Wind power prediction method based on wake perception condition residual error neural network

Abstract

The application discloses the technical field of wind turbine power prediction and intelligent modeling, and particularly relates to a wind power prediction method based on a wake perception condition residual error neural network. The method fully utilizes upstream and downstream unit monitoring data and wind farm geometric arrangement, constructs a condition residual error neural network structure with wake perception capability, and conditionally corrects wake residual error through a gate signal with clear physical meaning. Under the premise of ensuring that the prediction accuracy of the no-wake condition does not decrease, the prediction accuracy and robustness of the wake condition, especially the severe wake condition, are improved, and the method can be applied to the wind turbine power prediction scene of the minute level and shorter time scales.

Description

Technical Field

[0001] This invention relates to the field of wind turbine power prediction and intelligent modeling technology, and in particular to a wind power prediction method based on wake perception conditional residual neural network. Background Technology

[0002] As the scale of wind power grid connection continues to expand, the volatility and uncertainty of wind farm output power have an increasingly prominent impact on the safe and stable operation of the power system. To improve wind power absorption, reduce reserve costs, and alleviate active power regulation pressure, wind power prediction systems based on turbine monitoring data and meteorological data are widely deployed on both the grid and owner sides to estimate the output power of wind turbines over a future period, used for power generation planning, turbine operation optimization, and dispatch decisions. However, existing technologies have the following shortcomings:

[0003] 1) Most existing wind power prediction models treat the units as independent entities, using only the local wind speed and power characteristics for modeling. They do not make full use of upstream and downstream unit monitoring data and wake information such as unit geometry. The wake effect is mostly handled in the form of external engineering models or additional features. They lack conditional residual structures and gating mechanisms that are closely coupled with upstream and downstream unit monitoring data, unit geometry, and wake intensity indicators. This reduces the accuracy and robustness of downstream unit power prediction under complex wake conditions, making it difficult to accurately and timely characterize the power deviation of downstream units under wake conditions.

[0004] 2) Within large wind farms, the extraction of incoming wind energy by upstream turbines creates a significant wake effect at downstream turbines, causing the downstream turbines to exhibit multimodal and strongly nonlinear power-wind speed relationships under different wind direction sectors and wake intensities. Existing single network structures that directly superimpose wake-related features onto the input layer typically use a uniform loss function for overall optimization of all samples. This lacks differentiated modeling and residual correction mechanisms for wakeless and wake-based operating conditions, easily leading to problems such as systematic underfitting under severe wake-based conditions or a significant decrease in prediction accuracy under wakeless conditions after increasing the weight of wake samples.

[0005] 3) Existing methods generally lack gating signals and conditional residual channels constructed based on physical quantities such as wake intensity and wind direction sector in model structure and training design. It is difficult to form a wake perception and prediction framework that has both physical meaning and is easy to train stably, thus limiting the engineering application of achieving refined wind power prediction under complex wake conditions.

[0006] For example, the Chinese patent with authorization announcement number CN110188939B uses historical data of adjacent units to calculate wake model parameters and obtains the predicted values ​​of power of each unit and wind power at the field level through parameter prediction model. The wake is mainly reflected in the external engineering model parameters and does not explicitly distinguish different wake conditions in the neural network structure.

[0007] For example, the Chinese patent application with publication number CN114841077A improves the accuracy of single-unit power prediction under complex terrain and wind conditions by downscaling numerical weather forecasts in time and space and combining them with data-driven models. It focuses on improving the resolution of meteorological input, but does not adequately consider the wake coupling of upstream and downstream units within the same wind farm and the gating of operating conditions.

[0008] For example, the Chinese patent with authorization announcement number CN112270454B uses a quantile LSTM equal probability prediction model in the short-term load forecasting of power systems under extreme factors, which focuses on characterizing the distribution characteristics of load under extreme conditions, but does not combine the geometric layout of wind farms and wake intensity characteristics to construct a conditional residual structure for downstream units.

[0009] For example, Chinese patent application CN112329979A proposes an ultra-short-term wind power prediction method based on adaptive deep residual networks. The prediction accuracy is improved by optimizing residual units and adaptive learning rates. However, the residual structure is oriented towards the total power of a single wind farm. It does not distinguish between wake and non-wake conditions, nor does it combine the geometric layout of upstream and downstream units and wake intensity indicators for conditional modeling.

[0010] For example, Chinese patent application CN121072871A proposes a multimodal offshore wind power ultra-short-term prediction method. It uses a multi-scale wake perception graph spatiotemporal prediction model to characterize the spatiotemporal relationship of multiple units and the wake effect. However, the wake information is mainly reflected in the graph structure through a dynamic adjacency matrix. It does not adopt the conditional residual form of baseline subnetwork + wake residual subnetwork + gated subnetwork, nor does it explicitly introduce structural constraints that improve wake conditions and prevent degradation of non-wake conditions.

