Wind power climbing event prediction method and device, storage medium and electronic equipment
By introducing diffusion and XGBoost models into the prediction of wind power ramping events, the numerical weather prediction data is refined to correct errors, thus solving the oversmoothing problem of numerical weather prediction and improving the prediction accuracy and key feature recognition capability of wind power ramping events.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- STATE GRID LIAONING ELECTRIC POWER CO LTD
- Filing Date
- 2025-11-27
- Publication Date
- 2026-05-01
AI Technical Summary
Existing wind power ramping event prediction methods rely on numerical weather prediction, which suffers from oversmoothing and makes it difficult to accurately depict sudden changes in wind speed, resulting in insufficient prediction accuracy.
By introducing a pre-defined diffusion model to refine the error correction of numerical weather forecast data, and combining it with the XGBoost model, the accuracy and timeliness of meteorological input data are improved, and the complex mapping relationship between meteorological variables and wind power ramping events is explored.
It improves the prediction accuracy of wind power ramping events, especially under extreme weather conditions, and can more accurately identify key characteristics such as the start and end times of ramping and the rate of change.
Smart Images
Figure CN121965474A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of wind power technology, and in particular to a method, apparatus, storage medium and electronic equipment for predicting wind power ramping events. Background Technology
[0002] In recent years, the proportion of wind power's cumulative installed capacity in my country has gradually increased. However, wind power output exhibits significant volatility and randomness, especially during rapid wind speed changes, which can easily lead to pronounced uphill or downhill events, causing drastic fluctuations in wind power output within a short period. Such power fluctuations directly affect the active power balance of the power grid, causing grid frequency shifts, and in severe cases, potentially threatening the safe and stable operation of the power system. Against this backdrop, accurately predicting wind power uphill events is of great significance for the safe and stable operation of new power systems.
[0003] Currently, existing methods for predicting wind power ramping events typically use numerical weather prediction (NWP) as the primary input data. However, wind power ramping events usually occur during periods of rapid weather change, and numerical weather prediction generally suffers from "oversmoothing," making it difficult to accurately characterize the rapid fluctuations of key meteorological elements. This leads to significant deviations in the simulation and forecasting of extreme or abrupt weather processes, thus affecting the accuracy of wind power ramping event predictions. Summary of the Invention
[0004] In view of this, this application provides a method, apparatus, storage medium and electronic device for predicting wind power ramping events, which mainly enables fine correction of numerical weather forecast errors, thereby improving the prediction accuracy of wind power ramping events.
[0005] According to a first aspect of this application, a method for predicting wind power ramping events is provided, the method comprising: Obtain raw numerical weather forecast data for wind farms within a preset future time period; The original numerical weather forecast data is input into a preset diffusion model for fine error correction to obtain the corrected numerical weather forecast data. The corrected numerical weather forecast data is input into a preset wind power ramping event prediction model for prediction, and the ramping status label sequence of the wind farm in the preset future time period is obtained. Based on the climbing status label sequence, the wind power climbing events of the wind farm are determined.
[0006] According to a second aspect of this application, a wind power ramping event prediction device is provided, the device comprising: The acquisition unit is used to acquire raw numerical weather forecast data for the wind farm within a preset future time period. The correction unit is used to input the original numerical weather forecast data into a preset diffusion model for fine error correction, so as to obtain the corrected numerical weather forecast data. The prediction unit is used to input the corrected numerical weather forecast data into a preset wind power ramping event prediction model for prediction, and obtain the ramping status label sequence of the wind farm in the preset future time period. A determining unit is used to determine the wind power ramping event of the wind farm based on the ramping state label sequence.
[0007] According to a third aspect of this application, a storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the above-described wind power ramping event prediction method.
[0008] According to a fourth aspect of this application, an electronic device is provided, including a storage medium, a processor, and a computer program stored on the storage medium and executable on the processor, wherein the processor executes the program to implement the above-described wind power ramping event prediction method.
[0009] By employing the above technical solutions, this application provides a wind power ramping event prediction method, apparatus, storage medium, and electronic device. Compared with existing wind power ramping event prediction methods, this application introduces a preset diffusion model to refine the error correction of the original numerical weather forecast data. This solves the problem of excessive smoothing in numerical weather forecasts during periods of rapid weather change, making it difficult to accurately depict sudden wind speed changes. This improves the accuracy and timeliness of meteorological input data under extreme weather conditions, providing more realistic and reliable preliminary information for subsequent ramping event prediction and enhancing the prediction accuracy of wind power ramping events. Simultaneously, based on the corrected numerical weather forecast data, this application combines a wind power ramping event prediction machine learning model with strong nonlinear fitting. This fully explores the complex mapping relationship between meteorological variables and wind power ramping events, thereby improving the model's ability to identify key features such as ramping start and end times and rates of change, further enhancing the prediction accuracy of wind power ramping events.
