Convective cloud rain reduction operation analysis method and system based on deep learning

Abstract

The present application provides a deep learning-based convective cloud rain reduction operation analysis method and system, which is used to obtain a comprehensive convective cloud radar echo extrapolation data set, perform innovative data processing, accurately generate extrapolation results and formulate reasonable operation strategies, effectively solving the problems of insufficient data utilization and unscientific operation strategy formulation in existing technologies in convective cloud rain reduction operations, and significantly improving the accuracy and effectiveness of operations. The method includes: sparse sampling and spatiotemporal alignment processing of the convective cloud radar echo extrapolation data set in the target area to generate a spatiotemporally continuous radar echo extrapolation reconstruction data set; calling a spatiotemporal sequence prediction model to perform cyclic radar echo extrapolation processing on the radar echo extrapolation reconstruction data set to generate extrapolation results within a preset time range; generating a convective cloud rain reduction operation strategy based on the extrapolation results, and sending the convective cloud rain reduction operation strategy to the rain reduction operation service system for operation processing.

Description

Technical Field

[0001] The present application belongs to the field of meteorological processing technology, and specifically relates to a method and system for analyzing convective cloud rain reduction operations based on deep learning. Background Art

[0002] In the field of meteorology, convective cloud rain reduction operations are of great significance for ensuring the smooth progress of major events and reducing losses from natural disasters. With the development of meteorological science and technology, people's requirements for the accuracy and effectiveness of rain reduction operations are increasing. Traditional convective cloud rain reduction operations often lack comprehensive and accurate data analysis of convective clouds. Specifically, previous operational decisions have relied heavily on experience and simple observational data, failing to fully utilize the complex and changing convective cloud information. In terms of data processing, it is difficult to effectively integrate information from multiple time steps of convective cloud radar echo data, and the data consistency is poor, resulting in inaccurate judgments on convective cloud development trends. In terms of operational strategy formulation, it is impossible to accurately determine the location of the operation point and the timing of the operation, and there is a lack of scientific consideration of the characteristics of the operation status assessment. Summary of the Invention

[0003] This application provides a deep learning-based convective cloud rain reduction operation analysis method and system, which is used to obtain a comprehensive convective cloud radar echo extrapolation data set, perform innovative data processing, accurately generate extrapolation results and formulate reasonable operation strategies. It effectively solves the problems of insufficient data utilization and unscientific operation strategy formulation in convective cloud rain reduction operations in existing technologies, and significantly improves the accuracy and effectiveness of operations.

[0004] In a first aspect, an embodiment of the present application provides a deep learning-based convective cloud rain reduction operation analysis method, which is applied to a convective cloud rain reduction operation analysis system. The method includes: obtaining a convective cloud radar echo extrapolation dataset in a target area, the convective cloud radar echo extrapolation dataset including radar echo intensity distribution data of multiple time steps and corresponding convective cloud spatial distribution data; performing sparse sampling and spatiotemporal alignment processing on the convective cloud radar echo extrapolation dataset to generate a spatiotemporally continuous radar echo extrapolation reconstruction dataset; calling a spatiotemporal sequence prediction model to perform cyclic radar echo extrapolation processing on the radar echo extrapolation reconstruction dataset to generate an extrapolation result within a preset time range, the extrapolation result including the spatial evolution path and intensity change trend of the convective cloud; generating a convective cloud rain reduction operation strategy based on the extrapolation result, and sending the convective cloud rain reduction operation strategy to a rain reduction operation service system for operation processing, the convective cloud rain reduction operation strategy including operation point location, operation timing, and operation status evaluation characteristics.

[0005] In a second aspect, an embodiment of the present application provides a convective cloud rain reduction operation analysis system, which includes a processor and a memory, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor performs the steps of the above method.

[0006] In a third aspect, an embodiment of the present application provides a computer-readable storage medium, which includes a computer program. When the computer program is run on a convective cloud rain reduction operation analysis system, the computer program is used to enable the convective cloud rain reduction operation analysis system to perform the steps of the above method.

[0007] During the implementation of this application, on the one hand, by sparse sampling and spatiotemporal alignment processing of the convective cloud radar echo extrapolation dataset in the target area, the data can be effectively integrated to generate a spatiotemporally continuous radar echo extrapolation reconstruction dataset, which greatly improves the consistency and availability of the data, and enables the radar echo extrapolation reconstruction dataset to more accurately reflect the actual situation of convective clouds; on the other hand, calling the spatiotemporal sequence prediction model to perform cyclic radar echo extrapolation processing can deeply explore the potential laws in the data, and accurately generate extrapolation results containing the spatial evolution path and intensity change trend of convective clouds within a preset time range, providing a strong basis for subsequent decision-making; on the other hand, the convective cloud rain reduction operation strategy generated based on the extrapolation results can comprehensively consider the operation point location, operation timing and operation status evaluation characteristics, making the rain reduction operation more targeted and scientific, and sending the convective cloud rain reduction operation strategy to the rain reduction operation service system for operation processing, which can effectively optimize the operation process, improve the accuracy and efficiency of rain reduction operations, and provide reliable support for meteorological operations.

[0008] In summary, the embodiments of the present application obtain a comprehensive convective cloud radar echo extrapolation data set, perform innovative data processing, accurately generate extrapolation results and formulate reasonable operation strategies, effectively solving the problems of insufficient data utilization and unscientific operation strategy formulation in convective cloud rain reduction operations in the existing technology, and significantly improving the accuracy and effectiveness of the operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] Figure 1 A flowchart of a deep learning-based convective cloud rain reduction operation analysis method provided in an embodiment of the present application.

[0010] Figure 2 A structural diagram of a convective cloud rain reduction operation analysis system provided in an embodiment of the present application. DETAILED DESCRIPTION

[0011] To make the purpose, technical solutions, and advantages of the embodiments of this application more clear, the technical solutions of this application will be clearly and completely described below in conjunction with the drawings in the embodiments of this application. Obviously, the described embodiments are part of the embodiments of the technical solutions of this application, but not all of them. Based on the embodiments described in this application document, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the technical solutions of this application.

[0012] See also Figure 1 , which is a deep learning-based convective cloud rain reduction operation analysis method provided in an embodiment of the present application. This method can be applied to a convective cloud rain reduction operation analysis system. The specific process is as shown in steps 110 to 140.

[0013] Step 110: Acquire a convective cloud radar echo extrapolation dataset of the target area, wherein the convective cloud radar echo extrapolation dataset includes radar echo intensity distribution data of multiple time steps and corresponding convective cloud spatial distribution data.

[0014] In the embodiments of the present application, a meteorological monitoring area is used as an example. Multiple radar monitoring stations are set up in the area, and these radar monitoring stations continuously monitor convective clouds. Over a period of time, the radar monitoring stations collect data at set intervals, for example, once every minute. In each collected data, the radar echo intensity distribution data is presented in the form of a two-dimensional matrix. Each element in the matrix represents the radar echo intensity value at a different spatial location. This intensity value reflects the degree of reflection of the radar wave by the precipitation particles in the convective cloud.

[0015] Convective cloud spatial distribution data combines radar station location information with meteorological satellite data to determine the specific spatial extent and morphology of convective clouds. This data is also represented by a data structure, such as a set of polygon vertex coordinates. After sequentially collecting data for multiple time steps, a convective cloud radar echo extrapolation dataset is formed, which contains radar echo intensity distribution data for multiple time steps and the corresponding convective cloud spatial distribution data. This data collected from various radar stations is then aggregated and organized to ensure its accuracy and completeness.

[0016] Step 120: performing sparse sampling and spatiotemporal alignment processing on the convective cloud radar echo extrapolation dataset to generate a spatiotemporally continuous radar echo extrapolation reconstruction dataset.

[0017] In an embodiment of the present application, the acquired convective cloud radar echo extrapolation dataset is further subjected to sparse sampling and spatiotemporal alignment to generate a spatiotemporally continuous radar echo extrapolation and reconstruction dataset. As an optional embodiment, the sparse sampling and spatiotemporal alignment processing of the convective cloud radar echo extrapolation dataset to generate a spatiotemporally continuous radar echo extrapolation and reconstruction dataset includes:

[0018] Step 121: performing time alignment processing on the radar echo intensity distribution data of multiple time steps in the convective cloud radar echo extrapolation dataset according to a preset time resolution threshold, to obtain a radar echo intensity distribution sequence with uniform time intervals.

[0019] In this meteorological monitoring scenario, the preset time resolution threshold is determined based on actual needs and data analysis accuracy, for example, it is set to five minutes. When processing radar echo intensity distribution data of multiple time steps, each time step data in the data set is traversed. For cases where the time interval does not conform to five minutes, a time interpolation algorithm, such as a linear interpolation algorithm, is used. For example, the data collection time intervals of two adjacent time steps are four minutes and six minutes. The radar echo intensity distribution data at the five-minute time point is calculated by linear interpolation, that is, the intensity values ​​of the two adjacent time steps are weighted according to the time ratio to obtain the intensity value at the middle time point. Similarly, the data of all time steps are processed, and finally a radar echo intensity distribution sequence with a uniform time interval of five minutes is obtained. This sequence provides a unified time scale for subsequent analysis, allowing data from different time steps to be compared and processed within the same time frame.

[0020] Step 122: performing spatial interpolation processing on each radar echo intensity distribution data in the radar echo intensity distribution sequence with uniform time intervals, eliminating data missing areas in radar detection blind spots, and obtaining spatially continuous radar echo intensity distribution data.

