Wind power plant wind speed multi-strategy integrated prediction method and system and related equipment

Abstract

The invention discloses a wind power plant wind speed multi-strategy integrated prediction method and system and related equipment, and the method comprises the steps: obtaining the operation data and multi-source meteorological observation data of a wind power plant monitoring and data collection system, and carrying out the time alignment and data cleaning; constructing multi-dimensional meteorological data features based on the processed operation data and the multi-source meteorological observation data; the multi-dimensional meteorological data features are comprehensive feature spaces including time sequence features, physical features and microclimate features; carrying out clustering analysis on the multi-dimensional meteorological data features by adopting a K-means clustering algorithm, and automatically identifying a microclimate mode type; a Light GBM prediction model is constructed for each microclimate type; and fusing prediction results of the data driving model, the microclimate adaptive model and the physical constraint model, and generating a final wind speed prediction value through dynamic weight distribution. According to the method, the influence of microclimate condition space-time variability on the prediction performance is solved, and technical support is provided for wind power plant operation decision making, power generation capacity evaluation and power grid dispatching.

Description

Technical Field

[0001] This invention relates to the field of wind power generation technology, particularly to the field of wind speed prediction and machine learning technology, and more specifically to a multi-strategy integrated prediction method, system and related equipment for wind speed based on wind farm microclimate. Background Technology

[0002] As an important component of clean energy, wind power has experienced explosive growth in installed capacity. Wind speed forecasting plays a crucial role in wind farm operational efficiency, power generation assessment, and grid stability. Accurate wind speed forecasting can effectively improve the economic benefits of wind farms, optimize power generation scheduling plans, reduce grid fluctuation risks, and provide important data for predictive maintenance of wind power equipment.

[0003] However, wind speed forecasting faces numerous technical challenges. First, wind speed exhibits strong nonlinearity, high randomness, and complex spatiotemporal variability, with its variation patterns influenced by a combination of meteorological factors. Traditional linear forecasting methods struggle to effectively capture these complex nonlinear relationships. Second, the spatiotemporal variability of microclimate conditions significantly impacts the accuracy of wind speed forecasts. Different combinations of local meteorological conditions reveal marked differences in wind speed variation patterns, and existing forecasting methods generally lack the ability to adapt to these variations.

[0004] Existing wind speed forecasting methods are mainly divided into three categories: statistical methods, machine learning methods, and deep learning methods. Traditional statistical methods mainly include time series forecasting methods such as Autoregressive Integrated Moving Average (ARIMA) and Kalman filtering. These methods have the advantages of solid theoretical foundation and high computational efficiency, but they have obvious limitations in handling nonlinear and multivariate coupled problems, and are difficult to effectively capture the complex nonlinear variation characteristics of wind speed data.

[0005] Machine learning methods include Support Vector Machines (SVM), Random Forests, and Gradient Boosting Trees, which possess strong nonlinear modeling capabilities and good generalization performance. However, most existing machine learning methods adopt a general modeling strategy, that is, using a single model to handle prediction tasks under all meteorological conditions, lacking the ability to differentiate modeling for the characteristics of different meteorological conditions, resulting in unstable prediction performance under specific microclimate conditions.

[0006] Deep learning methods, such as Recurrent Neural Networks (RNNs), Long Short-Term Memory (LSTMs), Gated Recurrent Units (GRUs), and Transformers, possess powerful feature learning and sequence modeling capabilities, demonstrating certain advantages in wind speed prediction tasks. However, deep learning methods suffer from insufficient model interpretability, are prone to overfitting in small sample sizes, and have long training times and high computational resource consumption, making them difficult to meet the engineering application requirements of real-time prediction.

[0007] More importantly, most existing forecasting methods neglect the significant impact of microclimate conditions on wind speed prediction. Microclimate refers to localized meteorological conditions, including the combined state of various meteorological elements such as temperature, humidity, air pressure, wind direction, and cloud cover. These local meteorological conditions have differentiated effects on wind turbine performance and wind speed variation patterns. Under different microclimate conditions, wind speed variation patterns, the importance of influencing factors, and the applicability of forecasting models all differ significantly. However, current technologies lack systematic identification and specialized modeling research on these microclimate differences.

[0008] In terms of feature engineering, existing methods also have shortcomings. Most methods only consider simple time-series features or basic meteorological features, lacking in-depth exploration of wind energy physics mechanisms and systematic construction of microclimate features. At the same time, existing ensemble forecasting methods mostly adopt static weight allocation strategies, which are difficult to adapt to the dynamic changes in the performance of various forecasting models under different meteorological conditions.

[0009] Furthermore, existing wind speed forecasting systems also face challenges in engineering applications. On the one hand, while complex deep learning models offer high prediction accuracy, they suffer from long response times and high resource consumption, making it difficult to meet the demands of real-time forecasting. On the other hand, while simple statistical models are computationally efficient, their prediction accuracy is limited, failing to meet the high-precision forecasting requirements of modern wind farms.

[0010] Therefore, how to develop a wind speed prediction technology that can automatically identify microclimate patterns, achieve specialized modeling for different meteorological conditions, effectively integrate multiple prediction strategies, and have good engineering application performance is a technical problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0011] In view of this, the present invention provides a wind farm wind speed multi-strategy integrated prediction method, system and related equipment that at least solves some of the above-mentioned technical problems, and can provide accurate wind speed prediction support for wind farms.

