Wind power prediction method based on marine meteorological characteristics and wake flow interference

Abstract

The invention provides a wind power prediction method based on marine meteorological characteristics and wake flow interference, and aims to solve the problem that the existing wind power prediction does not fully consider the thermal and dynamic characteristics of a marine boundary layer, the wake flow interference effect of a fan group and the attenuation of a salt mist environment. According to the method, power prediction is realized by constructing a multi-source marine meteorological characteristic field, a wake flow interference field and an environment attenuation compensation mechanism. Specifically, a meteorological characteristic matrix is generated by combining an integral norm of wave height change, a position and wind direction characteristic matrix is constructed based on a fan position and a prevailing wind direction, wake flow interference intensity is quantified through a Gaussian kernel function, and meanwhile, an attenuation compensation item containing an error function is designed to process a salt spray corrosion effect. Establishing a bimodal wind speed probability distribution model, and combining a space-time adaptive quantile regression output power prediction interval; the method effectively solves the problems of marine meteorological dynamic characterization, wake flow interference quantization and environmental attenuation compensation, and significantly improves the precision and reliability of offshore wind power prediction.

Description

Technical Field

[0001] This invention relates to wind power prediction based on marine meteorological characteristics and wake interference. It addresses the challenges of marine meteorological complexity, wake effects, and salt spray environmental attenuation in wind power prediction by constructing a multi-source marine meteorological characteristic field, a wake interference field, and an environmental attenuation compensation mechanism. Background Technology

[0002] With the rapid development of the offshore wind power industry, wind power forecasting is becoming increasingly important for optimizing grid dispatch, improving energy efficiency, and ensuring the stable operation of wind farms. However, existing wind power forecasting methods have many limitations when facing the complex characteristics of the marine environment. On the one hand, marine meteorological conditions are complex and variable, and there are complex spatiotemporal correlations among multi-source data such as sea surface temperature, wave height, and vertical wind speed profiles. Traditional methods struggle to fully explore the intrinsic connections between these features, resulting in limited forecast accuracy. Existing technologies often focus on the analysis of single meteorological elements, failing to comprehensively capture the thermodynamic characteristics of the ocean boundary layer, thus affecting the accurate grasp of wind power variation trends. On the other hand, the wake interference effect between wind turbine clusters significantly affects wind power output. The wake can alter the wind speed environment of downstream turbines, thereby affecting their power generation efficiency. However, existing forecasting models are insufficient in quantifying the intensity of wake interference and simulating its spatiotemporal distribution, making it difficult to accurately assess the actual impact of wakes on wind power.

[0003] Furthermore, the salt spray environment encountered by offshore wind farms can also attenuate sensor measurement data, affecting the accuracy and reliability of the data. Current technologies lack effective dynamic compensation mechanisms, making it impossible to correct sensor data in a timely manner, further reducing the accuracy of wind power prediction. Summary of the Invention

[0004] This invention addresses the shortcomings of existing wind power forecasting methods, which fail to adequately consider the thermodynamic characteristics of the ocean boundary layer, the wake interference effect of wind turbine clusters, and the attenuation caused by salt spray. It proposes a wind power forecasting method based on ocean meteorological characteristics and wake interference. This invention achieves power forecasting by constructing a multi-source ocean meteorological characteristic field, a wake interference field, and an environmental attenuation compensation mechanism. Specifically, this invention generates a meteorological characteristic matrix using the integral norm of wave height variation, constructs a position and wind direction characteristic matrix based on wind turbine location and prevailing wind direction, quantifies the wake interference intensity using a Gaussian kernel function, designs an attenuation compensation term with an error function to handle the salt spray corrosion effect, establishes a dual-modal wind speed probability distribution model, and combines spatiotemporal adaptive quantile regression to output the power prediction interval. This invention effectively solves the problems of dynamic characterization of ocean meteorology, quantification of wake interference, and environmental attenuation compensation, significantly improving the accuracy and reliability of offshore wind power forecasting.

[0005] The present invention provides a wind power prediction method based on marine meteorological characteristics and wake interference, comprising the following steps:

[0006] S1: Acquire sea surface temperature field data, vertical wind speed profile data, and wave height field data for the target sea area; calculate the spatial gradient of the sea surface temperature field in the longitude and latitude directions, and multiply it with the first-order shear of the vertical wind speed; perform time integration on the gradient of the wave height field and generate wave dynamic features through arctangent function transformation, forming a meteorological feature matrix. Specifically, acquire the sea surface temperature field data of the target sea area, represented as a two-dimensional matrix. Where m represents the number of grid points in the longitude direction and n represents the number of grid points in the latitude direction; the spatial gradient of the temperature field is accurately calculated using the central difference method. The temperature change rate distribution is obtained by calculating the temperature difference between adjacent grid points in both the longitude and latitude directions and dividing it by the grid spacing.

[0007]

[0008] In this formula, i = 2, ..., m-1 and j = 2, ..., n-1 represent the point indices of longitude and latitude, respectively, and T s (i,j+1)-T s (i,j-1) and T s (i+1,j)-T s (i-1,j) represent the temperature difference between adjacent grid points at latitude and longitude, respectively, and Δx and Δy represent the grid spacing at latitude and longitude, respectively. This represents the gradient matrix along the longitude direction. This represents the gradient matrix along the latitudinal direction and also processes vertical wind speed profile data. Where l represents the vertical height layer, the wind speed change rate between adjacent height layers is calculated to obtain the vertical wind speed shear characteristics:

[0009]

[0010] In this formula, S z V represents the characteristics of vertical wind speed shear. z (i,j,k) represents the vertical wind speed profile data at the ith latitude grid and the ith longitude grid, at the ith layer height. The second derivative is calculated using the central difference scheme.

[0011]

[0012] In this formula, Let Δz represent the second derivative of vertical wind speed, Δz represent the vertical interlayer spacing, and k = 1, ..., l-1 represent the vertical height layer index, forming a three-dimensional feature field. Combined with sea surface temperature gradient and vertical wind speed characteristics, a three-dimensional thermodynamic feature field is generated. Feature fusion is achieved using element-wise multiplication.