[0011] The above-mentioned solutions either treat the wake effect as an external physical model parameter or adopt residual or probabilistic structures in the field of load forecasting. However, none of them combine the SCADA data, geometric relationships and wake intensity indicators of upstream and downstream units within the same wind farm to construct a gated conditional residual neural network to specifically constrain the power forecasting performance of downstream units under the condition of "wake condition improvement and non-wake condition no degradation". Summary of the Invention

[0012] The purpose of this invention is to address the problems in existing wind turbine power prediction models, such as insufficient wake perception capability, lack of conditional residual correction structure, and difficulty in balancing prediction accuracy under different operating conditions. This invention provides a wind power prediction method based on a wake perception conditional residual neural network. This method can fully utilize upstream and downstream turbine monitoring data and wind farm geometry to construct a conditional residual neural network structure with wake perception capability. It conditionally corrects the wake residual through physically meaningful gating signals, improving prediction accuracy and robustness under wake-free conditions, especially severe wake conditions, without compromising prediction accuracy. This method is applicable to wind turbine power prediction scenarios with timescales of minutes and shorter.

[0013] To achieve the above objectives, the present invention adopts the following technical solution:

[0014] A wind power prediction method based on a wake-sensing conditional residual neural network includes:

[0015] 1) Data Acquisition and Processing

[0016] Full-condition operational data of the floating wind-wave combined power generation system were acquired using sensors and a high-fidelity model. The raw data underwent preprocessing including normalization, denoising, and feature extraction. Training and testing sets were then created to provide a high-quality data foundation for subsequent modeling.

[0017] 2) Data acquisition and alignment of upstream and downstream units

[0018] Acquire monitoring data such as wind speed, wind direction, active power, and pitch angle of the target downstream unit and its main upstream units, as well as geometric parameters such as the distance between units, relative orientation, and rotor diameter; align the upstream and downstream data according to a unified time axis to construct baseline feature vectors, wake feature vectors, and operating condition feature vectors; the baseline feature vector includes basic input features such as historical power, wind speed, and wind direction of the downstream units.

[0019] 3) Wake condition identification and intensity feature extraction

[0020] The wake sector is divided according to the wind direction and the relative position of the unit to identify whether the sample is in the wake sector; the normalized wind speed attenuation index is calculated by combining the upstream and downstream wind speeds after advection alignment, and the samples are divided into no wake, general wake and severe wake, and the wake intensity is smoothly mapped to a continuous score of 0 to 1 as a gating supervision signal.

[0021] 4) Baseline prediction subnetwork construction

[0022] Using the baseline feature vector as input, a multi-layer feedforward neural network is constructed to output the baseline power prediction value at the corresponding time, which is used to characterize the typical power-wind speed relationship under conditions of no wake or weak wake. Multi-head structure can be set in the output layer to realize multi-step prediction.

[0023] 5) Construction of the wake residual subnetwork

[0024] Based on the baseline feature vector input, a wake feature vector composed of upstream unit wind speed, unit spacing, wind direction offset, and lateral overlap is introduced to construct a wake residual subnetwork. A Gaussian wake physical layer is embedded before or in the middle layer of the wake residual subnetwork to approximate the wind speed attenuation or power loss, and the wake residual is output to make additional corrections to the baseline prediction results.

[0025] 6) Fusion of wake gating subnetwork and conditional residual

[0026] A small number of operating condition features, such as sector marking, wake intensity score, and geometric features, are selected as the operating condition feature vector input to construct a lightweight wake gating sub-network. The output is a gating coefficient in the range of 0 to 1, which represents the weight of the wake residual in the correction. The outputs of the three sub-networks are fused according to the conditional residual form of "predicted power = baseline prediction + gating coefficient × wake residual", and the gating coefficient is shared for multi-step prediction scenarios.

[0027] 7) Joint Loss Function Design and Training

[0028] The joint loss function is designed, consisting of the overall prediction error loss, the consistency loss between the gating coefficient and the wake intensity score, and the wakeless consistency loss where the predicted value is close to the baseline value on the wakeless sample. The three losses are weighted according to preset weights, and the stochastic gradient descent algorithm is used to perform end-to-end joint training on the baseline prediction subnetwork, the wake residual subnetwork, and the gating subnetwork.

[0029] 8) Online Prediction and Application

[0030] The trained network is deployed in the wind farm operating environment, and updated upstream and downstream unit monitoring data and wake condition characteristics are input in real time to calculate the single-step or multi-step power prediction results of the target downstream unit. This is used for power generation planning, unit operation optimization, and coordinated control with energy storage, hydrogen production and other devices.

[0031] Compared with the prior art, the beneficial effects of the present invention are:

[0032] 1) Explicitly introduce wake sensing conditional residual structure

[0033] The predicted power of downstream units is expressed using a structure of "baseline subnetwork + wake residual subnetwork + gating subnetwork". It automatically weakens when there is no wake or a weak wake. f w Contribution, enhanced during severe wake. f w Compared to existing methods that directly output single-path predictions, this approach is more advantageous in simultaneously ensuring prediction accuracy under both normal and wake conditions.

[0034] 2) Ensure non-wake flow does not degrade through gating supervision and consistency constraints.

[0035] Incorporating both the gating factor and the wake intensity into the loss function D.U. * Consistency constraints, and on wakeless samples The consistency constraint enables the model to learn the ability to correct wake residuals while keeping the prediction of non-wake samples close to the baseline. Compared with the residual structure that only adds wake features at the input or lacks explicit constraints, it is easier to meet the requirement of "strengthening wake prediction without significantly deteriorating non-wake accuracy" in engineering.

[0036] 3) Embed the geometry and wake physical quantities of upstream and downstream units into the network structure.

