Online identification of pitch damping of floating wind turbines and full sea state update method

Abstract

The invention discloses an online identification and full sea condition updating method of pitch damping of floating wind turbines, which belongs to the field of offshore wind power. In order to realize the identification of damping parameters in the full sea condition by using limited floating foundation response and monitoring data of marine environment, and to realize the global updating of pitch damping parameters in the full sea condition under limited data, the technical key points are as follows: the state space and the covariance matrix of the state space under the current sea condition are predicted by the GPR model, wherein the state space represents the pitch damping ratio of the floating wind turbine; the mean of the updated state space and the covariance of the state space are predicted by the unscented Kalman filter; the mean of the updated state space and the covariance of the state space under the current sea condition are input into the GPR model for fitting by Gaussian process regression, and an updated GPR model is obtained; the above steps are executed repeatedly, and the effect is to realize a more accurate prediction of the pitch damping of the floating wind turbine under other sea conditions.

Description

Technical Field

[0001] The present invention relates to the field of offshore wind power, and in particular to a method for identifying and updating pitch damping parameters of a floating wind power structure dynamics system. Background Art

[0002] Offshore wind energy development has great development potential due to its more stable and higher wind energy resources and lower visual and noise pollution to humans, and the installed capacity of offshore wind power is expanding rapidly.

[0003] To ensure the maximization of benefits during the entire operating cycle of offshore wind turbines, the monitoring data during service is used to accurately model the offshore wind turbines, reducing the uncertainty of key parameters that affect the structural response during the modeling process. This can more accurately predict the motion response of the model and monitor the health of the structure, providing support for subsequent operation and maintenance strategies, thereby achieving a comprehensive improvement in operation and maintenance efficiency and safety.

[0004] However, offshore wind turbines will inevitably cause damage to key structures or changes in dynamic parameters during their operation cycle. Current research has not paid sufficient attention to the impact of viscous damping on the hydrodynamic response of offshore wind turbine floating platforms. System modeling of floating wind turbines usually uses potential flow theory to ignore the viscosity of the fluid for simplification, which will lead to errors and increased uncertainties in model parameters.

[0005] Among them, roll and pitch motions are considered to be the most challenging response quantities to predict, because the pitch motion near the natural period of the floating foundation is greatly affected by damping, and the potential flow theory ignores viscous damping, which will make the damping low, resulting in a larger predicted motion response amplitude, a conservative design for the foundation, and an increase in the construction cost and the uncertainty of the prediction results.

[0006] In order to better compensate for this part of nonlinear damping in actual modeling, an additional 5%-8% critical damping is added in frequency domain analysis and time domain analysis to replace viscous damping. However, viscous damping is a nonlinear function of sea conditions, and this simple linear model cannot meet the needs of accurately predicting the motion response of offshore wind turbines.

[0007] In order to model the viscous damping more accurately, the damping parameters are usually identified and corrected using the attenuation motion and harmonic stimulated pitch motion of the scaled physical model. However, due to the size effect of the scaled model, there is a significant difference between the identification effect and the pitch damping of the actual size structure. On the other hand, the pitch viscous damping varies with different sea conditions. In the scaled test, it is impossible to traverse enough sea conditions to provide sufficient data under different sea conditions, and the applicability of the pitch damping results is questioned.

[0008] Another traditional method to study the viscous effect is to use computational fluid dynamics (CFD) theory to simulate free decay motion, but the computational efficiency of high-fidelity simulation analysis is too low to meet the actual engineering needs. Therefore, there is an urgent need for a damping parameter identification technology that uses limited floating foundation response and marine environment monitoring data in the full sea state range to achieve global update of pitch damping parameters in the full sea state under limited data conditions. Summary of the invention

[0009] In order to achieve the above purpose of using limited floating foundation response and marine environment monitoring data to identify damping parameters in the full sea condition range, the global update of pitch damping parameters in the full sea condition under limited data conditions is realized.

[0010] In a first aspect, a method for online identification of pitch damping of a floating wind turbine and updating of full sea conditions is provided in an embodiment of the present application, comprising:

[0011] S100. Predicting the state space and the covariance matrix of the state space under the current sea condition by using the GPR model, wherein the state space represents the pitch damping ratio of the floating wind turbine;

[0012] S200. predicting the mean of the updated state space and the covariance of the state space of the floating wind turbine by using an unscented Kalman filter;

[0013] S300. By Gaussian process regression, the updated mean of the state space and the covariance of the state space under the current sea condition are input into the GPR model for fitting to obtain an updated GPR model;

[0014] S400. Loop through steps S100 to S300.

