Marine wind power fluctuation prediction method and system based on climbing identification

Abstract

The application discloses a sea wind power fluctuation prediction method and system based on climbing identification, relates to the related technical field of wind power prediction, and comprises the following steps: obtaining an excess sample data set based on scene characteristics of a target scene; performing fitting analysis on the excess sample data set, and calculating and obtaining an excess sample change rate curve set; performing time-frequency domain analysis by adopting wavelet transform, and performing clustering analysis; traversing a frequency domain response working condition set, evaluating an attention value of each frequency domain response working condition, and constructing a plurality of working condition mapping models; performing adaptive climbing identification, and performing real-time power fluctuation prediction according to a climbing identification prediction result. The application solves the technical problems that, in the prior art, a complex sea wind power with high nonlinearity and multi-scale characteristics cannot effectively respond to sudden power changes, and thus leads to low sea wind power fluctuation prediction accuracy and poor wind power operation stability, and achieves the technical effects of improving the accuracy of power fluctuation prediction and operation stability.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of fire linkage control, and particularly relates to a sea wind power fluctuation prediction method and system based on climbing identification. BACKGROUND

[0002] The output power of sea wind turbines is significantly volatile due to the influence of complex marine environment, wind speed change and meteorological conditions, etc. The power fluctuation not only affects the stability of the power grid, but also has an adverse effect on the operation management, economic benefits and equipment life of the wind farm. How to accurately predict the power fluctuation of sea wind power has become a key problem that needs to be solved in the current wind power field. At present, the prediction methods for wind power fluctuation are mainly based on statistical models and time series analysis, which can capture the relationship between wind speed change and power output to a certain extent. However, when dealing with the complex scene of sea wind power with high nonlinearity and multi-scale characteristics, the prediction accuracy and adaptability are still limited, especially in dealing with sudden large-scale power fluctuations, which affects the stability and reliability of sea wind power operation.

[0003] Therefore, in the related art, it is difficult to effectively respond to sudden power changes in the face of complex sea wind power with high nonlinearity and multi-scale characteristics, resulting in low prediction accuracy of sea wind power fluctuation and poor stability of wind power operation. SUMMARY

[0004] The present application provides a sea wind power fluctuation prediction method and system based on climbing identification, which solves the technical problem of low prediction accuracy of sea wind power fluctuation and poor stability of wind power operation in the prior art in the face of complex sea wind power with high nonlinearity and multi-scale characteristics, and achieves the technical effect of improving the accuracy of power fluctuation prediction and operation stability.

[0005] The present application provides a sea wind power fluctuation prediction method based on climbing identification, which comprises: collecting excess samples based on the scene characteristics of a target scene to obtain an excess sample data set; performing fitting analysis on the excess sample data set, and calculating a corresponding excess sample change rate curve set based on the fitting analysis result; performing time-frequency domain analysis on the excess sample change rate curve set using wavelet transform, extracting multi-dimensional frequency domain features therefrom, and performing clustering analysis on the multi-dimensional frequency domain features to output a clustering result as a frequency domain response working condition set; traversing the frequency domain response working condition set, evaluating the attention value of each frequency domain response working condition in combination with a statistical analysis method, and corresponding constructing a plurality of working condition mapping models, wherein the working condition mapping model is marked with an applicable interval; performing adaptive climbing identification based on the plurality of working condition mapping models, and performing real-time power fluctuation prediction according to the climbing identification prediction result.

[0006] In a possible implementation, the offshore wind power fluctuation prediction method based on climbing identification further performs the following processing: analyzing the excess sample data set to obtain corresponding sample data typical fluctuation indicators; and performing adaptive oversampling on the excess sample data set according to the sample data typical fluctuation indicators.

[0007] In a possible implementation, the offshore wind power fluctuation prediction method based on climbing identification further performs the following processing: determining the sizes of segmented windows corresponding to different indicator dimensions according to the time sequence data characteristics of a target scenario; performing filtering processing on the data of multiple indicator dimensions in the excess sample data set based on a sliding window method and in combination with the sizes of the segmented windows; fitting the multiple sets of segmented data after the filtering processing into continuous curves to form the excess sample rate of change curve set.

[0008] In a possible implementation, the offshore wind power fluctuation prediction method based on climbing identification further performs the following processing: analyzing the curves to determine frequency centers, energy distributions, and spectral entropies in combination with wavelet transform, and outputting the multiple-dimensional frequency domain characteristics; calculating the principal component contribution degrees of each frequency domain characteristic in the multiple-dimensional frequency domain characteristics to climbing identification based on a principal component analysis method; sequencing each frequency domain characteristic according to the principal component contribution degrees, and selecting the first K frequency domain characteristics to perform clustering analysis and obtain the clustering results; and defining the frequency domain response working conditions and the working condition intervals corresponding to each clustering cluster based on the clustering results.

[0009] In a possible implementation, the offshore wind power fluctuation prediction method based on climbing identification further performs the following processing: determining the climbing identification accuracy rates and the occurrence frequencies of the working condition categories based on the clustering results; weighting the climbing identification accuracy rates and the occurrence frequencies to determine the attention value of each frequency domain response working condition, and configuring the model construction constraint corresponding to each frequency domain response working condition according to the attention value; and constructing and training the plurality of working condition mapping models according to the model construction constraint.

[0010] In a possible implementation, the offshore wind power fluctuation prediction method based on climbing identification further performs the following processing: iteratively constructing a basic mapping model and storing the basic mapping model in a basic model library; repeatedly calling the basic model library according to the model construction constraint to generate a plurality of working condition basic model groups; performing random crossover on the plurality of working condition basic model groups, and performing supervised training and ensemble learning on the plurality of working condition basic model groups after the crossover based on the multiple-dimensional frequency domain characteristics and the excess sample data set to obtain the plurality of working condition mapping models.

