Method, device and equipment for wind power prediction of floating offshore wind farm

Abstract

The application discloses a floating offshore wind farm wind power prediction method, device and equipment, relates to the wind power prediction technical field, and can solve the technical problem that it is difficult to accurately predict wind power in the floating offshore wind farm scene. Including: constructing a three-dimensional time wake model of a floating wind turbine according to an inflow wind speed, an inflow environmental turbulence intensity and a working condition to obtain a wake wind speed; acquiring wind power at a previous moment, determining an environmental state according to the inflow wind speed, the inflow environmental turbulence intensity, the working condition, the wind power at the previous moment and the wake wind speed; and predicting wind power based on a trained policy network to obtain target wind power at a current moment, wherein the policy network is constructed based on a trained convolutional neural network and a trained long short-term memory network.

Description

Technical Field

[0001] This invention relates to the field of wind power prediction technology, and in particular to a method, apparatus and equipment for predicting wind power in a floating offshore wind farm. Background Technology

[0002] Wind energy, as an important member of the clean energy source, has become a key focus of development and utilization due to its large reserves and renewable nature. However, wind energy is an intermittent resource with poor stability and continuity, and its prediction is difficult, making it hard to meet the grid connection requirements. Improving the accuracy of wind power prediction is crucial for ensuring the safe and stable operation of the power system.

[0003] As onshore wind resources become increasingly saturated, wind farms are gradually expanding into offshore areas. Compared to onshore fixed wind turbines, offshore floating wind turbines operate in a more complex environment. The unique motion of floating platforms makes flow field distribution more difficult to predict. Therefore, there is an urgent need for a method to accurately predict wind power in floating offshore wind farm scenarios. Summary of the Invention

[0004] In view of this, the present invention provides a method, apparatus and equipment for predicting wind power in floating offshore wind farms, which can solve the technical problem of difficulty in accurately predicting wind power in floating offshore wind farm scenarios.

[0005] According to a first aspect of the present invention, a method for predicting wind power in a floating offshore wind farm is provided, the method comprising:

[0006] A three-dimensional time wake model of the floating wind turbine is constructed based on the inflow wind speed, the turbulence intensity of the inflow environment, and the operating conditions to obtain the wake wind speed.

[0007] Obtain the wind power at the previous moment, and determine the environmental state based on the inflow wind speed, the inflow environment turbulence intensity, the operating condition, the wind power at the previous moment, and the wake wind speed;

[0008] Based on the trained policy network, wind power is predicted for the environmental state to obtain the target wind power at the current moment. The policy network is constructed based on a trained convolutional neural network and a trained long short-term memory network.

[0009] Preferably, the step of constructing a three-dimensional time wake model of the floating wind turbine based on the inflow wind speed, the turbulence intensity of the inflow environment, and the operating conditions includes:

[0010] A three-dimensional double cosine wake model is constructed based on the inflow wind speed and the turbulence intensity of the inflow environment.

[0011] Based on the operating conditions and the aforementioned three-dimensional double cosine wake model, a three-dimensional wake model of a floating wind turbine under the action of wind and waves is constructed.

[0012] Calculate the wake delay time and add the wake delay time to the three-dimensional wake model to obtain the three-dimensional time wake model of the floating wind turbine.

[0013] Preferably, the step of constructing a three-dimensional double cosine wake model based on the inflow wind velocity and the turbulence intensity of the inflow environment includes:

[0014] A Jensen wake model was constructed based on the turbulence intensity of the inflow environment.

[0015] Construct a bicosine model for the wake wind speed distribution of a floating wind turbine when it has a bicosine shape;

[0016] A three-dimensional double cosine wake model is constructed based on the Jensen wake model, the double cosine model, and the inflow wind speed.

[0017] Preferably, the step of constructing a three-dimensional wake model of the floating wind turbine under wind and wave action based on the operating conditions and the three-dimensional double cosine wake model includes:

[0018] The average additional wind speed is constructed based on the operating conditions to determine the planar longitudinal motion of the wind turbine;

[0019] The additional average wind speed is added to the three-dimensional double cosine wake model to obtain the three-dimensional wake model of the floating wind turbine under the action of wind and waves.

[0020] Preferably, the wind power prediction based on the trained policy network to obtain the target wind power at the current moment includes:

[0021] The environmental state is input into a convolutional neural network to obtain output feature values;

[0022] Based on the gating mechanism of the Long Short-Term Memory network, the cell state at the previous time step is updated according to the output feature value to obtain the cell state at the current time step, and the target wind power at the current time step is output according to the cell state at the current time step.

[0023] Preferably, before predicting wind power based on the trained policy network for the environmental state, the method further includes:

[0024] Construct a policy network to be trained based on the convolutional neural network and the long short-term memory network to be trained.

[0025] Determine the wind power and historical environmental conditions at the first historical moment, and train the convolutional neural network to be trained based on the historical environmental conditions to obtain historical output feature values. The historical environmental conditions include the wind power at the second historical moment, which is the moment before the first moment.

[0026] Based on the gating mechanism of the long short-term memory network to be trained, the cell state at the second historical moment is updated according to the historical output feature value to obtain the cell state at the first historical moment, and the predicted wind power at the first historical moment is output according to the cell state at the first historical moment.

