A wind power prediction method and system based on wind speed correction
By combining a UAV wind measurement device with an improved least squares support vector machine model optimized by a quantum genetic algorithm and a BP neural network, a mapping model from the wind speed behind the wind turbine to the wind speed in front of the wind turbine is established. This solves the problem of insufficient wind power prediction accuracy caused by the measurement error of the wind speed behind the wind turbine, and realizes high-precision wind power prediction.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- HEFEI SIZHEN CHIP TECH CO LTD
- Filing Date
- 2026-04-30
- Publication Date
- 2026-07-24
AI Technical Summary
In existing wind power prediction methods, the large measurement error of wind speed behind the wind turbine leads to insufficient accuracy in wind power prediction. In particular, the accuracy of model prediction is further reduced under complex terrain conditions, and existing methods have poor adaptability to different wind fields.
A drone equipped with a wind measuring device is used to measure wind speed in front of the wind turbine. A mapping model between the wind speed behind the wind turbine and the wind speed in front of the wind turbine is established. The BP neural network is optimized by combining quantum genetic algorithm and an improved least squares support vector machine model to predict wind power. The prediction accuracy is improved by correcting the wind speed data.
It significantly improves the accuracy of wind power prediction, can dynamically adapt to changes in the wind field environment, maintains long-term high-precision prediction results, and reduces grid dispatch costs.
Smart Images
Figure CN122456469A_ABST
Abstract
Description
Technical Field
[0001] This application belongs to the field of new energy power generation, and specifically relates to a power prediction method based on wind speed. Background Technology
[0002] With the increasing depletion of fossil fuels and the growing prominence of environmental problems, wind power, as a clean and renewable energy source, has experienced rapid development globally. However, wind power is characterized by volatility, intermittency, and randomness, and the integration of large-capacity wind power into the grid poses a severe challenge to the safe and stable operation of the power system. Accurate wind power forecasting is an effective way to address this issue, enhancing system safety, reliability, and controllability. Based on the predicted output curves of wind farms, the output of conventional turbines can be optimized, reducing operating costs.
[0003] Currently, wind farm power prediction still faces some intractable technical problems. The wind speed in the wind turbine power curve should be the incoming wind speed at the hub height (i.e., the wind speed in front of the rotor), but in practice, the wind speed measured by the anemometer installed at the rear of the nacelle (i.e., the wind speed behind the rotor) is widely used. Due to the absorption of some wind energy by the rotor and the influence of the wake, the wind speed behind the rotor is lower than the wind speed in front of the rotor. Directly using the wind speed behind the rotor to establish a power prediction model will introduce systematic errors. The absolute average error of existing short-term wind speed prediction methods is relatively large (above 20%), and the prediction accuracy is difficult to meet the needs of grid dispatch. Commonly used statistical methods include Kalman filtering, time series methods, neural network methods, fuzzy logic methods, and support vector machine methods, but these methods are subject to strong regional limitations and have poor adaptability to different wind fields. In addition, traditional wind farm power prediction models usually use single-point wind speed as input, ignoring the three-dimensional spatial distribution characteristics of the wind field. Especially under complex terrain conditions such as mountains, the airflow is significantly affected by terrain lifting, acceleration, and separation, further reducing the prediction accuracy of the model.
[0004] Therefore, improving the accuracy of wind speed measurement in wind farms, and thus enhancing the accuracy of wind power prediction, has become a pressing technical problem to be solved in this field. Summary of the Invention
[0005] To address the accuracy issues in wind speed measurement and power prediction mentioned above, this application proposes a wind power prediction method based on wind speed correction, the specific scheme of which is as follows.