[0012] To achieve the above objectives, the present invention adopts the following technical solution:

[0013] In a first aspect, embodiments of the present invention provide a multi-strategy integrated prediction method for wind speed in wind farms, characterized by comprising the following steps:

[0014] S1. Acquire operational data from the wind farm monitoring and data acquisition system and multi-source meteorological observation data, and perform time alignment and data cleaning.

[0015] S2. Construct multidimensional meteorological data features based on the processed operational data and multi-source meteorological observation data; the multidimensional meteorological data features are a comprehensive feature space that includes temporal features, physical features and microclimate features;

[0016] S3. The K-means clustering algorithm is used to perform cluster analysis on the multidimensional meteorological data features to automatically identify the microclimate model type;

[0017] S4. Construct a LightGBM prediction model for each microclimate type;

[0018] S5. The prediction results of the data-driven model, the microclimate adaptive model, and the physical constraint model are integrated and the final wind speed prediction value is generated through dynamic weight allocation. The data-driven model is obtained by training on the complete dataset in steps S1 and S2. The microclimate adaptive model is obtained by training in steps S3 and S4. The physical constraint model is obtained by integrating physical knowledge into the machine learning model in step S2, combined with the wind profile law.

[0019] Further, step S1 specifically includes:

[0020] S11. Obtain wind turbine operation data from the wind farm monitoring and data acquisition system, including average wind speed, standard deviation of wind speed, active power and nacelle temperature;

[0021] S12. Obtain multi-source meteorological observation data through an open meteorological data application programming interface, including temperature, wind speed, relative humidity, and sea level pressure at different altitudes;

[0022] S13. Perform time synchronization processing on data sources with different frequencies;

[0023] S14. Perform outlier detection, missing value imputation, and data quality control.

[0024] Further, step S2 specifically includes:

[0025] S21. Extract historical wind speed lag variables from 1 hour to 24 hours, and calculate the statistical characteristics of mean, standard deviation, maximum and minimum values ​​within different time windows;

[0026] S22. Calculate air density, wind energy density, turbulence intensity, and wind direction decomposition characteristics based on wind energy engineering theory;

[0027] S23. Calculate the comfort index, temperature change rate, and interaction characteristics of meteorological elements;

[0028] S24. Use the Z-score standardization method to normalize all features.

[0029] Furthermore, step S3 specifically includes:

[0030] S31. Select temperature, wind speed, relative humidity, seasonal coding features, and time period features as clustering inputs;

[0031] S32. Determine the optimal number of clusters using the silhouette coefficient method;

[0032] S33. Automatically identify microclimate patterns by performing the K-means clustering algorithm through minimizing the objective function;

[0033] S34 uses principal component analysis for dimensionality reduction visualization to verify the effectiveness of the clustering results.

[0034] Further, step S4 specifically includes:

[0035] S41. Based on the microclimate clustering results, group the training data by type, and train a LightGBM regressor independently for each group of data.

[0036] S42. Optimize the hyperparameters of each LightGBM regressor using a grid search combined with cross-validation.

[0037] S43. Use the nearest neighbor classifier to identify the microclimate type for new input samples;

[0038] S44. Select the corresponding prediction model based on the identified microclimate type for prediction.

[0039] Further, step S5 specifically includes:

[0040] S51. Construct the basic LightGBM data-driven model, the microclimate adaptive MA-LightGBM model, and the physical constraint Physics-LightGBM model respectively.

[0041] S52. Calculate the dynamic weighting coefficients based on the historical prediction performance of each model within the sliding window;

[0042] S53. The final prediction result is generated by dynamically weighted fusion of three models.

[0043] In a second aspect, embodiments of the present invention also provide a wind farm wind speed multi-strategy integrated prediction system, which applies the wind farm wind speed multi-strategy integrated prediction method as described in any one of the first aspects to provide wind speed prediction for the wind farm. The system includes:

[0044] The data fusion module is used to acquire operational data from the wind farm monitoring and data acquisition system and multi-source meteorological observation data, and to perform time alignment and data cleaning.

[0045] The feature engineering module is used to construct multidimensional meteorological data features based on the processed operational data and multi-source meteorological observation data; the multidimensional meteorological data features are a comprehensive feature space that includes temporal features, physical features and microclimate features;

[0046] The microclimate identification module is used to cluster the multidimensional meteorological data features using the K-means clustering algorithm to automatically identify the microclimate pattern type;

[0047] The adaptive modeling module is used to build LightGBM prediction models for each microclimate type;

[0048] An integrated prediction module is used to fuse the prediction results of a data-driven model, a microclimate adaptive model, and a physical constraint model, and generate the final wind speed prediction value through dynamic weight allocation. The data-driven model is trained on the complete dataset of the data fusion module and the feature engineering module. The microclimate adaptive model is trained on the microclimate identification module and the adaptive modeling module. The physical constraint model is obtained by combining the feature engineering module with the wind profile law, integrating physical knowledge into the machine learning model.

