Meteorological auxiliary operation intelligent decision method, device and medium

Abstract

This invention provides a method, device, and medium for intelligent decision-making in meteorological auxiliary operations. First, Doppler radar data is acquired. Then, based on the Doppler radar data, the spatial distribution information of cloud clusters is determined. Next, based on the spatial distribution information and a clustering algorithm, the cloud clusters are adaptively segmented to obtain independent cloud cluster units. The neighborhood radius and core point threshold of the clustering algorithm are adaptively adjusted according to the spatial distribution information of the cloud clusters. Then, the cloud trajectory of the independent cloud cluster units is analyzed, and the current cloud cluster position is determined based on the cloud trajectory and Doppler radar data. Finally, based on the current cloud cluster position, auxiliary operation decision information is determined. This invention achieves intelligent identification of cloud spatial distribution through Doppler radar data and adaptively and dynamically adjusts the neighborhood radius and core point threshold of the clustering algorithm according to the actual spatial characteristics of the cloud clusters, achieving accurate segmentation of cloud clusters at different scales such as hail nuclei and rainbands, improving the accuracy of cloud cluster identification, and ensuring the effectiveness of operations.

Description

Technical Field

[0001] This invention relates to the field of meteorological operations technology, and in particular to an intelligent decision-making method, equipment and medium for meteorological auxiliary operations. Background Technology

[0002] As an important means of responding to extreme meteorological disasters such as drought, hail, and rainstorms, artificial weather modification plays an irreplaceable role in fields such as agricultural production, ecological protection, disaster prevention and mitigation, and water resource regulation.

[0003] Traditional weather warning and operational decision-making methods often employ strong center discrimination, echo height comparison, three-body scattering identification, and multi-factor weighting. These methods rely on fixed thresholds and static weights for judgment, making them unable to adapt to changes in cloud characteristics across different seasons, cloud systems, and geographical environments. This leads to issues such as missed warnings, false alarms, and warning location deviations. Furthermore, traditional storm identification algorithms, based on fixed topology and boundary assumptions, are prone to incorrectly merging multicellular and mixed cloud clusters. This results in key targets such as hail nuclei being diluted by rain cloud data, making it impossible to accurately identify independent operational units. Summary of the Invention

[0004] This invention provides a meteorological auxiliary operation intelligent decision-making method, equipment and medium to address the problems of existing technology comparison methods being unable to adapt to changes in cloud characteristics under different seasons, different cloud systems and different geographical environments, easily merging multicellular single clouds and mixed clouds, and failing to accurately identify independent operation units.

[0005] A first aspect of this invention provides a meteorological-assisted operational intelligent decision-making method, comprising: Acquire Doppler radar data; Based on Doppler radar data, determine the spatial distribution information of the cloud cluster; Based on the spatial distribution information of cloud clusters and clustering algorithms, cloud clusters are adaptively segmented to obtain independent cloud cluster units; among them, the neighborhood radius and core point threshold of the clustering algorithm are adaptively adjusted according to the spatial distribution information of cloud clusters. Analyze the cloud trajectory of independent cloud clusters, and determine the current cloud cluster location based on the cloud trajectory and Doppler radar data; Based on the current location of the cloud cluster, determine the information for auxiliary operation decision-making.

[0006] In one possible implementation, the cloud clusters are adaptively segmented based on their spatial distribution information and a clustering algorithm to obtain independent cloud cluster units, including: Based on the spatial distribution information of cloud clusters, the initial clustering parameters of the clustering algorithm are determined; Clustering algorithms were used to pre-segment the cloud clusters and point clouds corresponding to Doppler radar data to obtain preliminary cloud cluster segmentation results; The preliminary cloud segmentation results are compared with the spatial distribution information of the cloud clusters to determine their matching degree. Based on the matching results, the neighborhood radius and core point threshold of the clustering algorithm are adaptively adjusted. The clustering parameters were adjusted and the clustering was performed again to obtain independent cloud units.

[0007] In one possible implementation, the neighborhood radius and core point threshold of the clustering algorithm are adaptively adjusted based on the matching degree comparison results, including: Based on the matching results, the segmentation defect type and defect score are determined; Adjust the neighborhood radius and core point threshold based on the segmentation defect type and defect score.

[0008] In one possible implementation, the spatial distribution information of the cloud cluster is determined based on Doppler radar data, including: Generating radar composite reflectivity maps based on Doppler radar data; By using an AI-powered visual model to analyze the visual features of the radar composite reflectivity map, spatial distribution information of the cloud clusters can be obtained.

[0009] In one possible implementation, the cloud orientation of individual cloud cluster units is analyzed, including: Retrieve historical independent cloud cluster data within a preset time period prior to the current moment; Based on historical independent cloud cluster data and AI time series models, the trajectory of cloud clusters is determined.

[0010] In one possible implementation, the current location of the cloud cluster is determined based on the cloud cluster's trajectory and Doppler radar data, including: Based on the acquisition time of the Doppler radar data, the current time, and the cloud cluster's trajectory, the current location of the cloud cluster is estimated.

[0011] In one possible implementation, auxiliary operation decision information is determined based on the current cloud cluster location, including: Based on Doppler radar reflectivity data, the atmospheric vertical temperature profile was retrieved. Determine the height of the zero-degree layer based on the vertical temperature profile of the atmosphere; The optimal catalytic operation height was determined based on the zero-degree layer altitude and the atmospheric vertical temperature profile. Based on the current cloud location and the optimal catalytic operation altitude, the rocket launch azimuth, elevation angle, and payload parameters are calculated as auxiliary operational decision-making information.

