A Field-Level Control Strategy for Offshore Wind Farms Based on Distributed Rolling Optimization

Abstract

The present invention discloses an offshore wind farm field-level control strategy based on distributed rolling optimization, including: obtaining the wind farm flow field distribution information in real time and using it as the environmental input of the wind farm prediction proxy model, and predicting the dynamic operation characteristics of the wind farm under different control strategies by using the prediction proxy model; in closed-loop control, decomposing the centralized optimization problem of the large-scale wind farm system into multiple distributed local optimization sub-problems according to the unit spatial arrangement rule, establishing a multi-objective optimization function considering power optimization scheduling and fatigue load, and sequentially solving the local control optimal solutions of each wind farm subsystem within the unit optimization time domain period; repeating the above process in the next rolling optimization time domain period, and finally realizing the multi-objective rolling optimal control of the large-scale wind farm under the changing dynamic environmental wind conditions. The present invention can solve the problems of difficult optimization and high calculation cost of large-scale offshore wind farms under changing wind conditions, and realize active power optimization scheduling and load reduction.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent control of offshore wind farms, and particularly to an offshore wind farm field-level control strategy based on distributed rolling optimization. Background Art

[0002] Wind energy is an important renewable clean energy. However, in the application of wind power generation technology, in order to prevent the instability of wind energy from posing a hidden danger to the safety of the power grid and improve the stability of the system, it is required that the wind farms connected to the power grid should have a certain active power control ability, be able to control their active power output according to the requirements of the power grid dispatching, and improve the working stability level of the power system.

[0003] Currently, the traditional maximum power tracking technology still cannot accurately track the power grid dispatching curve. Therefore, it is necessary to develop a new intelligent active power control method to realize the dynamic power optimization dispatching of the wind farm while reducing the fatigue load on the key parts of the units in the field and improving the economic benefits of the wind farm system.

[0004] Chinese patent document with publication number CN104917204A discloses a method for optimizing the active power control of a wind farm. First, it collects the operation state data of the wind turbines in the wind farm in the current control period, the wind speed at the location of the wind turbines in the current control period, the output of the wind turbines in the current control period, and the predicted wind speed at the location of the wind turbines in the next control period, and receives the planned value of the active power of the wind farm issued by the dispatching center in real time. The active power control system of the wind farm reasonably arranges the output values of the wind turbines in the wind farm through an active power control optimization algorithm according to the collected data of each wind turbine and issues them to each wind turbine participating in the regulation, so as to realize the tracking of the planned value of the active power output value of the entire wind farm by the dispatching.

[0005] Chinese patent document with publication number CN103296701A discloses a method for controlling the active power of a wind farm, including the following steps: calculating the active power prediction error at each prediction point of the wind farm and establishing an active power prediction error distribution function of the wind farm; successively establishing an active power prediction error distribution model, an active power prediction confidence model and an active power control model of the wind farm; optimizing the active power control model of the wind farm to obtain the optimized confidence weights of each unit group, and then controlling the active power of the wind farm.

[0006] With the substantial increase in the installed capacity of wind power, the construction of wind farms is gradually showing large capacity and large scale. For large-scale wind farms, there are problems such as high optimization calculation cost and great difficulty in realizing their intelligent optimization control under changing environmental wind conditions. Summary of the Invention

[0007] The present invention provides a field-level control strategy for an offshore wind farm based on distributed rolling optimization, which can achieve dynamic intelligent optimization control of large-scale offshore wind farms under changing environmental wind conditions.

[0008] A field-level control strategy for an offshore wind farm based on distributed rolling optimization includes:

[0009] (1) Using the wind farm sensing system to obtain the internal flow field distribution of the wind farm in real time, and recording the dynamically changing environmental wind parameters and the operating state information of each unit at the current moment;

[0010] (2) Establishing a predictive agent model for the collaborative control of the offshore wind farm, including an environmental wind sub-model, a wind turbine sub-model, and a wake sub-model; using the internal flow field distribution of the wind farm obtained by the wind farm sensing system to correct the parameters of the predictive agent model;

[0011] (3) Taking the environmental wind parameters and the operating state information of each unit at the current moment obtained in step (1) as inputs, using the predictive agent model to characterize and predict the operating characteristics of the actual wind farm at future moments, and obtaining the time-domain changes of the output power and the key part loads of each unit under different collaborative control strategies;