[0015] According to a method for online identification of pitch damping of a floating wind turbine and updating of full sea conditions provided in an embodiment of the present application, step S200 includes:

[0016] S110. Based on the historical environmental data and the floating foundation motion data, a GPR model is established to predict the state space μ under the current sea conditions through the GPR model. k-1 and∑ k-1 The covariance matrix of the state space, where the state space represents the pitch damping ratio;

[0017] S120. Calculate the state space x k-1 The sigma point X[i], the weight coefficient w[i] of the mean corresponding to the sigma point X[i] m , weight coefficient w[i] of the covariance matrix c ;

[0018] S130. Calculate the predicted value of the sigma point from the sigma point X[i] through the state transfer function g According to the weight coefficient w[i] m , weight coefficient w[i] of the covariance matrix c , calculate the predicted state mean and the predicted state covariance

[0019] S140. Calculate the predicted measurement value of sigma point X[i] according to the measurement function h

[0020] S150. Based on the predicted measured value Calculate the predicted observed mean and the predicted observation covariance S k ;

[0021] S160. Obtain measured environmental data and floating foundation motion data, the measured environmental data is used to characterize the sea conditions, wherein the floating foundation motion data includes a time history curve of the pitching motion, and the observation z of the current sea conditions is calculated based on the floating foundation motion data. k , observe z k represents the standard deviation of the time course curve;

[0022] S170. Based on the predicted observed mean With observation z k Calculate the residual y k ; According to the weight coefficient w[i] of the covariance matrix c , predicted value Predicted state mean Predicted measurements and the predicted observed mean Compute the cross covariance between states and observations

[0023] S180. Based on the cross covariance between states and observations and the predicted observation covariance S k Calculate the Kalman gain K k ;

[0024] S190. According to the Kalman gain K k Update the mean of the state space and the covariance of the state space to get the updated mean μ of the state space k and the state space covariance ∑ k , where the state space x at time k is k By the mean μ of the state space k composition.

[0025] According to a method for online identification of pitch damping of a floating wind turbine and updating of full sea conditions provided in an embodiment of the present application, in step S120, the 2n+1 sigma points X[i] in the state space are expressed as follows:

[0026] X[0]=μ

[0027]

[0028] Where n represents the dimension of the state space, which consists of the pitch damping ratio of the floating wind turbine, n = 1;

[0029] in, is the i-th column of the square root of the matrix, λ is the scaling parameter that determines how far the sigma point is from the mean μ of the state space; where μ is the mean of the current state space, and ∑ is the covariance matrix of the current state space;

[0030] λ=α 2 (n+κ)-n

[0031] Where α represents the scaling factor, α = 0.15, and the hyperparameter κ is usually set to 3-n or 0;

[0032] The weight coefficient of the mean w[i] m , weight coefficient w[i] of the covariance matrix c It is expressed as:

[0033]

[0034] β represents a hyperparameter, and if the state space follows a Gaussian distribution, then β=2.

[0035] According to a method for online identification of pitch damping of a floating wind turbine and updating of full sea conditions provided in an embodiment of the present application, in step S130, the predicted value It is expressed as follows:

[0036]

[0037] represents the predicted value of sigma point X[i]; g represents the transfer function;

[0038] Predicted state mean It is expressed as follows:

[0039]

[0040] Predicted state covariance It is expressed as follows:

[0041]

[0042] Q kis the process noise ∈ k The covariance matrix of k ~N(0,Q k ).

[0043] According to a method for online identification of pitch damping of a floating wind turbine and updating of full sea conditions provided in an embodiment of the present application, in step S140, the predicted measurement value It is expressed as follows:

[0044]

[0045] h is the measurement function, which is obtained by considering the dynamic characteristics of the complex integrated wind turbine structure under the combined action of wind and waves and using a complex aerodynamic-hydrodynamic-servo-elastic fully coupled time domain analysis method.

[0046] According to a method for online identification of pitch damping of a floating wind turbine and updating of full sea conditions provided in an embodiment of the present application, in step S150, the predicted observation mean It is expressed as follows:

[0047]

[0048] Prediction observation covariance S k It is expressed as follows:

[0049]

[0050] R k is the measurement noise δ k The covariance matrix of the measurement noise δ k ~N(0,R k ).