[0012] In one possible implementation, after determining the auxiliary operation decision information, the method further includes: Based on real-time radar reflectivity data after the operation, the evolution of cloud intensity and range is predicted in time series to obtain radar reflectivity forecast values. Obtain the measured radar reflectivity values ​​within the same time period, compare the predicted values ​​with the measured values, and calculate the prediction accuracy. Based on the forecast accuracy, the clustering parameters are iteratively optimized.

[0013] A second aspect of the present invention provides a meteorological-assisted operation intelligent decision-making device, comprising: The acquisition module is used to acquire Doppler radar data; The determination module is used to determine the spatial distribution information of cloud clusters based on Doppler radar data; The segmentation module is used to adaptively segment the cloud clusters based on the spatial distribution information and clustering algorithm to obtain independent cloud cluster units; wherein, the neighborhood radius and core point threshold of the clustering algorithm are adaptively adjusted according to the spatial distribution information of the cloud clusters. The analysis module is used to analyze the cloud movement of independent cloud units and determine the current cloud location based on the cloud movement and Doppler radar data. The decision-making module is used to determine auxiliary operation decision information based on the current cloud location.

[0014] A third aspect of the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the meteorological auxiliary operation intelligent decision-making method of the first aspect above.

[0015] A fourth aspect of the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the meteorological-assisted operation intelligent decision-making method of the first aspect above.

[0016] Compared to traditional technologies, this invention provides a method, device, and medium for intelligent decision-making in meteorological auxiliary operations. First, Doppler radar data is acquired. Then, based on the Doppler radar data, the spatial distribution information of cloud clusters is determined. Next, based on the spatial distribution information and a clustering algorithm, the cloud clusters are adaptively segmented to obtain independent cloud cluster units. The neighborhood radius and core point threshold of the clustering algorithm are adaptively adjusted according to the spatial distribution information of the cloud clusters. The cloud trajectory of the independent cloud cluster units is then analyzed, and the current cloud cluster position is determined based on the cloud trajectory and Doppler radar data. Finally, based on the current cloud cluster position, auxiliary operation decision information is determined. This invention achieves intelligent identification of cloud spatial distribution using Doppler radar data and adaptively and dynamically adjusts the neighborhood radius and core point threshold of the clustering algorithm according to the actual spatial characteristics of the cloud clusters. This enables accurate segmentation of cloud clusters at different scales, such as hail nuclei and rainbands, improving the accuracy of cloud cluster identification and ensuring the effectiveness of operations. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating the implementation of the intelligent decision-making method for meteorological auxiliary operations provided in this embodiment of the invention. Figure 2 This is a schematic diagram of the structure of the intelligent decision-making device for meteorological auxiliary operations provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the structure of the electronic device provided in an embodiment of the present invention. Detailed Implementation

[0018] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0019] Figure 1 This is a flowchart illustrating the implementation of the intelligent decision-making method for meteorological auxiliary operations provided in this embodiment of the invention. Figure 1 As shown, the intelligent decision-making method for meteorological-assisted operations includes: S110, acquire Doppler radar data; S120 determines the spatial distribution information of the cloud cluster based on Doppler radar data; S130, Based on the spatial distribution information of the cloud clusters and the clustering algorithm, the cloud clusters are adaptively segmented to obtain independent cloud cluster units; wherein, the neighborhood radius and core point threshold of the clustering algorithm are adaptively adjusted according to the spatial distribution information of the cloud clusters. S140: Analyze the cloud cluster trajectory of independent cloud cluster units, and determine the current cloud cluster position based on the cloud cluster trajectory and Doppler radar data; S150 determines auxiliary operation decision information based on the current cloud cluster location.

[0020] In this embodiment of the invention, Doppler radar data includes, but is not limited to, radar echo reflectivity factor, radial velocity, velocity spectral width, signal-to-noise ratio, and corresponding spatial location information (azimuth, elevation, slant range) and timestamp information. After acquiring the raw data, the data is preprocessed, including removing ground clutter, electromagnetic interference, noise points, and missing measurements; spatial interpolation is performed to complete sparse areas; and the radar data in polar coordinates is converted into unified planar rectangular coordinates or three-dimensional spatial coordinates to obtain a standardized Doppler radar dataset.

[0021] A combined radar reflectivity map is generated based on standardized Doppler radar data. The combined radar reflectivity map is then analyzed using an AI visual model to extract and determine the spatial distribution information of the cloud clusters. This includes the cloud clusters' coverage area, boundary outline, and center location in the horizontal direction; cloud top height, cloud bottom height, and vertical thickness stratification in the vertical direction; and spatial characteristics such as echo intensity gradient, density, and continuity within the cloud clusters. This allows for a direct reflection of the overall layout, aggregation state, and structural density differences of the cloud clusters within the monitored airspace.

[0022] In traditional weather modification operations, cloud identification and segmentation often rely on fixed threshold discrimination and classical meteorological algorithms. For example, cloud extraction is performed based on single or combined thresholds such as echo intensity, top height, and vertical integral liquid water content. These methods are poorly adaptable to meteorological conditions, seasonal changes, and cloud types, and are prone to problems such as missed detection of weak echoes, misjudgment of strong echoes, and hail nucleus inundation.

[0023] After obtaining the spatial distribution information of cloud clusters, this invention determines the initial clustering parameters of the clustering algorithm based on the spatial distribution information of cloud clusters. The clustering algorithm is used to pre-segment the cloud point cloud corresponding to the Doppler radar data to obtain the preliminary cloud cluster segmentation result. The preliminary cloud cluster segmentation result is compared with the spatial distribution information of cloud clusters, and the neighborhood radius and core point threshold of the clustering algorithm are adaptively adjusted according to the comparison result. The clustering segmentation is performed again using the adjusted clustering parameters, thereby dividing the contiguous or adjacent cloud cluster regions into multiple independent cloud cluster units with clear boundaries and clear physical meaning.