[0049] Furthermore, the feature engineering module specifically includes:

[0050] The time-series feature extraction submodule is used to extract historical wind speed lag variables and statistical features at multiple time scales;

[0051] The physical characteristic calculation submodule is used to calculate physical characteristics such as air density, turbulence intensity, and wind direction decomposition based on wind energy engineering theory.

[0052] The microclimate feature generation submodule is used to calculate microclimate-sensitive features such as comfort index and temperature change rate;

[0053] The feature standardization submodule is used to perform Z-score standardization on all features.

[0054] Furthermore, the microclimate recognition module specifically includes:

[0055] The clustering feature selection submodule is used to select meteorological features such as temperature, wind speed, and relative humidity as clustering inputs.

[0056] The optimal number of clusters determination submodule is used to determine the optimal number of clusters using the silhouette coefficient method;

[0057] The K-means clustering submodule is used to perform unsupervised clustering algorithms to identify microclimate patterns;

[0058] The clustering result verification submodule is used to verify the validity of the clustering results through principal component analysis.

[0059] Furthermore, the adaptive modeling module specifically includes:

[0060] The group training submodule is used to group training data by type according to the microclimate clustering results and train specialized models.

[0061] The hyperparameter optimization submodule is used to optimize model hyperparameters using a grid search combined with cross-validation.

[0062] The microclimate type identification submodule is used to identify the microclimate type of new input samples;

[0063] The specialized model selection submodule is used to select the corresponding prediction model based on the identified microclimate type.

[0064] Furthermore, the integrated prediction module specifically includes:

[0065] The multi-strategy model building submodule is used to build predictive models using three modeling strategies: data-driven, microclimate adaptive, and physical constraint.

[0066] The dynamic weight calculation submodule is used to calculate dynamic weight coefficients based on the historical prediction performance of each model.

[0067] The weighted fusion submodule is used to generate the final prediction result through dynamic weight allocation.

[0068] Thirdly, embodiments of the present invention provide a computer-readable storage medium storing a computer program that, when executed by a processor, implements the wind farm wind speed multi-strategy integrated prediction method as described in any of the first aspects.

[0069] Fourthly, embodiments of the present invention also provide an electronic device, including a processor and a memory, wherein the memory stores machine-executable instructions that can be executed by the processor, and the processor executes the machine-executable instructions to implement the wind farm wind speed multi-strategy integrated prediction method as described in any of the first aspects.

[0070] The descriptions of the second to fourth aspects of this invention can be referred to the detailed description of the first aspect; and the beneficial effects described in the second to fourth aspects can be referred to the analysis of the beneficial effects of the first aspect, which will not be repeated here.

[0071] As can be seen from the above technical solution, compared with the prior art, the present invention has the following technical advantages:

[0072] This invention applies the concept of microclimate to the field of wind speed prediction. It automatically identifies microclimate patterns through unsupervised clustering. This method can automatically discover the regular differences under different meteorological conditions, providing a basis for specialized modeling.

[0073] This invention employs a specialized modeling strategy to construct independent prediction models for each microclimate type, thus solving the performance problem of general models under specific conditions.

[0074] The multidimensional integrated feature space constructed in this invention covers three dimensions: time series, physical, and microclimate. Among them, the microclimate features quantify the comprehensive impact of local meteorological conditions and provide input information for the prediction model.

[0075] This invention adopts a multi-strategy integrated architecture, which integrates three modeling methods: data-driven, physical constraint, and microclimate adaptive. It adjusts the weights of each model based on the historical performance of each model under different conditions through a dynamic weight allocation mechanism.

[0076] This invention has engineering application value, meets the needs of real-time prediction, and the lightweight model design based on LightGBM has high computational efficiency while ensuring prediction accuracy. The modular system design can be applied to different wind farm environments. Attached Figure Description

[0077] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0078] Figure 1 The flowchart of the wind speed multi-strategy integrated prediction method for wind farms provided by the present invention is shown.

[0079] Figure 2 The flowchart of the wind speed multi-strategy integrated prediction method for wind farms provided by the present invention is shown.

[0080] Figure 3 The flowchart for step S2, the feature engineering stage, is provided for this invention.

[0081] Figure 4The flowchart for step S3, the microclimate identification stage, is provided for this invention.

[0082] Figure 5 The flowchart for step S4, the adaptive modeling stage, is provided for this invention.

[0083] Figure 6 The flowchart for step S5, the integrated prediction stage, is provided for this invention.

[0084] Figure 7 This is a diagram illustrating the architecture of the wind farm wind speed multi-strategy integrated prediction system provided by the present invention.

[0085] Figure 8 This is a structural diagram of the electronic device provided by the present invention. Detailed Implementation

[0086] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0087] Example 1:

[0088] Reference Figure 1 As shown in the figure, an embodiment of the present invention provides a multi-strategy integrated prediction method for wind speed in wind farms, comprising the following steps:

[0089] S1. Acquire operational data from the Supervisory Control and Data Acquisition (SCADA) system and multi-source meteorological observation data, and perform time alignment and data cleaning.

