A wind power ramping prediction method and system under a complex scenario

Abstract

The application discloses a wind power climbing prediction method and system in a complex scene, and relates to the technical field of wind power prediction, and comprises the following steps: performing time sequence consistency fusion processing on historical meteorological data and historical power data after aggregation processing to generate a test data set; training an improved Informer model through the test data set to obtain a target prediction model, taking a real-time data set as the input of the target prediction model to obtain a native wind power prediction sequence; performing climbing detection on the native wind power prediction sequence and combining a bump event merging strategy to perform a first correction on an initial wind power prediction sequence to obtain a secondary wind power prediction sequence; and performing a second correction on extreme points in the secondary wind power prediction sequence based on an xLSTM model to obtain a target wind power prediction sequence. The application realizes accurate identification and early warning of the whole process of the climbing event, and effectively improves the overall precision of wind power climbing prediction in a complex scene.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of wind power prediction, in particular to a wind power ramping prediction method and system under complex scenarios. BACKGROUND

[0002] With the continuous growth of wind power installed capacity, its proportion in energy structure is increasing, but the inherent intermittency, volatility and randomness of wind power output also bring serious challenges to the stable operation of power grid. Especially in complex weather scenarios such as cold wave, typhoon and sandstorm, wind power is prone to rapid rise and fall in a short time, and if it cannot be accurately predicted, it will lead to difficulties in power grid dispatching, frequency fluctuation and even safety risk. At present, the research on wind power ramping prediction still faces some key technical bottlenecks: first, most traditional prediction models are based on stationary scenarios, and the capture ability of non-stationary and nonlinear time series characteristics under extreme weather conditions is limited, which is difficult to adapt to the dynamic changes of complex scenarios; second, the existing methods generally focus on the fitting accuracy of the overall power curve, and lack special optimization mechanism for extreme points in the ramping event, which leads to systematic deviation in the prediction of power peaks and valleys, directly affecting the accuracy of the judgment of the starting and ending time and the amplitude of the ramping; in addition, most ramping detection algorithms are prone to false detection and missed detection when facing frequent fluctuations, and lack effective filtering of short-term fluctuations and noise interference, further reducing the reliability of early warning. Therefore, developing a prediction method that can take into account long-term sequence dependence, accurately identify ramping events and effectively correct extreme value deviation has become the key to improving the ability of power grid to cope with complex scenarios.

[0003] The above information disclosed in the BACKGROUND section merely to enhance the understanding of the background of the present application, therefore it can include information that does not constitute prior art known to those of ordinary skill in the art. SUMMARY

[0004] The purpose of the present application is to solve the technical problem that the existing wind power ramping prediction method has poor adaptability to complex scenarios and low prediction accuracy, which leads to poor safety of power grid regulation. A wind power ramping prediction method and system under complex scenarios are proposed, which solves the problem of large wind power ramping prediction deviation under complex scenarios through a three-level cooperative mechanism of accurate power prediction-optimized ramping detection-extreme value deviation correction, breaks through the core limitations of weak feature capture, large extreme value deviation and low detection reliability of traditional methods under complex scenarios, realizes accurate identification and early warning of the whole process of ramping event, and effectively improves the overall accuracy of wind power ramping prediction under complex scenarios, providing reliable data guarantee for the safe operation of power grid.

[0005] In the first aspect, a technical solution provided by an embodiment of the present application is a wind power ramping prediction method under complex scenarios, comprising the following steps:

[0006] The historical meteorological data and the historical power data after the aggregation processing are subjected to time sequence consistency fusion processing to generate a test data set;

[0007] The improved Informer model is trained by using the test data set to obtain a target prediction model, and the original wind power prediction sequence is obtained by taking the real-time data set as the input of the target prediction model;

[0008] The initial wind power prediction sequence is corrected by using the bump event merging strategy to obtain a secondary wind power prediction sequence;

[0009] The extreme points in the secondary wind power prediction sequence are corrected by using the xLSTM model to obtain a target wind power prediction sequence.

[0010] Preferably, the historical meteorological data and the historical power data after the aggregation processing are subjected to time sequence consistency fusion processing to generate a test data set, which comprises the following steps:

[0011] The historical meteorological data and the historical power data on the time sequence are obtained by using the sampling frequency consistency;

[0012] The historical power data on the same time sequence is subjected to normalization processing by using the Min-Max normalization algorithm to obtain historical time sequence power data;

[0013] The historical meteorological data on the same time sequence is subjected to intensive representation by using the meteorological factor consistency to obtain historical time sequence meteorological data;

[0014] The historical time sequence meteorological data and the historical time sequence power data are subjected to time sequence consistency alignment by using the time sequence alignment algorithm to obtain a test data set.

[0015] Preferably, the improved Informer model is trained by using the test data set to obtain a target prediction model, which comprises the following steps:

[0016] The test data set is divided into a training set and a validation set according to a 1:G ratio;

[0017] The encoder and the decoder of the Informer model are reconstructed, the encoder is integrated with the ProbSparse self-attention mechanism and the self-attention distillation mechanism, and the decoder is a generative decoder;

[0018] The reconstructed Informer model is trained by using the training set to obtain an optimized prediction model, the optimized prediction model is subjected to reliability evaluation by using the validation set, and the optimized model after the reliability evaluation is taken as the target prediction model.

[0019] As preferred, the original wind power prediction sequence is detected for ramping and combined with a bump event merging strategy to modify the initial wind power prediction sequence to obtain a secondary wind power prediction sequence; comprising the following steps:

[0020] Based on the improved rotating door algorithm, the original wind power prediction sequence is detected for preliminary ramping to generate a plurality of power trend segments; according to the power change threshold δ, the plurality of power trend segments are classified to generate a sequence of original ramping trend segments;

[0021] The adjacent ramping trend segments in the sequence of original ramping trend segments are traversed and the secondary ramping trend segment sequence is generated according to the bump event merging strategy;

[0022] The starting point and the ending point in the secondary ramping trend segment sequence are extracted to construct an initial start-end point set; the nearest extreme point before the starting point and the nearest extreme point after the ending point corresponding to each ramping segment in the initial start-end point set are found;

[0023] According to the position relationship between the nearest extreme point before the starting point, the nearest extreme point after the ending point and the original start-end point, the scene correction and time sequence integration are performed to obtain a secondary wind power prediction sequence.

