Wind farm fatigue suppression active control method based on data mechanism hybrid modeling

Abstract

The application provides a wind farm fatigue suppression active control method based on data mechanism hybrid modeling, and belongs to the technical field of wind farms.The method comprises the following steps: a continuous state space model is established, a sampling period is used to discretize the continuous state space model, a discrete state space model is obtained, a threshold value of a switching control strategy of a wind speed region is calculated based on the discrete state space model, and then a prediction model of a main shaft torque and a prediction model of a tower thrust are obtained, which are denoted as a wind turbine linear prediction model; an equivalent fatigue load linearization model is obtained through main shaft torque time series data; after the equivalent fatigue load linearization model and the wind turbine linear prediction model are combined, a data mechanism double-driven fatigue load model is obtained; and the data mechanism double-driven fatigue load model is subjected to active optimization control solving, so that an optimal solution of active power distribution that meets the safety and stability constraints of the wind farm is obtained.

Description

Technical Field

[0001] The present invention belongs to the technical field of wind farms, and in particular relates to a wind farm fatigue suppression active power control method based on data mechanism hybrid modeling. Background Art

[0002] As wind power penetration continues to increase, the weak support and low inertia characteristics of wind power systems pose a significant challenge to the security and stability of the power grid. Wind power integration into the grid requires both inertia and primary frequency response capabilities. However, during frequency regulation, frequent fluctuations in output active power from wind farms increase fatigue loads on mechanical components such as the wind turbine shaft system and tower, leading to increased maintenance costs and decreased operational efficiency. Currently, wind farms primarily respond to system frequency fluctuations through a centralized, whole-station response approach. This involves using wind farm control systems to generate active power deviations based on frequency response strategies and distributing them to the turbines within the station.

[0003] Existing active power allocation methods rely primarily on the available power ratio of each wind turbine, ignoring the specific wind speed conditions and control inertia at the turbine's location. This approach not only fails to effectively optimize the allocation of wind farm resources, but also may further increase turbine wear, thereby affecting the long-term operational efficiency and economic viability of the entire wind farm.

[0004] To this end, the present invention proposes a wind farm fatigue suppression active power control method based on data mechanism hybrid modeling. Summary of the Invention

[0005] The present invention provides a wind farm fatigue suppression active power control method based on data mechanism hybrid modeling, so as to at least solve the problem that the existing active power distribution method cannot effectively utilize the optimal configuration of resources in the wind farm, aggravates the wear of the units, and thus affects the long-term operating efficiency and economy of the entire wind farm.

[0006] The present application provides a method for controlling active power of a wind farm based on data-mechanism hybrid modeling for fatigue suppression, the method comprising:

[0007] Step S1: establishing a continuous state space model, discretizing the continuous state space model using a sampling period to obtain a discrete state space model, and calculating the threshold of the switching control strategy of the wind speed region of the control system of the wind turbine generator system based on the discrete state space model;

[0008] Step S2: Based on the threshold of the switching control strategy in the wind speed region of the control system of the wind turbine generator system, a prediction model of the main shaft torque and a prediction model of the tower thrust are obtained, which are recorded as the wind turbine generator system linear prediction model;

[0009] Step S3: Obtain an equivalent fatigue load linearization model through the main shaft torque time series data, and combine the equivalent fatigue load linearization model with the wind turbine linear prediction model to obtain a fatigue load model driven by both data and mechanism;

[0010] Step S4: performing active power optimization control on the fatigue load model driven by both data and mechanism to obtain an optimal solution for active power distribution that satisfies the safety and stability constraints of the wind farm.

[0011] Furthermore, in step S1, a continuous state space model is established, specifically including:

[0012] Step S11: establishing a nonlinear mechanism model of the wind turbine;

[0013] Step S12: constructing a continuous state space model based on the nonlinear mechanism model of the wind turbine generator system.

[0014] Furthermore, in step S11, a nonlinear mechanism model of the wind turbine is established, including:

[0015] Step S111: Establish a derivative equation of the generator speed increment, which is expressed as follows:

[0016]

[0017] Where, represents the derivative of the generator speed increment, Indicates the gearbox transmission ratio, represents the equivalent mass of the blade mass and the engine mass combined, represents the generator torque increment, represents the aerodynamic torque increment, Indicates The generator torque measured at the moment, Indicates The aerodynamic torque measured at all times;

[0018] Step S112: Establish a derivative equation of the filtered generator speed increment, which is expressed as follows:

[0019]

[0020] Where, represents the derivative of the filtered generator speed increment, represents the filter time constant, represents the generator speed increment after filtering, Indicates the generator speed increment;

[0021] Step S113: Define increment get:

[0022]

[0023] Where, represents the pitch angle, represents the generator speed after filtering, Indicates the rated speed of the generator, represents the proportional gain of the fan PI controller, represents the integral gain of the fan PI controller, represents the Laplace operator;

[0024] Step S114: Based on the derivative equation of the generator speed increment after filtering established in step S112 and the increment obtained in step S113 , get the increment The derivative equation of :

[0025]

[0026] (4)

[0027] Where, Indicates increment The derivative of

[0028] Step S115: performing approximate calculation at the working point according to the Taylor series to obtain the aerodynamic torque increment and the generator torque increment;

[0029] The expression of aerodynamic torque increment is:

[0030]

[0031] Where, Indicates the The increase in the active power reference value of the typhoon generator;

[0032] The expression of the generator torque increment is:

[0033] (6)

[0034] Where, Indicates Generator torque at point Generator speed The partial derivative of is expressed as:

[0035] (7)

[0036] (8)

[0037] Where, , Indicates The wind speed value at the blade at the moment, Indicates leaves, express about function, represents the wind energy utilization coefficient;

[0038] Indicates Generator torque at point For increment The partial derivative of is expressed as:

[0039] (9)

[0040] Where, represents the air density, Indicates leaves;

[0041] Step S116: Calculation and ;

[0042] The expression is:

[0043]

[0044]

[0045] represents the tip speed ratio, represents the pitch angle, Represents the adjacent unit The difference, Represents the adjacent unit The difference, represents the wind energy utilization coefficient, represents the wind energy utilization coefficient at (n,m) in the lookup table, represents the wind energy utilization coefficient at (n,m+1) in the lookup table;

[0046] The expression is:

[0047] (11)

[0048] Where, Represents the wind energy utilization coefficient at (n+1,m) in the lookup table.

[0049] Furthermore, in step S12, a continuous state space model is constructed based on the nonlinear mechanism model of the wind turbine, including:

[0050] Step S121: Construct a continuous state space model under high wind speed conditions, the expression of which is:

[0051] (12)

[0052]

[0053]

[0054]

[0055] ;

[0056] Step S122: Based on the continuous state space model under high wind speed conditions, ignoring the terms related to the pitch angle, a continuous state space model under low wind speed conditions is obtained, which is expressed as:

[0057]

[0058]

[0059]

[0060] .

