New energy weather forecast method based on numerical weather forecast and AI revision

Abstract

The invention discloses a new energy weather forecast method based on numerical weather forecast and AI revision, and particularly relates to the technical field of weather forecast, and the method comprises the steps: obtaining the historical actually measured weather data of a new energy power station, correcting the numerical weather forecast data of multiple sources through employing an artificial intelligence algorithm, and building a correction model; inputting the current numerical weather forecast data into the correction model for correction, extracting the internal structure characteristics of the correction data in combination with a high-dimensional geometric mapping and manifold learning method, and evaluating the matching degree between the manifold distribution of the correction data and the historical data; analyzing and evaluating the data complexity by using multi-scale entropy, and ensuring that the data meets a preset standard; on the basis of matching degree and complexity evaluation, selecting suitable numerical weather forecast data for new energy short-term and medium-term meteorological parameter forecast; the method improves the prediction accuracy and applicability, and is suitable for the generation power prediction and optimization scheduling of the new energy power station.

Description

Technical Field

[0001] The present invention relates to the technical field of weather forecasting, and more specifically, to a new energy weather forecasting method based on numerical weather forecasting and AI revision. Background Art

[0002] With the rapid development of the new energy industry, weather forecasts play a vital role in the operation and management of new energy power stations. The power output of new energy power stations (such as wind farms, photovoltaic power stations, etc.) is directly affected by meteorological conditions, so accurate weather forecasts are crucial to the operation and scheduling of power stations, power generation predictions, and grid load management.

[0003] In the existing technology, the processing of meteorological forecast data is insufficient in terms of complexity and analysis of inherent structural characteristics, especially when facing multi-scale and multi-dimensional data. Due to the lack of comprehensive analysis of the complexity of meteorological data, the existing forecast model is difficult to accurately reflect the actual meteorological change trend, which will cause a large deviation between the forecast results and the actual situation, affecting the power generation prediction and grid dispatch of new energy power stations, and reducing the stability of the operation of new energy power stations.

[0004] In order to solve the above problems, a technical solution is now provided. Summary of the invention

[0005] In order to overcome the above-mentioned defects of the prior art, an embodiment of the present invention provides a new energy meteorological forecasting method based on numerical weather forecasting and AI revision to solve the problems raised in the above-mentioned background technology.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] A new energy weather forecasting method based on numerical weather forecast and AI revision includes the following steps:

[0008] S1: Obtain the historical measured meteorological data of the operating new energy power stations; Based on the historical measured meteorological data, use artificial intelligence algorithms to calibrate the historical numerical weather forecast data from multiple sources and establish a correction model;

[0009] S2: Input the current numerical weather forecast data from multiple sources into the correction model for correction to obtain the corrected weather forecast data;

[0010] S3: Use high-dimensional geometric mapping and manifold learning methods to reduce the dimension of the revised weather forecast data, perform frequency domain analysis and oscillation monitoring on the revised weather forecast data after dimension reduction, detect abnormal frequency components and oscillation patterns in the data, and determine whether there is abnormal oscillation;

[0011] S4: When there is no abnormal oscillation, based on the intrinsic structural characteristics of the revised meteorological forecast data, it is determined whether the distribution of the data manifold matches the manifold structure of the historical measured data above a preset threshold;

[0012] S5: Use the multi-scale entropy analysis method to evaluate the complexity of the revised weather forecast data at different scales and determine whether the complexity of the revised weather forecast data meets the preset standards;

[0013] S6: When the distribution of the data manifold matches the manifold structure of the historical measured data to a higher degree than the preset threshold, and the complexity of the revised meteorological forecast data meets the preset standard, the corresponding numerical weather forecast data is selected for the current new energy short-term and medium-term meteorological parameter forecast.

[0014] In a preferred embodiment, S1 specifically includes:

[0015] S101: extracting historical measured meteorological data for a specified time period from a data system of a new energy power station;

[0016] S102: Formatting the extracted historical measured meteorological data, unifying the data format, and eliminating incomplete and invalid records;

[0017] S103: Acquire historical numerical weather forecast data of corresponding time periods from different sources, and perform synchronization processing on the data in time and space;

[0018] S104: performing deviation analysis on the acquired numerical weather forecast data according to the historical measured meteorological data, and automatically correcting the forecast errors of each data source;

[0019] S105: Compare the corrected numerical weather forecast data with the historical measured meteorological data, and establish a correction model based on artificial intelligence.

[0020] In a preferred embodiment, S2 specifically includes:

[0021] S201: Acquire numerical weather forecast data from multiple sources to ensure that the data covers a time and space range consistent with historical measured meteorological data;

[0022] S202: Performing time and space synchronization processing on the acquired numerical weather forecast data to ensure that data from different sources are consistent at the same time node and spatial position;

[0023] S203: inputting the synchronously processed numerical weather forecast data into the correction model in sequence, and starting the model correction process;

[0024] S204: The correction model automatically adjusts the prediction results and performs error correction according to the deviation between the numerical weather forecast data from various sources and the historical measured data, and outputs the revised weather forecast data.

[0025] In a preferred embodiment, S3 specifically includes:

[0026] S301: pre-processing the revised weather forecast data;

[0027] S302: applying a high-dimensional geometric mapping method to perform dimensionality reduction processing on the pre-processed corrected weather forecast data to extract main features;

[0028] S303: using a manifold learning algorithm to further optimize the dimension reduction result and capture the intrinsic structural characteristics of the revised weather forecast data;

[0029] S304: performing frequency domain analysis on the revised weather forecast data after dimension reduction, and calculating frequency spectrum characteristics of the revised weather forecast data;

[0030] S305: Perform oscillation mode monitoring, detect abnormal frequency components, and determine whether there is abnormal oscillation in the revised weather forecast data.

