Wind power prediction method and system based on meteorological data

Abstract

The invention discloses a wind power prediction method and system based on meteorological data, and relates to the technical field of wind power prediction.The wind power prediction method comprises the steps that a basic database containing extreme meteorology, wind power and fan operation state data is constructed, key parameter dimensions are determined through core parameter sensitivity analysis, and a multi-scene parameter combination is generated; similar historical samples are screened, curves are aligned through a dynamic time warping algorithm, a mean value is taken, and a scene-power time sequence matrix is output; then dynamically adjusting the scene weight in combination with a historical occurrence frequency and a numerical weather forecast initial parameter deviation rate, fitting power time sequence data by adopting Cornish-Fisher series to construct a probability density function, and performing weighted fusion to obtain overall power probability distribution; and finally, extracting upper and lower quantiles of the double confidence levels to determine a wind power fluctuation domain, calculating a coverage rate and a bandwidth mean square root, and labeling confidence output. According to the method, precise scene coverage and dynamic weight adaptation are realized, the prediction reliability is improved, and scientific support is provided for power grid dispatching.

Description

Technical Field

[0001] This invention relates to the field of wind power forecasting technology, and specifically to a wind power forecasting method and system based on meteorological data. Background Technology

[0002] As the global energy structure transitions towards cleaner and lower-carbon energy, wind power, as a crucial component of renewable energy, has seen rapid and continuous growth in installed capacity and grid connection, becoming a key power source in the power system. During large-scale grid-connected operation of wind power, the output power of wind farms is significantly affected by meteorological conditions. In particular, the frequent occurrence of extreme weather events such as strong gusts, thunderstorms, and low-temperature icing can cause significant fluctuations in wind power output within a short period. Therefore, a wind power forecasting method and system based on meteorological data has emerged.

[0003] Existing technologies, such as the invention patent application CN119338068A, disclose a wind power forecasting method and system based on meteorological data, which relates to wind power forecasting technology. The method includes: pre-training a large model and training an LSTM model; obtaining geographical location description information of the target wind farm to construct a fine-tuning training set based on the key fields; fine-tuning the trained large model using the fine-tuning training set; obtaining current meteorological forecast data for the area covered by the target wind farm, and outputting wind farm forecast information for multiple distribution locations based on the fine-tuned large model; performing prediction using the trained LSTM model based on the wind farm forecast information to obtain power generation forecast information for multiple distribution locations; and calculating the power generation forecast result of the target wind farm based on the power generation forecast information for multiple distribution locations. This application combines a large model and LSTM, using the large model to train and predict meteorological data, and utilizing LSTM to predict wind farm power generation data, thereby improving the accuracy of wind power forecasting.

[0004] Regarding the above solutions, the applicant of this invention has found that the above technologies have at least the following technical problems: 1. Existing technologies, in the extreme weather scenario generation stage, are unable to achieve complete coverage of the possibilities of meteorological evolution. On the one hand, there are limitations in the selection of scenario types. Most methods only focus on common types such as strong gusts and thunderstorms, without fully considering new extreme weather scenarios brought about by climate non-stationarity, such as typhoon intensification and sudden strong inland convection events that exceed historical experience, resulting in insufficient adaptability of prediction models to "black swan" events. On the other hand, the selection of key parameter dimensions and gradient settings lack systematicity. Some technologies do not use professional meteorological models such as WRF for sensitivity analysis, but only rely on experience to select conventional parameters such as wind speed and temperature, ignoring core dimensions that significantly affect power fluctuations, such as the rate of wind speed rise and the wind direction deflection angle. Even if key parameters are selected, the gradient interval is mostly set subjectively, such as a fixed wind speed interval of 2 m / s, without dynamic adjustment based on the standard deviation characteristics of historical data. This results in scenario parameter combinations that are either too dense, causing computational redundancy, or too sparse, omitting key evolution states, and failing to achieve coverage of more than 90% of the extreme weather parameter value range.

[0005] 2. Existing technologies generally employ static weight allocation logic when determining scene weights, failing to align with the dynamic evolution of current weather patterns. Most methods determine prior probabilities solely based on the frequency of historical extreme weather events, directly using these probabilities as scene weights without considering the deviation between current initial parameters and historical averages in numerical weather prediction. For example, when the predicted initial wind speed is 20% higher than the historical average, the high weight of historical low-to-medium wind speed scenes is still applied, leading to conservative predictions that fail to reflect the power characteristics under actual weather intensity. While some technologies attempt to adjust weights, they lack quantitative adjustment logic, relying on vague statements such as appropriately increasing the weight of high-wind-speed scenes, without establishing a correlation between the deviation rate and the adjustment coefficient, resulting in significant subjectivity in weight adjustments. Furthermore, existing technologies fail to consider the correlation between scenes. For complex extreme scenarios such as strong gusts and thunderstorms, they simply use a single-scene weight superposition method, ignoring the coupling effect between meteorological parameters, further reducing the scientific rigor of weight allocation.

[0006] 3. Existing technologies suffer from insufficient model adaptability and impractical output formats in the construction and output of power probability distributions. Regarding probability fitting, most methods assume a standard normal distribution to construct the probability density function, failing to consider the asymmetric distribution characteristics of power data under extreme weather conditions—for example, power exhibits a right-skewed distribution under low-temperature icing scenarios, where the deviation of the traditional normal distribution fitting can exceed 15%. Even when some technologies employ correction methods such as Cornish-Fisher series, they do not strictly adhere to data standardization and inverse standardization processes, overlooking the implicit influence of mean and variance on quantile calculations, resulting in fitting errors exceeding 5%. In terms of output, existing technologies mostly output power ranges at a single confidence level, without simultaneously providing quantitative evaluation indicators such as coverage and root mean square bandwidth, making it impossible for grid dispatchers to determine the reliability and severity of fluctuations within the fluctuation domain. Summary of the Invention

[0007] To address the aforementioned technical shortcomings, the present invention aims to provide a wind power forecasting method and system based on meteorological data.

[0008] To solve the above-mentioned technical problems, the present invention adopts the following technical solution: In the first aspect, the present invention provides a wind power prediction method based on meteorological data, including the following steps: Step 1, generation of multi-scenario parameter combinations: determine the set of extreme weather types of the target wind farm, and construct a basic database of extreme weather-power, and then determine the key parameter dimensions of each extreme weather type according to the sensitivity of the core parameters of each extreme weather type, and generate multi-scenario parameter combinations.

