A wind power plant thunderstorm approaching early warning method, device and medium

Abstract

This invention relates to the field of meteorological forecasting technology and discloses a method, device, and medium for providing near-term thunderstorm warnings for wind farms. Based on current individual thunderstorm information, this invention uses a pre-constructed thunderstorm path prediction model to predict the thunderstorm path in the future, and obtains a first probability of thunderstorms occurring in a target area based on the thunderstorm path. Based on electric field data and meteorological environmental data within the target area, a second probability of thunderstorms occurring in the target area is calculated using a pre-constructed probability relationship model. The first and second probabilities are then fused to obtain a comprehensive probability, and a thunderstorm warning is issued based on this comprehensive probability. This invention achieves collaborative analysis of multimodal data by fusing the thunderstorm path prediction model and the atmospheric electric field probability relationship model, overcoming the problems of high false alarm rates and low effective alarm rates caused by the independent operation of a single sensor in existing technologies.

Description

Technical Field

[0001] This invention relates to the field of meteorological forecasting technology, specifically to a method, device, and medium for providing near-term thunderstorm warnings for wind farms. Background Technology

[0002] As the global energy structure accelerates its transition to green and low-carbon practices, new energy power plants such as wind farms, photovoltaic power stations, and energy storage bases are being built on a large scale and are widely distributed in complex terrain areas such as mountains, plains, and coastlines. However, due to their tall structures, metallic materials, and electrical sensitivity, new energy equipment such as wind turbine blades, photovoltaic modules, and converters are highly susceptible to lightning strikes during thunderstorms, posing a serious threat to the safe and stable operation of these power plants.

[0003] To address the need for thunderstorm early warning, existing systems mostly employ a single-sensor, independent operation mode, such as atmospheric electric field meters and lightning locators, lacking the ability for multi-modal data collaborative analysis. For example, a common method is to use radar echo data for extrapolation forecasting, but this method is limited by the radar's coverage and the limitations of echo analysis, resulting in significant blind spots in the prediction range. Another approach is a lightning early warning system based on atmospheric electric field meters. Although it performs reasonably well in terms of warning rate (POD) and false negative rate (FTWR), the electric field measurements of ground-based atmospheric electric field meters are highly susceptible to the influence of the surrounding environment and other external interference factors. This leads to an excessively high false alarm rate (FAR) and a very low effective warning rate (EAR, i.e., true warning accuracy), resulting in a high misjudgment rate for pre-warnings. Consequently, it is difficult to meet the actual requirements of new energy power plants for the accuracy and reliability of thunderstorm early warnings. Summary of the Invention

[0004] This invention provides a method, device, and medium for early warning of approaching thunderstorms at wind farms, in order to solve the technical problem that existing thunderstorm early warning technologies lack the ability of single sensors to perform multimodal data collaborative analysis, resulting in a high false alarm rate and a low effective alarm rate.

[0005] In a first aspect, the present invention provides a method for providing near-term thunderstorm warning for wind farms, comprising: acquiring information on individual cells of the current thunderstorm, electric field data of the target area, and meteorological environmental data, wherein the individual cell information includes the centroid coordinates, direction of movement, and distance of movement in adjacent time intervals; based on the current individual thunderstorm information, predicting the thunderstorm path in the future time period using a pre-constructed thunderstorm path prediction model, and obtaining a first probability of thunderstorm occurrence in the target area based on the thunderstorm path; wherein the thunderstorm path prediction model is a model trained using individual cell information of historical thunderstorms and used to predict future thunderstorm paths based on current individual thunderstorm information; based on the electric field data and meteorological environmental data in the target area, calculating a second probability of thunderstorm occurrence in the target area using a pre-constructed probability relationship model; wherein the probability relationship model is a model trained based on historical electric field data, meteorological environmental data, and thunderstorm occurrence events and used to calculate the probability of thunderstorm occurrence based on electric field data and meteorological environmental data; fusing the first probability and the second probability to obtain a comprehensive probability, and issuing a thunderstorm warning based on the comprehensive probability.

[0006] This invention achieves collaborative analysis of multimodal data by fusing a thunderstorm path prediction model and an atmospheric electric field probability relationship model, overcoming the problems of high false alarm rates and low effective alarm rates caused by the independent operation of a single sensor in existing technologies. Specifically, the thunderstorm path prediction model is trained based on individual historical thunderstorm information, enabling accurate prediction of future thunderstorm paths based on current thunderstorm data. The probability relationship model is trained based on historical electric field data, meteorological environmental data, and thunderstorm occurrence events, calculating the probability of thunderstorm occurrence. By fusing the first and second probabilities output by the two models, a comprehensive judgment of the likelihood of thunderstorm occurrence is made, significantly improving the accuracy and reliability of near-term thunderstorm warnings for wind farms.

[0007] In one optional implementation, the thunderstorm path prediction model is a Long Short-Term Memory (LSTM) network model. Its training process includes: acquiring individual information of historical thunderstorms and corresponding centroid coordinate labels for thunderstorm individuals in future time periods; converting the centroid coordinates of individual thunderstorms in the historical information into displacement vectors relative to the previous time period, and combining them with the movement direction and distance traveled in adjacent time periods from the individual information to form an input feature sequence; inputting the input feature sequence and the corresponding centroid coordinate labels for thunderstorm individuals in future time periods into a two-layer LSM network, where a Dropout layer is added between the two LSM layers for regularization, and a fully connected layer is connected after the last LSM layer for feature integration, thus training the thunderstorm path prediction model. In this implementation, by using a LSM network model to construct the thunderstorm path prediction model, the temporal sequence features of thunderstorm individual movement can be effectively captured. Specifically, during training, the centroid coordinates of historical thunderstorm individuals are converted into displacement vectors, and combined with the movement direction and distance traveled in adjacent time periods to form an input feature sequence, enabling the model to fully learn the temporal dependency between the spatial location changes and movement trends of thunderstorm individuals. The combination of a two-layer long short-term memory network with Dropout regularization and fully connected layer feature integration not only improves the model's prediction accuracy of future thunderstorm centroid coordinate sequences, but also enhances generalization ability and reduces the risk of overfitting, thus providing more reliable path prediction results for near-term thunderstorm warnings at wind farms.

