A Wind Farm Control Method Based on Cooperative Distributed Model Predictive Control

Abstract

The present invention discloses a wind farm control method based on cooperative distributed model predictive control, which mainly relates to the field of wind power technology; the method includes the steps of: establishing a cooperative distributed model predictive control system model; and using a cooperative iterative optimization algorithm to find an optimal solution; the present invention performs distributed optimization based on the cooperative optimization iterative algorithm, so that the state of the entire wind farm approaches the Pareto optimum, and finally an optimal solution under system constraints is obtained, achieving the goals of improving the output power of the wind farm and reducing the thrust load of wind turbines.

Description

Technical Field

[0001] The present invention relates to the technical field of wind power generation, and more particularly to a wind farm control method based on cooperative distributed model predictive control. Background Art

[0002] Power generation efficiency is the most concerned issue in the field of wind power technology. For a long time, in order to increase the power generation efficiency of wind turbines, the power generation strategy of a single unit generally adopts maximum power point tracking (MPPT) to maximize the power of a single unit. However, due to the existence of the wake effect, the power of the wind turbines in the downstream direction will be greatly reduced. Wind turbines are subject to various complex loads during operation, such as aerodynamic loads, mechanical loads, etc. Among them, the aerodynamic load is the main manifestation of the wind turbine load, which is directly related to the generation of power, and excessive loads will have a great impact on the working performance, mechanical structure, service life, etc. of the wind turbines. Therefore, when pursuing the power generation efficiency of the wind farm, the negative impact brought by the wind turbine load cannot be ignored.

[0003] For the sake of improving the power generation efficiency and safety of the wind farm, the active power control of the wind farm has high requirements for the control strategy. However, the wind farm is a complex system with strong non-linear characteristics, and the traditional control strategy cannot meet the control requirements.

[0004] Traditional wind farms use separately optimized control settings for wind turbines, which is called greedy control. The power generation efficiency of the wind farm under this control strategy is low because it ignores the influence of the wake effect. To improve the negative impact of the wake effect, there are two general control methods: axial induction control and wake redirection control.

[0005] The idea of axial induction control is to reduce the power generation of the upstream wind turbines by changing the axial induction factor, so that the downstream wind turbines can generate more electrical energy. However, axial induction control often only considers the power generation and ignores the load.

[0006] The idea of wake redirection control is that the wind turbines of the upstream wind farm deliberately do not face the incoming wind direction, so as to deflect the direction of wake diffusion, so that the downstream wind turbines will not completely or partially overlap with the wake of the upstream wind turbines. The deflection of the wake can be completed by adjusting the pitch angle and yaw angle. For wake redirection control, the pitch angle is effective in wake redirection control, but it will cause a large increase in the load. Summary of the Invention

[0007] The object of the present invention is to solve the problems of low power generation of wind farms and large power losses of downstream wind turbines under traditional control strategies, reduce the load of wind turbines as much as possible while increasing the power generation, and provide a wind farm control method based on cooperative distributed model predictive control aiming at improving the output power of the wind farm and reducing the thrust load of wind turbines. Distributed optimization is carried out based on a collaborative optimization iteration algorithm to make the state of the entire wind farm approach the Pareto optimum, and finally the optimal solution under system constraints is obtained to achieve the goal of improving the output power of the wind farm and reducing the thrust load of wind turbines.

[0008] To achieve the above object, the present invention is realized through the following technical solutions:

[0009] A wind farm control method based on cooperative distributed model predictive control, comprising the steps of:

[0010] S1. Establish a cooperative distributed model predictive control system model;

[0011] S2. Use a collaborative iterative optimization algorithm to find the optimal solution.

[0012] Preferably, step S1 includes:

[0013] S11. Regard each wind turbine as a subsystem and establish a subsystem prediction model;

[0014] S12. Determine the optimization objective function;

[0015] S13. Perform parallel optimization on the subsystem control input, and form a candidate wind farm control input sequence through a cooperation strategy.

[0016] Preferably, in step S11, the subsystem prediction model is:

[0017]

[0018] where u and v are the state variables of the wind farm, γ i is the control input of the wind turbine subsystem i;

[0019] γ -i =(γ 1 ,…,γ i-1 ,γ i+1 ,…,γ N ) T represents the coupled control input from other wind turbines.

[0020] Preferably, in step S12, the objective function is:

[0021] J = λ1(P ref -P)+λ2F;

[0022] Among them, λ1 and λ2 are weight parameters, and P ref is the reference power of the wind farm, P is the power output of the wind farm, and F is the thrust load of the wind turbine;

[0023]

[0024] Preferably, the optimization principle of the wind farm is to optimize the objective function by optimizing the control input of the wind turbine subsystem. Applying model predictive control to the objective function gives:

[0025]

[0026]

[0027] Among them, u0 and v0 are the current states of the wind farm; represents the input sequence of the previous iteration coupling. Preferably, in step S13, the candidate control input sequence of the wind farm is:

[0028]

[0029] γ(k + 1|k) = [γ 1 (k + 1|k), …, γ i (k + M|k)] T .