[0009] Step 2: Output of the scenario-power time series matrix: Select historical samples with the same extreme weather as the current target wind farm from the extreme weather-power basic database, extract the wind power time series data of the samples, align the curves using the dynamic time warping algorithm, take the mean value to obtain the change process of the extreme weather power of the current target wind farm over time, and output the scenario-power time series matrix.

[0010] Step 3: Obtaining the overall power probability distribution: The historical frequency of extreme weather events at the current target wind farm is statistically analyzed from the extreme weather-power database as the prior probability. The deviation rate is calculated by combining the initial parameters of extreme weather events at the current target wind farm in numerical weather forecasts. The prior probability is adjusted to obtain the posterior probability as the weight of the extreme weather scenario at the current target wind farm. The power time series data is fitted using Cornish-Fisher series to construct the probability density function. The weighted fusion is then used to obtain the overall power probability distribution of extreme weather events at the current target wind farm.

[0011] Step 4: Determination of wind power fluctuation domain: Extract the upper and lower quantiles of the overall power probability distribution of the current target wind farm under extreme weather conditions at 90% and 95% confidence levels through weighted fusion, thereby determining the wind power fluctuation domain of the current target wind farm. Calculate the root mean square of the fluctuation domain coverage and bandwidth of the current target wind farm. At the same time, label the confidence level and complete the output.

[0012] In a second aspect, the present invention provides a wind power forecasting system based on meteorological data, comprising the following modules: a multi-scenario parameter combination generation module: used to determine the set of extreme weather types of the target wind farm and construct a basic database of extreme weather-power, and then determine the key parameter dimensions of each extreme weather type according to the sensitivity of the core parameters of each extreme weather type, and generate multi-scenario parameter combinations.

[0013] The scenario-power time series matrix output module is used to filter historical samples of the same type as the current target wind farm's extreme weather from the extreme weather-power basic database, extract the sample wind power time series data, align the curves using a dynamic time warping algorithm, take the mean value to obtain the change process of the current target wind farm's extreme weather power over time, and output the scenario-power time series matrix.

[0014] The overall power probability distribution acquisition module is used to statistically analyze the historical frequency of extreme weather events at the current target wind farm from the extreme weather-power basic database as the prior probability. It calculates the deviation rate by combining the initial parameters of extreme weather events at the current target wind farm in numerical weather forecasts, adjusts the prior probability to obtain the posterior probability as the weight of the extreme weather scenario at the current target wind farm, uses Cornish-Fisher series to fit the power time series data to construct the probability density function, and weighted fusion to obtain the overall power probability distribution of extreme weather events at the current target wind farm.

[0015] The wind power fluctuation domain determination module is used to extract the upper and lower quantiles of the overall power probability distribution under extreme weather conditions of the current target wind farm obtained by weighted fusion at 90% and 95% confidence levels, thereby determining the wind power fluctuation domain of the current target wind farm, calculating the root mean square of the fluctuation domain coverage and bandwidth of the current target wind farm, and completing the output after labeling the confidence level.

[0016] The beneficial effects of this invention are as follows: 1. The embodiments of this invention can solidify the data support foundation and significantly improve the accuracy of the correlation between extreme weather and power. By collecting three core data types—extreme weather, wind power, and wind turbine operating status—from the target wind farm's location over the past 10 years, and strictly correlating them by timestamp to ensure data synchronization, and employing an outlier removal standard exceeding the historical mean by ±3 standard deviations, combined with a linear interpolation method with an interpolation interval ≤30 minutes and an interpolation error ≤5%, the invention effectively avoids parameter deviations caused by coarse data processing, providing a high-quality data foundation for subsequent predictions and solving the problem of weak data foundation in existing technologies.

[0017] 2. This invention improves the coverage of extreme weather scenarios, achieving scientific and comprehensive scenario generation. By using the WRF meteorological numerical model to screen extreme event samples that cause power fluctuations exceeding 30%, key parameter dimensions are selected based on correlation. Gradients are dynamically set at intervals of one-third of the historical parameter value range and historical standard deviation. Ultimately, at least 9 scenario combinations are generated for each extreme weather type, covering more than 90% of the historical value range of key parameters. This not only covers common scenarios but also adapts to the possibility of new extreme weather evolution, overcoming the limitations of existing technologies in terms of scenario scope and subjective gradients.

[0018] 3. In this embodiment of the invention, a dynamic weight adjustment mechanism is constructed, which can accurately adapt to real-time weather change trends. First, the historical frequency of extreme weather events is extracted from the basic database as a prior probability. Then, combined with the deviation rate between the current initial parameters of extreme weather and the historical average, a quantitative correspondence is established whereby the adjustment coefficient increases by 1% for every 10% increase in the deviation rate. The weights of high and low parameter values ​​are dynamically adjusted according to the positive or negative deviation rate, avoiding the conservative prediction problem caused by static weights, reducing the subjectivity of weight adjustment, and indirectly adapting to complex extreme scenarios, thereby improving the scientific nature of weight allocation.

[0019] 4. In this embodiment of the invention, the probability fitting and output format are optimized, which can accurately support power grid dispatching decisions. Cornish-Fisher series combined with skewness and kurtosis correction quantiles are used to adapt to the asymmetric power distribution characteristics under extreme weather conditions, reducing fitting bias. A 90% and 95% dual-confidence level fluctuation domain is output, and the root mean square of coverage and bandwidth is calculated simultaneously with confidence levels labeled. The results are presented through a time-power visualization curve, allowing dispatchers to intuitively obtain the fluctuation range, reliability, and severity without additional calculations, improving decision-making efficiency and meeting the actual dispatching needs of the power grid. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 This is a schematic diagram of the system structure connection of the present invention.

[0022] Figure 2 This is a schematic diagram of the implementation steps of the method of the present invention. Detailed Implementation

[0023] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0024] Examples of embodiments of the present invention Figure 1 As shown, a wind power forecasting method based on meteorological data includes the following steps: Step 1: Generation of multi-scenario parameter combinations: Determine the set of extreme weather types for the target wind farm and construct a basic database of extreme weather-power. Then, based on the sensitivity of the core parameters of each extreme weather type, determine the key parameter dimensions of each extreme weather type and generate multi-scenario parameter combinations.