[0012] (4) Considering comprehensively the operating characteristics of the wind farm within a unit optimization time domain period, establishing a multi-objective optimization function that minimizes the fatigue load under the condition of active power optimal scheduling, and realizing the active power optimal scheduling of the wind farm and the improvement of the fatigue life of the key components of the unit by minimizing the objective function;

[0013] (5) Decomposing the multi-objective optimization problem into sub-problems of several small-scale distributed local dynamic optimization subsystems, using intelligent optimization algorithms to solve the objective functions of each sub-problem, and comprehensively obtaining the distributed global optimum through the communication between the problems, and finally obtaining the optimal control within this optimization period;

[0014] (6) Sending the obtained optimal control parameter target values of each unit to the actuators of each wind turbine, and realizing the dynamic optimization scheduling of the wind farm power and the fatigue load optimization within a future limited time domain through single control or joint control methods;

[0015] (7) Repeating steps (1)-(5) in the next rolling optimization period, and executing the optimal control strategy in the current optimization period within its corresponding control period, so as to realize the multi-objective rolling optimization and optimal control of the wind farm under dynamically changing wind conditions.

[0016] The present invention obtains the flow field distribution information of a wind farm in real time according to a wind farm perception system, and uses it as the environmental input of a wind farm prediction agent model. The wind farm prediction agent model is used to obtain the dynamic operation characteristics of the wind farm under different control strategies. The centralized optimization problem of a large-scale wind farm system is decomposed into multiple distributed local optimization sub-problems according to the law of unit spatial arrangement. A multi-objective optimization function considering power optimization scheduling and fatigue load is established. The intelligent optimization algorithm is used to sequentially solve the local control optimal solutions of each wind farm subsystem within the unit optimization time domain period, so as to achieve the overall control optimum. And repeat the above optimization solution process in the next rolling optimization time domain period, and finally achieve the multi-objective rolling optimal control of the large-scale wind farm under the dynamic environmental wind condition changes.

[0017] Further, in step (1), the environmental wind parameters include wind information parameters under wake influence and without wake influence; the wind information parameters under wake influence include wake wind speed deficit, wake expansion, and wake meandering; and are obtained by measuring the wind speed distribution in the wake area downstream of the wind turbine.

[0018] The wind information parameters without wake influence include the average wind speed, wind direction, and turbulence intensity of the environmental wind; and are obtained by using lidar wind measurement technology to scan and inversely calculate the spatial distribution of the wind speed in the wind farm flow field in real time based on the Doppler principle.

[0019] The operating state information of each unit at the current moment includes the yaw angle and pitch angle, and is obtained by using the wind turbine SCADA system.

[0020] In step (2), the environmental wind sub-model uses the Taylor frozen turbulence hypothesis to complete the spatial propagation of the wind farm flow field; the wind turbine sub-model adopts a brake disc model, and sets the rotor surface radius and hub height parameters of the unit according to the simulation requirements; the wake sub-model is used to capture and predict the key wake characteristics related to the wind farm output power and wind turbine load, including wake wind speed deficit, wake expansion, and wake meandering.

[0021] In step (4), an optimization objective function considering both efficiency improvement and load reduction is established, so that the wind farm realizes the optimization scheduling of active power and minimizes the maximum fatigue load suffered by each unit, thereby improving the economic benefits and working life of the wind farm. The specific process of establishing the multi-objective optimization function is as follows:

[0022] (4-1) Calculate the output power of unit i in the wind farm, and the calculation formula is as follows:

[0023]

[0024] Where, P i is the output power (W) of the generator of unit i in the wind farm, T q,i(t) is the instantaneous torque (N-m) of the generator of unit i at time t, ω i is the instantaneous rotational speed (rpm) of the generator of unit i at time t, t k is the current time, and ΔT is the unit optimization time domain period. In the present invention, the pitch angle and the tip speed ratio can be changed through pitch control and torque control, so as to optimize the output power of the wind farm; or the ambient wind inflow angle can be changed through yaw control to optimize the operating performance of the wind farm.