[0010] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description
[0011] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1A flowchart illustrating a wind power ramping event prediction method provided in an embodiment of this application is shown. Figure 2 A schematic diagram of the training method for the diffusion model provided in an embodiment of this application is shown; Figure 3 A schematic diagram of the training method for the wind power ramping event prediction model provided in an embodiment of this application is shown. Figure 4 A schematic diagram of the structure of a wind power ramping event prediction device provided in an embodiment of this application is shown. Detailed Implementation
[0012] The present application will be described in detail below with reference to the accompanying drawings and embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in the embodiments of the present application can be combined with each other.
[0013] Existing manual image extraction methods are inefficient and rely too heavily on human experience. If human visual judgment is impaired, it is difficult to guarantee the accuracy of image extraction.
[0014] To address the aforementioned problems, embodiments of the present invention provide a method for predicting wind power ramping events, such as... Figure 1 As shown, the method includes: Step 10: Obtain raw numerical weather forecast data for the wind farm within a preset future time period.
[0015] The raw numerical weather forecast data includes wind speed, wind direction, and temperature. The preset time period can be set according to actual operational needs, such as obtaining raw numerical weather forecast data for the next day (24 hours).
[0016] In this embodiment of the invention, raw numerical weather prediction (NWP) data for a wind farm over a predetermined time period (e.g., 24 hours) can be obtained using numerical weather prediction (NWP) technology. Numerical weather prediction (NWP) is a method that uses mathematical models and computer technology to predict future weather conditions based on current meteorological observation data. Specifically, observational data such as temperature, humidity, air pressure, wind direction, and wind speed of the wind farm can be collected from meteorological stations, satellites, radars, etc. The collected observational data is then combined with the initial state of the numerical model to generate an initial field that more closely approximates the real atmospheric state. Next, a complex mathematical model is run using a supercomputer to solve partial differential equations describing atmospheric motion to simulate the evolution of the atmosphere. Finally, based on the simulation results, a weather forecast for a future period is generated, including meteorological elements such as temperature, precipitation, wind speed, and wind direction, which is used as the raw numerical weather prediction data.
[0017] Raw numerical weather forecast data specifically consists of time series vectors for a predetermined future time period, such as... This represents the wind speed forecast value every 15 minutes over the next 24 hours. This time series vector contains wind speed forecast values at 96 time points. Wind direction forecast values or temperature forecast values can also be represented in the same way. Finally, the raw numerical weather forecast data can be represented as a 3×96 matrix, specifically representing wind speed, wind direction and temperature every 15 minutes.
[0018] It should be noted that the future preset time period in the embodiments of the present invention is not limited to 24 hours, and the time interval is not limited to 15 minutes. Both can be adjusted according to actual business needs, and the embodiments of the present invention do not make specific limitations in this regard.
[0019] Step 20: Input the original numerical weather forecast data into the preset diffusion model for fine error correction to obtain the corrected numerical weather forecast data.
[0020] In this embodiment of the invention, the original numerical weather forecast data is used as the initial input, and then starting from time step T, it gradually returns to time step 0. Specifically, at each time step... t The pre-trained diffusion model is used to predict the noise in the original numerical weather forecast data. Based on the predicted noise and the parameters of the pre-trained diffusion model, the mean and variance are calculated. The data is sampled from the Gaussian distribution to obtain the next step of the data state. After T iterations, the final generated data is the corrected numerical weather forecast data output by the pre-trained diffusion model.
[0021] The corrected numerical weather forecast data is also a time series vector. For example, the corrected numerical weather forecast data has a dimension of 3×96, which represents the corrected wind speed forecast, corrected wind direction forecast, and corrected temperature forecast every 15 minutes in the next 24 hours.
[0022] This invention, through the introduction of a preset diffusion model for gradual denoising, refines the error correction of the original numerical weather forecast data. This addresses the problem of excessive smoothing in numerical weather forecasts during periods of rapid weather change, making it difficult to accurately depict sudden wind speed changes. Consequently, it improves the accuracy and timeliness of meteorological input data under extreme weather conditions, providing more realistic and reliable preliminary information for subsequent hill-climbing event prediction and enhancing the prediction accuracy of wind power hill-climbing events.
[0023] Step 30: Input the corrected numerical weather forecast data into the preset wind power ramping event prediction model for prediction, and obtain the ramping status label sequence of the wind power plant in the preset future time period.
[0024] Specifically, the preset wind power ramping event prediction model can be an extreme gradient boosting tree (XGBoost model). The ramping state label sequence includes ramping state labels corresponding to different time points within a preset future time period, and the ramping state labels include ramping up, ramping down, and no ramping.
[0025] For example, after obtaining the corrected numerical weather forecast data, it is organized into a 3×96 feature matrix and then input into the trained XGBoost model. The model outputs the ramp status label of the wind farm every 15 minutes in the next 24 hours, thereby realizing early warning of future wind power ramp events.