[0061] Furthermore, in step S1, the continuous state space model is discretized using a sampling period to obtain a discrete state space model, including:

[0062] Sampling period Discretization processing is performed on the continuous state space model under low wind speed and the continuous state space model under high wind speed respectively;

[0063] The continuous state space model under low wind speed is discretized to obtain the discrete state space model under low wind speed, which is expressed as follows:

[0064]

[0065]

[0066]

[0067] (13)

[0068] Where, represents the step index, for The state variables at time , for The state variables at time , 、 、 are all discrete state space matrices after sampling;

[0069] The continuous state space model under high wind speed is discretized to obtain the discrete state space model under high wind speed, which is expressed as follows:

[0070]

[0071]

[0072]

[0073] (14).

[0074] Furthermore, in step S1, based on the discrete state space model, the threshold of the switching control strategy of the wind speed region of the control system of the wind turbine generator system is calculated, which specifically includes:

[0075] From the low-speed shaft motion equation:

[0076]

[0077] Where, is the derivative of the rotor speed, and we can further obtain:

[0078]

[0079]

[0080]

[0081] therefore,

[0082]

[0083] When the current operating point is in the high wind speed area, that is, , Will follow the Active power reference value increment of typhoon generator decreases with the increase of , the control system of the wind turbine will transition to the low wind speed area. The threshold of the switching control strategy in this operating point area is , the calculation process is as follows:

[0084] According to formula (13), we can obtain:

[0085]

[0086] in and They are and The matrix elements of ,get:

[0087]

[0088] according to , it can be deduced that , whose expression is:

[0089]

[0090] When the current working point is in the low wind speed area, that is, hour, Will follow decreases and increases; if , the wind turbine control system will transition to high wind speed, and the threshold of the switching control strategy in the area where this operating point is located is , the calculation process is as follows:

[0091] According to formula (14),

[0092]

[0093] In the formula and They are and The matrix elements of Calculated , whose expression is:

[0094] .

[0095] Furthermore, in step S2, based on the threshold of the switching control strategy of the wind speed region of the control system of the wind turbine generator system, a prediction model of the main shaft torque and a prediction model of the tower thrust are obtained, including:

[0096] Calculate the spindle torque in the following four cases Increment of active power reference value The partial derivative of :

[0097] if , the wind turbine control system will stay at high wind speed. According to equations (5) and (6), equation (17) is transformed into:

[0098]

[0099]

[0100]

[0101] Based on formula (13) and formula (23), we can get:

[0102]

[0103] therefore,

[0104]

[0105] in, , and The subscript of represents the case index;

[0106] if , the wind turbine control system will stay at low wind speed. According to equations (5) and (6), equation (17) is transformed into:

[0107]

[0108] in,

[0109]

[0110]

[0111] According to formula (14) and formula (26), we can get:

[0112]

[0113] therefore,

[0114]

[0115] in ;

[0116] if , the wind turbine control system will transition from high wind speed to low wind speed. In this case, It is divided into two parts: and , Some operate in high wind speeds, while Some operate in low wind speeds;

[0117] Therefore, Approximate calculations yield:

[0118] (29)

[0119] therefore,

[0120]

[0121]

[0122] ;

[0123] If , the wind turbine control system will transit from low wind speed to high wind speed, is divided into two parts: and , part runs in low wind speed, part runs in high wind speed;

[0124] Therefore, for approximate calculation, we get:

[0125] (31)

[0126] Therefore,

[0127]

[0128]

[0129] .

[0130] Further, in step S2, based on the threshold of the wind speed region switching control strategy of the wind turbine control system, the prediction model of the main shaft torque and the prediction model of the tower thrust are obtained, and further comprising:

[0131] The tower thrust is calculated respectively under the following four conditions The partial derivative of the active power reference value increment :

[0132] If , the wind turbine control system remains in high wind speed, the increment of the tower thrust is calculated, and its expression is:

[0133]

[0134] ;

[0135] Based on equation (24) and equation (44), we get:

[0136]

[0137] Therefore,

[0138]

[0139] wherein, , and​ The subscript of represents the case index;

[0140] if , the wind turbine control system stays at low wind speed and calculates the tower thrust The increment of is expressed as:

[0141]

[0142] ;

[0143] Based on formula (14) and formula (36), we get:

[0144]

[0145] therefore,

[0146]

[0147] in, ;

[0148] if , the wind turbine control system transitions from high wind speed to low wind speed, Approximate calculations yield:

[0149] (39)

[0150] therefore,

[0151]

[0152] Where, ;

[0153] if , the wind turbine control system transitions from low wind speed to high wind speed, Approximate calculations yield:

[0154] (41)

[0155] therefore,

[0156]

[0157] Where, ;

[0158] According to formula (33) and formula (36), and They are and The function is:

[0159]

[0160]

[0161] function is nonlinear and is described by a lookup table derived from the blade geometry, whose input is and ;

[0162] according to and , determine the operating point by looking up the table The power coefficient at ,in, and are the corresponding row index and column index respectively;

[0163] calculate and :

[0164]

[0165]

[0166]

[0167]

[0168] The prediction model of spindle torque is:

[0169] The prediction model of tower thrust is:

[0170]

[0171] in, is the active power reference value of the wtth wind turbine, is the current active power output of the wind turbine, 、 、 、 All of them are derived from the formulas above. is the current tower thrust, is the current spindle torque, is the increment of the active power reference value of the wtth wind turbine.

[0172] Furthermore, in step S3, an equivalent fatigue load linearization model is obtained through the main shaft torque time series data. After combining the equivalent fatigue load linearization model with the wind turbine linear prediction model, a fatigue load model driven by both data and mechanism is obtained, which specifically includes:

[0173] The spindle torque time series data is generated by the spindle torque prediction model. The spindle torque time series data is sliced ​​with a sliding window of 10 to obtain the spindle torque time series data set. and , The data before the 9 seconds of the time window is used as historical input data. The 10th second data of the time window is used as the prediction data. The load cycle amplitude and mean of the main shaft torque data of each sliding window are calculated by the rain flow counting method to obtain the equivalent fatigue load:

[0174]

[0175] Where, is the load amplitude at each level, is the number of cycles of this level of load, m is the Wohler index of the SN curve, and N depends on the design life of the wind turbine. is the equivalent fatigue load of this load signal, and the equivalent fatigue load time series is obtained ;

[0176] Constructing the observation function , observation function Including the cubic polynomial basis function, the random Fourier basis function with fundamental frequencies of 0.1 and 0.01, and the delayed basis function with 10 steps of delay, the following matrix is ​​constructed:

[0177]

[0178]

[0179]

[0180] The state equation is constructed by Koopman operator, and its expression is:

[0181]

[0182] The Koopman matrices A and B are calculated using the extended dynamic modal decomposition (EDMD) method, namely:

[0183]

[0184] Vectorizing the optimization problem yields:

[0185]

[0186] in, is the Kronecker product, and by solving the pseudo-inverse, we get:

[0187]

[0188] Construct the output equation:

[0189]

[0190]

[0191]

[0192] in, is the error term, and the Koopman matrix C and D are calculated using the extended dynamic mode decomposition (EDMD) method, namely:

[0193]

[0194] By solving the pseudo-inverse, we get:

[0195]

[0196] Thus, the equivalent fatigue load linearization model based on the data-driven Koopman operator modeling is obtained, and its expression is:

[0197] After combining the equivalent fatigue load linearization model with the wind turbine linear prediction model, a data-mechanism dual-driven fatigue load model is obtained, and the model linearity is maintained. The input of the data-mechanism dual-driven fatigue load model is: wind speed, generator speed, output power, pitch angle, generator torque, active power reference, main shaft torque and tower thrust, and the output is the equivalent fatigue load.

[0198] Furthermore, in step S4, active power optimization control is performed on the fatigue load model driven by both data and mechanism to obtain an optimal solution for active power distribution that satisfies the safety and stability constraints of the wind farm, specifically including:

[0199] The optimization objective of active load shedding is constructed as follows:

[0200]

[0201] in, The equivalent fatigue load of each wind turbine is calculated based on the fatigue load model driven by the data mechanism. Obtained from the lookup table constructed later, the optimization problem constraints are:

[0202]

[0203] in, is the active power reference value, and are the maximum available power and minimum available power of a single machine, is the grid dispatch instruction, is the change in active power within 1 time step, is the maximum power ramp constraint for a single machine;

[0204] The data-mechanism dual-driven fatigue load model is used as a prediction model, and a distributed model predictive control algorithm is used to optimize and solve the data-mechanism dual-driven fatigue load model. The solution algorithm is a trust region constraint algorithm.

[0205] The central controller calculates the active power reference of each wind turbine in the wind farm using a traditional proportional allocation algorithm as the initial point. At the same time, the distributed controller of each wind turbine generates a wind turbine fatigue load lookup table based on the prediction model at a scale of 10W and sends it to the central controller. The central controller solves the optimization problem using a trust region constraint algorithm and sends the control strategy to each wind turbine controller.

[0206] The iterative steps of the trust region constraint algorithm specifically include:

[0207] Step S41: Set the initial iteration point according to the average distribution strategy , and set the initial trust domain radius , define the convergence threshold , gradient norm Stop iteration when

[0208] Step S42: At the current iteration point , construct a sub-problem including the objective function and constraints. The objective function uses a quadratic approximation model, and its expression is:

[0209]

[0210] Where, is the Hessian matrix, This is a trial step, and the constraints should be satisfied at this time:

[0211]

[0212] The constraints are linearized by Taylor's first-order expansion, and the expression is:

[0213]

[0214] ;

[0215] Step S43: Construct Lagrangian function:

[0216]

[0217] in, and is a Lagrange multiplier, and , by solving the partial derivatives of the Lagrangian function and combining the KKT conditions, the solution of the subproblem is obtained. The KKT conditions are:

[0218]

[0219] Step S44: Calculate the actual descent amount and predicted decline , and calculate the ratio of the actual decline to the predicted decline, the expression is:

[0220]

[0221] The ratio of the actual decline to the predicted decline is used to measure the degree of match between the predicted decline of the forecast model and the actual decline;

[0222] Step S45: Update the trust zone radius:

[0223] if , then the prediction model is considered to be of good quality, and the radius of the trust region is increased: ;

[0224] if , then the prediction model is considered to be of poor quality, and the trust region radius is reduced: ;

[0225] Otherwise, the trust region radius remains unchanged;

[0226] Step S46: Update the iteration point:

[0227] if , then accept the update, p, otherwise refuse to update, and stop the iteration when the iteration end condition is met;

[0228] After completing the iteration, the optimal solution for active power distribution that meets the safety and stability constraints of the wind farm is obtained.

[0229] It can be seen from the above technical solutions that the present invention has the following advantages:

[0230] In the wind farm fatigue suppression active power control method based on data-mechanism hybrid modeling provided in this application, under the premise of ensuring the safe and stable operation of the wind farm, the active power is accurately distributed according to the working status of each wind turbine and the specific wind speed and wind conditions, and the distributed model predictive control algorithm is used to optimize and solve the fatigue load model driven by both data and mechanism to meet the requirements of real-time calculation, thereby achieving an effective reduction in the fatigue load of the entire field.

[0231] The present invention introduces a spindle torque linearization prediction model based on physical mechanisms and a fatigue load calculation method based on data-driven methods, transforming the non-convex optimization problem into an easy-to-solve convex optimization problem, thereby significantly reducing the fatigue damage risk of wind turbines while ensuring the efficient operation of the wind farm.

[0232] The distributed controller generates a lookup table to accelerate the solution of the convex optimization problem, and the fatigue load of the entire wind farm is suppressed while meeting the safety and stability constraints of the wind farm. BRIEF DESCRIPTION OF THE DRAWINGS

[0233] In order to more clearly illustrate the technical solution of the present application, the following is a brief introduction to the drawings required for the description. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0234] Figure 1 This is a flow chart of the wind farm fatigue suppression active power control method based on data-mechanism hybrid modeling.

[0235] Figure 2 This is a diagram showing the predicted effect of the main shaft torque of the wind farm fatigue suppression active power control method based on data-mechanism hybrid modeling.

[0236] Figure 3 This is a diagram showing the prediction effect of the equivalent fatigue load linearization model of the wind farm fatigue suppression active power control method based on data-mechanism hybrid modeling. DETAILED DESCRIPTION

[0237] In order to make the application objectives, features, and advantages of this application more obvious and easy to understand, the technical solutions protected by this application will be clearly and completely described below using specific embodiments and drawings. Obviously, the embodiments described below are only part of the embodiments of this application, not all of them. Based on the embodiments in this patent, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this patent.

[0238] The technical solutions proposed in the embodiments of the present application are described in detail below with reference to the accompanying drawings.