[0030] Preferably, step S2 includes:

[0031] S21. Based on the cooperative distributed predictive control model in step S1, set the algorithm parameters;

[0032] S22. Use the cooperative iterative optimization algorithm to perform an exhaustive search on the feasible region to obtain the optimal solution.

[0033] Preferably, in step S21,

[0034] the optimal control input sequence of the wind farm is:

[0035]

[0036] The optimal output is: J best ;

[0037] The initial control input sequence of the wind turbine is the control input under the greedy control strategy, which is expressed as:

[0038]

[0039] The optimization output of the wind farm is J;

[0040] Let C Tbest = CT , γ best = γ, J best = J;

[0041] Set the search step ΔC for the wind turbine control input T , Δγ.

[0042] Preferably, in step S22,

[0043] At the beginning of the iteration, the No. 1 flagship machine performs an exhaustive search within the feasible region, and its control input becomes:

[0044]

[0045] γ 1 = (γ 1 ± nΔγ) ∈ [γ min , γ max , n ∈ Z;

[0046] For each change of n, perform an output check. If the current J ≤ J best , save and update J best ; If J > J best , then n = n + 1 and perform the next exhaustive search;

[0047] After the No. 1 flagship machine has completed the search, according to the search method of the No. 1 flagship machine group, perform an exhaustive search on the No. 2 to K flagship machines and update the corresponding output. After the No. 1 to K flagship machines have completed the exhaustive search, it represents the completion of one iteration. Check whether the algorithm converges to stop the iteration. The check algorithm is:

[0048] J p (k + 1) - J p-1 (k + 1) ≤ ε;

[0049] Where ε is a constant. If the algorithm converges, stop the iteration and apply the updated control input sequence to the wind farm. Otherwise, update the coupled input sequence to the current value and continue the iteration.

[0050] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0051] 1. For the wind farm control strategy under wake effect, the present invention proposes a wind farm control method based on cooperative distributed model predictive control. Considering the wake effect of the wind farm and combining the ideas of axial induction control and wake redirection control, by adjusting the yaw angle and the thrust coefficient based on the wind turbine, the power generation efficiency of the upstream wind turbines is reduced to a certain extent, so that the downstream wind turbines can capture more wind energy and increase the power generation efficiency to make up for the losses of the upstream wind turbines, and the power generation efficiency of the upstream and downstream wind turbines is adjusted to achieve the purpose of power optimization and reduction of the wind turbine load.

[0052] 2. For the wind farm control strategy, a cooperative model predictive control system is established, and the optimal control input sequence is obtained through iterative optimization.

[0053] 3. For the optimization iteration in the cooperative model predictive control system, a collaborative iterative optimization algorithm based on exhaustive search is proposed. The optimal solution is found through the exhaustive search of the feasible region, so that the state of the entire wind farm approaches the Pareto optimum.

[0054] 4. The problem that the traditional control method of the wind farm has low power generation efficiency due to ignoring the wake effect is solved, the negative impact of the wake effect is improved, the load of the wind turbines is reduced as much as possible, and the power optimization and load reduction of the wind farm are realized. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 is the flow chart of the present invention;

[0056] Figure 2 is the structure diagram of cooperative distributed model predictive control, taking the grouping result of the Danish Horns Rev I wind farm in the W wind direction as an example;

[0057] Figure 3 is the flow chart for searching the optimal solution based on the collaborative iterative algorithm. DETAILED DESCRIPTION OF THE INVENTION

[0058] The present invention will be further described below in conjunction with specific embodiments. It should be understood that these embodiments are only used to illustrate the present invention and not to limit the scope of the present invention. In addition, it should be understood that after reading the content taught by the present invention, those skilled in the art can make various changes or modifications to the present invention, and these equivalent forms also fall within the scope defined by this application.

[0059] Since model predictive control (MPC) was proposed in the 1970s, it has been successfully applied to wind power generation control in the past few decades because it can handle constrained multivariable control problems and consider various optimization factors. Due to its excellent handling ability of constraint conditions and the global optimization ability of rolling optimization, it has better control effects compared with traditional control methods and can meet the requirements of active power control in wind farms.

[0060] The concept of distributed model predictive control (DMPC) was proposed to reduce the computational load while solving optimization problems. The group-based distributed control method does not rely on the participation of the central controller of the wind farm. The distributed controller communicates with its corresponding wind turbine generator to obtain the operating status information of the corresponding wind turbine generator and communicates with the controllers of other wind turbine generators to obtain the control inputs of other wind turbine generators. Each wind turbine generator controller solves the control problem in parallel, thus effectively reducing the computational burden on the central controller.