[0024] As preferred, the adjacent ramping trend segments in the sequence of original ramping trend segments are traversed and the secondary ramping trend segment sequence is generated according to the bump event merging strategy, comprising the following steps:

[0025] The adjacent ramping trend segments in the sequence of original ramping trend segments are traversed and the power change amplitude of the adjacent two segments is calculated, if the power change amplitude is less than the set threshold and the trend directions of the adjacent two segments are consistent, it is determined as a bump event; otherwise, the subsequent adjacent segment is continuously traversed with the current trend segment endpoint as the starting point;

[0026] After the round of traversal of the sequence of original ramping trend segments, the adjacent ramping trend segments determined as the bump event are merged to generate a secondary ramping trend segment sequence.

[0027] As preferred, according to the position relationship between the nearest extreme point before the starting point, the nearest extreme point after the ending point and the original start-end point, the scene correction and time sequence integration are performed to obtain a secondary wind power prediction sequence, comprising the following steps:

[0028] If only the extreme point before the starting point and the ending point satisfy the ramping definition, the starting point is modified to the current extreme point before the starting point;

[0029] If only the extreme point after the starting point and the ending point satisfy the ramping definition, the ending point is modified to the current extreme point after the ending point;

[0030] If the overall between the starting point and the extreme point before the starting point and the extreme point after the ending point satisfies the climbing definition, the starting point and the ending point are modified to the corresponding extreme point before the starting point and the extreme point after the ending point.

[0031] The climbing definition refers to an interval composed of consecutive same-direction trend segments identified in the original wind power prediction sequence, and the absolute value of the power change of the starting point and the ending point exceeds the set power change threshold δ, and after the bump event merging and the starting point and ending point optimization correction, the starting point and the ending point are respectively located at the local minimum and the local maximum of the power fluctuation period.

[0032] As preferred, the xLSTM model is used to correct the extreme points in the secondary wind power prediction sequence to obtain a target wind power prediction sequence; including the following steps:

[0033] The starting extreme point and the ending extreme point of each climbing segment determined after the first correction are extracted from the secondary wind power prediction sequence to construct a set of extreme points to be corrected;

[0034] An input feature vector corresponding to each extreme point is constructed as an input sample set of the xLSTM model, and the input feature vector includes: the predicted power value at the current extreme point, the real-time meteorological data corresponding to the current extreme point, and the meteorological data of each of the three sampling points before and after the current extreme point;

[0035] The xLSTM model outputs an extreme correction result, and the corresponding extreme point power value in the secondary wind power prediction sequence is replaced or adjusted based on the extreme correction result to obtain a target wind power prediction sequence;

[0036] The correction result includes: the correction amount or the corrected power value of the target extreme point.

[0037] In the second aspect, the technical scheme provided in the embodiments of the present application is: a wind power climbing prediction system, which is suitable for the wind power climbing prediction method in a complex scenario as described above, and includes:

[0038] The data acquisition module: the historical meteorological data and the historical power data after the aggregation processing are subjected to time sequence consistency fusion processing to generate a test data set;

[0039] The model construction module: the improved Informer model is trained by using the test data set to obtain a target prediction model, and the real-time data set is used as the input of the target prediction model to obtain an original wind power prediction sequence;

[0040] The first correction module: the original wind power prediction sequence is subjected to climbing detection, and the initial wind power prediction sequence is subjected to first correction by combining a bump event merging strategy to obtain a secondary wind power prediction sequence;

[0041] The secondary correction module corrects extreme points in the secondary wind power prediction sequence based on the xLSTM model to obtain a target wind power prediction sequence.

[0042] In a third aspect, an electronic device is provided, which includes a memory and a processor, the memory storing a computer program, and the processor calling the computer program in the memory to implement the steps of the wind power ramping prediction method in a complex scenario as described above.

[0043] In a fourth aspect, a storage medium is provided, which stores computer executable instructions, and the computer executable instructions are loaded and executed by a processor to implement the steps of the wind power ramping prediction method in a complex scenario as described above.

[0044] The present application has the following beneficial effects:

[0045] (1) In view of the poor adaptability of the prior art in the complex scenarios such as cold wave and typhoon, and the difficulty in capturing non-stationary and nonlinear time series characteristics, which leads to low accuracy of the original power prediction, the present application proposes a three-level collaborative mechanism scheme of "accurate power prediction-optimized ramping detection-extreme value deviation correction", which performs sampling frequency unification, Min-Max normalization, weather factor set integration representation and time series alignment time series consistency fusion processing on historical weather data and historical power data after set integration processing, to ensure the homogeneity and time synchronization of the input data, and then reconstructs the Informer model integrating the ProbSparse self-attention mechanism and the self-attention distillation mechanism of the encoder and the generative decoder to avoid the error accumulation of traditional step-by-step prediction, and finally realizes the high-precision output of the original wind power prediction sequence in the complex scenario, breaking through the technical limitations of traditional models that are only suitable for stable scenarios and lack of capturing non-stationary fluctuation characteristics.

[0046] (2) In view of the problems that the existing climbing detection algorithm is prone to false detection and missed detection in the face of bump events (short-time small amplitude oscillation) in complex scenes, and the positioning deviation of the start and end points of the climbing section is large, the application proposes a one-time correction scheme based on the improved rotating door algorithm combined with bump event merging and start and end point optimization. The power trend section is generated by first detecting the original prediction sequence based on the improved rotating door algorithm, then the original climbing trend section sequence is obtained according to the power change threshold δ classification, then the power change amplitude is calculated by traversing adjacent trend sections, and the bump event is determined and merged combined with the trend direction consistency, eliminating the interference of short-term fluctuations on climbing detection, and finally finding the extreme points closest to the start and end points of each climbing section. According to the positional relationship between the extreme points and the original start and end points, the scene is corrected, the accurate identification of the climbing section in the secondary wind power prediction sequence is realized, the false detection rate and the missed detection rate of the climbing detection are significantly reduced, and the reliability of the spatio-temporal positioning of the climbing event is improved.