[0025] In a specific embodiment, the construction process of the basic database of extreme weather and power is as follows: S2.1, Data collection scope: Collect historical data of the target wind farm area for more than 10 years. The data types include extreme weather data, wind power data of the corresponding time period, and wind turbine operating status data. Among them, extreme weather data includes three types of meteorological data: strong gusts, thunderstorms, and low temperature icing. Wind power data includes the real-time power of the wind farm grid connection point. Wind turbine operating status data includes blade angle, speed, and pitch system data.

[0026] S2.2 Data Preprocessing: Outlier removal is performed on the collected historical data. The removal criteria are values ​​that exceed the historical mean ± 3 times the standard deviation of the data type. Missing data is supplemented by linear interpolation with an interpolation interval of ≤30 minutes and an interpolation error of ≤5%.

[0027] S2.3 Data Association and Storage: Extreme weather data, wind power data, and wind turbine operating status data are associated one by one according to timestamp to ensure data matching under the same timestamp; a relational database is used to store the associated data to build an extreme weather-power basic database corresponding to extreme weather parameters-wind power-wind turbine status.

[0028] It should be noted that during the data acquisition phase, extreme weather data needs to integrate monitoring data from wind towers at different heights, weather radars, and regional meteorological stations within the target wind farm. For strong gusts, the focus is on collecting 10-minute average wind speeds and instantaneous maximum wind speeds. For thunderstorms and strong winds, the data simultaneously records pressure fluctuations and precipitation intensity. For low-temperature icing, hourly temperature, humidity, and icing thickness observation data are included. Wind power data needs to be connected to the grid-connected point's second-level acquisition terminal of the power grid dispatching system. Wind turbine operating status data is captured in real time through the SCADA system, ensuring blade angle accuracy to 0.1° and speed data sampling frequency not less than 1Hz. In the preprocessing stage, for high-frequency fluctuating data such as strong gusts, historical mean and standard deviation need to be calculated first according to meteorological event type, and then outlier removal is performed. Before linear interpolation, the data continuity needs to be checked by root mean square error. For periods with interpolation errors exceeding 5%, secondary correction is performed using surrounding grid data output from the WRF model. Data association is uniformly calibrated using UTC timestamps, with deviations controlled within 1 minute. The relational database adopts a three-level table structure of "meteorological event-wind turbine cluster-time series". The stored fields include extreme values ​​of meteorological parameters, power fluctuation amplitude, and equipment status codes, and support multi-dimensional retrieval by event type, occurrence time, and wind turbine number.

[0029] In a specific embodiment, the process of determining the key parameter dimensions for each extreme weather type and generating multi-scenario parameter combinations is as follows: S3.1, Determining key parameter dimensions: Using the WRF meteorological numerical model, sensitivity analysis is performed on the core parameters of each extreme weather type. First, extreme event samples from historical data that cause wind power fluctuations exceeding 30% are selected. Then, the correlation between each meteorological parameter and power fluctuation in the samples is ranked, and the top three parameters with the highest correlation are taken as the key parameter dimensions for that extreme weather type. Thus, the key parameters for strong gusts are peak gust, duration, and wind speed rise rate; for thunderstorms, they are turbulence intensity, peak wind speed, and wind direction deflection angle; and for low-temperature icing, they are air temperature, relative humidity, and icing duration.

[0030] S3.2 Scenario parameter gradient setting: Based on the historical value range of each key parameter in the extreme weather-power database, the gradient value is set at one-third intervals of the historical standard deviation of each key parameter.

[0031] S3.3 Multi-scenario combination generation: The gradient values ​​of key parameters for each extreme weather type are fully permuted to generate multi-scenario parameter combinations; the number of scenarios for each extreme weather type is ≥9 sets, and they cover more than 90% of the historical value range of key parameters for each extreme weather type.

[0032] It should be noted that when using the WRF meteorological numerical model for sensitivity analysis, extreme meteorological data is first interpolated to the target wind farm area at a grid resolution of 1km×1km. Combined with the wind turbine hub height (80-120m), meteorological parameters at the corresponding height are extracted. The values ​​of individual parameters are adjusted one by one using the controlled variable method, while the remaining parameters are fixed at their historical averages. The fluctuation range of wind power caused by changes in each parameter is obtained. In setting the scenario parameter gradient, the historical value range of each key parameter in the basic database must first be statistically analyzed. The standard deviation of all data within the range is calculated. After determining the gradient interval by one-third of the standard deviation, if the difference between adjacent gradient values ​​is less than 0.5m / s or 0.5℃, adjacent gradients are merged to simplify the calculation. At the same time, it is ensured that the gradient values ​​cover the minimum and maximum values ​​of the historical value range without boundary truncation. When generating multiple scenarios, the gradient values ​​of each key parameter are first validated for reasonableness, and extreme gradient values ​​that exceed engineering realities are removed, such as gradient values ​​with an icing duration exceeding 72 hours. Then, a full permutation is performed. If the number of scenarios after permutation is too large, representative scenarios are selected based on the K-means clustering algorithm, and the scenario combinations corresponding to the cluster centers are retained to ensure that the number of scenarios is ≥9 while reducing computational complexity. After the combined generation, the historical matching degree of each scenario group needs to be verified. It is required that a single scenario can find at least 3 similar historical samples in the basic database to ensure the actual reference value of the scenario. Ultimately, this achieves comprehensive coverage of more than 90% of the historical value range of key parameters for each extreme weather type.

[0033] Step 2: Output of the scenario-power time series matrix: Select historical samples with the same extreme weather as the current target wind farm from the extreme weather-power basic database, extract the wind power time series data of the samples, align the curves using the dynamic time warping algorithm, take the mean value to obtain the change process of the extreme weather power of the current target wind farm over time, and output the scenario-power time series matrix.

[0034] In a specific embodiment, the extraction process of wind power time series data is as follows: From the extreme weather-power database, historical samples with the same extreme weather type as the current target wind farm and a deviation of ≤10% between key parameters and multi-scenario parameter combinations are obtained. Wind power time series data during the occurrence of extreme weather events are extracted from each historical sample, with the extraction duration consistent with the duration of the current scenario parameter combination. The extracted wind power time series data is converted into a unified format, with time on the horizontal axis and power value on the vertical axis.