[0037] In the feature and structural design, upstream and downstream SCADA data, as well as geometric quantities such as unit spacing, lateral offset, and wind direction deviation angle, are used simultaneously to determine the wake intensity. D.U. * The system directly inputs gating and residual subnetworks to the operating sector, rather than relying solely on external wake models or field-level meteorological downscaling. Compared to existing technologies, this improves the model's adaptability and transferability to different station layouts and wake intensity levels without requiring fine calibration of complex wake models. Attached Figure Description

[0038] Figure 1 This is a schematic diagram of the wind power prediction model based on a wake-sensing conditional residual neural network.

[0039] Figure 2 The flowchart shows the wind power prediction method based on wake sensing conditional residual neural network.

[0040] Figure 3 A bar chart comparing MAE of different prediction models under full-sample conditions;

[0041] Figure 4 A bar chart comparing the MAE of different prediction models under the wakeless condition;

[0042] Figure 5 A bar chart comparing the MAE of different prediction models under typical wake conditions;

[0043] Figure 6 A bar chart comparing the MAE of different prediction models under severe wake conditions. Detailed Implementation

[0044] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0045] The structures, proportions, and sizes illustrated in the accompanying drawings are merely for illustrative purposes and to aid those skilled in the art in understanding and reading the invention. They are not intended to limit the scope of the invention and therefore have no substantial technical significance. Any modifications to the structure, changes in proportions, or adjustments to size, provided they do not affect the effectiveness or purpose of the invention, should still fall within the scope of the technical content disclosed herein. Furthermore, the terms "upper," "lower," "left," "right," "middle," and "one" used in this specification are merely for clarity and not intended to limit the scope of the invention. Changes or adjustments to their relative relationships, without substantially altering the technical content, should also be considered within the scope of the invention's implementation.

[0046] The structure of the wind power prediction model based on the wake-sensing conditional residual neural network is as follows: Figure 1 As shown, the model consists of a baseline layer (NN), a wake layer (NN), and a soft-gated (switching) layer. Input features are divided into two parts: general data and wake data. General data includes historical power of downstream turbines, downstream turbine wind speed, wind direction, and the pitch angles of the three blades, and is input to both the baseline layer (NN) and the wake layer (NN). Wake data includes upstream turbine wind speed, wind direction, and geometric information such as the distance and relative offset between upstream and downstream units, and is input to the Gaussian wake model sublayer within the wake layer (NN). The baseline layer (NN) outputs the baseline power. P base The wake layer neural network outputs wake power with the assistance of physical characteristics generated by the Gaussian wake model. P wake Operating characteristics (such as whether it is in the wake sector, wake intensity) ΔU* Input the soft gating layer (etc.) to obtain the gating coefficient. g(z) ,right P wake After weighting and P base The predicted power output of the downstream wind turbine is obtained by superimposing the results at the adder node.

[0047] Figure 2 As shown, a wind power prediction method based on a wake perception conditional residual neural network is presented. The method mainly includes steps such as minute-level data processing, wake physical characteristic calculation, baseline power prediction, wake power prediction, gating coefficient calculation, and conditional residual superposition. First, SCADA data from upstream and downstream units are aggregated, cleaned, and time-aligned at 1-minute intervals to obtain minute-level data. Then, a Gaussian wake layer with learnable parameters is used to calculate the wake physical characteristics based on the minute-level data and geometric information, and this information is input into the wake layer neural network along with general data to obtain the desired results. P wake Simultaneously, the general data is input into the baseline layer NN to obtain... P baseThen mark the sectors. ΔU* , x / D、 y / D and other operating condition characteristics are input into the gating layer to obtain the gating coefficient. g(z) Implemented at the "×" node g(z) and P wake The product of the two, at the "+" node and P base Added together, the output of the downstream wind turbine is... t + D Power prediction results at time 1.

[0048] The detailed steps of the wind power prediction method are as follows:

[0049] 1. Data Acquisition and Preprocessing

[0050] 1.1 SCADA Data Acquisition

[0051] Raw SCADA data from upstream and downstream units is obtained from the existing monitoring system of the wind farm, with a typical sampling interval of 1 second, and includes at least, but not limited to, the following:

[0052] Wind speed signal: Upstream unit wind speed U upraw Downstream unit wind speed U downraw ;

[0053] Power signal: Active power of upstream units P upraw Downstream unit active power P downraw ;

[0054] Wind direction signal: Upstream unit wind direction Dir upraw Downstream unit wind direction Dir downraw ;

[0055] Blade pitch angle: Pitch angle of the three blades of the downstream unit Pitch 1 , Pitch 2 , Pitch 3 ;

[0056] Other optional operating parameters: relative wind direction of the nacelle, unit operating status code, etc.

[0057] 1.2 Data Cleaning and Aggregation

[0058] The raw SCADA data is coarsely cleaned to remove abnormal samples such as shutdowns, faults, disconnections, obvious power limitations, negative power, and negative wind speeds. The cleaned data is then aggregated according to time windows to obtain minute-level data sequences with a time resolution of 1 minute. The aggregation method can be either arithmetic mean or median. After aggregation, the following results are obtained: U up ( t ), U down ( t ), P up ( t ), P down ( t ), Dir up ( t ), Dir down ( t ), Pitch 1 ( t ), Pitch 2 ( t ), Pitch 3 ( t ) etc., among which t This indicates a minute-level time index.