[0024] For each independent cloud cluster unit, historical independent cloud cluster data within a preset time period prior to the current moment is acquired. A large-scale AI time-series model is used to analyze the motion characteristics of the time-series cloud cluster data, determining the cloud cluster's direction of movement, speed, and evolution trend, thus completing the cloud cluster trajectory analysis. Combining the cloud cluster trajectory, the timestamp of the latest radar data, and the time difference with the current moment, an empirical orthogonal decomposition combined with a genetic algorithm is used to perform high-precision estimation of the cloud cluster's position. This corrects for positional deviations caused by radar data delays, cloud cluster movement, and deformation, ultimately obtaining the precise and stable spatial position of each independent cloud cluster unit at the current moment, including horizontal coordinates and altitude levels, enabling accurate positioning and tracking of individual cloud clusters.

[0025] Based on the current spatial location, distribution range, movement direction, and intensity characteristics of each independent cloud cluster, the atmospheric vertical temperature profile is retrieved using Doppler radar reflectivity data to determine the zero-degree layer altitude and the optimal catalytic operation altitude between -5°C and -15°C. Combining the optimal catalytic operation altitude with the precise location of the cloud clusters, the azimuth, elevation, and payload parameters for rocket launch are calculated to generate corresponding auxiliary operational decision-making information. This auxiliary operational decision-making information includes, but is not limited to: assessment of the meteorological risk level of the operational area, suitable operational time period suggestions, operational location recommendations, launch parameter suggestions, and instructions for artificial rain enhancement / hail suppression operations.

[0026] In some embodiments, cloud clusters are adaptively segmented based on their spatial distribution information and a clustering algorithm to obtain independent cloud cluster units. This includes: determining initial clustering parameters for the clustering algorithm based on the spatial distribution information of the cloud clusters; pre-segmenting the cloud point cloud corresponding to the Doppler radar data using the clustering algorithm to obtain preliminary cloud cluster segmentation results; comparing the preliminary cloud cluster segmentation results with the spatial distribution information of the cloud clusters; adaptively adjusting the neighborhood radius and core point threshold of the clustering algorithm based on the matching degree comparison results; and performing clustering segmentation again using the adjusted clustering parameters to obtain independent cloud cluster units.

[0027] Traditional cloud segmentation commonly uses storm cell identification algorithms such as SCIT and TITAN, as well as clustering or region growing methods with fixed parameters. The key parameters such as neighborhood radius, growth threshold, and minimum cell area are all preset global constants that cannot be adaptively adjusted according to cloud density, scale, and morphology. This often leads to segmentation defects such as broken rainbands, multiple cells sticking together, and the inability to separate small-scale hail nuclei.

[0028] In this embodiment of the invention, the initial clustering parameters of the clustering algorithm are determined based on the spatial distribution information of the cloud clusters. The spatial distribution information of the cloud clusters includes features such as the cloud cluster location range, reflectivity intensity distribution, point cloud spatial density, individual cloud size, and cloud cluster morphology, obtained from Doppler radar reflectivity data inversion. Based on these spatial distribution features, the system automatically generates an initial neighborhood radius and an initial core point threshold that are suitable for the current cloud cluster, avoiding the use of fixed global parameters and making the initial clustering closer to the actual cloud cluster structure.

[0029] Using the aforementioned initial clustering parameters, the cloud point cloud corresponding to the Doppler radar data is pre-segmented. Each effective reflectivity data point in the radar echo is abstracted as a data point in the spatial point cloud. The connectivity between points is determined based on the initial neighborhood radius, and high-density core regions are identified based on the initial core point threshold. Through clustering and aggregation, echo points that are spatially adjacent and have similar densities are grouped into the same cloud cluster region, thus obtaining a preliminary cloud cluster segmentation result. This preliminary result is only a coarse segmentation and may have defects such as individual cell adhesion, rainband breakage, hail nucleus submersion, and noise point contamination.

[0030] Using the real cloud boundary, location, scale, and individual distribution identified from the radar composite reflectivity map by the AI ​​vision big model as a benchmark, the cloud outline, coverage area, and number of individual units obtained from the initial segmentation are aligned with the benchmark information one by one. The overlap, boundary error, integrity error, and separation error are calculated to form a quantitative matching result, thereby judging whether there are problems such as over-segmentation, under-segmentation, boundary offset, or target submersion in the current segmentation.

[0031] Using adaptively adjusted and optimized clustering parameters, the cloud point cloud is clustered and segmented again. This effectively removes isolated noise points, separates interconnected cloud clusters, maintains the structural integrity of the rainband, and accurately highlights high-density targets such as hail nuclei, ultimately resulting in independent cloud cluster units with clear boundaries, independent structures, and conformity to the actual cloud cluster distribution.

[0032] For example, the AI ​​vision large-scale model adopts an encoder structure based on a combination of convolutional neural networks and Transformers. Through multi-layer feature extraction, contextual modeling, and multi-scale fusion, it achieves visual understanding of radar composite reflectivity images. The principle is as follows: the radar composite reflectivity image is used as input. A shallow network extracts low-level features such as echo edges, texture, and intensity gradients. Then, a deep network captures high-level semantic features such as the overall outline, spatial distribution, and density of cloud clusters. An attention mechanism is used to focus on the effective cloud cluster area and suppress clutter interference. Finally, the spatial distribution information of the cloud cluster, including its location, boundaries, extent, and internal structure, is output.

