205, reduce the temperature according to the annealing formula: , wherein, and respectively represent the temperatures at the k+ 1-th and k -th iterations, K is the simulated annealing coefficient, and 0 < K < 1; S206, determine if the number of iterations has reached the preset threshold. If it has, output the global optimal axial induction factor for each wind turbine under different wind conditions; otherwise, jump to step S202 to continue the iteration.

[0009] Optionally, for each wind turbine, constructing an optimization problem for maximizing the power of the downstream wind turbine cluster affected by its wake in the scenario where different axial induction factors are adopted, and using an optimization algorithm to solve the axial induction factors of the downstream wind turbine cluster includes: S301. For each wind turbine, evenly divide its axial induction factor into M subintervals within the upper and lower limits, and randomly select a value from each subinterval, so as to obtain M axial induction factors; S302. For each wind turbine, according to the Jensen wake model, find the downstream wind turbine cluster affected by the wake of this wind turbine and set it as the set D , and the wind turbine cluster not affected by the wake of this wind turbine is set as the set N . Control different axial induction factors for this wind turbine respectively. The global optimal axial induction factors of the wind turbines belonging to the set N under different wind conditions are used to construct the objective function of the optimization problem for maximizing the total power of the wind turbine cluster belonging to the set D : , where and are respectively the lower limit and the upper limit of the axial induction factor of the wind turbine i ; For each wind turbine, construct an optimization problem for maximizing the power of the downstream wind turbine cluster affected by its wake in the scenario where different axial induction factors are adopted, and use an optimization algorithm to solve the axial induction factors of the downstream wind turbine cluster; S303. For each wind turbine, use the simulated annealing particle swarm optimization algorithm to solve the axial induction factors of the downstream wind turbine cluster.

[0010] Optionally, the generation of the training data set for this wind turbine includes: finding, for this wind turbine, among M different axial induction factors, the one that maximizes the total power of the wind turbine cluster belonging to the set D . Use this axial induction factor, the wind speed and the wind direction in the current scenario as inputs, and calculate the difference between the maximum total power and the total power of the wind turbine cluster belonging to the set D when the axial induction factor of this wind turbine is the preset minimum value. Use this difference as the output, and combine them into the training data set for this wind turbine.

[0011] Optionally, for each wind turbine, constructing a data-driven model for estimating the power loss of the downstream wind turbine cluster, and using the training data set of this wind turbine to train the data-driven model for estimating the power loss of the downstream wind turbine cluster affected by this wind turbine includes: S401. Construct a neuron shown in the following formula: , Among them, is the input of the neuron, is the output of the neuron, represents the dimensional space, n x and n y are respectively x in and z of the dimensions, is the weight matrix, is the bias, g represents the activation function; S402. Based on the neurons, construct a multi-layer perceptron model shown by the following formula: , , , Among them, represents the output of the k th layer and the j th neuron of the multi-layer perceptron model, is the activation function, and the activation function selects the hyperbolic tangent function tanh, represents the weight of the k th layer and the j th neuron of the multi-layer perceptron model, represents the output vector of the k -1 layer of the multi-layer perceptron model, represents the bias of the k th layer and the j th neuron of the multi-layer perceptron model, ~ respectively represent the weights of the 1st to k th layer and the N k th neurons of the multi-layer perceptron model, N k represents the number of neurons in the k th layer of the multi-layer perceptron model, ~ respectively represent the outputs of the 1st to k th layer and the N k th neurons of the multi-layer perceptron model; S403. For each fan, construct a data-driven model for estimating the power loss of the downstream fan cluster of the fan i according to the multi-layer perceptron model: , , , Among them, represents the model output of the multi-layer perceptron model, and the model output is the estimated value of the power loss of the downstream fan cluster of the fan i . x i = d i , v i , a i T represents the model input, d i , v i , a i respectively represent the inflow wind direction, inflow wind speed and axial induction factor of the fan i . b 1 to b 3 represent the biases of the first to third layers, w 1 to w 3 represent the weights of the first to third layers, w k represents the weight matrix of the k th layer, b k represents the bias vector of the k th layer; to respectively represent the weights of the first to k N k neurons in theth layer of the multi-layer perceptron model, to respectively represent the biases of the first to k N k neurons in theth layer of the multi-layer perceptron model; S404. For each fan, use the training data set of this fan to train the data-driven model for estimating the power loss of the downstream fan cluster affected by this fan.