[0027] The loss function is calculated based on the historical first-moment wind power and the historical first-moment predicted wind power. The training continues until the loss function is less than a preset threshold, resulting in a trained policy network. The policy network includes a trained convolutional neural network and a trained long short-term memory network.

[0028] Preferably, the method further includes:

[0029] The inflow wind speed, the inflow environment turbulence intensity, and the updated information in the operating condition are obtained. Based on the updated information and the three-dimensional time wake model, the updated wake wind speed is obtained.

[0030] Obtain the wind power at the previous moment, and determine the updated environmental state based on the updated information, the wind power at the previous moment, and the updated wake wind speed;

[0031] Based on the trained policy network, wind power is predicted for the updated environmental state to obtain the updated target wind power at the current moment.

[0032] According to a second aspect of the present invention, a wind power prediction device for a floating offshore wind farm is provided, the device comprising:

[0033] The module is used to construct a three-dimensional time wake model of the floating wind turbine based on the inflow wind speed, the turbulence intensity of the inflow environment, and the operating conditions, so as to obtain the wake wind speed.

[0034] The determination module is used to obtain the wind power at the previous moment and determine the environmental state based on the inflow wind speed, the inflow environment turbulence intensity, the operating condition, the wind power at the previous moment, and the wake wind speed.

[0035] The prediction module is used to predict wind power based on the environmental state based on the trained policy network, and obtain the target wind power at the current moment. The policy network is constructed based on the trained convolutional neural network and the trained long short-term memory network.

[0036] According to a third aspect of this application, a storage medium is provided that stores a computer program thereon, which, when executed by a processor, implements the above-described method for predicting wind power in a floating offshore wind farm.

[0037] According to a fourth aspect of this application, a computer device is provided, including a storage medium, a processor, and a computer program stored on the storage medium and executable on the processor, wherein the processor executes the program to implement the above-described method for predicting wind power in a floating offshore wind farm.

[0038] By means of the above technical solution, the present invention provides a method, apparatus and equipment for predicting wind power in a floating offshore wind farm. First, a three-dimensional time wake model of the floating wind turbine is constructed based on the inflow wind speed, the turbulence intensity of the inflow environment and the operating conditions to obtain the wake wind speed. Then, the wind power at the previous moment is obtained, and the environmental state is determined based on the inflow wind speed, the turbulence intensity of the inflow environment, the operating conditions, the wind power at the previous moment and the wake wind speed. Finally, the wind power is predicted based on the environmental state using a trained policy network to obtain the target wind power at the current moment. The policy network is constructed based on a trained convolutional neural network and a trained long short-term memory network. By utilizing the technical solution of this invention, since the operating conditions include information on the oscillating motion process under the floating offshore wind farm scenario, the determined wake wind speed is more consistent with the floating offshore wind farm scenario. This wake wind speed is used as a feature of the environmental state and input into the policy network. The convolutional neural network extracts the spatial features of the environmental state, and the long short-term memory network extracts the temporal series features of the environmental state. Thus, the target wind power can be accurately predicted from the feature information of the floating wind turbine wake in both the spatiotemporal dimensions. Attached Figure Description

[0039] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this invention, illustrate exemplary embodiments of the invention and are used to explain the invention, but do not constitute an undue limitation of this application. In the drawings:

[0040] Figure 1 A flowchart illustrating a wind power prediction method for a floating offshore wind farm provided by an embodiment of the present invention is shown.

[0041] Figure 2 A flowchart illustrating another method for predicting wind power in a floating offshore wind farm, provided by an embodiment of the present invention, is shown.

[0042] Figure 3 A schematic diagram of the structure of a wind power prediction device for a floating offshore wind farm provided in an embodiment of the present invention is shown.

[0043] Figure 4 A schematic diagram of the structure of another floating offshore wind farm wind power prediction device provided by an embodiment of the present invention is shown. Detailed Implementation

[0044] The present invention will be described in detail below with reference to the accompanying drawings and embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in the embodiments of the present invention can be combined with each other.

[0045] This embodiment provides a method for predicting wind power in floating offshore wind farms, such as... Figure 1 As shown, the method includes:

[0046] 101. Based on the inflow wind speed, the turbulence intensity of the inflow environment, and the operating conditions, a three-dimensional time wake model of the floating wind turbine is constructed to obtain the wake wind speed.

[0047] 102. Obtain the wind power at the previous moment, and determine the environmental state based on the inflow wind speed, the inflow environment turbulence intensity, the operating condition, the wind power at the previous moment, and the wake wind speed.

[0048] 103. Based on the trained policy network, wind power is predicted for the environmental state to obtain the target wind power at the current moment, wherein the policy network is constructed based on the trained convolutional neural network and the trained long short-term memory network.