[0013]

[0014] In this formula, T represents the Hadamard product. s (i,j) represents the sea surface temperature field data of the i-th longitude grid and the j-th latitude grid, Γ(i,j,k) represents the air-sea thermal and dynamic interaction effect at a specific location and altitude layer, and Γ represents the thermal and dynamic characteristic field; for wave height field data... Calculate the spatial gradient field:

[0015]

[0016] In this formula, Δx and Δy represent the latitude and longitude grid spacing, respectively. Let W(i,j+1)-W(i,j-1) represent the wave height gradient field, where W(i,j+1)-W(i,j-1) represent wave height field data at the same latitude but with a longitude difference of two grids. The gradient magnitude is quantified using the Frobenius norm.

[0017]

[0018] In this formula, ||·|| F Let Frobenius norm be denoted, Δt be the time window, and τ be the time variable. This represents the gradient time integral, which, through arctangent transformation, constrains the range of eigenvalues ​​and generates stable wave dynamic features.

[0019]

[0020] In this formula, D w Indicates the dynamic characteristics of waves. The gradient time integral is represented by arctan, which represents the arctangent function. This averages and compresses the thermodynamic feature field along the vertical dimension, reducing the three-dimensional feature field to a two-dimensional plane.

[0021]

[0022] In this formula, The vertical average thermodynamic characteristics are represented by Γ(i,j,k), where l represents the vertical height layer and Γ(i,j,k) represents the air-sea thermodynamic and dynamic interaction effect at a specific location and height layer. A variance-based dynamic weighting strategy is used to fuse thermodynamic and wave dynamic characteristics, and the weighting coefficients are adjusted according to the statistical properties of the characteristic field.

[0023]

[0024] In this formula, α represents the thermodynamic characteristic weighting coefficient, and D w Indicates the dynamic characteristics of waves. σ represents the vertical average thermodynamic characteristics, p represents the wave characteristic weighting coefficient, and σ represents the wave characteristic weighting coefficient. 2 Variance calculation is performed, and a comprehensive meteorological feature matrix is ​​generated through linear combination:

[0025]

[0026] In this formula, M(i,j) represents the meteorological characteristic matrix, and D w Indicates the dynamic characteristics of waves. It represents the vertical average thermodynamic characteristics and effectively captures abrupt changes in boundary layer wind speed.

[0027] S2: Input the latitude and longitude coordinates and prevailing wind direction vector of each wind turbine in the wind farm to construct the position and wind direction feature matrix; use the Gaussian kernel function to perform a nonlinear transformation on the position and wind direction feature matrix to obtain the wake interference intensity matrix; combine the wake interference intensity matrix with the vertical wind speed profile to generate a three-dimensional wake interference field.

[0028] S3: Collect time-series data of salt spray concentration and relative humidity in the sea area where the wind farm is located, calculate the exponential decay integral of salt spray concentration under the influence of relative humidity, and obtain the time-varying corrosion accumulation; input the corrosion accumulation into the error function for transformation, generate a saturated environmental decay compensation term, and dynamically correct the sensor measurement values.

[0029] S4: Establish a wind speed probability density function model with dual modes, use Weibull distribution to describe the basic wind speed characteristics and Gaussian distribution to characterize sudden turbulence events, and dynamically adjust the distribution parameters based on meteorological feature matrix and three-dimensional wake interference field to generate wind speed probability distribution;

[0030] S5: Input the predicted wind speed into the wind turbine power characteristic curve function to calculate the theoretical power output value. Use the spatiotemporal adaptive quantile regression algorithm and combine it with the wind speed probability distribution to generate the probability interval for power prediction. Output a structured prediction result containing timestamp, predicted power value, lower limit and upper limit of the probability interval.

[0031] According to a specific implementation of an embodiment of the present invention, the specific steps of S2 are as follows:

[0032] Based on the spatial distribution of the wind turbine cluster and the prevailing wind direction, a wake interference field is constructed to accurately quantify the wake influence effect between wind turbines. The latitude and longitude coordinates of all wind turbines in the wind farm are input to form a location feature matrix.

[0033]

[0034] In this formula, P represents the wind turbine location matrix, storing the longitude of each wind turbine. and latitude Coordinates, where i represents the wind turbine index number, ranging from 1 to N, and N represents the total number of wind turbines in the wind farm. Combined with the prevailing wind direction vector, this indicates the main wind direction characteristics of the current sea area.

[0035]

[0036] In this formula, d represents the dominant wind direction vector. x and d y Let represent the components of wind direction in the latitude and longitude directions, respectively. Calculate the actual geographical distance matrix between wind turbines, considering the influence of Earth's curvature, and use the Haversine formula to calculate the spherical distance:

[0037]

[0038] In this formula, D ij This represents the spherical distance matrix between wind turbines, where r represents the Earth's radius. Indicates latitude difference, Indicates the difference in longitude. Represent the latitude of the i-th wind turbine, calculate the cosine matrix of the angle between wind direction and position, and quantify the spatial relationship between the wind turbine line and the wind direction:

[0039]

[0040] In this formula, C ij p represents the cosine matrix of wind direction and position. j -p i This represents the position vector pointing from wind turbine i to wind turbine j. Indicates the dominant wind direction vector. This involves performing vector normalization to construct a location and wind direction feature matrix, and fusing distance and angle information.

[0041] R ij =D ij ×σ(C ij )

[0042] In this formula, R ij The matrix represents the location and wind direction features, σ represents the sigmoid activation function, and C... ij D represents the cosine matrix of wind direction and position. ij The matrix representing the spherical distance between wind turbines is used, and the intensity of wake interference is quantified using a Gaussian kernel function to simulate the attenuation characteristics of the wake effect.