[0051] According to a method for online identification of pitch damping of a floating wind turbine and updating of full sea conditions provided in an embodiment of the present application, in step S170, the residual y k It is expressed as follows:

[0052]

[0053] Cross covariance between states and observations It is expressed as follows:

[0054]

[0055] According to a method for online identification of pitch damping of a floating wind turbine and updating of full sea conditions provided in an embodiment of the present application, the Kalman gain K k It is expressed as follows:

[0056]

[0057] Updated state space mean μk It is expressed as follows:

[0058]

[0059] The covariance of the updated state space ∑ k It is expressed as follows:

[0060]

[0061] According to a method for online identification of pitch damping of a floating wind turbine and updating of full sea conditions provided in an embodiment of the present application, step S300 includes:

[0062] Assume that the training data D = {(x1, y1), (x2, y2), ..., (x n ,y n )};

[0063] The data comes from a noisy process:

[0064] y i =f(x i )+∈,

[0065] Among them, i represents the i-th sample point, x i Represents the sea state, which is an n-dimensional input vector; y i represents the mean μ of the updated state space k , the covariance matrix of the state space ∑ k ;∈ represents Gaussian noise term;

[0066] Aggregate the input and output into X = [x1, x2, ..., x n ] and y=[y1,y2,...,y n ];

[0067] Among them, X represents the ocean environment and wind environment. The ocean environment includes the wave height and spectrum peak period. The wind environment includes the average wind speed of turbulent wind. Y represents the updated state mean μ of the pitch damping ratio in the environment. k , covariance matrix∑ k ;

[0068] The kernel function is:

[0069]

[0070] represents the signal variance, l Hs is the length scale of the significant wave height, l Tp is the length scale of the spectrum peak period, l Wis the length scale of the average wind speed of turbulent wind; RBF means that the kernel function type is a vector odd function; length_scale means the length scale

[0071] For a given set of training data {X, y} and a test input x to predict sea conditions * , GPR defines a Gaussian prediction distribution whose output y * The mean of the state space is:

[0072]

[0073] The covariance of the state space is:

[0074]

[0075] Among them, the mean GP μ The prediction is a linear combination of the training output y; k * is the test input x * The kernel vector between X and the training input X; K(X,X) is the kernel matrix; Indicates the introduction of Gaussian noise; k(x * ,x * ) represents the test input x * The kernel vector is not updated;

[0076] The mean value of the state space is GP μ is the state space μ of the next sea state k-1 , represents the pitch damping ratio, and the covariance GP of the state space ∑ is the covariance matrix ∑ of the state space of the next sea state k-1 .

[0077] In a second aspect, an embodiment of the present application further provides an electronic device, comprising: one or more processors, a memory, and one or more programs; wherein the one or more programs are stored in the memory, and the one or more programs include instructions, which, when executed by the electronic device, enable the electronic device to execute the first aspect and any possible technical solution of the first aspect.

[0078] Beneficial effects: In the first aspect, the present invention is based on the unscented Kalman filter (UKF) and Gaussian process regression (GPR) method, and through the measured environment and floating foundation motion data, the pitch damping under the corresponding load and sea conditions is identified by the unscented Kalman filter, and then the identified sample points are fed back to the full sea state pitch damping Gaussian process regression model (GPR) to achieve the update of the full sea state GPR function of pitch damping. Using this updated GPR function with improved accuracy, the pitch damping of the floating wind turbine under other sea conditions is more accurately predicted, and then the motion response of the floating wind turbine under different environments is more accurately predicted.

[0079] On the second aspect, the identification method of the present invention introduces the average wind speed of turbulent wind into the environmental information, considers the influence of wind load on the nonlinear response of floating wind turbines, considers the integrated coupled dynamic response characteristics of floating wind turbines, and compared with other methods, the traceless transformation processing method will improve its update efficiency.

[0080] In the third aspect, the present invention can accurately model the pitch damping of floating wind turbines through the above-mentioned updating, fitting and prediction cycles with sufficient data; when the amount of data is insufficient, the GPR model can still use the kernel function based on prior knowledge and limited data sample points to reasonably update the pitch damping parameters under the conditions around the sample points. BRIEF DESCRIPTION OF THE DRAWINGS

[0081] Figure 1 Flowchart of the first UKF-GPR identification method.

