Wind power climbing event prediction method considering extreme weather and time-space correlation information

Abstract

The invention belongs to the technical field of wind power climbing event prediction, and particularly relates to a wind power climbing event prediction method considering extreme weather and time-space correlation information. The method comprises the following steps: acquiring historical actually measured meteorological data of each wind power station of a cluster; carrying out cold-wave weather event identification on historical actually measured meteorological data, generating an antagonistic network based on a time sequence, and carrying out cold-wave event sample expansion; an extreme learning machine is constructed, and cold-wave weather prediction is carried out; performing historical climbing event detection on historical power output results of each station of the cluster; dividing the climbing events into various climbing conditions with different severity degrees by using a K-Means clustering algorithm; and carrying out climbing event prediction. According to the method, sample support is provided for training of the climbing prediction model, the climbing events are clustered and divided by fusing the fan operation state and the climbing characteristics, and the harm degrees of different climbing events, especially the climbing events in extreme weather, are finely measured.

Description

Technical Field

[0001] This invention belongs to the field of wind power ramping event prediction technology, and particularly relates to a wind power ramping event prediction method that takes into account extreme weather and spatiotemporal correlation information. Background Technology

[0002] In recent years, with the rapid development of the wind power industry, the proportion of my country's cumulative installed wind power capacity in the total cumulative installed power generation capacity has been continuously increasing. However, the volatility and randomness of wind power output pose challenges to the safe and stable operation of the power grid; especially when wind speed fluctuates rapidly, wind power output will exhibit an upward or downward trend, requiring significant adjustments to the power grid dispatching strategy; accurate prediction of wind power ramp-up events can provide important information support for power grid dispatching.

[0003] Existing methods for predicting wind power ramping events typically rely on analyzing measured and predicted meteorological data from individual wind farms, as well as the correlation between wind power output and the timing of ramping events. However, current methods do not adequately consider factors such as the low probability of extreme weather events, the diverse variations in wind power output, and the rapid and hazardous nature of wind power ramping. This results in an inability to provide accurate and reliable ramping event predictions under extreme weather conditions. Furthermore, current methods do not fully utilize the spatiotemporal correlations between wind farms within a wind farm cluster, further limiting the accuracy of ramping event predictions.

[0004] Therefore, given the frequent occurrence of extreme weather events, there is an urgent need to develop a prediction method that can provide accurate and detailed predictions of climbing events, thus providing effective support for the operation and scheduling of the power grid. Summary of the Invention

[0005] To address the shortcomings of the existing technologies, this invention provides a wind power ramping event prediction method that considers extreme weather and spatiotemporal correlation information. The purpose is to accurately predict day-ahead wind power ramping events at various wind farms within a power grid cluster by meticulously considering the severity of different ramping events, especially those under extreme weather conditions, and the spatiotemporal correlations between different power farms, thereby providing effective data support for grid dispatching.

[0006] The technical solution adopted by the present invention to achieve the above objectives is as follows:

[0007] Methods for predicting wind power ramping events that consider extreme weather and spatiotemporal correlation information include:

[0008] Acquire historical measured meteorological data from each wind farm station in the cluster;

[0009] Cold wave weather events are identified based on historical meteorological data, and an adversarial network is generated based on time series data to expand the cold wave event sample.

[0010] Based on the historical measured meteorological data of each wind farm after sample expansion, an extreme learning machine is constructed to predict cold wave weather.

[0011] Historical climbing events are detected based on historical measured meteorological data from each wind farm in the cluster.

[0012] Based on the wind turbine operation characteristics during historical ramp events, the K-Means clustering algorithm is used to classify ramp events into different types of ramp situations with varying degrees of severity.

[0013] Based on measured meteorological data, power data, cold wave weather forecast data, and slope climbing conditions from various stations, a graph convolutional neural network is constructed to predict slope climbing events.

[0014] Furthermore, the acquisition of historical measured meteorological data from each wind farm in the cluster includes acquiring historical measured meteorological data, predicted meteorological data, and power data from each wind farm over the past year, including:

[0015] Historical output of n stations:

[0016] Historical measured wind speeds at n stations:

[0017] Historical predicted wind speeds for n stations:

[0018] Historical measured temperatures at n stations:

[0019] Historical predicted temperatures for n stations:

[0020] Historical measured humidity at n stations:

[0021] Historical predicted humidity for n stations:

[0022] Where: P i p represents the historical power output of station i from time t0 to time t0+r. i,τ V represents the output of station i at time τ; i V represents the historical measured wind speed at station i from time t0 to time t0+r. i,τ V represents the historical measured wind speed at station i at time τ; i nwp The historical predicted wind speed at station i from time t0 to time t0+r. T represents the historical predicted wind speed of station i at time τ; i The historical measured temperature of station i from time t0 to time t0+r, t i,τT represents the historical measured temperature of station i at time τ; i nwp The historical predicted temperature of station i from time t0 to time t0+r. H represents the historical predicted temperature of station i at time τ. i h represents the historical measured humidity of station i from time t0 to time t0+r. i,τ The historical measured humidity of station i at time τ; The historical predicted humidity of station i from time t0 to time t0+r. The historical predicted humidity of station i at time τ represents the humidity of the station i at time τ.

[0023] Furthermore, the process of identifying cold wave weather events based on historical meteorological data and generating an adversarial network based on time series data to expand the cold wave event sample includes:

[0024] Based on the definition of cold wave events, the historical cold wave occurrence periods of each station were selected according to the historical measured meteorological data of each station.