[0025] According to the above power calculation formula of a single unit, the total generated power of the wind farm system is the sum of the generated powers of all wind turbines in the wind farm system within a unit time It is:

[0026]

[0027] (4-2) Calculate the equivalent fatigue load on wind turbine i:

[0028] Use the prediction proxy model to obtain the time-domain variation of the load on the key parts of wind turbine i in the wind farm, including the load variations at the bottom of the tower and the root of the blade; through the rainflow counting method, perform equivalent analysis on the time-domain varying load on wind turbine i to obtain the equivalent fatigue load DEL of the key parts:

[0029]

[0030] In the formula, DEL is the equivalent fatigue load, is the equivalent cycle number within the time series ΔT, N ΔT,i is the number of occurrences of the i-th condition within the time series ΔT, L ΔT,i is the load range of the i-th condition within the time series ΔT, and m is the slope of the material S-N curve;

[0031] (4-3) Establish the objective function model for multi-objective optimization:

[0032] To achieve the optimal scheduling of the active power of the wind farm, make the output power of the wind farm meet the active power expectation value, and reduce the fatigue load on the key parts of the unit, the objective function is established as follows:

[0033]

[0034] In the formula, P des is the total output power expectation value of the wind farm system, is the total generated power of the wind farm system, is the normalized power deviation value, DEL normwhere \(F_{max}\) is the maximum fatigue load on the key parts of each normalized wind turbine generator, and \(\alpha\) is the weight coefficient.

[0035] The number of wind turbine generators in a large-scale wind farm is often huge. Using the objective function established in step (4) for centralized optimization calculation has a high cost. Step (5) aims to reduce the calculation cost of solving the optimal control for a large-scale wind farm. By using a distributed optimization strategy, the dynamic multi-objective optimization problem is decomposed into several small-scale distributed local dynamic optimization sub-problems. That is, for a large-scale offshore wind farm, according to the spatial arrangement characteristics of the generators, some adjacent generators in the spatial arrangement are selected to form a subsystem, the objective function of each subsystem is calculated, and the overall optimal solution is calculated by solving the optimal control strategies of each subsystem.

[0036] In step (5), the solution process of the sub-problems of several small-scale distributed local dynamic optimization subsystems is to sequentially complete the solution of all sub-problems in a specific order. First, fix the current optimal control and state of other subsystems as the initial value conditions, calculate and solve the minimization problem of the \(i\)-th distributed subsystem, and then sequentially solve multiple linear minimization problems to update the optimal control of other subsystems. When the optimal control parameters of each subsystem obtained after all subsystems complete one iteration of solution are compared with the results of the previous iteration, if the difference between the current optimal control obtained by a certain subsystem in the current iteration and the result obtained in the previous iteration is greater than the threshold, the non-converged subsystem enters the next iteration calculation process, and the converged subsystem no longer performs iterative solution, and the current solution order is updated. Repeat the above optimization solution until the optimal control of all subsystems converges.

[0037] Each distributed subsystem minimizes its own objective function, thereby achieving the optimal performance of the overall system while reducing the calculation cost of the overall field-level optimization. The expression form of the \(i\)-th distributed local dynamic optimization sub-problem within the optimization time \(t\in[k\Delta T,(k + P)\Delta T]\) is:

[0038]

[0039] where \(t\in[k\Delta T,(k + P)\Delta T]\), \(k\in\{0,\cdots,M\}\), \(x(t)\) and \(u(t)\) are the state variables and control variables of each wind farm subsystem respectively, and the system model is determined by the collaborative control prediction agent model of the offshore wind farm, and the obtained optimal control \(u(t)\) is executed and output within the control period \(t\in[k\Delta T,(k + L)\Delta T]\);

[0040] Among them, J is the objective function of the subsystem, ΔT is the sampling period, k represents the current sampling period, M is the number of sampling periods, P is the optimization time domain period, and L is the control time domain period; when performing the optimization solution of the above-mentioned i-th distributed local dynamic optimization subsystem in the k-th optimization period, other wind farm subsystems need to continuously iterate and provide their current optimal control as the initial value of the control input, and obtain the optimal control in the k-th optimization period through dynamic game optimization.

[0041] In step (5), the intelligent optimization algorithm includes genetic algorithm, particle swarm algorithm and game theory algorithm.

[0042] In step (6), the control parameters include but are not limited to pitch angle and yaw angle; by changing the control parameters of each unit through wind farm coordinated pitch control, torque control or yaw control, the real-time optimal scheduling of active power and the load reduction of the unit are realized within the control time domain L.

[0043] Compared with the prior art, the present invention has the following beneficial effects:

[0044] 1. The present invention completes the coordinated control of the offshore wind farm based on distributed rolling optimization, aiming to track the dynamic wind conditions in the wind farm, search for the optimal control solution in each optimization period and realize rolling optimization over time, so as to complete the intelligent control and optimization of the wind farm under complex environmental conditions.