[0008] In one optional implementation, the probabilistic relationship model is a logistic regression model. Its training process includes: acquiring historical electric field data, historical meteorological environment data, and corresponding thunderstorm event labels; extracting instantaneous electric field amplitude and electric field intensity difference values ​​from the historical electric field data, and extracting meteorological coupling indicators from the historical meteorological environment data to form an input feature vector; inputting the input feature vector and the corresponding thunderstorm event labels into the logistic regression model to train the probabilistic relationship model. This implementation, by using a logistic regression model to construct the probabilistic relationship model, can effectively integrate multi-dimensional features from atmospheric electric field data and meteorological environment data, establishing a probabilistic mapping relationship between them and thunderstorm events. During training, the instantaneous electric field amplitude and electric field intensity difference values ​​are extracted from the historical electric field data, and meteorological coupling indicators are extracted from the historical meteorological environment data to form the input feature vector, enabling the model to capture the instantaneous state and evolution trend of electric field changes. By training the aforementioned feature vectors together with thunderstorm event labels, the probabilistic relationship model can quantify the probability of thunderstorm occurrence under electric field and meteorological conditions. This overcomes the shortcomings of a single atmospheric electric field meter, which is susceptible to environmental interference and has a high false alarm rate. It also provides a reliable second probability input for subsequent weighted fusion with thunderstorm path prediction results.

[0009] In one alternative implementation, the meteorological coupling indicators include at least one of convective effective potential energy, temperature-dew point difference, and wind shear, thereby enabling a comprehensive characterization of the environmental trend of thunderstorm occurrence from multiple different dimensions, allowing the input features to more fully reflect the true atmospheric state, and thus improving the accuracy of the logistic regression model in predicting the probability of thunderstorm occurrence.

[0010] In one optional implementation, obtaining the first probability of a thunderstorm occurring within a target area based on the thunderstorm path includes: discretizing the predicted thunderstorm path into a series of trajectory points, defining a circular area with a preset radius centered on each trajectory point as the influence range of a single thunderstorm cell; dividing the target area into several sub-regions, and determining the probability of each sub-region being traversed by the thunderstorm as the first probability based on the intersection of each sub-region with the influence range. This implementation, by discretizing the predicted thunderstorm path into a sequence of trajectory points and defining the influence range of a single thunderstorm cell with a circular area of ​​a preset radius, can transform the abstract path prediction result into a strip-shaped influence area with spatial coverage significance. Based on this, dividing the target area into a sub-region grid and determining the probability of being traversed by the thunderstorm based on the intersection of each sub-region with the influence range ensures that the determination of the first probability considers both the spatial scale characteristics of the thunderstorm cell and achieves refined spatial probability assignment for the target area. This provides higher spatial resolution input conditions for subsequent weighted fusion with the second probability, which is beneficial for improving the accuracy of warning level classification.

[0011] In one optional implementation, the probability of each sub-region being crossed by a thunderstorm is determined based on the intersection of each sub-region with the affected area. This includes: determining whether each sub-region has been crossed by the thunderstorm, recording the probability value of the sub-region crossed by the thunderstorm as 1, and the probability value of the sub-region not crossed as 0, as the first probability of each sub-region. This implementation directly maps the intersection determination result of each sub-region with the thunderstorm's affected area to a binary first probability, i.e., recording the probability of the sub-region crossed by the thunderstorm as 1 and the probability of the sub-region not crossed as 0, so that the trajectory information of the thunderstorm path prediction can be transformed into a spatial probability distribution in a concise and clear manner.

[0012] In one optional implementation, the method for determining whether each sub-region has been traversed by a thunderstorm is as follows: for each sub-region, it is determined to have been traversed by a thunderstorm if one of the following conditions is met: there is at least one trajectory point within the sub-region; there is at least one trajectory point such that any vertex of the sub-region is within the influence range corresponding to that trajectory point; there is at least one trajectory point such that any edge of the sub-region intersects with the influence range corresponding to that trajectory point. Determining whether a sub-region has been traversed by a thunderstorm using these three dimensions comprehensively covers all possible spatial relationships between the thunderstorm's influence range and the sub-region, avoiding omissions or misjudgments that might occur with a single determination condition. This ensures the accuracy and completeness of the first probability spatial assignment and provides a reliable spatial basis for the subsequent precise classification of warning levels.

[0013] In one optional implementation, the first probability and the second probability are fused using a weighted fusion method. Thunderstorm warnings based on the comprehensive probability include comparing the comprehensive probability with multiple preset probability thresholds to determine the corresponding thunderstorm warning level. This implementation integrates the first and second probabilities using a weighted fusion method, allowing for flexible adjustment of the contribution ratios of thunderstorm path prediction results and atmospheric electric field and meteorological environment assessment results, making the comprehensive probability closer to the actual thunderstorm occurrence risk. Furthermore, comparing the comprehensive probability with preset multi-level probability thresholds to determine the corresponding thunderstorm warning level achieves standardized and tiered output of warning results, facilitating wind farm maintenance personnel to take differentiated protective measures based on different warning levels, thus improving the practicality and decision-making guidance value of the warnings.

[0014] Secondly, the present invention provides a wind farm thunderstorm proximity warning device, comprising: a data acquisition module for acquiring information on individual thunderstorm cells, electric field data of a target area, and meteorological environmental data, wherein the individual cell information includes the centroid coordinates, direction of movement, and distance traveled in adjacent time intervals; a first probability determination module for predicting the thunderstorm path in the future time period based on the current thunderstorm cell information using a pre-built thunderstorm path prediction model, and obtaining a first probability of thunderstorm occurrence in the target area based on the thunderstorm path; wherein the thunderstorm path prediction model is a model trained using historical thunderstorm cell information and used to predict future thunderstorm paths based on current thunderstorm cell information; a second probability determination module for calculating a second probability of thunderstorm occurrence in the target area based on the electric field data and meteorological environmental data of the target area using a pre-built probability relationship model; wherein the probability relationship model is a model trained based on historical electric field data, meteorological environmental data, and thunderstorm occurrence events and used to calculate the probability of thunderstorm occurrence based on the electric field data and meteorological environmental data; and a thunderstorm warning module for weighted fusion of the first and second probabilities to obtain a comprehensive probability, and issuing a thunderstorm warning based on the comprehensive probability.