[0082] Figure 2 Flowchart of the second UKF-GPR identification method.

[0083] Figure 3 Three-dimensional GPR effect diagram when the wind speed is 15m / s.

[0084] Figure 4 Heatmap of initial error and predicted error when wind speed is 15m / s.

[0085] Figure 5 Three-dimensional GPR effect diagram when the wind speed is 20m / s.

[0086] Figure 6 Heat map of initial error and predicted error when wind speed is 20m / s.

[0087] Figure 7 Three-dimensional GPR effect diagram when the wind speed is 25m / s.

[0088] Figure 8 Heatmap of initial error and predicted error when wind speed is 25m / s. DETAILED DESCRIPTION

[0089] like Figure 1 As shown in the figure, a method for online identification and full sea condition updating of pitch damping of floating wind turbines is based on the unscented Kalman filter (UKF) and Gaussian process regression (GPR) methods. Through the measured environment and floating foundation motion data, the pitch damping under corresponding load and sea conditions is identified by unscented Kalman filtering, and then the identified sample points are fed back to the Gaussian process regression model (GPR) of pitch damping in all sea conditions to realize the update of the GPR function of pitch damping in all sea conditions.

[0090] The present invention utilizes the updated GPR function with improved accuracy to achieve more accurate prediction of the pitch damping of the floating wind turbine under other sea conditions, thereby more accurately predicting the motion response of the floating wind turbine under different environments.

[0091] This identification method introduces the average wind speed of turbulent wind into the environmental information, considers the influence of wind load on the nonlinear response of floating wind turbines, considers the integrated coupled dynamic response characteristics of floating wind turbines, and compared with other methods, the traceless transformation processing method will improve its update efficiency.

[0092] like Figure 2 As shown, the process of the identification method can be simplified as follows:

[0093] 1. State space selection: Select the mean value of the pitch damping ratio to form the state space x k-1 If it is the first identification, the assumed initial pitch damping ratio μ0=0,∑0=0.01 is used. Otherwise, the GPR model is used to predict the pitch damping ratio μ under the current sea conditions. k-1 (state space),∑ k-1 (covariance matrix of state space).

[0094] 2. Untraceable transformation: for the state space x k-1 Calculate and get the state space x k-1 The sigma point X[i] and the corresponding mean weight coefficient w[i] m , weight coefficient w[i] of the covariance matrix c .

[0095] 3. Prediction: Calculate the predicted mean value by combining the state transfer function g and variance

[0096] 4. Update:

[0097] a) Calculate the predicted measurement value of each sigma point through the measurement function h to form the predicted measurement value Calculate the predicted observed mean Covariance S with predicted observationsk

[0098] b) Calculate the observed measurement value z through the monitoring data under the current sea conditions k

[0099] c) Calculate the residual:

[0100] d) Calculate the cross covariance between states and observations

[0101] e) Calculate the Kalman gain:

[0102] f) Update the state mean μ k and state covariance∑ k

[0103] 5. GPR fitting: Updated state mean μ under current sea conditions k and state covariance∑ k Input as training data into the GPR model for fitting.

[0104] 6. State space selection: Use the fitted GPR model to predict the next sea condition and obtain the pitch damping ratio μ k-1 (state space),∑ k-1 (The covariance matrix of the state space) will be used as prior knowledge for the next update.

[0105] like Figure 1-2 As shown, the floating wind turbine pitch damping online identification and full sea condition updating method of the present invention is described in detail:

[0106] UKF is an algorithm that updates the state of a dynamic system based on a series of observations and control information. It can handle nonlinear problems well and quantify the uncertainty of the system state. Specifically, the pitch damping of the floating wind turbine foundation as a system state that needs to be identified can be updated by using the standard deviation of the foundation pitch motion under the current sea conditions as an observation, where the change in pitch damping can be replaced by the pitch damping ratio:

[0107]

[0108] Wherein, damping ratio represents the pitch damping ratio, damping represents the pitch viscous damping, and criticaldamping represents the pitch critical damping.

[0109] The true value of the pitch damping ratio is a nonlinear function related to the wave height, spectrum peak period and average wind speed of turbulent wind. s is the significant wave height, T Pis the spectrum peak period, W is the turbulent wind speed, multiple groups of sea conditions are selected according to the average distribution, and the corresponding true value of the pitch damping ratio is calculated.