[0021] In the embodiment of the present application, due to the existence of blind spots in radar monitoring, there will be some areas of data missing in the data set. For each data in the radar echo intensity distribution sequence with uniform time intervals, that is, the two-dimensional radar echo intensity distribution matrix of each time step. For the data missing areas in the matrix, a spatial interpolation algorithm, such as the Kriging interpolation algorithm, is used. This algorithm takes into account the spatial correlation of the data and performs a comprehensive calculation based on factors such as the distance and direction of the surrounding known data points. For example, for a data missing point, known data points within a certain range around it are found. Different weights are assigned to these data points based on their distance from the missing point, with the closer the distance, the greater the weight. The radar echo intensity value of the missing point is then calculated by weighted averaging. This processing is performed on all missing areas in the entire matrix, thereby eliminating the data missing areas within the radar detection blind spots and obtaining spatially continuous radar echo intensity distribution data. This allows the radar echo intensity distribution of convective clouds in space to be fully presented, providing more accurate spatial information for subsequent analysis.

[0022] Step 123: determining a spatial sampling interval based on a moving speed threshold of the convective cloud, performing sparse sampling on the spatially continuous radar echo intensity distribution data according to the spatial sampling interval, and generating sparsely sampled radar echo intensity distribution data.

[0023] In the meteorological scenario of the embodiment of the present application, first, by analyzing the historical convective cloud data and the currently monitored convective cloud characteristics, a convective cloud movement speed threshold is determined, for example, set to 50 kilometers per hour. According to the movement speed threshold, combined with the spatial scale and analysis accuracy requirements, the spatial sampling interval is calculated. For example, the spatial sampling interval is calculated to be 1 kilometer. Then, the spatially continuous radar echo intensity distribution data is processed, and data points are selected in the two-dimensional radar echo intensity distribution matrix according to the spatial sampling interval of 1 kilometer. That is, the intensity value of a position is selected every 1 kilometer, and the data points at other positions are discarded to generate sparsely sampled radar echo intensity distribution data. This sparse sampling not only reduces the amount of data and reduces the computational complexity of subsequent processing, but also retains the key information of the convective cloud radar echo intensity distribution, so that the data can be more efficiently analyzed in subsequent analysis.

[0024] Step 124: performing spatial superposition processing on the sparsely sampled radar echo intensity distribution data and the convective cloud spatial distribution data to generate a temporally and spatially continuous radar echo extrapolation and reconstruction data set.

[0025] In this meteorological monitoring area scenario, the sparsely sampled radar echo intensity distribution data, a two-dimensional matrix containing intensity values ​​at specific spatial locations, is spatially superimposed with the convective cloud spatial distribution data, such as the convective cloud spatial range data represented by a set of polygon vertex coordinates. This is done by matching and corresponding each position in the radar echo intensity distribution data with the spatial range in the convective cloud spatial distribution data. For example, for a position in the radar echo intensity distribution matrix, if the position is within the polygonal range of the convective cloud spatial distribution, the intensity value of the position is associated with the convective cloud spatial information. By performing such processing on the entire data, a spatiotemporally continuous radar echo extrapolation reconstruction dataset is generated. This dataset integrates data that is evenly spaced in time, continuous in space, and superimposed with convective cloud spatial information, providing a comprehensive and accurate data foundation for subsequent convective cloud analysis.

[0026] Step 130: Calling the spatiotemporal sequence prediction model to perform cyclic radar echo extrapolation processing on the radar echo extrapolation reconstruction dataset to generate an extrapolation result within a preset time range, wherein the extrapolation result includes the spatial evolution path and intensity change trend of convective clouds.

[0027] In an embodiment of the present application, a trained spatiotemporal sequence prediction model is used to operate on a processed spatiotemporal continuous radar echo extrapolation and reconstruction dataset. The model learns and analyzes the characteristics of the time and space dimensions in the dataset, and predicts the convective cloud radar echo conditions within a preset time range in the future, thereby obtaining an extrapolation result that includes the spatial evolution path and intensity change trend of the convective cloud.

[0028] As another optional embodiment, calling the spatiotemporal sequence prediction model to perform cyclic radar echo extrapolation processing on the radar echo extrapolation reconstruction dataset to generate an extrapolation result within a preset time range includes:

[0029] Step 131: inputting the spatiotemporally continuous radar echo extrapolation and reconstruction data set into the encoder module of the spatiotemporal sequence prediction model to generate spatiotemporal coding features of radar echo intensity distribution data.

[0030] In this meteorological monitoring scenario, a spatiotemporally continuous radar echo extrapolation reconstructed dataset—namely, radar echo intensity distribution and convective cloud spatial distribution data that integrates temporal and spatial information—is input into the encoder module of the spatiotemporal series prediction model. The encoder module internally contains corresponding neural network layers and algorithms. For example, it may include a convolutional layer, which extracts features from the spatial information in the dataset through convolution operations, correlating and integrating radar echo intensity values ​​at different locations with surrounding spatial information. It may also include a recurrent neural network layer, which processes data in the temporal dimension and captures changing trends and patterns in the time series. These layers work together to extract and transform features from the input dataset, converting it into spatiotemporal encoded features of the radar echo intensity distribution data. These features are a highly abstract data representation that contains key temporal and spatial information, providing the basis for subsequent decoding and extrapolation.

[0031] Step 132: The decoder module of the spatiotemporal sequence prediction model performs time-step recursive decoding processing on the spatiotemporal coding features to generate radar echo intensity distribution prediction data for each time step within a preset time range.

[0032] Next, the decoder module of the spatiotemporal sequence prediction model processes the spatiotemporal coding features generated by the encoder module. The decoder module, which also includes a neural network structure and algorithm, uses a time-step recursive approach to predict the radar echo intensity distribution at each time step within a preset future time range, starting from the current moment. For example, by leveraging the recursive nature of a recurrent neural network, the module generates radar echo intensity distribution prediction data for the next time step through corresponding calculations and transformations based on the spatiotemporal coding features of the current time step and the prediction results of the previous time step. During this process, the decoder module continuously interprets and transforms the spatiotemporal coding features, converting the abstract features into specific predicted values ​​for the radar echo intensity distribution at each time step, thereby obtaining predicted data for each time step within the preset time range.

[0033] Step 133: performing spatial continuity constraint processing on the radar echo intensity distribution prediction data of each time step within the preset time range to generate spatially smoothed radar echo intensity distribution prediction data.

[0034] After obtaining the radar echo intensity distribution prediction data for each time step within a preset time range, spatial continuity constraint processing is performed to make this data more spatially reasonable and continuous. This processing uses corresponding algorithms and rules to operate on the two-dimensional radar echo intensity distribution prediction data matrix for each time step. For example, a spatial smoothing algorithm is used. For each position in the matrix, the intensity values ​​of its surrounding adjacent positions are considered and the intensity value at that position is adjusted through methods such as weighted averaging. If the intensity value of a position differs significantly from the intensity values ​​of its surrounding adjacent positions, the algorithm will correct it based on the surrounding values, making the intensity distribution across the entire space smoother and more continuous. Through this processing, spatially smoothed radar echo intensity distribution prediction data is generated, improving the spatial credibility and usability of the prediction data.

[0035] Step 134: Determine the spatial evolution path and intensity variation trend of the convective clouds based on the superposition result of the spatially smoothed radar echo intensity distribution prediction data and the convective cloud spatial distribution data, and generate the extrapolation result.

[0036] Finally, the spatially smoothed radar echo intensity distribution prediction data, obtained through spatial continuity constraint processing, is superimposed with the convective cloud spatial distribution data. During the superposition process, the radar echo intensity distribution prediction data for each time step is associated and matched with the corresponding convective cloud spatial range. For example, for one time step, the intensity value at different locations within the convective cloud spatial range is determined based on the radar echo intensity distribution prediction data. By superimposing and analyzing data from multiple time steps in this way, the changes in the spatial position and intensity of the convective clouds at different time steps are observed. This allows the spatial evolution path of the convective clouds to be determined—that is, the trajectory of the convective clouds in space over time—as well as the intensity change trend, such as whether the intensity is increasing or decreasing. Ultimately, an extrapolation result containing this information is generated.

[0037] In an alternative embodiment, a specific method and process for training a spatiotemporal series prediction model is provided. Through a series of operations and algorithms, the model can learn the characteristics and patterns of convective cloud radar echo data, thereby having the ability to accurately predict. Based on this, the training method of the spatiotemporal series prediction model includes:

[0038] Step 210: Obtain a historical convective cloud radar echo extrapolation dataset and corresponding real extrapolation result data, perform spatiotemporal alignment and noise filtering on the historical convective cloud radar echo extrapolation dataset, and generate a training dataset.

[0039] In the meteorological research scenario of the embodiment of the present application, a large amount of historical convective cloud radar echo extrapolation data sets and corresponding real extrapolation result data are first collected. These historical data come from monitoring records of convective clouds in different time periods in the past. Then, the historical convective cloud radar echo extrapolation data sets are processed. For the time dimension, a time alignment algorithm similar to step 121 is used to align the data at different time steps in the data set according to a set time resolution threshold, for example, set to ten minutes, to ensure that the data is consistent in time. For the spatial dimension, some spatial calibration and matching operations may also be involved so that data from different sources can accurately correspond in space. At the same time, the data is subjected to noise filtering using a filtering algorithm, such as a Gaussian filtering algorithm. For noise in the data set, such as outliers caused by errors in radar equipment or external interference, the data is smoothed by the algorithm to remove the influence of noise. After these spatiotemporal alignment and noise filtering processes, a training data set for training the spatiotemporal series prediction model is generated. The training data set has high-quality and consistent data characteristics, providing a reliable foundation for model training.

[0040] Step 220: Build an initial spatiotemporal sequence prediction model, which includes a spatiotemporal convolutional encoder, a time recursive decoder, and a spatial continuity constraint module.