[0090] S2. Construct multidimensional meteorological data features based on the processed operational data and multi-source meteorological observation data; the multidimensional meteorological data features are a comprehensive feature space that includes temporal features, physical features and microclimate features;

[0091] S3. The K-means unsupervised clustering algorithm is used to perform cluster analysis on the multidimensional meteorological data features to automatically identify different microclimate model types;

[0092] S4. Construct a LightGBM prediction model for each microclimate type; that is, based on the microclimate model type, construct a specialized Lightweight Gradient Boosting Machine (LightGBM) prediction model for each microclimate type.

[0093] S5. The prediction results of the data-driven model, the microclimate adaptive model, and the physical constraint model are integrated and the final wind speed prediction value is generated through dynamic weight allocation. The data-driven model is obtained by training on the complete dataset in steps S1 and S2. The microclimate adaptive model is obtained by training in steps S3 and S4. The physical constraint model is obtained by integrating physical knowledge into the machine learning model in step S2, combined with the wind profile law.

[0094] The aforementioned data-driven model is a predictive model built solely based on statistical patterns from historical data, without incorporating physical knowledge or microclimate characteristics. It directly uses the cleaned and aligned data from step S1, combined with the comprehensive feature space from step S2, as a complete dataset for training.

[0095] The Microclimate Adaptive Model (MA-LightGBM model) is a set of specialized models tailored for different microclimate types, adapting to changes in local meteorological conditions. Its construction logic relies on the clustering identification of microclimate types in step S3, and the output results of training the specialized LightGBM models by type in step S4.

[0096] The Physics-LightGBM model incorporates the physical laws of wind energy engineering (such as the wind profile law) as constraints into the machine learning model. It is obtained by introducing additional physical equations based on the physical features obtained in step S2, such as turbulence intensity and wind direction decomposition.

[0097] The following is combined with Figure 2 Each of the above steps will be explained in detail. Figure 2 This demonstrates a complete data processing flow, from inputting SCADA operation data from wind farms and multi-source meteorological data, through data fusion, feature engineering, microclimate identification, adaptive modeling, and integrated prediction, to finally outputting wind speed prediction results.

[0098] In this embodiment, step S1 is the data fusion stage, and its specific implementation method includes:

[0099] S11 acquires wind turbine operating data from the wind farm monitoring and data acquisition system (SCADA), specifically including wind turbine operating parameters such as the average wind speed, standard deviation of wind speed, active power, nacelle temperature, and ambient temperature from the SCADA system, for example, the data acquisition frequency is 15 minutes.

[0100] S12 acquires multi-source meteorological observation data through an open meteorological data application programming interface (API), including meteorological elements such as temperature at 2 meters altitude, wind speed at 10 meters altitude, relative humidity, sea level air pressure, and precipitation, with a data acquisition frequency of 1 hour.

[0101] S13 performs time synchronization processing on data sources of different frequencies, converting 15-minute frequency data into 1-hour frequency data. Time series resampling of SCADA data is performed using mean aggregation; the time alignment formula is as follows:

[0102]

[0103] Among them, V hour V represents the hourly average wind speed, in m / s. 15min,i This represents the wind speed data sampled in the i-th 15-minute period, in m / s, where i ranges from 1 to 4.

[0104] S14 involves outlier detection, missing value imputation, and data quality control. Outlier detection uses the 3σ principle, meaning data points exceeding the mean ± 3 standard deviations are identified as outliers. The criteria are expressed as follows:

[0105]

[0106] Among them, X i This represents the value of the i-th data point to be detected; σ represents the arithmetic mean of a data series. X The standard deviation of a data series is represented by the inequality X; when the inequality holds, the data point X... i It was identified as an outlier.

[0107] Step S2 is the feature engineering stage, refer to... Figure 3 As shown, its specific implementation methods include:

[0108] S21, extract historical wind speed lagged variables from 1 hour to 24 hours, and calculate the statistical characteristics of the mean, standard deviation, maximum, and minimum values ​​within different time windows. The lagged variable extraction is represented as follows:

[0109] F lag =[V t-1 V t-2 ,...,V t-24 ]

[0110] Among them, F lag V represents the eigenvector of a historical lagged variable; t-k This represents the wind speed value k hours before the current time t, in m / s, where k ranges from 1 to 24.

[0111] The statistical characteristics within different time windows are calculated and represented as follows:

[0112]

[0113] μ winIt represents the average wind speed within a time window, in m / s, and serves as a baseline for wind speed.

[0114] w represents the width of the time window, in hours, which determines the statistical scale (e.g., 6h / 12h / 24h).

[0115] j represents the time index variable, which is dimensionless and serves as a counter for all moments within the traversal window;

[0116] V t-j+1 This represents the historical wind speed sequence, in m / s, and is the actual wind speed observation value at the j-th moment within the window.

[0117] t represents the current time point, which is used as the calculation reference time (e.g., 12:00 on 2023-07-01);

[0118] σ win This indicates the intensity of wind speed fluctuations, expressed in m / s, and quantifies the degree of drastic change in wind speed.

[0119] S22 calculates air density, wind energy density, turbulence intensity, and wind direction decomposition characteristics based on wind energy engineering theory. Air density is calculated using the ideal gas law and expressed as:

[0120]

[0121] Where ρ represents air density, with units of kg / m³. 3 P represents atmospheric pressure, in Pa; R represents the gas constant of dry air, with a value of 287 J / (kg·K); T abs This indicates absolute temperature, and the unit is K.