[0006] This application discloses a wind power prediction method based on wind speed correction, including the following steps: S1 uses a drone equipped with wind measurement equipment to measure the actual wind speed time series data in front of the wind turbine rotor; S2, collect historical operating data of the wind turbine at the same time as step S1, including wind speed time series data measured by the wind turbine anemometer, wind turbine state parameter time series data, and actual output power time series data of the wind turbine, and synchronize and align with the data collected in step S1 in time. S3. Based on the synchronous data obtained in steps S1 and S2, establish a mapping model from the wind speed behind the wind turbine to the wind speed in front of the wind turbine. The input of the mapping model includes the wind speed behind the wind turbine and the state parameters of the wind turbine unit, and the output is the equivalent wind speed in front of the wind turbine. S4. Convert the historical wind speed behind the wind turbine in step S2 into the historical equivalent wind speed in front of the wind turbine through the mapping model established in step S3, and form a corrected historical wind speed dataset. S5. Based on the corrected historical wind speed dataset obtained in step S4, the wind turbine state parameters and actual output power collected in step S2, a calibrated wind power prediction model is trained. S6. In the real-time prediction stage, the wind speed behind the wind turbine and the state parameters of the wind turbine are collected in real time and input into the mapping model established in step S3 to obtain the equivalent wind speed in front of the wind turbine at the current moment. S7. Input the current equivalent wind speed in front of the wind turbine, the corrected historical wind speed dataset, the wind turbine state parameters and actual output power collected in step S2, and the real-time collected wind turbine state parameters into the calibrated wind power prediction model trained in step S5, and output the predicted wind power value for future times.
[0007] Preferably, in step S1, the wind measurement device carried by the UAV is a hovering multi-rotor UAV, which is equipped with one or more combinations of ultrasonic wind measurement, lidar or quantum radar.
[0008] Furthermore, the mapping model in step S3 employs a BP neural network model optimized by the quantum genetic algorithm (GQA), and its construction process includes: The structure of the input layer, hidden layer and output layer of the BP neural network is determined, where the input layer nodes include the wind speed behind the wind turbine, the blade rotation speed and the pitch angle, and the output layer nodes are the wind speed in front of the wind turbine. The initial weights and thresholds of the BP neural network are optimized using a quantum genetic algorithm (GQA). The optimized initial weights and thresholds are substituted into the BP neural network for training to obtain the final mapping model.
[0009] Furthermore, the establishment of the mapping model in step S3 is triggered by any of the following conditions: Perform periodic calibration at preset time intervals; Calibration is triggered when the average wind direction change in the wind field exceeds a preset threshold. Calibration is triggered when meteorological conditions change significantly, including one or more of the following: average temperature, rainfall, air pressure, and atmospheric particulate matter concentration.
[0010] Furthermore, the calibrated wind power prediction model in step S5 adopts a hybrid parameter input, including the corrected historical wind speed dataset, the wind speed behind the wind turbine, the wind turbine state parameters, and the actual output power. When the data in front of the wind turbine is occasionally missing, the time series data at that point only uses the wind speed behind the wind turbine.
[0011] Furthermore, the wind power prediction model in step S5 adopts an improved least squares support vector machine (LSSVM) model, the construction process of which includes: The inequality constraints in the traditional support vector machine are replaced with equality constraints, and the prediction model is obtained by solving a system of linear equations. The regularization parameter C and kernel function parameter σ of the least squares support vector machine are optimized using the quantum particle swarm optimization (QPSO) algorithm.
[0012] Furthermore, a wind speed prediction step is included before step S7: Based on historical equivalent wind turbine inlet wind speed time series data, a specific wind speed prediction model is used to predict the equivalent wind turbine inlet wind speed at future times, and the predicted value is input into the calibrated wind power prediction model for power prediction.
[0013] Furthermore, the specific wind speed prediction model employs a support vector machine-based prediction model, the construction process of which includes: Preprocess the corrected historical wind speed data to obtain the corrected historical wind speed time series dataset; Construct a prediction model based on support vector machine (SVM) and select a radial basis kernel function; Optimize the penalty factor and kernel width in the model parameters; Input time-series data of multiple meteorological variables such as temperature, air pressure, and wind direction, as well as corrected historical wind speed data, and train the wind speed prediction model. The trained wind speed prediction model is used to predict future wind speeds.
[0014] Furthermore, in step S1, both the UAV wind measurement equipment and the wind turbine are equipped with three-dimensional ultrasonic anemometers to acquire three-dimensional wind field data, including axial wind speed components, lateral wind speed components, and vertical wind speed components.