[0033] In some embodiments, the neighborhood radius and core point threshold of the clustering algorithm are adaptively adjusted based on the matching degree comparison results, including: determining the segmentation defect type and defect score based on the matching degree comparison results; and adjusting the neighborhood radius and core point threshold based on the segmentation defect type and defect score.

[0034] In this embodiment of the invention, the matching degree comparison is based on the real cloud spatial distribution information identified by the AI ​​visual large model. The preliminary cloud segmentation results obtained by clustering pre-segmentation are compared with this benchmark dimension by dimension to calculate a quantified matching degree index, including four core indicators: segmentation overlap, boundary error, region integrity, and target separation. Each indicator is weighted and summed according to preset weights to obtain the overall matching degree score. At the same time, based on the deviation of each indicator, the type of defect in the current segmentation is determined. The defect types are mainly divided into three categories, and each type of defect corresponds to a clear matching degree index characteristic: Oversegmentation defect: It manifests as a low matching score and a serious failure to meet the regional integrity index. In the initial segmentation results, a single complete cloud cluster (such as a continuous rain belt) is cut into multiple scattered fragments. The cloud clusters are broken and discontinuous. The core reason is that the clustering neighborhood radius is too small and the core point threshold is too high, resulting in a narrow clustering range. It is impossible to aggregate spatially adjacent and similarly dense cloud clusters into complete units. Undersegmentation defects: manifested as low matching scores, serious failure to meet target separation index, and multiple independent cloud clusters (such as hail nuclei and surrounding rain clouds) sticking together and unable to be separated in the preliminary segmentation results. Small-scale high-density targets (hail nuclei) are wrapped by large-scale cloud clusters and their features are submerged. The core reason is that the clustering neighborhood radius is too large and the core point threshold is too low, resulting in an excessively wide clustering range, which leads to the incorrect aggregation of cloud point clouds with different physical meanings. Boundary offset defect: It is characterized by a moderate matching score, a failure to meet the boundary error index, and a significant offset between the cloud cluster boundary and the real cloud cluster boundary in the preliminary segmentation result (towards the inside or outside). However, the integrity of individual cloud clusters and the separation of the target are basically qualified. The core reason is that the combination of cluster neighborhood radius and core point threshold is unreasonable, which makes it impossible for the segmentation boundary to accurately fit the real cloud cluster outline.

[0035] After determining the segmentation defect type, a corresponding defect score is calculated based on the deviation of each matching degree index. The defect score adopts a 1-10 scale, with a higher score indicating a more severe defect: the oversegmentation defect score is calculated based on the deviation rate of the region integrity, and the higher the deviation rate, the higher the score; the undersegmentation defect score is calculated based on the deviation rate of the target separation, and the higher the deviation rate, the higher the score; the boundary offset defect score is calculated based on the average value of the boundary error, and the higher the error, the higher the score. Finally, the overall defect score of the current segmentation is obtained, which serves as a quantitative basis for parameter adjustment.

[0036] Secondly, based on the segmentation defect type and defect score, the neighborhood radius (Eps) and core point threshold (MinPts) are precisely adjusted. The adjustment process follows the principles of "defect correspondence, score adaptation, and small-scale iteration" to ensure that the adjusted parameters can specifically address the corresponding segmentation defects. The specific adjustment rules are as follows: I. Addressing the defects of over-segmentation (fragmentation and insufficient integrity of cloud clusters).

[0037] Neighborhood radius (Eps) adjustment: Based on the oversegmentation defect score, Eps is increased by a preset ratio. The higher the defect score, the greater the increase ratio (e.g., for a score of 6-8, Eps increases by 10%-20%; for a score of 8-10, Eps increases by 20%-30%). This expands the spatial search range of clustering, enabling previously fragmented cloud clusters to reconnect and aggregate, restoring the integrity of the cloud clusters. Core Point Threshold (MinPts) Adjustment: MinPts is reduced synchronously according to a preset ratio to match the increase of Eps. The higher the defect score, the greater the reduction ratio (e.g., for a score of 6-8, MinPts is reduced by 5%-10%; for a score of 8-10, MinPts is reduced by 10%-15%). This reduces the density requirement of core points and avoids insufficient cloud aggregation due to too few core points, further solving the over-segmentation problem.

[0038] II. Addressing the defects of undersegmentation (cloud clustering, target submersion).

[0039] Neighborhood radius (Eps) adjustment: Based on the undersegmentation defect score, Eps is reduced by a preset ratio. The higher the defect score, the greater the reduction ratio (e.g., for a score of 6-8, Eps is reduced by 10%-20%; for a score of 8-10, Eps is reduced by 20%-30%). This narrows the spatial search range of clustering, limits the erroneous aggregation of point clouds from different cloud clusters, and enhances the ability to identify small-scale, high-density targets (such as hail nuclei). Core Point Threshold (MinPts) Adjustment: MinPts is increased synchronously according to a preset ratio to match the decrease in Eps. The higher the defect score, the greater the increase ratio (e.g., for a score of 6-8, MinPts increases by 5%-10%; for a score of 8-10, MinPts increases by 10%-15%). This increases the density requirements of core points, strengthens the identification of the core area of ​​high-density cloud clusters (hail nuclei), and enables accurate separation of adhering cloud clusters.

[0040] III. Addressing the boundary offset defect (segmentation boundary deviation).