[0026] Step 40: Based on the ramp status label sequence, determine the wind power ramp event of the wind farm.
[0027] Among them, wind power ramping events include uphill ramping events, downhill ramping events, and no ramping events.
[0028] In the embodiments of the present invention, based on the climbing status label sequence output by the preset wind power climbing event prediction model, it is possible to determine the wind power climbing events that may occur in the wind power plant within a preset time period in the future.
[0029] Based on the corrected high-precision meteorological data, this invention combines machine learning models with strong nonlinear fitting capabilities, such as XGBoost, to fully explore the complex mapping relationship between meteorological variables and wind power output ramping, thereby improving the model's ability to identify key features such as ramping start and end times and rate of change.
[0030] Furthermore, before performing refined error correction on the original numerical weather forecast data, this embodiment of the invention needs to train a preset diffusion model. The specific training process for the preset diffusion model is as follows: Figure 2 As shown, it includes: Step 50: Obtain numerical weather forecast data samples and measured meteorological data of the wind farm within a preset historical period.
[0031] The preset historical time period can be set according to actual business needs, and this embodiment of the invention does not impose specific limitations on it. Numerical weather forecast data samples specifically include wind speed data samples, wind direction data samples, and temperature data samples, etc., and the numerical weather forecast data samples are historical forecast data. Measured meteorological data includes measured wind direction, measured wind speed, and measured temperature, etc.
[0032] For example, obtaining a sample of wind speed data from the past year. Wind direction data sample and temperature data samples and measured wind direction Measured wind speed and measured temperature The data was collected every 15 minutes. The numerical weather prediction data samples and observed meteorological data were divided into daily units, with each numerical weather prediction data sample forming a time series vector. This represents a wind speed data sample, wind direction data sample, or temperature data sample taken every 15 minutes within a 24-hour period. Each input sample is ultimately integrated into a 3×96 matrix, which serves as the initial state for the diffusion model, enabling the forward diffusion process.
[0033] Step 60: Based on a preset number of time steps, Gaussian noise is added to the numerical weather forecast data samples step by step to obtain the noise sample corresponding to each time step.
[0034] The preset time steps can be set according to actual business needs, such as setting the time steps to 500. It should be noted that the embodiments of the present invention do not impose a specific limitation on the preset time steps.
[0035] In this embodiment of the invention, during the forward process, Gaussian noise is recursively added to the original data (numerical weather forecast data samples) over 500 time steps, causing the data distribution to gradually approach a standard normal distribution. Specifically, in the first... t step( t When the input noise is denoted as 1, 2, ..., 500, it is described by the following conditional probability distribution:
[0036] in, The preset noise variance scheduling coefficient controls the first... The noise intensity of the step, Indicates a Gaussian distribution. Let be the unit covariance matrix. intermediate state after adding noise The following was obtained by sampling from the above conditional Gaussian distribution:
[0037] Through the aforementioned forward diffusion process, numerical weather prediction data samples... After 500 steps of noise addition, it is converted into a pure noise sample. This provides a training basis for the subsequent reverse denoising process.
[0038] Step 70: Construct an initial diffusion model. Input the noise sample and its corresponding time step into the initial diffusion model for reverse denoising and predict the noise removal corresponding to each time step.
[0039] The initial diffusion model is a deep neural network.
[0040] In this embodiment of the invention, a reverse denoising process is constructed, which is implemented by a deep neural network. The network starts from an intermediate state after adding noise. and current time step t As input, the inverse denoising process is also assumed to be 500 steps, with each step predicting the noise that should be eliminated in that step.
[0041] Step 80: Based on the noise removal, reconstruct the meteorological data of the wind farm within a preset historical period, and construct a first loss function based on the reconstructed meteorological data and the measured meteorological data.
[0042] In this embodiment of the invention, a meteorological sequence approximating the measured values is indirectly reconstructed by noise elimination estimation. The first loss function is defined as the mean square error between the predicted reconstruction result and the measured meteorological data, specifically expressed as:
[0043] in, From the reverse process The final output sequence obtained by gradually denoising and restoring, i.e., the reconstructed meteorological data, These are actual measured meteorological data.
[0044] Step 90: Based on the first loss function, iteratively train the initial diffusion model to construct the preset diffusion model.
[0045] In this embodiment of the invention, a cosine noise scheduling strategy is adopted during training, with a batch size of 64, an AdamW optimizer, an initial learning rate of 0.0001, and 300 training epochs. After training, when correcting new NWP data (raw numerical weather forecast data), noise is first added to it to the [missing information - likely a specific level or metric]. Step to get , and then from Starting from this point, the corrected meteorological sequence is gradually generated through 500 backward iterations. This sequence, while preserving the large-scale weather trends of NWP, restores key details such as sudden increases in wind speed and abrupt changes in wind direction that were smoothed by the original model, significantly improving the authenticity and timeliness of meteorological input data, and providing high-quality data support for the accurate prediction of subsequent wind power ramp-up events.