[0110] 1. State space selection:

[0111] Take the first group of sea conditions to identify the pitch damping ratio, and the pitch damping ratio obeys the mean value μ k-1 and covariance is ∑ k-1 Gaussian distribution of , pitch damping ratio is used as the parameter to be identified to form the state space, x k-1 represents the state space of the system at time k-1, and its corresponding covariance matrix is ​​∑ k-1 Assume that the initial pitch damping ratio μ0=0, ∑0=0.01.

[0112] n represents the dimension of the state space. Since the state space consists only of the pitch damping ratio, n=1.

[0113] 2. Traceless transformation:

[0114] The unscented transform (UT) extracts the sigma point X. Select 2n+1 sigma points X[i] as follows:

[0115] X[0]=μ

[0116]

[0117] in, is the i-th column of the square root of the matrix, and λ is the scaling parameter that determines how far the sigma point is from the mean μ.

[0118] λ=α 2 (n+κ)-n

[0119] Where α represents the scaling factor, which is usually a small positive number. α=0.15 is recommended. The hyperparameter κ is usually set to 3-n or 0. The weight coefficient w[i] is m and w[i] c can be used to predict the state space and covariance matrix,

[0120]

[0121] where w[i] m The weight coefficient representing the mean, w[i] c represents the weight coefficient of the covariance matrix, and the hyperparameter β is set to 2 when the state space follows a Gaussian distribution.

[0122] 3. Prediction:

[0123] The predicted mean and variance are then calculated from the resulting sigma points in conjunction with the state transfer function:

[0124]

[0125] Indicates the predicted value of the sigma point;

[0126] Calculate the predicted state mean:

[0127]

[0128] w[i] m The weight coefficient representing the mean;

[0129] Compute the predicted state covariance:

[0130]

[0131] Process noise ∈ k is a covariance Q k Gaussian noise. That is, ∈ k ~N(0,Q k )

[0132] The present invention usually needs to use a complex process model when calculating the prediction mean and the prediction covariance matrix, so a nonlinear system transfer function g is introduced to represent the process model, and process noise ∈ k In this case, the state of the system is assumed to remain unchanged, so the system transfer function g can be simplified to a linear one, and the uncertainty of the system transfer function g is actually proportional to the intensity of the process noise Q k Get the value.

[0133] 4. Update:

[0134] The predicted measurement value of each sigma point is obtained by measuring the function h

[0135]

[0136] The measurement function h is usually a complex nonlinear function. Here, hydrodynamic analysis software is used instead. The calculated sigma point is input into the numerical model as the pitch damping ratio. The hydrodynamic analysis is performed under this sea condition, and the obtained standard deviation of the pitch of the floating wind turbine is used as the prediction measurement space.

[0137] Used to calculate the predicted observed mean and its predicted observation covariance S k , S k Indicates uncertainty.

[0138] Compute the predicted observed mean:

[0139]

[0140] Compute the predicted-observation covariance:

[0141]

[0142] Observation z k is the standard deviation of the pitch of the floating wind turbine under this sea condition obtained through real-time monitoring, which is affected by the measurement noise δ k , δ k ~N(0,R k ), R k is the covariance matrix of the measurement noise, subject to the measurement noise δ k , δ k ~N(0,R k )’s impact.

[0143] Observation z k Based on the following method: obtain the floating foundation motion data, the floating foundation motion data includes the time history curve of the pitching motion, calculate the observation z of the current sea condition based on the floating foundation motion data k , observe z k Represents the standard deviation of the time course curve.

[0144] Compute the residual:

[0145]

[0146] Compute the cross covariance between states and observations:

[0147]

[0148] Then use it to calculate the Kalman gain K k :

[0149]

[0150] Finally, a standard Kalman filter update is used to compute an updated state estimate.

[0151] Update state mean:

[0152]

[0153] Update state covariance:

[0154]

[0155] Update state mean μ k The state space x at time k is composed k , and its corresponding covariance matrix is ​​∑k

[0156] 5. Gaussian process regression (GPR) fitting:

[0157] Subsequently, the updated state space and state covariance are input into the Gaussian process regression model for fitting.

[0158] GPR is a powerful and nonparametric tool for learning regression functions from sample data. The main advantages of GPR are its modeling flexibility, its ability to provide uncertainty estimates, and its ability to learn noise and smoothness parameters from training data.