[0041] Neighborhood radius (Eps) adjustment: Based on the boundary offset direction and defect score, Eps is adjusted iteratively in small increments. If the boundary is too inward (the segmented range is smaller than the actual cloud), Eps is increased slightly (by 5%-10% each time); if the boundary is too outward (the segmented range is larger than the actual cloud), Eps is decreased slightly (by 5%-10% each time). The higher the defect score, the larger the adjustment range can be in a single step, but large adjustments should be avoided to prevent oversegmentation or undersegmentation. Core Point Threshold (MinPts) Adjustment: In conjunction with the slight adjustment of Eps, MinPts is fine-tuned simultaneously. If Eps increases, MinPts is slightly decreased (by 3%-5% each time); if Eps decreases, MinPts is slightly increased (by 3%-5% each time). This ensures that Eps and MinPts are properly matched, so that the segmentation boundary gradually approaches the actual cloud boundary, reducing boundary error.

[0042] After the parameters are adjusted, the adjusted neighborhood radius and core point threshold are substituted back into the clustering algorithm, and cloud segmentation is performed again to complete one adaptive adjustment. If the matching degree between the result of the second segmentation and the spatial distribution information of the cloud still does not reach the preset threshold, the above process is repeated until the segmentation result meets the requirements, and finally independent cloud units with clear boundaries, no mutual adhesion, and complete structure are obtained.

[0043] In some embodiments, determining the spatial distribution information of the cloud cluster based on Doppler radar data includes: generating a radar composite reflectivity map based on the Doppler radar data; and performing visual feature analysis on the radar composite reflectivity map using an AI visual large model to obtain the spatial distribution information of the cloud cluster.

[0044] In this embodiment of the invention, during real-time detection, the Doppler radar outputs raw data including radar echo reflectivity factor, radial motion velocity, velocity spectrum width, signal-to-noise ratio, and corresponding spatial location information (including azimuth, elevation, and slant range) and timestamp.

[0045] After acquiring the raw radar data, the system first preprocesses the data, including removing ground clutter, electromagnetic interference, isolated noise points and missing measurements, and performing spatial interpolation to complete sparse areas; at the same time, it converts the polar coordinate format radar data into a unified plane rectangular coordinate or three-dimensional spatial coordinate to obtain a standardized Doppler radar dataset.

[0046] Based on this, the system vertically integrates and superimposes Doppler radar reflectivity data from multiple elevation angles to generate a combined radar reflectivity map. This combined reflectivity map can comprehensively reflect the echo intensity distribution of cloud clusters at different altitudes, clearly presenting the overall morphology, vertical structure, intensity gradient, and spatial aggregation characteristics of the cloud clusters.

[0047] In some embodiments, analyzing the cloud cluster trajectory of independent cloud cluster units includes: acquiring historical independent cloud cluster data within a preset time period before the current moment; and determining the cloud cluster trajectory based on the historical independent cloud cluster data and the AI ​​time series model.

[0048] In terms of cloud movement and position estimation, traditional methods often use simple time-series extrapolation methods such as linear extrapolation and centroid translation. They rely on a single frame or a small amount of historical data to calculate the translation vector, which makes it difficult to capture complex evolution patterns such as nonlinear deformation, acceleration, and turning. Furthermore, they do not consider the position deviation caused by radar data delay, resulting in a significant misalignment between the target point and the actual cloud position.

[0049] In this embodiment of the invention, historical independent cloud cluster data within a preset time period prior to the current moment is obtained. This data is standardized data that has been clustered, segmented, and calibrated in the previous stage. Specifically, it includes the spatial location (horizontal coordinates, vertical height), echo intensity, morphological features (such as outline and density) of the historical independent cloud clusters, as well as the timestamp information of the corresponding time, to ensure the temporal continuity and integrity of the data. Moreover, all historical data corresponds one-to-one with the cloud cluster unit to be analyzed at present, avoiding data interference across cloud clusters and regions.

[0050] The aforementioned historical independent cloud cluster data were analyzed in depth using a pre-trained AI time series model. This AI time series model is based on the Transformer architecture and integrates a time series attention mechanism and a multi-scale feature extraction module. After being trained on a large number of cloud cluster time series change samples (including cloud cluster movement and evolution samples under different weather conditions), it has the ability to capture the time series evolution patterns of cloud clusters and can effectively mine the motion features in historical cloud cluster data.

[0051] Historical monitoring data of the same independent cloud cluster within a preset time period are arranged chronologically to construct a continuous temporal feature sequence, which is then input into an AI temporal model. The model first performs normalization preprocessing on the temporal sequence, identifies and removes abnormal data such as sudden changes in position, intensity jumps, and false contours caused by radar interference and data anomalies through motion consistency verification. At the same time, it performs spatiotemporal interpolation to complete the cloud cluster's coordinates, range, intensity, and other features at missing moments, ensuring that the temporal data is continuous, smooth, and conforms to the physical motion laws of the cloud cluster. Based on this, the model is built on a deep network structure with a temporal attention mechanism, extracting temporal features such as the cloud cluster's center position shift, movement rate changes, coverage expansion or contraction, echo intensity increase or decrease, and morphological evolution at different time scales. Through attention weights, it automatically highlights stable and continuous movement trends, suppresses non-dominant random changes such as short-term gusts, local disturbances, and noise fluctuations, and fits the overall movement vector of the cloud cluster by combining continuous movement trajectories from multiple historical frames. Finally, by integrating multi-dimensional information such as movement direction, movement speed, and development trend, the model accurately identifies the overall movement trend of the cloud cluster.

[0052] The AI ​​time series model outputs core information about the cloud's trajectory, including its direction of movement, speed of movement, and short-term future evolution trends (such as whether it will expand or contract, or maintain its original trajectory), thus determining the cloud's trajectory.

[0053] In some embodiments, determining the current location of a cloud cluster based on its trajectory and Doppler radar data includes estimating the current location of the cloud cluster based on the acquisition time of the currently acquired Doppler radar data, the current time, and the trajectory of the cloud cluster.