[0012] Optionally, for each fan, combining the data-driven model for estimating the power loss of the downstream fan cluster affected by the wake of this fan, and the mechanical power model and data-driven model based on a single fan to construct a local decentralized optimal control problem, and solving to obtain the optimal axial induction factor of this fan for controlling the fan includes: S501. Construct the fan i capture mechanical power P i ​Relationship with its axial induction factor a i is: , wherein, ρ is the air density, S = πR ² is the swept area of the fan impeller, R is the radius of the fan impeller, a i , v i are respectively the axial induction factor and the inflow wind speed of the fan i ; S502, combine the relationship between the mechanical power i captured by the fan P i and its axial induction factor a i with the power loss estimation data-driven model of the downstream fan cluster to construct the objective function of the local decentralized optimal control problem : , wherein, x i = d i , v i , a i T represents the model input, d i , v i , a i respectively represent the inflow wind direction, the inflow wind speed and the axial induction factor of the fan i ; b 1 to b 3 represent the offsets of the 1st to 3rd layers, w 1 to w 3 represent the weights of the 1st to 3rd layers, is the rated power of the fan i ; is the inflow wind speed of the fan i when it is v i and operates at the rated power, and its value is the numerical solution of the following formula: ; S503, the objective function of the local decentralized optimal control problem a i ​The univariate piecewise function is used to calculate the objective function of the local decentralized optimal control problem The extreme points within each segment, as well as at 0, a u 1 / 3, and the function values at the three endpoints. The axial induction factor corresponding to the maximum value of the objective function obj is taken as a i the optimal axial induction factor of the wind turbine i for controlling the wind turbine i .

[0013] In addition, the present invention also provides a decentralized dynamic control system for maximum power output of a wind farm, including a microprocessor and a memory connected to each other. The microprocessor is programmed or configured to execute the decentralized dynamic control method for maximum power output of the wind farm.

[0014] In addition, the present invention also provides a computer-readable storage medium storing a computer program or instruction, which is programmed or configured to execute the decentralized dynamic control method for maximum power output of the wind farm through a processor.

[0015] In addition, the present invention also provides a computer program product including a computer program or instruction, which is programmed or configured to execute the decentralized dynamic control method for maximum power output of the wind farm through a processor.

[0016] Compared with the prior art, the present invention can mainly achieve the following beneficial effects: The present invention proposes a data-model hybrid-driven decentralized dynamic control method for maximum power output of a wind farm. This method uses a deep neural network to train a power loss estimation data model of the downstream wind turbine cluster for each wind turbine, and converts the centralized optimization problem of the wind farm into a decentralized control problem for each wind turbine. This method can not only improve the calculation speed of the optimal control of large-scale wind farms, eliminate complex centralized communication, and improve fault tolerance, thereby making up for the deficiencies of traditional centralized control strategies such as slow calculation speed, heavy communication burden, and low fault tolerance when dynamically controlling large-scale wind farms, but also provides a new technical path for the maximum power output control of large-scale wind farms. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 is a schematic diagram of the basic process of the method according to the embodiment of the present invention.

[0018] Figure 2 is a flowchart of the simulated annealing particle swarm optimization algorithm in the embodiment of the present invention.

[0019] Figure 3 is a structural diagram of the simulation system of the simulation case in the embodiment of the present invention.

[0020] Figure 4 This is the total power output diagram of the wind farm under different control methods in the embodiments of the present invention.

[0021] Figure 5 This is the total power output diagram of the wind farm under different control methods when some wind turbines fail in the embodiments of the present invention. Detailed implementation manners

[0022] In order to enable those skilled in the art of the present technology to better understand the technical solutions of the present invention, the technical solutions of the present invention will be further described in detail below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0023] As Figure 1 shown, the decentralized dynamic control method for maximum power output of the wind farm in this embodiment includes the following steps: S1. Under the conditions of different incoming wind speeds and directions in the wind farm, construct an optimization problem for maximizing the total power of the wind farm, and use an optimization algorithm to solve the global optimal axial induction factor of each wind turbine under different wind conditions; S2. For each wind turbine, construct an optimization problem for maximizing the power of the downstream wind turbine cluster affected by its wake under the scenario of using different axial induction factors, and use an optimization algorithm to solve the axial induction factor of the downstream wind turbine cluster; S3. For each wind turbine, with the wind speed, wind direction and the axial induction factor of the wind turbine as inputs, and the difference between the maximum total power of the downstream wind turbine cluster and the total power when the axial induction factor of the wind turbine is a preset minimum value (for example, it can take the value of 0) as the output, generate the training data set of the wind turbine; S4. For each wind turbine, construct a data-driven model for estimating the power loss of the downstream wind turbine cluster, and use the training data set of the wind turbine to train the data-driven model for estimating the power loss of the downstream wind turbine cluster affected by the wind turbine; S5. For each wind turbine, combine the data-driven model for estimating the power loss of the downstream wind turbine cluster affected by its wake, and construct a local decentralized optimal control problem based on the mechanical power model and data-driven model of a single wind turbine, and solve to obtain the optimal axial induction factor of the wind turbine for controlling the wind turbine.