[0041] In this embodiment, the initial spatiotemporal series prediction model is constructed. The spatiotemporal convolutional encoder consists of multiple convolutional layers. Parameters such as kernel size and stride length are set based on data characteristics and model requirements. For example, kernel sizes vary: small kernels are used to capture local spatial features, while large kernels are used to capture broader spatial information. Through convolution operations, spatial features are extracted from the input data, transforming the spatial information into a more abstract feature representation. The temporal recursive decoder utilizes recurrent neural network structures, such as long short-term memory (LSTM) or gated recurrent units (GRU). These structures effectively process time series data, recursively learning and processing information in the temporal dimension to predict future scenarios based on past information. The spatial continuity constraint module includes algorithms and rules for ensuring spatial smoothness and continuity, such as the spatial smoothing algorithm mentioned above. These three modules are combined according to the predefined connections to form the overall architecture of the initial spatiotemporal series prediction model, providing a foundational framework for subsequent training and prediction.

[0042] Step 230: Input the training data set into the initial spatiotemporal sequence prediction model, extract the multi-scale spatiotemporal features of the radar echo intensity distribution training data in the training data set through the spatiotemporal convolution encoder, and generate predicted radar echo intensity distribution data based on the multi-scale spatiotemporal features through the time recursive decoder.

[0043] Optionally, the generated training dataset is input into the established initial spatiotemporal sequence prediction model. The spatiotemporal convolutional encoder begins processing the radar echo intensity distribution training data in the training dataset. Through convolution operations with convolution kernels of different sizes, spatiotemporal features of the data are extracted at different scales. For example, smaller convolution kernels can capture subtle variations in radar echo intensity within a local area, while larger kernels can capture more macroscopic spatial distribution features. Simultaneously, information along the temporal dimension is combined to extract multi-scale spatiotemporal features. The recursive time decoder then operates based on these multi-scale spatiotemporal features. Leveraging the recursive nature of recurrent neural networks, it gradually generates predicted radar echo intensity distribution data based on the extracted spatiotemporal features. For example, starting from the features of the current time step, combined with the prediction results and spatiotemporal features of the previous time step, it predicts the radar echo intensity distribution for the next time step, thereby generating the entire predicted data sequence.

[0044] Step 240: Calling the spatial continuity constraint module to perform spatial smoothing constraints on the predicted radar echo intensity distribution data, and calculating the spatiotemporal consistency loss between the predicted radar echo intensity distribution data and the actual extrapolated result data.

[0045] After generating the predicted radar echo intensity distribution data, the spatial continuity constraint module is called to process it. This module applies a corresponding algorithm, such as the spatial smoothing algorithm mentioned above, to spatially smooth the predicted radar echo intensity distribution data at each time step, ensuring the spatial continuity and rationality of the data. The spatial and temporal consistency loss between the predicted radar echo intensity distribution data and the true extrapolated result data is then calculated using a mean square error loss function or a cross-entropy loss function. By comparing the differences between the predicted and true data in time and space, a numerical value is calculated to represent the degree of inconsistency between the two. For example, the difference between the predicted and true values ​​at each time step and spatial location is calculated. These differences are then integrated using a loss function to obtain a spatial and temporal consistency loss value. This loss value reflects the degree of deviation between the model's predictions and the actual situation, providing a basis for subsequent model parameter optimization.

[0046] Step 250: Optimize the parameters of the initial spatiotemporal sequence prediction model based on the spatiotemporal consistency loss until the spatiotemporal consistency loss converges, thereby obtaining a trained spatiotemporal sequence prediction model.

[0047] Optionally, the parameters of the initial spatiotemporal series prediction model are optimized based on the calculated spatiotemporal consistency loss. Optimization algorithms, such as stochastic gradient descent or the adaptive moment estimation (Adam) algorithm, are used. These algorithms adjust model parameters, such as the weights and biases of the neural network layers, based on the magnitude and direction of the loss. For example, the gradient of the loss function with respect to the parameters is calculated, and the parameter values ​​are adjusted based on the direction and magnitude of the gradient, gradually reducing the loss. This process is repeated continuously, i.e., continuously inputting training data, calculating the loss, and optimizing the parameters, until the spatiotemporal consistency loss converges. When the loss value no longer decreases significantly and reaches a relatively stable state, it indicates that the model has learned the characteristics and patterns in the data. A trained spatiotemporal series prediction model is now obtained, which can more accurately predict and analyze convective cloud radar echo data.

[0048] Step 140: Generate a convective cloud rain reduction operation strategy based on the extrapolation result, and send the convective cloud rain reduction operation strategy to the rain reduction operation service system for operation processing. The convective cloud rain reduction operation strategy includes operation point location, operation timing, and operation status evaluation characteristics.

[0049] In an embodiment of the present application, based on the convective cloud extrapolation results obtained previously, a convective cloud rain reduction operation strategy is formulated through a series of analyses and calculations. The strategy covers key information such as the location of the operation point, the timing of the operation, and the evaluation characteristics of the operation status. The strategy is then sent to a dedicated rain reduction operation service system to guide the actual operation execution.

[0050] In a preferred embodiment, the specific steps and operation methods for generating a convective cloud rain reduction operation strategy based on the extrapolation results are further described in detail to provide more specific guidance for practical applications. Based on this, the generation of a convective cloud rain reduction operation strategy based on the extrapolation results includes:

[0051] Step 141: extracting the spatial evolution path and intensity variation trend of the convective clouds from the extrapolation results, and determining the movement direction and coverage area of ​​the convective clouds within a preset time range.

[0052] In this meteorological operation scenario, the spatial evolution path and intensity change trend of the convective clouds are carefully analyzed from the extrapolated results that have been obtained. By analyzing the spatial position information and intensity values ​​of the convective clouds at different time steps, the movement direction and coverage area of ​​the convective clouds within the preset time range are determined. For example, by observing the coordinate changes of the core area of ​​the convective clouds at different time steps in the extrapolated results, the direction vector of its movement is calculated to determine the movement direction. As for the coverage area, by analyzing the spatial range data of the convective clouds at each time step, the spatial range occupied by them at different times is determined. By integrating the data of multiple time steps, the coverage area within the preset time range is obtained. The coverage area may be an irregular polygonal area. Through a series of algorithms and data processing, its boundaries and range are accurately defined, providing an important basis for the subsequent determination of the location of the operation point and the timing of the operation.

[0053] Step 142: Generate an operation opportunity according to the radar echo intensity change trend in the coverage area, where the operation opportunity includes an operation start time and an operation duration.

[0054] After determining the coverage area of ​​convective clouds, we further analyze the changing trends in radar echo intensity within that area. We observe how radar echo intensity changes over time at different locations within the coverage area, as determined by the extrapolated results. For example, if radar echo intensity shows a rapid upward trend within a specific time period and reaches a certain threshold, this may indicate that the convective cloud has entered a critical stage in its development. Based on this intensity trend and in conjunction with the objectives and requirements of the rain reduction operation, we determine the operation start time. For example, the time when the radar echo intensity rises to near its peak and is about to begin to decline is set as the operation start time to maximize the effectiveness of the rain reduction operation. The operation duration is determined by analyzing factors such as the slope of the intensity trend and the movement speed of the convective clouds. If the intensity decreases slowly and the convective cloud movement is moderate, a longer operation duration may be set to ensure continuous intervention as the convective cloud passes through the target area. Through this analysis and calculation, an operation timing is generated, including the precise operation start time and duration.

[0055] Step 143: Calculate the rain reduction operation point location based on the movement direction and the coverage area, where the operation point location includes the area with the maximum gradient change and a preset interception point on the movement path of the convective cloud core.

[0056] Once the convective cloud's direction of movement and coverage area are known, the location of rain reduction operations is calculated. First, the gradient of radar echo intensity is analyzed. The intensity gradient is calculated by differencing the radar echo intensity at different locations within the coverage area. For example, for each location within the coverage area, the intensity difference between it and its adjacent locations is calculated to form an intensity gradient field. Within this gradient field, areas with the greatest gradient variation are identified. These areas typically correspond to locations experiencing more dramatic physical changes within the convective cloud and are key areas for rain reduction operations. Simultaneously, pre-set interception points are selected along the core path based on the convective cloud's direction of movement and its core movement path. For example, pre-set interception points are determined based on the convective cloud's movement vector at a certain distance ahead of its path, taking into account factors such as the boundaries and shape of the coverage area. These pre-set interception points are used to facilitate early intervention during convective cloud movement, thereby achieving optimal rain reduction results. The locations of rain reduction operations are determined by combining the areas with the greatest gradient variation with the pre-set interception points.

[0057] In one implementation, the specific steps for calculating the location of a rain reduction operation point based on the moving direction and the coverage area are further refined. Through a more detailed calculation and analysis method, the accuracy and effectiveness of the location of the operation point are ensured. Based on this, the calculation of the location of a rain reduction operation point based on the moving direction and the coverage area includes:

[0058] Step 1431: extracting a centroid coordinate sequence of the core area of ​​the convective cloud at consecutive time steps within a preset time window according to the spatial evolution path of the convective cloud, and calculating a moving direction vector and an average moving speed based on the centroid coordinate sequence.