[0015] Thirdly, the present invention provides a computer-readable storage medium storing computer instructions for causing a computer to execute the wind farm thunderstorm proximity warning method described in the first aspect or any corresponding embodiment thereof. Attached Figure Description

[0016] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0017] Figure 1 This is a schematic diagram of the first step of the wind farm thunderstorm proximity warning method according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the second process of the wind farm thunderstorm proximity warning method according to an embodiment of the present invention; Figure 3 This is a structural block diagram of a wind farm thunderstorm proximity early warning device according to an embodiment of the present invention; Figure 4 This is a schematic diagram of the hardware structure of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] It is understood that before using the technical solutions disclosed in the various embodiments of the present invention, users should be informed of the types, scope of use, and usage scenarios of the personal information involved in the present invention and their authorization should be obtained in accordance with relevant laws and regulations through appropriate means.

[0020] The terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0021] Existing thunderstorm warning systems mostly rely on single sensors such as atmospheric electric field meters and lightning locators to operate independently, lacking multimodal data collaborative analysis capabilities. Radar echo-based extrapolation forecasts are limited by coverage and echo analysis, resulting in prediction blind spots. While atmospheric electric field meter-based warning schemes perform reasonably well in terms of warning and false alarm rates, they are highly susceptible to environmental interference, leading to excessively high false alarm rates, low effective alarm rates, and high misjudgment rates for early warnings. These shortcomings make it difficult to meet the actual requirements of new energy power plants for the accuracy and reliability of thunderstorm warnings. Therefore, this invention provides a method, device, and medium for thunderstorm proximity warnings for wind farms, addressing the technical problems of high false alarm rates and low effective alarm rates caused by the lack of multimodal data collaborative analysis capabilities in existing thunderstorm warning technologies using single sensors.

[0022] According to an embodiment of the present invention, a method for early warning of approaching thunderstorms at wind farms is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0023] This embodiment provides a method for early warning of approaching thunderstorms at wind farms. Figure 1 This is a flowchart of a wind farm thunderstorm imminent warning method according to an embodiment of the present invention, such as... Figure 1 As shown, the process includes the following steps: Step S101: Obtain the cell information of the current thunderstorm, the electric field data of the target area, and the meteorological environment data. The cell information includes the centroid coordinates of the cell, the direction of movement, and the distance moved in adjacent time intervals.

[0024] This step is the data acquisition step, which aims to provide multimodal data support for subsequent thunderstorm path prediction and thunderstorm occurrence probability calculation.

[0025] Specifically, the current information of individual thunderstorm cells serves as the foundational input for thunderstorm path prediction models. The centroid coordinates of the individual cell describe its current spatial location, its direction of motion reflects its movement trend, and the distance traveled in adjacent time intervals characterizes its velocity. These three pieces of information collectively constitute the core elements describing the motion state of thunderstorm cells, enabling the model to capture the temporal dependency between changes in spatial location and movement trends.

[0026] The electric field data for the target area comes from equipment such as atmospheric electric field meters deployed within wind farms. This data reflects the instantaneous state and changing trends of the atmospheric electric field over the target area in real time, serving as a crucial precursor signal for determining the likelihood of thunderstorms. Meteorological environmental data provides background information for assessing the environmental potential for thunderstorms, including environmental parameters characterizing atmospheric stratification instability and humidity conditions. The combination of electric field data and meteorological environmental data provides comprehensive input information for probabilistic relationship models.

[0027] Step S102: Based on the current thunderstorm cell information, the thunderstorm path in the future period is predicted using a pre-built thunderstorm path prediction model, and the first probability of thunderstorms occurring in the target area is obtained based on the thunderstorm path; wherein, the thunderstorm path prediction model is a model trained using the cell information of historical thunderstorms and used to predict the future thunderstorm path based on the current thunderstorm cell information.

[0028] This step first inputs the acquired current thunderstorm cell information into a pre-constructed thunderstorm path prediction model. This model is a time-series prediction model trained using historical thunderstorm cell information, capable of learning the spatiotemporal evolution of thunderstorm cell motion states from historical thunderstorm cell centroid coordinate sequences, motion direction sequences, and adjacent time-series movement distance sequences. In the actual inference phase, based on the input cell motion feature sequence, the model outputs a predicted coordinate sequence of the thunderstorm cell centroid position for future time periods, forming a continuous thunderstorm prediction path.

[0029] Based on this, the predicted thunderstorm path is further transformed into the first probability of thunderstorms occurring within the target area. Considering that thunderstorm cells have a certain spatial scale and are not idealized point masses, this step discretizes the predicted path into a series of trajectory points, and uses a circular area with a preset radius centered on each trajectory point as the influence range of the thunderstorm cell at that moment. By performing an intersection analysis between this influence range and the spatial division of the target area, it can be determined whether each location within the target area is on the predicted path of the thunderstorm, thus obtaining the spatialized first probability distribution.

[0030] Step S103: Based on the electric field data and meteorological environment data in the target area, calculate the second probability of thunderstorms occurring in the target area using a pre-built probability relationship model; wherein, the probability relationship model is a model trained based on historical electric field data, meteorological environment data and thunderstorm occurrence events, used to calculate the probability of thunderstorm occurrence based on electric field data and meteorological environment data.

[0031] This step inputs the acquired electric field data and meteorological environmental data within the target area into a pre-constructed probabilistic model. This model is a probabilistic prediction model trained on historical electric field data, meteorological environmental data, and thunderstorm occurrence events. It can learn the evolution patterns of the atmospheric electric field and the statistical mapping relationship between meteorological environmental changes and thunderstorm occurrence events from multi-dimensional features such as instantaneous electric field amplitude, electric field intensity difference, and meteorological coupling indicators. In the actual inference stage, the model directly outputs the second probability of thunderstorms occurring within the target area based on the input real-time electric field characteristics and environmental feature vectors.