[0045] 2. Aiming at the problems of high cost and great difficulty in centralized optimization calculation of large-scale offshore wind farms, the present invention introduces a distributed optimization scheduling method, divides the wind farm system into multiple subsystems according to the layout law of the units, uses the intelligent optimization algorithm to complete the scheduling problem of the subsystems, and realizes global optimization by continuously optimizing the local optimum, improving the calculation flexibility of the system optimization scheduling. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 is the system structure diagram of a field-level control strategy for an offshore wind farm based on distributed rolling optimization of the present invention;

[0047] Figure 2 is the flow chart of distributed rolling optimization in the present invention;

[0048] Figure 3 is the schematic diagram of window rolling optimization based on time for an offshore wind farm in the present invention;

[0049] Figure 4 is the schematic diagram of a coordinated control prediction agent model for a 3×4 array distributed offshore wind farm in an embodiment of the present invention;

[0050] Figure 5Schematic diagram of the comparison of the operating state results of the prediction proxy model for an offshore wind farm before and after optimal control within the unit optimization time domain period. Detailed implementation manner

[0051] The present invention will be further described in detail below with reference to the drawings and embodiments. It should be noted that the following embodiments are intended to facilitate the understanding of the present invention and do not limit it in any way.

[0052] As Figure 1 shown, a field-level control strategy for an offshore wind farm based on distributed rolling optimization divides the offshore wind farm to be optimized into multiple subsystems according to the spatial distribution, and uses a real-time sensing system to complete the measurement of the internal flow field distribution of the wind farm. Based on the measurement data of the wind farm, a prediction proxy model of the wind farm is established to complete the simulation prediction of the wind farm system under different working conditions, and the operating state characteristics of each unit in the wind farm are obtained, including the time-domain variation of power and load, etc. A multi-objective optimization function is established, and an intelligent optimization algorithm is used to complete the optimal control search within the unit optimization time domain period, and an actuator is used to complete the control output within the current control period. The specific steps are as follows:

[0053] Step (1), use the sensing system of the offshore wind farm to obtain the wind speed distribution of the wind farm flow field in real time, and record the dynamically changing environmental wind parameters and the operating state information of each unit at the current moment. The distribution of environmental wind information in the offshore wind farm has high uncertainty. In order to obtain the dynamic distribution information of the internal flow field of the wind farm, the present invention uses lidar wind measurement technology to perform three-dimensional wind field detection based on the Doppler effect and then inversely calculate the environmental wind condition parameters such as the average wind speed, wind direction, and turbulence intensity of the input environmental wind, that is, the wind information parameters without wake influence; measure the wind speed distribution in the wake area downstream of the wind turbine to obtain wake characteristic parameters, including parameters such as quantifying the wake wind speed deficit, wake expansion, and wake meandering; use the wind turbine SCADA system to obtain the state parameters of the wind turbine at the current moment, including the yaw angle, pitch angle, etc.

[0054] The real-time wind information parameters obtained by the above sensing system are used for the simulation prediction in steps (2) and (3). In the embodiment of the present invention, the input average wind speed of the offshore wind farm is selected as 8 m / s, the input environmental wind direction is 270°, and the average turbulence intensity is 6%.

[0055] Step (2), establish a collaborative control prediction proxy model for the offshore wind farm, specifically including an environmental wind sub-model, a wind turbine sub-model, a wake sub-model, etc., and use the data measured by the wind farm sensing system in step (1) to correct the parameters of the established prediction proxy model.

[0056] The above-mentioned collaborative control prediction agent model for an offshore wind farm is used to predict the power performance and structural loads of each wind turbine in the wind farm. The environmental wind sub-model uses the Taylor frozen turbulence hypothesis to complete the spatial propagation of the wind farm flow field. The wind turbine sub-model adopts a brake disc model, and parameters such as the rotor plane radius and hub height of the wind turbine are set according to the simulation requirements. The wake sub-model is used to capture and predict the key wake characteristics related to the output power of the wind farm and the load of the wind turbine, including wake wind speed deficit, wake expansion, and wake meandering, etc. Under quasi-steady conditions, the wind farm uses the thin shear layer approximation of the Reynolds-averaged Navier-Stokes equations to simulate the dynamic changes of the wake wind speed deficit, and the turbulence closure model uses the eddy viscosity equation to describe it.