[0054] In this embodiment of the invention, since Doppler radar data has a fixed observation time interval, the cloud position corresponding to the latest frame of radar data is only the position at the historical observation time and cannot directly represent the current real position. Therefore, it is necessary to dynamically estimate and calibrate the cloud position in conjunction with the cloud's trajectory. First, the observation acquisition timestamp of the latest frame of Doppler radar data is recorded, and the current actual time of the system is obtained. The time difference between the two is calculated, which is the duration of the cloud's movement from the last radar observation to the current moment. Subsequently, based on the cloud trajectory information obtained from the previous AI time series model, including the cloud's stable movement direction, movement speed, and movement trend, the cloud center coordinates and height level at the radar observation time are used as the initial position reference. Extrapolation is performed according to the time difference and movement speed to obtain the theoretical position of the cloud at the current moment. On this basis, the radial movement speed, wind shear characteristics, and the cloud's own deformation trend contained in the radar data are combined to correct the preliminary estimated position, eliminating the positioning deviation caused by local airflow disturbances and changes in cloud morphology. Finally, the accurate and stable spatial position of the cloud at the current moment is obtained, including the horizontal coordinates, vertical height, and core area range.

[0055] Because Doppler radar data has a fixed update interval of 6 minutes, the latest radar data can only reflect the cloud position at historical observation times and cannot directly represent the current true position. Therefore, this invention introduces a high-precision real-time position estimation algorithm. The system first extracts the time series data of independent cloud clusters from the three time periods immediately preceding the current time. It then uses an AI time series model to fully analyze the cloud cluster's movement direction, speed, intensity evolution, and morphological change trends. Next, it uses an empirical orthogonal decomposition combined with a genetic algorithm, taking the difference between the latest radar data acquisition time and the current system time as the time variable. Combined with the cloud cluster's movement trajectory and deformation patterns, it performs high-precision extrapolation calculations, effectively correcting positioning errors caused by radar data delays, nonlinear cloud cluster movement, atmospheric turbulence disturbances, and other factors.

[0056] For example, based on the current time, data from the independent cloud clusters that have been segmented in the three adjacent time periods are extracted. The key features of each cloud cluster at different times, such as horizontal coordinates, echo intensity, centroid height, and coverage, are used to construct a standardized spatiotemporal data matrix according to time series and spatial grid. At the same time, the radar volume scan timestamp, the time difference between the data acquisition time and the current system time are included to form a complete modeling input dataset.

[0057] The aforementioned spatiotemporal data matrix is ​​subjected to empirical orthogonal decomposition, which decomposes the complex nonlinearly changing spatiotemporal field of cloud clusters into mutually orthogonal spatial feature vectors and time coefficient sequences. By selecting the principal components with the highest contribution rate through eigenvalue screening, random signals such as atmospheric disturbances and noise interference are removed, and stable dominant spatiotemporal features such as cloud cluster movement, expansion, contraction, and intensity evolution are extracted to obtain the core trend mode of cloud cluster motion, thereby achieving dimensionality reduction analysis of the nonlinear deformation and motion law of cloud clusters.

[0058] Based on the main spatiotemporal features extracted by empirical orthogonal decomposition and combined with the physical motion constraints of cloud clusters, a fitness function is constructed with the objectives of minimizing position prediction error, optimizing trajectory smoothness, and maximizing intensity evolution consistency. The parameters to be optimized, such as cloud cluster movement speed, direction correction coefficient, deformation coefficient, and time extrapolation weight, are used as decision variables for the genetic algorithm to establish an optimization model for high-precision estimation of cloud cluster real-time position.

[0059] The initialization process involves a population of several individuals, each corresponding to a set of cloud location estimation parameters. The population is iteratively evolved through selection, crossover, and mutation operations. The fitness value of each individual is calculated generation by generation. High-fitness, high-quality individuals are retained, while low-fitness individuals are eliminated. After multiple rounds of iteration, the population converges to the globally optimal parameter combination. The optimal spatiotemporal feature weights and motion correction coefficients are automatically matched to avoid the local optima defect of traditional linear extrapolation.

[0060] The optimal parameters obtained by the genetic algorithm are substituted into the spatiotemporal evolution model of the cloud cluster reconstructed by empirical orthogonal decomposition. The latest radar data time difference is combined for extrapolation calculation, and the horizontal position, vertical height and core area of ​​the cloud cluster are corrected simultaneously. Finally, the true and accurate spatial position of the cloud cluster at the current moment is output, eliminating the positioning deviation caused by the 6-minute radar data delay, nonlinear motion of the cloud cluster and atmospheric disturbance.

[0061] In some embodiments, auxiliary operational decision information is determined based on the current cloud location, including: retrieving the atmospheric vertical temperature profile based on Doppler radar reflectivity data; determining the zero-degree layer height based on the atmospheric vertical temperature profile; determining the optimal catalytic operation height based on the zero-degree layer height and the atmospheric vertical temperature profile; and calculating the rocket launch azimuth, elevation angle, and payload parameters as auxiliary operational decision information based on the current cloud location and the optimal catalytic operation height.