[0025] Based on the TimeGAN neural network model, the characteristics of the cold wave occurrence period of each station were mined, and samples were generated for cold wave events;

[0026] Through adversarial training of the generator and discriminator, a large amount of meteorological and power output simulation data of cold wave weather similar to the power output characteristics of extreme weather are generated, thereby expanding the cold wave sample of station i.

[0027] Furthermore, the method of constructing an extreme learning machine based on historical measured meteorological data from each wind farm after sample expansion to predict cold wave weather includes:

[0028] The cold wave weather prediction model uses the wind power output of each power station in the cluster for the three days prior to the prediction time. Measured meteorological data ( ), and the forecast weather data for the next day ( As input to the model, the weather type at each moment of a future day for a certain station in the cluster is used as the output, where, For station exist The measured wind speed during this period For station exist The measured temperature during this period For station exist The measured humidity during this period For station exist Forecast wind speeds for this period of time For station exist Predicted temperatures for this period of time For station exist Predicted humidity for this period;

[0029]

[0030] X = {X i |i=1,2,...,n}

[0031] Category i =X⊙Θ i,c

[0032] The weather type for station i in the next day is:

[0033] Category i =[category] i,1 ,category i,2 ,...,category i,96 ];

[0034] Where: category i,N ∈{cold wave weather, non-cold wave weather} represents the situation of the site at time N in the future; where Θ i,c For the corresponding parameters to be trained, ⊙ represents element-wise multiplication, X i The feature set of station i regarding cold wave weather forecast, and the feature set of each station X regarding cold wave weather forecast;

[0035] The model was trained using 80% of the cold wave weather dataset as the training set, and the model performance was tested using 20% ​​of the cold wave weather dataset as the test set.

[0036] Furthermore, the historical measured meteorological data results of each wind farm station in the cluster are used to detect historical ramp events;

[0037] This includes: constructing a ramp event detection framework, and using historical power data from each power station to detect historical ramp events. The steps are as follows:

[0038] If the power sampling within the sliding window at the current time point τ satisfies the following formula, then a ramp-up event is considered to have occurred at wind farm i, and the start time of the ramp-up event is recorded as τ. i,rs ;

[0039]

[0040] Where, p i,τ and p i,τ+Δτ These represent the power output of station i at times τ and τ+Δτ, respectively, where Δτ is the time width of the sliding window, taken as 4 hours. The threshold for ramp detection is set at 30% of the rated capacity of the wind farm;

[0041] When a certain moment τ satisfies the formula for cold wave weather prediction, the time window is shifted by one moment each time until the power sampling within the sliding window at the current time point satisfies the following formula. At this point, the climbing event is considered to have ended; the end time of the climbing event is recorded as τ. i,re The duration of the uphill climb is from τ i,rs To τ i,re ;

[0042]

[0043] Where, p i,τ and p i,τ-Δτ These represent the output of station i at times τ and τ-Δτ, respectively, where Δτ is the time width of the sliding window. This is the threshold for hill climbing detection.

[0044] Furthermore, based on the wind turbine operating characteristics during historical ramp events, the K-Means clustering algorithm is used to classify ramp events into various ramp situations of different severity, including:

[0045] The operating status of the wind turbines at each site during each ramping event is determined based on the physical parameters of wind turbine operation, and the ramping amplitude, ramping direction, ramping duration and average power output during the ramping event period are used as clustering features.

[0046] Taking station i as an example, the duration of the climbing event is from τ i,rs To τ i,re :

[0047]

[0048] in, and These represent the power output of the station at the start and end of the climb, respectively. upper The wind speed threshold t represents the wind speed of the wind turbine due to large wind shear. lower The temperature threshold representing the temperature at which the fan disconnects from the grid due to low temperature. The magnitude of the climb, representing the climb event. Represents the direction of climbing. Represents the duration of the climb, Represents the average wind power output during the ramp-up event, state i,τ Represents the operating state of the wind turbine at time τ. This represents the operating status of the wind turbine during a ramp-up event; features are extracted from each ramp-up event to obtain a ramp-up sample.

[0049] The K-Means clustering algorithm was used to classify the station ramp events into different types of ramp events with varying degrees of severity; the number of clusters was set to 6, and 6 samples were randomly selected as the initial cluster centers μ. I Based on the K-Means clustering algorithm, the types of hill climbing are divided into 6 categories:

[0050]

[0051] Where I represents the climbing category number, and F represents the sample. The number belonging to the climbing category, μ F Indicates sample Belongs to the cluster center, C I This represents the cluster of the I-th type of hill climbing, where the sample belongs to C. I This means that the distance of the sample is C. I Cluster center μ I It is closer to other cluster centers and is classified into a cluster under the first type of climbing category.

[0052] Furthermore, the method of constructing a graph convolutional neural network based on measured meteorological data, power data, cold wave weather forecast data, and climbing conditions of each station to predict climbing events includes constructing a graph convolutional neural network based on measured meteorological data, power data, forecast meteorological data, and climbing conditions of each station to predict climbing events.

[0053] The slope prediction model takes the wind power output, measured meteorological data, slope conditions, and forecast meteorological data for the next day as inputs for each station in the cluster three days before the prediction time, and the slope type of a certain station in the cluster at each time of the next day as output.

