Wind field prediction method and system based on big data
By defining interval weights and error sensitivity enhancement factors, the wind field prediction method is dynamically adjusted, which solves the problems of insufficient interval differences and error penalties in wind field prediction and improves the reliability and effectiveness of wind field prediction.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- 兰州中心气象台(兰州干旱生态环境监测预测中心)
- Filing Date
- 2026-03-05
- Publication Date
- 2026-06-09
AI Technical Summary
Existing wind field prediction methods cannot adapt to the wind speed distribution characteristics of different wind fields. Feature quantization is biased towards low-value ranges, and the error cost is not matched with the weight, resulting in poor reliability of wind field prediction. The error penalty for high-value ranges is insufficient, resulting in low decision-making practicality. Over-calibration or under-calibration leads to poor wind field prediction performance.
By defining interval weights, dynamically balancing the differences between different intervals, introducing an error sensitivity enhancement factor, and designing a unified loss framework, error amplification and loss enhancement are achieved, thereby improving the prediction accuracy and effectiveness of key intervals.
It improves the reliability and effectiveness of wind field forecasting, solves the problem of forecasting bias caused by interval differences in wind field forecasting, and enhances the decision-making practicality in high error intervals.
Smart Images
Figure CN122175074A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of data processing technology, specifically to a wind field prediction method and system based on big data. Background Technology
[0002] Wind field prediction methods are techniques that use multi-source information, such as historical wind field monitoring data and meteorological data, to mine wind characteristic patterns through statistical models or machine learning models, thereby predicting key parameters such as wind speed and direction within a wind field area at a specific future time scale. However, general wind field prediction methods suffer from several drawbacks: they cannot adapt to the wind speed distribution characteristics of different wind fields, feature quantization is biased towards low-value ranges, and the error cost is mismatched with the weight, leading to poor reliability in wind field predictions. Furthermore, general wind field prediction methods suffer from insufficient penalty for errors in high-value ranges, low decision-making practicality, and poor prediction performance due to over-calibration or under-calibration. Summary of the Invention
[0003] To address the aforementioned issues and overcome the shortcomings of existing technologies, this invention provides a wind field prediction method and system based on big data. Addressing the problems of general wind field prediction methods failing to adapt to the wind speed distribution characteristics of different wind fields, feature quantization biasing towards low-value intervals, and mismatch between error costs and weights, leading to poor prediction reliability, this solution defines interval weights to dynamically balance the differences between different intervals, avoiding the dominance effect of multiple sample intervals; defines an interval prediction error sensitivity index and designs an error sensitivity enhancement factor to convert loss differences into weight enhancement coefficients; and proactively tilts the feature quantization process towards intervals with high error costs, strengthening the feature representation of core intervals and improving the prediction accuracy of key intervals from the source. This approach aims to improve the reliability of final wind field predictions. Addressing the issues of insufficient penalty for errors in high-value intervals, low decision-making practicality, and poor prediction performance due to over-calibration or under-calibration in general wind field prediction methods, this solution defines the relative error of wind field prediction and introduces an error sensitivity enhancement factor to achieve nonlinear amplification of errors in high-error-sensitivity intervals. At the error measurement level, it transforms small absolute errors in core intervals such as high wind speeds into larger relative errors, providing a foundation for subsequent loss enhancement. A unified loss framework is designed to implement a dual mechanism of error amplification and loss enhancement, avoiding over-calibration or under-calibration and resolving the prediction bias caused by errors in high wind speed intervals, thereby improving wind field prediction performance.
[0004] The technical solution adopted by this invention is as follows: The wind field prediction method based on big data provided by this invention includes the following steps:
[0005] Step S1: Wind field data acquisition;
[0006] Step S2: Quantification of wind speed range characteristics;
[0007] Step S3: Construction of wind field prediction loss framework;
[0008] Step S4: Establishing the wind field prediction model;
[0009] Step S5: Real-time wind field prediction.
[0010] Further, in step S1, the wind field data acquisition involves obtaining historical wind field monitoring data, including wind turbine operation data, meteorological monitoring data, and numerical weather forecast data; performing preprocessing, including outlier handling, missing value completion, and data normalization; and labeling the data to indicate the actual wind speed at future time steps; thus obtaining the wind field dataset.