[0057] In order to simulate and predict the dynamic operation characteristics of an actual offshore wind farm, such as Figure 4 As shown, in the embodiment of the present invention, a collaborative control prediction agent model for an offshore wind farm with a 3×4 array distribution is established. This wind farm model consists of 12 NREL 5MW wind turbine sub-models, which is used to predict the power performance and structural loads of each wind turbine in the wind farm. Specifically, a dynamic wake meandering model is used to capture and predict the key wake characteristics related to the output power of the wind farm and the load of the wind turbine, including wake wind speed deficit, wake expansion, and wake meandering, etc.

[0058] Step (3), use the wind farm prediction agent model established in step (2) to characterize and predict the operating state of the actual wind farm within a limited future time, and obtain the time-domain changes of the output power and the loads of key parts of each wind turbine generator in the wind farm under the influence of wake interaction, so as to complete the evaluation of the operating performance of the wind farm within a unit optimization time domain period.

[0059] Step (4), comprehensively consider the operating characteristics of the wind farm within a unit optimization time domain period, establish a multi-objective optimization function that minimizes the fatigue load under the condition of realizing active power optimal scheduling, and improve the active power optimal scheduling of the wind farm and the fatigue life of the components of the wind turbine by minimizing the objective function.

[0060] Specifically, calculate the output power of the i-th wind turbine in the wind farm, and the calculation formula is as follows:

[0061]

[0062] In the formula, P i is the output power (W) of the generator of the i-th wind turbine in the wind farm, T q,i (t) is the instantaneous torque (N-m) of the generator of the i-th wind turbine at time t, ω i is the instantaneous rotational speed (rpm) of the generator of the i-th wind turbine at time t, t kLet \(t\) be the current moment, and \(\Delta T\) be the unit optimization time domain period. In the embodiment of the present invention, the optimization time domain period \(\Delta T\) is selected to be 630 s so that the offshore wind farm system reaches a stable flow field state. In the present invention patent, the pitch angle and the tip speed ratio can be changed through pitch control and torque control, thereby optimizing the output power of the wind farm; or the ambient wind inflow angle can be changed through yaw control to optimize the operating performance of the wind farm.

[0063] According to the above power calculation formula of a single unit, the total generated power of the wind farm system is the sum of the generated powers of all wind turbines in the wind farm system within a unit time. It is:

[0064]

[0065] 4-1) Calculate the equivalent fatigue load on wind turbine \(i\):

[0066] Use the wind farm collaborative control prediction proxy model to obtain the time-domain variation of the loads on the key parts of wind turbine \(i\) in the wind farm, including the load variations at the bottom of the tower and the root of the blade. Through the rainflow counting method, the time-domain varying loads on wind turbine \(i\) are equivalently analyzed to obtain the equivalent fatigue load DEL of the key parts:

[0067]

[0068] In the formula, DEL is the equivalent fatigue load, is the equivalent cycle number within the time series \(\Delta T\), \(N\) ΔT,i is the number of occurrences of the \(i\)-th working condition within the time series \(\Delta T\), \(L\) ΔT,i is the load range of the \(i\)-th working condition within the time series \(\Delta T\), and \(m\) is the slope of the material S-N curve.

[0069] 4-2) Establish the objective function model for multi-objective optimization:

[0070] To achieve the optimal dispatching of the active power of the wind farm, make the output power of the wind farm meet the expected value of the active power, and reduce the fatigue load on the key parts of the unit, the following objective function is established:

[0071]

[0072] In the formula, \(P\) des is the expected value of the total output power of the wind farm system, is the total generated power of the wind farm system, is the normalized power deviation value, DEL norm is the maximum fatigue load on the key parts of each wind turbine after normalization, and \(\alpha\) is the weight coefficient. In this embodiment, the weight coefficient \(\alpha\) is selected to be 0.5.

[0073] Step (5), based on the optimized objective function established in step (4), decomposes the centralized dynamic multi-objective optimization problem of the offshore wind farm into several small-scale distributed local dynamic optimization sub-problems, uses intelligent optimization algorithms to solve the multi-objective optimization functions of each sub-problem, and comprehensively obtains the distributed global optimum through the communication between each problem, and finally obtains the optimal control within a unit optimization period.

[0074] With the construction of large-scale offshore wind farms with tens of millions and gigabits, the centralized management and optimization of wind farms often incur high computational costs. As Figure 2 shown, the present invention selects a part of the wind turbines with adjacent spatial arrangements in a large-scale offshore wind farm according to the spatial arrangement characteristics of the wind turbines, calculates the objective functions of each subsystem, and calculates the global optimum by solving the optimal control strategies of each subsystem. The system models of each subsystem are obtained from the wind farm agent prediction model.