[0031] In a preferred embodiment, S303 specifically includes: for each data point after dimension reduction, finding multiple nearest neighboring data points to construct a neighborhood relationship;

[0032] Calculate the reconstruction error, which is expressed as: Among them, E recon is the reconstruction error, Z h is the representation of the hth data point in low-dimensional space, is the neighborhood set of data point h, C hv is the data point Z v Reconstruct Z h The contribution weight of h is the number of the data point, N points Represents the total number of all data points in the dataset.

[0033] In a preferred embodiment, S4 is specifically:

[0034] When there is no significant abnormal oscillation in the revised meteorological forecast data, the reconstruction error is obtained;

[0035] Use historical measured meteorological data to calculate the reconstruction error of historical measured data;

[0036] Compare the reconstruction error of the current revised weather forecast data with the reconstruction error of the historical measured data, and calculate the matching degree of the reconstruction error;

[0037] If the reconstruction error matching degree exceeds the corresponding preset threshold, then the distribution of the data manifold and the manifold structure matching degree of the historical measured data are not higher than the preset threshold;

[0038] If the reconstruction error matching degree does not exceed its corresponding preset threshold, then the distribution of the data manifold and the manifold structure matching degree of the historical measured data are higher than the preset threshold.

[0039] In a preferred embodiment, S5 specifically includes:

[0040] S501: performing segment preprocessing on the revised weather forecast data, and performing segment normalization for different meteorological elements;

[0041] S502: Based on the time resolution and characteristic changes of the meteorological forecast data, dynamically set time windows of different scales, and automatically generate a multi-scale entropy analysis range that adapts to the actual data;

[0042] S503: for each time scale, the sample entropy value of each meteorological element is calculated, and combined with the weight factor, the overall complexity index is calculated by weighting the entropy value to obtain a complete multi-scale entropy value sequence;

[0043] S504: Compare the multi-scale entropy value sequence of the revised meteorological forecast data with the entropy value of the historical measured data, calculate the complexity difference between the two, and generate a complexity difference curve;

[0044] S505: A dynamic early warning mechanism is set up to determine whether the complexity of the revised weather forecast data meets the preset standard based on the complexity difference curve.

[0045] In a preferred embodiment, S505 specifically includes:

[0046] Establish the early warning trigger condition. The early warning trigger condition formula is: Among them, Trigger is the output of the warning trigger condition, Threshold b is the preset complexity difference threshold;

[0047] When the output of the warning trigger condition is 1, it means that the trigger warning condition is met. At this time, the complexity of the revised weather forecast data does not meet the preset standard;

[0048] When the output of the warning trigger condition is 0, it means that the warning trigger condition is not met. At this time, the complexity of the revised weather forecast data meets the preset standard.

[0049] In a preferred embodiment, S6 is specifically:

[0050] Filter out the numerical weather forecast data that satisfies the reconstruction error matching degree not exceeding the corresponding preset threshold and the output of the warning trigger condition is 1;

[0051] Extract key meteorological parameters related to renewable energy power generation from the screened numerical weather forecast data, including wind speed, wind direction, solar radiation intensity, cloud cover, temperature and humidity;

[0052] Input the extracted key meteorological parameters into the short-term forecast model to generate short-term meteorological parameter forecast results;

[0053] The selected numerical weather forecast data and short-term forecast results are input into the medium-term forecast model, and the medium-term forecast model is used to generate medium-term meteorological parameter forecasts.

[0054] The technical effects and advantages of the new energy weather forecasting method based on numerical weather forecast and AI revision of the present invention are as follows:

[0055] 1. The present invention significantly improves the accuracy and practicality of new energy weather forecasts by combining artificial intelligence algorithms with numerical weather forecast data revision. First, the numerical weather forecasts from multiple sources are corrected using historical measured data to ensure the uniformity and accuracy of the data. The corrected data is reduced in dimension by using high-dimensional geometric mapping and manifold learning methods, and abnormal oscillations are further detected by frequency domain analysis to ensure the structural integrity and stability of the data during the dimensionality reduction process, avoiding the influence of noise or abnormal data. This process not only optimizes the use of data, but also provides a more reliable data basis for short-term and medium-term weather forecasts for new energy power stations.

[0056] 2. The present invention comprehensively evaluates the complexity of meteorological data at different scales by introducing a multi-scale entropy analysis method, ensuring that the performance of the data at multiple time scales meets the preset standards. Combined with the manifold matching evaluation, it can accurately determine whether the intrinsic structural characteristics of the data are consistent with the historical measured data, further improving the credibility of the weather forecast. The present invention can provide more accurate weather forecast support for the power generation prediction and operation scheduling of new energy power stations, effectively reduce the uncertainty of power station operation, and improve the stability of the power system. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 This is a schematic diagram of a new energy weather forecasting method based on numerical weather forecasting and AI revision in the present invention. DETAILED DESCRIPTION

[0058] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0059] Example

[0060] Figure 1 The present invention provides a new energy weather forecasting method based on numerical weather forecast and AI revision, which comprises the following steps:

[0061] S1: Obtain the historical measured meteorological data of the operating new energy power stations; based on the historical measured meteorological data, use artificial intelligence algorithms to calibrate the historical numerical weather forecast data from multiple sources and establish a correction model.