[0047] (3) In view of the problem that the existing method lacks special optimization of the extreme value point of the climbing event, resulting in large extreme value prediction deviation (peak / valley prediction is not accurate) and affecting the safety of power grid dispatching, the application proposes an extreme value secondary correction scheme based on an xLSTM model. The start and end extreme value points of the climbing section determined by the one-time correction are extracted from the secondary wind power prediction sequence to construct a to-be-corrected set, the complex correlation between the extreme value points and the surrounding time series meteorological factors is fully tapped, and then the xLSTM model with sLSTM and mLSTM double variants is used to learn the mapping relationship between the extreme value point prediction power and the actual power and output the correction amount or the corrected power value. The extreme value point in the secondary sequence is replaced and adjusted to finally obtain the target wind power prediction sequence, effectively reducing the extreme value point prediction deviation, providing reliable power data support for accurate power grid dispatching, and solving the technical bottleneck of the lack of extreme value optimization in traditional methods.

[0048] The above summary of the application is only a summary of the technical solutions of the application. In order to more clearly understand the technical means of the application, the application can be implemented according to the content of the specification, and in order to make the above and other purposes, features and advantages of the application more obvious and easy to understand, the specific embodiments of the application are described below. BRIEF DESCRIPTION OF DRAWINGS

[0049] Other features, objects and advantages of the application will become more apparent from the following detailed description of non-limiting embodiments made with reference to the accompanying drawings. The drawings are for the purpose of illustrating preferred embodiments only and are not to be construed as limiting the application. Moreover, the same reference numerals are used throughout the drawings to designate the same parts.

[0050] Figure 1 A flowchart of a wind power climbing prediction method in a complex scene.

[0051] Figure 2 This is a diagram of the reconstructed Informer model architecture of the present invention.

[0052] Figure 3 This is a comparison curve of the predicted and actual wind power output under cold wave weather conditions according to the present invention.

[0053] Figure 4 This is a comparison curve of predicted power and actual power under typhoon weather conditions, as presented in this invention.

[0054] Figure 5 This is a graph showing the predicted power curve and the actual power curve under sandstorm weather conditions, as presented in this invention.

[0055] Figure 6 This is a graph showing the predicted power curve and the actual power curve under thunderstorm weather conditions, as presented in this invention.

[0056] Figure 7 This is a bar chart illustrating the evaluation analysis of the reconstructed Informer model of this invention.

[0057] Figure 8 This is a graph showing the hill-climbing detection results without considering bump events in this invention.

[0058] Figure 9 The image shows the hill-climbing detection results considering the bump event in this invention.

[0059] Figure 10 For the present invention Figure 8 and Figure 9 A comparison chart showing the results under the two scenarios.

[0060] Figure 11 This is a framework diagram of the xLSTM model of this invention.

[0061] Figure 12 The images show the slope detection results before and after the modification of this invention.

[0062] Figure 13 This is a comparison chart of the wind power prediction curves of various models in this invention under cold wave weather.

[0063] Figure 14 This is a comparison chart of the wind power prediction curves of various models in this invention under sandstorm weather.

[0064] Figure 15 This is a comparison chart of the wind power prediction curves of various models in this invention under thunderstorm weather.

[0065] Figure 16 This is a comparison chart showing the results of slope prediction using the three slope prediction models of this invention.

[0066] Figure 17 This is a schematic diagram of the wind power ramp prediction system of the present invention. Detailed Implementation

[0067] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only one preferred embodiment of this invention and are only used to explain this invention. They do not limit the scope of protection of this invention. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0068] Before discussing the exemplary embodiments in more detail, it should be mentioned that some exemplary embodiments are described as processes or methods depicted as flowcharts. Although the flowcharts describe the operations (or steps) as sequential processes, many of the operations (or steps) can be performed in parallel, concurrently, or simultaneously. Furthermore, the order of the operations can be rearranged. The process can be terminated when its operation is completed, but it may also have additional steps not included in the figures; the process may correspond to a method, function, procedure, subroutine, subroutine, etc.

[0069] Example 1: As Figure 1 As shown, a wind power ramp-up prediction method in complex scenarios includes the following steps:

[0070] S1. Perform time-series consistency fusion processing on historical meteorological data and aggregated historical power data to generate a test dataset.

[0071] As an optional embodiment, the step of performing time-series consistency fusion processing on historical meteorological data and aggregated historical power data to generate a test dataset includes the following steps:

[0072] Historical meteorological and power data in time series are obtained by sampling frequency consistency;

[0073] Historical time-series power data are obtained by normalizing historical power data at the same time series using the Min-Max normalization algorithm.

[0074] Historical time-series meteorological data are obtained by intensively representing historical meteorological data of the same time series through the consistency of meteorological factors;

[0075] The test dataset was obtained by performing time-series consistency alignment on historical time-series meteorological data and historical time-series power data using a time-series alignment algorithm.

[0076] Understandably, this embodiment first establishes a consistent sampling frequency, unifying the time series granularity of historical meteorological data and historical power data (e.g., adjusting both to a 15-minute sampling interval), ensuring a preliminary matching basis in the time series dimension. Then, the Min-Max normalization algorithm is used to map historical power data under the same time series to the [0,1] interval, eliminating magnitude interference caused by differences in installed capacity. Simultaneously, based on the principle of meteorological factor consistency, core meteorological parameters strongly correlated with wind power (such as wind speed, wind direction, and temperature) are selected and standardized, resulting in historical meteorological data with a unified format and focused information, and historical power data is transformed into directly calculable historical time series power data. Finally, a time series alignment algorithm (such as linear interpolation completion based on timestamp matching) is used to calibrate the two data series. This approach addresses the potential for minor shifts in timestamps in data (such as time-series misalignments caused by sensor acquisition delays). It achieves precise temporal correspondence between historical time-series meteorological data and historical time-series power data, enabling deep fusion and consistent representation of multi-source heterogeneous data in the spatiotemporal dimension. This results in the construction of a high-quality, highly consistent test dataset, providing a high-quality, unbiased input data foundation for subsequent improvements to the Informer model training. It avoids model learning bias caused by inconsistent data timing or disordered formats, ensuring the initial accuracy of the model in predicting wind power in complex scenarios. It is important to note that when collecting raw meteorological data (including solar irradiance, temperature, humidity, etc.) and historical power data, it is necessary to ensure that the time span and sampling frequency of the data are consistent and cover a sufficiently long time period to include different weather conditions and seasonal variations.

[0077] Furthermore, the Min-Max normalization formula is calculated as follows:

[0078] (1);

[0079] Among them, y n The normalized wind power value is y, where y is the actual wind power value. max y represents the maximum wind power output from the original data. min This represents the minimum wind power output in the original data.