[0024] In step S1 of this embodiment, the construction of the optimization problem for maximizing the total power of the wind farm under the conditions of different incoming wind speeds and directions in the wind farm, and the use of an optimization algorithm to solve the global optimal axial induction factor of each wind turbine under different wind conditions include: S101. Calculate the mechanical power captured by each wind turbine in the wind farm according to the following formula i : , ​Among them, ρ is the air density, S = πR ² is the swept area of the fan impeller, R is the radius of the fan impeller, a i , v i are respectively the axial induction factor and the inflow wind speed of the fan i , and the power coefficient of the fan i is a function related to the axial induction factor and can be expressed as: ; S102. Due to the complex wake effect in the wind farm, when the fan j is affected by the wakes of multiple upstream fans at the same time, its inflow wind speed is calculated through the wake superposition model, and the Jensen wake model is used to calculate the wake wind speed: , , Among them, is the inflow wind speed of the fan i , is the inflow wind speed at the most upstream of the wind farm, represents the set of fans j that have a wake effect on the fan , i is the wake wind speed of the fan j on the fan , is the wake expansion coefficient of the Jensen wake model, i is the geographical distance between the fan j along the wind direction and , i is the coverage area of the wake caused by the fan j on the impeller plane of the fan S103. According to the wake model, the inflow wind speed j of the fan can be written as a function of the inflow wind speed at the most upstream of the wind farm and the axial induction factors of the upstream fans that have a wake effect on it : , The mechanical power j captured by the fan is written in the form of a function of the axial induction factor as shown in the following formula :

[0025] Among them, is the fan jAxial induction factor; According to the above two equations, the total power of the entire wind farm is the sum of the powers of all wind turbines, and thus can be written as a function of the upwind inflow velocity and the axial induction factors of all wind turbines. Therefore, the total power of the entire wind farm is the sum of the powers of all wind turbines Written as a function of the upwind inflow velocity and the axial induction factors of all wind turbines : , wherein, represents the set of axial induction factors of all wind turbines, n is the number of wind turbines in the wind farm; S104, construct the objective function for the optimization problem of maximizing the total power of the wind farm: , wherein, max represents taking the maximum value of the function ; S105, use the simulated annealing particle swarm optimization algorithm to solve the global optimal axial induction factor of each wind turbine under different wind conditions.

[0026] As Figure 2 shown, the steps of using the simulated annealing particle swarm optimization algorithm to solve in this embodiment include: S201, initialize the parameters of the particle swarm and simulated annealing, and the particle swarm is composed of the results of solving the problem; S202, update the velocity and position according to the iterative process of the particle swarm optimization algorithm: , wherein, and respectively represent the velocity and position of particle k+ at the i first iteration, and respectively represent the velocity and position of particle k at the i th iteration, w represents the inertia weight, c 1 and c 2 are non - negative acceleration coefficients, r 1 and r 2 are random numbers within the interval (0, 1) used to enhance the randomness of the search process, and respectively represent the individual optimal position of particle k and the global optimal position of the particle swarm after the i th iteration; S203, calculate the power output of the wind farm using the objective function of the problem to obtain the particle fitness of the current iteration; S204. Update the individual optimal position and value according to the fitness difference and annealing criterion:

[0027] Among them, p k+1 is the state of the system at the k -th iteration x k transferring to the state at the k +1-th iteration x k+1 acceptance probability, E k+1 and E k are the energy levels of the system at the k +1-th iteration and the k -th iteration respectively, T k represents the temperature of the simulated annealing mechanism; judge if the current particle is better than the global optimal solution, then update the global optimal position and value; S205. Reduce the temperature according to the annealing formula: , Among them, and are the temperatures at the k+ 1-st and k -th iterations respectively, K is the simulated annealing coefficient, and 0 < K < 1; S206. Judge if the number of iterations reaches the preset threshold. If it reaches the preset threshold, output the global optimal axial induction factor of each fan under different wind conditions. Otherwise, jump to step S202 to continue the iteration.

[0028] In step S2 of this embodiment, for each fan, constructing the power maximization optimization problem of the downstream fan cluster affected by its wake under the scenario of different axial induction factors, and using the optimization algorithm to solve the axial induction factor of the downstream fan cluster includes: S301. For each fan, evenly divide its axial induction factor into M sub-intervals within the upper and lower limits, and randomly take a value from each sub-interval, so as to obtain M axial induction factors; S302. For each fan, find the downstream fan cluster affected by the wake of this fan according to the Jensen wake model and set it as the set D , and set the fan cluster not affected by the wake of this fan as the set N, the control of different axial induction factors is carried out for this fan respectively, belonging to the set N of the global optimal axial induction factors of the fan under different wind conditions, and the objective function of the optimization problem with the maximum total power of the fan cluster belonging to the set D is constructed: , wherein, and are respectively the lower limit and the upper limit of the axial induction factor of the fan i ; For each fan, the optimization problem with the maximum power of the downstream fan cluster affected by its wake is constructed under the scenario of different axial induction factors adopted by it, and the axial induction factor of the downstream fan cluster is solved by using an optimization algorithm; S303, for each fan, the simulated annealing particle swarm optimization algorithm is used to solve the axial induction factor of the downstream fan cluster.