[0046] In some embodiments, in order to predict ramp events, it is necessary to train a preset wind power ramp event prediction model. For this training process, such as... Figure 3 As shown, it includes: Step 100: Obtain numerical weather forecast data samples and power data samples of the wind farm within a preset historical period.
[0047] In this embodiment of the invention, the process of obtaining numerical weather prediction data samples is exactly the same as step 50, and will not be described again here. For power data samples, such as obtaining power data samples from the past year... P The power data samples were collected every 15 minutes. The power data samples were divided into daily units, and each power data sample was a time series vector, representing the power at 15 minutes every 24 hours, for a total of 96 time points.
[0048] Step 110: Based on the power data samples and the sliding window size, determine the power sequence under each sliding window.
[0049] The size of the sliding window can be set according to actual business needs, and this embodiment of the invention does not impose specific limitations on it.
[0050] For example, a sliding window of size 5 time points (corresponding to 1 hour) is defined, and the power data sample (power sequence) is traversed by moving 1 time point (i.e., 15 minutes) at a time to obtain the power sequence within each sliding window. .
[0051] Step 120: Generate a historical ramp status label sequence of the same length as the power data sample based on the power sequence under each sliding window.
[0052] The historical ramp status label sequence contains historical ramp status labels corresponding to power at multiple time points. The historical ramp status labels include ramping up, ramping down, and no ramping.
[0053] This invention provides a method that combines a time sliding window with a multi-level threshold criterion to systematically analyze historical wind power data. This method introduces "fluctuation events" as intermediate identification units, combines the overall fluctuation intensity and local change frequency as dual standards, and integrates the ramping process within a continuous time period to finally generate a refined ramping status label for each time point.
[0054] In this embodiment of the invention, a historical ramp-up state label sequence of equal length to each power data sample (e.g., 96 time points) is generated based on the power sequence under each sliding window. Specifically, step 120 of generating the historical ramp-up state label sequence includes: calculating the difference between the maximum and minimum power in the power sequence under each sliding window; determining the number of fluctuation events corresponding to each sliding window based on the power sequence under each sliding window; identifying ramp-up events under each sliding window based on the difference between the maximum and minimum power and the number of fluctuation events, obtaining a ramp-up event identification result corresponding to each sliding window; if, based on the ramp-up event identification result corresponding to each sliding window, it is determined that two or more consecutive sliding windows contain ramp-up events with the same power change direction, then the ramp-up events corresponding to the two or more sliding windows are merged to obtain a ramp-up event set; and determining a historical ramp-up state label sequence of equal length to the power data sample based on the ramp-up event set.
[0055] When determining the number of fluctuation events corresponding to each sliding window, the power change rate is calculated based on the power of any two adjacent time points in the power sequence under each sliding window and the rated maximum power of the wind farm; based on the power change rate, it is determined whether a fluctuation event has occurred between any two adjacent time points; based on the fluctuation event determination result between any two adjacent time points, the number of fluctuation events corresponding to each sliding window is determined.
[0056] When identifying the ramping event under each sliding window, for any sliding window, if the difference between the maximum power and the minimum power corresponding to the sliding window is greater than or equal to a preset power, and the number of fluctuation events corresponding to the sliding window reaches a preset number, then it is determined that there is a ramping event in the sliding window.
[0057] The preset power and preset quantity can be set according to actual business needs, and the embodiments of the present invention do not impose specific limitations on them.
[0058] Specifically, in identifying the climbing events under each sliding window, the embodiments of the present invention need to consider two factors: the overall fluctuation intensity and the local change frequency. The overall fluctuation intensity can be determined by the difference between the maximum and minimum power under each sliding window, and the local change frequency can be determined by the number of fluctuation events under each sliding window.
[0059] If both the overall fluctuation intensity and the local frequency change simultaneously meet the following requirements, then a ramping event is determined to exist within the sliding window. Assume the preset power is the rated maximum power. P maxThe condition for overall fluctuation intensity is that the difference between the maximum and minimum power within the sliding window must be greater than or equal to 20% of the maximum rated power. The specific expression is as follows:
[0060] Meanwhile, assuming a preset quantity of 3, the condition to be met for local frequency changes is that at least 3 out of the 4 15-minute time intervals contained in the sliding window experience fluctuation events.
[0061] Regarding the specific process of fluctuation event detection, for the power sequence under each sliding window, a basic analysis unit is formed by two adjacent time points. For each basic analysis unit, its power change rate is calculated, and the specific calculation formula is as follows:
[0062] in, and The first and the Power values at each time point The rate of change of power, This represents the wind farm's rated maximum power. Assuming a preset power variation rate of 5%, if... If the value is ≥5%, then a "fluctuation event" is determined to have occurred within that 15-minute interval. This step can be used to identify localized, drastic changes in the power sequence, serving as a foundational feature for subsequent hill-climbing event identification.