[0062] In this embodiment of the invention, based on Doppler radar reflectivity data and its vertical layering structure, and combined with the correspondence between radar echo intensity and atmospheric temperature and humidity, a physical inversion model is used to invert and calculate the atmospheric vertical temperature profile within the monitoring area. This yields the atmospheric temperature distribution at different altitudes from near the ground to the cloud top, forming a continuous and refined atmospheric vertical temperature profile. Subsequently, based on the inverted atmospheric vertical temperature profile, the critical altitude at which the atmospheric temperature transitions from positive to negative is retrieved and determined as the zero-degree layer altitude. After obtaining the zero-degree layer altitude, the optimal catalytic operation altitude is further determined by combining the temperature gradient of each altitude layer in the atmospheric vertical temperature profile, the supercooled water distribution range, and the optimal activation temperature range of the catalyst. This altitude is typically located below the zero-degree layer, in a range where supercooled water content is relatively abundant and conducive to catalyst diffusion, thus maximizing the catalytic effect of artificial rain enhancement or hail suppression operations. Finally, using the currently accurately estimated spatial location of the cloud cluster as the operational target, and combining the optimal catalytic operation height, launch point geographical coordinates, ballistic constraints, cloud cluster coverage, echo intensity level, and other factors, the launch azimuth, launch elevation, and appropriate launch quantity of the rocket are calculated through conventional ballistic calculation models and operational parameter optimization models.

[0063] In some embodiments, after determining the auxiliary operation decision information, the method further includes: performing time-series prediction of cloud intensity and range evolution based on real-time radar reflectivity data after the operation to obtain radar reflectivity forecast values; obtaining measured radar reflectivity values ​​within the same time period, comparing the forecast values ​​with the measured values, and calculating the forecast accuracy; and iteratively optimizing clustering parameters based on the forecast accuracy.

[0064] In the decision-making stage of operational parameters, traditional methods rely heavily on radiosonde observations to obtain temperature profiles and zero-degree layer heights. However, radiosonde data has a low update frequency, sparse spatial resolution, and poor timeliness, making it impossible to match with radar data in real time. This results in inaccurate calculation of the optimal catalytic height and large deviations in rocket launch azimuth and elevation angle, directly affecting the catalytic effect and operational efficiency.

[0065] In this embodiment of the invention, based on the real-time Doppler radar reflectivity data after the operation is completed, combined with the existing cloud cluster temporal evolution characteristics and movement patterns, the AI ​​temporal prediction model is used to continuously extrapolate and predict the changes in echo intensity, horizontal range extension, vertical structure development and overall movement trend of the cloud cluster, so as to obtain the radar reflectivity forecast values ​​at different times within a preset future period.

[0066] The actual radar reflectivity values ​​obtained from radar detection are continuously acquired within the corresponding time period. The predicted values ​​and the actual values ​​at the same time and spatial location are compared point by point and region by region. The errors between the two in terms of intensity deviation, position deviation, and range overlap are calculated. The quantitative forecast accuracy is obtained by combining various error indicators.

[0067] The model is iteratively optimized based on the calculated forecast accuracy. When the forecast accuracy is lower than the preset threshold, it is determined that there is a deviation in the cloud cluster segmentation, direction prediction or location estimation. Then, the source is traced back and the neighborhood radius and core point threshold in the clustering algorithm are adaptively adjusted. At the same time, the parameters related to cloud cluster motion modeling are optimized to make subsequent cloud cluster segmentation more accurate and direction prediction more reliable.

[0068] Figure 2 This is a schematic diagram of the structure of the intelligent decision-making device for meteorological auxiliary operations provided in an embodiment of the present invention. Figure 2 As shown, the intelligent decision-making device for meteorological auxiliary operations includes: Acquisition module 210 is used to acquire Doppler radar data; The determination module 220 is used to determine the spatial distribution information of the cloud based on Doppler radar data; The segmentation module 230 is used to adaptively segment the cloud clusters based on the spatial distribution information and clustering algorithm to obtain independent cloud cluster units; wherein, the neighborhood radius and core point threshold of the clustering algorithm are adaptively adjusted according to the spatial distribution information of the cloud clusters. Analysis module 240 is used to analyze the cloud movement of independent cloud units and determine the current cloud position based on the cloud movement and Doppler radar data. The decision module 250 is used to determine auxiliary operation decision information based on the current cloud cluster location.

[0069] Optionally, the segmentation module 230 is used for: Based on the spatial distribution information of cloud clusters, the initial clustering parameters of the clustering algorithm are determined; Clustering algorithms were used to pre-segment the cloud clusters and point clouds corresponding to Doppler radar data to obtain preliminary cloud cluster segmentation results; The preliminary cloud segmentation results are compared with the spatial distribution information of the cloud clusters to determine their matching degree. Based on the matching results, the neighborhood radius and core point threshold of the clustering algorithm are adaptively adjusted. The clustering parameters were adjusted and the clustering was performed again to obtain independent cloud units.

[0070] Optionally, the segmentation module 230 is used for: Based on the matching results, the segmentation defect type and defect score are determined; Adjust the neighborhood radius and core point threshold based on the segmentation defect type and defect score.

[0071] Optionally, module 220 is used for: Generating radar composite reflectivity maps based on Doppler radar data; By using an AI-powered visual model to analyze the visual features of the radar composite reflectivity map, spatial distribution information of the cloud clusters can be obtained.

[0072] Optional, analysis module 240, used for: Retrieve historical independent cloud cluster data within a preset time period prior to the current moment; Based on historical independent cloud cluster data and AI time series models, the trajectory of cloud clusters is determined.

[0073] Optional, analysis module 240, used for: Based on the acquisition time of the Doppler radar data, the current time, and the cloud cluster's trajectory, the current location of the cloud cluster is estimated.

[0074] Optional, decision module 250, used for: Based on Doppler radar reflectivity data, the atmospheric vertical temperature profile was retrieved. Determine the height of the zero-degree layer based on the vertical temperature profile of the atmosphere; The optimal catalytic operation height was determined based on the zero-degree layer altitude and the atmospheric vertical temperature profile. Based on the current cloud location and the optimal catalytic operation altitude, the rocket launch azimuth, elevation angle, and payload parameters are calculated as auxiliary operational decision-making information.