[0239] Figure 1 This is a flow chart of a wind farm fatigue suppression active power control method based on data mechanism hybrid modeling provided in an embodiment of the present application. Figure 1 As shown, the embodiment of the present application provides a wind farm fatigue suppression active power control method based on data mechanism hybrid modeling, which specifically includes the following steps:

[0240] Step S1: Based on the findings from the previous investigation, the main shaft torque and tower thrust are the fatigue loads that cause wind turbine failures. Therefore, a continuous state space model is established, and the continuous state space model is discretized using a sampling period to obtain a discrete state space model. Based on the discrete state space model, the threshold of the switching control strategy of the wind speed area of ​​the control system of the wind turbine is calculated, that is, the sensitivity of the main shaft torque and the sensitivity of the tower thrust. The present invention calculates the sensitivity of the main shaft torque and the sensitivity of the tower thrust through an incremental state space model and discretization processing, and obtains the threshold of the switching control strategy in the wind speed area, thereby accurately calculating the changes in the main shaft torque and tower thrust of the wind turbine under different wind speed conditions, and can more accurately predict and control the fatigue load of the wind turbine in the process of participating in the grid frequency regulation, thereby reducing maintenance costs and extending the service life of the wind turbine.

[0241] Step S2: Based on the threshold of the switching control strategy in the wind speed region of the control system of the wind turbine generator system, a prediction model of the main shaft torque and a prediction model of the tower thrust are obtained, which are recorded as the wind turbine generator system linear prediction model;

[0242] Step S3: Obtain an equivalent fatigue load linearization model through the main shaft torque time series data, and combine the equivalent fatigue load linearization model with the wind turbine linear prediction model to obtain a fatigue load model driven by both data and mechanism;

[0243] Step S4: performing active power optimization control on the fatigue load model driven by both data and mechanism to obtain an optimal solution for active power distribution that satisfies the safety and stability constraints of the wind farm.

[0244] Set the working point time to , wind speed is a variable that can be measured and estimated. An estimate was made, The wind speed at the blade at this moment is , and is assumed to remain constant over a short control period. The power output of the generator measured at any moment ,exist Generator speed measured at all times ,exist The filtered generator speed measured at the moment ,exist Pitch angle measured at any moment , based on The wind speed at that moment is ,exist The power output of the generator measured at any moment ,exist Generator speed measured at all times ,exist The filtered generator speed measured at the moment ,exist Pitch angle measured at any moment , can be obtained in Pneumatic torque at time and in Generator torque at time .

[0245] Based on the linearization of the simplified nonlinear wind turbine model at the operating point, the incremental state-space model for high wind speeds is derived through the following steps. Terms related to the pitch angle are ignored, and the incremental state-space model for low wind speeds is derived from the incremental state-space model for high wind speeds. The symbol Δ represents the increment of a variable.

[0246] In step S1, a continuous state space model is established, which specifically includes:

[0247] Step S11: establishing a nonlinear mechanism model of the wind turbine;

[0248] Step S12: constructing a continuous state space model based on the nonlinear mechanism model of the wind turbine generator system.

[0249] In step S11, a nonlinear mechanism model of the wind turbine is established, including:

[0250] Step S111: Establish a derivative equation of the generator speed increment, which is expressed as follows:

[0251]

[0252] Where, represents the derivative of the generator speed increment, Indicates the gearbox transmission ratio, represents the equivalent mass of the blade mass and the engine mass combined, represents the generator torque increment, represents the aerodynamic torque increment, Indicates The generator torque measured at the moment, Indicates The aerodynamic torque measured at all times;

[0253] Step S112: Establish a derivative equation of the filtered generator speed increment, which is expressed as follows:

[0254]

[0255] Where, represents the derivative of the filtered generator speed increment, represents the filter time constant, represents the generator speed increment after filtering, Indicates the generator speed increment;

[0256] Step S113: Define increment get:

[0257]

[0258] Where, represents the pitch angle, represents the generator speed after filtering, Indicates the rated speed of the generator, represents the proportional gain of the fan PI controller, represents the integral gain of the fan PI controller, represents the Laplace operator;

[0259] Step S114: Based on the derivative equation of the generator speed increment after filtering established in step S112 and the increment obtained in step S113 , get the increment The derivative equation of :

[0260]

[0261] (4)

[0262] Where, Indicates increment The derivative of

[0263] The variables considered in the incremental state space model include: wind speed , power output , generator speed , filtered speed and pitch angle .

[0264] Step S115: Wind turbine generator system The calculation is nonlinear, and the aerodynamic torque increment and generator torque increment are obtained by approximate calculation at the working point according to Taylor series.

[0265] The expression of aerodynamic torque increment is:

[0266]

[0267] Where, Indicates the The increase in the active power reference value of the typhoon generator;

[0268] The expression of the generator torque increment is:

[0269] (6)

[0270] wherein, represents the wind speed value at the blade at the time instant t, represents the derivative of the generator torque with respect to the increment of the generator speed

[0271] (7)

[0272] (8)

[0273] wherein, , represents the wind speed value at the blade at the time instant t, represents the blade, represents the function of the wind energy exploitation coefficient; represents the derivative of the generator torque

[0274] with respect to the increment of the generator speed , whose expression is: (9)

[0275] wherein, represents the air density,

[0276] represents the blade; Step S116: calculating and

[0277] ;

[0278] whose expression is:

[0279]

[0280]

[0281] represents the tip speed ratio, represents the pitch angle, represents the difference of the of the adjacent cells, represents the difference of the of the adjacent cells, represents the wind energy exploitation coefficient, represents the wind energy exploitation coefficient at (n,m) in the lookup table,​​ represents the wind energy utilization coefficient at (n,m+1) in the lookup table;

[0282] The expression is:

[0283] (11)

[0284] Where, represents the wind energy utilization coefficient at (n+1,m) in the lookup table;

[0285] is a nonlinear function, The lookup table is generated based on the simulation results of the NREL 5MW wind turbine high-fidelity model. The input parameters are and , adjacent units The difference is , adjacent units The difference is ,according to and , working point The power coefficient at can be obtained by looking up the table, that is, ,in and are the corresponding row and column indices respectively.

[0286] In step S12, a continuous state space model is constructed based on the nonlinear mechanism model of the wind turbine, including:

[0287] Step S121: Based on formulas (1)-(11), a continuous state space model is constructed under high wind speed conditions, which is expressed as follows:

[0288] (12)

[0289]

[0290]

[0291] ;

[0292] Step S122: Based on the continuous state space model under high wind speed conditions, ignoring the terms related to the pitch angle, a continuous state space model under low wind speed conditions is obtained, which is expressed as:

[0293]

[0294]

[0295]

[0296] .

[0297] In step S1, the continuous state space model is discretized using a sampling period to obtain a discrete state space model, including:

[0298] Sampling period Discretization is performed on the continuous state space model under low wind speed and the continuous state space model under high wind speed respectively;

[0299] The continuous state space model under low wind speed is discretized to obtain the discrete state space model under low wind speed, which is expressed as follows:

[0300]

[0301]

[0302]

[0303] (13)

[0304] Where, represents the step index, for The state variables at time , for The state variables at time , 、 、 are all discrete state space matrices after sampling;

[0305] The continuous state space model under high wind speed is discretized to obtain the discrete state space model under high wind speed, which is expressed as follows:

[0306]

[0307]

[0308]

[0309] (14).