[0159] GPR can be viewed as a "Gaussian distribution on a function". More precisely, GPR describes a random process in which random variables are co-Gaussian distributed. A Gaussian process is fully described by its mean and covariance functions. Assume that the present invention has a set of training data D = {(x1, y1), (x2, y2), ..., (x n ,y n )}, x represents the sea state, y represents the updated state mean μ k , the covariance matrix is ​​∑ k .

[0160] These data come from a noisy process:

[0161] y i =f(x i )+∈,

[0162] where x i is an n-dimensional input vector representing the sea state, y i is a scalar output, representing the updated state mean and covariance matrix. The Gaussian noise term ∈ comes from N(0,σ 2 ). i represents the i-th sample point.

[0163] For convenience, the input and output are aggregated into X = [x1, x2, ..., x n ] and y=[y1,y2,...,y n ]. The joint distribution of the noisy output y is a function of the input X. It is a zero-mean multivariate Gaussian distribution of the form:

[0164]

[0165] where K(X,X) is the kernel matrix, whose elements are K ij = k(x i ,x j ). The kernel function k(x,x') is a measure of the "closeness" between the inputs. Gaussian noise is introduced. The key idea is to condition this Gaussian on known elements (training points). N() represents the Gaussian distribution.

[0166] This article uses the most popular radial basis function (RBF) as the kernel. Its form is:

[0167]

[0168] in is the signal variance, representing the prior knowledge about the variance of the random process. l is the length scale.

[0169] In this algorithm, the training data {X, y} is provided by the updated value of UKF, where X represents the environment such as the ocean environment (significant wave height, spectral peak period) and wind environment (average wind speed of turbulent wind), and y represents the updated value of the pitch damping ratio in this environment (updated state mean μ k , covariance matrix∑ k ).

[0170] The kernel function of the Gaussian process regression model is:

[0171]

[0172] represents the signal variance, l Hs is the length scale of the significant wave height, l Tp is the length scale of the spectrum peak period, l W is the length scale of the average wind speed of turbulent wind.

[0173] 6. State Space Selection

[0174] After the training data is input into the GPR model for fitting, a more accurate GPR model is obtained, which can predict the pitch damping ratio of the next environment. Therefore, the second sea condition is selected to predict its pitch damping ratio, and the UKF identification is performed based on the observed values ​​of the second sea condition.

[0175] For a given set of training data {X, y} and a test input x * (forecast sea conditions), GPR defines a Gaussian prediction distribution, whose output y * The mean of is:

[0176]

[0177] Its variance is:

[0178]

[0179] Here, k * is the test input x *and the kernel vector between the training input X. The above equation shows that the mean GP μ The prediction is a linear combination of the training outputs y, with each output weighted by the same factor as the test input x. * is directly related to the correlation between the training inputs X. Prediction uncertainty, given by the variance GP ∑ captures the process noise and the correlation between the test input and the training data. μ The pitch damping ratio μ of the next sea state is k-1 (state space), variance GP ∑ That is the ∑ of the next sea state k-1 (covariance matrix of state space)

[0180] After a sufficient amount of data is used to perform the above update, fitting and prediction cycles, the pitch damping of floating wind turbines can be accurately modeled; when the amount of data is insufficient, the GPR model can still use the kernel function based on prior knowledge and limited data sample points to reasonably update the pitch damping parameters under the conditions around the sample points.

[0181] In one example, a floating wind turbine is used to perform a specific example demonstration based on the identification method using simulated monitoring data.

[0182] Among them, H s The significant wave height is distributed averagely at (2,9) m, T P The spectrum peak period is averagely distributed in (9, 14) s, W is the turbulent wind speed averagely distributed in (15, 25) m / s, 48 ​​groups of sea conditions are randomly selected according to the average distribution, and the corresponding assumed true value of the pitch damping ratio is used to obtain the simulated monitoring data. The information of these 48 groups of sea conditions is shown in Table 1.

[0183] Table 148 Groups of sample point sea conditions and state space update data

[0184]

[0185]

[0186]

[0187] In the state space:

[0188] 1. State space selection:

[0189] The initial pitch damping ratio is x0=0,∑0=0.01 in the full sea state. After the first GPR model fitting.

[0190] 2. Traceless transformation:

[0191] The hyperparameter settings of the unscented transformation are shown in Table 2. The sigma point X[i] and the mean weight coefficient w[i] are obtained. m , the weight coefficient w[i] of the covariance matrix c .

