A method for offshore wind farm transient propagation in front of a wind turbine

Abstract

The present application relates to offshore wind turbine wind field technical field, specifically to a kind of offshore wind turbine front end wind field transient extension method, it includes the following steps: one, wind field model construction: the global prior wind field model including offshore wind turbine impeller rotation surface is established;Two, two-phase characteristic decomposition: according to the modal energy proportion of extended reconstruction wind field, offshore wind turbine front end prior wind field is decomposed into time coefficient and space base vector space-time two-phase characteristics;Three, nonlinear mapping model construction: according to the memory inference of recurrent neural network to time series and the logical inference of function relationship between complex information, the nonlinear mapping relationship from offshore wind turbine limited wind speed measuring point to wind field time coefficient is established;Four, transient expansion: according to the time series information of offshore wind turbine limited wind speed measuring point, global wind field is synchronously, transiently expanded, the present application has the effect that the front end wind field of offshore wind turbine can be quickly and accurately extended and reconstructed.

Description

TECHNICAL FIELD

[0001] The present application relates to the offshore wind turbine wind field technical field, and particularly relates to a kind of offshore wind turbine front end wind field transient extension method. BACKGROUND

[0002] At present, offshore wind energy meets the major needs of national energy sustainable development strategy, and has become the energy reserve strategic point that the world countries compete to develop. Offshore wind turbine is driven by wind in wind field, and the power mechanical energy generated by the rotation of impeller is converted into electric energy. For offshore wind turbine, wind field is an important power source for wind energy conversion, and accurate evaluation of wind field is related to efficient use of offshore wind energy, accurate evaluation of wind turbine power and reasonable analysis of local stress of blade. At present, there are only a limited number of wind speed time history data of measuring points in offshore wind power single machine position, which cannot effectively represent the dynamic information of the whole machine position wind field. Therefore, it is of great significance to carry out offshore global wind field dynamic extension and establish the mapping relationship from limited wind speed measuring points to global wind field for accurate evaluation of offshore wind field and efficient use of wind power resources.

[0003] In offshore wind field environment, wind speed has typical nonlinear characteristics of space-time distribution, and the average wind speed changes exponentially with height, and the turbulence characteristics become more complex with the increase of height. At the same height, with the evolution of non-directional flow of wind speed, the wind speed at different horizontal positions also has different fluctuation trends, which aggravates the nonlinearity of wind field information processing.

[0004] In the above technical solution, when offshore wind turbine front end wind field transient extension is carried out, the existing calculation method is difficult to accurately extend and reconstruct the front end wind field of offshore wind turbine. SUMMARY

[0005] To solve the problems in the prior art, the present application provides a kind of offshore wind turbine front end wind field transient extension method.

[0006] The technical scheme of the present application is as follows:

[0007] The present application provides a kind of offshore wind turbine front end wind field transient extension method, including the following steps:

[0008] I. Wind field model construction: according to the average wind exponential distribution form and the random change trend of turbulence wind speed, the offshore wind turbine front end wind field model is constructed, the fluctuating wind mutual power spectrum in wind field is calculated, and the global prior wind field model including the rotation surface of offshore wind turbine impeller is established;

[0009] II. Two-phase characteristic decomposition: the space-time characteristics of offshore wind turbine wind field are separated by intrinsic orthogonal decomposition method, i.e. the offshore wind turbine front end prior wind field is decomposed into time coefficient and space base vector space-time two-phase characteristics according to the mode energy proportion of extended and reconstructed wind field;

[0010] III. Nonlinear mapping model construction: according to the recurrent neural network, a nonlinear mapping model of offshore wind field expansion is constructed, and a nonlinear mapping relationship from the limited wind speed measurement point of the offshore wind turbine to the time coefficient of the wind field is established according to the memory reasoning of the recurrent neural network on the time sequence and the logical reasoning on the function relationship between complex information;

[0011] IV. Transient expansion: according to the spatial modal characteristics separated by the wind field expansion nonlinear mapping model and the prior wind field model, the transient expansion reconstruction of the offshore wind turbine front-end wind field is carried out, and the global wind field is expanded synchronously and transiently according to the time sequence information of the offshore wind turbine limited wind speed measurement point.

[0012] The beneficial effects achieved by the present application are: the space-time two-phase characteristics of the offshore wind turbine front-end prior wind field model are separated by the characteristic decomposition method, the separated wind field space-time characteristics are more obvious, the information redundancy of the front-end wind field transient expansion is avoided, and the implicit function relationship from the offshore wind turbine measured wind speed point to the time coefficient in the separated space-time two-phase characteristics is established by the recurrent neural network, which solves the problem of the limitation caused by the insufficient number of wind speed measurement points in the global wind field transient expansion reconstruction of the offshore wind turbine, the accuracy of the global wind field speed expansion by the limited wind speed measurement point is further improved through the offshore wind turbine front-end wind field transient expansion method, the effectiveness and synchronization of the global wind field transient expansion are ensured, and the front-end wind field of the offshore wind turbine is quickly and accurately expanded and reconstructed.