[0310] In step S1, based on the discrete state space model, the threshold value of the switching control strategy of the wind speed area of ​​the wind turbine control system is calculated, specifically including: calculating the threshold value of the switching control strategy of the wind turbine control system transitioning to the low wind speed area , the threshold value of the control system of the wind turbine to switch the control strategy to the high wind speed area ;

[0311] From the low-speed shaft motion equation:

[0312]

[0313] Where, is the derivative of the rotor speed, and we can further obtain:

[0314]

[0315]

[0316]

[0317] therefore,

[0318]

[0319] During the sampling period, the wind speed condition of the wind turbine may switch between high wind speed and low wind speed. When the current operating point is in the high wind speed area, that is, , Will follow the Active power reference value increment of typhoon generator decreases with the increase of , the control system of the wind turbine will transition to the low wind speed area. The threshold of the switching control strategy in this operating point area is , the calculation process is as follows:

[0320] According to formula (13), we can obtain:

[0321]

[0322] in and They are and The matrix elements of ,get:

[0323]

[0324] according to , it can be deduced that , whose expression is:

[0325]

[0326] It should be noted that the control system has two control strategies, one is high wind speed and the other is low wind speed. When the current operating point is in the high wind speed area, that is, , Will follow the The active power reference value increment of the wind turbine is reduced with the increase of the wind speed , the control system of the wind turbine will execute the control strategy of the low wind speed condition.

[0327] It should be further explained that the operating point refers to the operating point of switching the two control strategies of the wind turbine control system, i.e., the operating point of switching the high wind speed operating strategy and the low wind speed operating strategy. The wind turbine control system, in the case that other state quantities are determined, switches the control strategy only in relation to , and the threshold value of switching the control strategy is .

[0328] When the current operating point is located in the low wind speed region, i.e., , the will increase with the decrease of the wind speed ; if , the wind turbine control system will transit to the high wind speed, and the threshold value of the switching control strategy of the region where the operating point is located is , and the calculation process is as follows:

[0329] According to formula (14),

[0330]

[0331] In the formula, and are matrix elements of and , respectively, and is calculated to obtain , and the expression is as follows:

[0332] .

[0333] In step S2, based on the threshold value of the switching control strategy of the wind speed region of the control system of the wind turbine, the prediction model of the main shaft torque and the prediction model of the tower thrust are obtained, including:

[0334] The main shaft torque is calculated in the following four cases: The partial derivative of the active power reference value increment :

[0335] Case 1: high wind speed high wind speed.

[0336] If , the wind turbine control system will stay in the high wind speed, and according to formula (5) and formula (6), formula (17) is converted into:

[0337]

[0338]

[0339]

[0340] Based on formula (13) and formula (23), we can get:

[0341]

[0342] therefore,

[0343]

[0344] in, , and The subscript of represents the case index;

[0345] Case 2: Low wind speed Low wind speed.

[0346] if , the wind turbine control system will stay at low wind speed. According to equations (5) and (6), equation (17) is transformed into:

[0347]

[0348] in,

[0349]

[0350]

[0351] According to formula (14) and formula (26), we can get:

[0352]

[0353] therefore,

[0354]

[0355] in ;

[0356] Case 3: High wind speed Low wind speed.

[0357] if , the wind turbine control system will transition from high wind speed to low wind speed. In this case, It is divided into two parts: and , Some operate in high wind speeds, while Some operate in low wind speeds;

[0358] Therefore, Approximate calculations yield:

[0359] (29)

[0360] therefore,

[0361]

[0362]

[0363] ;

[0364] Case 4: Low wind speed High wind speed.

[0365] if , the wind turbine control system will transition from low wind speed to high wind speed, It is divided into two parts: and , Some operate in low wind speeds, Some operate in high wind speeds;

[0366] Therefore, Approximate calculations yield:

[0367] (31)

[0368] therefore,

[0369]

[0370]

[0371] .

[0372] For the sake of generality, omit and The subscript of , according to the measured value, is calculated and and send it to the wind farm controller to formulate the optimal dispatch algorithm.

[0373] In step S2, based on the threshold of the switching control strategy of the wind speed region of the control system of the wind turbine generator system, a prediction model of the main shaft torque and a prediction model of the tower thrust are obtained, which also includes:

[0374] Calculate the tower thrust under the following four conditions: Increment of active power reference value The partial derivative of :

[0375] Case 1: High wind speed High wind speed.

[0376] if , the wind turbine control system maintains high wind speed and calculates the tower thrust The increment of is expressed as:

[0377]

[0378] ;

[0379] Based on formula (24) and formula (44), we get:

[0380]

[0381] therefore,

[0382]

[0383] in, , and The subscript of indicates the case index;

[0384] Case 2: Low wind speed Low wind speed.

[0385] if , the wind turbine control system stays at low wind speed and calculates the tower thrust The increment of is expressed as:

[0386]

[0387] ;

[0388] Based on formula (14) and formula (36), we get:

[0389]

[0390] therefore,

[0391]

[0392] in, ;

[0393] Case 3: High wind speed Low wind speed.

[0394] if , the wind turbine control system transitions from high wind speed to low wind speed, Approximate calculations yield:

[0395] (39)

[0396] therefore,

[0397]

[0398] Where, ;

[0399] Case 4: Low wind speed High wind speed.

[0400] if , the wind turbine control system transitions from low wind speed to high wind speed, Approximate calculations yield:

[0401] (41)

[0402] therefore,

[0403]

[0404] Where, ;

[0405] According to formula (33) and formula (36), and They are and The function is:

[0406]

[0407]

[0408] function is nonlinear, and Similarly, it is described by a lookup table derived from the blade geometry, whose input is and ;

[0409] according to and , determine the operating point by looking up the table The power coefficient at ,in, and are the corresponding row index and column index respectively;

[0410] calculate and :

[0411]

[0412]

[0413]

[0414]

[0415] For generality, omit and The subscript of is calculated based on the measured value. and and sends it to the wind farm controller to formulate the optimal scheduling algorithm.

[0416] Figure 2 This is the predicted effect diagram of the main shaft torque.

[0417] The prediction model of spindle torque is:

[0418] The prediction model of tower thrust is:

[0419]

[0420] in, is the active power reference value of the wtth wind turbine, is the current active power output of the wind turbine, 、 、 、 All of them are derived from the above-derived formulas. is the current tower thrust, is the current spindle torque, is the increment of the active power reference value of the wtth wind turbine.

[0421] It should be noted that 、 、 、 The formula derived from , , , , , , , Come to.