[0029] In step S3 of this embodiment, the generation of the training data set of this fan includes: searching for M different axial induction factors for this fan, and selecting one that makes the total power of the fan cluster belonging to the set D the largest. Taking this axial induction factor, the wind speed and the wind direction under the current scenario as inputs, and calculating the difference between the maximum total power and the total power of the fan cluster belonging to the set D when the axial induction factor of this fan is the preset minimum value. Combining this difference as the output to form the training data set of this fan.

[0030] In step S4 of this embodiment, for each fan, a data-driven model for estimating the power loss of the downstream fan cluster is constructed. Using the training data set of this fan, the training of the data-driven model for estimating the power loss of the downstream fan cluster affected by this fan includes: S401, constructing a neuron shown by the following formula: , wherein, is the input of the neuron, is the output of the neuron, represents the dimensional space, n x and n y are respectively x in and z the dimensions of, is the weight matrix, is the bias, gdenote activation functions; Based on the need for explicit deep learning models, a multi-layer perceptron deep learning model is used to construct a power loss estimation model for the downstream fan cluster, and its basic building block is the above-mentioned neuron (artificial neuron); S402, through w and b weighted biases, the neuron can linearly combine the inputs, and through the activation function g the linear combination output can be non-linearly transformed. Based on this idea, a multi-layer perceptron model shown by the following formula is constructed based on neurons: , , , where, denotes the output of the k th layer and the j th neuron of the multi-layer perceptron model, is the activation function, and the activation function selects the hyperbolic tangent function tanh, denotes the weight of the k th layer and the j th neuron of the multi-layer perceptron model, denotes the output vector of the k -1 layer of the multi-layer perceptron model, denotes the bias of the k th layer and the j th neuron of the multi-layer perceptron model, ~ respectively denote the weights of the 1st to k th layer and the N k th neurons of the multi-layer perceptron model, N k denotes the number of neurons in the k th layer of the multi-layer perceptron model, ~ respectively denote the outputs of the 1st to k th layer and the N k th neurons of the multi-layer perceptron model; S403, for each fan, respectively construct a data-driven model for estimating the power loss of the downstream fan cluster of the fan i : , , , where, Represents the model output of a multi-layer perceptron model, and the model output is the power loss estimation value of the downstream fan cluster of the fan i of the downstream fan cluster, x i = d i , v i , a i T Represents the model input, d i 、 v i 、 a i respectively represent the inflow wind direction, inflow wind speed and axial induction factor of the fan i of the downstream fan cluster, b 1 to b 3 represent the biases of the first to third layers, w 1 to w 3 represent the weights of the first to third layers, w k Represents the weight matrix of the k th layer, b k Represents the bias vector of the k th layer; ~ respectively represent the weights of the first to k th layer of the multi-layer perceptron model N k neurons, ~ respectively represent the biases of the first to k th layer of the multi-layer perceptron model N k neurons; S404. For each fan, use the training data set of this fan to train the data-driven model for estimating the power loss of the downstream fan cluster affected by this fan.

[0031] In step S5 of this embodiment, the step of constructing a local decentralized optimal control problem by combining the data-driven model for estimating the power loss of the downstream fan cluster affected by the wake of this fan, and the mechanical power model and data-driven model based on a single fan, and solving to obtain the optimal axial induction factor of this fan for controlling the fan includes: S501. Construct the relationship between the mechanical power i captured by the fan P i and its axial induction factor a i shown in the following formula: ,​ Among them, ρ is the air density, S = πR ² is the swept area of the fan impeller, R is the radius of the fan impeller, a i , v i are respectively the axial induction factor and the inflow wind speed of the fan i ; S502. Combine the relationship between the mechanical power i captured by the fan P i and its axial induction factor a i with the power loss estimation data-driven model of the downstream fan cluster to construct the objective function of the local decentralized optimal control problem : , Among them, x i = d i , v i , a i T represents the model input, d i , v i , a i respectively represent the inflow wind direction, the inflow wind speed and the axial induction factor of the fan i ; b 1 to b 3 represent the offsets of the 1st to 3rd layers, w 1 to w 3 represent the weights of the 1st to 3rd layers, is the rated power of the fan i ; is the axial induction factor when the inflow wind speed of the fan i is v i and it works at the rated power. Its value is the numerical solution of the following formula: ; S503. The objective function of the local decentralized optimal control problem is a univariate piecewise function with respect to the axial induction factor a i , and the hyperbolic tangent function (activation function ​)( ) and the exponential function are both differentiable. Therefore, calculate the extreme points of the objective function of the local decentralized optimal control problem respectively, and the function values at the endpoints of 0, a u , 1 / 3. Select the axial induction factor corresponding to the maximum value of the objective function obj as the optimal axial induction factor of the wind turbine a i for controlling the wind turbine i . i .