[0032] Unlike step S102, which relies on extrapolating the spatial path of thunderstorms, the second probability in this step reflects the overall atmospheric potential of the target area. This probability estimate, derived from the perspective of charge accumulation and whether the atmospheric environment meets the conditions for thunderstorm occurrence, complements the first probability in step S102 regarding the location of individual thunderstorms. This step effectively integrates the instantaneous state and trends of electric field data through a probabilistic model, incorporating the combined influence of meteorological coupling indicators such as convective effective potential energy, temperature-dew point difference, and wind shear. This overcomes the shortcomings of single atmospheric electric field meters, which are susceptible to local environmental interference and have a high false alarm rate, providing background probability input based on environmental conditions for subsequent dual-probability fusion early warning.

[0033] Step S104: The first probability and the second probability are fused to obtain the comprehensive probability, and a thunderstorm warning is issued based on the comprehensive probability.

[0034] The first probability reflects the spatial movement trend of individual thunderstorm cells, while the second probability reflects the potential atmospheric trend of whether the current environment in the target area is conducive to thunderstorm occurrence. These two probabilities independently assess the likelihood of thunderstorm occurrence from different dimensions and are naturally complementary. When the overall atmospheric environment is unfavorable for thunderstorm occurrence, even if the predicted path indicates that thunderstorms may pass through certain areas, the combined probability will be lowered by the second probability, effectively reducing the false alarm rate. Conversely, when the atmospheric environment is favorable for thunderstorm occurrence, even if there is some uncertainty in the path prediction, the second probability, as a background probability, can still improve the overall level of the combined probability and reduce the risk of missed alarms.

[0035] In terms of fusion method, this step can perform weighted fusion of the first probability and the second probability. By adjusting the weight ratio of the two types of probabilities, it can flexibly adapt to the characteristics of thunderstorm activity in different regions and seasons. The comprehensive probability obtained by fusion combines the refined advantages of spatial path prediction with the global perspective of environmental condition judgment, realizing multimodal data synergy between thunderstorm motion information and atmospheric electric field and meteorological environment information.

[0036] Based on this, the overall probability is compared with preset multi-level probability thresholds to determine the corresponding thunderstorm warning level, achieving standardized and tiered output of warning results. For example, the higher the overall probability, the higher the warning level, meaning a greater likelihood and threat of thunderstorms. Wind farm maintenance personnel can take differentiated protective measures in advance according to different warning levels, such as activating lightning protection devices, suspending high-altitude operations, and switching equipment operating modes, thereby effectively avoiding lightning strike risks and ensuring the safety of farm equipment and personnel.

[0037] The wind farm thunderstorm imminent warning method provided in this embodiment achieves collaborative analysis of multimodal data by fusing a thunderstorm path prediction model and an atmospheric electric field probability relationship model. This overcomes the problems of high false alarm rate and low effective alarm rate caused by the independent operation of a single sensor in existing technologies. Specifically, the thunderstorm path prediction model is trained based on individual historical thunderstorm information and can accurately predict future thunderstorm paths based on current thunderstorm information. The probability relationship model is trained based on historical electric field data, meteorological environmental data, and thunderstorm occurrence events, thereby calculating the probability of thunderstorm occurrence. By fusing the first probability and the second probability output by the two models, the likelihood of thunderstorm occurrence is comprehensively judged, significantly improving the accuracy and reliability of the wind farm thunderstorm imminent warning.

[0038] This embodiment provides a method for early warning of approaching thunderstorms at wind farms. Figure 2 This is a flowchart of a wind farm thunderstorm imminent warning method according to an embodiment of the present invention, such as... Figure 2 As shown, the process includes the following steps: Step S201: Obtain the cell information of the current thunderstorm, the electric field data of the target area, and the meteorological environment data. The cell information includes the cell's centroid coordinates, direction of movement, and distance traveled in adjacent time intervals. For details, please refer to [link to relevant documentation]. Figure 1 Step S101 of the illustrated embodiment will not be described again here.

[0039] Step S202: Based on the current thunderstorm cell information, a pre-built thunderstorm path prediction model is used to predict the thunderstorm path in the future period, and the first probability of thunderstorms occurring in the target area is obtained based on the thunderstorm path; wherein, the thunderstorm path prediction model is a model trained using the cell information of historical thunderstorms and used to predict the future thunderstorm path based on the current thunderstorm cell information.

[0040] Specifically, step S202 above includes: Step S2021: Construct a thunderstorm path prediction model using individual historical thunderstorm information.

[0041] In one specific implementation, the thunderstorm path prediction model is a long short-term memory network model, trained using individual information from historical thunderstorms. The training process includes: Step a1: Obtain the individual information of historical thunderstorms and the corresponding centroid coordinates of thunderstorms in the future time period.

[0042] Specifically, the data first includes radar reflectivity data of the target area, centroid coordinates of thunderstorm cells, direction of movement of thunderstorm cells, and distance traveled by thunderstorm cells in adjacent time periods, all obtained from meteorological departments over the past five years. The acquired historical data undergoes preprocessing, including data cleaning, missing value imputation, and outlier handling, to ensure data integrity and accuracy.

[0043] After data preprocessing, thunderstorm cells are identified and their motion features are extracted. Specifically, a Cartesian coordinate system is established with the radar as the origin to identify the position of the three-dimensional centroid of the thunderstorm cell, and the three-dimensional centroid is projected onto a two-dimensional coordinate plane. The identification method for thunderstorm cells is as follows: thunderstorm cells are identified by analyzing thunderstorm information from different time periods within the thunderstorm development time series. The two-dimensional coordinates of the thunderstorm cell centroid are used as historical samples. Based on the historical centroid coordinates, a smooth curve is fitted using a Bezier curve to predict the direction of movement of the thunderstorm cell. The shortest distance between centroids at adjacent time points in the historical samples is used as the distance of each adjacent thunderstorm cell in the forecast result.