[0054]

[0055] T = {T i |i=1,2,...,n};

[0056]

[0057] Where Y is the output of GCN, considered as the spatiotemporal correlation between the mined sites, and T is the feature set of n sites; T i The features considered as those of wind farm i include the measured meteorological data of wind power output and the slope conditions (L) three days prior to the predicted time M. i,M:M-288 The input to the GCN is X, which is a vector composed of the features of each node, and Θ represents the convolution kernel parameter matrix. Let W be the degree matrix of the nodes, W be the adjacency matrix representing the stations, and I be the identity matrix.

[0058] Type i=Y⊙Θ i,s ;

[0059] Among them, the ramp-up situation of station i in the next day is Type i =[type i,1 ,type i,2 ,...,type i,96 ], type i,N ∈{1,2,3,4,5,6} represents the climbing situation of the station at time N in the future day; where Θ i,s For the corresponding parameters,

[0060] Indicates element-wise multiplication;

[0061] Using 80% of the obtained dataset as the training set, models were trained separately for cold wave weather and non-cold wave weather. Using 20% ​​of the obtained dataset as the test set, cold wave weather prediction models were used to predict cold wave events. If the day to be predicted was predicted to be a cold wave, the cold wave weather climbing identification model was used for prediction. If the day to be predicted was predicted to be a non-cold wave, the non-cold wave weather climbing identification model was used for prediction.

[0062] A wind power ramping event prediction device that considers extreme weather and spatiotemporal correlation information includes:

[0063] The acquisition module is used to acquire historical measured meteorological data from each wind farm station in the cluster.

[0064] The sample expansion module is used to identify cold wave weather events based on historical measured meteorological data and generate an adversarial network based on time series to expand the cold wave event sample.

[0065] The cold wave weather prediction module is used to construct an extreme learning machine based on the historical measured meteorological data of each wind farm after sample expansion, and to predict cold wave weather.

[0066] The historical ramp event detection module is used to detect historical ramp events based on the historical measured meteorological data of each wind farm in the cluster.

[0067] The ramp event classification module is used to classify ramp events into different types of ramp situations with varying degrees of severity based on the wind turbine operating characteristics when historical ramp events occur, using the K-Means clustering algorithm.

[0068] The slope event prediction module is used to construct a graph convolutional neural network based on measured meteorological data, power data, cold wave weather prediction data, and slope conditions from various stations to predict slope events.

[0069] A computer device includes a storage medium, a processor, and a computer program stored on the storage medium and executable on the processor, wherein the processor executes the computer program to implement the steps of any of the wind power ramping event prediction methods that take into account extreme weather and spatiotemporal correlation information.

[0070] A computer storage medium storing a computer program, wherein when the computer program is executed by a processor, the steps of the wind power ramping event prediction method considering extreme weather and spatiotemporal correlation information as described above are implemented.

[0071] The present invention has the following beneficial effects and advantages:

[0072] This invention fully considers the low probability and high severity of uphill events under extreme weather conditions. It establishes a TimeGAN network to expand data from samples under extreme weather conditions, providing sample support for training the uphill prediction model. Simultaneously, it integrates wind turbine operating status and uphill characteristics to cluster uphill events, refining the assessment of the severity of different uphill events, especially those under extreme weather conditions.

[0073] This invention also fully considers the low probability and high severity of uphill events under extreme weather conditions. It establishes a TimeGAN network to expand data from samples under extreme weather conditions, providing sample support for training the uphill prediction model. Simultaneously, it integrates wind turbine operating status and uphill characteristics to cluster uphill events, refining the assessment of the severity of different uphill events, especially those under extreme weather conditions.

[0074] This invention effectively solves the problem of insufficient prediction accuracy caused by the scarcity of extreme weather samples and the homogeneous processing of climbing events in existing technologies by adding two key steps: "extreme weather climbing event sample enhancement based on TimeGAN" and "clustering and partitioning that integrates wind turbine operating status and climbing features". The former uses generative adversarial networks to expand the data of high-hazard but low-frequency extreme climbing events, thereby improving the model's generalization ability. The latter achieves fine-grained identification and hierarchical prediction of different climbing types through multi-dimensional feature clustering, especially high-risk events under extreme weather conditions. This significantly improves the accuracy of day-ahead climbing event prediction and scheduling support capabilities of wind power clusters. Attached Figure Description

[0075] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the description of the embodiments taken in conjunction with the following drawings, in which:

[0076] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0077] To better understand the above-mentioned objectives, features, and advantages of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that, unless otherwise specified, the embodiments of the present invention and the features thereof can be combined with each other.

[0078] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and therefore the scope of protection of the invention is not limited to the specific embodiments disclosed below.

[0079] The following reference Figure 1 The technical solutions of some embodiments of the present invention are described below.

[0080] Example 1

[0081] This invention provides an embodiment of a wind power ramping event prediction method that considers extreme weather and spatiotemporal correlation information. For example... Figure 1 As shown, Figure 1 This is a flowchart of the wind power ramping event prediction method of the present invention, which takes into account extreme weather and spatiotemporal correlation information.

[0082] The method of the present invention includes the following steps:

[0083] Step 1. Obtain historical one-year measured meteorological data, forecasted meteorological data, and power data for each wind farm in the cluster.

[0084] Step 2. Identify cold wave weather events based on historical measured meteorological data from each station, and expand the cold wave event samples based on Time-series Generative Adversarial Network (TimeGAN).