[0011] Furthermore, in step S2, the quantification of wind speed range features specifically includes:
[0012] Adaptive division of wind speed intervals: For the wind field dataset, based on the wind speed distribution characteristics of the wind field dataset, the K-means clustering algorithm is used to divide the wind speed range into K continuous and non-overlapping intervals;
[0013] Interval sample size statistics; count the number of samples within each wind speed interval;
[0014] Interval weight calculation; For each wind speed interval, calculate the interval weight;
[0015] Definition of the interval prediction error sensitivity index; quantification of the actual loss weight corresponding to unit numerical error in different wind speed intervals;
[0016] The error sensitivity enhancement factor is defined as follows: based on the interval prediction error sensitivity index, the actual decision loss caused by the unit prediction error in each wind speed interval is quantified and converted into a weight enhancement coefficient.
[0017] Furthermore, in step S3, the construction of the wind field prediction loss framework specifically includes:
[0018] Definition of relative error in wind field prediction; The relative error in wind field prediction is defined, and an error sensitivity enhancement factor is introduced to obtain the relative error in wind field prediction.
[0019] A unified loss framework is designed; the framework consists of three parts: the loss subject, heterogeneous calibration coefficients, and prediction error calibration terms.
[0020] Scale transformation: By adjusting the hyperparameters, the framework is transformed into a loss function suitable for different prediction scales.
[0021] Furthermore, in step S4, the establishment of the wind field prediction model specifically includes:
[0022] Wind field prediction model architecture selection; The spatiotemporal fusion Transformer model was selected as the basic architecture for the wind field prediction model.
[0023] Hierarchical loss training: The unified loss framework is used as the loss function of the wind field prediction model. For each sample, the wind speed interval and weight are determined first. The loss value of each sample is calculated, and the average of the losses of all samples is used to obtain the global loss. The Adam optimizer is used to minimize the global loss and update the model parameters. During the training process, the interval weights and error sensitivity enhancement factors are embedded into the gradient calculation.
[0024] Model convergence determination; dividing the wind field dataset into training, validation, and test sets; and then establishing the wind field prediction model.
[0025] Furthermore, in step S5, the real-time wind field prediction is based on the established wind field prediction model. The wind field monitoring data is collected in real time, preprocessed, and then input into the wind field prediction model to obtain a normalized prediction value, which is then converted into the actual wind speed. The wind speed prediction value of the wind field is output as the real-time wind field prediction result.
[0026] The wind field prediction system based on big data provided by this invention includes a wind field data acquisition module, a wind speed interval feature quantification module, a wind field prediction loss framework construction module, a wind field prediction model establishment module, and a real-time wind field prediction module.
[0027] The wind field data acquisition module acquires historical wind field monitoring data, performs preprocessing, and obtains a wind field dataset.
[0028] The wind speed interval feature quantization module divides the wind field dataset into wind speed intervals and calculates interval weights and error sensitivity enhancement factors to achieve wind speed interval feature quantization.
[0029] The wind field prediction loss framework construction module defines the prediction relative error of the fusion error sensitivity enhancement factor and designs a unified loss framework;
[0030] The wind field prediction model building module uses a unified loss framework as the loss function, embeds interval weights and error sensitivity enhancement factors, and builds a wind field prediction model.
[0031] The real-time wind field prediction module predicts the wind field based on the established wind field prediction model and the real-time collected wind field monitoring data.
[0032] The beneficial effects achieved by the present invention using the above solution are as follows:
[0033] (1) To address the problems of general wind field prediction methods being unable to adapt to the wind speed distribution characteristics of different wind fields, feature quantization tilting towards low-value intervals, and the mismatch between error cost and weight, which leads to poor reliability of wind field prediction, this solution defines interval weights to dynamically balance the differences between different intervals and avoid the dominant effect of multiple sample intervals; defines an interval prediction error sensitivity index to design an error sensitivity enhancement factor, which transforms the loss difference into a weight enhancement coefficient; and allows the feature quantization process to actively tilt towards intervals with high error cost, strengthening the feature expression of core intervals and improving the prediction accuracy of key intervals from the source; thereby improving the reliability of the final wind field prediction.