[0059] 1.3 Partitioning of Training and Test Sets

[0060] While maintaining chronological order, minute-level data is divided into training, validation, and test sets based on dates or time periods. For example, the first few days of a month can be used as the training set, the middle few days as the validation set, and the last few days as the test set to prevent information leakage.

[0061] 1.4 Normalization Processing

[0062] The numerical features input to the network are normalized using standardization or interval scaling methods. Mean-variance standardization can be used. ,in, x Represent any input feature to be normalized. m ξ , s ξ These represent the mean and standard deviation of the feature on the training set, respectively, and remain consistent across the validation and test sets. Power output can be based on the unit's rated power. P r Normalization is performed to obtain This improves training stability.

[0063] 2. Geometric information of upstream and downstream units and classification of wake conditions

[0064] 2.1 Geometric Information Acquisition

[0065] Obtain the plane coordinates of upstream and downstream units based on wind farm design data. X up , Y up ), ( X down , Y down ), hub height H and impeller diameter D. Calculate the azimuth angle of the line connecting the centers of the upstream and downstream units. i geo and the projected distance of upstream and downstream units along the main wake axis x ud Horizontal offset y ud :

[0066] ;

[0067] lateral offset y ud It can be calculated from the (X,Y) difference by rotating the coordinates;

[0068] For a selected single upstream and downstream unit pair, if the line connecting the centers of the two units is taken as the geometric principal axis, then the lateral offset distance y ud It can be set to 0. The normalized distance is defined as... x D = x ud / D、 y D = y ud / D.

[0069] 2.2 Wake Main Sector Division

[0070] Connect the azimuth angles using geometric lines i geo Set the half-width of the symmetrical sector as the center. Dth ,For example Dth =15°, define the geometric wake main sector as [ i geo - Dth , i geo + Dth [Based on wind direction measured by downstream units] Dir down ( t Determine if it is in the geometric wake sector:

[0071] insect ( t )=1, when Dir down ( t )∈[ i geo - Dth , i geo + Dth ]hour;

[0072] otherwise insect ( t )=0.

[0073] 2.3 Advection Alignment and Wake Intensity Calculation

[0074] Considering the wake propagation delay, the projected distance between the upstream and downstream units along the main wake axis. x ud and upstream wind speed U up ( t Estimate advection time t ( t )= x ud / U up ( t ), and round it to the nearest minute. k This aligns the upstream signal with the downstream time:

[0075] U upshift ( t )= U up ( tk );

[0076] P upshift ( t )= P up ( tk ).

[0077] When only data from both upstream and downstream is available after alignment, the normalized wind speed attenuation is calculated for downstream units. ΔU* :

[0078] ΔU* ( t )=1- U down ( t ) / U upshift ( t (1)

[0079] when Uupshift ( t If the value is too small or an anomaly exists, discard the corresponding sample. Several operating condition ranges can be set based on ΔU*, for example:

[0080] ΔU* A value <0.05 is considered nonwake.

[0081] 0.05≤ ΔU* A value <0.20 is considered a normal insector.

[0082] ΔU* A value ≥0.20 is considered a severe wake.

[0083] To facilitate learning by the gating network, ΔU* is linearly or nonlinearly normalized to the interval of 0 to 1 to obtain the wake intensity score. s wake (t).

[0084] 3. Feature Vector Construction

[0085] 3.1 Baseline Feature Vector

[0086] At each time t Construct feature vectors for the baseline subnetwork x base ( t ), including but not limited to:

[0087] Current wind speed of downstream units U down ( t );

[0088] Current wind direction of downstream units Dir down ( t );

[0089] Current blade pitch angle of downstream unit Pitch 1 ( t ), Pitch 2 (t ), Pitch 3 ( t );

[0090] Current or historical power of downstream units P down ( t- 1) P down ( t- 2) ... to characterize short-term inertia and lag effects;

[0091] Optional parameters include air density estimates and wind speed variation rates.

[0092] 3.2 Wake Feature Vector

[0093] exist x base ( t Based on this, a wake feature vector is formed for the wake residual subnetwork. x wake ( t ), additionally includes:

[0094] Wind speed after upstream fan alignment U upshift ( t );

[0095] Power after upstream wind turbine horizontal alignment P upshift ( t );

[0096] Upstream and downstream geometric quantities x D , y D ;

[0097] Geometric wind direction error i err ( t )= Dir down ( t )- i geo (Normalized by 360°);

[0098] Physical characteristics calculated from the Gaussian wake model, such as wake wind speed ratio and overlap coefficient.

[0099] 3.3 Operating Condition Feature Vector

[0100] Constructing the operating condition feature vector for the gated subnetwork z ( t ),include:

[0101] Geometric sector marking insect ( t );

[0102] Wake intensity rating weak ( t );

[0103] x D , y D , i err ( tGeometric quantities such as )

[0104] Optional wind speed range markings (e.g., before / after rated speed).

[0105] 4. Neural Network Structure Design

[0106] The wake sensing conditional residual neural network in this embodiment mainly consists of a baseline prediction subnetwork. f 0 Wake residual subnetwork f w Gaussian wake physics layer and gated subnetwork g It consists of four parts.