[0080] S2. The improved Informer model is trained using the test dataset to obtain the target prediction model. The original wind power prediction sequence is obtained by using the real-time dataset as the input of the target prediction model.

[0081] As an optional embodiment, the step of training the improved Informer model to obtain the target prediction model using a test dataset includes the following steps:

[0082] The test dataset is divided into a training set and a validation set in a 1:G ratio.

[0083] The encoder and decoder of the Informer model are reconstructed. The encoder integrates a ProbSparse self-attention mechanism and a self-attention distillation mechanism, and the decoder is a generative decoder.

[0084] The reconstructed Informer model is trained using the training set to obtain an optimized prediction model; the optimized prediction model is then evaluated for reliability using the validation set, and the optimized model that passes the reliability evaluation is used as the target prediction model.

[0085] Understandably, this embodiment first separates the test dataset into a training set (for model parameter learning) and a validation set (for model performance verification) at a ratio of 1:G (e.g., 8:2), thus achieving a functional division of data for model "training-evaluation" and avoiding evaluation bias caused by data reuse; then, the core structure of the Informer model is reconstructed, such as... Figure 2 As shown, the encoder integrates a ProbSparse self-attention mechanism (which reduces redundancy in long-term data processing by focusing on the sparsity of the "dominant attention weights" and accurately captures the long-term dependence features of wind power in complex scenarios) and a self-attention distillation mechanism (which enhances the model's ability to extract non-stationary power fluctuation patterns by filtering key features between layers and filtering redundant information). The decoder uses a generative decoder (filling the target prediction sequence with 0-value placeholders to achieve a complete output of long-term power prediction results at once, avoiding the error accumulation problem of traditional stepwise prediction). Subsequently, the reconstructed Informer model is driven by the training set for iterative learning. The model parameters are continuously optimized through backpropagation to obtain a preliminarily optimized prediction model. By using a validation set (which shares the same data distribution as the training set but is independent) to evaluate the prediction accuracy (such as MAE and RMSE) and stability of the optimized model, only models that meet the preset accuracy threshold (such as validation set MAE ≤ preset value) are retained as the target prediction models. This achieves the technical effect of "balancing model training sufficiency and generalization ability". It ensures that the model fully learns the correlation between wind power and meteorological factors in complex scenarios through the training set, while effectively avoiding the risk of model overfitting through the validation set. This ensures that the original wind power prediction sequence output by the target prediction model has high initial accuracy, providing reliable basic data support for subsequent ramp detection and extreme value correction. This overcomes the limitations of traditional Informer models in long-term non-stationary power prediction, such as low computational efficiency and error accumulation.

[0086] Figure 3 This is a comparison chart of the predicted and actual wind power output over the three days following the cold wave. Figures 4-6Figures of predicted power curves and actual power curves under three types of extreme weather, typhoon, sandstorm and thunderstorm, respectively, Table 1 is the calculation results of mean absolute error and root mean square error of power prediction results of the reconstructed Informer model under four kinds of weather, Figure 7 The bar chart for evaluation and analysis of the reconstructed Informer model. The average absolute error of this model is within 10%, and the root mean square error is within 15% for the following results.

[0087] Table 1. Evaluation results of power prediction

[0088]

[0089] S3, the initial wind power prediction sequence is modified once to obtain a secondary wind power prediction sequence by combining the bump event merging strategy with the original wind power prediction sequence.

[0090] As an optional embodiment, the initial wind power prediction sequence is modified once to obtain a secondary wind power prediction sequence by combining the bump event merging strategy with the original wind power prediction sequence; comprising the following steps:

[0091] Based on the improved rotating door algorithm, the original wind power prediction sequence is preliminarily detected to generate a plurality of power trend segments; according to the power change threshold δ, the plurality of power trend segments are classified to generate a sequence of original climbing trend segments;

[0092] Iterate through adjacent climbing trend segments in the original climbing trend segment sequence and generate a secondary climbing trend segment sequence according to the bump event merging strategy;

[0093] Extract the starting point and the ending point in the secondary climbing trend segment sequence to construct an initial start-end point set; find the extreme value point closest to the starting point and the extreme value point closest to the ending point corresponding to each climbing segment in the initial start-end point set;

[0094] According to the position relationship of the extreme value point closest to the starting point, the extreme value point closest to the ending point and the original start-end point, the scene is corrected and the time sequence is integrated to obtain a secondary wind power prediction sequence.

[0095] It is important to note that the Spinning Door Transformation (SDT) algorithm is a highly efficient linear fitting algorithm that is fast and simple to execute. Essentially, it is a linear trend compression algorithm commonly used in the field of data compression. It achieves data compression by filtering and discarding some data points while retaining useful ones. The compression process must meet two principles: 1) Within the same compression segment, once the door is opened, it cannot be closed in the opposite direction; 2) During the door rotation, the maximum opening degree of the two doors is when the sum of the interior angles formed by the upper and lower doors is 180°. If the change angle is greater than 180°, the compression segment ends, and the previous data point is stored and retained.

[0096] Understandably, this embodiment first performs segmented compression and trend determination on the original wind power prediction sequence based on the improved rotating door algorithm to generate several power trend segments. Then, these power trend segments are classified according to the power change threshold δ (e.g., a normalized value of 0.02, corresponding to 2% of the wind farm's rated installed capacity), and the original climbing trend segment sequence that meets the climbing characteristics is selected. Subsequently, adjacent climbing trend segments in the original climbing trend segment sequence are traversed, and secondary climbing trend segment sequences are generated according to the bump event merging strategy (i.e., judging whether the power change amplitude of adjacent segments is less than a set threshold and whether the trend direction is consistent; if so, it is judged as a bump event and merged). Then, the starting point and ending point of the secondary climbing trend segment sequence are extracted to construct an initial set of starting and ending points. The algorithm identifies the nearest extreme point before the start point and the nearest extreme point after the end point of each climbing segment (the extreme point must be a local maximum or minimum value among the data points before and after it). Finally, based on the positional relationship between the nearest extreme point before the start point, the nearest extreme point after the end point, and the original start and end points, the start and end points of the climbing segments are corrected according to different scenarios. The corrected climbing segments are then integrated with the original stable segments according to timestamps. This achieves accurate detection and initial correction of climbing events in the original wind power prediction sequence, effectively eliminating the false climbing segmentation and start / end point positioning deviation caused by bump events. This enables the generated secondary wind power prediction sequence to more accurately represent the climbing characteristics of wind power in complex scenarios, providing high-quality basic sequence support for subsequent secondary correction of extreme points based on the xLSTM model.