[0044] The sample data was divided into training, validation, and test sets according to time sequence, with the test set data occurring later than the training set data. All features were normalized, including the two-dimensional coordinates of the thunderstorm cell centroid, the historical direction of thunderstorm cell movement, and the distance traveled by thunderstorm cells in adjacent time intervals.

[0045] Step a2: Convert the centroid coordinates of individual cells in the individual cell information of historical thunderstorms into displacement vectors relative to the previous time interval, and combine them with the direction of motion and the distance of movement in adjacent time intervals in the individual cell information to form the input feature sequence.

[0046] Specifically, the input features use historical time window data at times [t-τ, t-τ+1, ..., t-1, t], including the centroid coordinates (x, y) of thunderstorm cells at each time point. i ,y i ), and the corresponding direction of motion θ i And the corresponding movement distance d i The centroid coordinates (x) i ,y i ) is converted into a displacement vector (Δx) relative to the previous time step. i =x i -x i-1 ,Δy i =y i -y i-1The core input features are defined as follows: Input(shape=(T,F)), where T is the time step and F is the number of features, including four features: Δx, Δy, d, and θ. The corresponding labels are the centroid coordinates of thunderstorm cells within the future time period.

[0047] Step a3: Input the input feature sequence and the corresponding centroid coordinate labels of thunderstorm cells in the future time period into a two-layer long short-term memory network. A dropout layer is added between the two long short-term memory layers for regularization. The last long short-term memory network layer is connected to a fully connected layer for feature integration, and the thunderstorm path prediction model is trained.

[0048] Specifically, the main component of the LSTM structure is the forget gate f t Input gate i t and output gate o t .

[0049] The forgetting gate formula is: f t =σ(W f ·[h t-1 ,x t ]+b f ), where f t Output for the Forgot Gate, W f For linear coefficients, b i Here, σ represents the bias coefficient, and h represents the sigmoid function. t-1 It is the output from the previous moment, and the input signal x t .

[0050] Introducing an attention mechanism into the input gate, the modified formula is: i t =σ(W i ·[h t-1 ,x t ]+U i ·d t +b i ),in .

[0051] The LSTM cell state update formula is: , where C t Let represent the current LSTM cell state, and ⊙ represent the Hadamard product. Where i t and are the outputs of the Sigmoid activation function and the tanh activation function, respectively; W i W C b are linear coefficients; i b C This is the bias coefficient.

[0052] The output gate formula is: o t =σ(W o ·[h t-1,x t ]+b o ), h t =o t ⊙tanh(C t ), where o t For output gate output, h t Wo is the output of the hidden layer, and bo is the initial bias coefficient.

[0053] Applying LayerNorm after the linear transformation of each gate and before the activation function improves training stability and accelerates convergence.

[0054] In terms of network structure, a two-layer LSTM is used, with a Dropout layer added between the two LSTM layers for regularization to prevent overfitting. The last LSTM layer is followed by a fully connected layer for nonlinear transformation and feature integration, ultimately outputting the predicted coordinate sequence of the centroid of thunderstorm cells in the future time period.

[0055] The loss function combines mean squared error and mean absolute error. When the prediction error is less than the preset threshold δ, mean squared error is used, and when it is greater than δ, mean absolute error is used, thus balancing convergence speed and outlier resistance.

[0056] During training, hyperparameters such as learning rate, batch size, and number of iterations are determined, and the Adam optimization algorithm is used to optimize the model parameters. The model performance is evaluated on the validation set at regular intervals. If the model meets preset conditions, it is saved; otherwise, the model structure and hyperparameters are adjusted, and the model is retrained.

[0057] Step S2022: Input the current thunderstorm cell information into the thunderstorm path prediction model constructed through the above training process to predict the thunderstorm path in the future period.

[0058] Step S2023: Obtain the first probability of a thunderstorm occurring in the target area based on the thunderstorm path.

[0059] Specifically, the thunderstorm path predicted in step S2022 is transformed into a probability distribution of each sub-region within the target area being traversed by the thunderstorm, including: Step b1: Discretize the predicted thunderstorm path into a series of trajectory points, and use a circular area with a preset radius centered on each trajectory point as the influence range of the thunderstorm cell.

[0060] Specifically, the predicted coordinate sequence of the centroid of a thunderstorm cell within a future time period (e.g., the next 2 hours) output by the thunderstorm path prediction model is taken as a trajectory T. i The trajectory consists of a series of predicted points: P0, P1, ..., P k(Total k+1 points). Since thunderstorm cells in reality have a certain spatial scale and cannot be simply considered as point masses, the influence range of a thunderstorm cell is considered. That is, a circular area with a preset radius R centered on each trajectory point is taken as the coverage area of ​​the thunderstorm cell at that moment. The influence range of the entire predicted path is actually a strip-shaped area composed of the union of multiple circular areas and the area swept by the line segment connecting the centroids of adjacent time points.

[0061] To ensure the accuracy of discretization and avoid missing the gap between two adjacent circular regions, the trajectory T is... i Interpolation is performed, for example, every 1 km or every 10 seconds, to obtain a series of dense trajectory points: Q0, Q1, ..., Q m The interpolation intervals are sufficiently close, allowing the discrete point sequence to move approximately continuously, thus ensuring that the gaps between two circles are not missed and guaranteeing the completeness and accuracy of subsequent intersection determination.

[0062] Step b2: Divide the target area into several sub-regions. Based on the intersection of each sub-region with the affected area, determine the probability that each sub-region will be crossed by the thunderstorm as the first probability.

[0063] First, the target area is divided into several sub-regions based on factors such as its geographical location and terrain features. In practical applications, sub-regions are typically regular square grids, such as 10km × 10km squares.

[0064] Then, for each sub-region A, determine whether it has been crossed by a thunderstorm. The specific determination method is: iterate through all interpolated trajectory points Q. j (j=0,1,...,m), determine with Q j Does the circular influence region centered at point Q with radius R intersect with sub-region A? If there exists at least one trajectory point Q... j If the corresponding circular region intersects with sub-region A, then label I(T) i If I(T) = 1, it means that the sub-region was crossed by a thunderstorm; otherwise, I(T) = 1. i ,A)=0.