[0011] In a possible implementation, the offshore wind power fluctuation prediction method based on climbing identification further performs the following processing: a model complexity constraint for limiting the model size of a single base mapping model; a model number constraint for limiting the number of models in the working condition base model group; and a model structure constraint for limiting the structure category of the called base mapping model.

[0012] The application further provides an offshore wind power fluctuation prediction system based on climbing identification, comprising: an excessive sample collection module for collecting excessive samples based on the scene features of a target scene, and obtaining an excessive sample dataset; a dataset fitting analysis module for performing fitting analysis on the excessive sample dataset, and calculating a corresponding excessive sample change rate curve set based on the fitting analysis result; a frequency domain response working condition set output module for performing time-frequency domain analysis on the excessive sample change rate curve set by using wavelet transform, extracting multi-dimensional frequency domain features therefrom, and performing clustering analysis on the multi-dimensional frequency domain features, and outputting the clustering result as a frequency domain response working condition set; a working condition mapping model construction module for traversing the frequency domain response working condition set, evaluating the attention value of each frequency domain response working condition in combination with a statistical analysis method, and constructing a plurality of working condition mapping models correspondingly, wherein the working condition mapping model is marked with an applicable interval; and a power fluctuation prediction module for performing adaptive climbing identification based on the plurality of working condition mapping models, and performing real-time power fluctuation prediction according to the climbing identification prediction result.

[0013] The offshore wind power fluctuation prediction method and system based on climbing identification provided in the application obtain an excessive sample dataset based on the scene features of a target scene; perform fitting analysis on the excessive sample dataset, and calculate a corresponding excessive sample change rate curve set; perform time-frequency domain analysis by using wavelet transform, and perform clustering analysis; traverse the frequency domain response working condition set, evaluate the attention value of each frequency domain response working condition, and construct a plurality of working condition mapping models; perform adaptive climbing identification, and perform real-time power fluctuation prediction according to the climbing identification prediction result. The technical problems that the existing technology cannot effectively cope with sudden power changes of complex offshore wind power with high nonlinearity and multi-scale characteristics, and the offshore wind power fluctuation prediction precision is low and the wind power operation stability is poor are solved, and the technical effects of improving the accuracy of power fluctuation prediction and the operation stability are achieved. BRIEF DESCRIPTION OF DRAWINGS

[0014] In order to more clearly illustrate the technical solutions of the embodiments of the present disclosure, the drawings of the embodiments of the present disclosure will be briefly introduced below. In the present application, a flow chart is used to illustrate the operations performed by the system according to the embodiments of the present application. It should be understood that the foregoing or the following operations are not necessarily performed in sequence. On the contrary, various steps can be processed in reverse order or simultaneously according to needs. At the same time, other operations can be added to these processes, or one or more steps of operations can be removed from these processes.

[0015] Figure 1 The flow chart of the offshore wind power fluctuation prediction method based on climbing identification provided by the embodiments of the present application.

[0016] Figure 2 The structure schematic diagram of the offshore wind power fluctuation prediction system based on climbing identification provided by the embodiments of the present application.

[0017] Legend: Excess sample acquisition module 10, data set fitting analysis module 20, frequency domain response working condition set output module 30, working condition mapping model construction module 40, power fluctuation prediction module 50. DETAILED DESCRIPTION

[0018] The above description is only a summary of the technical solutions of the present application. In order to more clearly understand the technical means of the present application, the content of the specification can be implemented, and in order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the following specific embodiments of the present application are described.

[0019] In order to make the purposes, technical solutions and advantages of the present application more clear, the present application will be further described in detail below with reference to the drawings. The described embodiments should not be regarded as limiting the present application. All other embodiments obtained by those of ordinary skill in the art without making creative labor are within the scope of protection of the present application.

[0020] In the following description, "some embodiments" are referred to, which describe a subset of all possible embodiments, but it can be understood that "some embodiments" can be the same subset or different subsets of all possible embodiments, and can be combined with each other without conflict, and the term "first\second" referred to only distinguishes similar objects, and does not represent a specific order for the objects. The terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product or server including a series of steps or units does not have to be limited to those steps or units clearly listed, but can include other steps or modules not clearly listed or inherent to these processes, methods, products or devices. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as understood by those skilled in the art to which the present application belongs. The terms used herein are only for the purpose of describing the embodiments of the present application.

[0021] The embodiments of the present application provide a sea wind power fluctuation prediction method based on climbing recognition, as shown in the formula (1), the method comprises the steps of: Figure 1

[0022] Step S100, based on the scene characteristics of the target scene, excessive sample collection is carried out, and an excessive sample data set is obtained.