[0061] The present invention adopts a model predictive control framework to establish a cooperative distributed model predictive control system aiming at improving the output power of the wind farm and reducing the thrust load of the wind turbine generators, establish a distributed subsystem prediction model, design a subsystem controller, perform distributed optimization among all wind turbine generator subsystem controllers, find the optimal control input of the wind turbine generator subsystem according to the global state of the wind farm, form a candidate wind farm control input sequence through a cooperation strategy, evaluate this input sequence and decide whether to update and iterate, and finally find the optimal solution under system constraints based on the overall state of the wind farm. A collaborative iterative optimization algorithm based on exhaustive search is proposed, and the optimal solution is found by exhaustive search of the feasible region. The present invention can effectively improve the power generation of the wind farm, consider the wake effect, solve the problems of low power generation of the wind farm and large power loss of the downstream wind turbine generators under traditional control strategies, and reduce the load of the wind turbine generators as much as possible while increasing the power generation, so that it works within a safe range.

[0062] Embodiment: This embodiment takes an offshore wind farm as an example to illustrate the present invention. Specifically, as shown in the appendix Figures 1-3 The present invention relates to a wind farm control method based on cooperative distributed model predictive control, which establishes a cooperative distributed model predictive control system aiming at improving the output power of the wind farm and reducing the thrust load of the wind turbine generators, performs distributed optimization based on a collaborative optimization iteration algorithm to make the state of the entire wind farm approach the Pareto optimum, and finally obtains the optimal solution under system constraints to achieve the goal of improving the output power of the wind farm and reducing the thrust load of the wind turbine generators.

[0063] It includes the steps:

[0064] S1. Establish a cooperative distributed model predictive control system model, specifically including the steps:

[0065] S11. Establish a subsystem prediction model;

[0066] In an offshore wind farm, there are N wind turbines distributed, and each wind turbine is regarded as a subsystem. The state space model of the wind farm can be expressed as:

[0067]

[0068] where u and v are the state variables of the wind farm, γ i is the control input of the i-th wind turbine subsystem.

[0069] In cooperative DMPC, each wind turbine is regarded as a subsystem, and each wind turbine subsystem controller uses the model of the wind farm to optimize its corresponding input. During the optimization process, the coupling inputs from other adjacent wind turbine subsystems are regarded as fixed parameters. Therefore, the prediction model of subsystem i is:

[0070]

[0071] where, γ -i =(γ 1 ,…,γ i-1 ,γ i+1 ,…,γ N ) T represents the coupling control inputs from other wind turbines.

[0072] According to the prediction model, the next state of the system can be predicted:

[0073]

[0074] S12. Determine the optimization objective function;

[0075] In cooperative DMPC, all wind turbine sub-controllers use the global objective function of the wind farm. In the wind farm, considering both the power of the wind farm and the thrust load of the wind turbines, the objective function can be expressed as:

[0076] J = λ1(P ref - P) + λ2F;

[0077] where λ1 and λ2 are weight parameters, P ref is the reference power of the wind farm, P is the power output of the wind farm, and F is the thrust load of the wind turbine:

[0078]

[0079] The optimization principle of the wind farm is to optimize the objective function by optimizing the control input of the wind turbine subsystem. Applying model predictive control to the objective function gives:

[0080]

[0081] where u0, v0 are the current states of the wind farm; represents the previous iteration coupled input sequence.

[0082] S13. Perform parallel optimization on the subsystem control input;

[0083] For each wind turbine subsystem controller, the task at the next sampling time is to find the optimal wind turbine subsystem control input based on the global state of the wind farm. During this process, distributed optimization is performed among all wind turbine subsystem controllers, which requires multiple iterations to make decisions in a distributed manner. After receiving the predicted future states, each wind turbine subsystem controller sends the current problem conditions to its optimizer and then solves the corresponding non-linear problem. All controllers perform parallel optimization and couple the control inputs are regarded as fixed values during the optimization process.

[0084]

[0085] The above two equations respectively represent the coupled input sequences of wind turbine subsystem i in the prediction horizon k + j|k. Wind turbine subsystem control input sequence:

[0086]

[0087] When all controllers complete the calculation, they exchange the latest input values with each other and form a candidate wind farm control input sequence through a cooperation strategy:

[0088]

[0089] γ(k + 1|k) = [γ 1 (k + 1|k), …, γ i (k + M|k)] T ;

[0090] It contains the control input sequences of all subsystems.

[0091] S2. Use the cooperative iterative optimization algorithm to find the optimal solution, specifically including the steps:

[0092] S21. Set the algorithm parameters;

[0093] Considering that the wind farm has K groups, the optimal wind farm control input sequence is:

[0094]

[0095] The optimal output is J best .