[0034] (2) To address the problems of insufficient penalty for errors in high-value intervals, low decision-making practicality, and poor wind field prediction performance caused by over-calibration or under-calibration in general wind field prediction methods, this scheme defines the relative error of wind field prediction and introduces an error sensitivity enhancement factor to achieve nonlinear amplification of errors in high error sensitivity intervals. At the error measurement level, the small absolute errors in core intervals such as high wind speed are transformed into larger relative errors, providing a basis for subsequent loss enhancement. A unified loss framework is designed to realize the dual mechanism of error amplification and loss enhancement, avoiding over-calibration or under-calibration, and solving the pain point of prediction deviation caused by errors in high wind speed intervals; thereby improving the wind field prediction performance. Attached Figure Description
[0035] Figure 1 A flowchart illustrating the wind field prediction method based on big data provided by this invention;
[0036] Figure 2 This is a schematic diagram of the wind field prediction system based on big data provided by the present invention.
[0037] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof. Detailed Implementation
[0038] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.
[0039] In the description of this invention, it should be understood that the terms "upper", "lower", "front", "rear", "left", "right", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the system or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.
[0040] Example 1, see Figure 1 The wind field prediction method based on big data provided by this invention includes the following steps:
[0041] Step S1: Wind field data acquisition; acquire historical wind field monitoring data, perform preprocessing, and obtain a wind field dataset;
[0042] Step S2: Wind speed interval feature quantization; Divide the wind field dataset into wind speed intervals, calculate interval weights and error sensitivity enhancement factors to achieve wind speed interval feature quantization;
[0043] Step S3: Constructing the wind field prediction loss framework; Define the prediction relative error of the fusion error sensitivity enhancement factor and design a unified loss framework;
[0044] Step S4: Wind field prediction model establishment; Using a unified loss framework as the loss function, embedding interval weights and error sensitivity enhancement factors to establish a wind field prediction model;
[0045] Step S5: Real-time wind field prediction; Based on the established wind field prediction model, wind field prediction is achieved using the real-time collected wind field monitoring data.
[0046] Example 2, see Figure 1 This embodiment is based on the above embodiment. In step S1, wind field data acquisition involves obtaining historical wind field monitoring data, including wind turbine operation data, meteorological monitoring data, and numerical weather prediction data. The wind turbine operation data includes historical wind speed, output power, impeller speed, and yaw angle of a single / cluster wind turbine. The meteorological monitoring data includes wind direction, atmospheric pressure, ambient temperature, and humidity in the wind field area. The numerical weather prediction data includes predicted wind speed and wind direction for short-term (minute-level), medium-term (hour-level), and long-term (day-level). Preprocessing is performed, including outlier handling (3σ criterion), missing value completion, and data normalization (max-min normalization). Missing value completion uses LSTM interpolation to complete continuously missing data and neighbor sample mean to complete discrete missing data. Data labeling is performed, labeling the actual wind speed at future time steps, with labels and feature normalization synchronized. A wind field dataset is obtained.
[0047] Example 3, see Figure 1This embodiment is based on the above embodiment. In step S2, the wind speed interval feature quantization is due to the difference in the sample size ratio of wind speed intervals: the sample proportion of low wind speed intervals is high, and the sample proportion of high wind speed intervals is low. However, the prediction accuracy of high wind speed intervals has a more critical impact on wind farm operation and maintenance and scheduling decisions. Therefore, for the wind farm dataset, adaptive feature quantization for intervals with different sample sizes is achieved through data-driven wind speed interval division and interval weight calculation. Specifically, this includes:
[0048] Adaptive wind speed interval partitioning: For the wind field dataset, based on the wind speed distribution characteristics of the dataset, the K-means clustering algorithm is used to divide the wind speed range into K (values from 5 to 10) continuous and non-overlapping intervals V. k The samples within a given interval have high similarity, while the samples between intervals have low similarity, ensuring that the wind speed characteristics of each interval are consistent.
[0049] Interval sample size statistics; statistics on V for each wind speed interval k Number of samples M k ;
[0050] Interval weight calculation; For each wind speed interval, calculate the interval weight, expressed as: ;in, It is the interval weight of the k-th wind speed interval; and These are the global maximum and minimum values, respectively. It is a smoothing term, with a value of 10. -8 ~10 -6 ;
[0051] Definition of Interval Prediction Error Sensitivity Index: This index quantifies the actual loss weight corresponding to a unit numerical error in different wind speed intervals. The interval prediction error sensitivity index is expressed as: ;in, It is the average wind speed in the range of the wind turbine power-wind speed curve. The slope at a point reflects the degree to which changes in wind speed affect power; This refers to the rated power of the fan; It is the maximum output power corresponding to the average wind speed in the interval;
[0052] The error sensitivity enhancement factor is defined as follows: Based on the interval prediction error sensitivity index, it quantifies the actual decision-making loss caused by a unit prediction error in each wind speed interval and converts it into a weight enhancement coefficient, tilting the feature quantification process towards intervals with high error costs. Represented as:
[0053] .