[0085] Step 3. Construct an Extreme Learning Machine (ELM) for cold wave weather prediction based on measured meteorological data, power data, and predicted meteorological data from each station. This step utilizes the data augmented in Step 2. Considering the limited data samples during cold waves, training the model based on a small sample size would result in poor predictive performance. Therefore, the data augmented in Step 2 provides more data support for training the cold wave weather prediction model in Step 3.

[0086] Step 4. Perform historical ramp-up event detection on the historical power output results of each station in the cluster from Step 1.

[0087] Step 5. Based on the characteristics of the wind turbine operating status, climbing amplitude, climbing direction, and climbing duration when the historical climbing events occurred at each site in Step 4, the climbing events are divided into different types of climbing events with different degrees of severity using the K-Means clustering algorithm.

[0088] Step 6. Based on the measured meteorological data, power data, and predicted meteorological data of each wind farm, as well as the climbing event information from Step 5, construct a Graph Convolutional Neural Network (GCN) to predict climbing events. In this embodiment, it is assumed that there are n wind farms in the preset regional wind power cluster, and the historical power output data of each farm over the past year is used as the basis for modeling and prediction.

[0089] Step 1, which involves obtaining historical one-year measured meteorological data, forecasted meteorological data, and power data for each wind farm in the cluster, specifically includes:

[0090] Historical output of n stations:

[0091] Historical measured wind speeds at n stations:

[0092] Historical predicted wind speeds for n stations:

[0093] Historical measured temperatures at n stations:

[0094] Historical predicted temperatures for n stations:

[0095] Historical measured humidity at n stations:

[0096] Historical predicted humidity for n stations:

[0097] Where: P i p represents the historical power output of station i from time t0 to time t0+r. i,τ V represents the output of station i at time τ; i V represents the historical measured wind speed at station i from time t0 to time t0+r. i,τ V represents the historical measured wind speed at station i at time τ; i nwp The historical predicted wind speed at station i from time t0 to time t0+r. T represents the historical predicted wind speed of station i at time τ; i The historical measured temperature of station i from time t0 to time t0+r, t i,τ T represents the historical measured temperature of station i at time τ;i nwp The historical predicted temperature of station i from time t0 to time t0+r. H represents the historical predicted temperature of station i at time τ. i h represents the historical measured humidity of station i from time t0 to time t0+r. i,τ The historical measured humidity of station i at time τ; The historical predicted humidity of station i from time t0 to time t0+r. This represents the historical predicted humidity of station i at time τ. In this embodiment, the historical output data is the data from the most recent year, divided into 15-minute intervals, r = 365 * 24 * 60 / 15 = 35040.

[0098] Step 2 involves identifying cold wave weather events based on historical meteorological data from various stations and expanding the cold wave event samples using a Time-series Generative Adversarial Network (TimeGAN). The specific steps are as follows:

[0099] Step 2-1. Based on the definition of cold wave events, the historical cold wave occurrence periods of each station are selected according to the historical measured meteorological data of each station.

[0100] Based on the definition of a cold wave: Determine the time period of a cold wave event.

[0101] CW i ={(s q ,s w +288)|(max(t i,τ )-min(t i,τ )>8,t i,τ <4,τ∈[s,s+96])

[0102] or(max(t i,τ )-min(t i,τ )>10,t i,τ <4,τ∈[s,s+192])

[0103] or(max(t i,τ )-min(t i,τ )>12,t i,τ <4,τ∈[s,s+288]),s=s q ,s q +1,...,s w ,s w -s q ≤288}

[0104] CWi s represents the set of time periods of cold wave events occurring at the i-th station. q Representing the start of the cold wave event, s w +288 represents the end time of the cold wave event, t i,τ This represents the measured temperature value at station i at time τ. The formula means that a cold wave event occurs if the air temperature drops by more than 8°C within 24 hours, and the minimum temperature drops below 4°C; or the air temperature drops by more than 10°C within 48 hours, and the minimum temperature drops below 4°C; or the air temperature drops by more than 12°C within 72 hours, and the minimum temperature drops below 4°C; and a cold wave event lasts at least 3 days.

[0105] Step 2-2. Based on the TimeGAN neural network model, mine the characteristics of the cold wave occurrence period for each station, and generate samples for cold wave events. First, define the reconstruction loss function as follows:

[0106]

[0107] Where, x i Let x be the real sample space when a cold wave event occurs at the i-th station. i,τ This indicates that the cold wave event occurred in τ. q :τ w The sample space for the +288 time period includes measured meteorological data. Forecast meteorological data Wind power output data gather, For station i in τ q :τ w The measured wind speed during the +288 period For station i in τ q :τ w The actual temperature during the +288 period. For station i in τ q :τ w The measured humidity during the +288 period For station i in τ q :τ w Forecast wind speed during the +288 period For station i in τ q :τ w The predicted temperature during the +288 period. For station i in τ q :τ w Predicted humidity for this period: +288. L i,R Let E[·] be the loss function of the TimeGAN model, and let E[·] be the Euclidean distance used to calculate the similarity between vectors. i,τGiven time-series data obtained through autoencoder mapping-inverse mapping, this formula is used for autoencoder parameter optimization.

[0108] The supervised loss function is defined as follows:

[0109]

[0110] Among them, h i,τ Let g be the hidden layer vector mapped onto the latent space; X For the recurrent network within the generator; z i,τ Let be a random vector. This formula is used to learn the temporal dependencies of the samples.