[0059] In this meteorological operation scenario, a preset time window is selected from the spatial evolution path data of the convective cloud, for example, data within the next hour. Within this time window, the centroid coordinates of the convective cloud core region are extracted for consecutive time steps. For each time step of the convective cloud spatial extent data, the centroid coordinates of the core region are calculated using a corresponding algorithm. For example, for a convective cloud spatial extent represented by a polygon, the centroid coordinates are obtained by taking the weighted average of the polygon vertex coordinates. The centroid coordinates of these consecutive time steps are arranged sequentially to form a centroid coordinate sequence. The movement direction vector is then calculated based on this centroid coordinate sequence. By calculating the difference between the centroid coordinates of adjacent time steps, the displacement vector within each time interval is obtained. These displacement vectors are then averaged to obtain the movement direction vector, which represents the average movement direction of the convective cloud within that time period. Simultaneously, the average movement rate is calculated. The average movement rate of the convective cloud within the preset time window is obtained by summing the distances between adjacent centroids in the centroid coordinate sequence and dividing it by the total duration of the time window. These movement direction vectors and the average movement rate provide important kinematic parameters for subsequently determining the location of the operation point.

[0060] Step 1432: Perform a spatial projection operation on the boundary coordinates of the coverage area and the moving direction vector to determine the coordinates of the geometric center point of the downstream protection area, and generate a polar coordinate system with the geometric center point as the origin.

[0061] It can be understood that after obtaining the moving direction vector of the convective cloud, the boundary coordinates of the coverage area are processed. The coordinates of each boundary point of the coverage area are spatially projected with the moving direction vector. For example, for a point on the boundary, the point is projected onto the straight line where the moving direction vector is located through the vector projection formula to obtain the coordinates of the projection point. By performing projection operations on all boundary points, the range of the downstream protection area is determined. Then, the coordinates of the geometric center point of the downstream protection area are calculated. For example, for a polygonal downstream protection area, the coordinates of the geometric center point are obtained by calculating the average value of the coordinates of the polygon vertices. A polar coordinate system is established with the geometric center point as the origin. In the polar coordinate system, the position of each point can be expressed by polar diameter and polar angle. Such a coordinate system is more convenient for subsequent analysis and calculation of the position of the work point, and can more intuitively reflect the relationship between the point and the moving direction of the convective cloud and the protection area.

[0062] Step 1433: Generate radial detection paths along the moving direction vector in the polar coordinate system, calculate the time evolution step length on each path based on the average moving rate, and generate a spatiotemporally correlated gridded detection point array.

[0063] After the polar coordinate system is established, radial detection paths are generated along the moving direction vector with the origin as the center. These detection paths are evenly distributed in the space centered on the origin, covering the downstream protection area. The time evolution step on each path is calculated based on the average moving rate calculated previously. For example, based on the average moving rate and the preset time resolution, the position points corresponding to certain time intervals on each path are determined. For example, if the average moving rate is 60 kilometers per hour and the preset time resolution is one minute, then on each path, the corresponding moving distance is calculated every minute based on the moving rate to determine the position point on the path at that moment. By performing such calculations on all radial detection paths, a gridded detection point array with temporal and spatial correlation is generated. Each detection point in the array corresponds to a time and spatial position, providing a detailed spatial and temporal framework for subsequent analysis of radar echo intensity and determination of operation points.

[0064] Step 1434: Perform radar echo intensity gradient field analysis on each detection point in the gridded detection point array to extract the vertical integrated liquid water content change rate and horizontal echo intensity gradient modulus of each detection point within a preset time step.

[0065] Optionally, a detailed radar echo intensity gradient field analysis is performed for each detection point in the generated gridded detection point array. Within a preset time step, for example, data within the next ten minutes is selected to analyze the vertical integrated liquid water content change rate of each detection point. By integrating the liquid water content data of different altitude layers and then calculating the changes within the preset time step, the vertical integrated liquid water content change rate is obtained. At the same time, the horizontal echo intensity gradient modulus is calculated. For the horizontal space around each detection point, the gradient of the radar echo intensity is calculated, and then its modulus is taken to represent the degree of change of the radar echo intensity in the horizontal direction of the point. By performing such an analysis on each detection point, two important characteristic parameters about the point are obtained. These parameters reflect the physical characteristics and change trends of the convective cloud at that location, providing a basis for the subsequent determination of candidate operation points.

[0066] Step 1435: Perform dimensionless normalization on the vertically integrated liquid water content change rate and the horizontal echo intensity gradient modulus to generate a composite impact factor parameter, and select detection points whose composite impact factor parameters exceed the target threshold as candidate operation points.

[0067] In order to comprehensively consider the impact of the two parameters, the vertically integrated liquid water content change rate and the horizontal echo intensity gradient modulus, on the selection of the operation point, they are dimensionlessly normalized. A normalization algorithm is used, such as subtracting the respective means of the vertically integrated liquid water content change rate and the horizontal echo intensity gradient modulus from each other, and then dividing them by their respective standard deviations, to convert the two parameters into the same dimensional range. Then, the two standardized parameters are combined in some form, such as weighted summation, to generate a composite impact factor parameter. A target threshold is set, such as 0.8, and detection points whose composite impact factor parameters exceed the target threshold are screened out. These detection points have more significant characteristics in terms of the physical characteristics and changes of convective clouds and are considered to be possible candidate operation points, providing a candidate set for further determination of the operation point location.

[0068] Step 1436: Generate a density clustering model based on the spatial distribution density of the candidate operation points, extract the cluster center coordinates through the density clustering model, and perform position offset compensation according to the geometric topological characteristics of the downstream protection area to generate the location of the rain reduction operation point.

[0069] After obtaining the set of candidate work points, a density clustering model is generated based on the spatial distribution density of these candidate work points. A density clustering algorithm, such as the DBSCAN algorithm, is used. This algorithm divides density-connected points into different clusters based on the distribution density of the candidate work points in space. Through this algorithm, dense areas and sparse areas in the candidate work points can be found. Then, the cluster center coordinates are extracted from each cluster. These cluster centers represent the concentrated locations of the work points in the cluster. However, since the geometric topological characteristics of the downstream protection area may be more complex, such as the existence of irregular shapes or certain special areas that require key protection, it is necessary to compensate for the position offset of the cluster center coordinates based on these geometric topological characteristics. For example, if the cluster center is close to the boundary of the protection area and may exceed the effective working range, the cluster center coordinates are appropriately adjusted according to the shape and requirements of the protection area to place it in a more suitable working position. After such processing, the location of the rain reduction work point is finally generated.

[0070] Step 1437: performing a correlation check on the projection component of the moving direction vector of each operation point in the rain reduction operation point location and the coverage area parameter, and outputting an optimized operation point location coordinate sequence after eliminating redundant points in spatial distribution.

[0071] The generated rain reduction operation point locations are further optimized. The projection component of the moving direction vector of each operation point is calculated, that is, the moving direction vector of the operation point is projected onto one of the reference directions of the coverage area to obtain the projection component. At the same time, the coverage area parameters are considered, such as the total area of ​​the coverage area or the effective operation area. By analyzing the correlation between the projection component of the moving direction vector of the operation point and the coverage area parameter, it is judged whether the spatial distribution of the operation point is reasonable. If the relationship between the projection component of one of the operation points and the coverage area shows that it may be redundant in spatial distribution, for example, the operation point has little impact on the coverage area and overlaps with other operation points in function, it will be eliminated. After such correlation verification and redundant point elimination operations, the optimized operation point position coordinate sequence is output. These coordinates represent more reasonable and effective rain reduction operation point locations.

[0072] Step 144: Generate an operation status assessment feature based on the area with the maximum gradient change and the preset interception point. The operation status assessment feature includes an operation coverage matching degree and an operation effect sustainability index.

[0073] After determining the area with the maximum gradient change and the preset interception point, operation status assessment features are generated based on this. These features are used to evaluate the effectiveness and status of rain reduction operations during implementation, providing a basis for subsequent operation adjustments and optimization.

[0074] In another implementation, a specific operation and calculation method for generating an operation status assessment feature based on the area with the maximum gradient change and the preset interception point are described in detail to more accurately assess the operation status. Based on this, generating the operation status assessment feature based on the area with the maximum gradient change and the preset interception point includes:

[0075] Step 1441: extracting the spatiotemporal evolution parameters of the area with the maximum gradient change from the extrapolation results, including the echo intensity gradient modulus time series, the liquid water content spatial integral value, and the core movement path curvature radius.

[0076] Based on the previously obtained extrapolated results, detailed parameters are extracted for the areas with the greatest gradient changes. First, the time series of the echo intensity gradient modulus is extracted. At different time steps, the radar echo intensity gradient within the area with the greatest gradient change is calculated and its modulus is taken, forming a time-varying sequence that reflects the temporal evolution of the intensity of the radar echo intensity change within this area. Simultaneously, the spatial integral of the liquid water content is calculated. The liquid water content at different altitudes in the area with the greatest gradient change is integrated to obtain a value that comprehensively reflects the total amount of liquid water in this area. Changes in this value can reflect the changes in the water vapor content within the convective cloud. Furthermore, the curvature radius of the core moving path is extracted. The curvature radius of the convective cloud core moving path is calculated using a corresponding algorithm. The size of the curvature radius reflects the degree of curvature of the convective cloud moving path and is of great significance for analyzing the motion trend and stability of the convective cloud. These spatiotemporal evolution parameters describe the characteristics of the area with the greatest gradient change from different aspects, providing basic data for the subsequent generation of operational status assessment features.

[0077] Step 1442: Construct a three-dimensional spherical coordinate system within the spatial range of the preset interception point, and calculate the echo intensity flux accumulation value within the preset distance upstream of the interception point and the dissipation rate attenuation coefficient within the preset distance downstream.