[0049] For steps 101-103 of the embodiment, in order to accurately predict wind power in a floating offshore wind farm scenario, multiple features need to be extracted to determine the environmental state. This environmental state is then input into the trained policy network, which predicts wind power based on this state. Therefore, step 101 of the embodiment determines the wake wind speed based on real-time measured inflow wind speed, inflow environmental turbulence intensity, and operating conditions. Since the operating conditions include information about the oscillating motion process in the floating offshore wind farm scenario, the determined wake wind speed is more consistent with the floating offshore wind farm scenario. The wake wind speed is used as one feature of the environmental state, and step 102 of the embodiment determines other features of the environmental state. In the deep reinforcement learning framework (i.e., the policy network), the convolutional neural network (CNN) is responsible for extracting spatial features from the environmental state through convolutional operations. After the CNN extracts spatial features, the long short-term memory (LSTM) network is used to extract temporal series features from the environmental state input. Based on the spatial features extracted by the CNN and the temporal series information processed by the LSTM network, the policy network outputs the target wind power at the current moment.

[0050] This invention provides a method, apparatus, and equipment for predicting wind power in a floating offshore wind farm. First, a three-dimensional time wake model of the floating wind turbine is constructed based on the inflow wind speed, the turbulence intensity of the inflow environment, and the operating conditions to obtain the wake wind speed. Then, the wind power at the previous moment is obtained, and the environmental state is determined based on the inflow wind speed, the turbulence intensity of the inflow environment, the operating conditions, the wind power at the previous moment, and the wake wind speed. Finally, wind power is predicted based on the environmental state using a trained policy network to obtain the target wind power at the current moment. The policy network is constructed based on a trained convolutional neural network and a trained long short-term memory network. By utilizing the technical solution of this invention, since the operating conditions include information on the oscillating motion process under the floating offshore wind farm scenario, the determined wake wind speed is more consistent with the floating offshore wind farm scenario. This wake wind speed is used as a feature of the environmental state and input into the policy network. The convolutional neural network extracts the spatial features of the environmental state, and the long short-term memory network extracts the temporal series features of the environmental state. Thus, the target wind power can be accurately predicted from the feature information of the floating wind turbine wake in both the spatiotemporal dimensions.

[0051] Furthermore, as a refinement and extension of the specific implementation methods of the above embodiments, and to fully illustrate the specific implementation process in this embodiment, another method for predicting wind power in floating offshore wind farms is provided, such as... Figure 2 As shown, the method includes:

[0052] 201. Construct a three-dimensional double cosine wake model based on the inflow wind speed and the turbulence intensity of the inflow environment.

[0053] Among them, drones are used to measure the inflow wind speed and the intensity of inflow turbulence in the wind farm in real time.

[0054] The construction of a three-dimensional double cosine wake model based on the inflow wind speed and the turbulence intensity of the inflow environment includes: constructing a Jensen wake model based on the turbulence intensity of the inflow environment; constructing a double cosine model when the wake wind speed distribution of the floating wind turbine is a double cosine shape; and constructing a three-dimensional double cosine wake model based on the Jensen wake model, the double cosine model, and the inflow wind speed.

[0055] In this embodiment, taking the floating wind turbine as a reference when it is free from the influence of waves and swells and at the beginning of a wave cycle, the hub center is taken as the origin of the coordinate system. The x-direction represents the direction of the inflow wind speed (also the x-direction of the inflow wind speed), the y-direction represents the horizontal direction that is radially perpendicular to the x-axis (also the y-direction of the inflow wind speed), and the z-direction represents the vertical direction that is perpendicular to the xoy plane (also the z-direction of the inflow wind speed).

[0056] For this embodiment, the Jensen wake model is constructed based on the turbulence intensity of the inflow environment:

[0057]

[0058] Where a is the axial induction factor of the floating wind turbine, and r w It is the wake radius of a floating wind turbine, r d Where S is the rotor radius, S is the normalized downstream distance of the floating wind turbine, and C is the rotor radius. t I is the thrust coefficient of the floating wind turbine, and I0 is the turbulence intensity of the inflow environment at the hub height.

[0059] In this embodiment, a bicosine model is constructed when the wake velocity distribution of the floating wind turbine has a bicosine shape:

[0060]

[0061] Where u0 is the inflow wind speed at the height of the floating wind turbine hub, and z represents the vertical direction perpendicular to the xoy plane. h Here, α is the hub height of the floating wind turbine, α is the wind shear index, r is the radial distance from the center of the rotor, and θ is the distance between the wake centerline and the point of maximum velocity loss in the wake, θ = 0.6r. d r d It is the radius of the wind turbine, r′ w It is the distance r′ from the point where the wake velocity loss is maximum to the wake boundary. w =r w -θ, r w It is the wake radius of the floating wind turbine, and k is the period of the bicosine function, k = π / 2r′. w .

[0062] In this embodiment, a three-dimensional bicosine wake model is constructed based on the Jensen wake model, the bicosine model, and the inflow wind speed. This includes: the mass flux of the bicosine model is equal to the mass flux of the Jensen wake model. Based on the law of conservation of mass, the following first relationship can be derived:

[0063]

[0064] By combining the Jensen wake model, the bicosine model, and the first relation, a three-dimensional bicosine wake model is obtained:

[0065]

[0066] Where u(x,y,z) is the initial wake wind speed.

[0067] 202. Based on the working conditions and the aforementioned three-dimensional double cosine wake model, construct a three-dimensional wake model of the floating wind turbine under the action of wind and waves.