[0035] It should be noted that when extracting wind power time-series data from each historical sample, the start time of the extreme weather event should be used as the time anchor point—for strong gusts and thunderstorms, the starting point is the moment when the wind speed first reaches the extreme threshold, such as a gust peak of ≥17 m / s; for low-temperature icing, the starting point is the moment when the temperature first drops below 0℃ and the relative humidity is ≥85%. The extraction duration must strictly match the duration of the current scenario parameter combination. For example, if the scenario duration is 10 minutes, then 10 consecutive minutes of power data should be extracted from the start time. If the actual duration of the event is shorter than the extraction duration, the data should be extended to the set duration after the event ends, and the extended portion of the data should be marked as post-event stable data. When converting the extracted power time-series data into a unified format, it is necessary to specify that the horizontal axis time unit is minutes and the time interval is fixed at 1 minute. If the interval is less than 1 minute, it should be padded to the whole minute node by interpolation. The vertical axis power value should retain integer places, and the origin of the horizontal axis should be uniformly set to the start time of the duration of the scene parameter combination (t=0) to ensure that the time axis benchmark of different samples is consistent. At the same time, the wind turbine operating status corresponding to the historical samples should be checked synchronously during the extraction process. If there are power mutations caused by non-meteorological factors such as wind turbine shutdown or pitch failure during the sample period, and the mutation amplitude exceeds the normal fluctuation range by 50%, it should be marked as an abnormal sample and removed. Finally, it is necessary to ensure that the number of effective samples is ≥5 groups.

[0036] In a specific embodiment, the output scenario-power time series matrix is ​​output in the following specific process: S5.1 Curve alignment: Take the wind power time series curve of a certain set of historical samples in extreme weather of the target wind farm as the reference curve, denoted as time length T. Then take the power time series curves of other similar samples, denoted as time length T1, and adjust the power time series curves T1 of other similar samples to time length T. Then calculate the sum of squares of the power difference between the adjusted curve and the reference curve at each time point. Find the path that minimizes the sum of squares of the difference through dynamic programming to complete the curve alignment.

[0037] S5.2 Mean Calculation: For all aligned power time-series curves of the same type, calculate the power mean at each time point. The power mean calculated at each time point is the arithmetic mean of the power values ​​of all samples at each time point. This yields the time-power mean curve corresponding to the parameter combination of the extreme weather scenario of the target wind farm, which is the change process of the extreme weather power of the current target wind farm over time.

[0038] S5.3 Matrix Construction and Output: Combine the time-power mean curves corresponding to the extreme weather scenario parameters of the target wind farm, and construct a two-dimensional scenario-power time series matrix according to the dimensions of scenario number-time point-power mean; the rows of the power time series matrix represent scenario numbers, the columns represent time points, and the elements of the power time series matrix are the power mean of the corresponding scenario at the corresponding time point; output the scenario-power time series matrix in Excel or CSV format.

[0039] It should be noted that in the curve alignment process, the reference curve must be selected from the same historical samples with no power abrupt changes, data integrity ≥98%, and the smallest deviation from the scene parameter combination to ensure the reliability of the reference benchmark. When adjusting the length of the power time series curve T1 of other similar samples, if T1 < T, linear interpolation is used to complete the power values ​​at missing time points; if T1 > T, the core time period is extracted at equal time intervals, prioritizing the retention of power data segments corresponding to the peak values ​​of extreme weather intensity to avoid data interference from non-critical time periods. When dynamically planning to find the path with the minimum sum of squared differences, a threshold for the sum of squared differences must be set, ≤10% of the power mean of all samples. If the sum of squared differences between a sample and the reference curve after adjustment exceeds the threshold, it is marked as an abnormal curve and a new reference curve is selected or the abnormal sample is removed. In the mean calculation stage, before calculating the arithmetic mean at each time point, outliers exceeding ±2 standard deviations of the power mean of all samples at that time point must be removed. The mean is then calculated based on the valid samples to avoid extreme outliers affecting the accuracy of the curve trend. If the number of valid samples at a certain time point is less than 3, linear interpolation of the mean between two adjacent time points is used to supplement the curve, ensuring the continuity of the time-power mean curve. In the matrix construction and output stage, scene numbers are named according to the meteorological type code-scene sequence number rule, such as "QG-01, QG-02…" for strong gusts. Time point columns are labeled in the format t=0min, t=1min…t=Tmin, where T is the reference curve time length. The power mean is retained to one decimal place. The output Excel or CSV file must include a data description header, clearly defining the matrix dimensions, power units, time intervals, and data sources. It should also be stored according to meteorological type, such as creating a separate "strong gust scene-power time series matrix.csv" file for easy retrieval and data traceability during subsequent probability density function fitting.

[0040] Step 3: Obtaining the overall power probability distribution: The historical frequency of extreme weather events at the current target wind farm is statistically analyzed from the extreme weather-power database as the prior probability. The deviation rate is calculated by combining the initial parameters of extreme weather events at the current target wind farm in numerical weather forecasts. The prior probability is adjusted to obtain the posterior probability as the weight of the extreme weather scenario at the current target wind farm. The power time series data is fitted using Cornish-Fisher series to construct the probability density function. The weighted fusion is then used to obtain the overall power probability distribution of extreme weather events at the current target wind farm. In a specific embodiment, the deviation rate is calculated by combining the initial extreme weather parameters of the current target wind farm in numerical weather forecast, and the prior probability is adjusted to obtain the posterior probability as the scenario weight. The specific adjustment process is as follows: S6.1, Prior probability calculation: From the extreme weather-power base database, the historical occurrence frequency of the scenario parameter combination of the current target wind farm extreme weather type is counted. The ratio of the scenario occurrence frequency to the total historical occurrence frequency of extreme weather of the current target wind farm is used as the prior probability of the scenario parameter combination of the current target wind farm extreme weather type.

[0041] It should be noted that in the prior probability calculation stage, when counting the number of historical occurrences of scenario parameter combinations, the statistical period should be limited to the past 10 years, which is consistent with the data collection period of the basic database. Only valid extreme events under normal wind turbine operation should be counted, and non-meteorological driven events caused by wind turbine failure and maintenance should be excluded to ensure that the statistical results can truly reflect the historical frequency of the scenario. If a scenario parameter combination has no historical occurrence record, its prior probability is uniformly set to 0.5%.