[0078] A three-dimensional spherical coordinate system is constructed within the spatial range of the preset interception point to more accurately analyze the distribution of physical quantities in that area. A spherical space is constructed with the preset interception point as the sphere's center and a set radius. Within this three-dimensional spherical coordinate system, the cumulative echo intensity flux within a preset distance upstream of the interception point is calculated. By integrating the radar echo intensities at different locations within the preset upstream distance, taking into account factors such as spatial position and direction, the cumulative echo intensity flux is obtained. This value reflects the total amount of radar echo energy entering the area upstream of the interception point. Simultaneously, the dissipation rate attenuation coefficient is calculated within the preset downstream distance. The dissipation of radar echo intensity over time within the preset downstream distance is observed. By analyzing and calculating the intensity values ​​at different time steps, the dissipation rate attenuation coefficient is obtained. This coefficient represents the degree of attenuation of radar echo intensity in the downstream area over time and reflects the impact of rain reduction operations on convective clouds.

[0079] Step 1443: Perform time phase matching processing on the echo intensity gradient modulus time series and the dissipation rate attenuation coefficient to generate a time-varying correlation matrix and extract the principal component eigenvector.

[0080] Optionally, the extracted echo intensity gradient modulus time series and dissipation rate attenuation coefficient are subjected to time phase matching processing. Since the rhythm of these two parameters changing over time may be different, a time phase matching algorithm is used to better align and compare them in time. For example, a phase synchronization algorithm is used to adjust the time axes of the two sequences so that their key features can correspond in time. Then, a time-varying correlation matrix is ​​generated based on the matched sequence. The elements of the matrix represent the degree of correlation between the two parameters at different time points. The matrix elements are filled by calculating the correlation index of the two sequences at different time points. Next, a principal component analysis is performed on the time-varying correlation matrix to extract the principal component eigenvector. The principal component eigenvector can retain the information in the matrix to the greatest extent, reflecting the main change pattern and correlation characteristics between the echo intensity gradient modulus time series and the dissipation rate attenuation coefficient, and providing key information for the subsequent generation of operation status evaluation indicators.

[0081] Step 1444: Perform a dimensionless fusion operation on the liquid water content spatial integral value and the echo intensity flux cumulative value to obtain the operation energy intervention efficiency index, and calculate the path deviation correction factor based on the core moving path curvature radius.

[0082] Optionally, a dimensionless fusion operation is performed on the spatially integrated value of liquid water content and the cumulative value of echo intensity flux. A fusion algorithm is employed, such as normalizing the two parameters separately to bring them into the same dimensional range. Then, they are fused together through weighted summation or other appropriate computational methods to produce an operation energy intervention effectiveness index. This index comprehensively reflects the effect of the rain reduction operation on the energy and water vapor content within the convective cloud. A higher index indicates a greater impact on the convective cloud and a better effect. Simultaneously, a path deviation correction factor is calculated based on the curvature radius of the core movement path. For example, the correction factor is determined based on the size and variation of the curvature radius. If the curvature radius is small, indicating a significant curvature of the convective cloud movement path, a larger correction factor may be required to adjust the operation strategy to accommodate the convective cloud's motion changes. This path deviation correction factor is used to correct the operation status assessment, taking into account the impact of the convective cloud movement path.

[0083] Step 1445: construct a multidimensional state space based on the principal component eigenvector, and map the operation energy intervention efficiency index and the path deviation correction factor into the state space to generate a state trajectory curve.

[0084] Optionally, a multidimensional state space is constructed based on the extracted principal component eigenvectors. The dimension of the principal component eigenvectors determines the dimension of the state space. For example, if there are two principal component eigenvectors, a two-dimensional state space is constructed. The operation energy intervention effectiveness index and the path offset correction factor are respectively used as coordinate values ​​in the state space, and they are mapped to the multidimensional state space. As time goes by, these two parameters will change, forming a trajectory curve in the state space, namely the state trajectory curve. This curve reflects the comprehensive state of the operation at different time points. By analyzing the curve, we can understand the evolution of the operation effect over time, as well as the comprehensive impact of the operation energy intervention and the convective cloud motion path on the operation state.

[0085] Step 1446: Extract the time parameters and spatial coordinates corresponding to the extreme curvature points of the state trajectory curve, and generate a coverage matching evaluation matrix based on the operation time window of the preset interception point.

[0086] Optionally, the state trajectory curve is analyzed to extract the time parameters and spatial coordinates corresponding to the extreme points of curvature. The extreme points of curvature represent the points where the trend of the state trajectory curve changes significantly, and the time and space coordinates corresponding to these points are of great significance. Combined with the operation time window of the preset interception point, these time parameters and space coordinates are associated and matched with the operation time window. For example, determine whether the extreme point of curvature falls within the operation time window, and its specific position within the window. Based on this information, a coverage matching evaluation matrix is ​​generated. The elements of the matrix represent the degree of match between the operation coverage and the actual effect under different circumstances. For example, the rows of the matrix can represent different time points, and the columns can represent different spatial areas. The element values ​​are assigned according to the results of association and matching, so as to comprehensively evaluate the relationship between the operation coverage and the actual operation effect.

[0087] Step 1447: Perform singular value decomposition on the coverage matching evaluation matrix, extract the eigenvector corresponding to the maximum singular value as the operation coverage matching degree, and calculate the stability index of the eigenvector evolving over time as the operation effect sustainability parameter.

[0088] In this step, the generated coverage matching evaluation matrix is ​​subjected to singular value decomposition. Singular value decomposition is a matrix decomposition method that can decompose a matrix into the product of three matrices, where the singular values ​​reflect the important characteristics of the matrix. Through singular value decomposition, the eigenvector corresponding to the maximum singular value is extracted. This eigenvector integrates the key information in the matrix and is used as an indicator of the job coverage matching degree. It reflects the best matching relationship between the job coverage and the actual effect. At the same time, the stability index of the eigenvector evolving over time is calculated. The changes in the eigenvector at different time points are observed, and its stability is measured by calculating some statistical indicators of the eigenvector, such as variance. This stability index serves as a parameter for the continuity of the job effect, reflecting the degree of continuity of the job effect over time. The higher the stability, the longer the job effect lasts and the more stable the effect.

[0089] Step 1448: Perform linear weighted synthesis on the job coverage matching degree and the job effect persistence parameter to generate a comprehensive quantitative job status evaluation feature vector.

[0090] After obtaining the operation coverage matching degree and the operation effect persistence parameter, these two parameters are linearly weighted and synthesized to comprehensively evaluate the operation status. Based on actual needs and experience, the operation coverage matching degree and the operation effect persistence parameter are assigned corresponding weights, for example, weight w1 is assigned to the operation coverage matching degree, and weight w2 is assigned to the operation effect persistence parameter, where w1 + w2 = 1. Through linear weighted calculation, that is, the value of each dimension of the operation status evaluation feature vector is equal to the operation coverage matching degree multiplied by w1 plus the operation effect persistence parameter multiplied by w2, thus generating a comprehensive quantitative operation status evaluation feature vector. This vector comprehensively reflects the comprehensive performance of the operation in terms of coverage and effect persistence, providing a comprehensive and quantitative indicator for evaluating the status of convective cloud rain reduction operations, facilitating subsequent analysis and comparison of operation effects, as well as adjusting and optimizing operation strategies based on the evaluation results.

[0091] Step 145: Integrate the operation timing, operation point location, and operation status assessment features to generate the convective cloud rain reduction operation strategy.

[0092] After determining the operation timing, operation point location, and operation status assessment characteristics, these key elements are integrated to form a complete convective cloud rain reduction operation strategy. Specifically, the operation timing information, including the operation start time and operation duration, the calculated and optimized operation point location coordinate sequence, and the comprehensive and quantified operation status assessment feature vector are organically combined. This information is organized through a defined data structure or format, for example, a structure or data table containing these three components can be constructed (in actual implementation, the appropriate data structure is determined based on the specific programming environment and requirements). The resulting convective cloud rain reduction operation strategy comprehensively covers key aspects such as the time and location of operation implementation and the evaluation of operation effectiveness. This provides clear and detailed operational guidance for the rain reduction operation service system, ensuring that operations are carried out at the appropriate time and location, and enabling effective evaluation and monitoring of operation results, allowing for timely adjustment of the operation strategy and improving the efficiency and effectiveness of rain reduction operations.

[0093] In an exemplary embodiment, a detailed description is provided of how to convert a generated convective cloud rain reduction operation strategy into specific instructions and send them to a rain reduction operation service system to implement the actual operation execution process. Based on this, the convective cloud rain reduction operation strategy is sent to the rain reduction operation service system for execution, including:

[0094] Step 146: Generate a geographic coordinate instruction based on the operation point location in the convective cloud rain reduction operation strategy, the geographic coordinate instruction including latitude and longitude information and an operation altitude range; convert the operation opportunity into a time control instruction for the operation equipment, the time control instruction including the operation start time, operation duration and operation interval period; generate an operation resource scheduling instruction based on the operation status evaluation characteristics, the resource scheduling instruction including the catalyst release amount and equipment deployment priority; generate an operation instruction set based on the geographic coordinate instruction, the time control instruction and the resource scheduling instruction, and send the operation instruction set to the rain reduction operation service system to perform multi-device collaborative operation.

[0095] First, the location information of the work points in the convective cloud rain reduction operation strategy is processed. The work point location is represented by a certain coordinate format. A coordinate conversion algorithm converts this information into geographic coordinates, namely longitude and latitude information. Simultaneously, the operating altitude range for each work point is determined based on the operational requirements and the actual convective cloud conditions. For example, the operating altitude range for one work point might be from 1000 meters to 3000 meters above the ground. The longitude and latitude information and the operating altitude range are combined to form geographic coordinate instructions.

[0096] Next, the operation timing is converted. Information such as the operation start time and duration is converted into time control instructions that the operation equipment can recognize and execute, based on the control requirements of the operation equipment. For example, the operation start time is converted into the absolute start time of the equipment, and the operation duration is converted into the equipment's operating time setting. Furthermore, considering the possibility of multiple operation equipment working together, the operation interval period—the time interval between two consecutive operations—must be determined. This information is integrated to form the time control instructions for the operation equipment.