[0068] The step of constructing a three-dimensional wake model of a floating wind turbine under wind and wave action based on the operating conditions and the three-dimensional double cosine wake model includes: constructing an additional average wind speed for the planar swaying motion of the wind turbine based on the operating conditions; and adding the additional average wind speed to the three-dimensional double cosine wake model to obtain a three-dimensional wake model of a floating wind turbine under wind and wave action.

[0069] For this embodiment, the average additional wind speed for the planar oscillation motion of the wind turbine is constructed based on the operating conditions:

[0070] It should be noted that the swaying motion is the planar swaying motion of the wind turbine caused by wave motion in the floating offshore wind farm scenario. By adding the average wind speed to the three-dimensional double cosine wake model, the resulting three-dimensional wake model of the floating wind turbine under the action of wind and waves takes into account the unique motion characteristics of the floating platform. Based on this, the wake wind speed obtained by step 203 of the embodiment is more applicable to the floating offshore wind farm scenario. Therefore, the wake wind speed is used as part of the environmental state in step 204 of the embodiment. After the policy network is trained, the target wind power obtained at the current moment is more accurate.

[0071] The operating conditions include: the amplitude of the oscillating motion, the frequency of the oscillating motion, the phase angle in the initial stage, the average value of the oscillating motion displacement, and the period of the oscillating motion.

[0072]

[0073] Among them, A s It is the amplitude of the oscillating motion, f s It is the frequency of the oscillating motion. It is the phase angle in the initial stage. It is the average displacement of the oscillating motion, T s T1 and T2 represent the start and end times of the oscillation motion within a selected time period, respectively, and n is the number of periods. This is the average wind speed.

[0074] In this embodiment, the additional average wind speed is added to the three-dimensional double cosine wake model to obtain the three-dimensional wake model of the floating wind turbine under the action of wind and waves:

[0075]

[0076] 203. Calculate the wake delay time, add the wake delay time to the three-dimensional wake model to obtain the three-dimensional time wake model of the floating wind turbine, and obtain the wake wind speed.

[0077] For this embodiment, the wake delay time is calculated as follows:

[0078]

[0079] Where, τ ij It is the wake delay time from the i-th unit to the j-th unit.

[0080] In this embodiment, the wake delay time is added to the three-dimensional wake model to obtain a three-dimensional time wake model of the floating wind turbine:

[0081]

[0082] Where T is the propagation time of wind from the upstream wind turbine to the downstream wind turbine at time t, and u(x,y,z,T) is the wake wind speed, which, compared to u(x,y,z) (where u(x,y,z) is the initial wake wind speed), considers τ. ij , which is the wake delay time from the i-th unit to the j-th unit, means that the three-dimensional time wake model is obtained by correcting the three-dimensional wake model using the wake delay time.

[0083] It should be noted that the obtained wake wind speed u(x,y,z,T) represents the flow field characteristics of wind farms with different time and spatial dimensions.

[0084] 204. Obtain the wind power at the previous moment, and determine the environmental state based on the inflow wind speed, the inflow environment turbulence intensity, the operating condition, the wind power at the previous moment, and the wake wind speed.

[0085] In this embodiment, as one implementation method, a multi-source dataset is constructed, which includes real-time measured data (the real-time measured data are the inflow wind speed u0 and the inflow environmental turbulence intensity I0) and historical data (the historical data is the wind power P at the previous moment). t-1 ) and working condition φ t (The amplitude A of the oscillating motion is...) s The frequency f of the oscillating motion s Phase angle in the initial stage Average value of oscillating motion displacement and the period T of the oscillation motion s The wake velocity calculated using the multi-source dataset and steps 201-203 of the implementation example is used to determine the environmental state.

[0086] S t =[u0,I0,φ t ,P t-1 ,u(x,y,z,T)]

[0087] It should be noted that before constructing a multi-source dataset, the data in the multi-source dataset needs to undergo preprocessing such as data cleaning, format conversion, and time alignment.

[0088] 205. Based on the trained policy network, wind power is predicted for the environmental state to obtain the target wind power at the current moment, wherein the policy network is constructed based on the trained convolutional neural network and the trained long short-term memory network.

[0089] Before predicting wind power based on the trained policy network for the environmental state, the method further includes: constructing a policy network to be trained using a convolutional neural network and a long short-term memory network to be trained; determining the historical first-moment wind power and historical environmental state, training the convolutional neural network to be trained based on the historical environmental state to obtain historical output feature values, wherein the historical environmental state includes the historical second-moment wind power, and the second moment is the moment before the first moment; updating the historical second-moment cell state based on the gating mechanism of the long short-term memory network to be trained, obtaining the historical first-moment cell state, and outputting the historical first-moment predicted wind power based on the historical first-moment cell state; calculating a loss function based on the historical first-moment wind power and the historical first-moment predicted wind power, until the loss function is less than a preset threshold, to obtain a trained policy network, wherein the policy network includes a trained convolutional neural network and a trained long short-term memory network.