[0015] This application also provides a wind power prediction system for implementing the above method, characterized in that the prediction system includes: A drone equipped with a wind measurement device is used to collect the actual wind speed time series data in front of the wind turbine in step S1; The data acquisition module is used to collect historical operating data in step S2 and actual wind speed time series data in front of the wind turbine in step S1. The quantum-classical hybrid processing platform is used to establish the mapping model from the wind speed behind the wind turbine to the wind speed in front of the wind turbine in step S3 and the wind power prediction model in step S5. The quantum-classical hybrid processing platform is also used to perform the historical wind speed dataset correction in step S4 and the real-time wind power prediction in steps S6 and S7.
[0016] In summary, compared with existing technologies, the technical solutions conceived in this application achieve the following beneficial effects: By establishing a mapping model from the wind speed behind the wind turbine to the wind speed in front of the wind turbine, the wind speed measurement error caused by the wind turbine absorbing wind energy and the wake effect is effectively corrected. Furthermore, the corrected wind speed data is input into the power prediction model for retraining, making the input data of the power prediction model closer to the actual incoming wind speed and significantly improving prediction accuracy. This application combines UAV wind measurement technology with power prediction technology, which can be used for both the establishment of the initial mapping model and the periodic calibration and updating of the model, forming a complete closed-loop optimization system. Attached Figure Description
[0017] To more clearly illustrate the technical solutions in this embodiment or the prior art, the drawings used in the description of the embodiment or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0018] Figure 1 A flowchart illustrating a wind power prediction method based on wind speed correction provided in this application embodiment; Figure 2 A schematic diagram illustrating the construction process of the mapping model from the wind speed behind the wind turbine to the wind speed in front of the wind turbine provided in the embodiments of this application; Figure 3 A schematic diagram illustrating the process of establishing a power prediction model using an improved least squares support vector machine (LSSVM) for embodiments of this application; Figure 4 A system architecture diagram for wind power prediction based on wind speed correction provided in the embodiments of this application; Figure 5 A schematic diagram of the quantum-classical hybrid processing platform provided in the embodiments of this application. Detailed Implementation
[0019] To make the above-mentioned objectives, features, and advantages of this application more apparent and understandable, the embodiments of this application will be further described in detail below with reference to the accompanying drawings and specific implementation methods. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.
[0020] Many specific details are set forth in the following description in order to provide a full understanding of this application. However, this application may also be implemented in other ways different from those described herein. Those skilled in the art can make similar extensions without departing from the spirit of this application. Therefore, this application is not limited to the specific embodiments disclosed below.
[0021] This invention proposes a wind power prediction method and system based on wind speed correction. The invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.
[0022] This embodiment provides a wind power prediction method based on wind speed correction. The overall process is as follows: Figure 1 As shown, the specific steps include: S1 uses a drone equipped with wind measurement equipment to measure the actual wind speed time sequence data in front of the wind turbine rotor.
[0023] The wind measurement equipment carried by the drone is a hovering multi-rotor unmanned aerial vehicle, which is equipped with one or more combinations of ultrasonic wind meters, lidar or quantum radar.
[0024] While the drone measures the wind, it also collects the wind speed behind the wind turbine to align with the wind measurement data from the drone.
[0025] In practical terms, a multi-rotor drone equipped with an ultrasonic anemometer can be used to measure wind speed in front of the wind turbine rotor. During wind turbine operation, the drone is controlled to hover at the same height in front of the rotor and continuously measure the wind speed in front of the rotor, collecting data at a frequency of one data point every minute, hour, or day. The wind speed in front of the rotor measured by the drone is synchronized with the wind speed behind the rotor recorded by the wind turbine to form a one-to-one data pair.
[0026] In step S1, both the UAV wind measurement equipment and the wind turbine are equipped with three-dimensional ultrasonic anemometers to acquire three-dimensional wind field data, including axial wind speed components, lateral wind speed components and vertical wind speed components. The measurement height range is from 40 meters to 200 meters, which is the operating height range of the wind turbine.
[0027] S2, collect historical operating data of the wind turbine, including: wind speed time series data measured by the wind turbine anemometer, wind turbine state parameter time series data, and actual output power time series data of the wind turbine, and synchronize and align it with the data collected in step S1 in time.
[0028] The wind speed time series data measured by the anemometer behind the wind turbine is provided, with a time resolution of 1 minute, 1 hour, or 1 day depending on the required type (ultra-short-term, short-term, medium-term); wind turbine status parameters include blade speed, pitch angle, yaw angle, and turbulence intensity. Time series data and wind speed data are presented as time points, or blade speed data as time points, or actual power data as time points.