[0043]

[0044] In this formula, K ij Represents the wake interference intensity matrix. This represents the position parameter of the Gaussian kernel center. R represents the Gaussian kernel width parameter. ij The matrix representing location and wind direction characteristics is based on the vertical wind speed profile data V. z A three-dimensional wake interference field is generated, mapping the influence of the wind turbine's position onto a spatial grid:

[0045]

[0046] In this formula, This represents the three-dimensional wake interference field, where N represents the total number of wind turbines in the wind farm, x and y represent spatial grid coordinate indices, and z represents the vertical height layer index. and These represent the latitude and longitude of each wind turbine, K. ij The matrix represents the wake interference intensity matrix, and δ represents the Dirac function. Spatial discretization is achieved through the nearest neighbor approximation of the grid. The resulting three-dimensional interference field comprehensively represents the spatial distribution characteristics of the wake effect of the wind turbine group.

[0047] According to a specific implementation of an embodiment of the present invention, the specific steps of S3 are as follows:

[0048] To address the sensor degradation issue caused by marine salt spray environment, a dynamic compensation mechanism was constructed; time-series data of salt spray concentration in the sea area where the wind farm is located were collected. and relative humidity time series data h t ∈[0,100], where c t The concentration of salt spray at time point t, h t The relative humidity at time point t is represented, where t is the time index, ranging from 1 to T, and T is the length of the time series. The cumulative amount of salt spray corrosion is calculated, representing the time-varying corrosion effect of salt spray under humidity conditions.

[0049]

[0050] In this formula, A t The cumulative amount of time-varying corrosion is represented by T0, the cumulative time window is α, the humidity influence coefficient is γ, the decay coefficient is τ, the integration time variable is c, and the salt spray concentration is c. τ Based on the corrosion source, exp(ah) τ The accelerating effect of humidity on corrosion is quantified, and exp(-γ(t-τ)) represents the exponential decay characteristic of the corrosion effect. The cumulative corrosion amount is input into the error function and subjected to a nonlinear transformation to generate an environmental decay compensation term.

[0051] g t =tanh(W g ·[h t-1 ,x t ]+b g+κ·erf(dA t ))

[0052] In this formula, g t W represents the environmental degradation compensation term. g and b g Let A represent the trainable weights and bias parameters, respectively. t The cumulative amount of time-varying corrosion is represented by d, the scaling factor is represented by κ, the compensation intensity coefficient is represented by erf, and the Gaussian error function is represented by h. t-1 x represents the hidden state of the sensor at the previous time step. t To represent the original sensor measurements, a salt spray gating unit is designed to achieve dynamic correction of the sensor data.

[0053]

[0054] In this formula, x represents the corrected sensor data. t Represents the original sensor measurement value, g t This represents the environmental degradation compensation term.

[0055] According to a specific implementation of an embodiment of the present invention, the specific steps of S4 are as follows:

[0056] A dual-mode wind speed probability distribution model is established based on the meteorological feature matrix M(i,j) and the three-dimensional wake interferometry field to determine the marine meteorological characteristics and wake influence. At each location (i, j) in the spatial grid, four types of distribution parameters are extracted and calculated; the Weibull distribution parameters are generated through a feature mapping function to learn the nonlinear relationship between meteorological characteristics and wind speed statistical properties.

[0057]

[0058] In this formula, λ ij k represents the Weibull distribution scale parameter, reflecting the magnitude of the characteristic wind speed. ij This represents the shape parameter of the Weibull distribution, controlling the distribution skewness. and The parameter mapping function, with Gaussian distributed parameters directly derived from the wake interference field, represents the instantaneous fluctuation characteristics of wind speed.

[0059]

[0060] In this formula, μ ij σ represents the mean of a Gaussian distribution. ij The standard deviation of the Gaussian distribution is used to quantify the turbulence intensity. Representing the three-dimensional wake interference field, where l is the vertical height layer, calculate the dual-mode weighting coefficients and dynamically balance the contribution ratios of the two distributions:

[0061]

[0062] In this formula, φ ij w represents the weight coefficients of the Weibull distribution, with a value range of [0,1]. φ and b φ Let M(i,j) represent the trainable parameters, M(i,j) represent the meteorological feature matrix, and construct the mixture probability density function, defined at each spatial location (i,j):

[0063]

[0064] In this formula, f V Let v represent the wind speed probability density function, and μ represent the wind speed variable. ij σ represents the mean of a Gaussian distribution. ij φ represents the standard deviation of the Gaussian distribution, which quantifies the turbulence intensity. ij λ represents the weighting coefficients of the Weibull distribution. ij k represents the scaling parameter of the Weibull distribution. ij The shape parameters of the Weibull distribution are represented by a wake interference correction term, which addresses the distribution distortion in the vicinity of the wind turbine. The wake effect causes the wind speed distribution to exhibit a bimodal characteristic.

[0065]

[0066] In this formula, This represents the corrected probability density function. Let Frobenius norm be the gradient of the wake interference field, where Let η represent the spatial rate of change of the wake effect, ζ represent the correction magnitude coefficient, ζ represent the gradient scaling factor, and v represent the wind speed variable. During parameter training, the Weibull parameter mapping network is optimized through maximum likelihood estimation.

[0067]

[0068] In this formula, The historical wind speed is represented by W and u, which represent the weight matrix and bias vector of the neural network, respectively. H represents the number of training data points, and M represents the historical wind speed. t (i,j) represents the meteorological feature matrix at time point t. The training uses the Adam optimizer with a batch size of 128, 5000 iterations, and a dual-modal weight parameter w. φ and b φ Optimization is achieved by minimizing the Kolmogorov-Smirnov statistic:

[0069] min||F emp (v)-F model (v)|| ∞

[0070] In this formula, F emp Let F represent the empirical cumulative distribution function. model Let v represent the cumulative distribution function of the model, and v represent the wind speed variable.