[0090] The historical environmental conditions include historical inflow wind speed, historical inflow turbulence intensity, historical second-time wind power, and historical operating conditions. The historical second-time wind power is the actual wind power at the second time, and the historical first-time wind power is the actual wind power at the first time. Based on the historical environmental conditions including the historical second-time wind power, the training convolutional neural network and the training long short-term memory network are trained to obtain the historical first-time predicted wind power, which is the predicted wind power at the first time. By calculating the loss function between the historical first-time wind power and the historical first-time predicted wind power, if the loss function is greater than or equal to a preset threshold, then training continues until the loss function is less than the preset threshold. At this point, the trained policy network is obtained.

[0091] The specific training process is as follows: (1) Train the convolutional neural network to be trained according to the historical environment state to obtain historical output feature values. Specifically:

[0092]

[0093] In the formula, Z i,j,kIt is the historical output feature value of the k-th convolutional kernel at position (i,j); S is the historical environment state; W is the convolutional kernel, with a size of M×N; b k It is the bias term of the k-th convolution kernel.

[0094] (2) Based on the gating mechanism of the Long Short-Term Memory network to be trained, the cell state at the second historical moment is updated according to the historical output feature value to obtain the cell state at the first historical moment, and the predicted wind power at the first historical moment is output according to the cell state at the first historical moment. Specifically:

[0095] The cell state update formula is as follows:

[0096] C t =f t ·C t-1 +i t ·C tx

[0097] In the formula, C t It represents the cellular state at the very first moment in history, C t It uses a gating mechanism to dynamically adjust C based on historical output characteristic values. t-1 The result of the update is C. t-1 It is the cellular state at the second moment in history; f t It is the Gate of Oblivion; t It is an input gate; C tx It is a new candidate cell state.

[0098] The historical first-moment cell state is used to output the historical first-moment predicted wind power. Specifically, the historical first-moment cell state captures and stores long-term time-dependent information to assist the hidden state output of the historical first-moment predicted wind power in the long short-term memory network to be trained.

[0099] The method of predicting wind power based on the trained policy network to obtain the target wind power at the current moment includes: inputting the environmental state into a convolutional neural network to obtain output feature values; updating the cell state at the previous moment based on the output feature values ​​using the gating mechanism of a long short-term memory network to obtain the cell state at the current moment; and outputting the target wind power at the current moment based on the cell state at the current moment.

[0100]

[0101] In the formula, Z i,j,k It is the output feature value of the k-th convolutional kernel at position (i,j); S is the environment state; W is the convolutional kernel, with a size of M×N; b k It is the bias term of the k-th convolution kernel.

[0102] The cell state update formula is as follows:

[0103] C t =f t ·C t-1 +i t ·C tx

[0104] In the formula, C t It represents the current state of the cell, C. t It uses a gating mechanism to dynamically adjust C based on the output characteristic values. t-1 The result of the update is C. t-1 It represents the cell state at the previous moment; f t It is the Gate of Oblivion; t It is an input gate; C tx It is a new candidate cell state.

[0105] Preferably, after using the policy network to predict wind power based on the environmental state and obtain the target wind power at the current moment, the environmental state can also be input into the value network. The value network, by optimizing the decision path, can provide the policy network with a reward V(s) for the current state. t This value, or state value, helps the policy network make better choices during the decision-making process. In the policy optimization process, an experience replay technique is used, storing historical data in an experience pool and training with randomly selected samples to avoid the model over-reliance on the latest data, thus enhancing the stability of learning. Furthermore, an "ε-greedy policy" is used to help the model achieve a balance between exploring new policies and utilizing the currently known best policy, as shown in the following formula:

[0106]

[0107] In the formula, a t It represents the model's action at time t; Q(s) t a) is the current state s t The action value function for action a; ε is the model's exploration rate; argmaxQ(s) t a) represents choosing the action that brings the greatest expected return in the current state; random action represents choosing a random action with probability ε.

[0108] The policy network optimizes decision-making policies by maximizing long-term rewards, while the value network continuously optimizes the estimation of state rewards by minimizing the error between the current state value and the target value through temporal difference learning. Through these methods, both the policy network and the value network are continuously optimized, achieving higher prediction accuracy and stronger robustness.

[0109] 206. Obtain the inflow wind speed, the inflow environment turbulence intensity, and the updated information in the operating condition. Based on the updated information and the three-dimensional time wake model, obtain the updated wake wind speed.

[0110] 207. Obtain the wind power at the previous moment, and determine the updated environmental state based on the updated information, the wind power at the previous moment, and the updated wake wind speed.

[0111] 208. Based on the trained policy network, predict the wind power of the updated environment state to obtain the updated target wind power at the current moment.

[0112] For steps 201-205 of the embodiment, the inflow wind speed, the inflow environment turbulence intensity, and the operating conditions are all measured in real time, and the determined wake wind speed is also in real time. For steps 206-208 of the embodiment, if the inflow wind speed, the inflow environment turbulence intensity, and the operating conditions do not change, then the determined wake wind speed can be used directly. If at least one of the inflow wind speed, the inflow environment turbulence intensity, and the operating conditions changes, then the wake wind speed needs to be re-determined to obtain an updated wake wind speed. Correspondingly, the environmental conditions need to be re-determined to obtain an updated environmental condition. When predicting wind power, it is necessary to predict the updated environmental condition so that the updated target wind power obtained is consistent with the actual situation and has high accuracy.