[0075] Specifically, the embodiment of the present invention divides the collaborative control prediction agent model of the 3×4 array-distributed offshore wind farm established in step (2) into 4 subsystems according to the spatial arrangement, and each subsystem includes 3 wind turbine sub-models.

[0076] In the distributed optimization calculation, first fix the current optimal control and state of other subsystems as the initial value conditions, calculate and solve the minimization problem of the i-th (i = 1, 2, 3, 4) distributed subsystem, and then sequentially solve multiple linear minimization problems to update the optimal control of other subsystems. When the optimal control parameters of each subsystem obtained after all subsystems complete one iteration of solution are compared with the results of the previous iteration, if the difference between the optimal control obtained by a certain subsystem in the current iteration and the result obtained in the previous iteration is greater than the threshold, the non-converged subsystem enters the next iteration calculation process, and the converged subsystem no longer performs iterative solution, and updates the current solution order, and repeats the above optimization solution until the optimal control of all subsystems converges.

[0077] The above-mentioned small-scale distributed local dynamic optimization sub-problems jointly solve the optimized objective function established in step (4). Each distributed subsystem minimizes its own objective function, thereby achieving the optimal performance of the overall system and reducing the computational cost of the overall field-level optimization. Specifically, the expression form of the i-th distributed local dynamic optimization sub-problem within the optimization time t ∈ [kΔT, (k + P)ΔT] is:

[0078]

[0079] where \(t\in[k\Delta T,(k + P)\Delta T]\), \(k\in\{0,\cdots,M\}\), \(x(t)\) and \(u(t)\) are the state variables and control variables of each wind farm subsystem respectively, and the system model is determined by the wind farm simulation numerical model, and the obtained optimal control \(u(t)\) is executed and output within the control period \(t\in[k\Delta T,(k + L)\Delta T]\). Among them, \(J\) is the objective function of the subsystem, \(\Delta T\) is the sampling period, \(k\) represents the current sampling period, \(M\) is the number of sampling periods, \(P\) is the optimization time domain period, and \(L\) is the control time domain period. When optimizing and solving in the \(k\)-th optimization period of the \(i\)-th distributed local dynamic optimization subsystem above, other wind farm subsystems need to continuously iterate and provide their current optimal control as the initial value of the control input, and obtain the optimal control within the \(k\)-th optimization period through dynamic game optimization.

[0080] Based on the established optimization objective function of the distributed local dynamic optimization subsystem, use the intelligent optimization algorithm to solve the multi-objective optimization function of each sub-problem, and comprehensively obtain the distributed global optimum through the communication between each problem, and finally obtain the global optimal control within this optimization period. The embodiment of the present invention selects the genetic optimization algorithm to complete the search for the optimal solution of this example.

[0081] Step (6), send the system optimal control target sequence calculated by the intelligent optimization algorithm in step (5) to each fan, and the control parameters include but are not limited to pitch angle and yaw angle, etc. Change the control parameters of each unit through the coordinated pitch control, torque control or yaw control of the wind farm to achieve real-time optimal scheduling of active power within the control time domain \(L\) and load reduction of the unit. Among them, the constraint conditions of each control variable are:

[0082] 0° / s ≤ |ω yaw | ≤ 0.3° / s

[0083] 0° ≤ β ≤ 90°

[0084] 0° / s ≤ |ω pitch | ≤ 8° / s

[0085] T gen ≤ 47402.91 N·m

[0086] TrqRate ≤ 15000 N·m / s

[0087] In the formula, ω yaw is the yaw angular velocity of the unit nacelle, β is the pitch angle of the unit, ω pitch is the pitch angle angular velocity, T gen is the generator torque, and TrqRate is the generator torque change rate.

[0088] Step (7), taking the optimized time domain period ΔT as the time unit, as the sampling time of the wind farm sensing system progresses, repeat the above steps (1)-(5) for the online optimization control process, and execute and output the optimal control strategy within the current optimization period within the control time domain period, so as to achieve the multi-objective rolling optimization and optimal control of the wind farm under changing environmental wind conditions.

[0089] As Figure 3 shown, in the present invention, the objective function and the intelligent optimization algorithm are used to solve the optimal control within the unit optimization time domain interval of the offshore wind farm system. At the current time t k , the optimal control obtained in the first optimized time domain period is output for control. After the unit control period ΔT, new optimization calculations are started at t k + ΔT to solve the optimal control in the second optimized time domain period and output it for control, and so on. Among them, the prediction window step size of each optimized time domain period is ΔT P . Under the time-based periodic rolling mechanism, the number of rolling schedules and the adaptability of the wind farm system to dynamic factors are determined by ΔT, thereby completing the dynamic scheduling optimization of the offshore wind farm system under changing environmental wind conditions.