[0071] According to a specific implementation of an embodiment of the present invention, the specific steps of S5 are as follows:

[0072] The wind speed probability distribution is transformed into wind power prediction results, and point prediction and probability interval output are achieved through wind turbine power characteristic curves and spatiotemporal adaptive quantile regression; based on the wind speed probability distribution... The power characteristic function of the wind turbine, P(v), adopts a four-parameter model:

[0073]

[0074] In this formula, q represents the maximum power coefficient, b represents the transition zone slope parameter, c represents the rated wind speed threshold, s represents the power curve shape exponent, and v represents the wind speed variable. The desired power value at spatial location (i, j) is calculated. This is achieved through a weighted integral of wind speed probability density:

[0075]

[0076] In this formula, v min Indicates the wind speed at which the fan cuts in, v max Representing the wind turbine cutoff speed, a spatiotemporal adaptive quantile regression model is constructed to generate the probability interval for power prediction, and the quantile function is designed:

[0077]

[0078] In this formula, Q g This represents the g-quantile power prediction, where g represents the quantile value, β0 represents the basic quantile parameter, and Φ k Represents the spatiotemporal feature weighting function. Let I represent the spatiotemporal feature tensor, and let I represent the time-varying sensitivity coefficient. It consists of the following:

[0079]

[0080] In this formula, σ V The quantile parameter represents the turbulence intensity and is optimized using a sliding time window.

[0081]

[0082] In this formula, ρ g T represents the quantile loss function.w Indicates the length of the time window. Representing historical observed power values, the system generates structured prediction results, including timestamps, predicted power values, and probability intervals.

[0083]

[0084] In this formula, t represents the timestamp, and x and x' represent the spatial latitude and longitude grid coordinates. Q represents the desired power value. 0.05 Q represents the 5th percentile power lower limit. 0.95 This represents the 95th percentile power limit. The results are sorted in ascending order by time, and the spatial resolution is consistent with the input grid. Attached Figure Description

[0085] Figure 1 The flowchart shows a wind power prediction method based on marine meteorological characteristics and wake interference.

[0086] Figure 2 This is a diagram illustrating the architecture of a wind power prediction method based on marine meteorological characteristics and wake interference. Detailed Implementation

[0087] To facilitate understanding and implementation of the present invention by those skilled in the art, the present invention will be further described in detail below with reference to examples and accompanying drawings.

[0088] As attached Figure 1 and attached Figure 2 As shown, a wind power prediction method based on marine meteorological characteristics and wake interference includes the following steps:

[0089] Step 1: Acquire sea surface temperature field data, vertical wind speed profile data, and wave height field data for the target sea area; calculate the spatial gradient of the sea surface temperature field in the longitude and latitude directions, and multiply it with the first-order shear of the vertical wind speed; perform time integration on the gradient of the wave height field and generate wave dynamic features through arctangent function transformation, forming a meteorological feature matrix. Specifically, acquire the sea surface temperature field data of the target sea area, represented as a two-dimensional matrix. Where m represents the number of grid points in the longitude direction and n represents the number of grid points in the latitude direction; the spatial gradient of the temperature field is accurately calculated using the central difference method. The temperature change rate distribution is obtained by calculating the temperature difference between adjacent grid points in both the longitude and latitude directions and dividing it by the grid spacing.

[0090]

[0091] In this formula, i = 2, ..., m-1 and j = 2, ..., n-1 represent the point indices of longitude and latitude, respectively, and T s (i,j+1)-T s(i,j-1) and T s (i+1,j)-T s (i-1,j) represent the temperature difference between adjacent grid points at latitude and longitude, respectively, and Δz and Δy represent the grid spacing at latitude and longitude, respectively. This represents the gradient matrix along the longitude direction. This represents the gradient matrix along the latitudinal direction and also processes vertical wind speed profile data. Where l represents the vertical height layer, the wind speed change rate between adjacent height layers is calculated to obtain the vertical wind speed shear characteristics:

[0092]

[0093] In this formula, S z V represents the characteristics of vertical wind speed shear. z (i,j,k) represents the vertical wind speed profile data at the ith latitude grid and the ith longitude grid, at the ith layer height. The second derivative is calculated using the central difference scheme.

[0094]

[0095] In this formula, Let Δz represent the second derivative of vertical wind speed, Δz represent the vertical interlayer spacing, and k = 1, ..., l-1 represent the vertical height layer index, forming a three-dimensional feature field. Combined with sea surface temperature gradient and vertical wind speed characteristics, a three-dimensional thermodynamic feature field is generated. Feature fusion is achieved using element-wise multiplication.

[0096]

[0097] In this formula, ° represents the Hadamard product, and T s (i,j) represents the sea surface temperature field data of the i-th longitude grid and the j-th latitude grid, Γ(i,j,k) represents the air-sea thermal and dynamic interaction effect at a specific location and altitude layer, and Γ represents the thermal and dynamic characteristic field; for wave height field data... Calculate the spatial gradient field:

[0098]

[0099] In this formula, Δx and Δy represent the latitude and longitude grid spacing, respectively. Let W(i,j+1)-W(i,j-1) represent the wave height gradient field, where W(i,j+1)-W(i,j-1) represent wave height field data at the same latitude but with a longitude difference of two grids. The gradient magnitude is quantified using the Frobenius norm.

[0100]

[0101] In this formula, ||·|| FLet Frobenius norm be denoted, Δt be the time window, and τ be the time variable. This represents the gradient time integral, which, through arctangent transformation, constrains the range of eigenvalues ​​and generates stable wave dynamic features.

[0102]

[0103] In this formula, D w Indicates the dynamic characteristics of waves. The gradient time integral is represented by arctan, which represents the arctangent function. This averages and compresses the thermodynamic feature field along the vertical dimension, reducing the three-dimensional feature field to a two-dimensional plane.

[0104]

[0105] In this formula, The vertical average thermodynamic characteristics are represented by Γ(i,j,k), where l represents the vertical height layer and Γ(i,j,k) represents the air-sea thermodynamic and dynamic interaction effect at a specific location and height layer. A variance-based dynamic weighting strategy is used to fuse thermodynamic and wave dynamic characteristics, and the weighting coefficients are adjusted according to the statistical properties of the characteristic field.

[0106]

[0107] In this formula, α represents the thermodynamic characteristic weighting coefficient, and D w Indicates the dynamic characteristics of waves. σ represents the vertical average thermodynamic characteristics, p represents the wave characteristic weighting coefficient, and σ represents the wave characteristic weighting coefficient. 2 Variance calculation is performed, and a comprehensive meteorological feature matrix is ​​generated through linear combination:

[0108]

[0109] In this formula, M(i,j) represents the meteorological characteristic matrix, and D w Indicates the dynamic characteristics of waves. It represents the vertical average thermodynamic characteristics and effectively captures abrupt changes in boundary layer wind speed.