[0063] The embodiments of the present invention effectively eliminate misjudgments caused by single abrupt changes or slow trends by using the joint constraint of dual criteria, thereby ensuring that the identified climbing events have significant power change amplitudes and continuous dynamic activity.
[0064] Since the sliding window moves only one time point at a time, there is time overlap between adjacent windows. If two or more consecutive sliding windows are determined to be climbing events and their power changes in the same direction (e.g., both are power increases or power decreases), then the climbing events corresponding to these sliding windows are merged into a continuous climbing process, forming a complete climbing start and end time period.
[0065] Based on the merged set of ramp events, ramp status is labeled for each time point in the power sequence within the sliding window, generating a historical ramp status label sequence of the same length as the power data sample. The label status includes: ramping up, ramping down, and no ramping. Ramping up corresponds to time points during a period of significant power increase, ramping down corresponds to time points during a period of significant power decrease, and no ramping does not correspond to time points of any ramp event.
[0066] Step 130: Based on the numerical weather forecast data sample and the preset diffusion model, determine the corrected numerical weather forecast data sample.
[0067] In this embodiment of the invention, numerical weather forecast data samples are input into a preset diffusion model for fine error correction to obtain corrected numerical weather forecast data samples.
[0068] Step 140: Construct the preset wind power ramping event prediction model based on the corrected numerical weather forecast data sample and the historical ramping state label sequence.
[0069] Specifically, the preset wind power ramping event prediction model can be an extreme gradient boosting tree (XGBoost model).
[0070] In this embodiment of the invention, a wind power ramping event prediction model based on XGBoost is constructed. This model takes refined numerical weather forecast data corrected by a diffusion model as input and outputs a ramping state label sequence every 15 minutes for the next 24 hours, realizing the classification and prediction of ramping, ramping, and no ramping states. For the training process of the wind power ramping event prediction model, step 140 specifically includes: constructing an extreme gradient boosting tree and inputting the corrected numerical weather forecast data samples into the extreme gradient boosting tree for classification to obtain a predicted ramping state label sequence; constructing a second loss function based on the predicted ramping state label sequence and the historical ramping state label sequence; and training the extreme gradient boosting tree according to the second loss function to construct the preset wind power ramping event prediction model.
[0071] When constructing the second loss function, a label loss function is constructed based on the predicted climbing state label sequence and the historical climbing state label sequence; a regularization term is constructed for each decision tree based on the number of leaf nodes, leaf weights, and regularization coefficients corresponding to each decision tree in the extreme gradient boosting tree; and the second loss function is constructed based on the label loss function and the regularization term corresponding to each decision tree.
[0072] Specifically, the corrected numerical weather forecast data samples and historical climbing state label sequences are used as the dataset. Each corrected numerical weather forecast data sample corresponds to a historical climbing state label sequence. Each corrected numerical weather forecast data sample serves as an input feature, specifically a numerical weather forecast data sample corrected by the diffusion model for the next 24 hours (96 time points), with a dimension of 3×96, representing corrected wind speed, corrected wind direction, and corrected temperature samples every 15 minutes. The historical climbing state label sequence, with a dimension of 1×96, includes labels for uphill, downhill, and no uphill, and is generated in step 120. The above dataset is divided into a training set and a test set in chronological order, with a ratio of 8:2.
[0073] XGBoost is an efficient decision tree ensemble algorithm based on the gradient boosting framework. Its second objective function consists of a label loss function and a regularization term, expressed as:
[0074] in, To predict the climbing state label sequence Compared with historical climbing status label sequences (real label sequences) The loss function between them is multi-class cross-entropy loss (Softmax Loss) in this embodiment of the invention. For the first The regularization term of a decision tree. The number of leaf nodes. Leaf weight, and The regularization coefficient is used. For the first Individual base learners (decision trees); This represents the total number of decision trees.
[0075] The input to the wind power ramping event prediction model is the corrected numerical weather prediction data, specifically a three-dimensional meteorological feature matrix. The output is a sequence of climbing status labels, which specifically includes the label classification results for 96 time points. Each of them .
[0076] During model training, this embodiment of the invention iteratively learns the XGBoost model based on the training set data, setting the number of decision trees to 500, the learning rate to 0.1, the maximum depth to 6, and introducing an L2 regularization term to prevent overfitting. Furthermore, the regularization parameters are optimized through cross-validation. and The system incorporates an early stop mechanism (50 early stop rounds) to terminate training when performance on the test set no longer improves, thereby determining the optimal model parameters. During training, the macro-average F1 score is used as the evaluation metric to ensure that the model's predictive performance is balanced across the three categories of "climbing," "downhill," and "no climbing," ultimately resulting in a wind power climbing event prediction model with good generalization capabilities.