[0042] S6.2 Deviation Rate Calculation: Extract the initial parameters of the extreme weather of the current target wind farm from the numerical weather forecast, and calculate the deviation rate between the initial parameters of the extreme weather of the current target wind farm and the historical average of the initial parameters of the extreme weather type of the current target wind farm in the extreme weather-power base database. The deviation rate calculation formula is: Deviation rate = (current initial parameter value - historical average) / historical average × 100%.

[0043] It should be noted that in the deviation rate calculation stage, the initial parameters extracted from numerical weather prediction must be completely matched with the key parameter dimensions of the current extreme weather type. For example, if strong gusts are predicted, the initial values ​​of gust peak value, duration, and wind speed rise rate should be extracted. The extraction time is the numerical prediction data of 12 hours before the prediction date, and the validity of the initial parameters must be verified. Outliers exceeding the historical value range by ±20% should be removed, and linear interpolation of forecast data from adjacent time periods should be used to supplement them. When calculating the deviation rate, if there are multiple key initial parameters, the deviation rate of each parameter should be calculated separately, and the arithmetic mean of the deviation rates of all parameters should be taken as the final deviation rate to avoid adjustment deviation caused by the deviation of a single parameter.

[0044] S6.3, Posterior Probability Adjustment: If the deviation rate > 0, the posterior probability of the scenario with high parameter value = prior probability × (1 + adjustment coefficient). If the deviation rate ≤ 0, the posterior probability of the scenario with low parameter value = prior probability × (1 - adjustment coefficient). The adjustment coefficient is determined based on the deviation rate. For every 10% increase in the deviation rate, the adjustment coefficient increases by 1%. The range of the adjustment coefficient is 5%-10%.

[0045] It should be noted that in the posterior probability adjustment process, the criteria for defining high and low parameter value scenarios are based on the historical average of the key parameters for that extreme weather type. Scenarios with parameter values ​​higher than the historical average are considered high parameter value scenarios, while those lower than or equal to the historical average are considered low parameter value scenarios. When the adjustment coefficient is calculated by increasing the adjustment coefficient by 1% for every 10% increase in the deviation rate, if the deviation rate is negative, the adjustment coefficient is determined according to the rule of absolute value correspondence. For example, a deviation rate of -15% corresponds to an adjustment coefficient of 6.5%, and the adjustment coefficient is strictly limited to the range of 5%-10%. When the deviation rate exceeds the corresponding range, the adjustment coefficient is taken from the endpoint of the range. After adjustment, the posterior probabilities of all scenarios need to be normalized. If the posterior probability of a certain scenario after adjustment is ≤0.1%, it is merged into the scenario with the closest parameter value to ensure that the weight allocation both conforms to the real-time weather trend and meets the normative requirements of probability distribution.

[0046] In a specific embodiment, the weighted fusion obtains the overall power probability distribution of the current target wind farm under extreme weather conditions. The specific process is as follows: S7.1, Probability density function construction: The power time series data of the scene-power time series matrix is ​​fitted using the Cornish-Fisher series; first, the statistics of the scene power data are calculated, including skewness S and kurtosis K, and then substituted into the Cornish-Fisher series formula: ,in, To find the α quantiles of the standard normal distribution, with α taking values ​​of 90% and 95%, we obtain the α quantiles of the power probability density function at each time point. .

[0047] S7.2 Weighted Fusion: Based on the scene weights of the posterior probability obtained in 6.3, the power probability density functions of the current target wind farm at various time points under extreme weather conditions are weighted and fused; the fusion formula is: ,in, This indicates the number corresponding to each time point. , Let be the power probability density function after weighting the extreme weather scenarios for the target wind farm at time t. As the weight of the scene, Let t be the power probability density function of the target wind farm under extreme weather conditions at time t.

[0048] S7.3 Distributed Output: By Time Point The output is the overall power probability density function after fusion. The function is presented in the form of a table of time-power value-probability density, or in the form of a probability distribution curve for visualization, which is the overall power probability distribution of the current target wind farm under extreme weather conditions.

[0049] Step 4: Determination of wind power fluctuation domain: Extract the upper and lower quantiles of the overall power probability distribution of the current target wind farm under extreme weather conditions at 90% and 95% confidence levels through weighted fusion, thereby determining the wind power fluctuation domain of the current target wind farm. Calculate the root mean square of the fluctuation domain coverage and bandwidth of the current target wind farm. At the same time, label the confidence level and complete the output.

[0050] In a specific embodiment, the determination of the wind power fluctuation domain of the current target wind farm is specifically determined as follows: S8.1, Quantile extraction: For the weighted fusion of the overall power probability distribution of the current target wind farm under extreme weather conditions, extract the lower and upper quantiles at a 90% confidence level and the lower and upper quantiles at a 95% confidence level at each time point; the quantile extraction is based on the corrected quantiles obtained by Cornish-Fisher series fitting.

[0051] It should be noted that in the quantile extraction process, when using the corrected quantiles obtained from Cornish-Fisher series fitting, it is necessary to first confirm that the fitting error is ≤3%, and then extract the quantiles one by one according to the time point. At a 90% confidence level, the lower quantile corresponds to a power value with a cumulative probability of 10%, and the upper quantile corresponds to a power value with a cumulative probability of 90%. At a 95% confidence level, the lower quantile corresponds to a power value with a cumulative probability of 5%, and the upper quantile corresponds to a power value with a cumulative probability of 95%. During the extraction process, the power value is retained to one decimal place, and the quantile must fall within the range of 0-120% of the rated power of the wind farm. If it exceeds the range, it is corrected according to the boundary value to avoid power values ​​that are not physically meaningful.

[0052] S8.2, Fluctuation Domain Definition: The lower quantile to the upper quantile at the 90% confidence level at the same time point is defined as the current target wind farm's wind power fluctuation domain at the 90% confidence level, and the lower quantile to the upper quantile at the 95% confidence level is defined as the current target wind farm's wind power fluctuation domain at the 95% confidence level.

[0053] It should be noted that during the wave range definition process, logical verification must be performed on the quantiles at the same time point: the lower quantile at the 90% confidence level must not be lower than the lower quantile at the 95% confidence level, and the upper quantile at the 90% confidence level must not be higher than the upper quantile at the 95% confidence level. If a logical conflict occurs, the original corrected quantiles fitted by the Cornish-Fisher series should be used as the benchmark to re-verify the fusion validity of the overall power probability distribution to ensure that the inclusion relationship of the wave range is reasonable.