[0023] Preferably, the scene characteristics of the target scene refer to specific characteristics closely related to the actual operation environment and working conditions of the sea wind farm, which can include geographical location characteristics (the sea area position, altitude, marine climate characteristics, etc. where the wind farm is located), meteorological conditions (such as wind speed, wind direction, temperature and humidity, air pressure, wave height, sea surface temperature, etc. Environmental parameters changing with time), wind turbine characteristics (operation parameters, power curve, load response characteristics, etc. of different types of units), grid connection conditions (load demand, frequency stability, grid connection scheduling strategy, etc. of the power grid); According to these scene characteristics, excessive sample collection is carried out, that is, on the basis of conventional data collection, more diversified, extreme or boundary condition data is collected to cover more possible working conditions. Specifically, data can be collected from historical operation data, real-time monitoring data, simulation data, etc. and data in extreme scenes such as abnormal fluctuation, strong wind weather and equipment failure is collected to ensure that the model also has good prediction ability in complex environment, including wind speed data (wind speed change at different time points), power output data (actual power generation data of wind turbine under different environmental conditions), temperature and humidity data (change of environmental temperature and humidity may affect air density), air pressure data (air pressure change will have certain influence on aerodynamic characteristics) and wave height, wind direction angle change, grid load data, etc. to form an excessive sample data set to more comprehensively describe the wind farm operation condition, and ensure that the power fluctuation prediction model has high reliability and adaptability under different working conditions. ​

[0024] Step S200, fitting analysis is performed on the excess sample data set, and the corresponding excess sample change rate curve set is calculated based on the fitting analysis result.

[0025] Preferably, based on the collected excess sample data (such as wind speed, power output, temperature and humidity, air pressure, etc.), fitting analysis is performed to obtain the fitting analysis result, so as to reveal the internal correlation and change trend between different variables. Specifically, the excess sample data is preprocessed, including cleaning, removing outliers and noise data, and standardization processing. The relationship between wind speed and power, meteorological data and fluctuation trend is fitted by using a regression model, or the time series characteristics of power output are fitted by using time series analysis such as ARIMA model, exponential smoothing, etc. The dynamic correlation characteristics between different variables are extracted, and the mapping relationship between wind speed, meteorological conditions and power output fluctuation is particularly focused on. Then, the rate of change of the excess sample data with time (i.e. the change rate) is further calculated according to the fitting analysis result, and these rate data are represented in the form of a curve to form an excess sample change rate curve set. For example, the difference between adjacent time points is calculated using the difference method, and the derivative of the fitting curve is calculated to obtain the continuous change rate, which represents the instantaneous change rate of power with time. Similarly, the change rates of wind speed, air pressure, temperature and humidity, and other variables are calculated to form a multi-dimensional change rate curve set, which more directly reflects the fluctuation characteristics of the data in different time periods, especially the mutation points, climbing sections or severe fluctuation regions, which helps to improve the sensitivity of the prediction model to sudden fluctuations.

[0026] Further, step S200 further includes step S201 of analyzing the excess sample data set to obtain corresponding sample data typical fluctuation indicators; and step S202 of performing adaptive oversampling on the excess sample data set according to the sample data typical fluctuation indicators.

[0027] Preferably, statistical analysis is performed on the data of the excess sample data set to quantify the fluctuation characteristics of the data at different time periods or operating conditions. Specifically, the excess sample data is segmented according to time sequence or operating condition characteristics (e.g., by hour, by wind speed interval, etc.), and for each data segment, key indicators that can reflect the instability or fluctuation degree of the data are extracted, referred to as typical fluctuation indicators, which can include variance (measuring the degree of data deviation from the mean value, the greater the variance, the more intense the data fluctuation), standard deviation (square root of variance, more intuitively reflecting the fluctuation range of the data, the greater the standard deviation, the more unstable the data fluctuation), skewness (describing the symmetry of data distribution, high skewness indicating data distribution skewed to one direction, which can exist abnormal fluctuation), kurtosis (measuring the sharpness of data distribution, high kurtosis indicating the existence of sudden extreme fluctuation in data), and moving average difference (calculating local mean change through sliding window, suitable for detecting short-term intense fluctuation), etc. The data segments with intense fluctuation (e.g., standard deviation higher than a certain threshold) are analyzed to identify potential key fluctuation areas.

[0028] Preferably, the sampling density is automatically adjusted according to the data fluctuation, the sampling frequency is increased in data regions with large fluctuations to obtain more data points, and the learning ability of the model for complex fluctuation patterns is improved, i.e., according to the fluctuation indicators (such as variance, standard deviation), each data segment is assigned a sampling weight, the regions with large fluctuations (high variance, high standard deviation) have increased sampling frequency to obtain more samples to capture rapid changes, and the regions with small fluctuations (low variance, low standard deviation) have reduced sampling frequency to reduce redundant data and optimize computational efficiency. For example, in regions with intense fluctuations, interpolation techniques such as linear interpolation and spline interpolation are used to generate new data points; for data sparse regions, synthetic data can be generated through SMOTE algorithm to enhance sample diversity; for high fluctuation regions, overlapping sliding windows are applied to extract more fine-grained data segments, thereby improving the prediction ability for complex power fluctuation patterns and effectively improving the data processing efficiency and model generalization performance.

[0029] Further, step S200 further comprises step S210 of determining the segment window size corresponding to different indicator dimensions according to the time series data characteristics of the target scene; step S220 of filtering the data of multiple indicator dimensions in the excess sample data set based on the sliding window method combined with the segment window size; and step S230 of fitting the filtered multiple sets of segmented data into a continuous curve to form the excess sample change rate curve set.

[0030] Preferably, for different index dimensions of the target scene (such as wind speed, power, temperature and humidity, air pressure, etc.), the corresponding segmented window size is set according to the fluctuation characteristics and data change frequency, so as to more accurately capture the data trend and fluctuation characteristics. The segmented window size determines the analysis granularity of the data in each sliding window. Specifically, short-term rapid fluctuation indicators (such as wind speed and power output) have high-frequency change characteristics, and a smaller window size (such as 5-10 minutes) is suitable for capturing sudden fluctuations or climbing changes. Long-term stable change indicators (such as temperature and humidity, air pressure) change relatively smoothly, and a larger window size (such as 30 minutes to 1 hour) is suitable for reducing noise effects. For example, based on autocorrelation function (ACF) analysis, the periodicity of the data is identified to determine the optimal window length, or wavelet analysis is used to extract frequency characteristics, and the appropriate time window is matched. For example, the wind speed window size is 10 minutes (adapted to the rapid change characteristics of wind speed), the power output window size is 15 minutes (balanced real-time and fluctuation capture ability), and the temperature and air pressure window size is 60 minutes (data fluctuation is smooth, and the processing frequency is reduced).