[0065] Subregion A is usually a square polygon. The method to determine whether a circular region (center O, radius R) intersects with subregion A is as follows: If one of the following conditions is met, it is considered an intersection: Condition 1: The center of the circle is located within the sub-region, meaning there exists at least one trajectory point within the sub-region; Condition 2: Any vertex of the sub-region is inside the circle, that is, there exists at least one trajectory point such that the distance from any vertex of the sub-region to the trajectory point is less than or equal to the preset radius R, that is, there exists at least one trajectory point such that any vertex of the sub-region is within the influence range corresponding to the trajectory point. Condition 3: Any edge of the sub-region intersects the circle, that is, there exists at least one trajectory point such that any edge of the sub-region intersects the circular region corresponding to the trajectory point, that is, the distance from the center of the circle to the edge is less than or equal to the radius R.

[0066] In one optional implementation, the probability of each sub-region being crossed by the thunderstorm is determined based on the intersection of each sub-region with the affected area. This includes: determining whether each sub-region has been crossed by the thunderstorm, recording the probability value of the sub-region crossed by the thunderstorm as 1, and the probability value of the sub-region not crossed as 0, as the first probability of each sub-region. Therefore, for I(T) i The probability that a subregion with A) = 1 will be crossed by a thunderstorm is the probability of the corresponding trajectory occurring.

[0067] Step S203: Based on the electric field data and meteorological environment data in the target area, calculate the second probability of thunderstorms occurring in the target area using a pre-built probability relationship model; wherein, the probability relationship model is a model trained based on historical electric field data, meteorological environment data and thunderstorm occurrence events, used to calculate the probability of thunderstorm occurrence based on electric field data and meteorological environment data.

[0068] Specifically, step S203 above includes: Step S2031: Construct a probabilistic relationship model using historical electric field data, meteorological environment data, and information on thunderstorm occurrence events.

[0069] In one specific implementation, the probabilistic relationship model is a logistic regression model, and its training process includes: Step c1: Obtain historical electric field data, historical meteorological environment data, and corresponding thunderstorm event tags.

[0070] Specifically, historical electric field intensity data and synchronous meteorological parameters are collected. For each time point t and each sub-region k, a training sample is constructed, and the feature values ​​in the sample include: Instantaneous electric field amplitude E: The electric field strength at the current time t; Electric field strength difference ΔE: The difference in electric field strength between the current time t and the previous time, or the average rate of change over a period of time; Meteorological coupling index: an index calculated based on meteorological parameters, such as at least one of convective effective potential energy (CAPE), temperature-dew point difference (T-Td), and wind shear.

[0071] The corresponding label is the thunderstorm occurrence event label, indicating whether a thunderstorm actually occurred at that time and in that sub-region.

[0072] Step c2 involves extracting the instantaneous amplitude and electric field intensity difference from historical electric field data, and extracting meteorological coupling indicators from historical meteorological and environmental data to form the input feature vector.

[0073] Specifically, the instantaneous amplitude E of the electric field, the difference in electric field intensity ΔE, and meteorological coupling indicators are used to construct the input feature vector. The difference in electric field intensity ΔE refers to the difference in electric field intensity between the current moment and the previous moment (or the average rate of change over a period of time). The meteorological coupling indicators include at least one of the following: convective available potential energy (CAPE), temperature-dew point difference (T-Td), and wind shear.

[0074] Step c3: Input the input feature vector and the corresponding thunderstorm event label into the logistic regression model to train the probability relationship model.

[0075] The logistic regression model maps feature vectors to the probability of thunderstorms occurring. The model formula is as follows: ; in = ,Right now ; These are weighting coefficients. It is the intercept (bias term). Given a feature vector The probability of thunderstorms occurring under certain conditions.

[0076] Through the above training process, the logistic regression model can learn the instantaneous state of the electric field, the trend of electric field changes, and the statistical mapping relationship between meteorological environmental conditions and thunderstorm events. Thus, in the actual prediction stage, based on the input real-time electric field data and meteorological environmental data, it can output the second probability of thunderstorms occurring in the target area.

[0077] Step S2032: Input the electric field data and meteorological environment data in the target area into the probability relationship model constructed through the above training process, and calculate the second probability of thunderstorms occurring in the target area.

[0078] Step S204: The first probability and the second probability are fused to obtain the comprehensive probability, and a thunderstorm warning is issued based on the comprehensive probability.

[0079] In one specific implementation, the method of fusing the first probability and the second probability is weighted fusion.

[0080] Specifically, for each sub-region, the first probability and the second probability of that sub-region are weighted and summed to obtain the overall probability of that sub-region. The formula for weighted fusion is: P 综合，k =w1·P 1，k 10w2·P2 Among them, P 综合，k Let P be the overall probability of the k-th sub-region. 1，k P1 is the first probability of the k-th sub-region (with a value of 0 or 1, indicating whether the sub-region is crossed by a thunderstorm), P2 is the second probability of the entire target area (the probability of thunderstorm occurrence predicted based on electric field and meteorological data), w1 and w2 are the first weight and the second weight, respectively, and w1+w2=1.

[0081] In one specific implementation, the first weight w1 is set to 0.7, and the second weight w2 is set to 0.3. It should be noted that the above weights are preset and can be flexibly adjusted according to the characteristics of thunderstorm activity in different regions and seasons in practical applications.

[0082] After obtaining the combined probability of each sub-region, the combined probability is compared with multiple preset probability thresholds to determine the corresponding thunderstorm warning level. In one specific implementation, the preset probability thresholds and corresponding warning levels are as follows: A red alert is issued when the overall probability is above 0.9. When the overall probability is between 0.7 and 0.9, it is designated as an orange alert; When the overall probability is between 0.5 and 0.7, it is designated as a yellow alert; When the overall probability is below 0.5, it is designated as a blue alert.