[0062] S2: Input the current numerical weather forecast data from multiple sources into the correction model for correction to obtain the corrected weather forecast data.

[0063] S3: Use high-dimensional geometric mapping and manifold learning methods to reduce the dimension of the revised weather forecast data, perform frequency domain analysis and oscillation monitoring on the revised weather forecast data after dimension reduction, detect abnormal frequency components and oscillation patterns in the data, and determine whether there is abnormal oscillation.

[0064] S4: When there is no abnormal oscillation, based on the intrinsic structural characteristics of the revised meteorological forecast data, determine whether the distribution of the data manifold matches the manifold structure of the historical measured data above a preset threshold.

[0065] S5: Use the multi-scale entropy analysis method to evaluate the complexity of the revised weather forecast data at different scales and determine whether the complexity of the revised weather forecast data meets the preset standards.

[0066] S6: When the distribution of the data manifold matches the manifold structure of the historical measured data to a higher degree than the preset threshold, and the complexity of the revised meteorological forecast data meets the preset standard, the corresponding numerical weather forecast data is selected for the current new energy short-term and medium-term meteorological parameter forecast.

[0067] S1 specifically includes:

[0068] S101: Extract historical measured meteorological data for a specified period of time from the data system of the new energy power station to ensure the integrity and accuracy of the data:

[0069] Through the access interface of the new energy power station data system, the historical measured meteorological data within the specified time period is obtained. The historical measured meteorological data usually includes temperature, humidity, wind speed, wind direction, atmospheric pressure, solar radiation intensity and other meteorological elements. These data must cover the actual operation of the new energy power station in time and space to ensure that the data is representative and accurate. In order to ensure data integrity, possible missing values ​​or outliers need to be excluded during data extraction. During the extraction process, the time range of the data must be clearly specified to ensure that the extracted meteorological data meets the needs of subsequent correction and modeling.

[0070] S102: Format the extracted historical measured meteorological data, unify the data format, remove incomplete and invalid records, and ensure the uniformity of the data:

[0071] Format the historical measured meteorological data. The historical measured meteorological data comes from multiple sensors or different recording systems, and the data formats may be inconsistent. Standardize it and convert it into a unified format, including time format, unit unification, etc. For example, the temperature can be unified in degrees Celsius, the wind speed in meters per second, and the solar radiation intensity in watts per square meter. During the formatting process, incomplete and invalid data records need to be eliminated to ensure the accuracy of subsequent analysis. For incomplete data, missing values ​​can be filled by data interpolation or other reasonable data recovery techniques to minimize information loss. Through these processes, it is ensured that the historical measured meteorological data can be consistent with the numerical weather forecast data so that accurate corrections can be made in subsequent steps.

[0072] S103: Obtain historical numerical weather forecast data for corresponding time periods from different sources to ensure the diversity of data sources, and perform synchronization processing on them in time and space:

[0073] Numerical weather forecast data for the corresponding time period are obtained from multiple sources. These sources can include national meteorological departments, commercial meteorological service providers, or satellite remote sensing data. The datasets from each source may have differences in spatial and temporal resolution, and these data are processed synchronously.

[0074] Time synchronization refers to aligning the timestamps of different data sources to ensure that comparison and correction are performed at the same time node. Spatial synchronization uses interpolation algorithms to project data of different resolutions onto the same grid system. Through these processes, numerical weather forecast data can be directly compared and analyzed with measured meteorological data.

[0075] S104: Using artificial intelligence algorithms, the obtained numerical weather forecast data is analyzed for deviations based on historical measured meteorological data, and the forecast errors of various data sources are automatically corrected:

[0076] The artificial intelligence algorithm is used to correct the acquired numerical weather forecast data. First, by constructing a deviation model, the deviation between the historical measured meteorological data and the numerical weather forecast data of the corresponding time period is analyzed. This deviation can be described by the following formula: ΔY = Y obs -Y nwp ; Among them, ΔY represents the forecast error, Y obs represents the historical measured meteorological data, Y nwpRepresents numerical weather forecast data. By statistically analyzing the distribution of forecast errors in historical data, the sources of systematic and random errors in each data source are identified.

[0077] In artificial intelligence algorithms, commonly used models include deep neural networks, support vector machines, and random forests. These models can automatically extract key factors that affect the accuracy of weather forecasts by learning from a large amount of historical data, and make targeted corrections to numerical weather forecast data from different sources. Taking deep neural networks as an example, the input of the model is historical numerical weather forecast data and the corresponding measured meteorological data, and the output is the corrected numerical weather forecast data. The model continuously adjusts the weights during the training process, and ultimately makes the corrected forecast data closer to the actual observed values.

[0078] The corrected numerical weather forecast data is expressed by the following formula: corrected =Y nwp +ΔY adjusted ; Among them, Y corrected is the corrected forecast data, ΔY adjusted It is the error value adjusted by artificial intelligence algorithm.