[0032] To verify the effectiveness of the decentralized dynamic control method for maximum power output of the wind farm in this embodiment, in order to verify the effectiveness of the decentralized dynamic control method for maximum power output of the wind farm proposed in this embodiment, tests are carried out in the wind farm as Figure 3 shown. The test wind farm includes 7 feeders, each feeder is connected to 7 wind turbines arranged in series, and the spacing of each wind turbine in the longitudinal and transverse directions is 400 meters. The electric energy generated by the wind turbines is collected through 33kV / 155kV and 155kV / 380kV substations and transmitted to the external power grid. The 0° wind direction is Figure 3 from left to right in, parallel to the connection line of wind turbine 1 and wind turbine 43, and the positive direction is defined as the counterclockwise direction when looking down on the wind farm.

[0033] Figure 5 shows the box plot of the estimation error of the power loss estimation model of the downstream wind turbine cluster of wind turbines 1 to 42 when the wind direction is between 0° and 45°. Since there are no downstream wind turbines for wind turbines 43 to 49, the box plots of these 7 wind turbines are not shown. It can be seen that the average value of the output error of the power loss estimation model of the downstream wind turbine cluster of each wind turbine is below 50kW, and the average value of most units is below 10kW. Therefore, it can be proved that the trained model shows good accuracy. The error distribution range of the front-row wind turbines is significantly larger than that of the rear-row wind turbines, because the number of wind turbines in the downstream wind turbine cluster of the front-row wind turbines is more, resulting in deviation of the model accuracy. In addition, under the operating conditions of the wind direction between 0° and 45°, the root mean square error (RMSE) of the power loss estimation model of the downstream wind turbine cluster of each wind turbine is shown in Table 1.

[0034] Table 1 Root Mean Square Error of Power Loss Estimation Model of Downstream Wind Turbine Cluster

[0035] Figure 4The total power output of the wind farm under different control schemes is shown, and the difference between the total power output of the wind farm under normal working conditions and different control methods is shown in Table 2. The greedy control method refers to that all wind turbines work in the maximum power point tracking mode, and the centralized method refers to the method of solving the optimization problem in a centralized manner based on the SA-PSO algorithm.

[0036] Table 2 The difference between the total power output of the wind farm of the method proposed in this embodiment and different control methods under normal working conditions

[0037] Depend on Figure 4 As can be seen from Table 2, compared with the conventional greedy control method (i.e., each wind turbine works in the maximum power tracking mode), the method proposed in this embodiment and the centralized method can significantly improve the total power output of the wind farm. In addition, the total power output of the wind farm in some periods of time is higher than that of the centralized method. This is because the centralized method may still fall into the local optimum during these periods, while the method proposed in this embodiment just finds a better solution than the centralized method. The average calculation time of the method proposed in this embodiment is only 4.82×10 -4 seconds, while the centralized method requires 7.02 seconds. It can be proved that the method proposed in this embodiment can have similar or even better control performance than the centralized method based on the extremely short calculation time.

[0038] Figure 5 The difference between the total power output of the wind farm of the proposed method and the other two different control methods is shown when wind turbines 1, 16, 34, and 49 fail respectively. Table 3 shows the difference between the total power output of the wind farm of the proposed method in this embodiment and the different control methods under fault conditions.

[0039] Table 3 The difference between the total power output of the wind farm of the method proposed in this embodiment and different control methods under fault conditions

[0040] from Figure 5 It can be seen from Table 3 that in the four fault scenarios, the maximum, average and minimum values ​​of the difference between the method proposed in this embodiment and the greedy control method are still large, and the average values ​​of the centralized method are all greater than zero, which can prove the superiority of the method proposed in this embodiment. In addition, when fan 1 fails, the average difference between the method proposed in this embodiment and the centralized method is the highest, while the average difference when fan 49 fails is the lowest. Compared with Table 2, it can be found that when fan 1 fails, the average difference between the method proposed in this embodiment and the centralized method is higher than that under normal conditions, which shows that the method proposed in this embodiment shows stronger adaptability to the failure of the front row fan.

[0041] In summary, the decentralized dynamic control method for maximum power output of the wind farm in this embodiment includes constructing an optimization problem for maximizing the total power of the wind farm under different inflow wind speeds and directions of the wind farm, and using an optimization algorithm to solve the global optimal axial induction factor of each wind turbine under different wind conditions; selecting a certain wind turbine, constructing an optimization problem for maximizing the power of the downstream wind turbine cluster affected by its wake under the scenario of using different axial induction factors, and using an optimization algorithm to solve the axial induction factor of the downstream wind turbine cluster; using the wind speed, wind direction and the axial induction factor of this wind turbine as inputs, and subtracting the total power of the downstream wind turbine cluster when the axial induction factor of this wind turbine is 0 from the maximum value of the total power of the downstream wind turbine cluster as the output to generate the training data set of this wind turbine; for each wind turbine, training a data-driven model for estimating the power loss of the downstream wind turbine cluster affected by its wake based on a deep learning model; for each wind turbine, constructing a local decentralized optimal control problem based on the mechanical power model and the data-driven model of a single wind turbine. The decentralized dynamic control method for maximum power output of the wind farm in this embodiment can dynamically improve the overall power generation of the wind farm while considering the influence of the wake effect, bringing higher economic benefits to the operator; it does not require a central controller, reducing the communication complexity; the calculation speed is extremely fast, ensuring real-time performance during actual engineering applications, and having fault tolerance.