[0097] As an optional embodiment, traversing adjacent climbing trend segments in the original climbing trend segment sequence and generating a secondary climbing trend segment sequence according to the bump event merging strategy includes the following steps:

[0098] Traverse adjacent climbing trend segments in the original climbing trend segment sequence and calculate the power change amplitude between adjacent segments. If the power change amplitude is less than the set threshold and the trend direction of adjacent segments is consistent, it is determined to be a bump event; otherwise, continue to traverse subsequent adjacent segments starting from the end point of the current trend segment.

[0099] After the original climbing trend segment sequence is traversed, adjacent climbing trend segments determined as bump events are merged to generate a secondary climbing trend segment sequence.

[0100] It can be understood that, in this embodiment, the power change amplitude between adjacent two segments is calculated synchronously by traversing the adjacent two segments in the original climbing trend segment sequence one by one, and the power change amplitude is compared with the preset threshold value, and it is checked whether the trend directions of the adjacent two segments are consistent (both are rising or both are falling). If both conditions are met (the power change amplitude is less than the set threshold value and the trend directions are consistent), the adjacent segment combination is determined as a bump event. If not, the end point of the current trend segment is taken as a new starting point, and the subsequent adjacent climbing trend segment is continuously traversed. After the complete round of traversal of the original climbing trend segment sequence is completed, all adjacent climbing trend segments determined as bump events are integrated and merged, and finally a secondary climbing trend segment sequence is generated, realizing effective filtering and merging of false segmentation caused by short-time small-amplitude fluctuations (bump events) in the original climbing trend segment sequence, eliminating the interference of redundant segmentation on the overall trend representation of the climbing event, and significantly improving the trend continuity and accuracy of the secondary climbing trend segment sequence, providing a more accurate trend segment basis for subsequent initial start and end point set construction and climbing segment boundary correction.

[0101] As an optional embodiment, the secondary wind power prediction sequence obtained by performing scene correction and time sequence integration according to the positional relationship between the extreme value point before the starting point, the extreme value point after the ending point and the original start and end points comprises the following steps:

[0102] If only the extreme value point before the starting point and the ending point satisfy the climbing definition, the starting point is corrected to the extreme value point before the current starting point.

[0103] If only the extreme value points before and after the starting point and the ending point satisfy the climbing definition, the ending point is corrected to the extreme value point after the current ending point.

[0104] If the extreme value points before and after the starting point and the ending point satisfy the climbing definition, the starting point and the ending point are both corrected to the extreme value points before and after the starting point and the ending point.

[0105] The climbing definition refers to an interval composed of continuous trend segments in the original wind power prediction sequence, the absolute value of the power change of the starting point and the ending point of the interval exceeds the set power change threshold δ, and after the bump event merging and start and end point optimization correction, the starting point and the ending point are located at the local minimum value and the local maximum value of the power fluctuation period.

[0106] It can be understood that the embodiment first defines the climbing as "an interval composed of continuous same-direction trend segments identified in the original wind power prediction sequence, the absolute value of the power change of the starting point and the ending point of the interval exceeds the set power change threshold δ, and after the bump event is combined and the starting and ending points are optimized and corrected, the starting point and the ending point are respectively the local minimum value and the local maximum value of the power fluctuation period", and then for each climbing segment, it is checked whether the three intervals of "the nearest extreme point before the starting point and the original ending point", "the original starting point and the nearest extreme point after the ending point", and "the nearest extreme point before the starting point and the nearest extreme point after the ending point" meet the above climbing definition: if only the first interval meets, the starting point of the climbing segment is corrected to the nearest extreme point before the starting point; if only the second interval meets, the ending point of the climbing segment is corrected to the nearest extreme point after the ending point; if the third interval meets, the starting and ending points of the climbing segment are respectively corrected to the nearest extreme point before the starting point and the nearest extreme point after the ending point, and finally all the corrected climbing segments and the original flat segments are integrated in time sequence according to the time stamp order, realizing the accurate calibration of the starting and ending points of the climbing segment, effectively eliminating the problem of inaccurate representation of the space-time range of the climbing event caused by the deviation of the starting and ending points in the traditional climbing detection, and making the generated secondary wind power prediction sequence more accurately reflect the actual climbing characteristics of the wind power under complex scenarios, providing clear boundary and reliable feature support for the subsequent secondary correction of extreme points based on the xLSTM model.

[0107] It should be noted that the calculation principle of the improved rotating door algorithm is shown in formulas (2), (3) and (4):

[0108] (2);

[0109] wherein k1 and k2 are the upper and lower door slopes between two points; P0 is the starting point power value in the current compressed segmented data sequence; P1 is the power value of the next data point; t0 is the starting point time of the current compressed segmented data sequence; t1 is the time of the next data point; and ε is a gating parameter.

[0110] (3);

[0111] wherein k 1n and k 2n are the upper and lower door slopes between point n and the last stored point; t n is the time corresponding to the new data point, and P n is the power value of the new data point.

[0112] (4);

[0113] Formula (4) compares and fixes the appropriate slope to meet the compression principle 1). If k If so, store the previous data point and use it as the new starting point; If so, continue compressing this data segment.

[0114] Secondly, the initial slope detection results are processed by merging bump events to correct for the bump events. For example: given the original slope trend segment sequence fitted by the rotating door algorithm: Y={(T1,P1),(T2,P2),…(T… n ,P n )}, where n is the number of sequence data points, T is the time index, and P is the wind power value; y={(t1,p1),(t2,p2),…(t m ,p m )} represents the merged secondary climbing trend segment sequence; m is the number of data points in the secondary climbing trend segment sequence; t is the time index, and p m Let m be the predicted wind power value corresponding to data point m in the sequence. The above two climbing trend segments satisfy the following: The conditions for merging adjacent climbing sections are:

[0115] (5);

[0116] (6);

[0117] Among them, P avt Let be the threshold value for the amplitude change of wind power, and 0 <i<n。

[0118] If the climbing period simultaneously satisfies formulas (5) and (6), then data point i is merged, i.e., t i =T i+1 p i =P j+1 After point i is merged, merging continues forward along the sequence. If formulas (5) and (6) do not hold simultaneously, it indicates that the direction of the climb is inconsistent or the climb definition cannot be met after merging. In this case, the merging of the current adjacent trend segments is terminated, and t j =T j+1 p j+1 =P i+1 Continue merging forward along the sequence.