[0031] Preferably, the sliding window method is a technique commonly used in time series data analysis. By sliding a fixed-size window over the data sequence, data segments are extracted and processed step by step. Filtering is used to smooth the data, remove noise, and retain the main trend or fluctuation characteristics of the data. Specifically, a certain window size is selected, and the step size of the window, i.e. the time interval of each slide, is defined. The window is slid over the data sequence, and the data subset in each window is extracted. Then, the average value of the data in each window is calculated to smooth the data fluctuations. Alternatively, different weights are assigned to data at different time points to enhance the influence of recent data. Filtering algorithms are applied independently or jointly to each indicator (wind speed, power, temperature and humidity, air pressure, etc.) to ensure consistent processing of fluctuation characteristics between different data dimensions, filter out high-frequency noise in the data, retain key trends and fluctuation information, and reduce data fluctuations to make subsequent fitting models more stable. Finally, the filtered segmented data is fitted, and the change rate curves of different indicators (such as wind speed, power, air pressure, etc.) are integrated to generate a continuous change rate curve set. Each curve represents the dynamic change characteristics of a specific indicator, not only retaining the key dynamic characteristics of the original data, but also greatly improving the accuracy and robustness of power fluctuation prediction.

[0032] In step S300, wavelet transform is used to perform time-frequency domain analysis on the excess sample change rate curve set, extract multi-dimensional frequency domain features, and perform clustering analysis on the multi-dimensional frequency domain features. The clustering result is a frequency domain response working condition set.

[0033] Preferably, the wavelet transform is a powerful signal processing tool that can simultaneously analyze the time domain and frequency domain characteristics of the signal, and is suitable for processing non-stationary signals such as wind power fluctuation data. That is, the wavelet transform is used for local analysis in different time scales and frequency ranges to capture the characteristics of sudden changes, ramping and irregular changes in power fluctuations. Specifically, according to the offshore wind power fluctuation signal with sudden change and ramping characteristics, a suitable wavelet function (such as db4, db6) is selected to decompose the rate of change curve into different frequency levels to obtain the approximation coefficients (reflecting the low-frequency trend and capturing the long-term stable change) and the detail coefficients (reflecting the high-frequency fluctuation and capturing the short-term fluctuation and sudden change). According to the needs, some frequency bands are retained or filtered out, and the signal is reconstructed to highlight the key frequency components, remove noise, and retain features valuable for prediction, so as to decompose the complex power fluctuation signal into sub-signals of different scales and reveal the dynamic characteristics of power changes in different time scales and frequency ranges.

[0034] Preferably, after wavelet transform, key parameters reflecting signal characteristics are extracted from different frequency levels as multi-dimensional frequency domain features to describe the frequency distribution, energy distribution, fluctuation intensity, etc. of power fluctuations, which may include energy features (calculating the energy of different wavelet decomposition layers to reflect the intensity of power fluctuations in different frequency bands), frequency band energy ratio (the proportion of energy in different frequency bands to help determine the main frequency range of fluctuations), entropy value features (features describing signal complexity and uncertainty, such as wavelet entropy), instantaneous frequency (describing the local frequency change of the signal at different time points to reveal sudden changes or abnormal fluctuations), peak frequency and bandwidth (peak frequency reflects the main fluctuation frequency, and bandwidth represents the frequency distribution range of fluctuations), etc.

[0035] Preferably, the multi-dimensional frequency domain features are subjected to cluster analysis, i.e. data with similar features are divided into different groups or categories to identify different frequency domain response working conditions. Specifically, the clustering method (such as K-means) and the number of clusters k are determined, the clustering effect is evaluated by indicators such as silhouette coefficient, the multi-dimensional frequency domain features are input into the clustering model, the clustering label of each sample is output, and the clustering result is visually displayed through PCA (principal component analysis) or t-SNE dimensionality reduction visualization. Then, the frequency domain response working condition set is output, i.e. based on the clustering analysis result, the power fluctuation data is divided into several representative working condition categories, each category represents a specific power fluctuation mode or operating state, thereby realizing deep mining and pattern recognition of offshore wind power fluctuation data, which is helpful for accurate and real-time power fluctuation prediction.

[0036] Further, step S300 further comprises step S310 of analyzing the curve to determine a frequency center, an energy distribution, and a spectrum entropy in combination with wavelet transform, and outputting the multi-dimensional frequency domain features; step S320 of calculating and obtaining a principal component contribution degree of each frequency domain feature in the multi-dimensional frequency domain features to the climbing identification based on a principal component analysis method; step S330 of serializing each frequency domain feature according to the principal component contribution degree and selecting a first preset number K of frequency domain features for cluster analysis to obtain the cluster result; and step S340 of defining a frequency domain response working condition and a working condition interval corresponding to each cluster in the cluster result.