[0013] Further, the wind field model construction further comprises: establishing a prior wind field model of the offshore wind turbine rotating plane front-end based on OpenFAST-TurbSim.

[0014] Through the above scheme, OpenFAST-TurbSim is a random, full-field, turbulent wind simulator, which is suitable for establishing a wind field model.

[0015] Further, the calculation of the fluctuating wind mutual power spectrum in the wind field comprises:

[0016] The average wind speed of any point is represented by an exponential wind profile model, and the wind profile model is

[0017]

[0018] In the formula, z is the height of the wind speed point to be solved, V hub is the reference wind speed at the hub of the wind turbine, z hub is the corresponding height, and alpha represents the wind shear factor.

[0019] According to the Kaimal spectrum, the simulation representation of the fluctuating wind speed spectrum is:

[0020]

[0021] where S u (ω) is the Kaimal auto-power spectral density function, V ave is the mean wind speed of the simulation point, ω is the frequency, L u is the scale parameter;

[0022] The correlation function of the wind speed time history at different positions is:

[0023]

[0024] where Δx, Δy, Δz are the spatial distance components in the x, y, z directions, respectively, V ave (z i ) and V ave (z j ) are the mean wind speeds of the simulation wind speed points; C x , C y , C z represent the attenuation coefficients of the spatial wind field in the x, y, z directions;

[0025] The fluctuating wind cross-power spectrum S ij (ω) is calculated based on the spatial correlation of different wind speed measurement points in the wind field as follows:

[0026]

[0027] where S ii (ω) and S jj (ω) are the fluctuating wind auto-power spectral density functions of the spatial simulation wind speed points i and j.

[0028] Through the above scheme, the wind speed in the wind field can be divided into two parts, the mean wind speed and the fluctuating wind speed. For offshore wind fields, the wind speed time histories at different positions in space do not exist independently, and are correlated with each other. The correlation mainly reflects two aspects: one is the correlation between the spatial distances, and the other is the correlation of the frequency characteristics between two points. The correlation function has a typical distance correlation, that is, the closer the distance between the wind speeds of two points in space, the higher the degree of correlation between them. At present, the harmonic superposition method is mostly used for the simulation of fluctuating wind speed time history, and the core idea is to equivalent the discrete spectrum to the target fluctuating wind speed spectrum.

[0029] Further, the establishing of the global prior wind field model of the impeller rotation surface of the offshore wind turbine includes:

[0030] The fluctuating wind cross-power spectrum S ij (ω) is calculated based on the phase angle difference of different wind speed measurement points as follows:

[0031]

[0032] where I is the imaginary unit; is the phase angle of the two points i and j, which can be expressed as:

[0033]

[0034] where Random is a random value between [-π, π]; f * is a dimensionless parameter;

[0035] The fluctuating wind power spectrum matrix is decomposed by Cholesky decomposition as:

[0036] S(ω) = H(ω)H * (ω) T

[0037] where H(ω) is the lower triangular matrix obtained by decomposition, H * (ω) T is the corresponding conjugate transpose matrix;

[0038] The fluctuating wind speed time history is calculated based on the superposition principle of trigonometric functions as:

[0039]

[0040] where k = 1, 2, …, n; l = 1, 2, …, N, n is the number of simulated wind speed points in space, N is the total amount of frequency division of the fluctuating wind spectrum, θ km is the amplitude angle of H km (ω l ), Δω = (ω u - ω d ) / N, ω u and ω d are the upper and lower cut-off frequencies of the simulated Kaimal spectrum, φ l is a random number in [0, 2π], ω l and ω ml are the corresponding frequency variable components, ω l and ω ml can be expressed as:

[0041]

[0042]

[0043] The equivalent fluctuating wind speed time history is obtained by fast Fourier transform as:

[0044]

[0045] where p = 0, 1, …, 2N 2 -1; q is the remainder of ; Gkm (qΔt) can be represented as:

[0046]

[0047] In the formula, B km (lΔw) can be represented as:

[0048]

[0049] In the formula, According to the above formula, the fluctuating wind field is generated, and the front-end wind field with a turbulence intensity of 12% is constructed as a prior wind field model.

[0050] According to the above scheme, from the field prediction data of the fluctuating wind speed time history, the fluctuating wind mutual power spectral density matrix in the offshore wind field is not completely symmetric positive definite, and the fluctuations of the fluctuating wind speed at different spatial positions are different, which requires considering the influence of the phase angle difference when simulating the fluctuating wind speed. Affected by the number of spatial wind speed simulation points and the number of fluctuating wind frequency partitions, the calculation amount of the fluctuating wind speed time history simulation using the harmonic superposition method is huge, and the equivalent using fast Fourier transform can reduce the calculation amount and improve the efficiency of the wind speed simulation. This step specifies the prior wind field data under the specified turbulence intensity as the basis data for feature decomposition and deep learning, which helps to provide a prior basis for building a subsequent wind field expansion nonlinear mapping model.