[0111] The unsupervised loss function is defined as follows:

[0112]

[0113] In the formula: y i,τ and y i,τ These represent the discriminator's classification of real samples and synthetic samples, respectively.

[0114] Steps 2-3. The parameter optimization process for the autoencoder is as follows:

[0115]

[0116] The parameter optimization process in adversarial generative networks:

[0117]

[0118] In the formula: θ e θ r θ g and θ d Let represent the parameters in the embedding function, recovery function, generator, and discriminator, respectively; λ and η are both greater than 0, used to balance the loss functions in the autoencoder and adversarial generative network, respectively. Adversarial learning makes the discriminator L... i,S and generator L i,U The goal is to achieve Nash equilibrium, meaning that the generator output is roughly the same as the original data distribution.

[0119] Steps 2-4. Through adversarial training of the generator and discriminator, a large amount of meteorological and power output simulation data of cold wave weather similar to the power output characteristics of extreme weather are generated, thereby expanding the cold wave sample of station i.

[0120] The expansion of cold wave samples at the wind farm effectively alleviates the problem of insufficient model training samples caused by the scarcity of cold wave events by generating cold wave-related wind power ramp data with real time-series characteristics. With the support of a large amount of data, the prediction accuracy of the cold wave weather prediction model in step 3 can be improved; at the same time, the recognition accuracy and prediction reliability of wind power ramp events under extreme cold wave weather in step 6 can be improved.

[0121] Step 3 involves constructing an Extreme Learning Machine (ELM) based on the measured meteorological data, power data, and predicted meteorological data from each station in Step 2 to predict cold wave weather. This provides more data support for training the cold wave weather prediction model in Step 3 and the hill climb identification model in Step 6. Specifically, this includes the following steps:

[0122] Step 3-1. The cold wave weather prediction model uses the wind power output p of each station in the cluster three days before the prediction time. i,M:M-288 Measured meteorological data (v i,M:M-288 ,t i,M:M-288 ,h i,M:M-288 ), and the forecast weather data for the next day. As input to the model, the weather type at each moment of a future day for a certain station in the cluster is used as the output. Where v i，M：M-288 For the measured wind speed at station i during the time interval M-288:M, t i，M：M-288 h represents the measured temperature of station i during the time interval M-288:M. i,M：M-288 The measured humidity at station i during the period from M-288:M is... The predicted wind speed for station i during the time interval M:M+96. The predicted temperature for station i during the time interval M:M+96. The predicted humidity for station i during the time interval M:M+96.

[0123]

[0124] X = {X i |i=1,2,...,n}

[0125] Category i =X⊙Θ i,c

[0126] The weather type for station i in the next day is:

[0127] Category i =[category] i,1 ,category i,2 ,...,category i,96 ];

[0128] Where: category i,N ∈{cold wave weather, non-cold wave weather} represents the situation of the site at time N in the future; where Θ i,c For the corresponding parameters to be trained, ⊙ represents element-wise multiplication, X i Station i represents the feature set of cold wave weather forecasts, and X represents the feature set of cold wave weather forecasts for each station.

[0129] Step 3-2. Train the model using 80% of the cold wave weather dataset as the training set. Test the model performance using 20% ​​of the cold wave weather dataset as the test set.

[0130] Step 4 involves detecting historical ramp-up events based on the historical power output results of each wind farm in the cluster obtained in Step 1. The results are used in Steps 5 and 6. Specifically, this includes constructing a ramp-up event detection framework and detecting historical ramp-up events using the historical power data of each farm based on this framework. The specific steps are as follows:

[0131] Step 4-1. If the power sampling within the sliding window at the current time point τ satisfies the following formula, then it is considered that a ramping event has occurred at wind farm i, and the start time of the ramping event is recorded as τ. i,rs .

[0132]

[0133] Where, p i,τ and p i,τ+Δτ These represent the power output of station i at times τ and τ+Δτ, respectively. Δτ is the time width of the sliding window, taken as 4 hours. The threshold for ramp detection is set to 30% of the wind farm's rated capacity.

[0134] Step 4-2. When a certain moment τ satisfies the formula defined in Step 3-1, the time window is shifted by one moment at a time until the power sampling within the sliding window at the current time point satisfies the following formula. Then, the climbing event is considered to have ended. The end time of the climbing event is recorded as τ. i,re The duration of the uphill climb is from τ i,rs To τ i,re .

[0135]

[0136] Where, p i,τ and p i,τ-Δτ These represent the power output of station i at times τ and τ-Δτ, respectively. Δτ is the time width of the sliding window. This is the threshold for hill climbing detection.

[0137] Step 5, based on the characteristics obtained in Step 4, such as the wind turbine operating status, ramp amplitude, ramp direction, and ramp duration at the time of the ramp event, uses the K-Means clustering algorithm to classify the ramp events into different types of ramp events with varying degrees of severity. The specific steps are as follows:

[0138] Step 5-1. Based on the physical parameters of wind turbine operation, determine the operating status of wind turbines in each climbing event at each site, and use the climbing amplitude, climbing direction, climbing duration and average power output during the climbing event period as clustering features.