[0090] As Figure 5 shown, in the embodiment of the present invention, the optimized time domain period ΔT is selected to be 630 s so that the offshore wind farm system reaches a stable flow field state. The results show the time domain change performance parameters of the offshore wind farm model between 30 s and 630 s, specifically including the environmental wind input wind speed of the offshore wind farm prediction agent model, the total output power of the wind farm system, and the tower pitch moment and blade pitch moment received by the unit. Among them, the non-optimized control condition is the operation result of the wind farm system under the application of the greedy algorithm. Through comparison, it is found that under the same environmental wind input, the wind farm system reaches the wake stable state between 400 s and 600 s, and the optimized control can significantly achieve the optimized scheduling of power. In this embodiment, the power can be increased by 12.43%. The rain flow counting method is used to calculate the maximum fatigue load of the tower and blade of the unit within the unit optimized time domain period. From the results, it can be obtained that the maximum fatigue load of the tower can be reduced by 0.3%.

[0091] The above embodiments have described in detail the technical solutions and beneficial effects of the present invention. It should be understood that the above are only specific embodiments of the present invention and are not used to limit the present invention. Any modifications, supplements, and equivalent replacements made within the scope of the principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A field-level control strategy for an offshore wind farm based on distributed rolling optimization, characterized in that, it includes: (1) Using the wind farm perception system to obtain the internal flow field distribution of the wind farm in real time, and recording the dynamically changing environmental wind parameters and the current operating state information of each unit; (2) Establishing a predictive agent model for the collaborative control of the offshore wind farm, including an environmental wind sub-model, a wind turbine sub-model, and a wake sub-model; using the internal flow field distribution of the wind farm obtained by the wind farm perception system to correct the parameters of the predictive agent model; (3) Taking the environmental wind parameters and the current operating state information of each unit obtained in step (1) as inputs, using the predictive agent model to characterize and predict the operating characteristics of the actual wind farm at future times, and obtaining the time-domain changes of the output power and key part loads of each unit under different collaborative control strategies; (4) Considering comprehensively the operating characteristics of the wind farm within a unit optimization time domain period, establishing a multi-objective optimization function for minimizing fatigue loads under active power optimal scheduling, and realizing the active power optimal scheduling of the wind farm and the improvement of the fatigue life of the key components of the unit by minimizing the objective function; (5) Decomposing the multi-objective optimization problem into sub-problems of several small-scale distributed local dynamic optimization subsystems, using intelligent optimization algorithms to solve the objective functions of each sub-problem, and comprehensively obtaining the distributed global optimum through communication between the problems, and finally obtaining the optimal control within the optimization period; (6) Sending the obtained optimal control parameter target values of each unit to the actuators of each wind turbine, and realizing the dynamic optimal scheduling of the wind farm power and the fatigue load optimization within a future finite time domain through single control or joint control methods; (7) Repeating steps (1)-(5) in the next rolling optimization period, and executing the optimal control strategy in the current optimization period within its corresponding control period, so as to realize the multi-objective rolling optimization and optimal control of the wind farm under dynamically changing wind conditions.

2. The field-level control strategy for an offshore wind farm based on distributed rolling optimization according to claim 1, characterized in that, in step (1), the environmental wind parameters include wind information parameters under wake influence and without wake influence; the wind information parameters under wake influence include wake wind speed deficit, wake expansion, and wake meandering; they are obtained by measuring the wind speed distribution in the wake area downstream of the wind turbine; the wind information parameters without wake influence include the average wind speed, wind direction, and turbulence intensity of the environmental wind; they are obtained by using lidar wind measurement technology and inversely calculating based on scanning the spatial wind speed distribution of the wind farm flow field in real time using the Doppler principle; the current operating state information of each unit includes the yaw angle and pitch angle, which are obtained by using the wind turbine SCADA system.

3. The field-level control strategy for an offshore wind farm based on distributed rolling optimization according to claim 1, characterized in that, In step (2), the environmental wind sub-model uses the Taylor frozen-turbulence hypothesis to complete the spatial propagation of the wind farm flow field; the wind turbine sub-model adopts a brake disc model, and sets parameters of the rotor plane radius and hub height of the unit according to simulation requirements; the wake sub-model is used to capture and predict key wake characteristics related to the wind farm output power and wind turbine loads, including wake wind speed deficit, wake expansion, and wake meandering.