[0051] Further, the decomposition of the offshore wind turbine front-end prior wind field into time coefficients and spatial basis vectors in space-time two-phase characteristics according to the modal energy proportion of the expanded reconstructed wind field includes:

[0052] The spatial basis function of the fluctuating wind speed time history and the spatial position of the wind speed point are related, which is represented as:

[0053]

[0054] In the formula, σ i (t) represents the time characteristic vector of the prior wind field, and λ is the characteristic value corresponding to the spatial basis function; C(t, t') represents the correlation function of the fluctuating wind speed in the prior wind field, which can be represented as:

[0055] C(t, t') = ∫ A u'(h, t) · u'(h, t') dh

[0056] In the formula, A is the area of the impeller rotating region;

[0057] When the prior wind field model is mapped to the modal space, the time coefficient corresponding to the wind field can be represented as:

[0058] α i (t) = ∫ Ω u'(h, t) · ψi (h)dh

[0059] wherein, a i (t) is the wind field time coefficient corresponding to the wind field projected by the prior wind field model to the i-th order space basis vector;

[0060] Based on the obtained wind field time coefficient, the prior wind field can be decomposed into corresponding space-time two-phase characteristics by intrinsic orthogonal decomposition as:

[0061]

[0062] wherein, N represents the total order of the spatial decomposition of the prior wind field;

[0063]

[0064] Through the above scheme, on the basis of ensuring the energy proportion of the wind field mode, the complex characteristics in the prior wind field are separated into space-time two-phase characteristics, and the separated wind field characteristics are more obvious, which is helpful to establish the implicit function relationship of the limited wind speed measuring point to the time sequence characteristics of the prior wind field.

[0065] Further, the nonlinear mapping model construction comprises:

[0066] The space-time two-phase characteristics of the prior wind field are combined with the time sequence memory characteristics of the recurrent neural network to construct the implicit nonlinear function relationship of the limited wind speed measuring point to the time characteristics, and the memory reasoning mechanism of the recurrent neural network is represented as:

[0067]

[0068] wherein, g represents the relu activation function of the output layer of the recurrent neural network, f is the relu activation function of the hidden layer, and the relu activation function is represented as:

[0069]

[0070] wherein, x represents the summed output after the weight ratio of the neurons of the recurrent neural network, and based on the space-time two-phase characteristics of the prior wind field decomposition, the nonlinear mapping function of the recurrent neural network is combined to construct the wind field expansion nonlinear mapping model.

[0071] Through the above scheme, the prior wind field is decomposed into space-time two-phase characteristics, wherein the space characteristics are only related to the spatial position of each wind speed point in the wind field, and the time characteristics are both the same number of time sequences of the wind speed of the limited wind speed measuring point, and the implicit nonlinear function relationship of the limited wind speed measuring point to the time characteristics is constructed by combining the time sequence memory characteristics of the recurrent neural network.

[0072] Further, the spatio-temporal two-phase feature based on prior wind field decomposition, combined with the nonlinear mapping function of the recurrent neural network, constructs a wind field expansion nonlinear mapping model, which includes:

[0073] The spatio-temporal two-phase feature based on prior wind field decomposition selects four limited wind speed measurement points in accordance with the actual offshore wind turbine wind speed measurement equipment installation engineering as the wind field transient expansion data source.

[0074] The four limited wind speed measurement points are selected as the input of the recurrent neural network model, the time coefficient in the spatio-temporal two-phase feature is selected as the output of the recurrent neural network model, the mean square error between the predicted time coefficient and the true time coefficient is taken as the network loss function, and the wind field transient expansion model network structure is built.

[0075] The implicit nonlinear mapping relationship existing in the wind field transient expansion is learned and approximated.

[0076] A recurrent neural network model is established based on PyTorch, combined with the separated spatio-temporal two-phase feature in the prior wind field, to construct a wind field transient expansion nonlinear mapping model.

[0077] Through the above scheme, the wind field transient expansion nonlinear mapping model is constructed, the nonlinear mapping relationship between the limited wind speed measurement points in accordance with the actual offshore wind turbine installation and the time coefficient feature separated in the prior wind field is established, the chaotic nature of the limited wind speed measurement points to the prior wind field feature is eliminated by the recurrent neural network, and the time coefficient in the global wind field can be expanded through only the wind speed data of the four measurement points.

[0078] Further, the learning and approximation of the implicit nonlinear mapping relationship existing in the wind field transient expansion includes:

[0079] The first 800s of data in the 1000s wind speed time series data in the prior wind field are selected for recurrent neural network model training, and the last 200s are used to verify the accuracy of the network predicted time coefficient, wherein the recurrent neural network parameters are: the input layer dimension is set to 9, the hidden layer dimension is set to 26, the output layer dimension is set to 6, the linear layer is set to 3, the learning rate is set to 0.001, and the optimization rate is set to Adam.