[0110] Step 2: Construct a wake interference field based on the spatial distribution of the wind turbine cluster and the prevailing wind direction to accurately quantify the wake influence effect between wind turbines. Input the latitude and longitude coordinates of all wind turbines in the wind farm to form a location feature matrix:

[0111]

[0112] In this formula, P represents the wind turbine location matrix, storing the longitude of each wind turbine. and latitude Coordinates, where i represents the wind turbine index number, ranging from 1 to N, and N represents the total number of wind turbines in the wind farm. Combined with the prevailing wind direction vector, this indicates the main wind direction characteristics of the current sea area.

[0113]

[0114] In this formula, d represents the dominant wind direction vector. x and d y Let represent the components of wind direction in the latitude and longitude directions, respectively. Calculate the actual geographical distance matrix between wind turbines, considering the influence of Earth's curvature, and use the Haversine formula to calculate the spherical distance:

[0115]

[0116] In this formula, D ij This represents the spherical distance matrix between wind turbines, where r represents the Earth's radius. Indicates latitude difference, Indicates the difference in longitude. Represent the latitude of the i-th wind turbine, calculate the cosine matrix of the angle between wind direction and position, and quantify the spatial relationship between the wind turbine line and the wind direction:

[0117]

[0118] In this formula, C ij p represents the cosine matrix of wind direction and position. j -p i This represents the position vector pointing from wind turbine i to wind turbine j. Indicates the dominant wind direction vector. This involves performing vector normalization to construct a location and wind direction feature matrix, and fusing distance and angle information.

[0119] R ij =D ij ×σ(C ij )

[0120] In this formula, R ij The matrix represents the location and wind direction features, σ represents the sigmoid activation function, and C... ij D represents the cosine matrix of wind direction and position. ij The matrix representing the spherical distance between wind turbines is used, and the intensity of wake interference is quantified using a Gaussian kernel function to simulate the attenuation characteristics of the wake effect.

[0121]

[0122] In this formula, K ij Represents the wake interference intensity matrix. This represents the position parameter of the Gaussian kernel center. R represents the Gaussian kernel width parameter. ij The matrix representing location and wind direction characteristics is based on the vertical wind speed profile data V. z A three-dimensional wake interference field is generated, mapping the influence of the wind turbine's position onto a spatial grid:

[0123]

[0124] In this formula, This represents the three-dimensional wake interference field, where N represents the total number of wind turbines in the wind farm, x and y represent spatial grid coordinate indices, and z represents the vertical height layer index. and These represent the latitude and longitude of each wind turbine, K. ij δ represents the wake interference intensity matrix, and δ represents the Dirac function. Spatial discretization is achieved through grid nearest neighbor approximation. The generated three-dimensional interference field comprehensively represents the spatial distribution characteristics of the wake effect of the wind turbine group.

[0125] Step 3: To address the sensor attenuation issue caused by marine salt spray environment, a dynamic compensation mechanism is constructed; time-series data of salt spray concentration in the sea area where the wind farm is located are collected. and relative humidity time series data h t ∈[0,100], where c t The concentration of salt spray at time point t, h t The relative humidity at time point t is represented, where t is the time index, ranging from 1 to T, and T is the length of the time series. The cumulative amount of salt spray corrosion is calculated, representing the time-varying corrosion effect of salt spray under humidity conditions.

[0126]

[0127] In this formula, A t The cumulative amount of time-varying corrosion is represented by T0, the cumulative time window is α, the humidity influence coefficient is γ, the decay coefficient is τ, the integration time variable is c, and the salt spray concentration is c. τ Based on the corrosion source, exp(ah) τ The accelerating effect of humidity on corrosion is quantified, and exp(-γ(t-τ)) represents the exponential decay characteristic of the corrosion effect. The cumulative corrosion amount is input into the error function and subjected to a nonlinear transformation to generate an environmental decay compensation term.

[0128] g t =tanh(W g ·[h t-1 ,x t ]+b g +κ·erf(dA t ))

[0129] In this formula, gt W represents the environmental degradation compensation term. g and b g Let A represent the trainable weights and bias parameters, respectively. t The cumulative amount of time-varying corrosion is represented by d, the scaling factor is represented by κ, the compensation intensity coefficient is represented by erf, and the Gaussian error function is represented by h. t-1 x represents the hidden state of the sensor at the previous time step. t To represent the original sensor measurements, a salt spray gating unit is designed to achieve dynamic correction of the sensor data.

[0130]

[0131] In this formula, x represents the corrected sensor data. t Represents the original sensor measurement value, g t This indicates an environmental degradation compensation term;

[0132] Step 4: Establish a dual-mode wind speed probability distribution model based on marine meteorological characteristics and wake influence, using the meteorological feature matrix M(i,j) and the three-dimensional wake interferometric field. At each location (i, j) in the spatial grid, four types of distribution parameters are extracted and calculated; the Weibull distribution parameters are generated through a feature mapping function to learn the nonlinear relationship between meteorological characteristics and wind speed statistical properties.

[0133]

[0134] In this formula, λ ij k represents the Weibull distribution scale parameter, reflecting the magnitude of the characteristic wind speed. ij This represents the shape parameter of the Weibull distribution, controlling the distribution skewness. and The parameter mapping function, with Gaussian distributed parameters directly derived from the wake interference field, represents the instantaneous fluctuation characteristics of wind speed.

[0135]

[0136] In this formula, μ ij σ represents the mean of a Gaussian distribution. ij The standard deviation of the Gaussian distribution is used to quantify the turbulence intensity. Representing the three-dimensional wake interference field, where l is the vertical height layer, calculate the dual-mode weighting coefficients and dynamically balance the contribution ratios of the two distributions:

[0137]

[0138] In this formula, φ ij w represents the weight coefficients of the Weibull distribution, with a value range of [0,1]. φand b φ Let M(i,j) represent the trainable parameters, M(i,j) represent the meteorological feature matrix, and construct the mixture probability density function, defined at each spatial location (i,j):

[0139]

[0140] In this formula, f V Let v represent the wind speed probability density function, and μ represent the wind speed variable. ij σ represents the mean of a Gaussian distribution. ij φ represents the standard deviation of the Gaussian distribution, which quantifies the turbulence intensity. ij λ represents the weighting coefficients of the Weibull distribution. ij k represents the scaling parameter of the Weibull distribution. ij The shape parameters of the Weibull distribution are represented by a wake interference correction term, which addresses the distribution distortion in the vicinity of the wind turbine. The wake effect causes the wind speed distribution to exhibit a bimodal characteristic.