[0079] S105: Compare the corrected numerical weather forecast data with the historical measured meteorological data, establish a correction model based on artificial intelligence, and save the model for subsequent meteorological data correction:

[0080] After the correction is completed, the corrected numerical weather forecast data is compared with the historical measured meteorological data to evaluate the effectiveness of the correction model. In order to evaluate the effect of the correction model, we calculate the mean square error by comparing the historical measured meteorological data and the corrected forecast data. Specifically, the measured data at each time point is compared with the corrected forecast data, the square of the difference between them is calculated, and the data of all time points are averaged. In this way, a mean square error that reflects the overall forecast accuracy is obtained. The smaller the mean square error, the smaller the difference between the corrected forecast data and the actual situation, and the better the correction effect of the model. The mean square error can well measure the accuracy of the correction model over the entire time period.

[0081] By calculating the mean square error, the correction effect of the correction model over the entire time period can be evaluated. If the mean square error is low, it means that the correction model is effective and the model can be saved for subsequent meteorological data correction.

[0082] It is worth noting that the correction model is not only used for a single correction, but can also be used for continuous correction of future numerical weather forecast data. Therefore, the saved model can be repeatedly called in subsequent forecasts to further optimize the power forecast and meteorological parameter forecast of new energy power stations.

[0083] S2 specifically includes:

[0084] S201: Obtain numerical weather forecast data from multiple sources to ensure that the data covers a time and space range consistent with historical measured meteorological data:

[0085] Acquire numerical weather forecast data for the current time period from multiple sources. Data sources may include national meteorological departments, third-party meteorological service platforms, or other numerical weather forecast providers. Each dataset must cover the same time range and spatial distribution as historical measured meteorological data. When acquiring data, ensure that the time interval and spatial resolution of the data are consistent with previous historical data so that the correction model in subsequent steps can perform consistency analysis. All datasets must be clearly marked with timestamps and spatial grid information to ensure that they can be compared at the same time node and the same geographical area.

[0086] S202: Perform time and space synchronization processing on the acquired numerical weather forecast data to ensure that data from different sources are consistent at the same time node and spatial position:

[0087] Since numerical weather forecast data from different sources may have different temporal and spatial resolutions, the data must be synchronized in time and space. The purpose of time synchronization is to align weather forecast data from various sources to a unified time node. If the temporal resolution of data from a certain source is coarse or does not match the forecast data from other sources, it can be supplemented by interpolation so that data from all sources have consistent records at each time node.

[0088] Spatial synchronization is achieved by re-interpolating data of different spatial resolutions into the same spatial grid.

[0089] Using a uniform grid size, data from different sources may use larger or smaller grid spacing, and all forecast data are uniformly projected to the same spatial resolution through bilinear interpolation or other spatial interpolation algorithms. The interpolated data can be expressed by the following formula: Among them, P interp (x, y, t) represents the forecast data after interpolation, P(x i ,y i ,t) indicates that at the neighboring point x i ,y i and the numerical weather forecast value at t, n represents the number of neighboring points for interpolation, x and y are used to represent the location coordinates of the numerical weather forecast data in geographic space, x is longitude, y is latitude, t represents the time node, x i is the longitude of the ith neighboring point, y i is the latitude of the i-th neighboring point.

[0090] In this way, the consistency of weather forecast data from all sources in time and space is ensured.

[0091] S203: Input the synchronously processed numerical weather forecast data into the correction model in sequence, and start the model correction process:

[0092] The processed numerical weather forecast data is sequentially input into the established correction model. The correction model is trained based on the deviation between the historical measured meteorological data and the numerical weather forecast data. Therefore, when the model receives new forecast data, it can identify the deviation trend between it and the historical data. At this time, the input of the correction model is the weather forecast data that has been synchronously processed, and the output is the corrected forecast result. Artificial intelligence algorithms such as neural networks or random forests are usually used in the correction model. The specific correction formula can be expressed as: P corrected (x, y, t) = f AI (P nwp (x, y, t), θ); where, P corrected (x, y, t) represents the corrected forecast data, P nwp (x, y, t) is the numerical weather forecast data after synchronization processing, f AI represents the calibration model, and θ represents the model parameters.

[0093] The correction model automatically adjusts the prediction results of the current input data based on the weights and parameters trained with historical data.

[0094] S204: The correction model automatically adjusts the prediction results and performs error correction according to the deviation between the numerical weather forecast data from each source and the historical measured data, and outputs the corrected weather forecast data:

[0095] During the model calibration process, the model will automatically adjust the current forecast results based on the deviation between the historical measured meteorological data and the numerical weather forecast data. This process relies on the error correction mechanism learned in the calibration model constructed in the early stage. Specifically, the model adjusts the forecast value of the current data by comparing the difference between the historical measured data and the historical forecast data.

[0096] The core mechanism of error correction can be described by the following formula: ΔP = P obs (x, y, t)-P nwp (x, y, t); where ΔP represents the prediction error, P obs (x, y, t) represents the historical measured meteorological data.

[0097] The correction model adjusts the current weather forecast data according to the forecast error. The adjusted results are as follows: corrected (x, y, t) = P nwp (x,y,t)+fadjust (ΔP); where f adjust is a function used to correct the error.

[0098] After the correction process is completed, the corrected weather forecast data is output. These data are more accurate than the original numerical weather forecast data. The corrected weather forecast data includes all meteorological elements such as temperature, humidity, wind speed, wind direction, etc. At the same time as the output, the original uncorrected data also needs to be saved for comparison and further optimization in subsequent analysis.

[0099] The corrected weather forecast data can be recorded by the following expression: corrected (x1,y1,t),P corrected (x2,y2,t),…}.

[0100] S3 specifically includes:

[0101] S301: Preprocessing the revised weather forecast data:

[0102] When there are missing values ​​in the revised weather forecast data, linear interpolation method is used to fill them.