[0113] This invention provides a method, apparatus, and equipment for predicting wind power in a floating offshore wind farm. First, a three-dimensional time wake model of the floating wind turbine is constructed based on the inflow wind speed, the turbulence intensity of the inflow environment, and the operating conditions to obtain the wake wind speed. Then, the wind power at the previous moment is obtained, and the environmental state is determined based on the inflow wind speed, the turbulence intensity of the inflow environment, the operating conditions, the wind power at the previous moment, and the wake wind speed. Finally, wind power is predicted based on the environmental state using a trained policy network to obtain the target wind power at the current moment. The policy network is constructed based on a trained convolutional neural network and a trained long short-term memory network. By utilizing the technical solution of this invention, since the operating conditions include information on the oscillating motion process under the floating offshore wind farm scenario, the determined wake wind speed is more consistent with the floating offshore wind farm scenario. This wake wind speed is used as a feature of the environmental state and input into the policy network. The convolutional neural network extracts the spatial features of the environmental state, and the long short-term memory network extracts the temporal series features of the environmental state. Thus, the target wind power can be accurately predicted from the feature information of the floating wind turbine wake in both the spatiotemporal dimensions.

[0114] Furthermore, as Figure 1 and Figure 2The specific implementation of the method shown in this invention provides a wind power prediction device for a floating offshore wind farm, such as... Figure 3 As shown, the device includes: a construction module 31, a determination module 32, and a prediction module 33;

[0115] Module 31 is used to construct a three-dimensional time wake model of the floating wind turbine based on the inflow wind speed, the turbulence intensity of the inflow environment and the operating conditions, so as to obtain the wake wind speed.

[0116] The determination module 32 is used to obtain the wind power at the previous moment and determine the environmental state based on the inflow wind speed, the inflow environment turbulence intensity, the operating condition, the wind power at the previous moment, and the wake wind speed.

[0117] The prediction module 33 is used to predict the wind power of the environment based on the trained policy network to obtain the target wind power at the current moment. The policy network is constructed based on the trained convolutional neural network and the trained long short-term memory network.

[0118] Accordingly, in order to construct a three-dimensional time wake model of the floating wind turbine based on the inflow wind speed, the turbulence intensity of the inflow environment, and the operating conditions, module 31 is constructed. Specifically, it can be used to construct a three-dimensional double cosine wake model based on the inflow wind speed and the turbulence intensity of the inflow environment; construct a three-dimensional wake model of the floating wind turbine under the action of wind and waves based on the operating conditions and the three-dimensional double cosine wake model; calculate the wake delay time, add the wake delay time to the three-dimensional wake model, and obtain the three-dimensional time wake model of the floating wind turbine.

[0119] Accordingly, in order to construct a three-dimensional double cosine wake model based on the inflow wind speed and the turbulence intensity of the inflow environment, module 31 is constructed, which can be used to construct a Jensen wake model based on the turbulence intensity of the inflow environment; construct a double cosine model when the wake wind speed distribution of the floating wind turbine is in a double cosine shape; and construct a three-dimensional double cosine wake model based on the Jensen wake model, the double cosine model and the inflow wind speed.

[0120] Accordingly, in order to construct a three-dimensional wake model of a floating wind turbine under the action of wind and waves based on the working conditions and the three-dimensional double cosine wake model, the construction module 31 can be used to construct the additional average wind speed of the wind turbine planar sway motion based on the working conditions; the additional average wind speed is added to the three-dimensional double cosine wake model to obtain the three-dimensional wake model of the floating wind turbine under the action of wind and waves.

[0121] Accordingly, in order to predict wind power based on the trained policy network and obtain the target wind power at the current moment, the prediction module 33 can specifically be used to input the environmental state into the convolutional neural network to obtain the output feature value; based on the gating mechanism of the long short-term memory network, update the cell state at the previous moment according to the output feature value to obtain the cell state at the current moment, and output the target wind power at the current moment according to the cell state at the current moment.

[0122] In specific application scenarios, such as Figure 4 As shown, a wind power prediction device for a floating offshore wind farm includes: a training module 34, specifically used to construct a policy network to be trained based on a convolutional neural network and a long short-term memory network to be trained; determine the historical wind power at a first historical moment and the historical environmental state; train the convolutional neural network to be trained based on the historical environmental state to obtain historical output feature values, wherein the historical environmental state includes the wind power at a second historical moment, the second moment being the moment before the first moment; based on the gating mechanism of the long short-term memory network to be trained, update the cell state at the second historical moment according to the historical output feature values ​​to obtain the cell state at the first historical moment; output the predicted wind power at the first historical moment according to the cell state at the first historical moment; calculate a loss function based on the wind power at the first historical moment and the predicted wind power at the first historical moment, until the loss function is less than a preset threshold, to obtain a trained policy network, wherein the policy network includes a trained convolutional neural network and a trained long short-term memory network.