[0037] Preferably, the wavelet transform is used to perform multi-scale decomposition on the excess sample change rate curve, that is, time domain signals (such as power fluctuation change rate curves) are decomposed into different frequency ranges, from which key frequency domain features are extracted and determined, including a frequency center (representing a dominant frequency of energy concentration of the signal), an energy distribution (describing the distribution of different frequency components in the total energy of the signal), and a spectrum entropy (used to measure the complexity and uncertainty of the signal, and the greater the entropy value, the more complex the signal and the more violent the fluctuation), and then output as multi-dimensional frequency domain features; the principal component analysis is used to identify the most representative features in high-dimensional data, such as projecting the data into a new coordinate system to find the principal component that explains the most variance of the data, that is, to calculate the contribution degree of each frequency domain feature to the climbing identification, and to screen out the most important features. Specifically, the data is standardized, a covariance matrix of the multi-dimensional frequency domain features is constructed to reflect the correlation between the features, the covariance matrix is then subjected to eigenvalue decomposition to obtain eigenvalues and eigenvectors (principal component directions), wherein the greater the eigenvalue, the stronger the principal component in explaining the data variance, and then the variance contribution rate of each principal component is calculated, and at the same time, the load of each original frequency domain feature on the principal component is evaluated to quantify the principal component contribution degree.

[0038] Preferably, after obtaining the contribution degree of the frequency domain feature, the features are sorted according to the contribution degree sequence, and the top K most representative features are selected for cluster analysis to reduce the calculation complexity and improve the clustering effect. Specifically, the contribution degrees calculated by PCA are sorted from high to low to form a feature priority sequence, K features with a preset number are determined to form a feature matrix after dimension reduction, the data is divided into k clusters, the variance within the cluster is minimized, and then the clustering effect is evaluated using indicators such as the silhouette coefficient to determine the optimal clustering number and obtain the clustering result. Finally, based on each cluster in the clustering result, the frequency domain response working condition and the working condition interval are defined. Specifically, the typical working condition state is summarized by classifying and summarizing data samples with similar frequency domain features. Each cluster represents a specific power fluctuation mode, and the corresponding working condition interval is defined, for example, a smooth working condition (low-frequency fluctuation dominates, energy is concentrated in the low-frequency band, power fluctuation is smooth, suitable for normal operation state), a climbing working condition (mid-frequency fluctuation is enhanced, frequency change is relatively sharp, reflecting power climbing or rapid change state), an abnormal fluctuation working condition (high-frequency component accounts for a large proportion, entropy value is high, which may correspond to equipment failure, extreme weather or other abnormal conditions), etc. Then, according to the characteristic value range of the frequency center and the energy distribution, the numerical interval of each working condition is defined, the frequency center interval (Hz) is 0.1-0.3 Hz (smooth working condition), 0.4-0.6 Hz (climbing working condition), the energy distribution ratio is low-frequency energy ratio > 80% (smooth), mid-high frequency ratio > 50% (climbing), the spectral entropy is entropy value < 1.5 (smooth), entropy value > 2.5 (abnormal working condition), which realizes the processing from the original power fluctuation data to the typical working condition recognition, and is helpful for realizing real-time power fluctuation prediction and adaptive control.

[0039] Step S400, traverse the frequency domain response working condition set, combine statistical analysis method to evaluate the attention value of each frequency domain response working condition, and correspondingly build a plurality of working condition mapping models, wherein the working condition mapping model is marked with an applicable interval.

[0040] Preferably, the frequency domain response working conditions are traversed and analyzed, data samples, frequency domain features (such as frequency center, energy distribution, frequency spectrum entropy) and power fluctuation patterns related thereto are extracted, and then a statistical analysis method is combined to evaluate the attention value of each frequency domain response working condition, that is, to quantify the importance of each frequency domain response working condition in power fluctuation prediction. Specifically, the attention value of each frequency domain response working condition is calculated by weighting the attention evaluation indexes such as fluctuation intensity, abnormal probability, fluctuation frequency and energy concentration degree, and then a plurality of working condition mapping models are constructed based on different frequency domain response working conditions to describe the mapping relationship between input variables (such as wind speed, air pressure, etc.) and power fluctuation under a specific working condition. Specifically, according to the working condition characteristics, a suitable model type (such as a regression model, a random forest, a neural network, an LSTM model, etc.) is selected, the working condition data with high attention value is used for model training to ensure that the data covers the main fluctuation characteristics of the working condition, and then a plurality of working condition mapping models are obtained, wherein each working condition mapping model is marked with an applicable interval to represent the applicable range in actual application, which helps to automatically switch the most suitable prediction model under different working conditions, thereby significantly improving the prediction accuracy of offshore wind power fluctuation and the reliability of system operation.

[0041] Further, step S400 further comprises step S410 of statistically analyzing and determining the ramp recognition accuracy and occurrence frequency of the working condition categories based on the clustering results; step S420 of weighting the ramp recognition accuracy and the occurrence frequency to determine the attention value of each frequency domain response working condition, and configuring the model construction constraint corresponding to each frequency domain response working condition according to the attention value; and step S430 of constructing and training a plurality of working condition mapping models according to the model construction constraint.

[0042] Preferably, according to the clustering results, that is, in each cluster (working condition category), the identification results of the climbing events are statistically analyzed to calculate the climbing identification accuracy (true rate and false positive rate) and the occurrence frequency of the working condition category, wherein the climbing identification accuracy measures the ability of the model to accurately identify the rapid rise or fall of power (i.e., the climbing phenomenon) under different working conditions, the occurrence frequency indicates the probability of a specific frequency domain response working condition appearing in historical data or real-time data, reflecting its representativeness and influence range, the true rate refers to the proportion of correctly identified climbing events to all actual climbing events, and the false positive rate refers to the proportion of incorrectly identified smooth events as climbing events, and then the climbing identification accuracy and the occurrence frequency corresponding to each working condition are output; the climbing identification accuracy and the occurrence frequency are weighted to calculate the attention value of each frequency domain response working condition, and the model construction constraints of each frequency domain response working condition are configured according to different attention working conditions, that is, the parameter limits of model design and training are set to optimize the model performance and the allocation of computing resources, such as configuring the model type, data volume ratio, learning rate, iteration number, etc. according to the working condition category and the attention value, with the progress of model training, the model performance is monitored in real time, and the model constraint parameters are dynamically adjusted to adapt to the changes in data distribution.