[0080] Further, the transient expansion further includes:

[0081] The spatial mode in the spatio-temporal two-phase feature is selected, and the time coefficient predicted by the recurrent neural network is weighted and summed to perform offshore wind turbine front-end wind field transient expansion. The accuracy of the wind field transient expansion model in expanding wind speed at different points is verified by comparison with the conventional wind profile model.

[0082] The above scheme solves the complex mapping problem from the limited wind speed measuring point to the global wind field wind speed measuring point, and further improves the accuracy of the extension of the limited wind speed measuring point to the global wind field.

[0083] The offshore wind turbine front-end wind field transient extension method has the following advantages:

[0084] 1. The offshore wind turbine front-end prior wind field model is separated in time and space by feature decomposition, the separated wind field time and space features are more obvious, the information redundancy of the front-end wind field transient extension is avoided, and the recurrent neural network establishes an implicit function relationship between the offshore wind turbine measured wind speed point and the time coefficient in the separated time and space two-phase features, solves the limitation problem caused by the insufficient number of wind speed measuring points in the reconstruction of the offshore wind turbine global wind field transient extension, and further improves the accuracy of the extension of the global wind field wind speed from the limited wind speed measuring point through the offshore wind turbine front-end wind field transient extension method, ensures the effectiveness and synchronization of the global wind field transient extension, and thus quickly and accurately extends and reconstructs the front-end wind field of the offshore wind turbine. DETAILED DESCRIPTION

[0085] In order to facilitate those skilled in the art to understand the present application, the specific embodiments of the present application will be described below with reference to the accompanying drawings.

[0086] In the description of the present application, it should be noted that the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inner", "outer" and the like indicate the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present application and simplifying the description, and therefore cannot be understood as indicating or implying that the devices or elements referred to must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as limiting the present application. In addition, the terms "first", "second", "third" are only for descriptive purposes and cannot be understood as indicating or implying relative importance.

[0087] In the description of the present application, it should be noted that unless otherwise explicitly specified and limited, the terms "mounting", "connection", "connection" should be understood in a broad sense, for example, it can be fixedly connected, or it can be detachably connected, or integrally connected, it can be mechanically connected, or it can be electrically connected, it can be directly connected, or indirectly connected through an intermediate medium, and it can be the communication between two elements. For those skilled in the art, the specific meaning of the above terms in the present application can be understood according to the specific circumstances.

[0088] The present application provides an offshore wind turbine front-end wind field transient extension method, and the specific steps are as follows:

[0089] Step one, wind field model construction:

[0090] S100, according to the average wind index type distribution form and the random change trend of turbulent wind speed, the offshore wind turbine front end wind field model is constructed, the fluctuating wind mutual power spectrum in the wind field is calculated, and the global prior wind field model including the offshore wind turbine impeller rotating surface is established. The prior wind field model of the rotating plane front end of offshore wind turbine is established based on OpenFAST-TurbSim. OpenFAST-TurbSim is a random, full-field, turbulent wind simulator, which is suitable for establishing wind field model.

[0091] S101, the wind speed in the wind field can be divided into two parts: average wind speed and fluctuating wind speed. The average wind speed at any point is represented by an exponential wind profile model, and the wind profile model is:

[0092]

[0093] In the formula, z is the height of the wind speed point to be solved, V hub is the reference wind speed at the hub of the wind turbine, z hub is the corresponding height, and α represents the wind shear factor.

[0094] S102, the fluctuating wind speed refers to the wind speed fluctuation process within 10 minutes, and its change trend conforms to the stationary Gaussian process. At present, the Kaimal spectrum which conforms to the spatial and temporal variation law of offshore wind speed is specified in IEC-61400-3 standard to simulate the fluctuating wind speed spectrum. The simulation representation of the fluctuating wind speed spectrum obtained according to the Kaimal spectrum is:

[0095]

[0096] In the formula, S u (ω) is the Kaimal self-power spectrum density function, V ave is the average wind speed of the simulation point, ω is the frequency, and L u is the scale parameter.

[0097] S103, for offshore wind field, the wind speed time history at different positions in space does not exist independently, and they are correlated with each other. The correlation mainly reflects two aspects: one is the correlation between spatial distance, and the other is the correlation between the frequency characteristics of two points. The correlation function has a typical distance correlation, that is, the closer the distance between the wind speeds of two points in space, the higher the correlation degree. Therefore, the correlation function of wind speed time history at different positions is:

[0098]

[0099] In the formula, Δx, Δy, Δz are the spatial distance components in x, y, z directions respectively, V ave (zi ) and V ave (z j ) is the average wind speed of the simulated wind speed point; C x , C y , C z represent the attenuation coefficients of the spatial wind field in the x, y, and z directions.