[0054] By performing the above operations, this solution addresses the problems of general wind field prediction methods, such as inability to adapt to the wind speed distribution characteristics of different wind fields, feature quantization bias towards low-value intervals, and mismatch between error costs and weights, leading to poor wind field prediction reliability. This solution defines interval weights to dynamically balance the differences between different intervals, avoiding the dominant effect of multiple sample intervals; it defines an interval prediction error sensitivity index and designs an error sensitivity enhancement factor to convert loss differences into weight enhancement coefficients; it allows the feature quantization process to actively favor intervals with high error costs, strengthening the feature representation of core intervals and improving the prediction accuracy of key intervals from the source; thus improving the final reliability of wind field prediction.
[0055] Example 4, see Figure 1 This embodiment is based on the above embodiment. In step S3, the construction of the wind field prediction loss framework is necessary because wind field prediction needs to cover different time scales (short-term, medium-term, and long-term) and different spatial scales (single wind turbine, wind turbine cluster, and overall wind field). Changing the loss function according to the scale would lead to a complex prediction model architecture and poor generalization ability. Therefore, a single loss integrated architecture is constructed based on interval weights. By adjusting the hyperparameters, it can be transformed into a loss function suitable for different scales, achieving a single loss that fits all scenarios. Specifically, it includes:
[0056] Definition of relative prediction error; let the model's predicted value for the input features be... , representing the normalized wind speed, with the true value being y. t The relative error of wind field prediction is defined, and an error sensitivity enhancement factor is introduced. For high error sensitivity intervals, the same absolute error will be nonlinearly amplified into a higher relative error, achieving severe penalty for small errors in high-value intervals; for low error sensitivity intervals, a linear error metric is maintained to avoid excessive penalty; the relative error of wind field prediction is expressed as: ;in, It is the relative error of wind field prediction at time step t. This indicates that the prediction was completely accurate; This indicates that the prediction bias has reached the global range level;
[0057] A unified loss framework is designed; the framework consists of three parts: the loss subject, heterogeneous calibration coefficients, and prediction error calibration terms, represented as follows: ;in, It is the loss value of the sample at time step t; It is a heterogeneous calibration coefficient that integrates interval weights and prediction errors to achieve heterogeneous calibration for different intervals; The first is the prediction error calibration term, which avoids excessive contribution of simplified samples (such as low wind speed sections with small deviations) to the loss; the second is the calibration nonlinearity amplitude control factor, which adjusts the degree of nonlinearity of the calibration coefficient as a function of the prediction relative error, with a value of 0 to 1.0; the third is the base calibration amplitude control factor, which sets the base amplitude of the calibration coefficient, a reference value that does not change with the prediction error, with a value of 0.1 to 1.0; the fourth is the error correction amplitude control factor, which adjusts the amplitude of the prediction error calibration term to control the degree of loss contribution of simplified samples, with a value of 0.05 to 0.5.
[0058] In Nonlinear amplification for error measurement. and In Weight enhancement for loss calibration further strengthens the weight of calibration coefficients and calibration terms on the basis of amplified error, ensuring that the loss in the high-value range accounts for a higher proportion of the global loss; thus realizing a synergistic optimization strategy of error amplification + loss enhancement; in wind field prediction, small errors in the high wind speed range can lead to exponential deviations in power prediction, and the dual mechanism can accurately solve this pain point.
[0059] Scale transformation; by adjusting hyperparameters, the framework is transformed into a loss function applicable to different prediction scales. The core transformation relationship is: for short-term single-wind turbine prediction, the nonlinear adjustment term of the calibration coefficient is turned off, i.e., a=0. For long-term overall wind field forecasting, the weights of all wind speed intervals are unified, ignoring differences in interval sample size and error sensitivity, focusing on the overall wind speed trend forecast. ,at this time .