[0103] Statistical methods were used to calculate the mean and standard deviation of each meteorological element, and outliers exceeding three times the standard deviation were eliminated.

[0104] In order to ensure the comparability of various meteorological elements, the revised meteorological forecast data are standardized.

[0105] S302: Apply high-dimensional geometric mapping method to reduce the dimension of the pre-processed corrected weather forecast data and extract the main features:

[0106] The high-dimensional geometric mapping (such as principal component analysis, PCA) is used to reduce the dimension of the pre-processed revised weather forecast data. The specific operations are as follows:

[0107] Calculate the covariance matrix of the revised weather forecast data, the formula is: Among them, M cov is the covariance matrix, W k is the kth corrected weather forecast data sample, μ W is the mean of the revised weather forecast data, R is the number of revised weather forecast data samples, k is the number of revised weather forecast data samples, Represents the transpose operation of a vector or matrix.

[0108] Perform eigenvalue decomposition on the covariance matrix to obtain the eigenvalues ​​and corresponding eigenvectors.

[0109] According to the cumulative variance contribution rate, the first m principal components are selected so that the variance contribution rate reaches more than 95%. The calculation formula for the cumulative variance contribution rate is: Among them, Var cum (m) represents the cumulative variance contribution rate of the first m principal components, s is the total number of eigenvalues, p is the number of eigenvalues, and m is the number of selected principal components.

[0110] Project the original data into low-dimensional space to obtain the corrected weather forecast data after dimensionality reduction: Z proj =WQ m Among them, Z proj is the revised weather forecast data after dimension reduction, W is the original data matrix, Q m are the first m eigenvectors.

[0111] S303: Use the manifold learning algorithm to further optimize the dimensionality reduction results and capture the inherent structural characteristics of the revised weather forecast data:

[0112] For each data point after dimensionality reduction, find its nearest multiple neighboring data points and build a neighborhood relationship. The Euclidean distance calculation formula between two points is: Among them, D uv is the data point Z u and Z v The distance between them, d is the number of dimensions of the data after dimensionality reduction of the revised weather forecast data, u and v represent the indexes of the two data points whose distance needs to be calculated, g is the dimension index, and Z u，g is the data point Z u The coordinate value in the g-th dimension, Z v，g is the data point Z v Coordinate value in the gth dimension.

[0113] The reconstruction error is calculated by minimizing the reconstruction error and maintaining the linear structure of the data in the local neighborhood. Its expression is: Among them, E recon is the reconstruction error, Z h is the representation of the hth data point in low-dimensional space, is the neighborhood set of data point h, C hv is the data point Z v Reconstruct Z h The contribution weight (the weight is based on Z v With Z h and its relative position in the neighborhood), h is the number of the data point, N points Represents the total number of all data points in the dataset.

[0114] Through eigenvalue decomposition, the low-dimensional embedding result of the reduced-dimensional data is obtained, and the optimized reduced-dimensional data is obtained.

[0115] S304: Perform frequency domain analysis on the revised weather forecast data after dimension reduction, and calculate the frequency spectrum characteristics of the revised weather forecast data:

[0116] The corrected weather forecast data after dimensionality reduction are arranged in chronological order to form a time series.

[0117] The frequency domain analysis is performed using Fast Fourier Transform (FFT) to extract the frequency components.

[0118] According to the Fourier transform results, the amplitude and phase of each frequency are calculated to obtain the spectrum characteristics.

[0119] S305: Perform oscillation mode monitoring, detect abnormal frequency components, and determine whether there is abnormal oscillation in the corrected weather forecast data:

[0120] According to the spectrum characteristics of historical meteorological data, the threshold of the normal frequency range is set.

[0121] Calculate the amplitude of each frequency component and determine whether any frequency exceeds the threshold. If the amplitude of a frequency exceeds the threshold, it is considered an abnormal oscillation.

[0122] Based on the amplitudes of all frequencies, determine whether there are significant abnormal oscillations in the revised weather forecast data.

[0123] Among them, significant abnormal oscillation usually means that the amplitude of some frequency components significantly exceeds the normal range, which is manifested as abnormal periodic fluctuations in the data. It is usually judged by the following standards:

[0124] The amplitude exceeds a set threshold (for example, two to three standard deviations above the historical mean).

[0125] The energy distribution of a certain frequency significantly exceeds the accumulated energy in the normal range.

[0126] The frequency distribution pattern does not match the expected one, with unexpected high amplitude frequencies appearing.

[0127] The abnormal frequency oscillation lasts for a long time and has significant persistence.

[0128] Abnormal resonance occurs in multiple frequency components at the same time.

[0129] S4 is specifically:

[0130] Based on the reconstruction error obtained in S303, it is determined whether the distribution of the data manifold and the manifold structure matching degree of the historical measured data is higher than a preset threshold.

[0131] When the result of the judgment in S305 is that there is no significant abnormal oscillation in the corrected weather forecast data, the reconstruction error E obtained by the manifold learning algorithm in S303 is used.recon , which reflects the difference in local geometric relationship between the revised weather forecast data after dimensionality reduction and the original high-dimensional data. The smaller the reconstruction error, the better the manifold is maintained after dimensionality reduction.

[0132] Similarly, the reconstruction error of the historical measured meteorological data is calculated using the historical measured data as a reference for manifold matching.