[0077] This invention provides a method for predicting wind power climbing events. Compared with existing methods, this method introduces a pre-defined diffusion model to refine the error correction of the original numerical weather forecast data. This addresses the problem of over-smoothing in numerical weather forecasts during periods of rapid weather change, making it difficult to accurately depict sudden wind speed changes. This improves the accuracy and timeliness of meteorological input data under extreme weather conditions, providing more realistic and reliable preliminary information for subsequent climbing event prediction and enhancing the prediction accuracy of wind power climbing events. Furthermore, based on the corrected numerical weather forecast data, this invention combines a wind power climbing event prediction machine learning model with strong nonlinear fitting. This fully explores the complex mapping relationship between meteorological variables and wind power climbing events, improving the model's ability to identify key features such as the start and end times of climbing and the rate of change, further enhancing the prediction accuracy of wind power climbing events.
[0078] Furthermore, as Figures 1 to 3 The specific implementation of the method shown in this embodiment provides a wind power ramping event prediction device, such as... Figure 4 As shown, the system includes: an acquisition unit 101, a correction unit 102, a prediction unit 103, and a determination unit 104.
[0079] The acquisition unit 101 can be used to acquire raw numerical weather forecast data for a future preset time period of the wind farm.
[0080] The correction unit 102 can be used to input the original numerical weather forecast data into a preset diffusion model for fine error correction, so as to obtain the corrected numerical weather forecast data.
[0081] The prediction unit 103 can be used to input the corrected numerical weather forecast data into a preset wind power ramping event prediction model for prediction, and obtain the ramping status label sequence of the wind power plant within the preset future time period.
[0082] The determining unit 104 can be used to determine the wind power ramping event of the wind farm based on the ramping state label sequence.
[0083] In some embodiments, the apparatus further includes a first building unit.
[0084] The first training unit is used to acquire numerical weather forecast data samples and measured meteorological data of the wind farm within a preset historical period; based on a preset number of time steps, gradually add Gaussian noise to the numerical weather forecast data samples to obtain noise samples corresponding to each time step; construct an initial diffusion model, input the noise samples and their corresponding time steps into the initial diffusion model for reverse denoising, and predict the noise elimination corresponding to each time step; based on the noise elimination, reconstruct the meteorological data of the wind farm within the preset historical period; construct a first loss function based on the reconstructed meteorological data and the measured meteorological data; and iteratively train the initial diffusion model based on the first loss function to construct the preset diffusion model.
[0085] In some embodiments, the generation unit and the second building unit.
[0086] The acquisition unit 101 can also be used to acquire numerical weather forecast data samples and power data samples of the wind farm within a preset historical period.
[0087] The determining unit 104 can also be used to determine the power sequence under each sliding window based on the power data sample and the sliding window size.
[0088] The generation unit can be used to generate a historical ramp status label sequence of the same length as the power data sample, based on the power sequence under each sliding window.
[0089] The determining unit 104 can also be used to determine the corrected numerical weather forecast data sample based on the numerical weather forecast data sample and the preset diffusion model.
[0090] The second construction unit can be used to construct the preset wind power ramping event prediction model based on the corrected numerical weather forecast data sample and the historical ramping state label sequence.
[0091] In some embodiments, the generation unit includes: a calculation module, a determination module, an identification module, and a merging module.
[0092] The calculation module can be used to calculate the difference between the maximum power and the minimum power in the power sequence under each sliding window.
[0093] The determining module can be used to determine the number of fluctuation events corresponding to each sliding window based on the power sequence under each sliding window.
[0094] The identification module can be used to identify the climbing events under each sliding window based on the difference between the maximum power and the minimum power, as well as the number of fluctuation events, and obtain the climbing event identification result corresponding to each sliding window.
[0095] The merging module can be used to merge the climbing events corresponding to the two or more sliding windows if, based on the climbing event identification results corresponding to each sliding window, it is determined that two or more consecutive sliding windows have climbing events and the power change direction is consistent, to obtain a climbing event set.
[0096] The determining module can also be used to determine a historical climbing status label sequence of the same length as the power data sample based on the climbing event set.
[0097] In some embodiments, the determining module may be specifically used to calculate the power change rate based on the power of any two adjacent time points in the power sequence under each sliding window and the rated maximum power of the wind farm; determine whether a fluctuation event occurs between any two adjacent time points based on the power change rate; and determine the number of fluctuation events corresponding to each sliding window based on the fluctuation event determination result between any two adjacent time points.
[0098] In some embodiments, the identification module can be specifically used to determine that a ramping event exists in any sliding window if the difference between the maximum power and the minimum power corresponding to any sliding window is greater than or equal to a preset power, and the number of fluctuation events corresponding to any sliding window reaches a preset number.
[0099] In some embodiments, the second building unit includes a classification module and a building module.
[0100] The classification module can be used to construct an extreme gradient boosting tree and input the corrected numerical weather forecast data sample into the extreme gradient boosting tree for classification to obtain a predicted climbing state label sequence.
[0101] The construction module can be used to construct a second loss function based on the predicted climbing state label sequence and the historical climbing state label sequence.