[0141]

[0142] In this formula, This represents the corrected probability density function. Let Frobenius norm be the gradient of the wake interference field, where Let η represent the spatial rate of change of the wake effect, ζ represent the correction magnitude coefficient, ζ represent the gradient scaling factor, and v represent the wind speed variable. During parameter training, the Weibull parameter mapping network is optimized through maximum likelihood estimation.

[0143]

[0144] In this formula, The historical wind speed is represented by W and u, which represent the weight matrix and bias vector of the neural network, respectively. H represents the number of training data points, and M represents the historical wind speed. t (i,j) represents the meteorological feature matrix at time point t. The training uses the Adam optimizer with a batch size of 128, 5000 iterations, and a dual-modal weight parameter w. φ and b φ Optimization is achieved by minimizing the Kolmogorov-Smirnov statistic:

[0145] min||F emp (v)-F model (v)|| ∞

[0146] In this formula, F emp Let F represent the empirical cumulative distribution function. model Let v represent the cumulative distribution function of the model, and v represent the wind speed variable.

[0147] Step 5: Convert the wind speed probability distribution into wind power prediction results, and achieve point prediction and probability interval output through wind turbine power characteristic curves and spatiotemporal adaptive quantile regression; based on the wind speed probability distribution The power characteristic function of the wind turbine, P(v), adopts a four-parameter model:

[0148]

[0149] In this formula, q represents the maximum power coefficient, b represents the transition zone slope parameter, c represents the rated wind speed threshold, s represents the power curve shape exponent, and v represents the wind speed variable. The desired power value at spatial location (i, j) is calculated. This is achieved through a weighted integral of wind speed probability density:

[0150]

[0151] In this formula, v min Indicates the wind speed at which the fan cuts in, v max Representing the wind turbine cutoff speed, a spatiotemporal adaptive quantile regression model is constructed to generate the probability interval for power prediction, and the quantile function is designed:

[0152]

[0153] In this formula, Q g This represents the g-quantile power prediction, where g represents the quantile value, β0 represents the basic quantile parameter, and Φ k Represents the spatiotemporal feature weighting function. Let I represent the spatiotemporal feature tensor, and let I represent the time-varying sensitivity coefficient. It consists of the following:

[0154]

[0155] In this formula, σ V The quantile parameter represents the turbulence intensity and is optimized using a sliding time window.

[0156]

[0157] In this formula, ρ g T represents the quantile loss function. w Indicates the length of the time window. Representing historical observed power values, the system generates structured prediction results, including timestamps, predicted power values, and probability intervals.

[0158]

[0159] In this formula, t represents the timestamp, and x and x' represent the spatial latitude and longitude grid coordinates. Q represents the desired power value. 0.05 Q represents the 5th percentile power lower limit. 0.95 This represents the 95th percentile power limit. The results are sorted in ascending order by time, and the spatial resolution is consistent with the input grid.