[0133] Compare the reconstruction error of the current revised weather forecast data with the reconstruction error of the historical measured data, and calculate the reconstruction error matching degree. The reconstruction error matching degree can be calculated by the following formula: ΔE = |E recon -E recon_hist |; where ΔE is the reconstruction error matching degree, E recon_hist is the reconstruction error of the historical measured data.

[0134] If the reconstruction error matching degree exceeds the corresponding preset threshold, the distribution of the data manifold and the manifold structure matching degree of the historical measured data are not higher than the preset threshold, indicating that there are significant differences between the manifold structure of the data and the historical measured data, and the manifold matching degree is low.

[0135] If the reconstruction error matching degree does not exceed its corresponding preset threshold, the distribution of the data manifold and the manifold structure matching degree of the historical measured data are higher than the preset threshold, and it is considered that the manifold distribution of the current data and the manifold structure matching degree of the historical data are high.

[0136] Among them, the preset threshold corresponding to the reconstruction error matching degree is usually set through historical data statistics and experimental verification. First, use a large amount of historical measured data to calculate its reconstruction error and obtain its statistical characteristics, such as mean and standard deviation. Then, set the threshold according to the actual application requirements. Usually, the mean of the error plus two to three times the standard deviation is selected as the preset threshold to ensure that a small deviation can be accepted within the normal data fluctuation range. At the same time, the setting of the threshold can also be based on the test data in actual use. Through repeated adjustment and optimization, the optimal threshold that can balance accuracy and stability can be found.

[0137] S5 specifically includes:

[0138] S501: Perform segmented preprocessing on the revised weather forecast data, and perform segmented normalization for different meteorological elements:

[0139] The different meteorological elements (such as temperature, humidity, wind speed, wind direction, etc.) in the weather forecast data are normalized in sections.

[0140] Since the dimensions of each meteorological element are different, direct analysis may lead to the exaggeration or neglect of the impact of some elements. Therefore, it is necessary to preprocess these data first and convert the data of each meteorological element into dimensionless data for subsequent multi-scale entropy analysis.

[0141] S502: Based on the time resolution and characteristic changes of the meteorological forecast data, dynamically set time windows of different scales and automatically generate a multi-scale entropy analysis range that adapts to the actual data:

[0142] Different time scale windows are dynamically set based on the time resolution of weather forecast data (such as hourly, daily, etc.) and changes in data characteristics.

[0143] The setting of the time scale directly affects the effect of multi-scale entropy analysis, and a time window suitable for the data is automatically generated based on the actual data.

[0144] The setting formula of the time window is: ΔT b =T final -T start +θ·δ t ; where ΔT b is the window length of the bth time scale, T start and T final are the start and end time points of the data, respectively, t is the time resolution, θ is the scale index, and b is the number of the time scale.

[0145] By dynamically generating time windows, the analysis can capture both the short-term fluctuation characteristics of the data and the long-term trends, ensuring that the multiscale entropy analysis adapts to the multiscale characteristics of actual meteorological forecast data.

[0146] S503: For each time scale, the sample entropy value of each meteorological element is calculated, and combined with the weight factor, the overall complexity index is calculated by weighting the entropy value to obtain a complete multi-scale entropy value sequence:

[0147] At each time scale, the sample entropy value of each meteorological element is calculated separately. Sample entropy is a common indicator for evaluating data complexity and can be used to quantify the complexity of meteorological data at different scales. In order to obtain a comprehensive complexity evaluation result, the sample entropy value of each meteorological element is weighted to obtain the overall multi-scale complexity index.

[0148] Calculate the sample entropy value of meteorological elements, and its expression is: Among them, S sample，b，ais the sample entropy value of the meteorological element, indicating the sample entropy of the a-th meteorological element at the b-th time scale; a is the number of the meteorological element; #matches indicates the number of data pairs that meet the matching conditions; #total_pairs is the total number of data pairs; # is a symbol used to indicate "quantity", similar to the meaning of "number" or "count".

[0149] Calculate the overall entropy value of meteorological elements, and its expression is: Among them, S total，b is the overall entropy value of meteorological elements, indicating the overall entropy value at the bth time scale, W a is the weight factor of the ath meteorological element, and mr represents the total number of meteorological elements.

[0150] By combining the complexity assessment results of each meteorological element, a multi-scale entropy value sequence is obtained based on the overall entropy value of the meteorological element.

[0151] S504: Compare the multi-scale entropy value sequence of the revised weather forecast data with the entropy value of the historical measured data, calculate the complexity difference between the two, and generate a complexity difference curve:

[0152] The multi-scale entropy value sequence of the revised meteorological forecast data is compared with the entropy value of the historical measured data, and their complexity differences at different time scales are calculated. A complexity difference curve is generated to intuitively show the difference between the two.

[0153] The calculation complexity difference value is expressed as: Among them, ΔS b is the complexity difference value, indicating the complexity difference at the bth time scale, is the overall entropy value of the revised weather forecast data, is the overall entropy value of historical measured data.

[0154] The complexity difference values ​​calculated under all time scales are arranged in sequence, and a complexity difference curve is drawn with the time scale as the horizontal axis and the complexity difference value as the vertical axis.

[0155] S505: A dynamic early warning mechanism is set up to determine whether the complexity of the revised weather forecast data meets the preset standard based on the complexity difference curve:

[0156] The early warning trigger condition formula is:

[0157] Among them, Trigger is the output of the warning trigger condition. It is a Boolean value that indicates whether the warning is triggered. Its value is 1 or 0 depending on whether the condition is met.