[0100] S104, the simulation of the fluctuating wind speed time history is currently mostly carried out using the harmonic superposition method, the core idea of which is to equivalently convert the discrete spectrum into the target fluctuating wind speed spectrum. When the spatial correlation of different wind speed measuring points in the wind field is considered, the fluctuating wind cross-power spectral density matrix S ij (ω) is:

[0101]

[0102] In the formula, S ii (ω) and S jj (ω) are the fluctuating wind speed self-power spectral density functions of two points i and j of the spatial simulated wind speed.

[0103] S105, from the field prediction data of the fluctuating wind speed time history, the fluctuating wind cross-power spectral density matrix in the offshore wind field is not completely positive definite, and the fluctuation of the fluctuating wind speed at different spatial positions is different, which requires that the influence of the phase angle difference be considered when simulating the fluctuating wind. When the phase angle difference of different wind speed measuring points is considered, the fluctuating wind cross-power spectral density S ij (ω) is:

[0104]

[0105] In the formula, I is an imaginary unit; is the phase angle of two points i and j, can be expressed as:

[0106]

[0107] In the formula, Random is a random value in [-π, π]; f * is a dimensionless parameter.

[0108] S106, the fluctuating wind power spectral matrix is decomposed by Cholesky as:

[0109] S(ω) = H(ω)H * (ω) T

[0110] In the formula, H(ω) is the lower triangular matrix obtained by decomposition, H * (ω) T is the corresponding conjugate transpose matrix;

[0111] S107, the time history of fluctuating wind speed is calculated based on the superposition principle of trigonometric functions:

[0112]

[0113] where k = 1, 2, …, n; l = 1, 2, …, N, n is the number of simulated wind speed points in space, N is the total amount of fluctuating wind spectrum frequency division, θ km is the amplitude angle of H km (ω l ), Δω = (ω u - ω d ) / N, ω u and ω d are the upper and lower cut-off frequencies of the simulated Kaimal spectrum, φ l is a random number in [0, 2π], ω l and ω ml are the corresponding frequency variable components, and ω l and ω ml can be expressed as:

[0114]

[0115]

[0116] S108, influenced by the number of space wind speed simulation points and the number of fluctuating wind frequency division, the calculation amount of fluctuating wind speed time history simulation using the harmonic superposition method is huge. Fast Fourier transform can reduce the calculation amount and improve the efficiency of wind speed simulation. Through fast Fourier transform, the equivalent fluctuating wind speed time history is:

[0117]

[0118] where p = 0, 1, …, 2N 2 -1; q is the remainder of ; G km (qΔt) can be expressed as:

[0119]

[0120] where B km (lΔw) can be expressed as:

[0121]

[0122] where According to the above formula, the fluctuating wind field is generated, and the front-end wind field with a turbulence intensity of 12% is constructed as a prior wind field model.

[0123] Step one: Prior wind field data under specified turbulence intensity is taken as the basic data for feature decomposition and deep learning, which helps to provide a priori basis for subsequent wind field expansion nonlinear mapping model.

[0124] Step two, two-phase feature decomposition:

[0125] S201, the spatiotemporal characteristics of offshore wind field are separated by eigenvalue orthogonal decomposition method, that is, the offshore wind field prior to the front end is decomposed into time coefficient and space basis vector spatiotemporal two-phase characteristics according to the mode energy proportion of the expansion reconstruction wind field. The spatiotemporal characteristics of offshore wind field are complex, the wind speed time series fluctuates with turbulence, and the wind speed amplitude changes with the increase of spatial height. It is very difficult to directly use limited wind speed measurement points for global wind field mapping. The separation of spatiotemporal characteristics of wind field by eigenvalue orthogonal decomposition can effectively simplify the redundancy of offshore wind field expansion reconstruction at the front end.

[0126] S202, the correlation between the spatial basis function of fluctuating wind speed time history and the spatial position of wind speed point is expressed as:

[0127]

[0128] In the formula, σ i (t) represents the time characteristic vector of the prior wind field, λ is the characteristic value corresponding to the spatial basis function; C(t,t') represents the correlation function of the fluctuating wind speed in the prior wind field, which can be expressed as:

[0129] C(t,t') = ∫ A u'(h,t)·u'(h,t') dh

[0130] In the formula, A is the area of the impeller rotation region.

[0131] S203, when the prior wind field model is mapped to the modal space, the corresponding time coefficient of the wind field can be expressed as:

[0132] α i (t) = ∫ Ω u'(h,t)·ψ i (h) dh

[0133] In the formula, α i (t) is the wind field time coefficient corresponding to the projection of the prior wind field model to the i-th spatial basis vector.