Claims

1. A wind power prediction method based on marine meteorological characteristics and wake interference, characterized in that... Includes the following steps: S1: Acquire sea surface temperature field data, vertical wind speed profile data, and wave height field data for the sea area; calculate the spatial gradient of the sea surface temperature field in the longitude and latitude directions; perform time integration on the gradient of the wave height field and generate wave dynamic features through arctangent function transformation to form a meteorological feature matrix. Specifically, acquire the sea surface temperature field data of the target sea area, represented as a two-dimensional matrix. Where m represents the number of grid points in the longitude direction and n represents the number of grid points in the latitude direction; The spatial gradient of the temperature field is accurately calculated using the central difference method. The temperature change rate distribution is obtained by calculating the temperature difference between adjacent grid points in both the longitude and latitude directions and dividing by the grid spacing. In this formula, i = 2, ..., m-1 and j = 2, ..., n-1 represent the point indices of longitude and latitude, respectively, and T s (i,j+1)-T s (i,j-1) and T s (i+1,j)-T s (i-1,j) represent the temperature difference between adjacent grid points at latitude and longitude, respectively, and Δx and Δy represent the grid spacing at latitude and longitude, respectively. This represents the gradient matrix along the longitude direction. This represents the gradient matrix along the latitudinal direction and also processes vertical wind speed profile data. Where l represents the vertical height layer, the wind speed change rate between adjacent height layers is calculated to obtain the vertical wind speed shear characteristics: In this formula, S z V represents the characteristics of vertical wind speed shear. z (i,j,k) represents the vertical wind speed profile data at the ith latitude grid and the ith longitude grid, at the ith layer height. The second derivative is calculated using the central difference scheme. In this formula, Let Δz represent the second derivative of vertical wind speed, Δz represent the vertical interlayer spacing, and k = 1, ..., l-1 represent the vertical height layer index, forming a three-dimensional feature field. Combined with sea surface temperature gradient and vertical wind speed characteristics, a three-dimensional thermodynamic feature field is generated. Feature fusion is achieved using element-wise multiplication. In this formula, T represents the Hadamard product. s (i,j) represents the sea surface temperature field data of the i-th longitude grid and the j-th latitude grid, Γ(i,j,k) represents the air-sea thermal and dynamic interaction effect at a specific location and altitude layer, and Γ represents the thermal and dynamic characteristic field; for wave height field data... Calculate the spatial gradient field: In this formula, Δx and Δy represent the latitude and longitude grid spacing, respectively. Let W(i,j+1)-W(i,j-1) represent the wave height gradient field, where W(i,j+1)-W(i,j-1) represent wave height field data at the same latitude but with a longitude difference of two grids. The gradient magnitude is quantified using the Frobenius norm. In this formula, ||·|| F Let Frobenius norm be denoted, Δt be the time window, and τ be the time variable. This represents the gradient time integral, which, through arctangent transformation, constrains the range of eigenvalues ​​and generates stable wave dynamic features. In this formula, D w Indicates the dynamic characteristics of waves. The gradient time integral is represented by arctan, which represents the arctangent function. This averages and compresses the thermodynamic feature field along the vertical dimension, reducing the three-dimensional feature field to a two-dimensional plane. In this formula, The vertical average thermodynamic characteristics are represented by Γ(i,j,k), where l represents the vertical height layer and Γ(i,j,k) represents the air-sea thermodynamic and dynamic interaction effect at a specific location and height layer. A variance-based dynamic weighting strategy is used to fuse thermodynamic and wave dynamic characteristics, and the weighting coefficients are adjusted according to the statistical properties of the characteristic field. In this formula, α represents the thermodynamic characteristic weighting coefficient, and D w Indicates the dynamic characteristics of waves. σ represents the vertical average thermodynamic characteristics, p represents the wave characteristic weighting coefficient, and σ represents the wave characteristic weighting coefficient. 2 Variance calculation is performed, and a comprehensive meteorological feature matrix is ​​generated through linear combination: In this formula, M(i,j) represents the meteorological characteristic matrix, and D w Indicates the dynamic characteristics of waves. It represents the vertical average thermodynamic characteristics and effectively captures abrupt changes in boundary layer wind speed. S2: Input the latitude and longitude coordinates and prevailing wind direction vector of each wind turbine in the wind farm to construct the position and wind direction feature matrix; use the Gaussian kernel function to perform a nonlinear transformation on the position and wind direction feature matrix to obtain the wake interference intensity matrix; combine the wake interference intensity matrix with the vertical wind speed profile to generate a three-dimensional wake interference field. S3: Collect time-series data of salt spray concentration and relative humidity in the sea area where the wind farm is located, calculate the exponential decay integral of salt spray concentration under the influence of relative humidity, and obtain the time-varying corrosion accumulation; input the corrosion accumulation into the error function for transformation, generate a saturated environmental decay compensation term, and dynamically correct the sensor measurement values. S4: Establish a dual-modal wind speed probability density function model, using the Weibull distribution to describe the basic wind speed characteristics and the Gaussian distribution to characterize sudden turbulence events. Based on the meteorological feature matrix and the three-dimensional wake interference field, dynamically adjust the distribution parameters to generate the wind speed probability distribution. S5: Input the predicted wind speed into the wind turbine power characteristic curve function to calculate the theoretical power output value. Use the spatiotemporal adaptive quantile regression algorithm and combine it with the wind speed probability distribution to generate the probability interval for power prediction. Output a structured prediction result containing timestamp, predicted power value, lower limit and upper limit of the probability interval.

2. The wind power prediction method based on marine meteorological characteristics and wake interference as described in claim 1, characterized in that... The specific method for step S2 is as follows: Based on the spatial distribution of the wind turbine cluster and the prevailing wind direction, a wake interference field is constructed to accurately quantify the wake influence effect between wind turbines. The latitude and longitude coordinates of all wind turbines in the wind farm are input to form a location feature matrix. In this formula, P represents the wind turbine location matrix, storing the longitude of each wind turbine. and latitude Coordinates, where i represents the wind turbine index number, ranging from 1 to N, and N represents the total number of wind turbines in the wind farm. Combined with the prevailing wind direction vector, this indicates the main wind direction characteristics of the current sea area. In this formula, d represents the dominant wind direction vector. x and d y Let represent the components of wind direction in the latitude and longitude directions, respectively. Calculate the actual geographical distance matrix between wind turbines, considering the influence of Earth's curvature, and use the Haversine formula to calculate the spherical distance: In this formula, D ij This represents the spherical distance matrix between wind turbines, where r represents the Earth's radius. Indicates latitude difference, Indicates the difference in longitude. Represent the latitude of the i-th wind turbine, calculate the cosine matrix of the angle between wind direction and position, and quantify the spatial relationship between the wind turbine line and the wind direction: In this formula, C ij p represents the cosine matrix of wind direction and position. j -p i This represents the position vector pointing from wind turbine i to wind turbine j. This represents the dominant wind direction vector. This involves performing vector normalization to construct a location and wind direction feature matrix, and fusing distance and angle information. R ij =D ij ×σ(C ij ) In this formula, R ij The matrix represents the location and wind direction features, σ represents the sigmoid activation function, and C... ij D represents the cosine matrix of wind direction and position. ij The matrix representing the spherical distance between wind turbines is used, and the intensity of wake interference is quantified using a Gaussian kernel function to simulate the attenuation characteristics of the wake effect. In this formula, K ij This represents the wake interference intensity matrix. This represents the position parameter of the Gaussian kernel center. R represents the Gaussian kernel width parameter. ij The matrix representing location and wind direction characteristics is based on the vertical wind speed profile data V. z A three-dimensional wake interference field is generated, mapping the influence of the wind turbine's position onto a spatial grid: In this formula, This represents the three-dimensional wake interference field, where N represents the total number of wind turbines in the wind farm, x and y represent spatial grid coordinate indices, and z represents the vertical height layer index. and These represent the latitude and longitude of each wind turbine, K. ij The matrix represents the wake interference intensity matrix, and δ represents the Dirac function. Spatial discretization is achieved through the nearest neighbor approximation of the grid. The resulting three-dimensional interference field comprehensively represents the spatial distribution characteristics of the wake effect of the wind turbine group.