[0097] Then, job resource scheduling instructions are generated based on the job status assessment characteristics. The job status assessment characteristics reflect the expected results and requirements of the job. Based on this information, the required catalyst delivery amount for each job site is determined. For example, if the job site is located in an area with strong convective cloud development, a larger amount of catalyst may be required. At the same time, equipment deployment priority is determined based on factors such as the importance of the job and the performance of the equipment. For example, for key job sites, higher-performance equipment is prioritized. Information such as catalyst delivery amount and equipment deployment priority is combined to form job resource scheduling instructions.

[0098] Finally, the geographic coordinate instructions, time control instructions, and resource scheduling instructions are integrated to generate a set of job instructions. This set contains all the key information required for job execution. This set of job instructions is then sent to the rain reduction operation service system via network communication or other appropriate means. After receiving the set of instructions, the rain reduction operation service system coordinates multiple operating devices based on the information contained therein, enabling multi-device collaborative operations. For example, devices are deployed to designated locations according to geographic coordinate instructions, their startup and operating times are controlled according to time control instructions, and catalyst delivery and equipment deployment sequence are allocated according to resource scheduling instructions, ensuring that rain reduction operations are executed accurately and efficiently according to predetermined strategies.

[0099] As an optional embodiment, the method further includes:

[0100] Step 300: Acquire operation execution data fed back by the rain reduction operation service system in real time, wherein the operation execution data includes the actual operation point location and radar echo monitoring data after the operation; perform a matching degree analysis on the actual operation point location and the operation point location in the convective cloud rain reduction operation strategy to generate a position deviation correction feature; generate an effect deviation correction feature based on the difference between the radar echo monitoring data after the operation and the extrapolated result; and perform online parameter adjustment on the spatiotemporal series prediction model based on the position deviation correction feature and the effect deviation correction feature.

[0101] This optional embodiment introduces how to adjust and optimize the spatiotemporal series prediction model according to actual feedback data during the execution of the operation to improve the accuracy of the model and the operation effect. During the execution of the rain reduction operation, the operation execution data fed back by the rain reduction operation service system is obtained in real time. The actual operation point position is the location information where the operation equipment actually arrives and performs the operation. By comparing and analyzing it with the operation point position in the operation strategy, the degree of matching between the two is calculated. For example, a distance measurement algorithm is used to calculate the Euclidean distance or other suitable distance measurement value between the actual operation point and the planned operation point, and then the preset threshold is used to determine whether the matching degree meets the requirements. If there is a deviation between the actual operation point position and the planned operation point position, a position deviation correction feature is generated based on information such as the size and direction of the deviation.

[0102] At the same time, radar echo monitoring data is obtained after the operation, reflecting the actual state of the convective clouds. This data is then compared and analyzed with the extrapolated results generated by the previous spatiotemporal series prediction model. Differences between the two are compared in terms of radar echo intensity distribution and the spatial extent of convective clouds. For example, the difference in radar echo intensity at different locations is calculated, and changes in the spatial extent of convective clouds are statistically analyzed. Based on these differences, effect deviation correction features are generated, which reflect the gap between the actual and expected effects of the operation.

[0103] Finally, based on the position deviation correction features and the effect deviation correction features, the spatiotemporal series prediction model undergoes online parameter adjustments. Using online learning algorithms, such as stochastic gradient descent, the model parameters are adjusted based on the deviation correction features. If the position deviation is large, model parameters related to spatial positioning may be adjusted; if the effect deviation is significant, parameters related to radar echo intensity prediction and convective cloud evolution simulation are adjusted. By continuously adjusting model parameters based on actual feedback data, the spatiotemporal series prediction model can more accurately reflect the actual convective cloud conditions, improve the model's prediction accuracy, and thus optimize subsequent convective cloud rain reduction strategies.

[0104] As a non-limiting embodiment, after sending the convective cloud rain reduction operation strategy to the rain reduction operation service system for operation processing, the method further includes:

[0105] Step 400: Acquire spatial trajectory data of the operating equipment and measured liquid water content data from the meteorological monitoring station in real time, perform moving direction vector decomposition on the spatial trajectory data to generate equipment motion trajectory characteristic parameters; perform time domain difference operation based on the measured liquid water content data and the predicted liquid water content data in the extrapolated results to generate an influence factor correction coefficient for each operating point; calculate the coverage overlap rate between adjacent operating points based on the equipment motion trajectory characteristic parameters, and generate a target weight matrix in combination with the influence factor correction coefficient; perform density resampling on the operating points in the uncovered area using the target weight matrix to generate a derived operating point coordinate set and update it to the rain reduction operation service system.

[0106] This embodiment describes how, after the operation is executed, the operation strategy is further optimized and adjusted based on the spatial trajectory data of the operating equipment and the measured data of the liquid water content of the meteorological monitoring station, so as to improve the coverage effect and pertinence of the operation. In detail, the spatial trajectory data of the operating equipment during the operation is collected in real time. These data record the movement path of the operating equipment in space. The spatial trajectory data is decomposed into the moving direction vector. By analyzing the position changes of the equipment at different times, the moving direction vector and speed parameters of the equipment are calculated to generate the equipment motion trajectory characteristic parameters. These parameters can accurately describe the motion state of the operating equipment and provide a basis for subsequent analysis of the operation coverage.

[0107] At the same time, measured liquid water content data from meteorological monitoring stations is obtained and then subjected to temporal difference calculations with the predicted liquid water content data from the extrapolated results of the spatiotemporal series prediction model. For each operating point, the difference between the measured and predicted liquid water content at different time steps is calculated. Based on these differences, a correction coefficient for the impact factor at each operating point is generated. This coefficient reflects the degree of difference between the actual liquid water content and the predicted value and can be used to correct operating strategies.

[0108] Based on the characteristic parameters of the equipment's motion trajectory, the coverage overlap ratio between adjacent work points is calculated. By analyzing the equipment's motion trajectory, the presence and degree of overlap in the coverage areas of adjacent work points are determined. For example, the coverage overlap ratio is calculated by calculating the ratio of the intersection area to the union area of ​​the coverage areas of two work points. Combined with the impact factor correction coefficient, a target weight matrix is ​​generated based on the importance and actual effectiveness of different work points. The elements in the matrix represent the weight of each work point, which is determined by combining the coverage overlap ratio and the impact factor correction coefficient.

[0109] Finally, the target weight matrix is ​​used to density-resample the operating points in uncovered areas. For areas that are not fully covered, the sampling probability of each operating point is determined based on the weight matrix. Operating points with greater weights have a higher probability of being sampled. A resampling algorithm generates a set of derived operating point coordinates. These new operating point coordinates are generated based on actual operating conditions and data feedback, making them more targeted and reasonable. This derived operating point coordinate set is updated to the rain reduction operation service system, enabling the system to adjust operations based on the new operating points, improving coverage and overall efficiency.

[0110] As a non-limiting embodiment, after sending the convective cloud rain reduction operation strategy to the rain reduction operation service system for operation processing, the method further includes:

[0111] Step 500: extract the three-dimensional wind field vector data of each height layer in the operating area, perform vector superposition operation on the wind field vector data and the spatial evolution path in the extrapolated result to generate a corrected moving path; calculate the catalyst diffusion rate compensation factor according to the curvature parameter of the corrected moving path, and generate a layered delivery rate control instruction in combination with the operating height range; perform parabolic equation fitting on the catalyst settling trajectory of each operating point, and adjust the equation coefficient based on the diffusion rate compensation factor to generate a three-dimensional diffusion model; based on the layered delivery rate control instruction and the three-dimensional diffusion model, regulate the catalyst release pulse frequency and injection elevation angle in real time.

[0112] This embodiment illustrates how to use the three-dimensional wind field vector data and extrapolation results in the operating area to precisely control the placement of the catalyst after the operation is executed, so as to improve the diffusion effect of the catalyst and the operating efficiency. First, the three-dimensional wind field vector data of different height layers in the operating area are extracted. These data reflect the wind direction and wind speed information at different heights in the operating area. The wind field vector data is vector-superimposed with the spatial evolution path of the convective cloud in the extrapolation result of the spatiotemporal sequence prediction model. For example, for the spatial position and movement direction of the convective cloud at a certain moment, the wind field vector at that position is added to obtain a corrected movement path. The corrected movement path takes into account the influence of the wind field on the movement of the convective cloud and more accurately reflects the actual movement trajectory of the convective cloud.

[0113] A catalyst diffusion rate compensation factor is calculated based on the curvature parameter of the corrected movement path. The curvature parameter reflects the degree of curvature of the corrected movement path; the greater the curvature, the greater the potential impact on catalyst diffusion. Furthermore, a compensation factor is calculated based on the curvature to adjust the catalyst diffusion rate. Taking into account the operating altitude range, meteorological conditions at different altitudes, and catalyst diffusion requirements, the catalyst release rate for each altitude is determined based on the altitude range, and the layered release rate control instructions are generated.

[0114] A parabolic equation is fitted to the catalyst settling trajectory at each operating point. Actual catalyst settling data is collected and, using curve fitting algorithms such as the least squares method, a parabolic equation is fitted to describe the catalyst settling trajectory. Based on the previously calculated diffusion rate compensation factor, the coefficients of the parabolic equation are adjusted to generate a three-dimensional diffusion model that more accurately reflects the catalyst's diffusion under current meteorological conditions.