[0107] 4.1 Baseline Prediction Subnetwork

[0108] Baseline prediction subnetwork

[0109] f 0 Implemented using a multi-layer feedforward neural network, with x base ( t (Input) is the downstream unit's position at the predicted time. t + D Baseline power p base ( t + D Let the input dimension be... d base The network contains L 0 There are 1 fully connected hidden layer, each with a width of 1. H 0 The activation function can be ReLU, LeakyReLU, etc. Its forward propagation form is:

[0110] h 0 ( 0 )= x base ( t (2)

[0111] h 0 ( l )= s ( W 0 ( l ) h 0 ( l -1)+ b 0 ( l )), l =1,2,…,L 0 (3)

[0112] p base ( t + D )= W 0 ( out ) h 0 ( L 0 )+ b 0 ( out (4)

[0113] in, x base ( t )express t The baseline input feature vector of the downstream wind turbine at any given time, such as the historical power of the wind turbine, wind speed, wind direction and blade pitch angle, etc. h 0 ( 0 () represents the input layer output. h 0 ( l ) is the first l The output vector of the hidden layer; W 0 ( l ) for connecting the first l -1st floor and the l The weight matrix of the layer, b 0 ( l ) represents the corresponding bias vector. l =1,2,…, L 0 ; s (·) represents an element-wise nonlinear activation function, such as ReLU, tanh, or other common activation functions. L 0 This represents the number of hidden layers in the baseline subnetwork. W 0 ( out ) represents the output layer weight matrix. b 0 ( out ) represents the output layer bias term. p base ( t + D ) as the baseline subnetwork in t + DThe downstream wind turbine power prediction value is provided at any given time (a scalar for single-step prediction, and a vector composed of multiple prediction step sizes for multi-step prediction). It can be extended to a multi-head structure at the output layer as needed, for example, outputting simultaneously... D =1~5 multi-step prediction sequence.

[0114] 4.2 Gaussian wake physical layer

[0115] The Gaussian wake physics layer is embedded before or in the middle layer of the wake residual subnetwork input, in order to x wake ( t In ) U upshift ( t ), x D , y D Using these parameters as inputs, the downstream incoming wind speed attenuation and velocity distribution are approximated using the Gaussian wake model. Taking the classic Bastankhah Gaussian model as an example, let the upstream turbine thrust coefficient be... C T The half-width of the Gaussian wake is s ( x ), then the distance x The normalized velocity deficit at a given point can be expressed as:

[0116] (5),

[0117] in, x This represents the downstream position coordinates along the main wake axis in the wake model; when calculating the wake loss at the target downstream unit location, x = x. ud When the wheel hub heights are approximately the same, the radial distance r It can be approximated as the absolute value of the lateral offset distance, i.e., r≈y ud ; d U ( x,r ) indicates the downstream location ( x,r Normalized velocity loss at ) U ( x,r () indicates the actual wind speed at that location; U upshift This indicates the upstream airflow velocity after alignment with the advection delay. r The radial distance relative to the wake centerline is obtained by combining the lateral and vertical offsets; e The base of the natural constant. Based on equation (5), the local wind speed ratio can be further constructed. Speed ​​loss d U ( x,r The wake physical characteristics are denoted as follows: f wake ( t ), and with x wake ( t The enhanced wake input vector is formed by splicing the vectors together. .

[0118] 4.3 Wake Residual Subnetwork

[0119] Wake residual subnetwork f w Accept the enhanced wake input vector As input, a multi-layer feedforward network structure is used to output the wake correction power component. p wake ( t + D Its structural form is similar to f 0 But number of floors L w and width h w Different:

[0120] (6)

[0121] h w ( l )= s ( W w ( l ) h w ( l -1)+ b w ( l )), l =1,2,…, L w (7)

[0122] p wake ( t + D )= W w ( out ) h w ( L g )+ b w ( out (8)

[0123] in, Indicates in tThe wake-related input feature vector constructed at each moment includes upstream and downstream unit wind speeds, wind directions, unit geometry, and physical features obtained from the Gaussian wake layer. f wake ( t ); h w ( 0 () represents the input layer output of the wake residual subnetwork. h w ( l ) is the first l Hidden layer output; W w ( l ), b w ( l ) are respectively the first l Layer weight matrix and bias vector; s (·) represents the element-wise nonlinear activation function; L w is the number of hidden layers in the wake residual subnetwork; L g The index of the last layer connected to the output layer (usually taken as...) L g =L w ); W w ( out ), b w ( out These are the output layer weight matrix and bias term, respectively. p wake ( t + D ) is the wake residual subnetwork in t + D The power correction component output at each step is a scalar in single-step prediction and can be expanded into a vector in multi-step prediction for use with the baseline power. p base ( t + D The final predicted power is obtained by adding the two together.

[0124] Like the baseline subnetwork, it can be expanded to a multi-head output for multi-step prediction.