[0029] For example, the actual wind speed time series data in front of the wind turbine is continuously measured within 24 hours, and the wind speed time series data behind the wind turbine, the wind turbine's state parameter time series data, and the wind turbine's actual output power time series data are collected from the wind turbine's anemometer within these 24 hours as the historical operating data of the wind turbine referred to in step S2.
[0030] S3. Based on the synchronous data obtained in steps S1 and S2, establish a mapping model from the wind speed behind the wind turbine to the wind speed in front of the wind turbine. The input of the mapping model includes the wind speed behind the wind turbine and the state parameters of the wind turbine unit, and the output is the equivalent wind speed in front of the wind turbine.
[0031] In some embodiments of this application, this step uses a BP neural network optimized by a quantum genetic algorithm to establish a mapping model from the wind speed behind the wind turbine to the wind speed in front of the wind turbine, as follows: Figure 2 As shown.
[0032] First, the structure of the BP neural network was determined. Considering the influence of the rotating wind turbine on wind speed, the wind speed behind the turbine, blade rotation speed, and propeller pitch angle were used as inputs, and the wind speed in front of the turbine, measured by an ultrasonic anemometer mounted on a multi-rotor UAV, was used as the output. A three-layer network structure was adopted, including an input layer, one hidden layer, and an output layer. A neural network with one hidden layer can fit a continuous function within a closed interval. The number of neurons in the hidden layer was determined through multiple experiments to achieve optimal network convergence.
[0033] Secondly, the initial weights and thresholds of the BP neural network are optimized using a quantum genetic algorithm (GQA). The genetic algorithm employs real-number encoding, sequentially encoding the weights between the input and hidden layers, the hidden and output layers, and the thresholds of the hidden and output layer nodes to form an individual in the population. The encoding length is calculated as: L = NP + PM + P + M, where N is the number of input nodes, P is the number of hidden nodes, and M is the number of output nodes.
[0034] The main steps of a genetic algorithm are as follows: (1) A set of initial individuals is randomly generated to form an initial population. Each individual represents the initial solution to the problem in a specific encoding form, and each individual is evaluated by a goodness-of-fit value.
[0035] (2) If the convergence requirement of the algorithm is met, output the search result; otherwise, proceed to the next step.
[0036] (3) Perform genetic operations based on fitness levels.
[0037] (4) Return to step (2).
[0038] In a quantum computer, the state of a qubit can be represented as: In the formula, and Representing states respectively and state The amplitude constant satisfies .in, , These represent the probabilities of the observed quantum state value being 0 and 1, respectively.
[0039] In quantum genetic algorithm (GQA), the chromosome structure of quantum genetic algorithm using qubit amplitude encoding expands the search space. Mutation and updating are performed through quantum rotation gates, which is more efficient than the crossover and mutation operations of traditional genetic algorithms and has a significantly faster convergence speed than standard genetic algorithms.
[0040] In some embodiments of this application, a quantum-classical hybrid processing platform is used to predict wind power. In the process of optimizing the BP neural network using the quantum genetic algorithm (GQA), the quantum genetic algorithm (GQA) process can be performed using a quantum computing module, while the BP neural network is run in the classical computing module, thereby enabling the quantum computing module and the classical computing module to work together.
[0041] The optimal solution obtained from the genetic algorithm is substituted into the backpropagation neural network for training. 80% of the data is randomly selected as training samples, and the remaining 20% is used as test samples. Finally, a GA-BP mapping model of wind speed from behind the wind turbine to in front of the wind turbine is obtained.
[0042] The establishment of the mapping model in step S3 is triggered by any of the following conditions: Periodic triggering, performing periodic calibration at preset time intervals; Triggered by wind direction change: Calibration is triggered when the average wind direction change in the wind field exceeds a preset threshold. The calibration is triggered when meteorological conditions change significantly. Meteorological conditions include one or more of the following: average temperature, rainfall, air pressure, and atmospheric particulate matter concentration.