[0060] Example 5, see Figure 1 This embodiment is based on the above embodiment. In step S4, the wind field prediction model is established because even within the same macroscopic wind speed range, the sample size of different sub-ranges or different wind turbines still differs. To avoid the problems of overcalibration due to a large sample size in a sub-range and undercalibration due to a small sample size in a sub-range, the hierarchical distribution characteristics of the interval weights are utilized to achieve differentiated training for different wind speed ranges. This ensures that the samples in each interval obtain parameter update intensity that matches their own sample size, and then a wind field prediction model is established based on the wind field dataset. Specifically, it includes:
[0061] Wind field prediction model architecture selection: The spatiotemporal fusion Transformer model is selected as the basic architecture of the wind field prediction model, while capturing: temporal characteristics: short-term fluctuations and long-term trends of wind speed; spatial characteristics: wake effect between wind turbines and wind field topography influence.
[0062] Hierarchical loss training: A unified loss framework is used as the loss function for the wind field prediction model. For each sample, the wind speed interval and weight are first determined. The loss value of each sample is calculated, and the average loss of all samples is taken to obtain the global loss. The Adam optimizer is used to minimize the global loss and update the model parameters. During training, the interval weights and error sensitivity enhancement factors are embedded in the gradient calculation. The magnitude of the gradient is positively correlated with the interval weights, expressed as: ;
[0063] The larger the sample size, the larger the gradient and the stronger the parameter update intensity (the stronger the calibration); conversely, the smaller the sample size, the weaker the update intensity (the weaker the calibration), thus achieving hierarchical heterogeneous training.
[0064] Model convergence determination; the wind field dataset is divided into training set, validation set and test set; the training set is used for updating the parameters of the wind field prediction model, the validation set is used to monitor overfitting, determine convergence and adjust overfitting parameters, and the test set is used for final performance evaluation; thus, the wind field prediction model is established.
[0065] By performing the above operations, this solution addresses the problems of insufficient penalty for errors in high-value intervals, low decision-making practicality, and poor wind field prediction performance due to over-calibration or under-calibration in general wind field prediction methods. It defines the relative error of wind field prediction and introduces an error sensitivity enhancement factor to achieve nonlinear amplification of errors in high error sensitivity intervals. At the error measurement level, it transforms the small absolute errors in core intervals such as high wind speeds into larger relative errors, providing a foundation for subsequent loss enhancement. A unified loss framework is designed to implement a dual mechanism of error amplification and loss enhancement, avoiding over-calibration or under-calibration and resolving the prediction bias caused by errors in high wind speed intervals, thereby improving wind field prediction performance.
[0066] Example 6, see Figure 1 This embodiment is based on the above embodiment. In step S5, the real-time wind field prediction is based on the established wind field prediction model. The wind field monitoring data is collected in real time, preprocessed and then input into the wind field prediction model to obtain the normalized prediction value, which is then converted into the actual wind speed. The short-term, medium-term and long-term wind speed prediction values of the wind field are output as the real-time wind field prediction results.
[0067] Example 7, see Figure 2 Based on the above embodiments, the wind field prediction system based on big data provided by the present invention includes a wind field data acquisition module, a wind speed interval feature quantification module, a wind field prediction loss framework construction module, a wind field prediction model establishment module, and a real-time wind field prediction module.
[0068] The wind field data acquisition module acquires historical wind field monitoring data, performs preprocessing, and obtains a wind field dataset.
[0069] The wind speed interval feature quantization module divides the wind field dataset into wind speed intervals and calculates interval weights and error sensitivity enhancement factors to achieve wind speed interval feature quantization.
[0070] The wind field prediction loss framework construction module defines the prediction relative error of the fusion error sensitivity enhancement factor and designs a unified loss framework;
[0071] The wind field prediction model building module uses a unified loss framework as the loss function, embeds interval weights and error sensitivity enhancement factors, and builds a wind field prediction model.
[0072] The real-time wind field prediction module predicts the wind field based on the established wind field prediction model and the real-time collected wind field monitoring data.
[0073] It should be noted that, in this document, the terms “comprising,” “including,” or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.
[0074] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention.
[0075] The present invention and its embodiments have been described above. This description is not restrictive, and the accompanying drawings are only one embodiment of the present invention; the actual structure is not limited thereto. In conclusion, if those skilled in the art are inspired by this description and design similar structures and embodiments without departing from the spirit of the invention, such designs should fall within the protection scope of the present invention.