[0042] In addition, this embodiment also provides a decentralized dynamic control system for maximum power output of a wind farm, including a microprocessor and a memory connected to each other, and the microprocessor is programmed or configured to execute the decentralized dynamic control method for maximum power output of the wind farm.

[0043] In addition, this embodiment also provides a computer-readable storage medium, in which a computer program or instruction is stored, and the computer program or instruction is programmed or configured to execute the decentralized dynamic control method for maximum power output of the wind farm through a processor.

[0044] In addition, this embodiment also provides a computer program product, including a computer program or instruction, and the computer program or instruction is programmed or configured to execute the decentralized dynamic control method for maximum power output of the wind farm through a processor.

[0045] Those skilled in the art should understand that the technical solutions provided by the present invention can be in the form of a method, a system, or a computer program product. Therefore, the present invention can be implemented in the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can be in the form of a computer program product implemented on one or more computer-readable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code. The present invention is described with reference to the flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to the embodiments of the present invention. It should be understood that each flow and / or block in the flowchart and / or block diagram, as well as the combination of flows and / or blocks in the flowchart and / or block diagram, can be realized by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate a device for realizing the functions specified in the process Figure 1 one process or multiple processes and / or blocks Figure 1 one block or multiple blocks. These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable memory generate a manufactured product including an instruction device, and the instruction device realizes the functions specified in the process Figure 1 one process or multiple processes and / or blocks Figure 1 one block or multiple blocks. These computer program instructions can also be loaded onto a computer or other programmable data processing device, so that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process. Therefore, the instructions executed on the computer or other programmable device provide steps for realizing the functions specified in the process Figure 1 one process or multiple processes and / or blocks Figure 1 one block or multiple blocks.

[0046] The above is only the preferred embodiment of the present invention, and the protection scope of the present invention is not limited to the above embodiments. All technical solutions within the idea of the present invention belong to the protection scope of the present invention. It should be noted that for those of ordinary skill in the art in this technical field, several improvements and refinements made without departing from the principle of the present invention should also be regarded as within the protection scope of the present invention.

Claims

1. A decentralized dynamic control method for maximum power output of a wind farm, characterized in that, Including the following steps: Under the conditions of different incoming wind speeds and directions in the wind farm, construct an optimization problem for maximizing the total power of the wind farm, and use an optimization algorithm to solve the global optimal axial induction factor of each wind turbine under different wind conditions; For each wind turbine, construct an optimization problem for maximizing the power of the downstream wind turbine cluster affected by its wake under the scenario of using different axial induction factors, and use an optimization algorithm to solve the axial induction factor of the downstream wind turbine cluster; For each wind turbine, use the wind speed, wind direction, and the axial induction factor of the wind turbine as inputs, and subtract the total power when the axial induction factor of the wind turbine is the preset minimum value from the maximum total power of the downstream wind turbine cluster as the output to generate the training dataset of the wind turbine; For each wind turbine, construct a data-driven model for estimating the power loss of the downstream wind turbine cluster, and use the training dataset of the wind turbine to train the data-driven model for estimating the power loss of the downstream wind turbine cluster affected by the wind turbine; For each wind turbine, combine the data-driven model for estimating the power loss of the downstream wind turbine cluster affected by its wake, and construct a local decentralized optimal control problem based on the mechanical power model and data-driven model of a single wind turbine, and solve to obtain the optimal axial induction factor of the wind turbine for controlling the wind turbine.