[0125] 4.4 Gated Subnetwork

[0126] Gated subnets g Based on working condition characteristics z ( t () is the input, and the output is the gating coefficient. g ( t + D)∈[0,1], adopting a lightweight fully connected structure, the number of hidden layers L g Fewer (e.g., 1-2 layers), width of each layer H g It is relatively small. Its forward propagation form is:

[0127] h g ( 0 ) =z ( t (9)

[0128] h g (l)=σ(W g ( l ) h g (l - 1) +b g ( l )) ,l=1,2,…,L g (10)

[0129] g ( t + D )= s gate ( W g ( out ) h g ( L g )+ b g ( out (11)

[0130] in, z ( t ) indicates in t The operating condition characteristic vector constructed at any given time includes an indication of whether it is in the wake sector and the wake intensity Δ. U* Operating condition-related characteristics such as wind direction and sector number; h g ( 0 ) represents the input layer output of the gated subnetwork. h g ( 0 ) is the first l Hidden layer output; W g ( l ), b g ( l ) are respectively the first lLayer weight matrix and bias vector; s (·) represents the element-wise nonlinear activation function; L g is the number of hidden layers in the gated subnetwork; W g ( out ), b g ( out These are the output layer weight matrix and bias term, respectively. s gate (·) is the gated output activation function, preferably the Sigmoid function, used to restrict the gate coefficient to the interval [0,1]; g ( t +Δ) is in t The gating coefficient at time +Δ indicates that the closer the value is to 1, the greater the weight of the wake residual correction; the closer the value is to 0, the more the prediction results of the baseline subnetwork are used. In the case of multi-step prediction, the output of the gating subnetwork can be expanded into a vector of length Horizon, providing the corresponding gating coefficients for different prediction step sizes.

[0131] 4.5 Conditional Residual Output

[0132] Final predicted normalized power It is given by the following formula:

[0133] (12);

[0134] When restoring to actual power, multiply by the rated power. P r, Equation (12) embodies the conditional residual structure: when g When g approaches 0, the model degenerates into a single baseline prediction; when g approaches 1 and p wake When the value is negative, the model provides a significant power correction under severe wake conditions.

[0135] 5. Joint Loss Function and Training Method

[0136] 5.1 Prediction Error Loss

[0137] For each sample ( t , D Define prediction error loss. L pred , can be adopted L 1 loss or L 2 Loss, for example:

[0138] (13)

[0139] in, P is the predicted power of downstream units output by the model. down ( t + D () represents the measured power at the corresponding time, and t represents the current sample time. D The step size is used for prediction. Formula (13) represents the magnitude of the error between the predicted and measured values, which is used to constrain the overall fitting accuracy of the model on the entire sample. The smaller the error, the closer the model's prediction result is to the true power; the larger the error, the more obvious the prediction deviation of the model on the corresponding sample. During training, this loss is averaged over all samples and used as an important component of the total loss function.

[0140] 5.2 Gated monitoring losses

[0141] To guide the gating network output to match the actual wake intensity, a gating supervision loss is introduced.

[0142] L gate Wake intensity can be scored. s wake ( t + D Normalized to [0,1] as soft labels, and then... L 2 Or in cross-entropy form:

[0143] L gate =( g ( t + D )- s wake ( t + D )) 2 (14)

[0144] Based on this, it is possible to select only samples within the geometric sector or... ΔU* Samples exceeding a certain threshold are weighted and calculated.

[0145] 5.3 No-wake consistency loss

[0146] To ensure that the prediction of non-wake samples does not deteriorate, in ΔU* ( t + D Smaller (e.g., <0.05) and insect ( t + D On samples where )=0, constraints With baseline subnetwork output Pbase Approximately. Let the set of wakeless samples be:

[0147] S cons ={( t , D )∣ ΔU* ( t + D <0.05, insert( t + D )=0},

[0148] right S cons Define a single-sample loss for each sample. l cons :

[0149] ,

[0150] Then there is no wake consistency loss L cons for

[0151] (15)

[0152] Among them, P base ( t + D () is the baseline subnetwork f 0 The predicted output power. This loss is calculated only on samples with no wake or a weak wake, thus the gating coefficient is used for such conditions. g ( t + D When the value approaches 0, the wake residual channel does not participate in the correction.

[0153] 5.4 Total Loss Function

[0154] During training, the three types of losses mentioned above are weighted according to coefficients. l pred , l gate , l cons We obtain the total loss by weighted summation. L total :

[0155] L total = l pred L pred + l gate L gate +λ cons L cons (16)

[0156] in l pred >0、 l gate ≥0、 l cons ≥0 can be determined through empirical or validation set adjustments, for example... l pred =1, l gate =0.2~0.5, l cons =0.5~1.

[0157] 5.5 Training Steps

[0158] (1) Initialize network parameters W 0 , b w , W w , b w , W g , b g Learnable parameters in the Gaussian wake layer can be initialized using Xavier or Kaiming.

[0159] (2) Construct batch samples on the training set in chronological order, each sample containing x base ( t ), , z ( t ),Label p down ( t + D and wake intensity score s wake ( t + D );

[0160] (3) x base ( t ), , z ( t They are respectively sent to the baseline subnetwork. f 0. Wake residual subnetwork f w and gated subnets gThe baseline prediction is obtained by forward calculation according to equations (2)-(12). p base ( t + D Wake correction term p wake ( t + D ), Gating coefficient g ( t + D ), and obtain the total predicted output. ;

[0161] (4) Calculate according to formulas (13) to (16). L total Backpropagation and parameter updates are performed using optimization algorithms such as Adam.

[0162] (5) Monitoring on the validation set L pred Alternatively, partition MAE and select the optimal training rounds based on the early stopping strategy;

[0163] (6) Save the optimal model parameters for subsequent online prediction.