[0139] Taking station i as an example, the duration of the climbing event is from τ i,rs To τ i,re :

[0140]

[0141] in, and These represent the power output of the station at the start and end of the climb, respectively. upper The wind speed threshold t represents the wind speed of the wind turbine due to large wind shear. lower This represents the temperature threshold at which the fan disconnects from the grid due to low temperature. The magnitude of the climb, representing the climb event. Represents the direction of climbing. Represents the duration of the climb, Represents the average wind power output during the ramp-up event, state i,τ Represents the operating state of the wind turbine at time τ. This represents the operating status of the wind turbine during a ramp-up event. Features are extracted from each ramp-up event to obtain a ramp-up sample.

[0142] Step 5-2. Use the K-Means clustering algorithm to classify the station ramp-up events into different types of ramp-up events with varying severity. Set the number of clusters to 6, and randomly select 6 samples as the initial cluster centers μ. I Based on the K-Means clustering algorithm, the types of hill climbing are divided into 6 categories:

[0143]

[0144] Where I represents the climbing category number, and F represents the sample. The possible climbing category number, μ F Indicates sample The possible cluster center, C I This represents the cluster of the I-th type of hill climbing, where the sample belongs to C. I This means that the distance of the sample is C. I Cluster center μI Because it is closer to other cluster centers, it is assigned to the cluster of the I-th climbing category. The algorithm iteratively adjusts the cluster centers to minimize the Euclidean distance E between all samples and their corresponding cluster centers, thereby achieving sample clustering and classifying all samples into 6 climbing categories.

[0145] Step 6 involves constructing a graph convolutional neural network (GCN) based on the measured meteorological data, power data, and predicted meteorological data from each station, as well as the climbing event information from Step 5, to predict climbing events. Specifically, this includes constructing a graph convolutional neural network (GCN) based on the measured meteorological data, power data, predicted meteorological data, and climbing event information from each station to predict climbing events.

[0146] Step 6-1. The slope prediction model takes the wind power output, measured meteorological data, slope conditions, and forecast meteorological data for the next day as inputs for each station in the cluster three days before the prediction time, and the slope type of a certain station in the cluster at each moment of the next day as output.

[0147]

[0148] T = {T i |i=1,2,...,n}

[0149]

[0150] Where Y is the output of GCN, considered as the spatiotemporal correlation between the mined sites, and T is the feature set of n sites. i The features considered as those of wind farm i include the measured meteorological data of wind power output and the slope conditions (L) three days prior to the predicted time M. i,M:M-288 The input to the GCN is X, which is a vector composed of the features of each node, and Θ represents the convolution kernel parameter matrix. Let W be the degree matrix of the nodes, W be the adjacency matrix representing the stations, and I be the identity matrix.

[0151] Type i =Y⊙Θ i,s

[0152] Among them, the ramp-up situation of station i in the next day is Type i =[type i,1 ,type i,2 ,...,type i,96 ], type i,N∈{1,2,3,4,5,6} represents the climbing situation of the station at time N in the future day; where Θ i,s For the corresponding parameters,

[0153] This represents element-wise multiplication.

[0154] Step 6-2 involves using 80% of the obtained dataset as the training set to train models separately for cold wave and non-cold wave weather. Since the hill-climbing prediction model constructed in Step 6 of this invention is a type-based prediction, it is divided into predictions for normal weather and predictions for cold wave weather. Therefore, it is first necessary to know whether the predicted day is a normal weather or cold wave weather.

[0155] Using 20% ​​of the obtained dataset as the test set, the cold wave weather prediction model constructed in step 3 is used to predict cold wave events. If the day to be predicted is predicted to be a cold wave, the cold wave weather climbing identification model is used for prediction. If the day to be predicted is predicted to be a non-cold wave, the non-cold wave weather climbing identification model is used for prediction.

[0156] Example 2

[0157] The present invention provides another embodiment of a wind power ramping event prediction device that considers extreme weather and spatiotemporal correlation information, used to implement the steps of the wind power ramping event prediction method considering extreme weather and spatiotemporal correlation information described in Embodiment 1, specifically including:

[0158] The acquisition module is used to acquire historical measured meteorological data from each wind farm station in the cluster.

[0159] The sample expansion module is used to identify cold wave weather events based on historical measured meteorological data and generate an adversarial network based on time series to expand the cold wave event sample.

[0160] The cold wave weather prediction module is used to construct an extreme learning machine based on the historical measured meteorological data of each wind farm after sample expansion, and to predict cold wave weather.

[0161] The historical ramp event detection module is used to detect historical ramp events based on historical measured meteorological data of each wind farm in the cluster, including power output results.

[0162] The ramp event classification module is used to classify ramp events into different types of ramp situations with varying degrees of severity based on the wind turbine operating characteristics when historical ramp events occur, using the K-Means clustering algorithm.

[0163] The slope event prediction module is used to construct a graph convolutional neural network based on measured meteorological data, power data, cold wave weather prediction data, and slope conditions from various stations to predict slope events.

[0164] Example 3

[0165] Based on the same inventive concept, embodiments of the present invention also provide a computer device, including a storage medium, a processor, and a computer program stored on the storage medium and executable on the processor. When the processor executes the computer program, it implements the steps of any of the wind power ramping event prediction methods considering extreme weather and spatiotemporal correlation information described in Embodiment 1 or 2.

[0166] Example 4

[0167] Based on the same inventive concept, this embodiment of the invention also provides a computer storage medium storing a computer program, which, when executed by a processor, implements the steps of any of the wind power ramping event prediction methods that consider extreme weather and spatiotemporal correlation information as described in Embodiment 1 or 2.