[0043] Preferably, according to the model construction constraints, a plurality of working condition mapping models are constructed for different working conditions to adapt to complex power fluctuation scenarios, that is, the data set is divided according to the working condition to ensure that the training data of each model fully covers its applicable working condition, and the hyperparameters (such as learning rate, batch size, regularization parameter, etc.) are set according to the model construction constraints. For climbing identification, the model evaluation is additionally focused on the climbing detection rate and the missed detection rate, and finally a plurality of working condition mapping models are trained to have high precision, strong adaptability and good real-time performance, which can effectively support the prediction and scheduling control of offshore wind power fluctuations.

[0044] Further, step S430 further comprises step S431 of iteratively constructing a basic mapping model and storing it in a basic model library; step S432 of repeatedly calling the basic model library according to the model construction constraints to generate a plurality of working condition basic model groups; and step S433 of performing random crossover on the plurality of working condition basic model groups and performing supervised training and ensemble learning on the crossed plurality of working condition basic model groups based on the multi-dimensional frequency domain features and the excess sample data set to obtain a plurality of working condition mapping models.

[0045] Preferably, the basic mapping model is a basic prediction model constructed based on initial data and simple algorithms for different frequency domain response conditions and stored in a basic model library for subsequent calling, combination and optimization. Specifically, different algorithms (such as linear regression, decision tree, random forest, support vector machine, etc.) are used to construct the basic model, and different data division strategies (such as Bootstrap sampling), feature selection methods, model structures, etc. are used to increase the diversity of the model. The trained basic model (including model parameters, training data configuration, evaluation index, etc.) is stored in the basic model library. According to the model construction constraints, the models in the basic model library are repeatedly called to generate a plurality of condition basic model groups, that is, a model cluster composed of several model combinations selected from the basic model library, which is used to process specific frequency domain response conditions, ensures the rationality of the model group structure, avoids model redundancy or excessive complexity, and improves the adaptability to complex conditions.

[0046] Preferably, part of the models in different model groups are randomly selected for cross combination (such as model replacement cross or fusion cross) to generate new model groups, increase model diversity and optimize performance, and improve overall prediction performance. Then use the labeled data (such as the actual value of power fluctuation) to train the model to minimize the error between the predicted value and the true value. Specifically, use multi-dimensional frequency domain features (such as frequency center, energy distribution, spectral entropy, etc.) and excess sample data sets as model inputs, and independently supervise the training of each model group to ensure that each model group can fully learn the rules in the data and improve the accuracy of power fluctuation prediction. Then obtain a plurality of condition mapping models based on ensemble learning, for example, take the average or voting result of the prediction results of multiple basic models, which is suitable for reducing model variance and improving stability. Combine the multiple condition basic models after random cross and supervised training into an integrated model to generate multiple high-performance condition mapping models, each of which is suitable for different conditions or scenarios and has stronger prediction ability and adaptability, which can effectively support the accurate prediction and intelligent control of offshore wind power fluctuation.

[0047] Step S432 further includes step A, model complexity constraint, for limiting the model size of a single basic mapping model; step B, model number constraint, for limiting the number of models in the condition basic model group; and step C, model structure constraint, for limiting the structure category of the called basic mapping model.

[0048] Preferably, the model construction constraints include model complexity constraints for limiting the size of individual base models (such as the number of layers, the number of parameters of neural networks, the depth of decision trees, etc.), avoiding overfitting or waste of computing resources caused by overly complex models, for example, limiting the number of layers of neural network models to no more than 3, and the depth of decision trees to no more than 10; model number constraints for limiting the number of models in each operating condition base model group to balance the diversity of models and computing overhead, for example, each model group contains 3-5 base models; model structure constraints for limiting the types of callable base models or combination methods to ensure reasonable structural diversity of the model group, for example, requiring at least one linear model and one nonlinear model in the model group.

[0049] Step S500, adaptive ramping identification is performed based on the plurality of operating condition mapping models, and real-time power fluctuation prediction is performed according to the ramping identification prediction result.

[0050] Preferably, adaptive ramping identification is performed based on the plurality of operating condition mapping models, that is, the most suitable operating condition mapping model is dynamically selected under different operating conditions to accurately identify the rapid rise or fall (i.e., "ramping") of offshore wind power, wherein the ramping event refers to a phenomenon that the power output changes significantly in a short time, which can be rapid rise or rapid fall. Specifically, real-time monitoring data (wind speed, air pressure, temperature and humidity, historical power, etc.) and multi-dimensional frequency domain features (such as frequency center, energy distribution, spectral entropy, etc.) are input into the operating condition mapping model as input data, and the closest frequency domain response operating condition is matched, and then the optimal operating condition mapping model in the base model library is automatically called for ramping identification, including extracting key features (such as power change rate) from the input data, and if the power change rate is greater than a predetermined threshold, it indicates that a ramping event is detected. Then, based on the identification result and model prediction, the power output fluctuation trend of the offshore wind turbine in the future short time is predicted in real time after the ramping identification is completed, specifically, real-time monitoring data (wind speed, power historical data, weather data, etc.), ramping identification results (ramping type, amplitude, duration, etc.), and multi-dimensional frequency domain features (frequency center, energy distribution, spectral entropy, etc.) are input as input, and ARIMA, LSTM (Long Short-Term Memory Network), random forest, etc. are used to build and train prediction models for real-time power fluctuation prediction, which can include short-term power prediction (power fluctuation trend in the next 5 minutes, 10 minutes, 30 minutes) and fluctuation amplitude prediction (predicted maximum, minimum and fluctuation range of power change), etc. When the prediction result shows that the power will soon fluctuate sharply, an early warning signal is automatically triggered to assist grid dispatching and wind farm control, effectively dealing with complex offshore environments and variable wind power operating conditions, and ensuring the safe and efficient operation of the wind farm.