[0123] In specific application scenarios, such as Figure 4 As shown, a wind power prediction device for a floating offshore wind farm is disclosed. The device further includes an update module 35, which is specifically used to acquire the inflow wind speed, the inflow environment turbulence intensity, and update information in the operating condition; obtain the updated wake wind speed based on the update information and the three-dimensional time wake model; acquire the wind power at the previous moment; determine the updated environmental state based on the update information, the wind power at the previous moment, and the updated wake wind speed; and predict the wind power based on the updated environmental state using a trained policy network to obtain the updated target wind power at the current moment.

[0124] It should be noted that other corresponding descriptions of the functional units involved in the wind power prediction device for a floating offshore wind farm provided in this embodiment can be found in [reference needed]. Figures 1 to 2 The corresponding description will not be repeated here.

[0125] Based on the above, Figures 1 to 2Accordingly, this embodiment also provides a storage medium, which may be volatile or non-volatile, storing a computer program that, when executed by a processor, implements the above-described method. Figures 1 to 2 The method for predicting wind power in a floating offshore wind farm is shown.

[0126] Based on this understanding, the technical solution of the present invention can be embodied in the form of a software product, which can be stored in a storage medium (such as a CD-ROM, USB flash drive, portable hard drive, etc.) and includes several instructions to cause a computer device (such as a personal computer, server, or network device, etc.) to execute the methods of various implementation scenarios of the present invention.

[0127] Based on the above, Figures 1 to 2 The method shown and Figure 3 , Figure 4 To achieve the above objectives, this embodiment also provides a computer device, which includes a storage medium and a processor; the storage medium is used to store a computer program; the processor is used to execute the computer program to implement the above-described virtual device embodiment. Figures 1 to 2 The method for predicting wind power in a floating offshore wind farm is shown.

[0128] Optionally, the computer device may also include a user interface, a network interface, a camera, radio frequency (RF) circuitry, sensors, audio circuitry, a Wi-Fi module, etc. The user interface may include a display screen, input units such as a keyboard, etc., and optional user interfaces may also include USB interfaces, card reader interfaces, etc. The network interface may optionally include standard wired interfaces, wireless interfaces (such as Wi-Fi interfaces), etc.

[0129] Those skilled in the art will understand that the computer device structure provided in this embodiment does not constitute a limitation on the physical device, and may include more or fewer components, or combine certain components, or have different component arrangements.

[0130] The storage medium may also include an operating network communication module. An operating system is a program that manages the hardware and software resources of the aforementioned computer device, supporting the operation of information processing programs and other software and / or programs. The network communication module is used for communication between the various components within the storage medium, as well as for communication with other hardware and software within the information processing device.

[0131] Through the above description of the embodiments, those skilled in the art can clearly understand that the present invention can be implemented by means of software plus necessary general-purpose hardware platform, or it can be implemented by hardware.

[0132] This invention provides a method, apparatus, and equipment for predicting wind power in a floating offshore wind farm. First, a three-dimensional time wake model of the floating wind turbine is constructed based on the inflow wind speed, the turbulence intensity of the inflow environment, and the operating conditions to obtain the wake wind speed. Then, the wind power at the previous moment is obtained, and the environmental state is determined based on the inflow wind speed, the turbulence intensity of the inflow environment, the operating conditions, the wind power at the previous moment, and the wake wind speed. Finally, wind power is predicted based on the environmental state using a trained policy network to obtain the target wind power at the current moment. The policy network is constructed based on a trained convolutional neural network and a trained long short-term memory network. By utilizing the technical solution of this invention, since the operating conditions include information on the oscillating motion process under the floating offshore wind farm scenario, the determined wake wind speed is more consistent with the floating offshore wind farm scenario. This wake wind speed is used as a feature of the environmental state and input into the policy network. The convolutional neural network extracts the spatial features of the environmental state, and the long short-term memory network extracts the temporal series features of the environmental state. Thus, the target wind power can be accurately predicted from the feature information of the floating wind turbine wake in both the spatiotemporal dimensions.

[0133] Those skilled in the art will understand that the accompanying drawings are merely schematic diagrams of a preferred embodiment, and the modules or processes shown in the drawings are not necessarily essential for implementing the present invention. Those skilled in the art will understand that the modules in the apparatus of the embodiment can be distributed within the apparatus of the embodiment as described, or they can be located in one or more apparatuses different from this embodiment, with corresponding changes. The modules of the above-described embodiment can be combined into one module, or further divided into multiple sub-modules.

[0134] The serial numbers used above are for descriptive purposes only and do not represent the superiority or inferiority of the implementation scenarios. The above disclosures are merely a few specific implementation scenarios of the present invention; however, the present invention is not limited thereto, and any variations conceived by those skilled in the art should fall within the protection scope of the present invention.