[0122] Wind energy density is calculated as follows:

[0123]

[0124] Among them, P wind Wind energy density is expressed in W / m³. 2 ρ represents air density, with units of kg / m³. 3 V wind This indicates wind speed, measured in m / s.

[0125] Turbulence intensity is defined as the ratio of the standard deviation of wind speed to the mean wind speed, expressed as:

[0126]

[0127] Where TI represents turbulence intensity, which is dimensionless; σ wind This represents the standard deviation of wind speed, in m / s. This represents the average wind speed, expressed in m / s.

[0128] S23 calculates the comfort index, temperature change rate, and interaction characteristics of meteorological elements. The comfort index combines the combined effects of temperature, humidity, and air pressure, and is expressed as:

[0129] CI = a·T air +b·H rel +c·P atm +d

[0130] Where CI represents the comfort index, which is dimensionless; T air H represents air temperature, expressed in °C. rel Relative humidity is expressed in % (%); P atm The pressure is represented by hPa; a, b, c, and d are empirical coefficients, which are dimensionless.

[0131] The rate of temperature change is calculated as follows:

[0132]

[0133] in, T represents the rate of temperature change, with units of °C / h; current T represents the current temperature, in °C. previous This indicates the temperature at the previous moment, in °C.

[0134] S24 uses the Z-score standardization method to normalize all features. The standardization formula is expressed as:

[0135]

[0136] Where Z represents the standardized eigenvalue, which is dimensionless; X represents the original eigenvalue; μ X σ represents the arithmetic mean of the characteristic sequence. X The standard deviation of the characteristic sequence is represented.

[0137] In this embodiment, Figure 3 The feature engineering phase is demonstrated, encompassing a feature engineering architecture across three dimensions: temporal features, physical features, and microclimate features. Temporal features include 1-24 hour lagged variables and statistical characteristics; physical features include air density, wind energy density, turbulence intensity, and wind direction decomposition characteristics; and microclimate features include comfort index, temperature change rate, and interaction characteristics of meteorological elements. All features are normalized and standardized using Z-scores to form a multi-dimensional comprehensive feature space.

[0138] Step S3 is the microclimate identification stage, referring to... Figure 4 As shown, its specific implementation methods include:

[0139] S31 selects meteorological elements such as temperature, wind speed, and relative humidity, as well as seasonal coding features and time periodic features as clustering inputs. The seasonal coding features encode monthly information using sine and cosine transforms, and are represented as follows:

[0140]

[0141] Among them, S sin The sinusoidal component representing seasonal characteristics is dimensionless; S cos The cosine component represents the seasonal characteristics and is dimensionless; M represents the month, with a value ranging from 1 to 12.

[0142] S32, the silhouette coefficient method is used to determine the optimal number of clusters. The silhouette coefficient is calculated as follows:

[0143]

[0144] Among them, s i The profile coefficient of sample i is represented, and its value ranges from [-1, 1]; a i b represents the average distance from sample i to other samples of the same class; i This represents the average distance from sample i to the nearest sample of the opposite class.

[0145] S33 automatically identifies microclimate patterns by minimizing the objective function using the K-means clustering algorithm.

[0146]

[0147] Where J represents the objective function value; k represents the number of clusters; C j Denotes the j-th cluster; x i μ represents the vector of the i-th sample point in the cluster. j Let represent the center vector of the j-th cluster.

[0148] S34 uses Principal Component Analysis (PCA) for dimensionality reduction visualization to verify the effectiveness of the clustering results. By projecting the high-dimensional clustering results onto a two-dimensional plane, the distribution of different microclimate types and the clustering effect are visually displayed.

[0149] The specific implementation of the K-means clustering algorithm in step S3 above includes:

[0150] ①The silhouette coefficient is used to evaluate the clustering quality. When the silhouette coefficient is close to 1, it indicates that the clustering effect is good, and when it is close to -1, it indicates that the clustering effect is poor.

[0151] ② Use the elbow rule to help determine the optimal number of clusters, and determine the appropriate value of k by observing the trend of the objective function as the number of clusters changes.

[0152] ③ By randomly initializing cluster centers multiple times, the clustering result with the smallest objective function value is selected to avoid local optima.

[0153] ④ Introduce a clustering validity verification mechanism, and use principal component analysis to project the high-dimensional clustering results onto a two-dimensional plane for visual verification.

[0154] In this embodiment, Figure 4 This paper demonstrates the complete workflow for microclimate pattern recognition using the K-means unsupervised clustering algorithm. The workflow includes key steps such as cluster feature selection (temperature, wind speed, relative humidity, seasonality, and time-period features), determining the optimal number of clusters using the silhouette coefficient method, minimizing the objective function using K-means clustering to identify microclimate patterns, and dimensionality reduction visualization and cluster effectiveness verification using principal component analysis (PCA).

[0155] Step S4 is the adaptive modeling stage, referring to... Figure 5 As shown, its specific implementation methods include:

[0156] S41, based on the microclimate clustering results, group the training data by type, and train a LightGBM regressor independently for each group. Assuming the clustering results divide the data into k groups, train k specialized models: M1, M2, ..., M k Each specialized model is optimized for specific microclimate conditions.