[0158] Threshold bis the preset complexity difference threshold at the bth time scale.

[0159] When the output of the warning trigger condition is 1, it means that the trigger warning condition is met. At this time, the complexity of the revised meteorological forecast data does not meet the preset standard, that is, at the bth time scale, the complexity difference value exceeds the preset complexity difference threshold. At this time, it is considered that the complexity of the data is abnormal and needs further processing or correction.

[0160] When the output of the warning trigger condition is 0, it means that the warning trigger condition is not met. At this time, the complexity of the corrected meteorological forecast data meets the preset standard. At this time, the data complexity is normal and no further operation is required.

[0161] Among them, the preset complexity difference threshold is usually set based on the statistical analysis of historical data. First, the complexity difference at each time scale is calculated from a large amount of historical measured data to obtain its mean and standard deviation. Then, when setting the threshold, choose to add two to three times the standard deviation on the basis of the mean, so that most of the normal fluctuation range can be covered while identifying abnormal situations. The threshold can also be adjusted through feedback in actual use to ensure that it can capture abnormal complexity in the data in a timely manner without being overly sensitive to normal fluctuations.

[0162] S6 is specifically:

[0163] The numerical weather forecast data that satisfies the reconstruction error matching degree not exceeding the corresponding preset threshold and the output of the warning trigger condition is 1 are screened.

[0164] From the screened numerical weather forecast data, key meteorological parameters related to renewable energy power generation are extracted, including but not limited to:

[0165] Wind speed and direction: affect the power output of a wind farm.

[0166] Solar radiation intensity and cloud cover: affect the power generation efficiency of photovoltaic power stations.

[0167] Temperature and humidity: May have an impact on meteorological conditions and equipment operation.

[0168] The extracted key meteorological parameters are input into the short-term forecast model, which makes predictions based on current data and historical patterns, usually covering meteorological conditions for the next few hours to a day, specifically:

[0169] Model input: Input the selected key meteorological parameters into the short-term forecast model and set the time range.

[0170] Run the model: Generate short-term meteorological parameter forecast results through optimization algorithms or neural networks.

[0171] Result output: Output forecast results, including changes in key meteorological parameters within the next few hours to a day.

[0172] After the short-term forecast, medium-term meteorological parameter forecasts are carried out, usually predicting the meteorological conditions for the next few days to a week. The data at this stage is more complex and the meteorological change cycle considered is longer.

[0173] Model input: Input the selected numerical weather forecast data and short-term forecast results into the medium-term forecast model.

[0174] Run medium-term models: Use medium-term forecast models to generate longer-term forecasts of medium-term meteorological parameters, including wind speed, solar radiation intensity, etc.

[0175] Output medium-term forecast results: Generate forecasts of meteorological parameters for the next few days and use them to predict the power generation capacity of renewable energy.

[0176] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters and thresholds in the formula are set by technicians in this field according to actual conditions.

[0177] The above embodiments can be implemented in whole or in part by software, hardware, firmware or any other combination. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from one website site, computer, server or data center to another website site, computer, server or data center by wired (e.g., infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that contains one or more available media sets. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state hard disk.

[0178] Those of ordinary skill in the art will appreciate that the modules and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0179] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and modules described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0180] In the several embodiments provided in the present application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the modules is only a logical function division. There may be other division methods in actual implementation, such as multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or modules, which can be electrical, mechanical or other forms.

[0181] The modules described as separate components may or may not be physically separated, and the components shown as modules may or may not be physical modules, and may be located in one place or distributed on multiple network modules. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0182] In addition, each functional module in each embodiment of the present application may be integrated into one processing module, or each module may exist physically separately, or two or more modules may be integrated into one module.

[0183] If the functions are implemented in the form of software function modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0184] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

[0185] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A new energy weather forecasting method based on numerical weather forecast and AI revision, characterized in that: The steps include: S1: Obtain the historical measured meteorological data of the operating new energy power stations; Based on the historical measured meteorological data, use artificial intelligence algorithms to calibrate the historical numerical weather forecast data from multiple sources and establish a correction model; S2: Input the current numerical weather forecast data from multiple sources into the correction model for correction to obtain the corrected weather forecast data; S3: Use high-dimensional geometric mapping and manifold learning methods to reduce the dimension of the revised weather forecast data, perform frequency domain analysis and oscillation monitoring on the revised weather forecast data after dimension reduction, detect abnormal frequency components and oscillation patterns in the data, and determine whether there is abnormal oscillation; S4: When there is no abnormal oscillation, based on the intrinsic structural characteristics of the revised meteorological forecast data, it is determined whether the distribution of the data manifold matches the manifold structure of the historical measured data above a preset threshold; S5: Use the multi-scale entropy analysis method to evaluate the complexity of the revised weather forecast data at different scales and determine whether the complexity of the revised weather forecast data meets the preset standards; S6: When the distribution of the data manifold matches the manifold structure of the historical measured data to a higher degree than the preset threshold, and the complexity of the revised meteorological forecast data meets the preset standard, the corresponding numerical weather forecast data is selected for the current new energy short-term and medium-term meteorological parameter forecast.