4. The field-level control strategy for an offshore wind farm based on distributed rolling optimization according to claim 1, characterized in that in step (4), the specific process of establishing the multi-objective optimization function is as follows: (4-1) Calculate the output power of unit i in the wind farm, and the calculation formula is as follows: Wherein, P i is the output power of the generator of unit i in the wind farm, T q,i (t) is the instantaneous torque of the generator of unit i at time t, ω i is the instantaneous speed of the generator of unit i at time t, t k is the current time, and ΔT is the unit optimization time domain period; According to the power calculation formula for a single unit above, the total power generation of the wind farm system is solved, which is the sum of the power generation of all wind turbines in the wind farm system per unit time. It is: (4-2) Calculate the equivalent fatigue load on wind turbine generator set i: Use the prediction proxy model to obtain the time-domain variation of the loads on key parts of wind turbine generator set i in the wind farm, including the load variations at the bottom of the tower and the root of the blade; through the rain-flow counting method, perform equivalent analysis on the time-domain varying loads on wind turbine generator set i to obtain the equivalent fatigue load DEL of the key parts: where DEL is the equivalent fatigue load, is the equivalent number of cycles within the time series ΔT, N ΔT,i is the number of occurrences of the i-th working condition within the time series ΔT, L ΔT,i is the load range of the i-th working condition within the time series ΔT, and m is the slope of the material S-N curve; (4-3) Establish the objective function model for multi-objective optimization: To achieve the optimal scheduling of the active power of the wind farm, make the output power of the wind farm meet the expected value of the active power, and reduce the fatigue loads on the key parts of the unit, the following objective function is established: Wherein, P des is the expected value of the total output power of the wind farm system, is the total power generation of the wind farm system, is the power deviation value after normalization, DEL norm is the maximum fatigue load on the key parts of each wind turbine after normalization, and α is the weight coefficient.

5. The field-level control strategy for an offshore wind farm based on distributed rolling optimization according to claim 1, characterized in that in step (5), the solution process of the sub-problems of several small-scale distributed local dynamic optimization subsystems sequentially completes the solution of all sub-problems in a specific order. When the optimal control parameters of each subsystem obtained after all subsystems complete one iteration of solution are compared with the results of the previous iteration, if the difference between the optimal control obtained in the current iteration of a certain subsystem and the result obtained in the previous iteration is greater than the threshold, the non-converged subsystem enters the next iteration calculation process, and the converged subsystems no longer perform iterative solution, and update the current solution order, and repeat the above optimization solution until the optimal control of all subsystems converges.

6. The field-level control strategy for an offshore wind farm based on distributed rolling optimization according to claim 1, characterized in that in step (5), the expression form of the i-th distributed local dynamic optimization sub-problem within the optimization time t∈[kΔT,(k + P)ΔT] is: where \(t\in[k\Delta T,(k + P)\Delta T]\), \(k\in\{0,\ldots,M\}\), \(x(t)\) and \(u(t)\) are the state variables and control variables of each wind farm subsystem respectively, and the system model is determined by the cooperative control prediction agent model of the offshore wind farm, and the obtained optimal control \(u(t)\) is executed and output within the control period \(t\in[k\Delta T,(k + L)\Delta T]\); where J is the objective function of the subsystem, ΔT is the sampling period, k represents the current sampling period, M is the number of sampling periods, P is the optimization time domain period, and L is the control time domain period; when solving the optimization of the i-th distributed local dynamic optimization subsystem in the k-th optimization period, it is necessary for other wind farm subsystems to continuously iterate and provide their current optimal control as the initial value of the control input, and obtain the optimal control in the k-th optimization period through dynamic game optimization.

7. The field-level control strategy for an offshore wind farm based on distributed rolling optimization according to claim 1, characterized in that in step (5), the intelligent optimization algorithms include genetic algorithms, particle swarm algorithms, and game theory algorithms.

8. The field-level control strategy for an offshore wind farm based on distributed rolling optimization according to claim 1, characterized in that, in step (6), the control parameters include but are not limited to pitch angle and yaw angle; by means of wind farm coordinated pitch control, torque control or yaw control, the control parameters of each unit are changed to achieve real-time optimal scheduling of active power within the control time domain L and load reduction of the units.

Images