[0043] To ensure the mapping model can adapt to changes in wind field conditions and maintain the effectiveness of calibration, an adaptive calibration method triggered by multiple conditions is established. Regular triggering can be set to automatically calibrate at 3:00 AM daily, with a full recalibration performed on the 1st of each month. During operation triggered by wind direction changes, the average wind direction change is monitored in real time; calibration is triggered when the average wind direction change exceeds 15° within 10 consecutive minutes. The wind direction change threshold is set based on the included angle of the wind turbine hub axis to ensure that the mapping relationship is updated promptly when the degree of influence of the wind turbine on the anemometer behind the rotor changes.
[0044] The specific thresholds for meteorological conditions can be set with reference to the following empirical standards: average temperature change exceeding 5℃; rainfall exceeding 10mm / hour; air pressure change exceeding 10hPa; atmospheric particulate matter concentration (PM2.5 / PM10) change exceeding 50%.
[0045] The calibration process is as follows: A drone is launched and positioned in front of the wind turbine rotor to measure wind speed data, simultaneously collecting wind speed and state parameter data behind the rotor. The newly collected data is then merged with historical data to form an updated training set. The existing GA-BP mapping model is fine-tuned, retaining existing model knowledge while adapting to new wind conditions. The accuracy of the calibrated model is verified; if a significant improvement is observed, the online model is updated; otherwise, the original model is retained. Through a multi-condition triggered adaptive calibration mechanism, this method can dynamically adapt to changes in the wind farm environment, maintaining the long-term effectiveness of the mapping model and ensuring consistently high accuracy in wind power prediction.
[0046] S4. Convert the historical wind speed behind the wind turbine in step S2 into the historical equivalent wind speed in front of the wind turbine through the mapping model established in step S3, forming a corrected historical wind speed dataset.
[0047] The historically collected wind speed data behind the wind turbine is converted into an equivalent wind speed in front of the wind turbine using the mapping model established in step S3, thus forming a corrected historical wind speed dataset.
[0048] S5. Based on the corrected historical wind speed dataset obtained in step S4, the wind turbine state parameters and actual output power collected in step S2, a calibrated wind power prediction model is trained.
[0049] In some embodiments of this application, this step uses an improved least squares support vector machine (LSSVM) to establish a power prediction model, as follows: Figure 3 As shown.
[0050] This step uses an improved least squares support vector machine (LSSVM) to build a power prediction model, and the process is as follows: Figure 3As shown.
[0051] First, a training sample set is constructed. The inputs for each sample include: the corrected equivalent wind speed in front of the rotor, the original wind speed behind the rotor, the blade rotation speed, the pitch angle, the yaw angle, the turbulence intensity, and the time characteristics (time of day, season); the output is the actual output power.
[0052] Secondly, Least Squares Support Vector Machine (LSSVM) is used for modeling. LSSVM replaces the inequality constraints in traditional SVM with equality constraints, obtaining the prediction model by solving a system of linear equations, thus reducing space complexity and improving computational speed. The LSSVM model can be represented as: Let C be the weight vector and C be the penalty factor. It is a relaxation factor.
[0053] Introducing the Lagrange multipliers, we finally obtain the regression function: For the prediction function, b are constants, and the kernel function uses Gaussian radial basis functions. , These are training samples.
[0054] Next, the regularization parameter C and kernel function parameter σ of LSSVM are optimized using the quantum particle swarm optimization (QPSO) algorithm. After optimization, the optimal parameter combination is obtained.
[0055] Finally, the calibrated power prediction model was trained using all the training data.
[0056] The calibrated wind power prediction model in step S5 uses mixed parameter input, including the corrected historical wind speed dataset, the wind speed behind the wind turbine, the wind turbine state parameters, and the actual output power. When the data in front of the wind turbine is occasionally missing, the time series data at that point only uses the wind speed behind the wind turbine.
[0057] S6. In the real-time prediction stage, the wind speed behind the wind turbine and the state parameters of the wind turbine are collected in real time and input into the mapping model established in step S3 to obtain the equivalent wind speed in front of the wind turbine at the current moment.
[0058] S7. Input the current equivalent wind speed in front of the wind turbine, the corrected historical wind speed dataset, the wind turbine state parameters and actual output power collected in step S2, and the real-time collected wind turbine state parameters into the calibrated wind power prediction model trained in step S5, and output the predicted wind power value for future times.