[0134] S204, based on the obtained wind field time coefficient, the prior wind field can be decomposed into corresponding spatiotemporal two-phase characteristics by eigenvalue orthogonal decomposition:

[0135]

[0136] In the formula, N represents the total order of spatial decomposition of the prior wind field;

[0137] S205, for the prior wind field model, all orders can reconstruct all characteristics of complex wind field. But in actual engineering application, the energy ratio of high order mode in prior wind field is often small, and it represents more complex turbulent nonlinear characteristics, so under the condition of ensuring a certain energy ratio, the high order mode energy is often truncated, at this time the time and space two-phase characteristics of the prior wind field are properly reduced. After truncating the high order mode energy, the time and space two-phase characteristics of the prior wind field are:

[0138]

[0139] Based on the selected mode order, the time and space two-phase characteristics of the prior wind field can be separated, compared with the original wind speed time series in the wind field, the separated wind field characteristics are more obvious.

[0140] Step two, on the basis of ensuring the energy ratio of wind field mode, the complex characteristics in the prior wind field are separated into time and space two-phase characteristics, and the separated wind field characteristics are more obvious, which is helpful to establish the implicit function relationship between the limited wind speed measuring point and the time sequence characteristics of the prior wind field.

[0141] Step three, nonlinear mapping model construction:

[0142] S301, according to the recurrent neural network, the extended nonlinear mapping model of offshore wind field is constructed, and the nonlinear mapping relationship between the limited wind speed measuring point of offshore wind turbine and the time coefficient of wind field is established according to the memory reasoning of recurrent neural network to time sequence and the logical reasoning to the function relationship between complex information.

[0143] S302, deep learning network is a common tool for processing complex information mapping relationship, its core idea is to simulate artificial neuron structure to carry out information weight distribution, nonlinear mapping processing, etc., its structure can be divided into input layer, hidden layer and output layer. Recurrent neural network is good at processing time sequence information, and different from traditional deep learning network, it can remember the hidden layer characteristics of previous layers through the cycle layer. The spatial characteristics of prior wind field time and space two-phase characteristics combined with the memory characteristics of recurrent neural network to time sequence construct the implicit nonlinear function relationship between limited wind speed measuring point and time characteristics, and the memory reasoning mechanism of recurrent neural network is represented as:

[0144]

[0145] In the formula, g represents the relu activation function of the output layer of recurrent neural network, f is the relu activation function of the hidden layer, and the relu activation function is represented as:

[0146] relu(x) = max(0, x)

[0147] In the formula, x represents the sum output after the neural network neuron weight ratio of the recurrent neural network, the spatiotemporal two-phase characteristics based on the prior wind field decomposition, and the nonlinear mapping function of the recurrent neural network are combined to construct a wind field expansion nonlinear mapping model.

[0148] S303, based on the spatiotemporal two-phase characteristics of the prior wind field decomposition, four limited wind speed measurement points conforming to the actual offshore wind turbine wind speed measurement equipment installation engineering are selected as the wind field transient expansion data source.

[0149] S304, the selected four limited wind speed measurement points are used as the input of the recurrent neural network model, the time coefficient in the spatiotemporal two-phase characteristics is selected as the output of the recurrent neural network model, and the mean square error between the predicted time coefficient and the actual time coefficient is used as the network loss function to build the wind field transient expansion model network structure.

[0150] S305, the implicit nonlinear mapping relationship existing in the wind field transient expansion is learned and approximated.

[0151] S306, a recurrent neural network model is established based on PyTorch, the spatiotemporal two-phase characteristics separated in the prior wind field are combined, and a wind field transient expansion nonlinear mapping model is constructed. The first 800s of data in the 1000s wind speed time series data in the prior wind field are selected for recurrent neural network model training, and the last 200s are used for accuracy verification of the network prediction time coefficient. The recurrent neural network parameters are: the input layer dimension is set to 9, the hidden layer dimension is set to 26, the output layer dimension is set to 6, the linear layer is set to 3, the learning rate is set to 0.001, and the optimization rate is set to Adam.

[0152] Step three, by constructing the wind field transient expansion nonlinear mapping model, the nonlinear mapping relationship between the limited wind speed measurement points conforming to the actual offshore wind turbine installation and the time coefficient characteristics separated in the prior wind field is established. The chaotic nature of the limited wind speed measurement points to the prior wind field characteristics is eliminated by the recurrent neural network, and the time coefficient in the global wind field can be expanded by only four measurement points.

[0153] Step four, transient expansion:

[0154] S401, according to the spatial modal characteristics separated from the wind field expansion nonlinear mapping model and the prior wind field model, the offshore wind turbine front-end wind field is transiently expanded and reconstructed, and the global wind field is synchronized and transiently expanded according to the time series information of the offshore wind turbine limited wind speed measurement points.

[0155] S402, the spatial modal in the spatiotemporal two-phase characteristics is selected, and the time coefficient predicted by the recurrent neural network is weighted and summed to perform offshore wind turbine front-end wind field transient expansion. The accuracy of the wind field transient expansion model for different point wind speed expansion is verified by comparison with the conventional wind profile model.