3. The wind power prediction method based on marine meteorological characteristics and wake interference as described in claim 1, characterized in that... The specific method in step S3 is as follows: To address the sensor degradation issue caused by marine salt spray environment, a dynamic compensation mechanism was constructed; time-series data of salt spray concentration in the sea area where the wind farm is located were collected. and relative humidity time series data h t ∈[0,100], where c t The concentration of salt spray at time point t, h t The relative humidity at time point t is represented, where t is the time index, ranging from 1 to T, and T is the length of the time series. The cumulative amount of salt spray corrosion is calculated, representing the time-varying corrosion effect of salt spray under humidity conditions. In this formula, A t The cumulative amount of time-varying corrosion is represented by T0, the cumulative time window is α, the humidity influence coefficient is γ, the decay coefficient is τ, the integration time variable is c, and the salt spray concentration is c. τ Based on the corrosion source, exp(ah) τ The accelerating effect of humidity on corrosion is quantified, and exp(-γ(t-τ)) represents the exponential decay characteristic of the corrosion effect. The cumulative corrosion amount is input into the error function and subjected to a nonlinear transformation to generate an environmental decay compensation term. g t =tanh(W g ·[h t-1 ,x t ]+b g +κ·erf(dA t )) In this formula, g t W represents the environmental degradation compensation term. g and b g Let A represent the trainable weights and bias parameters, respectively. t The cumulative amount of time-varying corrosion is represented by d, the scaling factor is represented by κ, the compensation intensity coefficient is represented by erf, and the Gaussian error function is represented by h. t-1 x represents the hidden state of the sensor at the previous time step. t To represent the original sensor measurements, a salt spray gating unit is designed to achieve dynamic correction of the sensor data. In this formula, x represents the corrected sensor data. t Represents the original sensor measurement value, g t This represents the environmental degradation compensation term.

4. The wind power prediction method based on marine meteorological characteristics and wake interference as described in claim 1, characterized in that... The specific steps in step S4 are as follows: A dual-mode wind speed probability distribution model is established based on the meteorological feature matrix M(i,j) and the three-dimensional wake interferometry field to determine the marine meteorological characteristics and wake influence. At each location (i, j) in the spatial grid, four types of distribution parameters are extracted and calculated; the Weibull distribution parameters are generated through a feature mapping function to learn the nonlinear relationship between meteorological characteristics and wind speed statistical properties. In this formula, λ ij k represents the Weibull distribution scale parameter, reflecting the magnitude of the characteristic wind speed. ij This represents the shape parameter of the Weibull distribution, controlling the distribution skewness. and The parameter mapping function, with Gaussian distributed parameters directly derived from the wake interference field, represents the instantaneous fluctuation characteristics of wind speed. In this formula, μ ij σ represents the mean of a Gaussian distribution. ij The standard deviation of the Gaussian distribution is used to quantify the turbulence intensity. Representing the three-dimensional wake interference field, where l is the vertical height layer, calculate the dual-mode weighting coefficients and dynamically balance the contribution ratios of the two distributions: In this formula, φ ij w represents the weight coefficients of the Weibull distribution, with a value range of [0,1]. φ and b φ Let M(i,j) represent the trainable parameters, M(i,j) represent the meteorological feature matrix, and construct the mixture probability density function, defined at each spatial location (i,j): In this formula, f V Let v represent the wind speed probability density function, and μ represent the wind speed variable. ij σ represents the mean of a Gaussian distribution. ij φ represents the standard deviation of the Gaussian distribution, which quantifies the turbulence intensity. ij λ represents the weighting coefficients of the Weibull distribution. ij k represents the scaling parameter of the Weibull distribution. ij The shape parameters of the Weibull distribution are represented by a wake interference correction term, which addresses the distribution distortion in the vicinity of the wind turbine. The wake effect causes the wind speed distribution to exhibit a bimodal characteristic. In this formula, This represents the corrected probability density function. Let Frobenius norm be the gradient of the wake interference field, where Let η represent the spatial rate of change of the wake effect, ζ represent the correction magnitude coefficient, ζ represent the gradient scaling factor, and v represent the wind speed variable. During parameter training, the Weibull parameter mapping network is optimized through maximum likelihood estimation. In this formula, The historical wind speed is represented by W and u, which represent the weight matrix and bias vector of the neural network, respectively. H represents the number of training data points, and M represents the historical wind speed. t (i,j) represents the meteorological feature matrix at time point t. The training uses the Adam optimizer with a batch size of 128, 5000 iterations, and a dual-modal weight parameter w. φ and b φ Optimization is achieved by minimizing the Kolmogorov-Smirnov statistic: min||F emp (v)-F model (v)|| ∞ In this formula, F emp Let F represent the empirical cumulative distribution function. model Let v represent the cumulative distribution function of the model, and v represent the wind speed variable.

5. The wind power prediction method based on marine meteorological characteristics and wake interference according to claim 1, characterized in that... The specific steps in step S5 are as follows: The wind speed probability distribution is transformed into wind power prediction results, and point prediction and probability interval output are achieved through wind turbine power characteristic curves and spatiotemporal adaptive quantile regression; based on the wind speed probability distribution... The power characteristic function of the wind turbine, P(v), adopts a four-parameter model: In this formula, q represents the maximum power coefficient, b represents the transition zone slope parameter, c represents the rated wind speed threshold, s represents the power curve shape exponent, and v represents the wind speed variable. The desired power value at spatial location (i, j) is calculated. This is achieved through a weighted integral of wind speed probability density: In this formula, v min Indicates the fan cut-in wind speed, v max Representing the wind turbine cutoff speed, a spatiotemporal adaptive quantile regression model is constructed to generate the probability interval for power prediction, and the quantile function is designed: In this formula, Q g This represents the g-quantile power prediction, where g represents the quantile value, β0 represents the basic quantile parameter, and Φ k Represents the spatiotemporal feature weighting function. Let I represent the spatiotemporal feature tensor, and let I represent the time-varying sensitivity coefficient. It consists of the following: In this formula, σ V The quantile parameter represents the turbulence intensity and is optimized using a sliding time window. In this formula, ρ g T represents the quantile loss function. w Indicates the length of the time window. Representing historical observed power values, the system generates structured prediction results, including timestamps, predicted power values, and probability intervals. In this formula, t represents the timestamp, and x and x' represent the spatial latitude and longitude grid coordinates. Q represents the desired power value. 0.05 Q represents the 5th percentile power lower limit. 0.95 This represents the 95th percentile power limit. The results are sorted in ascending order by time, and the spatial resolution is consistent with the input grid.

Images