[0157] S42 employs a grid search combined with cross-validation to optimize the hyperparameters of each specialized model. Key optimization parameters include the learning rate, tree depth (max_depth), number of leaf nodes (num_leaves), and feature sampling ratio (feature_fraction). Cross-validation is used in the optimization process, with root mean square error (RMSE) as the evaluation metric.

[0158] S43, a nearest neighbor classifier is used to identify the microclimate type for the new input samples. The specific process includes calculating the Euclidean distance from the new sample to each cluster center, expressed as:

[0159]

[0160] Where, d j The x represents the Euclidean distance from the new sample to the j-th cluster center; N represents the feature dimension; x n This represents the nth feature value of the new sample. μ j,n The nth dimension coordinates represent the cluster centers of the j-th cluster, which are determined by the characteristics and characterize the typical features of this type of microclimate.

[0161] The cluster center with the smallest distance is selected as the microclimate type for the new sample:

[0162] j * =argmin j d j

[0163] Where, j * This indicates the identified microclimate type number.

[0164] S44, based on the identified microclimate type, selects the corresponding specialized prediction model for prediction, realizing the transformation from general modeling to specialized modeling. The specific implementation of the specialized modeling strategy includes:

[0165] ① Perform independent feature selection and engineering on the data for each microclimate type to identify the most important feature variables for that type.

[0166] ② Stratified sampling is used to ensure that the training data for each specialized model is representative.

[0167] ③Evaluate the performance of each specialized model through cross-validation to ensure that the model has good generalization ability.

[0168] ④ Establish a model selection mechanism to automatically select the appropriate specialized model for prediction based on the microclimate type of the new sample.

[0169] In this embodiment, Figure 5 This diagram illustrates the process of grouping training data by type based on microclimate clustering results and constructing a specialized Lightweight Gradient Boosting Machine (LightGBM) prediction model for each microclimate type. The diagram includes sub-steps such as grouping training data by type based on microclimate clustering results, independently training a LightGBM regressor for each microclimate type, optimizing hyperparameters using grid search combined with cross-validation, identifying the microclimate type using a nearest neighbor classifier for new input samples, and selecting the corresponding specialized prediction model to perform predictions based on the identified microclimate type.

[0170] Step S5 is the integrated prediction stage, referring to... Figure 6 As shown, its specific implementation methods include:

[0171] S51, construct the basic LightGBM data-driven model, the microclimate adaptive MA-LightGBM model, and the physical constraint Physics-LightGBM model respectively. Among them, the basic LightGBM data-driven model uses all training data to train a single LightGBM model; the microclimate adaptive MA-LightGBM model is based on the aforementioned set of specialized models; and the physical constraint Physics-LightGBM model incorporates constraints from physical knowledge such as the wind profile law.

[0172] In a physical constraint model, the wind profile law is expressed as:

[0173]

[0174] Among them, V h V represents the wind speed at height h, in m / s. ref Indicates the reference height h ref The wind speed at the location is expressed in m / s; α represents the surface roughness coefficient, which is dimensionless.

[0175] S52, calculate the dynamic weight coefficients based on the historical prediction performance of each model within the sliding window. Let the sliding window length be W. slide Then the weights of the m-th model are expressed as:

[0176]

[0177] Among them, w m (t) represents the dynamic weights of the m-th model at time t; γ represents the root mean square error of the m-th model within the time window; γ represents the weight adjustment factor.

[0178] S53 generates the final prediction result through dynamic weight allocation. The weighted fusion formula is expressed as:

[0179]

[0180] in, This represents the final integrated prediction result, in m / s. This represents the prediction result of the m-th model, in m / s.

[0181] In this embodiment, Figure 6 This module demonstrates a complete architecture integrating three modeling strategies: the LightGBM data-driven model, the MA-LightGBM microclimate adaptive model, and the Physics-LightGBM physical constraint model. It calculates the real-time weights of each model through prediction, dynamic weight calculation, and dynamic weight allocation, and generates the final wind speed prediction result through weighted fusion, showcasing the technical advantages of multi-strategy integration.

[0182] As will be apparent to those skilled in the art from the above embodiments, the present invention provides a multi-strategy integrated prediction method for wind speed in wind farms. First, the present invention applies the concept of microclimate to the field of wind speed prediction, automatically identifying microclimate patterns through unsupervised clustering, thus automatically discovering the regular differences under different meteorological conditions and providing a basis for specialized modeling. Second, the present invention employs a specialized modeling strategy, constructing independent prediction models for each microclimate type, solving the performance problem of general models under specific conditions. Furthermore, the multi-dimensional comprehensive feature space constructed by the present invention covers three dimensions: temporal, physical, and microclimate, where microclimate features quantify the comprehensive impact of local meteorological conditions. Moreover, the present invention adopts a multi-strategy integrated architecture, integrating three modeling methods: data-driven, physically constrained, and microclimate adaptive, adjusting weights based on the historical performance of each model under different conditions through a dynamic weight allocation mechanism. Finally, the present invention has engineering application value, meeting real-time prediction requirements; the lightweight model design based on LightGBM ensures high computational efficiency while maintaining prediction accuracy.