[0422] In step S3, an equivalent fatigue load linearization model is obtained through the main shaft torque time series data. After combining the equivalent fatigue load linearization model with the wind turbine linear prediction model, a fatigue load model driven by both data and mechanism is obtained, which specifically includes:

[0423] Slice the spindle torque time series data using a sliding window with a window length of 10 to obtain the spindle torque time series data set and , The data before the 9 seconds of the time window is used as historical input data. The 10th second data of the time window is used as the prediction data. The load cycle amplitude and mean of the main shaft torque data of each sliding window are calculated by the rain flow counting method to obtain the equivalent fatigue load:

[0424]

[0425] Where, is the load amplitude at each level, is the number of cycles of this level of load, m is the Wohler index of the SN curve, and N depends on the design life of the wind turbine. is the equivalent fatigue load of this load signal, and the equivalent fatigue load time series is obtained ;

[0426] Constructing the observation function , observation function Including the cubic polynomial basis function, the random Fourier basis function with the fundamental frequencies of 0.1 and 0.01, and the delayed basis function with 10 steps of delay, the following matrix is ​​constructed:

[0427]

[0428]

[0429]

[0430] The state equation is constructed by Koopman operator, and its expression is:

[0431]

[0432] The Koopman matrices A and B are calculated using the extended dynamic modal decomposition (EDMD) method, namely:

[0433]

[0434] Vectorizing the optimization problem yields:

[0435]

[0436] in, is the Kronecker product, and by solving the pseudo-inverse, we get:

[0437]

[0438] Construct the output equation:

[0439]

[0440]

[0441]

[0442] in, is the error term, and the Koopman matrix C and D are calculated using the extended dynamic mode decomposition (EDMD) method, namely:

[0443]

[0444] By solving the pseudo-inverse, we get:

[0445]

[0446] Thus, the equivalent fatigue load linearization model based on the Koopman operator modeling method of data-driven method is obtained. Figure 3 This is the prediction effect diagram of the equivalent fatigue load linearization model, and its expression is:

[0447]

[0448]

[0449] The load data for the first 9 seconds of the next time window, The load data for the first 9 seconds of the current time window, Forecast load data for the 10th second of the current time window, is the observation function.

[0450] After combining the equivalent fatigue load linearization model with the wind turbine linear prediction model, a data-mechanism dual-driven fatigue load model is obtained, and the model linearity is maintained. The input of the data-mechanism dual-driven fatigue load model is: wind speed, generator speed, output power, pitch angle, generator torque, active power reference, main shaft torque and tower thrust, and the output is the equivalent fatigue load.

[0451] The linear prediction model of the wind turbine here includes a prediction model of the main shaft torque and a prediction model of the tower thrust.

[0452] In step S4, active power optimization control is performed on the fatigue load model driven by both data and mechanism to obtain the optimal solution for active power distribution that meets the safety and stability constraints of the wind farm. Specifically, the solution includes:

[0453] The optimization objective of active load shedding is constructed as follows:

[0454]

[0455] in, The equivalent fatigue load of each wind turbine is calculated based on the fatigue load model driven by the data mechanism. Obtained from the lookup table. The optimization problem constraints are:

[0456]

[0457] in, is the active power reference value, and are the maximum available power and minimum available power of a single machine, is the grid dispatch instruction, is the change in active power within 1 time step, is the maximum power ramp constraint for a single machine;

[0458] The data-mechanism dual-driven fatigue load model is used as a prediction model, and a distributed model predictive control algorithm is used to optimize and solve the data-mechanism dual-driven fatigue load model. The solution algorithm is a trust region constraint algorithm.

[0459] The central controller calculates the active power reference of each wind turbine in the wind farm as the initial point based on the traditional proportional allocation algorithm. At the same time, the distributed controller of each wind turbine generates a wind turbine fatigue load lookup table based on the prediction model at a scale of 10W and sends it to the central controller. The central controller solves the optimization problem using the trust domain constraint algorithm and sends the control strategy to each wind turbine controller. A dual-loop communication method is adopted to improve the reliability of data transmission.

[0460] The iterative steps of the trust region constraint algorithm specifically include:

[0461] Step S41: Set the initial iteration point according to the average distribution strategy , and set the initial trust domain radius , define the convergence threshold , gradient norm Stop iteration when

[0462] Step S42: At the current iteration point , construct a sub-problem including the objective function and constraints. The objective function uses a quadratic approximation model, and its expression is:

[0463]

[0464] Where, is the Hessian matrix, This is a trial step, and the constraints should be satisfied at this time:

[0465]

[0466] The constraints are linearized by Taylor's first-order expansion, and the expression is:

[0467]

[0468] ;

[0469] Step S43: Construct Lagrangian function:

[0470]

[0471] in, and is a Lagrange multiplier, and , by solving the partial derivatives of the Lagrangian function and combining the KKT conditions, the solution of the subproblem is obtained. The KKT conditions are:

[0472]

[0473] Step S44: Calculate the actual descent amount and predicted decline , and calculate the ratio of the actual decline to the predicted decline, the expression is:

[0474]

[0475] The ratio of the actual decline to the predicted decline is used to measure the degree of match between the predicted decline of the forecast model and the actual decline;

[0476] Step S45: Update the trust zone radius:

[0477] if , then the prediction model is considered to be of good quality, and the radius of the trust region is increased: ;

[0478] if , then the prediction model is considered to be of poor quality, and the trust region radius is reduced: ;

[0479] Otherwise, the trust region radius remains unchanged;

[0480] Step S46: Update the iteration point:

[0481] if , then accept the update, p, otherwise refuse to update, and stop the iteration when the iteration end condition is met;

[0482] After completing the iteration, the optimal solution for active power distribution that meets the safety and stability constraints of the wind farm is obtained and sent to each wind turbine for execution.

[0483] By comparing with the traditional proportional allocation algorithm, the present invention achieves a 35% reduction in the fatigue load of the entire main shaft in a 10×5MW wind farm while meeting the operation constraints of the wind farm.

[0484] After adopting the lookup table calculation acceleration, the present invention can achieve real-time solution speed (with 1 s as the control time step, the solution time for 10 wind turbines is 0.54 s).

[0485] The above description of the disclosed embodiments is intended to enable one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be readily apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention is not limited to the embodiments shown herein but is intended to conform to the widest scope consistent with the principles and novel features disclosed herein.

[0486] These changes, modifications, substitutions and variations to the embodiments without departing from the principles and spirit of the present invention are still within the scope of protection of the present invention.