[0168] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0169] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0170] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0171] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0172] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A wind power ramping event prediction method that considers extreme weather and spatiotemporal correlation information, characterized by: include: Acquire historical measured meteorological data from each wind farm station in the cluster; Cold wave weather events are identified based on historical meteorological data, and an adversarial network is generated based on time series data to expand the cold wave event sample. Based on the historical measured meteorological data of each wind farm after sample expansion, an extreme learning machine is constructed to predict cold wave weather. Historical climbing events are detected based on historical measured meteorological data from each wind farm in the cluster. Based on the wind turbine operation characteristics during historical ramp events, the K-Means clustering algorithm is used to classify ramp events into different types of ramp situations with varying degrees of severity. Based on measured meteorological data, power data, cold wave weather forecast data, and slope climbing conditions from various stations, a graph convolutional neural network is constructed to predict slope climbing events.

2. The wind power ramping event prediction method considering extreme weather and spatiotemporal correlation information according to claim 1, characterized in that: The acquisition of historical measured meteorological data for each wind farm in the cluster includes acquiring historical measured meteorological data, predicted meteorological data, and power data for each wind farm over the past year, including: Historical output of n stations: Historical measured wind speeds at n stations: Historical predicted wind speeds for n stations: Historical measured temperatures at n stations: Historical predicted temperatures for n stations: Historical measured humidity at n stations: Historical predicted humidity for n stations: Where: P i p represents the historical power output of station i from time t0 to time t0+r. i,τ V represents the output of station i at time τ; i V represents the historical measured wind speed at station i from time t0 to time t0+r. i,τ V represents the historical measured wind speed at station i at time τ; i nwp The historical predicted wind speed at station i from time t0 to time t0+r. T represents the historical predicted wind speed of station i at time τ; i The historical measured temperature of station i from time t0 to time t0+r, t i,τ T represents the historical measured temperature of station i at time τ; i nwp The historical predicted temperature of station i from time t0 to time t0+r. H represents the historical predicted temperature of station i at time τ. i h represents the historical measured humidity of station i from time t0 to time t0+r. i,τ The historical measured humidity of station i at time τ; The historical predicted humidity of station i from time t0 to time t0+r. The historical predicted humidity of station i at time τ represents the humidity of the station i at time τ.

3. The wind power ramping event prediction method considering extreme weather and spatiotemporal correlation information according to claim 1, characterized in that: The process of identifying cold wave weather events based on historical meteorological data and generating an adversarial network based on time series data to expand the cold wave event sample includes: Based on the definition of cold wave events, the historical cold wave occurrence periods of each station were selected according to the historical measured meteorological data of each station. Based on the TimeGAN neural network model, the characteristics of the cold wave occurrence period of each station were mined, and samples were generated for cold wave events; Through adversarial training of the generator and discriminator, a large amount of meteorological and power output simulation data of cold wave weather with similar characteristics to extreme weather power output are generated, thereby expanding the cold wave sample of station i.

4. The wind power ramping event prediction method considering extreme weather and spatiotemporal correlation information according to claim 1, characterized in that: The method involves constructing an extreme learning machine based on historical measured meteorological data from various wind farms after sample expansion, to predict cold wave weather; including: The cold wave weather prediction model uses the wind power output p of each power station in the cluster three days before the prediction time. i,M:M-288 Measured meteorological data (v i,M:M-288 ,t i,M:M-288 ,h i,M:M-288 ), and the forecast weather data for the next day. As input to the model, the weather type at each moment of a future day for a certain station in the cluster is used as the output, where v i,M：M-288 For the measured wind speed at station i during the time interval M-288:M, t i,M：M-288 h represents the measured temperature of station i during the time interval M-288:M. i,M:M-288 The measured humidity at station i during the period from M-288:M is... The predicted wind speed for station i during the time interval M:M+96. The predicted temperature for station i during the time interval M:M+96. The predicted humidity for station i during the time interval M:M+96; X={X i |i=1,2,...,n} Category i =X⊙Θ i,c The weather type for the station i in the next day is: Category i =[category i,1 ,category i,2 ,...,category i,96 ]; Where: category i,N ∈{cold wave weather, non-cold wave weather} represents the situation of the station at time N in the future; where Θ i,c For the corresponding parameters to be trained, ⊙ represents element-wise multiplication, X i The feature set of station i regarding cold wave weather forecast, and the feature set of each station X regarding cold wave weather forecast; The model was trained using 80% of the cold wave weather dataset as the training set, and the model performance was tested using 20% ​​of the cold wave weather dataset as the test set.

5. The wind power ramping event prediction method considering extreme weather and spatiotemporal correlation information according to claim 1, characterized in that: The method involves detecting historical climbing events based on historical measured meteorological data from each wind farm station in the cluster. This includes: constructing a ramp event detection framework, and using historical power data from each power station to detect historical ramp events. The steps are as follows: If the power sampling within the sliding window at the current time point τ satisfies the following formula, then a ramp-up event is considered to have occurred at wind farm i, and the start time of the ramp-up event is recorded as τ. i,rs ; Where, p i,τ and p i,τ+Δτ These represent the power output of station i at times τ and τ+Δτ, respectively, where Δτ is the time width of the sliding window, taken as 4 hours. The threshold for ramp detection is set at 30% of the rated capacity of the wind farm; When a certain moment τ satisfies the formula for cold wave weather prediction, the time window is shifted by one moment each time until the power sampling within the sliding window at the current time point satisfies the following formula. At this point, the climbing event is considered to have ended; the end time of the climbing event is recorded as τ. i,re The duration of the uphill climb is from τ i,rs To τ i,re ; Where, p i,τ and p i,τ-Δτ These represent the output of station i at times τ and τ-Δτ, respectively, where Δτ is the time width of the sliding window. This is the threshold for hill climbing detection.