2. The decentralized dynamic control method for maximum power output of a wind farm according to claim 1, wherein The step of constructing an optimization problem for maximizing the total power of the wind farm under the conditions of different incoming wind speeds and directions in the wind farm and using an optimization algorithm to solve the global optimal axial induction factor of each wind turbine under different wind conditions includes: S101, calculate the mechanical power captured by each wind turbine in the wind farm according to the following formula i :​ ， Among them, ρ is the air density, S = πR ² is the swept area of the fan impeller, R is the radius of the fan impeller, a i 、 v i are respectively the axial induction factor and the inflow wind speed of the fan i , is the power coefficient of the fan i ; ; S102, use the Jensen wake model to calculate the wake wind speed: ， ， Among them, is the inflow wind speed of the wind turbine i , and is the inflow wind speed at the most upstream of the wind farm. denotes the set of wind turbines j that have a wake influence on the wind turbine . i The wake wind speed of the wind turbine j on the wind turbine is the wake expansion coefficient of the Jensen wake model. is the geographical distance between the wind turbines i and j along the wind direction. is the coverage area of the wake generated by the wind turbine i on the impeller plane of the wind turbine j . S103, write the inflow wind speed of the fan j as a function of the inflow wind speed at the far upstream of the wind farm and the axial induction factor of the upstream fan that has a wake effect on it, as shown in the following formula :​ ， The captured mechanical power of the wind turbine j is written as a function of the axial induction factor in the form shown below : ​ Among them, is the axial induction factor of the fan j ; The total power of the entire wind farm is the sum of the powers of all the wind turbines written as a function of the upwind inflow wind speed and the axial induction factors of all the wind turbines : ， Among them, represents the set of all axial induction factors of the wind turbines, n is the number of wind turbines in the wind farm; S104, construct the objective function of the optimization problem for maximizing the total power of the wind farm: ， where max represents taking the maximum value of the function ; S105, use the simulated annealing particle swarm optimization algorithm to solve the global optimal axial induction factor of each wind turbine under different wind conditions.

3. The decentralized dynamic control method for maximum power output of a wind farm according to claim 2, characterized in that, The steps of using the simulated annealing particle swarm optimization algorithm to solve include: S201, initialize the parameters of the particle swarm and simulated annealing, and the particle swarm is composed of the results of solving the problem; S202, update the speed and position according to the iterative process of the particle swarm optimization algorithm: ， Among them, and respectively represent the velocity and position of the particle k+ at the i first iteration, and respectively represent the velocity and position of the particle k at the i th iteration, w represents the inertia weight, c 1 and c 2 are non - negative acceleration coefficients, r 1 and r 2 are random numbers within the interval (0, 1) used to enhance the randomness of the search process, and respectively represent the individual best position of the particle k after the i th iteration and the global best position of the particle swarm; S203, use the objective function of the problem to calculate the power output of the wind farm to obtain the fitness of the particle at the current iteration; S204, update the individual optimal position and value according to the fitness difference and annealing criterion: Among them, p k+1 is the state of the system at the k th iteration x k transferred to the state at the k +1 th iteration x k+1 acceptance probability, E k+1 and E k are the energy levels of the system at the k +1 th iteration and at the k th iteration respectively, T k represents the temperature of the simulated annealing mechanism; judge if the current particle is better than the global optimal solution, then update the global optimal position and value; S205, reduce the temperature according to the annealing formula: ， Among them, and are respectively the temperatures at the k+ 1st and k iteration, K is the simulated annealing coefficient, and 0 < K < 1; S206, determine whether the number of iterations reaches the preset threshold. If it reaches the preset threshold, output the global optimal axial induction factor of each wind turbine under different wind conditions, otherwise jump to step S202 to continue the iteration.

4. The decentralized dynamic control method for maximum power output of a wind farm according to claim 1, characterized in that, The step of constructing an optimization problem for maximizing the power of the downstream wind turbine cluster affected by its wake under the scenario of using different axial induction factors for each wind turbine and using an optimization algorithm to solve the axial induction factor of the downstream wind turbine cluster includes: S301. For each fan, evenly divide its axial induction factor into M subintervals within the upper and lower limits, and randomly select a value from each subinterval, thereby obtaining M axial induction factors; S302. For each wind turbine, find the downstream wind turbine clusters affected by the wake of this wind turbine according to the Jensen wake model and set them as a set D , and set the wind turbine clusters not affected by the wake of this wind turbine as a set N . Control this wind turbine with different axial induction factors respectively. For the global optimal axial induction factors of the wind turbines belonging to the set N , construct the objective function of the optimization problem with the maximum total power of the wind turbine clusters belonging to the set D : ， Among them, and are the lower limit and upper limit of the axial induction factor of the fan i respectively; For each wind turbine, construct an optimization problem for maximizing the power of the downstream wind turbine cluster affected by its wake under the scenario of using different axial induction factors, and use an optimization algorithm to solve the axial induction factor of the downstream wind turbine cluster; S303, for each wind turbine, use the simulated annealing particle swarm optimization algorithm to solve the axial induction factor of the downstream wind turbine cluster.

5. The decentralized dynamic control method for maximum power output of a wind farm according to claim 4, wherein The training data set for generating the wind turbine includes: searching for M one of the different axial induction factors for the wind turbine that maximizes the total power of the wind turbine cluster belonging to the set D . Taking this axial induction factor, the wind speed and wind direction in the current scenario as inputs, and calculating the difference between the maximum total power and the total power of the wind turbine cluster belonging to the set D when the axial induction factor of this wind turbine is the preset minimum value. Using this difference as the output, and combining them to form the training data set for this wind turbine.