[0192] 3. Prediction:

[0193] In this case, the state of the system is assumed to remain unchanged, so the system transfer function g can be simplified to a linear value of Q = 0.01 2 . Get the predicted mean Prediction covariance matrix

[0194] 4. Update:

[0195] First, the predicted measurement value of each sigma point is obtained by measuring the function h Obtain observation z based on simulated monitoring data k In order to better simulate the actual situation, white noise with a signal-to-noise ratio of 100 is added, and the measurement noise R takes the intensity of the white noise. After calculation, the updated state mean μ is obtained k , update state covariance∑ k The state means and covariances before and after the update are shown in Table 1.

[0196] 5. Gaussian process regression (GPR) fitting:

[0197] The ocean environment is compared with the updated state mean μ k , update state covariance∑ k The training values ​​are input into the GPR model for fitting, and the GPR hyperparameter settings are shown in Table 2.

[0198] Table 2 UKF and GPR hyperparameter values

[0199]

[0200] 6. State space selection:

[0201] The predicted value of the GPR model is selected as the state space for other subsequent sea conditions.

[0202] After the above process, a relatively accurate GPR model was obtained by fitting the GPR model using 48 sample points. Figure 3-8As shown in the figure, in the three-dimensional GPR effect diagram, the wind speed is fixed, the significant wave height and the spectrum peak period are the x-axis, and the y-axis represents the pitch damping ratio. The red three-dimensional surface represents the assumed true value of the pitch damping ratio, the blue three-dimensional surface represents the predicted value of the pitch damping ratio, and the gray three-dimensional surface represents the surface of the predicted value of the pitch damping ratio plus or minus 3 times the standard deviation. The initial error heat map represents the error between the initial pitch damping ratio and the assumed true value at a fixed wind speed, and the predicted error heat map represents the error between the predicted pitch damping ratio and the assumed true value at a fixed wind speed. By comparing the two, it can be seen that the identification algorithm is effective.

[0203] Based on the above embodiments, an embodiment of the present application further provides an electronic device, and when the computer program is run on the electronic device, the electronic device executes the method provided in the above embodiments.

[0204] Based on the above embodiments, the embodiments of the present application further provide a computer storage medium, in which a computer program is stored. When the computer program is executed by a computer, the computer executes the method provided in the above embodiments.

[0205] The storage medium may be any available medium that can be accessed by a computer. For example, but not limited to, a computer-readable medium may include RAM, ROM, EEPROM, CD-ROM or other optical disk storage, magnetic disk storage medium or other magnetic storage device, or any other medium that can be used to carry or store the desired program code in the form of instructions or data structures and can be accessed by a computer.

[0206] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0207] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0208] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0209] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0210] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application is also intended to include these modifications and variations.

Claims

1. A method for online identification of pitch damping of a floating wind turbine and updating of full sea conditions, characterized in that: include S100. Predicting the state space and the covariance matrix of the state space under the current sea condition by using the GPR model, wherein the state space represents the pitch damping ratio of the floating wind turbine; S200. predicting the mean of the updated state space and the covariance of the state space of the floating wind turbine by using an unscented Kalman filter; S300. By Gaussian process regression, the updated mean of the state space and the covariance of the state space under the current sea condition are input into the GPR model for fitting to obtain an updated GPR model; S400. Loop through steps S100 to S300; The step S200 includes: S110. Based on the historical environmental data and floating foundation movement data, a GPR model is established to predict the state space under the current sea conditions through the GPR model. as well as The covariance matrix of the state space, where the state space represents the pitch damping ratio; S120. Calculate state space Sigma point , and sigma point The corresponding weight coefficient of the mean , weight coefficients of the covariance matrix ; S130. From sigma point Through the state transfer function Calculate the predicted value of sigma point , according to the weight coefficient , weight coefficients of the covariance matrix , calculate the predicted state mean and the predicted state covariance ; S140. According to the measurement function Calculate sigma points The predicted measurement value of ; S150. Based on the predicted measured value Calculate the predicted observed mean and the predicted observation covariance ; S160. Obtain measured environmental data and floating foundation motion data, the measured environmental data is used to characterize the sea conditions, wherein the floating foundation motion data includes a time history curve of the pitching motion, and the observation of the current sea conditions is calculated based on the floating foundation motion data. , observation represents the standard deviation of the time course curve; S170. Based on the predicted observed mean With observation Calculate residuals ; According to the weight coefficient of the covariance matrix , predicted value , predicted state mean , predict the measured value and the predicted observed mean Compute the cross covariance between states and observations ; S180. Based on the cross covariance between states and observations and the predicted-observation covariance Calculate Kalman gain ; S190. According to Kalman gain Update the mean of the state space and the covariance of the state space to get the updated mean of the state space and the state space covariance , where time The state space By the mean of the state space composition.