6. The wind power ramping event prediction method considering extreme weather and spatiotemporal correlation information according to claim 1, characterized in that: Based on the wind turbine operating characteristics during historical ramp events, the K-Means clustering algorithm is used to classify ramp events into different ramp situations of varying severity, including: The operating status of the wind turbines at each site during each ramping event is determined based on the physical parameters of wind turbine operation, and the ramping amplitude, ramping direction, ramping duration and average power output during the ramping event period are used as clustering features. Taking station i as an example, the duration of the climbing event is from τ i,rs To τ i,re : in, and These represent the power output of the station at the start and end of the climb, respectively. upper The wind speed threshold representing the wind turbine due to large wind shear, t lower The temperature threshold representing the temperature at which the fan disconnects from the grid due to low temperature. The magnitude of the climb, representing the climb event. Represents the direction of climbing. Represents the duration of the climb, Represents the average wind power output during the ramp-up event, state i,τ Represents the operating state of the wind turbine at time τ. This represents the operating status of the wind turbine during a ramp-up event; features are extracted from each ramp-up event to obtain a ramp-up sample. The K-Means clustering algorithm was used to classify the station ramp events into different types of ramp events with varying degrees of severity; the number of clusters was set to 6, and 6 samples were randomly selected as the initial cluster centers μ. I Based on the K-Means clustering algorithm, the types of hill climbing are divided into 6 categories: Where I represents the climbing category number, and F represents the sample. The number belonging to the climbing category, μ F Indicates sample Belongs to the cluster center, C I This represents the cluster of the I-th type of hill climbing, where the sample belongs to C. I This means that the distance of the sample is C. I Cluster center μ I It is closer to other cluster centers and is classified into a cluster under the first type of climbing category.

7. The wind power ramping event prediction method considering extreme weather and spatiotemporal correlation information according to claim 2, characterized in that: The method of constructing a graph convolutional neural network based on measured meteorological data, power data, cold wave weather forecast data, and slope conditions of each station to predict slope events includes: constructing a graph convolutional neural network based on measured meteorological data, power data, forecast meteorological data, and slope conditions of each station to predict slope events. The slope prediction model takes the wind power output, measured meteorological data, slope conditions, and forecast meteorological data for the next day as inputs for each station in the cluster three days before the prediction time, and the slope type of a certain station in the cluster at each time of the next day as output. T={T i |i=1,2,...,n}; Where Y is the output of GCN, considered as the spatiotemporal correlation between the mined sites, and T is the feature set of n sites; T i The features considered as those of wind farm i include the measured meteorological data of wind power output and the slope conditions (L) three days prior to the predicted time M. i,M:M-288 The input to the GCN is X, which is a vector composed of the features of each node, and Θ represents the convolution kernel parameter matrix. Let W be the degree matrix of the nodes, W be the adjacency matrix representing the stations, and I be the identity matrix. Type i =Y⊙Θ i,s ; Among them, the ramp-up situation of station i in the next day is Type i =[type i,1 ,type i,2 ,...,type i,96 ], type i,N ∈{1,2,3,4,5,6} represents the climbing situation of the station at time N in the future day; where Θ i,s For the corresponding parameters, ⊙ represents element-wise multiplication; Using 80% of the obtained dataset as the training set, models were trained separately for cold wave weather and non-cold wave weather. Using 20% ​​of the obtained dataset as the test set, cold wave weather prediction models were used to predict cold wave events. If the day to be predicted was predicted to be a cold wave, the cold wave weather climbing identification model was used for prediction. If the day to be predicted was predicted to be a non-cold wave, the non-cold wave weather climbing identification model was used for prediction.

8. A wind power ramping event prediction device that considers extreme weather and spatiotemporal correlation information, characterized in that: include: The acquisition module is used to acquire historical measured meteorological data from each wind farm station in the cluster. The sample expansion module is used to identify cold wave weather events based on historical measured meteorological data and generate an adversarial network based on time series to expand the cold wave event sample. The cold wave weather prediction module is used to construct an extreme learning machine based on the historical measured meteorological data of each wind farm after sample expansion, and to predict cold wave weather. The historical ramp event detection module is used to detect historical ramp events based on the historical measured meteorological data of each wind farm in the cluster. The ramp event classification module is used to classify ramp events into different types of ramp situations with varying degrees of severity based on the wind turbine operating characteristics when historical ramp events occur, using the K-Means clustering algorithm. The slope event prediction module is used to construct a graph convolutional neural network based on measured meteorological data, power data, cold wave weather prediction data, and slope conditions from various stations to predict slope events.

9. A computer device, comprising a storage medium, a processor, and a computer program stored on the storage medium and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the wind power ramping event prediction method that considers extreme weather and spatiotemporal correlation information as described in any one of claims 1-6.

10. A computer storage medium, characterized in that: The computer storage medium contains a computer program, which, when executed by a processor, implements the steps of the wind power ramping event prediction method that considers extreme weather and spatiotemporal correlation information as described in any one of claims 1-6.

Images