2. According to claim 1, a new energy weather forecasting method based on numerical weather forecast and AI revision is characterized in that: S1 specifically includes: S101: extracting historical measured meteorological data for a specified time period from a data system of a new energy power station; S102: Formatting the extracted historical measured meteorological data, unifying the data format, and eliminating incomplete and invalid records; S103: Acquire historical numerical weather forecast data of corresponding time periods from different sources, and perform synchronization processing on the data in time and space; S104: performing deviation analysis on the acquired numerical weather forecast data according to the historical measured meteorological data, and automatically correcting the forecast errors of each data source; S105: Compare the corrected numerical weather forecast data with the historical measured meteorological data, and establish a correction model based on artificial intelligence.

3. A new energy weather forecasting method based on numerical weather forecast and AI revision according to claim 2, characterized in that: S2 specifically includes: S201: Acquire numerical weather forecast data from multiple sources to ensure that the data covers a time and space range consistent with historical measured meteorological data; S202: Performing time and space synchronization processing on the acquired numerical weather forecast data to ensure that data from different sources are consistent at the same time node and spatial position; S203: inputting the synchronously processed numerical weather forecast data into the correction model in sequence, and starting the model correction process; S204: The correction model automatically adjusts the prediction results and performs error correction according to the deviation between the numerical weather forecast data from various sources and the historical measured data, and outputs the revised weather forecast data.

4. The new energy weather forecasting method based on numerical weather forecast and AI revision according to claim 3 is characterized in that: S3 specifically includes: S301: pre-processing the revised weather forecast data; S302: applying a high-dimensional geometric mapping method to perform dimensionality reduction processing on the pre-processed corrected weather forecast data to extract main features; S303: using a manifold learning algorithm to further optimize the dimension reduction result and capture the intrinsic structural characteristics of the revised weather forecast data; S304: performing frequency domain analysis on the revised weather forecast data after dimension reduction, and calculating frequency spectrum characteristics of the revised weather forecast data; S305: Perform oscillation mode monitoring, detect abnormal frequency components, and determine whether there is abnormal oscillation in the revised weather forecast data.

5. The new energy weather forecasting method based on numerical weather forecast and AI revision according to claim 4 is characterized in that: S303 specifically includes: for each data point after dimensionality reduction, finding multiple nearest neighboring data points to establish a neighborhood relationship; Calculate the reconstruction error, which is expressed as: Among them, E recon is the reconstruction error, Z h is the representation of the hth data point in low-dimensional space, is the neighborhood set of data point h, C hυ is the data point Z υ Reconstruct Z h The contribution weight of h is the number of the data point, N points Represents the total number of all data points in the dataset.

6. The new energy weather forecasting method based on numerical weather forecast and Al revision according to claim 5 is characterized in that: S4 is specifically: When there is no significant abnormal oscillation in the revised meteorological forecast data, the reconstruction error is obtained; Use historical measured meteorological data to calculate the reconstruction error of historical measured data; Compare the reconstruction error of the current revised weather forecast data with the reconstruction error of the historical measured data, and calculate the matching degree of the reconstruction error; If the reconstruction error matching degree exceeds the corresponding preset threshold, then the distribution of the data manifold and the manifold structure matching degree of the historical measured data are not higher than the preset threshold; If the reconstruction error matching degree does not exceed its corresponding preset threshold, then the distribution of the data manifold and the manifold structure matching degree of the historical measured data are higher than the preset threshold.

7. A new energy weather forecasting method based on numerical weather forecast and AI revision according to claim 6, characterized in that: S5 specifically includes: S501: performing segment preprocessing on the revised weather forecast data, and performing segment normalization for different meteorological elements; S502: Based on the time resolution and characteristic changes of the meteorological forecast data, dynamically set time windows of different scales, and automatically generate a multi-scale entropy analysis range that adapts to the actual data; S503: for each time scale, the sample entropy value of each meteorological element is calculated, and the overall complexity index is calculated by weighting the entropy value in combination with the weight factor to obtain a complete multi-scale entropy value sequence; S504: Compare the multi-scale entropy value sequence of the revised meteorological forecast data with the entropy value of the historical measured data, calculate the complexity difference between the two, and generate a complexity difference curve; S505: A dynamic early warning mechanism is set up to determine whether the complexity of the revised weather forecast data meets the preset standard based on the complexity difference curve.

8. The new energy weather forecasting method based on numerical weather forecast and AI revision according to claim 7 is characterized in that: S505 is specifically: Establish the early warning trigger condition. The early warning trigger condition formula is: Among them, Trigger is the output of the warning trigger condition, Threshold b is the preset complexity difference threshold; When the output of the warning trigger condition is 1, it means that the trigger warning condition is met. At this time, the complexity of the revised weather forecast data does not meet the preset standard; When the output of the warning trigger condition is 0, it means that the warning trigger condition is not met. At this time, the complexity of the revised weather forecast data meets the preset standard.

9. A new energy weather forecasting method based on numerical weather forecast and AI revision according to claim 8, characterized in that: S6 is specifically: Filter out the numerical weather forecast data that satisfies the reconstruction error matching degree not exceeding the corresponding preset threshold and the output of the warning trigger condition is 1; Extract key meteorological parameters related to renewable energy power generation from the screened numerical weather forecast data, including wind speed, wind direction, solar radiation intensity, cloud cover, temperature and humidity; Input the extracted key meteorological parameters into the short-term forecast model to generate short-term meteorological parameter forecast results; The selected numerical weather forecast data and short-term forecast results are input into the medium-term forecast model, and the medium-term forecast model is used to generate medium-term meteorological parameter forecasts.

Images