[0156] The fourth step solves the complex mapping problem from the limited wind speed measuring point to the global wind field wind speed measuring point, the accuracy of the limited wind speed measuring point to the global wind field expansion is further improved, and the effectiveness and accuracy of the offshore wind turbine front-end wind field transient expansion are verified through comparison with the conventional wind speed expansion model.

[0157] The present application provides a kind of offshore wind turbine front-end wind field transient expansion method implementation principle is: by the way of characteristic decomposition, offshore wind turbine front-end prior wind field model is separated in time and space two-phase characteristics, the wind field time-space characteristics after separation is more obvious, avoids the information redundancy of front-end wind field transient expansion, while recurrent neural network establishes the implicit function relationship between offshore wind turbine measured wind speed point and the time coefficient in the separated time-space two-phase characteristics with engineering installation actual, solve the problem of the limitation caused by the insufficient number of wind speed measuring points in the reconstruction of offshore wind turbine global wind field transient expansion, through offshore wind turbine front-end wind field transient expansion method, the accuracy of global wind field wind speed expansion by limited wind speed measuring point is further improved, the effectiveness and synchronism of global wind field transient expansion are guaranteed, so as to quickly and accurately expand and reconstruct the front-end wind field of offshore wind turbine.

[0158] The above-mentioned embodiments of the present application do not constitute a limitation on the scope of protection of the present application. Any modifications, equivalent replacements and improvements made within the spirit and principles of the present application shall be included in the scope of protection of the claims of the present application.

Claims

1. A method of offshore wind farm transient propagation in front of a wind turbine, characterized in that, The method comprises the following steps: I. Wind field model construction: according to the average wind index type distribution form and the random variation trend of turbulent wind speed, a wind field model in front of an offshore wind turbine is constructed, the fluctuating wind mutual power spectrum in the wind field is calculated, and a global prior wind field model including a rotating surface of a wind turbine impeller is established; II. Two-phase characteristic decomposition: the space-time characteristics of the wind field of the offshore wind turbine are separated by the intrinsic orthogonal decomposition method, that is, the prior wind field in front of the offshore wind turbine is decomposed into space-time two-phase characteristics of time coefficients and spatial basis vectors according to the mode energy proportion of the extended reconstructed wind field; III. Nonlinear mapping model construction: an offshore wind field expansion nonlinear mapping model is constructed according to a recurrent neural network, and a nonlinear mapping relationship from the limited wind speed measurement point of the offshore wind turbine to the time coefficient of the wind field is established according to the memory reasoning of the recurrent neural network on the time series and the logical reasoning on the function relationship between complex information; IV. Transient expansion: the wind field in front of the offshore wind turbine is reconstructed by transient expansion according to the time coefficient predicted by the wind field expansion nonlinear mapping model and the spatial basis vector characteristics separated from the prior wind field model, and the global wind field is synchronously and transiently expanded according to the time series information of the limited wind speed measurement point of the offshore wind turbine.

2. A method of offshore wind farm transient propagation at the front end of a wind farm according to claim 1, characterized in that, The wind field model construction further comprises: establishing a prior wind field model in front of a rotating plane of the offshore wind turbine based on OpenFAST-TurbSim.

3. A method of offshore wind farm transient propagation at the front end of a wind farm according to claim 1, characterized in that, The calculation of the fluctuating wind mutual power spectrum in the wind field comprises: The average wind speed at any point is represented by an exponential wind profile model, and the wind profile model is where z is the height of the wind speed point to be solved, is the reference wind speed at the fan hub, is the corresponding height, and a represents the wind shear factor; The simulation representation of the fluctuating wind speed spectrum is obtained according to the Kaimal spectrum: wherein is the Kaimal's power spectral density function, is the mean wind speed of the simulated point, is the frequency, is the scale parameter; The correlation function of the wind speed time history at different positions is: wherein are the spatial distance components in the x, y, z directions, respectively, and is the average wind speed of the simulated wind speed points; , , represent the attenuation coefficients of the spatial wind field in the x, y, z directions. Based on the spatial correlation of different wind speed measuring points in the wind field, the fluctuating wind mutual power spectrum is calculated Is: wherein and is the power spectral density function of the fluctuating wind velocity between two points i and j in space.