2. The method for online identification of pitch damping of floating wind turbines and updating of full sea conditions according to claim 1 is characterized in that: In step S120, the state space Sigma points It is expressed as follows: in, represents the dimension of the state space, which consists of the pitch damping ratio of the floating wind turbine, ; in, is the square root of the matrix List, It is the mean value that determines the distance of the sigma point from the state space The scaling parameters for near and far distances; is the mean of the current state space, is the covariance matrix of the current state space; in, represents the scaling factor, , hyperparameters Set to 3-n or 0; Weight coefficient of mean , weight coefficients of the covariance matrix It is expressed as: represents the hyperparameter, and the state space follows a Gaussian distribution.

2.

3. The method for online identification of pitch damping of floating wind turbines and updating of full sea conditions according to claim 1 is characterized in that: In step S130, the predicted value It is expressed as follows: Indicates sigma point The predicted value of represents the transfer function; Predicted state mean It is expressed as follows: Predicted state covariance It is expressed as follows: is the process noise The covariance matrix of .

4. The method for online identification of pitch damping of floating wind turbines and updating of full sea conditions according to claim 1 is characterized in that: In step S140, the measured value is predicted It is expressed as follows: It is a measurement function, which is obtained by considering the dynamic characteristics of the complex integrated wind turbine structure under the combined action of wind and waves and using a complex aerodynamic-hydrodynamic-servo-elastic fully coupled time domain analysis method.

5. The method for online identification of pitch damping of floating wind turbines and updating of full sea conditions according to claim 1 is characterized in that: In step S150, the observed mean is predicted It is expressed as follows: Prediction-Observation Covariance It is expressed as follows: is the measurement noise The covariance matrix of the measurement noise .

6. The method for online identification of pitch damping of floating wind turbines and updating of full sea conditions according to claim 1 is characterized in that: In step S170, Residual It is expressed as follows: Cross covariance between states and observations It is expressed as follows: 。 7. The method for online identification of pitch damping of floating wind turbines and updating of full sea conditions according to claim 1 is characterized in that: Kalman Gain It is expressed as follows: The mean of the updated state space It is expressed as follows: The covariance of the updated state space It is expressed as follows: 。 8. The method for online identification of pitch damping of a floating wind turbine and updating of full sea conditions according to any one of claims 1 to 6, characterized in that: The step S300 includes: Assume that the training data is obtained in step S200 ; The data comes from a noisy process: in, Indicates Sample points, represents the sea state, which is an n-dimensional input vector; represents the mean of the updated state space , the covariance matrix of the state space ; represents the Gaussian noise term; Aggregate the input and output into and ; in, Represents the ocean environment and wind environment. The ocean environment includes the wave height and spectrum peak period. The wind environment includes the average wind speed of turbulent wind. The updated state mean of the pitch damping ratio in the representative environment , covariance matrix ; The kernel function is: represents the signal variance, is the length scale of the significant wave height, is the length scale of the spectral peak period, is the length scale of the average wind speed of turbulent wind; Indicates that the kernel function type is a vector odd function; Indicates length scale For a given set of training data and a test input for predicted sea conditions , GPR defines a Gaussian prediction distribution whose output The mean of the state space is: The covariance of the state space is: Among them, the mean Predictions are the training output Linear combinations of is the test input With training input The kernel vector between ; is the kernel matrix; Indicates the introduction of Gaussian noise; Represents test input The kernel vector is not updated; The mean of the state space The state space for the next sea state , represents the pitch damping ratio, and the covariance of the state space is the covariance matrix of the state space of the next sea state .

9. An electronic device, comprising: One or more processors, a memory, and one or more programs; wherein the one or more programs are stored in the memory, and the one or more programs include instructions, which, when executed by the electronic device, enable the electronic device to perform any one of the methods described in claims 1-8.

Images