[0059] In some other embodiments of this application, in addition to inputting the current equivalent wind speed in front of the wind turbine obtained in step S6 and the real-time collected wind turbine state parameters into the calibrated wind power prediction model trained in step S5, meteorological forecast wind speed data is also added and input into the wind power prediction model, and the wind power prediction value for future time is output after processing.
[0060] In some other embodiments of this application, a wind speed prediction step is included before step S7: Based on historical equivalent wind turbine inlet wind speed time series data, a specific wind speed prediction model is used to predict the equivalent wind turbine inlet wind speed at future times, and this predicted value is used to replace the weather forecast wind speed and input into the calibrated wind power prediction model for power prediction.
[0061] The specific wind speed prediction model uses a support vector machine-based prediction model, and its construction process includes: Preprocessed and corrected historical wind speed data yielded corrected historical wind speed time-series data; Construct a prediction model based on support vector machine (SVM) and select radial basis function (RBF); Optimize the penalty factor and kernel width in the model parameters; Input time-series data of multiple meteorological variables such as temperature, air pressure, and wind direction, as well as corrected historical wind speed data, and train the wind speed prediction model. The trained wind speed prediction model is used to predict future wind speeds.
[0062] In the optimization process, the Quantum Positive Swarm Optimization (QPSO) algorithm can be used to iteratively optimize parameters such as the penalty factor and kernel width. The penalty factor is generally used to control the degree of punishment for misclassified samples, affecting the model's generalization ability; the kernel function parameter determines the influence range of the Gaussian kernel function, affecting the model's complexity. The optimization process needs to maximize cross-validation accuracy while avoiding overfitting. QPSO introduces quantum mechanical principles into traditional particle swarm optimization, where particle positions are described by wave functions rather than traditional velocity and position; particles are in a quantum bound state, oscillating around a local attractor. The algorithm includes a group of "quantum particles," each representing a set of possible parameter combinations. The entire algorithm evolution process is also a process of evaluating and iterating the fitness of particles. After the iteration is complete, the parameter combination corresponding to the globally optimal position is the optimal parameter we are looking for.
[0063] Compared to the traditional particle swarm optimization algorithm, the quantum swarm optimization (QPSO) algorithm requires fewer parameters to be adjusted, is easier to use, has stronger global search capabilities, converges faster, and can efficiently find the optimal parameter combination for the support vector machine, thereby significantly improving the model's predictive performance and generalization ability.
[0064] In this embodiment, the Quantum Profit Group Optimization (QPSO) algorithm is used to iteratively optimize parameters such as the penalty factor and kernel width in the support vector machine model. The QPSO algorithm process can be performed using a quantum computing module, while the support vector machine model itself is run in a classical computing module. Thus, by running the QPSO algorithm, the quantum computing module accelerates the entire support vector machine model operation process, thereby improving the overall efficiency of the wind speed prediction model.
[0065] In another embodiment of this application, a three-dimensional wind field enhancement model for wind speed prediction is introduced. The wind speed data in front of the wind turbine includes three dimensions: axial wind speed component, lateral wind speed component, and vertical wind speed component. In step S1, both the UAV wind measurement device and the wind turbine are equipped with three-dimensional ultrasonic anemometers to acquire three-dimensional wind field data, including the axial wind speed component (X direction, i.e., wind direction), the lateral wind speed component (Y direction, horizontal direction perpendicular to the wind direction), and the vertical wind speed component (Z direction, vertical direction). The measurement height range is from 40 meters to 200 meters, within the wind turbine's operating height range. The data acquisition frequency is 1 Hz (i.e., once per second), forming a three-dimensional spatiotemporal grid data, with dimensions including longitude, latitude, altitude, and time.
[0066] In this embodiment, the input layer of the BP neural network receives a three-dimensional wind field spatiotemporal tensor with dimensions of (number of longitude grids × number of latitude grids × number of altitude layers × time step), as well as basic parameters such as wind speed behind the wind turbine, blade rotation speed, and blade pitch angle. The output layer outputs the predicted power value for future moments. It has only one neuron and uses a linear activation function. When the wind direction changes drastically, traditional models often fail (because the training data does not cover the complex flow patterns under this specific wind direction). However, this model can reasonably estimate the wind speed distribution under this wind direction based on terrain features and fluid dynamics principles, maintaining high prediction accuracy.