4. A method of offshore wind farm transient propagation at the front end of a wind farm according to claim 3, characterized in that, The establishment of the global prior wind field model including the rotating surface of the wind turbine impeller comprises: Based on the phase angle difference of different wind speed measuring points to calculate the fluctuating wind mutual power spectrum Is: where I is the imaginary unit; the phase angle of the two points i and j, may be expressed as: In the formula, Random is a random value in [-π, π]; is a dimensionless parameter; The fluctuating wind power spectrum matrix is decomposed by Cholesky: In the formula, H(ω) is a lower triangular matrix obtained by decomposition, H * (ω) T is a corresponding conjugate transpose matrix; The fluctuating wind speed time history is calculated based on the superposition principle of trigonometric functions: where k = 1, 2, …, n; l = 1, 2, …, N, n is the number of simulated wind speed points in the space, and N is the total amount of frequency division of the fluctuating wind spectrum, is the amplitude angle, / N, and are the upper and lower cut-off frequencies of the simulated Kaimal spectrum, respectively, is a random number within [0, 2π], and are the corresponding frequency variable components, and can be expressed as: The equivalent fluctuating wind speed time history is obtained by fast Fourier transform: where p = 0, 1,..., 2 -1; q is the remainder; may be expressed as: wherein may be represented as: In the formula, According to the above formula, the pulsating wind field is generated, and the front-end wind field with a turbulence intensity of 12% is constructed as a prior wind field model.

5. A method of offshore wind farm transient propagation at the front end of a wind farm according to claim 1, characterized in that, The prior wind field in front of the offshore wind turbine is decomposed into space-time two-phase characteristics of time coefficients and spatial basis vectors according to the mode energy proportion of the extended reconstructed wind field, which comprises: The spatial basis function of the fluctuating wind speed time history is related to the spatial position of the wind speed point, and is represented as: wherein represents a time characteristic vector of the prior wind field, are eigenvalues corresponding to the spatial basis functions; represents a correlation function of the fluctuating wind speed in the prior wind field, which can be expressed as: In the formula, A is the area of the rotating region of the impeller; When the prior wind field model is mapped to the modal space, the time coefficient corresponding to the wind field can be represented as: In the formula, is the wind field time coefficient corresponding to the wind field model projected to the i-th order spatial basis vector. Based on the obtained wind field time coefficient, the prior wind field can be decomposed into corresponding space-time two-phase characteristics by intrinsic orthogonal decomposition as: In the formula, N represents the total order of the spatial decomposition of the prior wind field; After truncating the high-order modal energy, the space-time two-phase characteristics of the prior wind field are: 。 6. A method of offshore wind farm transient propagation at the front end of a wind farm according to claim 1, characterized in that, The nonlinear mapping model construction comprises: The spatial characteristics of the space-time two-phase characteristics of the prior wind field combine the memory characteristics of the recurrent neural network to construct an implicit nonlinear function relationship from the limited wind speed measurement point to the time characteristics, and the memory reasoning mechanism of the recurrent neural network is represented as: In the formula, g represents the relu activation function of the output layer of the recurrent neural network, f is the relu activation function of the hidden layer, and the relu activation function is represented as: In the formula, x represents the sum output after the weight ratio of the recurrent neural network neuron, the spatiotemporal two-phase feature based on the prior wind field decomposition, and the nonlinear mapping function of the recurrent neural network, a wind field expansion nonlinear mapping model is constructed.

7. A method of offshore wind farm transient propagation at the front end of a wind farm according to claim 6, characterized in that, The spatiotemporal two-phase feature based on the prior wind field decomposition, and the nonlinear mapping function of the recurrent neural network, a wind field expansion nonlinear mapping model is constructed. The spatiotemporal two-phase feature based on the prior wind field decomposition selects four limited wind speed measurement points in accordance with the actual offshore wind turbine wind speed measurement equipment installation engineering as the wind field transient expansion data source. The four limited wind speed measurement points are selected as the input of the recurrent neural network model, the time coefficient in the spatiotemporal two-phase feature is selected as the output of the recurrent neural network model, the mean square error between the predicted time coefficient and the real time coefficient is taken as the network loss function, and the wind field transient expansion model network structure is built. The implicit nonlinear mapping relationship existing in the wind field transient expansion is learned and approximated. A recurrent neural network model is established based on PyTorch, and a wind field transient expansion nonlinear mapping model is constructed based on the separated spatiotemporal two-phase feature in the prior wind field.

8. A method of offshore wind farm transient propagation at the front end of a wind farm according to claim 7, characterized in that, The implicit nonlinear mapping relationship existing in the wind field transient expansion is learned and approximated. The first 800s of data in the 1000s wind speed time series data in the prior wind field are selected for recurrent neural network model training, and the last 200s are used to verify the accuracy of the network prediction time coefficient, wherein the recurrent neural network parameters are: the input layer dimension is set to 9, the hidden layer dimension is set to 26, the output layer dimension is set to 6, the linear layer is set to 3, the learning rate is set to 0.001, and the optimization rate is set to Adam.

9. A method of offshore wind farm transient propagation at the front end of a wind farm according to claim 1, characterized in that, The transient expansion further includes: The spatial basis vector in the spatiotemporal two-phase feature is selected, and the time coefficient predicted by the recurrent neural network is weighted and summed to perform offshore wind turbine front-end wind field transient expansion, and the accuracy of the wind field transient expansion model for different point wind speed expansion is verified by comparison with the conventional wind profile model.

Images