[0102] The construction module can also be used to train the extreme gradient boosting tree according to the second loss function to construct the preset wind power ramping event prediction model.
[0103] In some embodiments, the construction module may be specifically used to construct a label loss function based on the predicted climbing state label sequence and the historical climbing state label sequence; construct a regularization term corresponding to each decision tree based on the number of leaf nodes, leaf weights, and regularization coefficients corresponding to each decision tree in the extreme gradient boosting tree; and construct a second loss function based on the label loss function and the regularization term corresponding to each decision tree.
[0104] It should be noted that other corresponding descriptions of the functional units involved in the wind power ramping event prediction device provided in this embodiment can be found in [reference]. Figures 1 to 3 The corresponding descriptions in [the document] will not be repeated here.
[0105] Based on the above, Figures 1 to 3 Accordingly, this embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements the above-described method. Figures 1 to 3 The wind power ramping event prediction method is shown.
[0106] Based on this understanding, the technical solution of this application can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (such as CD-ROM, USB flash drive, mobile hard drive, etc.) and includes several instructions to cause an electronic device (such as personal computer, server, or network device, etc.) to execute the methods of various implementation scenarios of this application.
[0107] Based on the above, Figures 1 to 3 The method shown, and Figure 4 To achieve the above objectives, the present application also provides an electronic device, specifically a personal computer, tablet computer, server, or other network device, as shown in the virtual device embodiment. This device includes a storage medium and a processor; the storage medium stores a computer program; the processor executes the computer program to achieve the above-described objectives. Figures 1 to 3 The wind power ramping event prediction method is shown.
[0108] Optionally, the aforementioned physical devices may also include a user interface, a network interface, a camera, radio frequency (RF) circuitry, sensors, audio circuitry, a Wi-Fi module, etc. The user interface may include a display screen, input units such as a keyboard, etc., and optional user interfaces may also include USB interfaces, card reader interfaces, etc. The network interface may optionally include standard wired interfaces, wireless interfaces (such as Wi-Fi interfaces), etc.
[0109] Those skilled in the art will understand that the physical device structure provided in this embodiment does not constitute a limitation on the physical device, and may include more or fewer components, or combine certain components, or have different component arrangements.
[0110] The storage medium may also include an operating system and a network communication module. The operating system is a program that manages the hardware and software resources of the aforementioned physical device, supporting the operation of information processing programs and other software and / or programs. The network communication module is used to enable communication between the various components within the storage medium, as well as communication with other hardware and software in the information processing physical device.
[0111] Through the above description of the embodiments, those skilled in the art can clearly understand that this application can be implemented by means of software plus necessary general-purpose hardware platform, or it can be implemented by hardware.
[0112] This invention, through the introduction of a pre-defined diffusion model to refine the error correction of raw numerical weather forecast data, addresses the problem of over-smoothing in numerical weather forecasts during periods of rapid weather change, which makes it difficult to accurately depict sudden wind speed changes. This improves the accuracy and timeliness of meteorological input data under extreme weather conditions, providing more realistic and reliable preliminary information for subsequent wind power climbing event prediction, thus enhancing the prediction accuracy of wind power climbing events. Furthermore, based on the corrected numerical weather forecast data, this invention combines a wind power climbing event prediction machine learning model with strong nonlinear fitting to fully explore the complex mapping relationship between meteorological variables and wind power climbing events. This enhances the model's ability to identify key features such as the start and end times of climbing and the rate of change, further improving the prediction accuracy of wind power climbing events.
[0113] Those skilled in the art will understand that the accompanying drawings are merely schematic diagrams of a preferred embodiment, and the modules or processes shown in the drawings are not necessarily essential for implementing this application. Those skilled in the art will understand that the modules in the apparatus of the embodiment can be distributed within the apparatus of the embodiment as described, or can be modified to be located in one or more apparatuses different from this embodiment. The modules of the above-described embodiment can be combined into one module, or further divided into multiple sub-modules.
[0114] The serial numbers in this application are for descriptive purposes only and do not represent the superiority or inferiority of any particular implementation scenario. The above disclosures are merely a few specific implementation scenarios of this application; however, this application is not limited thereto, and any variations conceived by those skilled in the art should fall within the protection scope of this application.
Claims
1. A method for predicting wind power ramping events, characterized in that, include: Obtain raw numerical weather forecast data for wind farms within a preset future time period; The original numerical weather forecast data is input into a preset diffusion model for fine error correction to obtain the corrected numerical weather forecast data. The corrected numerical weather forecast data is input into a preset wind power ramping event prediction model for prediction, and the ramping status label sequence of the wind farm in the preset future time period is obtained. Based on the climbing status label sequence, the wind power climbing events of the wind farm are determined.