[0067] This application also provides a wind power prediction system for implementing the wind power prediction method based on wind speed correction proposed in the above embodiments. The wind power prediction system includes: A drone equipped with a wind measurement device is used to collect the actual wind speed time series data in front of the wind turbine in step S1; Data acquisition module, see reference Figure 4 This is used to collect historical operating data in step S2 and actual wind speed time series data in front of the wind turbine in step S1. like Figure 4 As shown, a drone equipped with wind measurement equipment hovers in the air to monitor wind speed in front of the wind turbine and transmits data in real time to a ground-based data acquisition module via wireless communication. The information collected and stored in the data acquisition module is interconnected with a quantum-classical hybrid processing platform.
[0068] Quantum-classical hybrid processing platforms, such as Figure 5 As shown, the quantum-classical hybrid processing platform is used to establish the mapping model from the wind speed behind the wind turbine to the wind speed in front of the wind turbine in step S3 and the wind power prediction model in step S5. It is also used to perform the historical wind speed dataset correction in step S4 and the real-time wind power prediction in steps S6 and S7.
[0069] The quantum-classical hybrid processing platform includes the aforementioned quantum computing module and classical computing module. The quantum computing module is used to run the quantum genetic algorithm (GQA) in step S3 and the quantum particle swarm optimization (QPSO) algorithm in step S5. Furthermore, a wind speed prediction step is included before step S7, and the QPSO algorithm in the wind speed prediction model can also be performed in the quantum computing module. Other steps, such as the training and running of the BP neural network model and the support vector machine model, are all performed in the classical computing module. The quantum algorithm (quantum computing process) acts as a heterogeneous acceleration unit to accelerate specific steps of the wind speed prediction model and the wind power prediction model. The quantum and classical algorithms work together to improve the computational efficiency and accuracy of wind power prediction.
[0070] This invention provides a wind power prediction method and system based on wind speed correction, which can be widely applied to various types of wind farms, especially large grid-connected wind farms with high prediction accuracy requirements. By improving the accuracy of wind power prediction, it helps grid dispatching departments to more accurately arrange power generation plans, reduce spinning reserve capacity, lower power system operating costs, and increase the wind power penetration limit. Simultaneously, this invention can also be used in wind turbine performance evaluation, wind resource assessment, and wind farm optimized operation, possessing broad industrial application prospects and significant economic and social benefits.
[0071] The various embodiments in this specification are described in a progressive, parallel, or combined manner. Each embodiment focuses on the differences from other embodiments, and the same or similar parts between the embodiments can be referred to each other.
[0072] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that an article or apparatus comprising a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such an article or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the article or apparatus that includes the aforementioned element.
[0073] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
Claims
1. A wind power prediction method based on wind speed correction, characterized in that, Includes the following steps: S1 uses a drone equipped with wind measurement equipment to measure the actual wind speed time series data in front of the wind turbine rotor; S2, collect historical operating data of the wind turbine at the same time as step S1, including wind speed time series data measured by the wind turbine anemometer, wind turbine state parameter time series data, and actual output power time series data of the wind turbine, and synchronize and align with the data collected in step S1 in time. S3. Based on the synchronous data obtained in steps S1 and S2, establish a mapping model from the wind speed behind the wind turbine to the wind speed in front of the wind turbine. The input of the mapping model includes the wind speed behind the wind turbine and the state parameters of the wind turbine, and the output is the equivalent wind speed in front of the wind turbine. S4. Convert the historical wind speed behind the wind turbine in step S2 into the historical equivalent wind speed in front of the wind turbine through the mapping model established in step S3, and form a corrected historical wind speed dataset. S5. Based on the corrected historical wind speed dataset obtained in step S4, the wind turbine state parameters and actual output power collected in step S2, a calibrated wind power prediction model is trained. S6. In the real-time prediction stage, the wind speed behind the wind turbine and the state parameters of the wind turbine are collected in real time and input into the mapping model established in step S3 to obtain the equivalent wind speed in front of the wind turbine at the current moment. S7. Input the current equivalent wind speed in front of the wind turbine, the corrected historical wind speed dataset, the wind turbine state parameters and actual output power collected in step S2, and the real-time collected wind turbine state parameters into the calibrated wind power prediction model trained in step S5, and output the predicted wind power value for future times.