[0119] From the perspective of wind power data trend changes, this embodiment classifies the trend segments generated by SDT detection:

[0120] (7);

[0121] Wherein, δ is the power change threshold, which is generally 2% of the rated installed capacity. After normalization, the value of δ can be set to 0.02, and the time interval between the two sampling points is 15 minutes.

[0122] In formula (7), when T=1, it is an upward trend; when T=0, it is a stable trend; and when T=-1, it is a downward trend. Continuous upward or downward trend segments will contain bump events.

[0123] Finally, the start and end points of the ramp section are corrected. Assume a set of ramp detection data for wind power output is Q={(s1,e1),(s2,e2),…(s…} k ,e k )}, where s represents the starting point of a certain uphill section, e is the ending point, and k is the total number of uphill events. In the uphill section, the extreme points must satisfy formula (8):

[0124] (8);

[0125] Based on the location information of the endpoints on both sides of each uphill segment, the nearest extreme point is found. Then, it is determined whether this extreme point meets the definition of an uphill slope, and whether adjustments should be made. Finally, the final value of the endpoint of this uphill segment is determined.

[0126] To determine whether corrections should be made at the endpoint of the uphill section, let p be the extreme point before the starting point of the target uphill section. k1 The extreme point after the termination point of the target uphill section is p. k2 Whether further revisions are needed depends on the specific circumstances described below:

[0127] 1) [p] k1 ,e k Within the segment, the wind power at both ends meets the definition of ramping, but in [p] k1 ,p k2 If the condition is not met, the start and end points of the climb are corrected to p. k1 e k ;

[0128] 2) [s] k ,p k2 Within the segment, the wind power at both ends meets the definition of ramping, but in [p] k1 ,p k2 If the condition is not met, the start and end points of the climb are corrected to s. k p k2 ;

[0129] 3) [p] k1 ,p k2 Within the segment, the wind power corresponding to both ends meets the definition of ramping, at which point the ramp start and end points are corrected to p. k1 p k2 .

[0130] The determination of the climb event prediction result is shown in Table 2. TP represents correct prediction of the occurrence of a climb event; TN represents correct prediction of the non-occurrence of a climb event; FP represents false prediction of the occurrence of a climb event; and FN represents false prediction of the non-occurrence of a climb event.

[0131] Table 2. Climb event prediction results

[0132]

[0133] Figure 8 For the climb detection result without considering the bump event, Figure 9 For the climb detection considering the bump event, Figure 10 For the contrast image in the case of not considering the bump event and considering the bump event, the evaluation index comparison before and after the improvement of the rotating door is shown in Table 3. It can be known from the chart that the evaluation results of the improved rotating door algorithm are all better than those of the unimproved rotating door algorithm.

[0134] Table 3. Climb event detection evaluation results

[0135]

[0136] Among them, the recall rate R: (9);

[0137] The precision rate P: (10);

[0138] The key success index C: (11);

[0139] The accuracy rate A CC : (12).

[0140] S4, based on the xLSTM model, the extreme value points in the secondary wind power prediction sequence are twice corrected to obtain a target wind power prediction sequence.

[0141] As an optional embodiment, the xLSTM model is used to twice correct the extreme value points in the secondary wind power prediction sequence to obtain a target wind power prediction sequence; comprising the following steps:

[0142] The starting extreme value point and the ending extreme value point of each climb segment determined after the first correction are extracted from the secondary wind power prediction sequence to construct a to-be-corrected extreme value point set;

[0143] An input feature vector corresponding to each extreme value point is constructed as an input sample set of the xLSTM model, and the input feature vector includes: a predicted power value at the current extreme value point, real-time meteorological data corresponding to the current extreme value point, and meteorological data of each of three sampling points before and after the current extreme value point.

[0144] The extreme value correction result is output by the xLSTM model, and a corresponding extreme point power value in the secondary wind power prediction sequence is replaced or adjusted based on the extreme value correction result to obtain a target wind power prediction sequence.

[0145] The correction result includes a correction amount of a target extreme point power value or a corrected power value.

[0146] It can be understood that, in the embodiment, first, the determined starting extreme point and ending extreme point of each climbing section are screened from the once corrected secondary wind power prediction sequence to construct a to-be-corrected extreme point set, so as to ensure that the correction object focuses on the extreme points that are critical to the determination of the climbing event; then, an exclusive input feature vector is constructed for each extreme point in the set, the vector not only includes the predicted power value at the current extreme point to reflect the power characteristics of the extreme point, but also incorporates real-time meteorological data (such as wind speed, wind direction, temperature, etc.) corresponding to the current extreme point, and meteorological data of each of the three sampling points before and after the current extreme point to capture the correlation rule between the extreme point and the surrounding time-series meteorological factors, so as to form an input sample set that comprehensively represents the characteristics of the extreme point; then, the sample set is input into the xLSTM model, and the model is used to output the correction result (which can be a correction amount of the extreme point power value or directly output the corrected power value) corresponding to each extreme point by using the deep modeling capability of the model on the time-series correlation characteristics; finally, the power value of the corresponding extreme point in the secondary wind power prediction sequence is replaced or adjusted according to the correction result to generate a target wind power prediction sequence, so as to realize accurate correction of the predicted deviation of the extreme point in the secondary sequence and effectively improve the representation accuracy of the target wind power prediction sequence for the peak and valley of the wind power in a complex scenario, thereby providing reliable data support for power grid dispatching, regulation and control based on accurate power prediction and ensuring operation safety.

[0147] It should be noted that the xLSTM is based on the LSTM model, which is expanded into multiple LSTM variants, including sLSTM and mLSTM variants; as Figure 11 As shown in FIG. 1, the framework diagram of the xLSTM model, from left to right in the figure are the original LSTM memory cell, the new memory cell, the xLSTM module, and the xLSTM architecture. Among them, starting from the original LSTM, by introducing new technical means such as exponential gating, matrix memory, and parallel training, two enhanced memory cells, sLSTM and mLSTM, are derived; then these memory cells are encapsulated into a residual block to form an xLSTM module; finally, the overall architecture of the xLSTM is obtained by stacking the fast detection, which finally solves the limitations of the traditional LSTM model in long sequence, parallel training, and high-dimensional memory representation.