[0075] Optionally, the meteorological-assisted intelligent decision-making device also includes an optimization module for: Based on real-time radar reflectivity data after the operation, the evolution of cloud intensity and range is predicted in time series to obtain radar reflectivity forecast values. Obtain the measured radar reflectivity values ​​within the same time period, compare the predicted values ​​with the measured values, and calculate the prediction accuracy. Based on the forecast accuracy, the clustering parameters are iteratively optimized.

[0076] Figure 3 This is a schematic diagram of the structure of the electronic device provided in an embodiment of the present invention. For example... Figure 3 As shown, the electronic device 3 of this embodiment includes a memory 31, a processor 30, and a computer program 32 stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the meteorological auxiliary operation intelligent decision-making method as described in the above embodiment.

[0077] The computer-readable storage medium of this invention stores a computer program, which, when executed by a processor, implements the steps of the meteorological-assisted operation intelligent decision-making method as described in the above embodiments.

[0078] In the above embodiments, the descriptions of each embodiment have their own emphasis. Parts not detailed or described in a particular embodiment can be referred to in the relevant descriptions of other embodiments. Unless otherwise specified or in conflict with logic, the terminology and / or descriptions between different embodiments are consistent and can be referenced interchangeably. Technical features in different embodiments can be combined to form new embodiments based on their inherent logical relationships.

[0079] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A meteorological-assisted operation intelligent decision-making method, characterized in that, include: Acquire Doppler radar data; Based on the Doppler radar data, the spatial distribution information of the cloud cluster is determined; Based on the spatial distribution information of the cloud clusters and the clustering algorithm, the cloud clusters are adaptively segmented to obtain independent cloud cluster units; wherein, the neighborhood radius and core point threshold of the clustering algorithm are adaptively adjusted according to the spatial distribution information of the cloud clusters. Analyze the cloud trajectory of independent cloud clusters, and determine the current cloud cluster location based on the cloud trajectory and Doppler radar data; Based on the current cloud cluster location, auxiliary operation decision information is determined.

2. The intelligent decision-making method for meteorological-assisted operations according to claim 1, characterized in that, Based on the spatial distribution information and clustering algorithm of the cloud clusters, the cloud clusters are adaptively segmented to obtain independent cloud cluster units, including: Based on the spatial distribution information of cloud clusters, the initial clustering parameters of the clustering algorithm are determined; Clustering algorithms were used to pre-segment the cloud clusters and point clouds corresponding to Doppler radar data to obtain preliminary cloud cluster segmentation results; The preliminary cloud segmentation results are compared with the spatial distribution information of the cloud clusters to determine their matching degree. Based on the matching results, the neighborhood radius and core point threshold of the clustering algorithm are adaptively adjusted. The clustering parameters were adjusted and the clustering was performed again to obtain independent cloud units.

3. The intelligent decision-making method for meteorological-assisted operations according to claim 2, characterized in that, Based on the matching results, the neighborhood radius and core point threshold of the clustering algorithm are adaptively adjusted, including: Based on the matching results, the segmentation defect type and defect score are determined; Adjust the neighborhood radius and core point threshold based on the segmentation defect type and the defect score.

4. The intelligent decision-making method for meteorological-assisted operations according to claim 1, characterized in that, Based on the Doppler radar data, the spatial distribution information of the cloud cluster is determined, including: A combined radar reflectivity map is generated based on the Doppler radar data; The spatial distribution information of the cloud clusters is obtained by performing visual feature analysis on the radar composite reflectivity map using an AI visual big data model.

5. The intelligent decision-making method for meteorological-assisted operations according to claim 4, characterized in that, Analysis of the cloud trajectory of independent cloud cluster units includes: Retrieve historical independent cloud cluster data within a preset time period prior to the current moment; Based on the historical independent cloud cluster data and the AI ​​time series model, the cloud cluster trajectory is determined.

6. The intelligent decision-making method for meteorological-assisted operations according to claim 5, characterized in that, Based on the cloud cluster's trajectory and Doppler radar data, determine the current location of the cloud cluster, including: Based on the acquisition time of the Doppler radar data, the current time, and the direction of the cloud cluster, the current location of the cloud cluster is estimated.

7. The intelligent decision-making method for meteorological-assisted operations according to claim 1, characterized in that, Based on the current cloud cluster location, auxiliary operation decision information is determined, including: Based on Doppler radar reflectivity data, the atmospheric vertical temperature profile was retrieved. The height of the zero-degree layer is determined based on the aforementioned atmospheric vertical temperature profile; The optimal catalytic operation height is determined based on the zero-degree layer altitude and the atmospheric vertical temperature profile. Based on the current cloud location and the optimal catalytic operation altitude, the rocket launch azimuth, elevation angle, and payload parameters are calculated as auxiliary operational decision-making information.

8. The intelligent decision-making method for meteorological-assisted operations according to claim 1, characterized in that, After determining the auxiliary operation decision information, the method further includes: Based on real-time radar reflectivity data after the operation, the evolution of cloud intensity and range is predicted in time series to obtain radar reflectivity forecast values. Obtain the measured radar reflectivity values ​​within the same time period, compare the predicted values ​​with the measured values, and calculate the prediction accuracy. Based on the forecast accuracy, the clustering parameters are iteratively optimized.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the intelligent decision-making method for meteorological auxiliary operations as described in any one of claims 1 to 8.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the steps of the intelligent decision-making method for meteorological auxiliary operations as described in any one of claims 1 to 8.

Images