[0083] The thunderstorm imminent warning method for wind farms provided in this embodiment, by combining a thunderstorm path prediction model and atmospheric electric field data analysis, can not only predict the time and location of thunderstorms, but also determine the corresponding warning level based on the probability of thunderstorms occurring in each sub-region, providing a basis for subsequent defense decisions and improving the reliability and practicality of thunderstorm warnings. By acquiring data such as the two-dimensional coordinates of the centroid of historical thunderstorm cells, the direction of movement of historical thunderstorm cells, and the movement distance of thunderstorm cells in adjacent time periods to construct a training sample set, and training the LSTM model based on the training sample set, an accurate thunderstorm path prediction model can be obtained. This overcomes the shortcomings of existing technologies that rely on radar data and lightning data, which have lag and uncertainty, and improves the accuracy of thunderstorm path prediction.

[0084] This embodiment also provides a wind farm thunderstorm proximity early warning device, which is used to implement the above embodiments and preferred embodiments, and will not be repeated as already described. As used below, the term "module" can be a combination of software and / or hardware that implements a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0085] This embodiment provides a thunderstorm proximity early warning device for wind farms, such as... Figure 3 As shown, it includes: The data acquisition module 301 is used to acquire the cell information of the current thunderstorm, the electric field data of the target area and the meteorological environment data. The cell information includes the centroid coordinates of the cell, the direction of movement and the distance moved in adjacent time intervals. The first probability determination module 302 is used to predict the thunderstorm path in the future period based on the current thunderstorm cell information and a pre-built thunderstorm path prediction model, and to obtain the first probability of thunderstorms occurring in the target area based on the thunderstorm path; wherein, the thunderstorm path prediction model is a model trained using the cell information of historical thunderstorms and used to predict the future thunderstorm path based on the current thunderstorm cell information. The second probability determination module 303 is used to calculate the second probability of thunderstorms occurring in the target area based on electric field data and meteorological environment data in the target area using a pre-built probability relationship model; wherein, the probability relationship model is a model trained based on historical electric field data, meteorological environment data and thunderstorm occurrence events, used to calculate the probability of thunderstorm occurrence based on electric field data and meteorological environment data. The thunderstorm warning module 304 is used to perform weighted fusion based on the first probability and the second probability to obtain a comprehensive probability, and to issue a thunderstorm warning based on the comprehensive probability.

[0086] In some alternative implementations, the first probability determination module 302 includes: The thunderstorm path prediction model building unit is used to construct a thunderstorm path prediction model using individual information from historical thunderstorms.

[0087] The thunderstorm path prediction unit is used to input the current individual thunderstorm information into the thunderstorm path prediction model constructed through the above training process to predict the thunderstorm path in the future period.

[0088] The first probability calculation unit is used to obtain the first probability of a thunderstorm occurring in the target area based on the thunderstorm path.

[0089] In some alternative implementations, the second probability determination module 303 includes: The probabilistic relationship model building unit is used to construct probabilistic relationship models using information such as historical electric field data, meteorological environment data, and thunderstorm occurrence events.

[0090] The second probability calculation unit is used to input the electric field data and meteorological environment data in the target area into the probability relationship model constructed through the above training process, and calculate the second probability of thunderstorms occurring in the target area.

[0091] The wind farm thunderstorm proximity early warning device provided in this embodiment of the invention can execute the wind farm thunderstorm proximity early warning method provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects of the method. Further functional descriptions of the above modules and units are the same as in the corresponding embodiments described above, and will not be repeated here.

[0092] Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention.

[0093] The following is a detailed reference. Figure 4 This diagram illustrates a structural schematic suitable for implementing an electronic device according to embodiments of the present invention. The electronic device may include a processor (e.g., a central processing unit, graphics processor, etc.) 401, which can perform various appropriate actions and processes according to a program stored in read-only memory (ROM) 402 or a program loaded from memory 408 into random access memory (RAM) 403. The RAM 403 also stores various programs and data required for the operation of the electronic device. The processor 401, ROM 402, and RAM 403 are interconnected via a bus 404. An input / output (I / O) interface 405 is also connected to the bus 404.

[0094] Typically, the following devices can be connected to I / O interface 405: input devices 406 including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices 407 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; memory devices 408 including, for example, magnetic tapes, hard disks, etc.; and communication devices 409. Communication device 409 allows electronic devices to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 4 Electronic devices with various devices are shown, but it should be understood that it is not required to implement or have all of the devices shown, and more or fewer devices may be implemented or have instead.

[0095] In particular, according to embodiments of the present invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of the present invention include a computer program product comprising a computer program carried on a non-transitory computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device 409, or installed from a memory 408, or installed from a ROM 402. When the computer program is executed by the processor 401, it performs the functions defined in the wind farm thunderstorm proximity warning method of the embodiments of the present invention.

[0096] Figure 4The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of use of the embodiments of the present invention.

[0097] This invention also provides a computer-readable storage medium. The methods described above according to embodiments of the invention can be implemented in hardware or firmware, or implemented as computer code that can be recorded on a storage medium, or implemented as computer code downloaded via a network and originally stored on a remote storage medium or a non-transitory machine-readable storage medium and then stored on a local storage medium. Thus, the methods described herein can be processed by software stored on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. The storage medium can be a magnetic disk, optical disk, read-only memory, random access memory, flash memory, hard disk, or solid-state drive, etc.; further, the storage medium can also include combinations of the above types of memory. It is understood that computers, processors, microprocessor controllers, or programmable hardware include storage components capable of storing or receiving software or computer code. When the software or computer code is accessed and executed by the computer, processor, or hardware, the wind farm thunderstorm proximity warning method shown in the above embodiments is implemented.

[0098] A portion of this invention can be applied as a computer program product, such as computer program instructions, which, when executed by a computer, can invoke or provide the methods and / or technical solutions according to the invention through the operation of the computer. Those skilled in the art will understand that the forms in which computer program instructions exist in a computer-readable medium include, but are not limited to, source files, executable files, installation package files, etc. Correspondingly, the ways in which computer program instructions are executed by a computer include, but are not limited to: the computer directly executing the instructions, or the computer compiling the instructions and then executing the corresponding compiled program, or the computer reading and executing the instructions, or the computer reading and installing the instructions and then executing the corresponding installed program. Here, the computer-readable medium can be any available computer-readable storage medium or communication medium accessible to a computer.