6. The decentralized dynamic control method for maximum power output of a wind farm according to claim 1, characterized in that For each wind turbine, constructing a data-driven model for estimating the power loss of the downstream wind turbine cluster, and using the training data set of this wind turbine to train the data-driven model for estimating the power loss of the downstream wind turbine cluster affected by this wind turbine includes: S401, constructing a neuron shown in the following formula: ， Among them, is the input of the neuron, is the output of the neuron, represents the dimensional space, n x and n y are respectively x in and z the dimensions of, is the weight matrix, is the bias, g represents the activation function; S402, constructing a multi-layer perceptron model shown in the following formula based on the neuron: ， ， ， Among them, represents the output of the k -th neuron in the j -th layer of the multi-layer perceptron model. is the activation function, and the activation function selects the hyperbolic tangent function tanh. represents the weight of the k -th neuron in the j -th layer of the multi-layer perceptron model. represents the output vector of the k -1-th layer of the multi-layer perceptron model. represents the bias of the k -th neuron in the j -th layer of the multi-layer perceptron model. ~ respectively represent the weights of the 1st to k -th neurons in the N k -th layer of the multi-layer perceptron model. N k represents the number of neurons in the k -th layer of the multi-layer perceptron model. ~ respectively represent the outputs of the 1st to k -th neurons in the N k -th layer of the multi-layer perceptron model. S403. For each fan, respectively construct a data-driven model for estimating the power loss of the downstream fan cluster of the fan according to the multi-layer perceptron model i : ， ， ， Among them, represents the model output of the multi-layer perceptron model, and the model output is the estimated value of the power loss of the downstream fan cluster of the fan i . x i = d i , v i , a i T represents the model input, d i , v i , a i respectively represent the inflow wind direction, inflow wind speed and axial induction factor of the fan i . b 1 to b 3 represent the biases of the first to third layers, w 1 to w 3 represent the weights of the first to third layers, w k represents the weight matrix of the k th layer, b k represents the bias vector of the k th layer; to respectively represent the weights of the first to k th neurons of the N k th layer of the multi-layer perceptron model, to respectively represent the biases of the first to k th neurons of the N k th layer of the multi-layer perceptron model; S404, for each wind turbine, using the training data set of this wind turbine to train the data-driven model for estimating the power loss of the downstream wind turbine cluster affected by this wind turbine.

7. The decentralized dynamic control method for maximum power output of a wind farm according to claim 1, characterized in that For each wind turbine, combining the data-driven model for estimating the power loss of the downstream wind turbine cluster affected by the wake of this wind turbine, and constructing a local decentralized optimal control problem based on the mechanical power model and the data-driven model of a single wind turbine, and solving to obtain the optimal axial induction factor of this wind turbine for controlling the wind turbine includes: S501, construct a fan as shown in the following formula i Capture mechanical power P i With its axial induction factor a i Relationship: ， Among them, ρ is the air density, S = πR ² is the swept area of the fan impeller, R is the radius of the fan impeller, a i and v i are respectively the axial induction factor and the inflow wind speed of the fan i ; S502, the fan i Capture mechanical power P i Its axial induction factor a i The relationship with, combined with the data-driven model for estimating the power loss of the downstream fan cluster, constructs the objective function of the local decentralized optimal control problem : ， Among them, x i = d i , v i , a i T represents the model input, d i 、 v i 、 a i respectively represent the incoming flow wind direction, incoming flow wind speed and axial induction factor of the wind turbine i ; 1 to b 1 to b 3 represent the offsets of the first to third layers, w 1 to w 3 represent the weights of the first to third layers, is the rated power of the wind turbine i ; is the axial induction factor when the incoming flow wind speed of the wind turbine i is v i and it operates at the rated power, and its value is the numerical solution of the following formula:​ ； S503, the objective function of the local decentralized optimal control problem is a piecewise function of a single variable with respect to the axial induction factor a i . Calculate the extreme points of the objective function of the local decentralized optimal control problem in each segment, as well as the function values at the three endpoints of 0, a u , and 1 / 3. Take the axial induction factor obj corresponding to the maximum value of the objective function a i as the optimal axial induction factor of the fan i for controlling the fan i .

8. A distributed dynamic control system for maximum power output of a wind farm, comprising a microprocessor and a memory connected to each other, characterized in that, The microprocessor is programmed or configured to execute the decentralized dynamic control method for maximum power output of the wind farm according to any one of claims 1 to 7.

9. A computer-readable storage medium storing a computer program or instructions, characterized in that, The computer program or instruction is programmed or configured to execute the decentralized dynamic control method for maximum power output of the wind farm according to any one of claims 1 to 7 through a processor.

10. A computer program product, comprising a computer program or instructions, characterized in that, The computer program or instruction is programmed or configured to execute the decentralized dynamic control method for maximum power output of the wind farm according to any one of claims 1 to 7 through a processor.

Images