[0051] In the foregoing, reference is made to Figure 1A method for predicting offshore wind power fluctuation based on climbing identification according to an embodiment of the present application is described in detail. Next, a system for predicting offshore wind power fluctuation based on climbing identification according to an embodiment of the present application will be described with reference to Figure 2 A system for predicting offshore wind power fluctuation based on climbing identification according to an embodiment of the present application is described.

[0052] The system for predicting offshore wind power fluctuation based on climbing identification according to an embodiment of the present application is used to solve the technical problem that the existing technology cannot effectively deal with sudden power changes in the face of complex offshore wind power with highly nonlinear and multi-scale characteristics, resulting in low prediction accuracy of offshore wind power fluctuation and poor stability of wind power operation. As shown in Figure 2 The system for predicting offshore wind power fluctuation based on climbing identification includes an excess sample collection module 10, a data set fitting analysis module 20, a frequency domain response working condition set output module 30, a working condition mapping model construction module 40, and a power fluctuation prediction module 50.

[0053] The excess sample collection module 10 is used to collect excess samples based on the scene characteristics of the target scene and obtain an excess sample data set. The data set fitting analysis module 20 is used to perform fitting analysis on the excess sample data set and calculate a corresponding excess sample change rate curve set based on the fitting analysis result. The frequency domain response working condition set output module 30 is used to perform time-frequency domain analysis on the excess sample change rate curve set using wavelet transform, extract multi-dimensional frequency domain features therefrom, and perform clustering analysis on the multi-dimensional frequency domain features to output a clustering result as a frequency domain response working condition set. The working condition mapping model construction module 40 is used to traverse the frequency domain response working condition set, evaluate the attention value of each frequency domain response working condition in combination with a statistical analysis method, and construct a plurality of working condition mapping models correspondingly, wherein the working condition mapping models are marked with applicable intervals. The power fluctuation prediction module 50 is used to perform adaptive climbing identification based on the plurality of working condition mapping models and perform real-time power fluctuation prediction according to the climbing identification prediction result.

[0054] Next, the specific configuration of the data set fitting analysis module 20 will be described in detail. The data set fitting analysis module 20 further includes: parsing the excess sample data set to obtain corresponding sample data typical fluctuation indicators; and performing adaptive oversampling on the excess sample data set according to the sample data typical fluctuation indicators.

[0055] Next, the specific configuration of the data set fitting analysis module 20 will be described in detail. The data set fitting analysis module 20 further comprises: determining the segmentation window size corresponding to different index dimensions according to the time sequence data characteristics of the target scene; filtering the data of multiple index dimensions in the excess sample data set based on the sliding window method combined with the segmentation window size; fitting the multiple sets of segmented data after filtering into continuous curves to form the excess sample change rate curve set.

[0056] Next, the specific configuration of the frequency domain response working condition set output module 30 will be described in detail. The frequency domain response working condition set output module 30 further comprises: analyzing the frequency center, energy distribution and spectral entropy based on wavelet transform, and outputting the multi-dimensional frequency domain features; calculating the principal component contribution of each frequency domain feature in the multi-dimensional frequency domain features to the hill climbing identification based on the principal component analysis method; sequencing each frequency domain feature according to the principal component contribution, and selecting the first K frequency domain features for clustering analysis to obtain the clustering result; defining the frequency domain response working condition and the working condition interval corresponding to each clustering cluster based on the clustering result.

[0057] Next, the specific configuration of the working condition mapping model construction module 40 will be described in detail. The working condition mapping model construction module 40 further comprises: statistically analyzing and determining the hill climbing identification accuracy and occurrence frequency of the working condition category based on the clustering result; weighting the hill climbing identification accuracy and the occurrence frequency to determine the attention value of each frequency domain response working condition, and configuring the model construction constraint corresponding to each frequency domain response working condition according to the attention value; constructing and training multiple working condition mapping models according to the model construction constraint.

[0058] Next, the specific configuration of the working condition mapping model construction module 40 will be described in detail. The working condition mapping model construction module 40 further comprises: iteratively constructing a basic mapping model and storing it in a basic model library; generating multiple working condition basic model groups by repeatedly calling the basic model library according to the model construction constraint; performing random crossover on multiple working condition basic model groups, and performing supervised training and ensemble learning on the multiple working condition basic model groups after the crossover based on the multi-dimensional frequency domain features and the excess sample data set, to obtain multiple working condition mapping models.

[0059] Next, the specific configuration of the working condition mapping model construction module 40 will be described in detail. The working condition mapping model construction module 40 further comprises: a model complexity constraint for limiting the model size of a single basic mapping model; a model number constraint for limiting the number of models in the working condition basic model group; and a model structure constraint for limiting the structure category of the called basic mapping model.

[0060] The offshore wind power fluctuation prediction system based on climbing identification provided by the embodiment of the present application can execute the offshore wind power fluctuation prediction method based on climbing identification provided by any embodiment of the present application, and has the corresponding function modules and beneficial effects of the execution method.