[0115] Finally, based on the layered delivery rate control instructions and the three-dimensional diffusion model, the catalyst release pulse frequency and injection elevation angle are controlled in real time. Based on the layered delivery rate control instructions, the catalyst delivery amount is determined at different altitudes and time points, and the delivery amount is controlled by adjusting the catalyst release pulse frequency. Simultaneously, the injection elevation angle is adjusted based on the three-dimensional diffusion model and actual operational requirements to ensure that the catalyst is dispersed in the appropriate location and direction, improving catalyst utilization efficiency and the effectiveness of rain reduction operations. This ensures efficient and accurate catalyst delivery under various meteorological conditions and convective cloud states.

[0116] As a non-limiting embodiment, after sending the convective cloud rain reduction operation strategy to the rain reduction operation service system for operation processing, the method further includes:

[0117] Step 600: Obtain a mapping relationship table between the Doppler radar reflectivity factor and the catalyst quantity of the operating area, and search for the optimal catalyst quantity gradient in the mapping relationship table based on the real-time reflectivity change rate; calculate the resource utilization index according to the remaining catalyst reserves of the operating equipment and the optimal catalyst quantity gradient, and generate the resource allocation status characteristics of each operating equipment; perform sliding average processing on the resource allocation status characteristics within a preset time window to generate a target resource scheduling sequence; perform attenuation control on the delivery quantity of non-critical operating points based on the target resource scheduling sequence, and allocate excess delivery tasks to critical operating points.

[0118] The embodiment of the present application illustrates how, after the operation is executed, the catalyst delivery can be reasonably scheduled based on the mapping relationship between the Doppler radar reflectivity factor of the operation area and the catalyst delivery amount, as well as the resource situation of the operation equipment, so as to optimize resource utilization and operation results. First, a mapping relationship table between the Doppler radar reflectivity factor of the operation area and the catalyst delivery amount is obtained. This table is established through a large amount of experiments and historical data, and it records the optimal catalyst amount corresponding to different reflectivity factors. Based on the real-time monitored Doppler radar reflectivity change rate, the corresponding optimal catalyst amount gradient is searched in the mapping relationship table. For example, if the reflectivity change rate is fast, it means that the convective cloud is developing rapidly, and according to the mapping relationship table, it is determined that a larger catalyst amount gradient is required at this time.

[0119] A resource utilization index is calculated based on the remaining catalyst reserves of each operating unit and the optimal catalyst quantity gradient. For example, the resource utilization index is calculated by calculating the ratio of the remaining catalyst reserves of each operating unit to the required catalyst reserves calculated according to the optimal catalyst quantity gradient. This index reflects the resource utilization of each operating unit and generates resource allocation status characteristics for each operating unit. These characteristics can be represented by data vectors or other suitable data structures, comprehensively describing the resource status of each operating unit.

[0120] Within a preset time window, for example, ten minutes, a sliding average of resource allocation status characteristics is performed. By continuously updating the data within the time window, the average of the resource allocation status characteristics is calculated to generate a target resource scheduling sequence. This sequence reflects the average status and trend of resource allocation for operating equipment over a period of time.

[0121] Finally, based on the target resource scheduling sequence, the work sites are classified and resources are scheduled. Key and non-key work sites are identified based on the importance and effectiveness of the work. For non-key work sites, catalyst delivery is reduced based on the target resource scheduling sequence and available resources. For key work sites, excess delivery tasks are allocated to ensure operational effectiveness, prioritizing catalyst supply to key sites. This resource scheduling approach optimizes catalyst delivery, improves resource utilization efficiency, and ensures optimal rain reduction results within limited resources.

[0122] It is understandable that after completing the above operations such as formulation and execution of the convective cloud rain reduction operation strategy, in order to further improve the effect and efficiency of the convective cloud rain reduction operation, it can also be implemented in combination with the following embodiments.

[0123] First, when using radar base data from a single radar station for many years in the data preparation phase, the Python radar library wradlib is used to read the data and perform appropriate quality control on all elevation echoes using a texture filter to ensure data quality. Subsequently, the data is interpolated to a spatial resolution of 100 using the four nearest points. The combined radar reflectivity product is formed in a Cartesian grid. In order to facilitate storage and subsequent processing, the data is stored as a single-channel grayscale image in PNG format using the radar's maximum detection range of 230 km. The image resolution is set to Next, all severe convective processes were selected from these data to construct a severe convective radar echo extrapolation dataset. The screening criteria were set to ensure that the maximum echo value of each sample (20 frames, 2 hours) was above 50dBZ and the maximum echo value of each frame was above 30dBZ. These strict criteria were used to screen out severe convective processes that met the requirements. Continuous sampling was used, that is, a radar image was taken every 6 minutes, and the window size was set to 20 frames (10 frames were input and 10 frames were predicted). The dataset was constructed using a sliding window method to provide sufficient and appropriate data support for subsequent model training.

[0124] Furthermore, during the model selection and optimization process, a variety of mainstream spatiotemporal series prediction models were used, and key model parameters such as the number of stacked layers, number of hidden neurons, and learning rate were set for each model. These mainstream models include ConvLSTM, PredRNN, PredRNN++, MIM, PredRNN-V2, SimVP-V1, SimVP-V2, and TAU. These models were trained and optimized based on a constructed dataset of severe convective radar echo extrapolation. The optimal model was selected by comparing their performance. For example, different models have their strengths and weaknesses in capturing the spatiotemporal characteristics of convective clouds and in terms of prediction accuracy. Extensive experiments and evaluations led to the identification of the most suitable model for this dataset and task. Furthermore, adjustments to the sampling method also significantly impacted the results. Given the long time it takes for operational effects to appear and the excessive attenuation characteristics of the model, the sampling method was modified to avoid masking operational effects and improve extrapolation performance. Sparse sampling is used, where a radar chart is taken every 12 minutes to reconstruct the dataset. Based on the original dataset, every other radar chart is taken to form a sparsely sampled severe convection radar echo extrapolation dataset. This sampling method reduces the data volume while better adapting to operational requirements and model characteristics.

[0125] Furthermore, based on a sparsely sampled severe convection radar echo extrapolation dataset, key parameters were set using the optimal model and then trained and optimized to obtain the weights for the optimal training results. Because the error in the model's extrapolation results increases dramatically with increasing time steps, to avoid masking the effectiveness of the operation and relying on the model's learning of the normal development and dissipation rates of clouds, only the first frame of the extrapolation results, i.e., the 12-minute extrapolation results, were extracted. Extrapolation was then performed on a 12-minute cycle to obtain longer-term extrapolation results. From the outset of the operation, the optimal result weights obtained from training and optimization based on the sparsely sampled severe convection radar echo extrapolation dataset were used to perform radar echo extrapolation on a 12-minute cycle, extracting only the first frame, i.e., the 12-minute extrapolation results. This extrapolation result serves as a reference for the normal development and dissipation rates of clouds.

[0126] During the operation, it is crucial to identify the target area for analysis. Due to the uncertainty of the shadow operation's impact zone, significant changes may occur around the area directly affected by the operation. This embodiment of the present application focuses solely on changes in the area directly affected by the operation. The diffusion range of silver iodide is estimated based on the location of the operation point and the direction and speed of cloud movement. This defines the direct impact zone of the rain reduction operation as the target area for analysis, providing a clear scope for subsequent effect evaluation.

[0127] During the operational effectiveness evaluation phase, the difference between the extrapolated radar chart and the actual radar chart can be used to evaluate and analyze the combined radar reflectivity changes in the direct impact area for each 12-minute period. This method allows for qualitative determination of the location of cloud radar reflectivity reduction, and qualitative determination of cloud development trends based on the entire movement of clouds within the direct impact area. Furthermore, the recommended ZR relationship for convective cloud precipitation in the US summer is used to calculate rainfall in the cloud clusters directly impacted by the operation, using the extrapolated radar chart. This is then compared with the actual rainfall to quantitatively analyze the effectiveness of the rain reduction operation. This combined qualitative and quantitative evaluation approach provides a comprehensive and accurate understanding of the actual effectiveness of the rain reduction operation, providing a strong basis for adjusting and optimizing subsequent operational strategies.

[0128] In addition, in the actual implementation process, the missing data in the radar blind area can be eliminated by adaptively adjusting the semi-variogram function parameters in the Kriging interpolation algorithm, and the spatiotemporal alignment can be achieved by combining the dimensional unification of the convective cloud movement speed and spatial resolution (for example, converting the speed threshold of 50km / h to m / s and matching it with the spatial sampling interval); when constructing the spatiotemporal convolution encoder, a multi-scale convolution kernel combination (such as 3×3 and 5×5 convolution in parallel) is used to extract local and global features, and the sequence dependency is captured through the time recursive mechanism of the gated recurrent unit (GRU); for the spatial continuity constraint processing, the Laplace-based The dynamic generation of the neighborhood weight matrix in the smoothing algorithm achieves spatial regularization of the echo intensity distribution. In the construction of the density clustering model, the optimal cluster center is determined through the adaptive calculation of the ε neighborhood radius in the DBSCAN algorithm (based on the standard deviation of the spatial distribution of candidate operation points). For the fusion operation of liquid water content and echo intensity, the entropy method is used to determine the objective weight coefficient of the characteristic parameter to avoid subjective experience bias. At the same time, the projection component of the moving path is corrected based on the three-dimensional wind field data assimilation technology in the WRF meteorological model, and the catalyst diffusion rate is combined with the Reynolds number correction factor of the parabolic equation to achieve three-dimensional diffusion model optimization.