Claims

1. A wind farm fatigue suppression active power control method based on data mechanism hybrid modeling, characterized in that: The method comprises: Step S1: establishing a continuous state space model, discretizing the continuous state space model using a sampling period to obtain a discrete state space model, and calculating the threshold of the switching control strategy of the wind speed region of the control system of the wind turbine generator system based on the discrete state space model; Step S2: Based on the threshold of the switching control strategy in the wind speed region of the control system of the wind turbine generator system, a prediction model of the main shaft torque and a prediction model of the tower thrust are obtained, which are recorded as the wind turbine generator system linear prediction model; Step S3: Obtain an equivalent fatigue load linearization model through the main shaft torque time series data, and combine the equivalent fatigue load linearization model with the wind turbine linear prediction model to obtain a fatigue load model driven by both data and mechanism; Step S4: performing active power optimization control on the fatigue load model driven by both data and mechanism to obtain an optimal solution for active power distribution that satisfies the safety and stability constraints of the wind farm; In step S1, a continuous state space model is established, which specifically includes: Step S11: establishing a nonlinear mechanism model of the wind turbine; Step S12: constructing a continuous state space model based on the nonlinear mechanism model of the wind turbine; When the current operating point is in the high wind speed area, that is, , Will follow the Active power reference value increment of typhoon generator decreases with the increase of , the control system of the wind turbine will transition to the low wind speed area. The threshold of the switching control strategy in this operating point area is ; Indicates The pitch angle measured at any moment, represents the pitch angle increment; When the current working point is in the low wind speed area, that is, hour, Will follow decreases and increases; if , the wind turbine control system will transition to high wind speed, and the threshold of the switching control strategy in the area where this operating point is located is , Indicates The filtered generator speed measured at the moment, represents the generator speed increment after filtering, Indicates the rated speed of the generator.

2. The wind farm fatigue suppression active power control method based on data mechanism hybrid modeling according to claim 1 is characterized in that: In step S11, a nonlinear mechanism model of the wind turbine is established, including: Step S111: Establish a derivative equation of the generator speed increment, which is expressed as follows: Where, represents the derivative of the generator speed increment, Indicates the gearbox transmission ratio, represents the equivalent mass of the blade mass and the engine mass combined, represents the generator torque increment, represents the aerodynamic torque increment, Indicates The generator torque measured at the moment, Indicates The aerodynamic torque measured at every moment; Step S112: Establish a derivative equation of the filtered generator speed increment, which is expressed as follows: Where, represents the derivative of the filtered generator speed increment, represents the filter time constant, represents the generator speed increment after filtering, Indicates the generator speed increment; Step S113: Define increment get: Where, represents the pitch angle, represents the generator speed after filtering, Indicates the rated speed of the generator, represents the proportional gain of the fan PI controller, represents the integral gain of the fan PI controller, represents the Laplace operator; Step S114: Based on the derivative equation of the generator speed increment after filtering established in step S112 and the increment obtained in step S113 , get the increment The derivative equation of : （4） Where, Indicates increment The derivative of Step S115: performing approximate calculation at the working point according to the Taylor series to obtain the aerodynamic torque increment and the generator torque increment; The expression of aerodynamic torque increment is: Where, Indicates the The increase in the active power reference value of the typhoon generator; The expression of the generator torque increment is: （6） Where, Indicates Generator torque at point Generator speed The partial derivative of is expressed as: （7） （8） Where, , Indicates The wind speed value at the blade at the moment, Indicates leaves, express about function, represents the wind energy utilization coefficient; Indicates Generator torque at point For increment The partial derivative of is expressed as: （9） Where, represents the air density, Indicates leaves; Step S116: Calculation and ; The expression is: represents the tip speed ratio, represents the pitch angle, Represents the adjacent unit The difference, Represents the adjacent unit The difference, represents the wind energy utilization coefficient, represents the wind energy utilization coefficient at (n,m) in the lookup table, represents the wind energy utilization coefficient at (n,m+1) in the lookup table; The expression is: （11） Where, Represents the wind energy utilization coefficient at (n+1,m) in the lookup table.

3. The wind farm fatigue suppression active power control method based on data mechanism hybrid modeling according to claim 2 is characterized in that: In step S12, a continuous state space model is constructed based on the nonlinear mechanism model of the wind turbine, including: Step S121: Construct a continuous state space model under high wind speed conditions, the expression of which is: （12） ； Step S122: Based on the continuous state space model under high wind speed conditions, ignoring the terms related to the pitch angle, a continuous state space model under low wind speed conditions is obtained, which is expressed as: 。 4. The wind farm fatigue suppression active power control method based on data mechanism hybrid modeling according to claim 3 is characterized in that: In step S1, the continuous state space model is discretized using a sampling period to obtain a discrete state space model, including: Sampling period Discretization processing is performed on the continuous state space model under low wind speed and the continuous state space model under high wind speed respectively; The continuous state space model under low wind speed is discretized to obtain the discrete state space model under low wind speed, which is expressed as follows: （13） Where, represents the step index, for The state variables at time , for The state variables at time , 、 、 are all discrete state space matrices after sampling; The continuous state space model under high wind speed is discretized to obtain the discrete state space model under high wind speed, which is expressed as follows: （14）。 5. The wind farm fatigue suppression active power control method based on data-mechanism hybrid modeling according to claim 4 is characterized in that: In step S1, based on the discrete state space model, the threshold of the switching control strategy of the wind speed region of the control system of the wind turbine generator system is calculated, which specifically includes: From the low-speed shaft motion equation: Where, is the derivative of the rotor speed, and we can further obtain: therefore, When the current operating point is in the high wind speed area, that is, , Will follow the Active power reference value increment of typhoon generator decreases with the increase of , the control system of the wind turbine will transition to the low wind speed area. The threshold of the switching control strategy in this operating point area is , the calculation process is as follows: According to formula (13), we can obtain: in and They are and The matrix elements of ,get: according to , it can be deduced that , whose expression is: When the current working point is in the low wind speed area, that is, hour, Will follow decreases and increases; if , the wind turbine control system will transition to high wind speed, and the threshold of the switching control strategy in the area where this operating point is located is , the calculation process is as follows: According to formula (14), In the formula and They are and The matrix elements of Calculated , whose expression is: 。 6. The wind farm fatigue suppression active power control method based on data-mechanism hybrid modeling according to claim 5 is characterized in that: In step S2, based on the threshold of the switching control strategy of the wind speed region of the control system of the wind turbine generator system, a prediction model of the main shaft torque and a prediction model of the tower thrust are obtained, including: Calculate the spindle torque in the following four cases Increment of active power reference value The partial derivative of : if , the wind turbine control system will stay at high wind speed. According to equations (5) and (6), equation (17) is transformed into: Based on formula (13) and formula (23), we can get: therefore, in, , and The subscript of indicates the case index; if , the wind turbine control system will stay at low wind speed. According to equations (5) and (6), equation (17) is transformed into: in, According to formula (14) and formula (26), we can get: therefore, in ; if , the wind turbine control system will transition from high wind speed to low wind speed. In this case, It is divided into two parts: and , Some operate in high wind speeds, while Some operate in low wind speeds; Therefore, Approximate calculations yield: (29) therefore, ； if , the wind turbine control system will transition from low wind speed to high wind speed, It is divided into two parts: and , Some operate in low wind speeds, Some operate in high wind speeds; Therefore, Approximate calculations yield: (31) therefore, 。 7. The wind farm fatigue suppression active power control method based on data-mechanism hybrid modeling according to claim 6 is characterized in that: In step S2, based on the threshold of the switching control strategy of the wind speed region of the control system of the wind turbine generator system, a prediction model of the main shaft torque and a prediction model of the tower thrust are obtained, which also includes: Calculate the tower thrust under the following four conditions: Increment of active power reference value The partial derivative of : if , the wind turbine control system maintains high wind speed and calculates the tower thrust The increment of is expressed as: ； Based on formula (24) and formula (44), we get: therefore, in, , and The subscript of indicates the case index; if , the wind turbine control system stays at low wind speed and calculates the tower thrust The increment of is expressed as: ； Based on formula (14) and formula (36), we get: therefore, in, ; if , the wind turbine control system transitions from high wind speed to low wind speed, Approximate calculations yield: (39) therefore, Where, ; if , the wind turbine control system transitions from low wind speed to high wind speed, Approximate calculations yield: (41) therefore, Where, ; According to formula (33) and formula (36), and They are and The function is: function is nonlinear and is described by a lookup table derived from the blade geometry, whose input is and ; according to and , determine the operating point by looking up the table The power coefficient at ,in, and are the corresponding row index and column index respectively; calculate and : The prediction model of spindle torque is: The prediction model of tower thrust is: in, is the active power reference value of the wtth wind turbine, is the current active power output of the wind turbine, 、 、 、 All of them are derived from the formulas above. is the current tower thrust, is the current spindle torque, is the increment of the active power reference value of the wtth wind turbine.