[0061] Although the present application makes various references to certain modules in the system according to the embodiments of the present application, however, any number of different modules can be used and run on the user terminal and / or server, and the various units and modules are only divided according to the functional logic, but are not limited to the above division, as long as the corresponding functions can be realized; in addition, the specific names of the functional units are only for the convenience of mutual differentiation, and are not used to limit the protection scope of the present application.

[0062] The above specific embodiments do not constitute a limitation on the protection scope of the present application. Those skilled in the art should understand that various modifications, combinations and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A method for offshore wind power fluctuation prediction based on climbing identification, characterized in that, The method comprises: Based on the scene characteristics of the target scene, over-sampling is performed to obtain an over-sampling data set; Perform fitting analysis on the over-sampling data set, and calculate the corresponding over-sampling change rate curve set based on the fitting analysis result; Perform time-frequency domain analysis on the over-sampling change rate curve set using wavelet transform, extract multi-dimensional frequency domain features therefrom, and perform clustering analysis on the multi-dimensional frequency domain features, and output the clustering result as a frequency domain response working condition set; Iterate through the frequency domain response working condition set, evaluate the attention value of each frequency domain response working condition by combining statistical analysis method, and correspondingly construct multiple working condition mapping models, wherein the working condition mapping model is marked with an applicable interval; Based on the multiple working condition mapping models, perform adaptive hill climbing identification, and perform real-time power fluctuation prediction according to the hill climbing identification prediction result; Wherein, iterating through the frequency domain response working condition set, combining statistical analysis method to evaluate the attention value of each frequency domain response working condition, and correspondingly constructing multiple working condition mapping models, comprising: Based on the clustering result, statistically analyze and determine the hill climbing identification accuracy and occurrence frequency of the working condition category; Weight the hill climbing identification accuracy and the occurrence frequency to determine the attention value of each frequency domain response working condition, and configure the model construction constraint corresponding to each frequency domain response working condition according to the attention value; According to the model construction constraint, construct and train multiple working condition mapping models; Wherein, according to the model construction constraint, constructing and training multiple working condition mapping models, comprising: Iteratively construct a basic mapping model and store it in a basic model library; According to the model construction constraint, repeatedly call the basic model library to generate multiple working condition basic model groups; Randomly cross multiple working condition basic model groups, and based on the multi-dimensional frequency domain features and the over-sampling data set, supervise the training and integrated learning of the crossed multiple working condition basic model groups to obtain multiple working condition mapping models; Wherein, the model construction constraint comprises: Model complexity constraint for limiting the model size of a single basic mapping model; Model number constraint for limiting the number of models in the working condition basic model group; Model structure constraint for limiting the structure category of the called basic mapping model.

2. The method of claim 1, wherein the method comprises: Before the fitting analysis on the over-sampling data set and the calculation of the corresponding over-sampling change rate curve set based on the fitting analysis result, it further comprises: Analyzing the over-sampling data set to obtain corresponding sample data typical fluctuation indicators; According to the sample data typical fluctuation indicators, perform adaptive over-sampling on the over-sampling data set.

3. The method of claim 2, wherein the method comprises: The fitting analysis on the over-sampling data set and the calculation of the corresponding over-sampling change rate curve set based on the fitting analysis result, comprising: According to the time series data characteristics of the target scene, determine the segmentation window size corresponding to different index dimensions; Based on the sliding window method, filter the data in multiple index dimensions in the over-sampling data set according to the segmentation window size; Fitting the filtered multi-component segmented data into continuous curves to form the set of excess sample change rate curves.

4. The method of claim 3, wherein the method comprises: Performing time-frequency domain analysis on the set of excess sample change rate curves using wavelet transform, extracting multi-dimensional frequency domain features therefrom, and performing cluster analysis on the multi-dimensional frequency domain features to output a cluster result as a frequency domain response working condition set, including: In combination with wavelet transform, analyzing curves to determine frequency center, energy distribution, and spectral entropy, and outputting the multi-dimensional frequency domain features; Based on principal component analysis method, calculating and obtaining principal component contribution degrees of each frequency domain feature in the multi-dimensional frequency domain features to the climbing recognition; According to the principal component contribution degrees, serializing each frequency domain feature, and selecting a pre-set number K of frequency domain features for cluster analysis to obtain the cluster result; Based on each cluster in the cluster result, defining a frequency domain response working condition and a working condition interval corresponding thereto.

5. A sea-based wind power fluctuation prediction system based on climbing recognition, characterized in that, The system is used to implement the offshore wind power fluctuation prediction method based on climbing recognition according to any one of claims 1 to 4, and the system includes: An excess sample acquisition module configured to acquire excess sample data set based on scene features of a target scene; A data set fitting analysis module configured to perform fitting analysis on the excess sample data set and calculate and obtain a corresponding set of excess sample change rate curves based on the fitting analysis result; A frequency domain response working condition set output module configured to perform time-frequency domain analysis on the set of excess sample change rate curves using wavelet transform, extract multi-dimensional frequency domain features therefrom, and perform cluster analysis on the multi-dimensional frequency domain features to output a cluster result as a frequency domain response working condition set; A working condition mapping model construction module configured to traverse the frequency domain response working condition set, evaluate attention value of each frequency domain response working condition in combination with statistical analysis method, and construct a plurality of working condition mapping models corresponding thereto, wherein the working condition mapping models are labeled with applicable intervals; A power fluctuation prediction module configured to perform adaptive climbing recognition based on the plurality of working condition mapping models, and perform real-time power fluctuation prediction according to the climbing recognition prediction result.

Images