2. The wind power prediction method based on wind speed correction according to claim 1, characterized in that, In step S1, the wind measurement equipment carried by the UAV is a hovering multi-rotor UAV, which is equipped with one or more combinations of ultrasonic wind measuring instruments, lidar or quantum radar.
3. The wind power prediction method based on wind speed correction according to claim 1, characterized in that, The mapping model in step S3 uses a BP neural network model optimized by the quantum genetic algorithm (GQA), and its construction process includes: The structure of the input layer, hidden layer and output layer of the BP neural network is determined, where the input layer nodes include the wind speed behind the wind turbine, the blade rotation speed and the pitch angle, and the output layer nodes are the wind speed in front of the wind turbine. The initial weights and thresholds of the BP neural network are optimized using a quantum genetic algorithm (GQA). The optimized initial weights and thresholds are substituted into the BP neural network for training to obtain the final mapping model.
4. The wind power prediction method based on wind speed correction according to claim 1, characterized in that, The establishment of the mapping model in step S3 is triggered by any of the following conditions: Perform periodic calibration at preset time intervals; Calibration is triggered when the average wind direction change in the wind field exceeds a preset threshold. Calibration is triggered when meteorological conditions change significantly, including one or more of the following: average temperature, rainfall, air pressure, and atmospheric particulate matter concentration.
5. The wind power prediction method based on wind speed correction according to claim 1, characterized in that, The calibrated wind power prediction model in step S5 uses mixed parameter input, including the corrected historical wind speed dataset, the wind speed behind the wind turbine, the wind turbine state parameters, and the actual output power. When the data before the wind turbine is occasionally missing, the time series data at that point only uses the wind speed behind the wind turbine.
6. The wind power prediction method based on wind speed correction according to claim 5, characterized in that, The wind power prediction model in step S5 adopts an improved least squares support vector machine (LSSVM) model, and its construction process includes: The inequality constraints in the traditional support vector machine are replaced with equality constraints, and the prediction model is obtained by solving a system of linear equations. The regularization parameter C and kernel function parameter σ of the least squares support vector machine are optimized using the quantum particle swarm optimization (QPSO) algorithm.
7. The wind power prediction method based on wind speed correction according to claim 1, characterized in that, The wind speed prediction step is included before step S7: Based on historical equivalent wind turbine inlet wind speed time series data, a specific wind speed prediction model is used to predict the equivalent wind turbine inlet wind speed at future times, and the predicted value is input into the calibrated wind power prediction model for power prediction.
8. The wind power prediction method based on wind speed correction according to claim 7, characterized in that, The specific wind speed prediction model adopts a support vector machine-based prediction model, and its construction process includes: Preprocess and correct the historical wind speed data to obtain the corrected historical wind speed time series data; Construct a prediction model based on support vector machine (SVM) and select radial basis function (RBF); Optimize the penalty factor and kernel width in the model parameters; Input time-series data of multiple meteorological variables such as temperature, air pressure, and wind direction, as well as corrected historical wind speed data, and train the wind speed prediction model. The trained wind speed prediction model is used to predict future wind speeds.
9. The wind power prediction method based on wind speed correction according to claim 1, characterized in that, In step S1, both the UAV wind measurement equipment and the wind turbine are equipped with three-dimensional ultrasonic anemometers to acquire three-dimensional wind field data, including axial wind speed components, lateral wind speed components, and vertical wind speed components.
10. A wind power prediction system for implementing the method as described in any one of claims 1-7, characterized in that, include: A drone equipped with a wind measurement device is used to collect the actual wind speed time series data in front of the wind turbine in step S1; The data acquisition module is used to collect historical operating data in step S2 and actual wind speed time series data in front of the wind turbine in step S1. The quantum-classical hybrid processing platform is used to establish the mapping model from the wind speed behind the wind turbine to the wind speed in front of the wind turbine in step S3 and the wind power prediction model in step S5. The quantum-classical hybrid processing platform is also used to perform the historical wind speed dataset correction in step S4 and the real-time wind power prediction in steps S6 and S7.