[0183] Example 2:

[0184] Based on the same inventive concept, embodiments of the present invention also provide a multi-strategy integrated prediction system for wind speed in wind farms. This system includes five main functional modules: a data fusion module, a feature engineering module, a microclimate identification module, an adaptive modeling module, and an integrated prediction module. Using the multi-strategy integrated prediction method for wind speed in wind farms as described in Embodiment 1, refer to… Figure 7 As shown, this paper illustrates a four-layer system architecture design comprising a data layer, a processing layer, a model layer, and an application layer. The data layer contains SCADA operational data and multi-source meteorological observation data; the processing layer includes a data fusion module, a feature engineering module, and a microclimate identification module; the model layer includes an adaptive modeling module and an integrated prediction module; and the application layer outputs wind speed prediction results to provide technical support for wind farm operation decisions, power generation assessment, and grid dispatch.

[0185] Specifically, the system includes:

[0186] The data fusion module is used to acquire operational data from the wind farm monitoring and data acquisition system and multi-source meteorological observation data, and to perform time alignment and data cleaning.

[0187] The feature engineering module is used to construct multidimensional meteorological data features based on the processed operational data and multi-source meteorological observation data; the multidimensional meteorological data features are a comprehensive feature space that includes temporal features, physical features and microclimate features;

[0188] The microclimate identification module is used to cluster the multidimensional meteorological data features using the K-means clustering algorithm to automatically identify the microclimate pattern type;

[0189] The adaptive modeling module is used to build LightGBM prediction models for each microclimate type;

[0190] An integrated prediction module is used to fuse the prediction results of a data-driven model, a microclimate adaptive model, and a physical constraint model, and generate the final wind speed prediction value through dynamic weight allocation. The data-driven model is trained on the complete dataset of the data fusion module and the feature engineering module. The microclimate adaptive model is trained on the microclimate identification module and the adaptive modeling module. The physical constraint model is obtained by combining the feature engineering module with the wind profile law, integrating physical knowledge into the machine learning model.

[0191] The wind speed multi-strategy integrated prediction system for wind farms provided in this embodiment of the invention produces the same technical effects as the aforementioned method embodiment. For the sake of brevity, any parts not mentioned in this embodiment can be referred to the corresponding content in the aforementioned method embodiment, and will not be repeated here.

[0192] Example 3:

[0193] This invention provides another electronic device, comprising:

[0194] At least one processor; and a memory communicatively connected to said at least one processor;

[0195] The memory stores instructions that can be executed by the at least one processor, which are executed by the at least one processor to enable the at least one processor to execute the wind farm wind speed multi-strategy integrated prediction method of Embodiment 1.

[0196] Reference Figure 8 As shown, the device may include: a processor 10, a communication interface 11, a memory 12, and a communication bus 13, wherein the processor 10, the communication interface 11, and the memory 12 communicate with each other via the communication bus 13. The processor 10 can call logical instructions in the memory 12 to execute a multi-strategy integrated prediction method for wind speed in a wind farm, the method including:

[0197] S1. Acquire operational data from the wind farm monitoring and data acquisition system and multi-source meteorological observation data, and perform time alignment and data cleaning.

[0198] S2. Construct multidimensional meteorological data features based on the processed operational data and multi-source meteorological observation data; the multidimensional meteorological data features are a comprehensive feature space that includes temporal features, physical features and microclimate features;

[0199] S3. The K-means clustering algorithm is used to perform cluster analysis on the multidimensional meteorological data features to automatically identify the microclimate model type;

[0200] S4. Construct a LightGBM prediction model for each microclimate type;

[0201] S5. The prediction results of the data-driven model, the microclimate adaptive model, and the physical constraint model are integrated and the final wind speed prediction value is generated through dynamic weight allocation. The data-driven model is obtained by training on the complete dataset in steps S1 and S2. The microclimate adaptive model is obtained by training in steps S3 and S4. The physical constraint model is obtained by integrating physical knowledge into the machine learning model in step S2, combined with the wind profile law.

[0202] In some embodiments, the processor 10 may be composed of integrated circuits, such as a single packaged integrated circuit or multiple integrated circuits packaged with the same or different functions, including combinations of one or more central processing units (CPUs), microprocessors, digital processing chips, graphics processors and various control chips.

[0203] Any content or technical means not mentioned in the embodiments of this invention can be obtained by referring to the prior art. This disclosure does not limit the scope of the invention and therefore will not be elaborated further.

[0204] Example 4:

[0205] This invention also provides a storage medium storing instructions. When the instructions are executed on a terminal, they can realize a wind farm wind speed multi-strategy integrated prediction method as described in Embodiment 1.

[0206] Computer-readable storage media can be, for example, but not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatuses, or devices, or any combination thereof. More specific examples (a non-exhaustive list) of computer-readable storage media include: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM), registers, hard disks, optical fibers, portable...

[0207] Compact Disc Read-Only Memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof, or any other form of computer-readable storage medium known in the art. An exemplary storage medium is coupled to a processor, enabling the processor to read information from and write information to the storage medium. Of course, the storage medium can also be a component of the processor. The processor and the storage medium can reside in an Application-Specific Integrated Circuit (ASIC). In this embodiment, the computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0208] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to the method section.