[0054] S8.3 Fluctuation Domain Integration: In chronological order, from the start time of the prediction to the end time of the prediction, the wind power fluctuation domains of all time points are integrated to form a continuous fluctuation domain curve of time-90% confidence fluctuation domain and 95% confidence fluctuation domain. The horizontal axis of the curve is time and the vertical axis is power value, which intuitively presents the change of the wind power fluctuation domain of the current target wind farm over time.

[0055] It should be noted that in the fluctuation domain integration process, the time sequence is arranged in minute intervals from the prediction start time to the prediction end time, consistent with the time interval of the scenario-power time series matrix, and ≤10 minutes. If the fluctuation domain upper and lower limits change by ≥20% between adjacent time points, linear interpolation is used to smooth the transition, and the change after smoothing is ≤15% to avoid abrupt changes in the fluctuation domain curve. The integrated continuous fluctuation domain curve needs to be labeled with the time axis unit, power axis unit, and confidence level indicator, and supplementary curve data descriptions should be provided to clarify the calculation basis and fitting error of each time point quantile, so as to provide traceable technical support for subsequent coverage calculation and scheduling decisions.

[0056] It should also be noted that determining the power fluctuation range of the current target wind farm has several advantages. First, it provides a quantified power fluctuation boundary for grid dispatch—a fluctuation range with 90% and 95% dual confidence levels. This allows dispatchers to clearly define the probable range of wind power changes under extreme weather conditions, facilitating the precise reservation of reserve capacity for regulating resources such as thermal power and energy storage. This reduces the risk of grid frequency fluctuations caused by sudden power changes and improves the stability of the power system.

[0057] Secondly, by integrating the fluctuation domain at different time points to form a continuous curve, the temporal characteristics of power fluctuations can be intuitively presented, such as the rapid rise in power at the beginning of strong gusts and the slow decline in power at the end of icing. This helps wind farms to formulate wind turbine operation and maintenance strategies in advance, such as adjusting the pitch angle before the power surge, reducing the impact of extreme weather on equipment, and improving the safety of wind turbine operation.

[0058] Finally, by combining quantitative indicators such as fluctuation domain coverage and root mean square bandwidth, the reliability and volatility of the prediction results can be objectively assessed: coverage reflects the actual matching degree of the fluctuation domain, and root mean square bandwidth reflects the average dispersion of fluctuations. Together, they provide data support for the iterative optimization of the prediction method and continuously improve the practical value of wind power prediction.

[0059] In a specific embodiment, the calculation of the current target wind farm fluctuation domain coverage and bandwidth root mean square, and the output after labeling the confidence level, are as follows: S9.1, Coverage calculation: From the extreme weather-power database, the predicted samples of the current target wind farm under extreme weather types in the past year are selected. The predicted fluctuation domain of each sample is compared with the measured wind power value in the same period. The number of times the measured power value falls into the predicted fluctuation domain is counted. The ratio of the number of times it falls into the predicted fluctuation domain to the total number of samples is used as the current target wind farm fluctuation domain coverage.

[0060] S9.2 Root Mean Square Calculation of Bandwidth: For the predicted wind power fluctuation domain of the current target wind farm, calculate the fluctuation domain bandwidth at each time point. The fluctuation domain bandwidth = upper quantile - lower quantile. Then calculate the square mean of the bandwidth at all time points, and take the square root of the square mean to obtain the root mean square of the bandwidth. The calculation formula is as follows: Where bandwidth t is the fluctuation domain bandwidth at time t, and n is the total number of time points.

[0061] S9.3 Result Labeling and Output: In the wind power fluctuation domain curve of the current target wind farm, clearly label the 90% confidence level and 95% confidence level of the fluctuation domain. At the same time, label the fluctuation domain coverage and root mean square bandwidth; and represent them through visualization charts.

[0062] Examples of embodiments of the present invention Figure 2 As shown, a wind power forecasting system based on meteorological data includes the following modules: a multi-scenario parameter combination generation module: used to determine the set of extreme weather types of the target wind farm and build a basic database of extreme weather-power, and then determine the key parameter dimensions of each extreme weather type according to the sensitivity of the core parameters of each extreme weather type, and generate multi-scenario parameter combinations.

[0063] The scenario-power time series matrix output module is used to filter historical samples of the same type as the current target wind farm's extreme weather from the extreme weather-power basic database, extract the sample wind power time series data, align the curves using a dynamic time warping algorithm, take the mean value to obtain the change process of the current target wind farm's extreme weather power over time, and output the scenario-power time series matrix.

[0064] The overall power probability distribution acquisition module is used to statistically analyze the historical frequency of extreme weather events at the current target wind farm from the extreme weather-power basic database as the prior probability. It calculates the deviation rate by combining the initial parameters of extreme weather events at the current target wind farm in numerical weather forecasts, adjusts the prior probability to obtain the posterior probability as the weight of the extreme weather scenario at the current target wind farm, uses Cornish-Fisher series to fit the power time series data to construct the probability density function, and weighted fusion to obtain the overall power probability distribution of extreme weather events at the current target wind farm.

[0065] The wind power fluctuation domain determination module is used to extract the upper and lower quantiles of the overall power probability distribution under extreme weather conditions of the current target wind farm obtained by weighted fusion at 90% and 95% confidence levels, thereby determining the wind power fluctuation domain of the current target wind farm, calculating the root mean square of the fluctuation domain coverage and bandwidth of the current target wind farm, and completing the output after labeling the confidence level.

[0066] The examples described in this invention are not limited to the specific embodiments listed above. The examples are merely illustrative to facilitate understanding of the invention and do not constitute a limitation on the scope of protection of this invention. Any modifications, equivalent substitutions, etc., made within the spirit and principles of this invention should be included within the scope of protection.

[0067] The above description is merely an example and illustration of the concept of the present invention. Those skilled in the art can make various modifications or additions to the specific embodiments described or use similar methods to replace them, as long as they do not deviate from the concept of the invention or exceed the scope defined in this specification, they should all fall within the protection scope of the present invention.