2. The method according to claim 1, characterized in that, Before inputting the raw numerical weather forecast data into a preset diffusion model for refined error correction to obtain the corrected numerical weather forecast data, the method further includes: Acquire numerical weather forecast data samples and measured meteorological data of the wind farm within a preset historical period; Based on a preset number of time steps, Gaussian noise is gradually added to the numerical weather forecast data samples to obtain the noise sample corresponding to each time step. An initial diffusion model is constructed, and the noise samples and their corresponding time steps are input into the initial diffusion model for reverse denoising to predict the noise removal corresponding to each time step. Based on the noise removal, the meteorological data of the wind farm within a preset historical period are reconstructed. Based on the reconstructed meteorological data and the measured meteorological data, a first loss function is constructed; Based on the first loss function, the initial diffusion model is iteratively trained to construct the preset diffusion model.
3. The method according to claim 1, characterized in that, Before inputting the corrected numerical weather forecast data into a preset wind power ramping event prediction model for prediction to obtain the ramping state label sequence of the wind farm within the preset future time period, the method further includes: Obtain numerical weather forecast data samples and power data samples of the wind farm within a preset historical period; Based on the power data samples and the sliding window size, the power sequence under each sliding window is determined; Based on the power sequence under each sliding window, a historical ramp status label sequence of the same length as the power data sample is generated; Based on the numerical weather forecast data sample and the preset diffusion model, the corrected numerical weather forecast data sample is determined; Based on the corrected numerical weather forecast data sample and the historical ramp-up state label sequence, the preset wind power ramp-up event prediction model is constructed.
4. The method according to claim 3, characterized in that, The step of generating a historical ramp status label sequence of the same length as the power data sample based on the power sequence under each sliding window includes: Calculate the difference between the maximum and minimum power in the power sequence under each sliding window; Based on the power sequence under each sliding window, determine the number of fluctuation events corresponding to each sliding window; Based on the difference between the maximum and minimum power and the number of fluctuation events, the climbing events under each sliding window are identified, and the climbing event identification results corresponding to each sliding window are obtained. If, based on the hill-climbing event identification results corresponding to each sliding window, it is determined that two or more consecutive sliding windows have hill-climbing events and the power change direction is consistent, then the hill-climbing events corresponding to the two or more sliding windows are merged to obtain a hill-climbing event set. Based on the set of climbing events, a historical climbing status label sequence of the same length as the power data sample is determined.
5. The method according to claim 4, characterized in that, The step of determining the number of fluctuation events corresponding to each sliding window based on the power sequence under each sliding window includes: The power change rate is calculated based on the power at any two adjacent time points in the power sequence under each sliding window, and the rated maximum power of the wind farm. Based on the power change rate, determine whether a fluctuation event occurs between any two adjacent time points; Based on the fluctuation event determination results between any two adjacent time points, determine the number of fluctuation events corresponding to each sliding window; and / or The step of identifying the ramp events under each sliding window based on the difference between the maximum and minimum power and the number of fluctuation events, and obtaining the ramp event identification result corresponding to each sliding window, includes: For any sliding window, if the difference between the maximum power and the minimum power corresponding to the sliding window is greater than or equal to a preset power, and the number of fluctuation events corresponding to the sliding window reaches a preset number, then it is determined that there is a ramping event in the sliding window.
6. The method according to claim 3, characterized in that, The step of constructing the preset wind power ramping event prediction model based on the corrected numerical weather forecast data sample and the historical ramping state label sequence includes: An extreme gradient boosting tree is constructed, and the corrected numerical weather forecast data samples are input into the extreme gradient boosting tree for classification to obtain the predicted climbing state label sequence. Based on the predicted climbing state label sequence and the historical climbing state label sequence, a second loss function is constructed; The extreme gradient boosting tree is trained according to the second loss function to construct the preset wind power ramping event prediction model.
7. The method according to claim 6, characterized in that, The second loss function is constructed based on the predicted climbing state label sequence and the historical climbing state label sequence, including: Based on the predicted climbing state label sequence and the historical climbing state label sequence, a label loss function is constructed; Based on the number of leaf nodes, leaf weights, and regularization coefficients of each decision tree in the extreme gradient boosting tree, a regularization term is constructed for each decision tree. The second loss function is constructed based on the label loss function and the regularization term corresponding to each decision tree.
8. A wind power ramping event prediction device, characterized in that, include: The acquisition unit is used to acquire raw numerical weather forecast data for the wind farm within a preset future time period. The correction unit is used to input the original numerical weather forecast data into a preset diffusion model for fine error correction, so as to obtain the corrected numerical weather forecast data. The prediction unit is used to input the corrected numerical weather forecast data into a preset wind power ramping event prediction model for prediction, and obtain the ramping status label sequence of the wind farm in the preset future time period. A determining unit is used to determine the wind power ramping event of the wind farm based on the ramping state label sequence.
9. A storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1 to 7.
10. An electronic device comprising a storage medium, a processor, and a computer program stored on the storage medium and executable on the processor, characterized in that, When the processor executes the computer program, it implements the method of any one of claims 1 to 7.