[0148] Table 4 shows a comparison of wind power prediction errors before and after correction. Comparing the prediction data before and after extreme value correction, it is found that the data error is significantly reduced after extreme value correction, and the predicted value at the extreme value after correction is closer to the actual detected value.

[0149] Table 4. Comparison of Wind Power Prediction Performance Before and After Correction

[0150]

[0151] Figure 12 Figure 5 shows the slope detection results before and after the mechanism correction, and Table 5 shows the slope event prediction and evaluation table before and after the correction. After extreme value correction, all indicators of slope detection are significantly better than before the correction.

[0152] Table 5. Evaluation Table of Climbing Event Prediction Before and After Extreme Value Correction

[0153]

[0154] Comparative analysis was conducted under various extreme weather conditions, and the final wind power prediction results are as follows: Figures 13-15 As shown, the Informer-SDT-xLSTM prediction algorithm and other ramp prediction algorithms are analyzed and evaluated in terms of power prediction performance. Two ramp prediction models, TCN-Informer and Transformer-LSTM, are selected to compare the prediction performance with the model in this application. Ramp predictions are performed on the data for the next ten days when a cold wave occurs in a certain area. Table 6 shows the evaluation of the prediction effects of each model.

[0155] Table 6. Evaluation of 10-day power predictions for each model

[0156]

[0157] like Figure 16 Table 7 shows a comparison of the ramp prediction results of the three ramp prediction models, along with a comparison of the Pionjia indexes for ramp prediction. As can be seen from the figures, all evaluation indicators of this invention are higher than those of the proposed ramp prediction algorithm. The Informer-SDT-xLSTM wind power ramp prediction method has better ramp prediction performance, can adapt to complex weather conditions, and can predict more accurate results.

[0158] Table 7. Comparison of Slope Prediction and Evaluation

[0159]

[0160] Example 2: This embodiment of the invention also provides a wind power ramping prediction system, applicable to the wind power ramping prediction method in a complex scenario as described above, such as... Figure 17 As shown, it includes:

[0161] The data acquisition module 101: the historical meteorological data and the historical power data after the aggregation processing are subjected to time sequence consistency fusion processing to generate a test data set;

[0162] The model construction module 102: the improved Informer model is trained by using the test data set to obtain a target prediction model, and the original wind power prediction sequence is obtained by taking the real-time data set as the input of the target prediction model.

[0163] The first correction module 103: the initial wind power prediction sequence is subjected to a first correction by using a bump event merging strategy to obtain a secondary wind power prediction sequence after the original wind power prediction sequence is subjected to a ramp detection.

[0164] The second correction module 104: the extreme value points in the secondary wind power prediction sequence are subjected to a second correction by using an xLSTM model to obtain a target wind power prediction sequence.

[0165] It can be understood that the wind power ramp prediction system is constructed by using the modular system architecture in the embodiment, the time sequence consistency fusion processing of the multi-source heterogeneous data is realized by using the data acquisition module, and a high-quality input base is provided for the prediction model; the ProbSparse self-attention mechanism of the improved Informer model and the generative decoder are integrated by using the model construction module, and the efficient and accurate prediction of the long sequence power data is realized; the ramp detection and trend optimization of the initial power sequence are completed by using the improved rotating door algorithm and the bump event merging strategy in combination with the first correction module; finally, the extreme value points identified are subjected to a special correction by using the xLSTM model in the second correction module, and the whole-process automatic processing from the data preprocessing, power prediction, ramp detection to extreme value optimization is realized. The technical scheme realizes the coordinated operation of the four functional modules, and a complete "data fusion-prediction-detection-correction" technical closed loop is constructed, the multiple technical bottlenecks such as data inconsistency, low prediction accuracy, poor detection reliability and large extreme value deviation of the traditional method in the complex scene are effectively overcome, and the overall accuracy and system robustness of the wind power ramp prediction are significantly improved.

[0166] In the embodiment of the present application, a technical scheme is provided: an electronic device, comprising a memory and a processor, the memory stores a computer program, and the processor calls the computer program in the memory to realize the steps of the wind power ramp prediction method in the complex scene as described above.

[0167] In the embodiment of the present application, a technical scheme is provided: a storage medium, the storage medium stores computer executable instructions, and the computer executable instructions are loaded and executed by a processor to realize the steps of the wind power ramp prediction method in the complex scene as described above.

[0168] Through the description of the above embodiments, those skilled in the art can understand that, for the convenience and brevity of description, only the division of the above functional modules is exemplified, and in actual application, the above functions can be completed by different functional modules according to needs, that is, the internal structure of the specific device is divided into different functional modules to complete all or part of the functions described above.

[0169] In the embodiments provided in the present application, it should be understood that the disclosed structures and methods can be implemented in other manners. For example, the above-described embodiments of the structure are merely illustrative. For example, the division of the modules or units is only a logical function division, and there can be another division manner in actual implementation. For example, a plurality of units or components can be combined or integrated into another structure, or some features can be ignored or not executed. In addition, the displayed or discussed coupling or direct coupling or communication connection between the units can be indirect coupling or communication connection through some interfaces, structures or units, and can be electrical, mechanical or in other forms.

[0170] The units described as separated components can or can not be physically separated, and the components displayed as units can be one physical unit or a plurality of physical units, that is, can be located in one place, or also can be distributed to a plurality of different places. Some or all of the units can be selected according to actual needs to achieve the purposes of the embodiments.

[0171] In addition, each functional unit in the embodiments of the present application can be integrated in one processing unit, or each unit can exist physically, or two or more units can be integrated in one unit. The integrated unit can be implemented in the form of hardware or in the form of software function unit.

[0172] If the integrated unit is implemented in the form of software function unit and sold or used as an independent product, it can be stored in a readable storage medium. Based on such understanding, the technical solutions of the embodiments of the present application essentially or said part making contributions to the prior art or all or part of the technical solutions can be embodied in the form of a software product. The software product is stored in a storage medium, including a plurality of instructions to make a device (which can be a single-chip microcomputer, a chip, etc.) or a processor execute all or part of the steps of the embodiments of the methods of the present application. The foregoing storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various program codes that can store the medium.

[0173] The above specific embodiments are the preferred embodiments of the wind power climbing prediction method and system in a complex scene of the present application, and do not limit the specific implementation range of the present application. The scope of the present application includes but is not limited to the specific embodiments. Any equivalent changes made according to the shape and structure of the present application are within the protection scope of the present application.