Claims

1. A wind power forecasting method based on meteorological data, characterized in that, Includes the following steps: Step 1: Generation of multi-scenario parameter combinations: Determine the set of extreme weather types for the target wind farm and build a basic database of extreme weather and power. Then, based on the sensitivity of the core parameters of each extreme weather type, determine the key parameter dimensions of each extreme weather type and generate multi-scenario parameter combinations. Step 2, Output of the Scene-Power Time Series Matrix: Select historical samples with the same extreme weather as the current target wind farm from the extreme weather-power database, extract the wind power time series data of the samples, align the curves using the dynamic time warping algorithm, take the mean value to obtain the change process of the extreme weather power of the current target wind farm over time, and output the scene-power time series matrix. Step 3: Obtaining the overall power probability distribution: The historical frequency of extreme weather events at the current target wind farm is statistically analyzed from the extreme weather-power database as the prior probability. The deviation rate is calculated by combining the initial parameters of extreme weather events at the current target wind farm in numerical weather forecasts. The prior probability is adjusted to obtain the posterior probability as the weight of the extreme weather scenario at the current target wind farm. The power time series data is fitted using Cornish-Fisher series to construct the probability density function. The weighted fusion is then used to obtain the overall power probability distribution of extreme weather events at the current target wind farm. Step 4: Determination of wind power fluctuation domain: Extract the upper and lower quantiles of the overall power probability distribution of the current target wind farm under extreme weather conditions at 90% and 95% confidence levels through weighted fusion, thereby determining the wind power fluctuation domain of the current target wind farm. Calculate the root mean square of the fluctuation domain coverage and bandwidth of the current target wind farm. At the same time, label the confidence level and complete the output.

2. The wind power forecasting method based on meteorological data according to claim 1, characterized in that, The specific construction process for building the basic database of extreme weather and power is as follows: S2.1 Data Collection Scope: Collect historical data for the past 10 years or more for the area where the target wind farm is located. Data types include extreme weather data, wind power data for the corresponding time period, and wind turbine operating status data. Among them, extreme weather data includes three types of meteorological data: strong gusts, thunderstorms, and low temperature icing. Wind power data includes the real-time power of the wind farm's grid connection point. Wind turbine operating status data includes blade angle, speed, and pitch system data. S2.2 Data Preprocessing: Outlier removal is performed on the collected historical data. The removal criterion is that the value exceeds the historical mean of the data type by ±3 times the standard deviation. Missing data is supplemented by linear interpolation with an interpolation interval of ≤30 minutes and an interpolation error of ≤5%. S2.3 Data Association and Storage: Extreme weather data, wind power data, and wind turbine operating status data are associated one by one according to timestamp to ensure data matching under the same timestamp; a relational database is used to store the associated data to build an extreme weather-power basic database corresponding to extreme weather parameters-wind power-wind turbine status.

3. The wind power forecasting method based on meteorological data according to claim 2, characterized in that, The process of determining the key parameter dimensions for each extreme weather type and generating multi-scenario parameter combinations is as follows: S3.1 Determination of Key Parameter Dimensions: Sensitivity analysis of core parameters for each extreme weather type was conducted using the WRF meteorological numerical model. First, extreme event samples from historical data that caused wind power fluctuations exceeding 30% were selected. Then, the correlation between each meteorological parameter and power fluctuation in the samples was ranked, and the top three parameters with the highest correlation were taken as the key parameter dimensions for that extreme weather type. Thus, the key parameters for strong gusts were peak gust, duration, and wind speed rise rate; for thunderstorms, they were turbulence intensity, peak wind speed, and wind direction deflection angle; and for low-temperature icing, they were air temperature, relative humidity, and icing duration. S3.2, Scene parameter gradient setting: Based on the historical value range of each key parameter in the extreme weather-power database, the gradient value is set at one-third intervals of the historical standard deviation of each key parameter; S3.3 Multi-scenario combination generation: The gradient values ​​of key parameters for each extreme weather type are fully permuted to generate multi-scenario parameter combinations; the number of scenarios for each extreme weather type is ≥9 sets, and they cover more than 90% of the historical value range of key parameters for each extreme weather type.

4. The wind power forecasting method based on meteorological data according to claim 3, characterized in that, The specific extraction process for the wind power time series data is as follows: From the extreme weather-power database, historical samples with the same extreme weather type as the current target wind farm and a deviation of ≤10% between key parameters and multi-scenario parameter combinations are obtained. Wind power time series data during the extreme weather event is extracted from each historical sample, with the extraction duration consistent with the duration of the current scenario parameter combination. The extracted wind power time series data is converted into a unified format, with time on the horizontal axis and power value on the vertical axis.

5. The wind power forecasting method based on meteorological data according to claim 4, characterized in that, The output scenario – the power timing matrix – is output in the following specific process: S5.1 Curve Alignment: Using the wind power time series curve of a certain set of historical samples in extreme weather conditions of the target wind farm as the reference curve, denoted as time length T, the power time series curves of other similar samples are denoted as time length T1, and the power time series curves T1 of other similar samples are adjusted to time length T; then the sum of squares of the power difference between the adjusted curve and the reference curve at each time point is calculated, and the path that minimizes the sum of squares of the difference is found through dynamic programming to complete the curve alignment; S5.2 Mean Calculation: For all aligned power time-series curves of the same type of sample, calculate the power mean at each time point. The power mean calculated at each time point is the arithmetic mean of the power values ​​of all samples at each time point. This yields the time-power mean curve corresponding to the parameter combination of the extreme weather scenario of the target wind farm, which is the change process of the extreme weather power of the current target wind farm over time. S5.3 Matrix Construction and Output: Combine the time-power mean curves corresponding to the extreme weather scenario parameters of the target wind farm, and construct a two-dimensional scenario-power time series matrix according to the dimensions of scenario number-time point-power mean; the rows of the power time series matrix represent scenario numbers, the columns represent time points, and the elements of the power time series matrix are the power mean of the corresponding scenario at the corresponding time point; output the scenario-power time series matrix in Excel or CSV format.