[0164] 6. Online Prediction and Application Deployment

[0165] 6.1 Online Data Access

[0166] In the actual operating environment of a wind farm, the real-time, second-level data output by the SCADA system is connected to a server or embedded computing device, and subjected to coarse cleaning and minute-level aggregation according to the rules in the embodiment to generate real-time updated data. U up ( t ), U down ( t ), P down ( t ), Dir up ( t ), Dir down ( t ), Pitch 1 ( t )…wait for the signal.

[0167] 6.2 Operating Condition Identification and Feature Generation

[0168] For each latest time step t Calculate using geometric relationships and wind direction insect (t), calculated using historically aligned upstream and downstream wind speeds. ΔU* (t The wake intensity score is obtained according to equation (1) and the threshold values ​​for each interval. s wake ( t Simultaneously construct x base ( t ), , z ( t ), and standardized using the same normalization method as the training phase.

[0169] 6.3 Network Forward Inference

[0170] Will x base ( t ), , z ( t Input the trained baseline subnetwork f 0 Wake residual subnetwork f w and gated networks g ,get p base ( t + D ), p wake ( t + D ), g ( t + D Then, the normalized predicted value is obtained according to equation (12). and multiplied by the rated power P r get .

[0171] 6.4 Multi-step rolling prediction

[0172] Multiple prediction step sizes can be set according to actual needs. D The power prediction sequence for the next few minutes is obtained by using a multi-head output or cyclic network call method, ∈{1,2,3,4,5}min. The prediction results are output to the wind farm dispatch system, the unit local control system, or the superior dispatch center for rolling power generation plan preparation, ramp constraint checking, and power coordination control with energy storage, hydrogen production and other equipment.

[0173] 6.5 Model Update

[0174] During the long-term operation of the system, newly added SCADA data can be periodically added to the training set, and the network parameters can be retrained or fine-tuned according to the above steps to adapt to factors such as seasonal changes, equipment aging, and adjustments to operating strategies, thereby improving the long-term stability and generalization ability of the model.

[0175] like Figure 3 The bar chart comparing the MAE of different prediction models under full-sample conditions is shown. The horizontal axis represents different prediction models, including the method of this invention (conditional residual neural network), wake-only structure, wakeless structure, full wakeless structure, as well as RF, SVR, TCN, LSTM, RNN, GRU, CNN, TFT, etc. The wake-only structure means that only the wake layer output is retained as the final prediction result; the wakeless structure means that only the baseline layer output is retained as the final prediction result, its equivalent form being soft-gated and always 0; the full wakeless structure means that the baseline layer and the wake layer are directly added to obtain the final prediction result, its equivalent form being soft-gated and always 1. The vertical axis represents the mean absolute error (MAE) (kW) on the full-sample test set under the condition of Δ=1. The red bars on the left represent the conditional residual model of this invention, the green columns represent ablation experimental models under different structural settings, and the blue columns on the right represent the comparison models of traditional machine learning and deep learning, used to demonstrate the predictive performance advantage of this invention on the overall sample.

[0176] like Figure 4 The bar chart comparing the MAE of different prediction models under wakeless conditions is shown. It compares the MAE of the proposed method, wake-only structures, wakeless structures, full-wake structures, and models such as RF, SVR, TCN, LSTM, RNN, GRU, CNN, and TFT on wakeless condition samples. In the chart, the red bars still correspond to the proposed model, the green columns represent the ablation results of different wake processing methods, and the blue columns represent the performance of various comparative models, illustrating that the proposed method maintains or outperforms existing methods in predictive accuracy under wakeless conditions.

[0177] like Figure 5 The bar chart comparing the MAE of different prediction models under typical wake conditions shows a comparison of the test MAE of each model on a typical wake condition sample. The chart reflects that under moderate wake disturbance, the conditional residual structure of this invention fully utilizes wake information while suppressing unnecessary downward correction, resulting in lower error compared to using only the baseline model or a simple wake correction model.

[0178] like Figure 6The bar chart comparing the MAE of different prediction models under severe wake conditions shows that the MAE of each model was compared on a sample of severe wake conditions. It can be seen that when ΔU* is large and the wake effect is significant, the method of this invention, corresponding to the red bar, has the lowest error among the compared models. This indicates that introducing a Gaussian wake physical layer in the form of gated residuals can more accurately capture the power loss caused by severe wakes and improve the prediction reliability of downstream units under extreme conditions.

[0179] While the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of the present invention are still within the scope of protection of the present invention.