Claims

1. A wind power ramping prediction method under complex scenarios, characterized in that: The method comprises the following steps: The historical meteorological data and the historical power data after the aggregation processing are subjected to time sequence consistency fusion processing to generate a test data set; The improved Informer model is trained through the test data set to obtain a target prediction model, and the original wind power prediction sequence is obtained by taking the real-time data set as the input of the target prediction model; The original wind power prediction sequence is subjected to a ramp detection and combined with a bump event merging strategy to modify the initial wind power prediction sequence to obtain a secondary wind power prediction sequence; The extreme points in the secondary wind power prediction sequence are subjected to a secondary modification based on an xLSTM model to obtain a target wind power prediction sequence. The historical meteorological data and the historical power data after the aggregation processing are subjected to time sequence consistency fusion processing to generate a test data set, which comprises the following steps: The historical meteorological data and the historical power data on the time sequence are obtained through the sampling frequency consistency; The historical power data on the same time sequence is subjected to normalization processing through a Min-Max normalization algorithm to obtain historical time sequence power data; The historical meteorological data on the same time sequence is subjected to intensive representation through meteorological factor consistency to obtain historical time sequence meteorological data; The historical time sequence meteorological data and the historical time sequence power data are subjected to time sequence consistency alignment through a time sequence alignment algorithm to obtain a test data set; The original wind power prediction sequence is subjected to a ramp detection and combined with a bump event merging strategy to modify the initial wind power prediction sequence to obtain a secondary wind power prediction sequence, which comprises the following steps: Based on the improved rotating door algorithm, a preliminary ramp detection is performed on the original wind power prediction sequence to generate a plurality of power trend segments; according to a power change threshold Delta A trend classification is performed on the plurality of power trend segments to generate an original ramp trend segment sequence; Adjacent ramp trend segments in the original ramp trend segment sequence are traversed, and the power change amplitudes of the adjacent two segments are calculated, and if the power change amplitude is less than a set threshold and the trend directions of the adjacent two segments are consistent, the bump event is determined; otherwise, the subsequent adjacent segment is continuously traversed from the end point of the current trend segment as the starting point; After the round of traversal of the original ramp trend segment sequence, the adjacent ramp trend segments determined as the bump event are merged to generate a secondary ramp trend segment sequence; The starting point and the end point in the secondary ramp trend segment sequence are extracted to construct an initial start-stop point set; The nearest extreme point before the starting point and the nearest extreme point after the end point corresponding to each ramp segment in the initial start-stop point set are found; According to the position relationship among the nearest extreme point before the starting point, the nearest extreme point after the end point and the original start-stop point, a scene correction and time sequence integration are performed to obtain a secondary wind power prediction sequence.

2. The wind power ramp prediction method under a complex scene according to claim 1, characterized in that: The improved Informer model is trained through the test data set to obtain a target prediction model, which comprises the following steps: The test data set is divided into a training set and a validation set according to a 1:G ratio; The encoder and the decoder of the Informer model are reconstructed, the encoder is integrated with a ProbSparse self-attention mechanism and a self-attention distillation mechanism, and the decoder is a generative decoder; The reconstructed Informer model is trained by the training set to obtain an optimized prediction model; the reliability of the optimized prediction model is evaluated by the validation set, and the optimized model after the reliability evaluation is taken as the target prediction model.

3. The wind power ramping prediction method under a complex scenario according to claim 1, characterized in that: The secondary wind power prediction sequence is obtained by modifying the starting point and the ending point according to the position relationship between the extreme point before the starting point, the extreme point after the ending point and the original starting and ending points, and integrating the time sequence. If only the extreme point before the starting point and the extreme point between the starting point and the ending point satisfy the ramping definition, the starting point is modified to the extreme point before the current starting point. If only the extreme point after the starting point and the extreme point after the ending point satisfy the ramping definition, the ending point is modified to the extreme point after the current ending point. If the extreme point before the starting point and the extreme point after the ending point satisfy the ramping definition as a whole, the starting point and the ending point are both modified to the extreme point before the starting point and the extreme point after the ending point. The ramping definition refers to an interval composed of continuous trend segments in the original wind power prediction sequence, and the absolute value of the power change of the starting point and the ending point of the interval exceeds the set power change threshold δ.

4. The wind power ramping prediction method under a complex scenario according to claim 1, characterized in that: The target wind power prediction sequence is obtained by modifying the extreme points in the secondary wind power prediction sequence based on the xLSTM model; including the following steps: The starting extreme point and the ending extreme point of each ramping segment are extracted from the secondary wind power prediction sequence after the first modification to construct a set of extreme points to be modified; An input feature vector corresponding to each extreme point is constructed as an input sample set of the xLSTM model, and the input feature vector includes the predicted power value at the current extreme point, the real-time meteorological data corresponding to the current extreme point and the meteorological data of each of the three sampling points before and after the current extreme point; The extreme modification result is output by the xLSTM model, and the corresponding extreme point power value in the secondary wind power prediction sequence is replaced or adjusted based on the extreme modification result to obtain the target wind power prediction sequence; The modification result includes the modification amount or the modified power value of the target extreme point.

5. A wind power ramping prediction system suitable for a wind power ramping prediction method in a complex scenario according to any one of claims 1-4, characterized in that: It includes: A data acquisition module: the historical meteorological data and the historical power data after the aggregation processing are fused in time sequence consistency to generate a test data set; A model construction module: the improved Informer model is trained by the test data set to obtain a target prediction model, and the original wind power prediction sequence is obtained by taking the real-time data set as the input of the target prediction model; A first modification module: the initial wind power prediction sequence is modified once by combining the ramping detection and the bump event merging strategy to obtain the secondary wind power prediction sequence; The secondary correction module corrects extreme points in the secondary wind power prediction sequence based on an xLSTM model to obtain a target wind power prediction sequence.

6. An electronic device, comprising: The memory stores a computer program, and the processor invokes the computer program in the memory to implement the steps of the wind power ramping prediction method under a complex scenario according to any one of claims 1 to 4.

7. A storage medium characterized by: The storage medium stores computer executable instructions, and the computer executable instructions are loaded and executed by the processor to implement the steps of the wind power ramping prediction method under a complex scenario according to any one of claims 1 to 4.

Images