Claims
1. A wind field prediction method based on big data, characterized by: The method includes the following steps: Step S1: Wind field data acquisition; acquire historical wind field monitoring data, perform preprocessing, and obtain a wind field dataset; Step S2: Wind speed interval feature quantization; Divide the wind field dataset into wind speed intervals, calculate interval weights and error sensitivity enhancement factors to achieve wind speed interval feature quantization; Step S3: Constructing the wind field prediction loss framework; Define the prediction relative error of the fusion error sensitivity enhancement factor and design a unified loss framework; Step S4: Wind field prediction model establishment; Using a unified loss framework as the loss function, embedding interval weights and error sensitivity enhancement factors to establish a wind field prediction model; Step S5: Real-time wind field prediction; Based on the established wind field prediction model, wind field prediction is achieved using the real-time collected wind field monitoring data.
2. The wind field prediction method based on big data according to claim 1, characterized in that: In step S2, the quantization of wind speed range features specifically includes: Adaptive division of wind speed intervals: For the wind field dataset, based on the wind speed distribution characteristics of the wind field dataset, the K-means clustering algorithm is used to divide the wind speed range into K continuous and non-overlapping intervals; Interval sample size statistics; count the number of samples within each wind speed interval; Interval weight calculation; For each wind speed interval, calculate the interval weight; Definition of the interval prediction error sensitivity index; quantification of the actual loss weight corresponding to unit numerical error in different wind speed intervals; Definition of error sensitivity enhancement factor.
3. The wind field prediction method based on big data according to claim 2, characterized in that: The error sensitivity enhancement factor is defined based on the interval prediction error sensitivity index, which quantifies the actual decision loss caused by a unit prediction error in each wind speed interval and converts it into a weight enhancement coefficient.
4. The wind field prediction method based on big data according to claim 3, characterized in that: In step S3, the construction of the wind field prediction loss framework specifically includes: Definition of relative error in wind field prediction; The relative error in wind field prediction is defined, and an error sensitivity enhancement factor is introduced to obtain the relative error in wind field prediction. A unified loss framework is designed; the framework consists of three parts: the loss subject, heterogeneous calibration coefficients, and prediction error calibration terms. Scale transformation: By adjusting the hyperparameters, the framework is transformed into a loss function suitable for different prediction scales.
5. The wind field prediction method based on big data according to claim 4, characterized in that: In step S4, the establishment of the wind field prediction model specifically includes: Wind field prediction model architecture selection; The spatiotemporal fusion Transformer model was selected as the basic architecture for the wind field prediction model. Hierarchical loss training: The unified loss framework is used as the loss function of the wind field prediction model. For each sample, the wind speed interval and weight are determined first. The loss value of each sample is calculated, and the average of the losses of all samples is used to obtain the global loss. The Adam optimizer is used to minimize the global loss and update the model parameters. During the training process, the interval weights and error sensitivity enhancement factors are embedded into the gradient calculation. Model convergence determination; dividing the wind field dataset into training, validation, and test sets; and then establishing the wind field prediction model.
6. The wind field prediction method based on big data according to claim 5, characterized in that: In step S1, the wind field data acquisition involves obtaining historical wind field monitoring data, including wind turbine operation data, meteorological monitoring data, and numerical weather forecast data. Preprocessing is performed, including outlier handling, missing value completion, and data normalization; And the data is labeled to indicate the actual wind speed at future time steps; Obtain the wind field dataset.
7. The wind field prediction method based on big data according to claim 6, characterized in that: In step S5, the real-time wind field prediction is based on the established wind field prediction model. The wind field monitoring data is collected in real time, preprocessed, and then input into the wind field prediction model to obtain a normalized prediction value, which is then converted into the actual wind speed. Output the predicted wind speed value of the wind field; As a result of real-time wind field prediction.
8. A wind field prediction system based on big data, used to implement the wind field prediction method based on big data as described in any one of claims 1-7, characterized in that: It includes a wind field data acquisition module, a wind speed interval feature quantification module, a wind field prediction loss framework construction module, a wind field prediction model establishment module, and a real-time wind field prediction module; The wind field data acquisition module acquires historical wind field monitoring data, performs preprocessing, and obtains a wind field dataset. The wind speed interval feature quantization module divides the wind field dataset into wind speed intervals and calculates interval weights and error sensitivity enhancement factors to achieve wind speed interval feature quantization. The wind field prediction loss framework construction module defines the prediction relative error of the fusion error sensitivity enhancement factor and designs a unified loss framework; The wind field prediction model building module uses a unified loss framework as the loss function, embeds interval weights and error sensitivity enhancement factors, and builds a wind field prediction model. The real-time wind field prediction module predicts the wind field based on the established wind field prediction model and the real-time collected wind field monitoring data.