[0096] The initial control input sequence of the wind turbine is the control input under the greedy control strategy, expressed as:

[0097] γ = [γ 1 , …, γ K T ;

[0098] The optimized output of the wind farm is J

[0099] Let C Tbest = C T , γ best = γ, J best = J

[0100] Set the search step sizes ΔC T , Δγ

[0101] S22. Iteratively search for the optimal solution;

[0102] At the beginning of the iteration, the No. 1 flagship machine conducts an exhaustive search within the feasible region, and its control input becomes:

[0103]

[0104] γ 1 = (γ 1 ± nΔγ) ∈ [γ min , γ max , n ∈ Z;

[0105] For each change in n, perform an output check. If the current J ≤ J best , save and update γ 1 , J best . If J > J best , then n = n + 1 and perform the next exhaustive search

[0106] After the No. 1 flagship machine has completed its search, perform an exhaustive search on the No. 2 to Kth flagship machines and update the corresponding outputs according to the search method of the No. 1 flagship machine group. Note that the Kth flagship machine, being in the most downstream position of the wake, maintains the optimal input under the greedy strategy and does not require its control input to be optimized. When the exhaustive search of the No. 1 to Kth flagship machines is completed, it represents one iteration. Check whether the algorithm has converged to stop the iteration through the following formula:

[0107] J p (k + 1) - J p-1 ​(k + 1) ≤ ε;

[0108] where ε is a constant. If the algorithm converges, stop the iteration and apply the updated control input sequence to the wind farm; otherwise, update the coupled input sequence to the current value and continue the iteration. Since a fixed search step size is set, through this iterative method, the wind farm can finally approach the Pareto optimal state.

Claims

1. A wind farm control method based on cooperative distributed model predictive control, characterized in that, Including the steps: S1. Establish a cooperative distributed model predictive control system model; S2. Use a cooperative iterative optimization algorithm to find the optimal solution; Step S1 includes: S11. Treat each wind turbine as a subsystem and establish a subsystem prediction model; The subsystem prediction model is: where u and v are the state variables of the wind farm, γ i is the control input of the wind turbine subsystem i; γ -i = (γ 1 , …, γ i-1 , γ i+1 , …, γ N ) T represents the coupled control input from other wind turbine units; S12. Determine the optimization objective function; The objective function is: J = λ1(P ref - P) + λ2F; where λ1 and λ2 are weight parameters, P ref is the reference power of the wind farm, P is the power output of the wind farm, and F is the thrust load of the wind turbine; The optimization principle of the wind farm is to optimize the control input of the wind turbine subsystems to find the optimal solution for the objective function. Applying model predictive control to the objective function gives: Among them, u0 and v0 are the current states of the wind farm; represents the coupled input sequence of the previous iteration; S13. Perform parallel optimization on the subsystem control inputs and form a candidate wind farm control input sequence through a cooperation strategy; The candidate wind farm control input sequence is: γ(k+1|k) = [γ 1 (k+1|k), …, γ i (k+M|k)] T 。 2. The wind farm control method based on cooperative distributed model predictive control according to claim 1, characterized in that, Step S2 includes: S21. Based on the cooperative distributed prediction control model in Step S1, set the algorithm parameters; S22. Use the cooperative iterative optimization algorithm to perform an exhaustive search on the feasible region to obtain the optimal solution.

3. The wind farm control method based on cooperative distributed model predictive control according to claim 2, characterized in that, In Step S21, The optimal control input sequence of the wind farm is: The optimal output is: J best ; The initial control input sequence of the wind turbines is the control input under the greedy control strategy, denoted as: The optimization output of the wind farm is J; Let C Tbest = C T , γ best = γ, J best = J; Set the search step size ΔC for the wind turbine control input T , Δγ.

4. A wind farm control method based on cooperative distributed model predictive control according to claim 2, characterized in that, In Step S22, At the beginning of the iteration, Flagship 1 performs an exhaustive search within the feasible region, and its control input becomes: γ 1 = (γ 1 ± nΔγ) ∈ [γ min , γ max , n ∈ Z; For each change in n, perform an output check. If the current J ≤ J best , save and update γ 1 , J best ; if J > J best , then n = n + 1 and perform the next exhaustive search; After Flagship 1 has completed the search, following the search method of Flagship 1, perform an exhaustive search on Flagships 2 to K and update the corresponding outputs. When the exhaustive search of Flagships 1 to K is completed, it represents one iteration. Stop the iteration by checking whether the algorithm converges. The checking algorithm is: J p (k + 1)-J p-1 (k + 1) ≤ ε; where ε is a constant. If the algorithm converges, stop the iteration and apply the updated control input sequence to the wind farm. Otherwise, update the coupled input sequence to the current value and continue the iteration.

Images