6. The wind power forecasting method based on meteorological data according to claim 5, characterized in that, The deviation rate is calculated by combining the initial extreme weather parameters of the current target wind farm in numerical weather prediction, and the prior probability is adjusted to obtain the posterior probability as the scene weight. The specific adjustment process is as follows: S6.1 Prior probability calculation: From the extreme weather-power base database, count the historical occurrences of the current target wind farm extreme weather type scenario parameter combination, and use the ratio of the scenario occurrences to the total historical occurrences of extreme weather at the current target wind farm as the prior probability of the current target wind farm extreme weather type scenario parameter combination; S6.2 Deviation Rate Calculation: Extract the initial parameters of the extreme weather of the current target wind farm from the numerical weather forecast, and calculate the deviation rate between the initial parameters of the current target wind farm's extreme weather and the historical average of the initial parameters of the current target wind farm's extreme weather type in the extreme weather-power base database. The deviation rate calculation formula is: Deviation Rate = (Current Initial Parameter Value - Historical Average) / Historical Average × 100%; S6.3, Posterior Probability Adjustment: If the deviation rate > 0, the posterior probability of the scenario with high parameter value = prior probability × (1 + adjustment coefficient). If the deviation rate ≤ 0, the posterior probability of the scenario with low parameter value = prior probability × (1 - adjustment coefficient). The adjustment coefficient is determined based on the deviation rate. For every 10% increase in the deviation rate, the adjustment coefficient increases by 1%. The range of the adjustment coefficient is 5%-10%.

7. The wind power forecasting method based on meteorological data according to claim 6, characterized in that, The weighted fusion yields the overall power probability distribution of the current target wind farm under extreme weather conditions, and the specific process is as follows: S7.1, Probability Density Function Construction: For the power time series data of the scene-power time series matrix, Cornish-Fisher series is used for fitting; first, the statistics of the scene power data are calculated, including skewness S and kurtosis K, and then substituted into the Cornish-Fisher series formula: ,in, To find the α quantiles of the standard normal distribution, with α taking values ​​of 90% and 95%, we obtain the α quantiles of the power probability density function at each time point. ; S7.2 Weighted Fusion: Based on the scene weights of the posterior probability obtained in 6.3, the power probability density functions of the current target wind farm at various time points under extreme weather conditions are weighted and fused; the fusion formula is: ,in, This indicates the number corresponding to each time point. , Let be the power probability density function after weighting the extreme weather scenarios for the target wind farm at time t. As the weight of the scene, Let t be the power probability density function of the target wind farm under extreme weather conditions at time t; S7.3 Distributed Output: By Time Point The output is the overall power probability density function after fusion. The function is presented in the form of a table of time-power value-probability density, or in the form of a probability distribution curve for visualization, which is the overall power probability distribution of the current target wind farm under extreme weather conditions.

8. The wind power forecasting method based on meteorological data according to claim 7, characterized in that, The specific process for determining the wind power fluctuation range of the current target wind farm is as follows: S8.1, Quantile Extraction: For the overall power probability distribution of the current target wind farm under extreme weather conditions obtained by weighted fusion, extract the lower and upper quantiles at the 90% confidence level and the lower and upper quantiles at the 95% confidence level at each time point. Quantile extraction was based on the corrected quantiles obtained from Cornish-Fisher series fitting; S8.2, Fluctuation Domain Definition: The lower quantile to the upper quantile at the 90% confidence level at the same time point is defined as the current target wind farm's wind power fluctuation domain at the 90% confidence level, and the lower quantile to the upper quantile at the 95% confidence level is defined as the current target wind farm's wind power fluctuation domain at the 95% confidence level. S8.3 Fluctuation Domain Integration: In chronological order, from the start time of the prediction to the end time of the prediction, the wind power fluctuation domains of all time points are integrated to form a continuous fluctuation domain curve of time-90% confidence fluctuation domain and 95% confidence fluctuation domain. The horizontal axis of the curve is time and the vertical axis is power value, which intuitively presents the change of the wind power fluctuation domain of the current target wind farm over time.

9. A wind power forecasting method based on meteorological data according to claim 8, characterized in that, The calculation of the current target wind farm's fluctuation domain coverage and root mean square bandwidth, along with the labeling of confidence levels, is completed and output. The specific output process is as follows: S9.1 Coverage Calculation: From the extreme weather-power base database, select the predicted samples of the current target wind farm under extreme weather types in the past year, compare the predicted fluctuation range of each sample with the measured wind power value in the same period, count the number of times the measured power value falls into the predicted fluctuation range, and use the ratio of the number of times it falls into the predicted fluctuation range to the total number of samples as the fluctuation range coverage of the current target wind farm. S9.2 Root Mean Square Calculation of Bandwidth: For the wind power fluctuation domain of the current target wind farm predicted now, calculate the fluctuation domain bandwidth at each time point. Fluctuation domain bandwidth = upper quantile - lower quantile. Next, calculate the square mean of the bandwidth at all time points, and take the square root of the square mean to obtain the root mean square of the bandwidth. The calculation formula is as follows: Where bandwidth t is the fluctuation domain bandwidth at time t, and n is the total number of time points; S9.3 Result Labeling and Output: In the wind power fluctuation domain curve of the current target wind farm, clearly label the 90% confidence level and 95% confidence level of the fluctuation domain. At the same time, label the fluctuation domain coverage and root mean square bandwidth; and represent them through visualization charts.

10. A system that utilizes the wind power forecasting method based on meteorological data according to any one of claims 1-9, characterized in that, Includes the following modules: Multi-scenario parameter combination generation module: used to determine the set of extreme weather types of the target wind farm and build a basic database of extreme weather-power. Then, based on the sensitivity of the core parameters of each extreme weather type, it determines the key parameter dimensions of each extreme weather type and generates multi-scenario parameter combinations. Scene-Power Time Series Matrix Output Module: This module is used to filter historical samples similar to the extreme weather of the current target wind farm from the extreme weather-power database, extract the wind power time series data of the samples, align the curves using a dynamic time warping algorithm, take the mean value to obtain the change process of the extreme weather power of the current target wind farm over time, and output the scene-power time series matrix. The overall power probability distribution acquisition module is used to statistically analyze the historical frequency of extreme weather events at the current target wind farm from the extreme weather-power basic database as the prior probability, calculate the deviation rate by combining the initial parameters of extreme weather events at the current target wind farm in numerical weather forecast, adjust the prior probability to obtain the posterior probability as the weight of the extreme weather scenario at the current target wind farm, use Cornish-Fisher series to fit the power time series data to build the probability density function, and obtain the overall power probability distribution of extreme weather events at the current target wind farm by weighted fusion. The wind power fluctuation domain determination module is used to extract the upper and lower quantiles of the overall power probability distribution under extreme weather conditions of the current target wind farm obtained by weighted fusion at 90% and 95% confidence levels, thereby determining the wind power fluctuation domain of the current target wind farm, calculating the root mean square of the fluctuation domain coverage and bandwidth of the current target wind farm, and completing the output after labeling the confidence level.

Images