Claims

1. A wind power prediction method based on a wake-sensing conditional residual neural network, characterized in that, include: (1) Data acquisition and processing; Acquire full-condition operating data of the floating wind and wave combined power generation system and preprocess the raw data; Divide the dataset into training and testing sets; (2) Acquisition and alignment of upstream and downstream unit data: acquire monitoring data and geometric parameters of the target downstream unit and its main upstream units; perform horizontal alignment of upstream and downstream unit data according to a unified time axis, and construct baseline feature vector, wake feature vector and operating condition feature vector; (3) Wake condition identification and intensity feature extraction: Divide the main wake sector according to the wind direction and the relative position of the unit, and identify whether the sample is in the wake sector; By combining the wind speeds of upstream and downstream units after advection alignment, a normalized wind speed attenuation index is calculated. The samples are divided into no wake, general wake, and severe wake, and the wake intensity is smoothly mapped to a continuous score of 0 to 1 as a gated monitoring signal. (4) Baseline prediction subnetwork construction: Using the baseline feature vector as input, a multi-layer feedforward neural network is constructed to output the baseline power prediction value at the corresponding time, which is used to characterize the typical power-wind speed relationship under conditions of no wake or weak wake. A multi-head structure is set in the output layer to realize multi-step prediction. (5) Construction of wake residual subnetwork: Based on the baseline feature vector input, the wake feature vector is introduced to construct the wake residual subnetwork; Gaussian wake physical layer is embedded before or in the middle layer of the wake residual subnetwork to approximate the wind speed attenuation or power loss, and the wake residual is output to make additional corrections to the baseline prediction results. (6) The wake gating subnetwork and conditional residual are fused. The working condition feature vector is used as input and the gating coefficient is output. A lightweight fully connected structure is adopted to construct a lightweight wake gating subnetwork and output a gating coefficient in the range of 0 to 1, which represents the weight of the wake residual in the correction. The outputs of the three subnetworks are fused in the form of conditional residual "predicted power = baseline prediction + gating coefficient × wake residual" and the gating coefficient is shared for multi-step prediction scenarios. (7) Design and training of joint loss function: Design a joint loss function, weight the three losses according to the preset weights, and use the stochastic gradient descent algorithm to perform end-to-end joint training of the baseline prediction subnetwork, the wake residual subnetwork and the gated subnetwork. (8) Online prediction and application: Deploy the trained network in the wind farm operating environment, input the updated upstream and downstream unit monitoring data and wake condition characteristics in real time, calculate the single-step or multi-step power prediction results of the target downstream unit, and use them for power generation planning, unit operation optimization and coordinated control with energy storage and hydrogen production devices.

2. The wind power prediction method based on wake-sensing conditional residual neural network as described in claim 1, characterized in that, The data acquisition and processing in step (1) specifically involves acquiring full-condition operating data of the floating wind and wave combined power generation system through sensors and high-fidelity models; performing preprocessing such as normalization, denoising, and feature extraction on the raw data; and dividing the data into training and testing sets to provide a data foundation for subsequent modeling.

3. The wind power prediction method based on wake-sensing conditional residual neural network as described in claim 1, characterized in that, The monitoring data in step (2) includes wind speed, wind direction, active power and pitch angle; the geometric parameters include the distance between units, relative orientation and impeller diameter; the basic input characteristics of the downstream units include historical power, wind speed, pitch angle and wind direction.

4. The wind power prediction method based on wake-sensing conditional residual neural network as described in claim 1, characterized in that, The advection alignment in step (2) specifically involves taking into account the wake propagation delay, estimating the advection time based on the upstream and downstream distances and the upstream wind speed, and rounding it to the nearest minute to align the upstream signal with the downstream time.

5. The wind power prediction method based on wake-sensing conditional residual neural network as described in claim 1, characterized in that, The specific steps in step (3) of dividing the wake main sector are as follows: taking the azimuth angle of the geometric connection line as the center, setting the half-width of the symmetrical sector, defining the geometric wake main sector, and determining whether it is in the geometric wake sector based on the wind direction measured by the downstream unit.

6. The wind power prediction method based on wake-sensing conditional residual neural network as described in claim 1, characterized in that, In step (5), a wake residual subnetwork is constructed, and a multi-layer feedforward network structure is adopted to output the wake correction power component, which is expanded into a multi-head output for multi-step prediction.

7. The wind power prediction method based on wake-sensing conditional residual neural network as described in claim 1, characterized in that, The joint loss function in step (7) consists of the overall prediction error loss, the consistency loss between the gating coefficient and the wake intensity score, and the wakeless consistency loss where the predicted value is close to the baseline value on the wakeless sample.

8. The wind power prediction method based on wake-sensing conditional residual neural network as described in claim 7, characterized in that, The total loss function of the joint loss function in step (7) is obtained by weighting and summing the prediction error loss, the consistency loss of the gating coefficient and the wake intensity score, and the consistency loss without wake according to the weight coefficients during the training process.

9. The wind power prediction method based on wake-sensing conditional residual neural network as described in claim 7, characterized in that, The end-to-end joint training in step (7) includes: (a) Initialize network parameters and learnable parameters in the Gaussian wake layer; (b) Construct batch samples in chronological order on the training set; (c) The baseline input feature vector, enhanced wake feature, and operating condition feature are fed into the baseline sub-network, wake residual sub-network, and gate sub-network, respectively. The baseline prediction, wake correction term, and gate coefficient are obtained by forward calculation, and the total prediction output is obtained. (d) Calculate the total loss and use the Adam optimization algorithm for backpropagation and parameter update; (e) Monitor the prediction error loss or partition mean absolute error on the validation set and select the optimal training rounds based on the early stopping strategy; (f) Save the optimal model parameters for subsequent online prediction.

10. The wind power prediction method based on wake-sensing conditional residual neural network as described in claim 1, characterized in that, The multi-step rolling prediction in step (8) specifically involves setting multiple prediction step lengths according to actual needs, and using a multi-head output or cyclic network call method to obtain a power prediction sequence for the next few minutes; the prediction results are output to the wind farm dispatch system, the unit local control system or the upper-level dispatch center for rolling power generation plan preparation, ramp constraint check and power coordination control with energy storage and hydrogen production equipment.

Images