Claims

1. A method for predicting wind power in a floating offshore wind farm, characterized in that, The method includes: A three-dimensional time wake model of the floating wind turbine is constructed based on the inflow wind speed, the turbulence intensity of the inflow environment, and the operating conditions to obtain the wake wind speed. The operating conditions include: the amplitude of the sway motion, the frequency of the sway motion, the phase angle of the initial stage, the average value of the sway motion displacement, and the period of the sway motion. Obtain the wind power at the previous moment, and determine the environmental state based on the inflow wind speed, the inflow environment turbulence intensity, the operating condition, the wind power at the previous moment, and the wake wind speed; Based on the trained policy network, wind power is predicted for the environmental state to obtain the target wind power at the current moment. The policy network is constructed based on a trained convolutional neural network and a trained long short-term memory network. The construction of a three-dimensional time wake model for the floating wind turbine based on the inflow wind speed, the turbulence intensity of the inflow environment, and the operating conditions includes: A three-dimensional double cosine wake model is constructed based on the inflow wind speed and the turbulence intensity of the inflow environment. Based on the operating conditions and the aforementioned three-dimensional double cosine wake model, a three-dimensional wake model of a floating wind turbine under the action of wind and waves is constructed. Calculate the wake delay time and add the wake delay time to the three-dimensional wake model to obtain the three-dimensional time wake model of the floating wind turbine.

2. The method according to claim 1, characterized in that, The construction of a three-dimensional double cosine wake model based on the inflow wind speed and the turbulence intensity of the inflow environment includes: A Jensen wake model was constructed based on the turbulence intensity of the inflow environment. Construct a bicosine model for the wake wind speed distribution of a floating wind turbine when it has a bicosine shape; A three-dimensional double cosine wake model is constructed based on the Jensen wake model, the double cosine model, and the inflow wind speed.

3. The method according to claim 1, characterized in that, The construction of a three-dimensional wake model of a floating wind turbine under wind and wave action based on the operating conditions and the three-dimensional double cosine wake model includes: The average additional wind speed is constructed based on the operating conditions to determine the planar longitudinal motion of the wind turbine; The additional average wind speed is added to the three-dimensional double cosine wake model to obtain the three-dimensional wake model of the floating wind turbine under the action of wind and waves.

4. The method according to claim 1, characterized in that, The wind power prediction based on the trained policy network for the environmental state, to obtain the target wind power at the current moment, includes: The environmental state is input into a convolutional neural network to obtain output feature values; Based on the gating mechanism of the Long Short-Term Memory network, the cell state at the previous time step is updated according to the output feature value to obtain the cell state at the current time step, and the target wind power at the current time step is output according to the cell state at the current time step.

5. The method according to claim 1, characterized in that, Before predicting wind power based on the trained policy network for the environmental state, the method further includes: Construct a policy network to be trained based on the convolutional neural network and the long short-term memory network to be trained. Determine the wind power and historical environmental conditions at the first historical moment, and train the convolutional neural network to be trained based on the historical environmental conditions to obtain historical output feature values. The historical environmental conditions include the wind power at the second historical moment, which is the moment before the first moment. Based on the gating mechanism of the long short-term memory network to be trained, the cell state at the second historical moment is updated according to the historical output feature value to obtain the cell state at the first historical moment, and the predicted wind power at the first historical moment is output according to the cell state at the first historical moment. The loss function is calculated based on the historical first-moment wind power and the historical first-moment predicted wind power. The training continues until the loss function is less than a preset threshold, resulting in a trained policy network. The policy network includes a trained convolutional neural network and a trained long short-term memory network.

6. The method according to claim 1, characterized in that, The method further includes: The inflow wind speed, the inflow environment turbulence intensity, and the updated information in the operating condition are obtained. Based on the updated information and the three-dimensional time wake model, the updated wake wind speed is obtained. Obtain the wind power at the previous moment, and determine the updated environmental state based on the updated information, the wind power at the previous moment, and the updated wake wind speed; Based on the trained policy network, wind power is predicted for the updated environmental state to obtain the updated target wind power at the current moment.

7. A wind power prediction device for a floating offshore wind farm, characterized in that, The device includes: The construction module is used to construct a three-dimensional time wake model of the floating wind turbine based on the inflow wind speed, the turbulence intensity of the inflow environment, and the operating conditions to obtain the wake wind speed. The operating conditions include: the amplitude of the sway motion, the frequency of the sway motion, the phase angle of the initial stage, the average value of the sway motion displacement, and the period of the sway motion. The determination module is used to obtain the wind power at the previous moment and determine the environmental state based on the inflow wind speed, the inflow environment turbulence intensity, the operating condition, the wind power at the previous moment, and the wake wind speed. The prediction module is used to predict the wind power of the environment based on the trained policy network to obtain the target wind power at the current moment. The policy network is constructed based on the trained convolutional neural network and the trained long short-term memory network. The construction module is used to construct a three-dimensional double cosine wake model based on the inflow wind speed and the turbulence intensity of the inflow environment; construct a three-dimensional wake model of the floating wind turbine under the action of wind and waves based on the operating conditions and the three-dimensional double cosine wake model; calculate the wake delay time, add the wake delay time to the three-dimensional wake model, and obtain a three-dimensional time wake model of the floating wind turbine.

8. A storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the wind power prediction method for the floating offshore wind farm according to any one of claims 1 to 6.

9. A computer device, comprising a storage medium, a processor, and a computer program stored on the storage medium and executable on the processor, characterized in that, When the processor executes the computer program, it implements the wind power prediction method for a floating offshore wind farm as described in any one of claims 1 to 6.

Images