[0099] Although embodiments of the invention have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the invention, and such modifications and variations all fall within the scope defined by the appended claims.

Claims

1. A wind farm station thunderstorm approaching early warning method, characterized in that, The method includes: Acquire the individual cell information of the current thunderstorm, the electric field data of the target area, and the meteorological environment data, wherein the individual cell information includes the centroid coordinates of the individual cell, the direction of movement, and the distance moved in adjacent time intervals; Based on the current thunderstorm cell information, a pre-built thunderstorm path prediction model is used to predict the thunderstorm path in the future time period, and a first probability of thunderstorms occurring in the target area is obtained based on the thunderstorm path; wherein, the thunderstorm path prediction model is a model trained using the cell information of historical thunderstorms and used to predict the future thunderstorm path based on the current thunderstorm cell information. Based on the electric field data and meteorological environment data within the target area, a second probability of thunderstorms occurring within the target area is calculated using a pre-constructed probability relationship model; wherein, the probability relationship model is a model trained based on historical electric field data, meteorological environment data, and thunderstorm occurrence events, used to calculate the probability of thunderstorm occurrence based on electric field data and meteorological environment data; The first probability and the second probability are fused to obtain a comprehensive probability, and a thunderstorm warning is issued based on the comprehensive probability.

2. The method for early warning of approaching thunderstorms at wind farms according to claim 1, characterized in that, The thunderstorm path prediction model is a long short-term memory network model, and its training process includes: Obtain historical information on individual thunderstorm cells and their corresponding centroid coordinates for future thunderstorm cells. The centroid coordinates of individual cells in the historical thunderstorm cell information are converted into displacement vectors relative to the previous time interval, and together with the direction of motion and the distance of movement in adjacent time intervals in the cell information, they form an input feature sequence. The input feature sequence and the corresponding centroid coordinates of thunderstorm cells in the future time period are input into a two-layer long short-term memory network. A dropout layer is added between the two long short-term memory layers for regularization. The last long short-term memory layer is followed by a fully connected layer for feature integration, and the thunderstorm path prediction model is trained to obtain the model.

3. The wind farm site thunderstorm nowcast method of claim 1, wherein, The probabilistic relationship model is a logistic regression model, and its training process includes: Acquire historical electric field data, historical meteorological environment data, and corresponding thunderstorm event tags; The instantaneous amplitude of the electric field and the difference in electric field intensity are extracted from the historical electric field data, and meteorological coupling indicators are extracted from the historical meteorological environment data to form an input feature vector; The input feature vector and the corresponding thunderstorm event label are input into the logistic regression model to train the probability relationship model.

4. The wind farm site thunderstorm nowcast method of claim 3, wherein, The meteorological coupling index includes at least one of convective effective potential energy, temperature-dew point difference, and wind shear.

5. The wind farm site thunderstorm nowcast method of claim 1, wherein, The first probability of a thunderstorm occurring in the target area based on the thunderstorm path includes: The predicted thunderstorm path is discretized into a series of trajectory points, and a circular area with a preset radius centered on each trajectory point is taken as the influence range of the thunderstorm unit. The target area is divided into several sub-regions. Based on the intersection of each sub-region with the influence range, the probability that each sub-region is passed through by the thunderstorm is determined as the first probability.

6. The wind farm site thunderstorm nowcast method of claim 5, wherein, Based on the intersection of each sub-region with the aforementioned influence range, the probability of each sub-region being penetrated by the thunderstorm is determined, including: Determine whether each sub-region has been crossed by a thunderstorm. The probability value of a sub-region that has been crossed by a thunderstorm is recorded as 1, and the probability value of a sub-region that has not been crossed by a thunderstorm is recorded as 0. These are used as the first probability of each sub-region.

7. The wind farm site thunderstorm nowcast method of claim 6, wherein, The method for determining whether each sub-region has been passed through by a thunderstorm is as follows: for each sub-region, it is determined to have been passed through by a thunderstorm if one of the following conditions is met: There exists at least one trajectory point within the sub-region; There exists at least one trajectory point such that any vertex of the sub-region is located within the influence range corresponding to that trajectory point; There exists at least one trajectory point such that any edge of the sub-region intersects with the influence range corresponding to that trajectory point.

8. The method for early warning of approaching thunderstorms at wind farms according to claim 1, characterized in that: The method of fusing the first probability and the second probability is weighted fusion; The step of issuing a thunderstorm warning based on the comprehensive probability includes: comparing the comprehensive probability with multiple preset probability thresholds to determine the corresponding thunderstorm warning level.

9. A thunderstorm proximity early warning device for wind farms, characterized in that, The device includes: The data acquisition module is used to acquire the individual cell information of the current thunderstorm, the electric field data of the target area, and the meteorological environment data. The individual cell information includes the centroid coordinates of the individual cell, the direction of movement, and the distance moved in adjacent time intervals. The first probability determination module is used to predict the thunderstorm path in the future time period based on the current thunderstorm cell information using a pre-built thunderstorm path prediction model, and to obtain the first probability of thunderstorms occurring in the target area based on the thunderstorm path; wherein, the thunderstorm path prediction model is a model trained using the cell information of historical thunderstorms and used to predict the future thunderstorm path based on the current thunderstorm cell information. The second probability determination module is used to calculate the second probability of thunderstorms occurring in the target area based on electric field data and meteorological environment data in the target area using a pre-constructed probability relationship model; wherein, the probability relationship model is a model trained based on historical electric field data, meteorological environment data and thunderstorm occurrence events, used to calculate the probability of thunderstorm occurrence based on electric field data and meteorological environment data; The thunderstorm warning module is used to perform weighted fusion based on the first probability and the second probability to obtain a comprehensive probability, and to issue a thunderstorm warning based on the comprehensive probability.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions for causing the computer to execute the wind farm thunderstorm proximity warning method according to any one of claims 1 to 8.

Images