[0209] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A multi-strategy integrated prediction method for wind speed in wind farms, characterized in that, Includes the following steps: S1. Acquire operational data from the wind farm monitoring and data acquisition system and multi-source meteorological observation data, and perform time alignment and data cleaning. S2. Construct multidimensional meteorological data features based on the processed operational data and multi-source meteorological observation data; the multidimensional meteorological data features are a comprehensive feature space that includes temporal features, physical features and microclimate features; S3. The K-means clustering algorithm is used to perform cluster analysis on the multidimensional meteorological data features to automatically identify the microclimate model type; S4. Construct a LightGBM prediction model for each microclimate type; S5. The prediction results of the data-driven model, the microclimate adaptive model, and the physical constraint model are integrated and the final wind speed prediction value is generated through dynamic weight allocation. The data-driven model is obtained by training on the complete dataset in steps S1 and S2. The microclimate adaptive model is obtained by training in steps S3 and S4. The physical constraint model is obtained by integrating physical knowledge into the machine learning model in step S2, combined with the wind profile law.

2. The wind farm wind speed multi-strategy integrated prediction method according to claim 1, characterized in that, Step S1 specifically includes: S11. Obtain wind turbine operation data from the wind farm monitoring and data acquisition system, including average wind speed, standard deviation of wind speed, active power and nacelle temperature; S12. Obtain multi-source meteorological observation data through an open meteorological data application programming interface, including temperature, wind speed, relative humidity, and sea level pressure at different altitudes; S13. Perform time synchronization processing on data sources with different frequencies; S14. Perform outlier detection, missing value imputation, and data quality control.

3. The wind farm wind speed multi-strategy integrated prediction method according to claim 1, characterized in that, Step S2 specifically includes: S21. Extract historical wind speed lag variables from 1 hour to 24 hours, and calculate the statistical characteristics of mean, standard deviation, maximum and minimum values ​​within different time windows; S22. Calculate air density, wind energy density, turbulence intensity, and wind direction decomposition characteristics based on wind energy engineering theory; S23. Calculate the comfort index, temperature change rate, and interaction characteristics of meteorological elements; S24. Use the Z-score standardization method to normalize all features.

4. The multi-strategy integrated prediction method for wind speed in a wind farm according to claim 1, characterized in that, Step S3 specifically includes: S31. Select temperature, wind speed, relative humidity, seasonal coding features, and time period features as clustering inputs; S32. Determine the optimal number of clusters using the silhouette coefficient method; S33. Automatically identify microclimate patterns by performing the K-means clustering algorithm through minimizing the objective function; S34 uses principal component analysis for dimensionality reduction visualization to verify the effectiveness of the clustering results.

5. The wind farm wind speed multi-strategy integrated prediction method according to claim 1, characterized in that, Step S4 specifically includes: S41. Based on the microclimate clustering results, group the training data by type, and train a LightGBM regressor independently for each group of data. S42. Optimize the hyperparameters of each LightGBM regressor using a grid search combined with cross-validation. S43. Use the nearest neighbor classifier to identify the microclimate type for new input samples; S44. Select the corresponding prediction model based on the identified microclimate type for prediction.

6. The multi-strategy integrated prediction method for wind speed in a wind farm according to claim 1, characterized in that, Step S5 specifically includes: S51. Construct the basic LightGBM data-driven model, the microclimate adaptive MA-LightGBM model, and the physical constraint Physics-LightGBM model respectively. S52. Calculate the dynamic weighting coefficients based on the historical prediction performance of each model within the sliding window; S53. The final prediction result is generated by dynamically weighted fusion of three models.

7. A multi-strategy integrated prediction system for wind speed in wind farms, characterized in that, The system, employing the multi-strategy integrated prediction method for wind speed in wind farms as described in any one of claims 1-6, provides wind speed predictions for wind farms. The system comprises: The data fusion module is used to acquire operational data from the wind farm monitoring and data acquisition system and multi-source meteorological observation data, and to perform time alignment and data cleaning. The feature engineering module is used to construct multidimensional meteorological data features based on the processed operational data and multi-source meteorological observation data; the multidimensional meteorological data features are a comprehensive feature space that includes temporal features, physical features and microclimate features; The microclimate identification module is used to cluster the multidimensional meteorological data features using the K-means clustering algorithm to automatically identify the microclimate pattern type; The adaptive modeling module is used to build LightGBM prediction models for each microclimate type; An integrated prediction module is used to fuse the prediction results of a data-driven model, a microclimate adaptive model, and a physical constraint model, and generate the final wind speed prediction value through dynamic weight allocation. The data-driven model is trained on the complete dataset of the data fusion module and the feature engineering module. The microclimate adaptive model is trained on the microclimate identification module and the adaptive modeling module. The physical constraint model is obtained by combining the feature engineering module with the wind profile law, integrating physical knowledge into the machine learning model.

8. A computer-readable storage medium storing a computer program thereon, characterized in that, When the computer program is executed by the processor, it implements the wind farm wind speed multi-strategy integrated prediction method as described in any one of claims 1-6.

9. An electronic device, characterized in that, The method includes a processor and a memory, the memory storing machine-executable instructions that can be executed by the processor, the processor executing the machine-executable instructions to implement the wind farm wind speed multi-strategy integrated prediction method as described in any one of claims 1-6.

Images