8. The wind farm fatigue suppression active power control method based on data-mechanism hybrid modeling according to claim 7 is characterized in that: In step S3, an equivalent fatigue load linearization model is obtained through the main shaft torque time series data. After combining the equivalent fatigue load linearization model with the wind turbine linear prediction model, a fatigue load model driven by both data and mechanism is obtained, which specifically includes: The main shaft torque time series data is generated by the main shaft torque prediction model. The main shaft torque and tower thrust time series data are sliced ​​with a sliding window of 10 to obtain the main shaft torque and tower thrust time series data sets. and , The data before the 9 seconds of the time window is used as historical input data. The 10th second data of the time window is used as the prediction data. The load cycle amplitude and mean of the main shaft torque data of each sliding window are calculated by the rain flow counting method to obtain the equivalent fatigue load: Where, is the load amplitude at each level, is the number of cycles of this level of load, m is the Wohler index of the SN curve, and N depends on the design life of the wind turbine. is the equivalent fatigue load of this load signal, and the equivalent fatigue load time series is obtained ; Constructing the observation function , observation function Including the cubic polynomial basis function, the random Fourier basis function with the fundamental frequencies of 0.1 and 0.01, and the delayed basis function with 10 steps of delay, the following matrix is ​​constructed: The state equation is constructed by Koopman operator, and its expression is: The Koopman matrices A and B are calculated using the extended dynamic modal decomposition (EDMD) method, namely: Vectorizing the optimization problem yields: in, is the Kronecker product, and by solving the pseudo-inverse, we get: Construct the output equation: in, is the error term, and the Koopman matrix C and D are calculated using the extended dynamic mode decomposition (EDMD) method, namely: By solving the pseudo-inverse, we get: Thus, the equivalent fatigue load linearization model based on the data-driven Koopman operator modeling is obtained, and its expression is: The load data for the first 9 seconds of the next time window, The load data for the first 9 seconds of the current time window, Forecast load data for the 10th second of the current time window, is the observation function; After combining the equivalent fatigue load linearization model with the wind turbine linear prediction model, a data-mechanism dual-driven fatigue load model is obtained, and the model linearity is maintained. The input of the data-mechanism dual-driven fatigue load model is: wind speed, generator speed, output power, pitch angle, generator torque, active power reference, main shaft torque and tower thrust, and the output is the equivalent fatigue load.

9. The wind farm fatigue suppression active power control method based on data-mechanism hybrid modeling according to claim 8, characterized in that: In step S4, active power optimization control is performed on the fatigue load model driven by both data and mechanism to obtain the optimal solution for active power distribution that meets the safety and stability constraints of the wind farm. Specifically, the solution includes: The optimization objective of active load shedding is constructed as follows: in, The equivalent fatigue load of each wind turbine is calculated based on the fatigue load model driven by the data mechanism. The lookup table is obtained, and the optimization problem constraints are: in, is the active power reference value, and are the maximum available power and minimum available power of a single machine, is the grid dispatch instruction, is the change in active power within 1 time step, is the maximum power ramp constraint for a single machine; The data-mechanism dual-driven fatigue load model is used as a prediction model, and a distributed model predictive control algorithm is used to optimize and solve the data-mechanism dual-driven fatigue load model. The solution algorithm is a trust region constraint algorithm. The central controller calculates the active power reference of each wind turbine in the wind farm using a traditional proportional allocation algorithm as the initial point. At the same time, the distributed controller of each wind turbine generates a wind turbine fatigue load lookup table based on the prediction model at a scale of 10W and sends it to the central controller. The central controller solves the optimization problem using a trust region constraint algorithm and sends the control strategy to each wind turbine controller. The iterative steps of the trust region constraint algorithm specifically include: Step S41: Set the initial iteration point according to the average distribution strategy , and set the initial trust domain radius , define the convergence threshold , gradient norm Stop iteration when Step S42: At the current iteration point , construct a sub-problem including the objective function and constraints. The objective function uses a quadratic approximation model, and its expression is: Where, is the Hessian matrix, This is a trial step, and the constraints should be satisfied at this time: The constraints are linearized by Taylor's first-order expansion, and the expression is: ； Step S43: Construct Lagrangian function: in, and is a Lagrange multiplier, and , by solving the partial derivatives of the Lagrangian function and combining the KKT conditions, the solution of the subproblem is obtained. The KKT conditions are: Step S44: Calculate the actual descent amount and predicted decline , and calculate the ratio of the actual decline to the predicted decline, the expression is: The ratio of the actual decline to the predicted decline is used to measure the degree of match between the predicted decline of the forecast model and the actual decline; Step S45: Update the trust zone radius: if , then the prediction model is considered to be of good quality, and the radius of the trust region is increased: ; if , then the prediction model is considered to be of poor quality, and the trust region radius is reduced: ; Otherwise, the trust region radius remains unchanged; Step S46: Update the iteration point: if , then accept the update, p, otherwise refuse to update, and stop the iteration when the iteration end condition is met; After completing the iteration, the optimal solution for active power distribution that meets the safety and stability constraints of the wind farm is obtained.

Images