[0129] On the one hand, the embodiment of the present application can effectively integrate data by sparse sampling and spatiotemporal alignment processing of the convective cloud radar echo extrapolation dataset in the target area, generating a spatiotemporally continuous radar echo extrapolation reconstruction dataset, greatly improving the consistency and availability of the data, and making the radar echo extrapolation reconstruction dataset more accurately reflect the actual situation of convective clouds; on the other hand, calling the spatiotemporal sequence prediction model to perform cyclic radar echo extrapolation processing can deeply explore the potential patterns in the data and accurately generate extrapolation results that include the spatial evolution path and intensity change trend of convective clouds within a preset time range, providing a strong basis for subsequent decision-making; on the other hand, the convective cloud rain reduction operation strategy generated based on the extrapolation results can comprehensively consider the operation point location, operation timing and operation status evaluation characteristics, making the rain reduction operation more targeted and scientific, and sending the convective cloud rain reduction operation strategy to the rain reduction operation service system for operation processing, which can effectively optimize the operation process, improve the accuracy and efficiency of rain reduction operations, and provide reliable support for meteorological operations.

[0130] In summary, the embodiments of the present application obtain a comprehensive convective cloud radar echo extrapolation data set, perform innovative data processing, accurately generate extrapolation results and formulate reasonable operation strategies, effectively solving the problems of insufficient data utilization and unscientific operation strategy formulation in convective cloud rain reduction operations in the existing technology, and significantly improving the accuracy and effectiveness of the operation.

[0131] Based on the same inventive concept, the present application also provides a convective cloud rain reduction operation analysis system. Figure 2 As shown, it is a structural schematic diagram of a possible convective cloud rain reduction operation analysis system provided in an embodiment of the present application. Figure 2 In the present invention, convective cloud rain reduction operation analysis system 200 includes: a processor 210 and a memory 220. The memory 220 stores a computer program executable by the processor 210. By executing the instructions stored in the memory 220, the processor 210 can perform the steps of the above-mentioned deep learning-based convective cloud rain reduction operation analysis method.

[0132] Based on the same inventive concept, an embodiment of the present application provides a computer-readable storage medium, which includes a computer program. When the computer program is run on a convective cloud rain reduction operation analysis system, the computer program is used to enable the convective cloud rain reduction operation analysis system to execute the steps of the above-mentioned convective cloud rain reduction operation analysis method based on deep learning. In some possible implementations, various aspects of the convective cloud rain reduction operation analysis method based on deep learning provided by the present application can also be implemented in the form of a program product, which includes a computer program. When the program product is run on a convective cloud rain reduction operation analysis system, the computer program is used to enable the convective cloud rain reduction operation analysis system to execute the steps of the above-mentioned convective cloud rain reduction operation analysis method based on deep learning. For example, the convective cloud rain reduction operation analysis system can execute the following steps: Figure 1 Follow the steps shown in .

[0133] In the technical solutions involved in the above-mentioned embodiments of the present invention, whether it is performing comparison calculations of multi-dimensional features or constructing composite parameters, if there are problems caused by significant differences in the number of dimensions, dimensional units and semantic meanings of different features, technical personnel in this field, based on their professional knowledge and past practical experience, are fully able to understand that these differences need to be properly handled so that the calculation results are accurate and comparable, and avoid situations such as logical confusion and unclear mathematical meaning.

[0134] In detail, when faced with features with different numbers of dimensions, in order to accurately calculate the similarity, matching degree or feature distance between different features, technical personnel in this field can use a variety of strategies, including but not limited to feature selection, feature extraction, and kernel function processing.

[0135] When processing the comparison of multi-dimensional features, in order to achieve comparable alignment of feature spaces, those skilled in the art may adopt a variety of existing common technical means, including but not limited to standardization preprocessing, mapping conversion, and space projection.

[0136] In the process of constructing composite parameters (such as loss function values), different parameter items often have different dimensions. Those skilled in the art can adopt normalization processing or an adaptive weight allocation mechanism based on distribution characteristics.

[0137] The general technical approaches described above for solving the feature matching and loss balancing problems are common knowledge in the field. These techniques have been fully validated and widely used in numerous practical applications, and those skilled in the art can skillfully and flexibly apply these methods to address similar dimensional discrepancies.

[0138] The formulas and calculation processes involved in the embodiments of the present application, whether used for multidimensional feature comparison or composite loss function construction, strictly follow the principle of dimensional correspondence. The variables in each formula have clear and definite physical meanings, and their operation logic is also fully consistent with basic mathematical and physical logic. The operation results must be the reasonable results expected by this application. Those skilled in the art have the ability to comprehensively apply the above-mentioned general technical means according to specific data conditions and business needs, and effectively solve the various problems caused by the number of dimensions, dimensional differences, etc. in the multidimensional feature comparison calculation and composite loss function construction in the embodiments, and ensure the accuracy, reliability and feasibility of the technical solution of the present invention.

Claims

1. A deep learning-based convective cloud rain reduction analysis method, characterized in that: include: Acquire a convective cloud radar echo extrapolation dataset of a target area, wherein the convective cloud radar echo extrapolation dataset includes radar echo intensity distribution data of multiple time steps and corresponding convective cloud spatial distribution data; performing sparse sampling and spatiotemporal alignment processing on the convective cloud radar echo extrapolation dataset to generate a spatiotemporally continuous radar echo extrapolation reconstruction dataset; Calling a spatiotemporal sequence prediction model to perform cyclic radar echo extrapolation processing on the radar echo extrapolation reconstruction dataset to generate an extrapolation result within a preset time range, wherein the extrapolation result includes the spatial evolution path and intensity change trend of convective clouds; A convective cloud rain reduction operation strategy is generated based on the extrapolation result, and the convective cloud rain reduction operation strategy is sent to a rain reduction operation service system for operation processing. The convective cloud rain reduction operation strategy includes operation point location, operation timing, and operation status evaluation characteristics.

2. The method according to claim 1, characterized in that The step of performing sparse sampling and spatiotemporal alignment processing on the convective cloud radar echo extrapolation dataset to generate a spatiotemporally continuous radar echo extrapolation reconstruction dataset includes: performing time alignment processing on radar echo intensity distribution data of multiple time steps in the convective cloud radar echo extrapolation dataset according to a preset time resolution threshold to obtain a radar echo intensity distribution sequence with uniform time intervals; Performing spatial interpolation processing on each radar echo intensity distribution data in the radar echo intensity distribution sequence with uniform time intervals, eliminating data missing areas in radar detection blind spots, and obtaining spatially continuous radar echo intensity distribution data; determining a spatial sampling interval based on a moving speed threshold of the convective cloud, and sparsely sampling the spatially continuous radar echo intensity distribution data according to the spatial sampling interval to generate sparsely sampled radar echo intensity distribution data; The sparsely sampled radar echo intensity distribution data and the convective cloud spatial distribution data are spatially superimposed to generate a temporally and spatially continuous radar echo extrapolation and reconstruction data set.

3. The method according to claim 1, characterized in that The calling of the spatiotemporal sequence prediction model to perform cyclic radar echo extrapolation processing on the radar echo extrapolation reconstruction data set to generate an extrapolation result within a preset time range includes: Inputting the spatiotemporally continuous radar echo extrapolation and reconstruction data set into an encoder module of a spatiotemporal sequence prediction model to generate spatiotemporal coding features of radar echo intensity distribution data; The decoder module of the spatiotemporal sequence prediction model performs time-step recursive decoding processing on the spatiotemporal coding features to generate radar echo intensity distribution prediction data for each time step within a preset time range; Performing spatial continuity constraint processing on the radar echo intensity distribution prediction data of each time step within the preset time range to generate spatially smoothed radar echo intensity distribution prediction data; According to the superposition result of the spatially smoothed radar echo intensity distribution prediction data and the convective cloud spatial distribution data, the spatial evolution path and intensity change trend of the convective cloud are determined to generate the extrapolation result.

4. The method according to claim 3, characterized in that The training method of the spatiotemporal sequence prediction model includes: Acquire a historical convective cloud radar echo extrapolation dataset and corresponding real extrapolation result data, perform spatiotemporal alignment and noise filtering on the historical convective cloud radar echo extrapolation dataset to generate a training dataset; Building an initial spatiotemporal sequence prediction model, the initial spatiotemporal sequence prediction model including a spatiotemporal convolutional encoder, a time recursive decoder, and a spatial continuity constraint module; Inputting the training data set into the initial spatiotemporal sequence prediction model, extracting multi-scale spatiotemporal features of radar echo intensity distribution training data in the training data set through the spatiotemporal convolutional encoder, and generating predicted radar echo intensity distribution data based on the multi-scale spatiotemporal features through the time recursive decoder; Calling the spatial continuity constraint module to perform spatial smoothing constraints on the predicted radar echo intensity distribution data, and calculating the spatiotemporal consistency loss between the predicted radar echo intensity distribution data and the true extrapolation result data; Parameters of the initial spatiotemporal sequence prediction model are optimized based on the spatiotemporal consistency loss until the spatiotemporal consistency loss converges, thereby obtaining a trained spatiotemporal sequence prediction model.

5. The method according to claim 1, wherein Generating a convective cloud rain reduction operation strategy based on the extrapolation result includes: Extracting the spatial evolution path and intensity change trend of the convective clouds from the extrapolated results, and determining the movement direction and coverage area of ​​the convective clouds within a preset time range; generating an operation opportunity according to a change trend of the radar echo intensity within the coverage area, wherein the operation opportunity includes an operation start time and an operation duration; Calculating the location of the rain reduction operation point based on the movement direction and the coverage area, wherein the location of the operation point includes the area with the maximum gradient change and a preset interception point on the movement path of the convective cloud core; Generate an operation status assessment feature based on the maximum gradient change area and the preset interception point, wherein the operation status assessment feature includes an operation coverage matching degree and an operation effect sustainability index; The convective cloud rain reduction operation strategy is generated by integrating the operation timing, operation point location and operation status assessment characteristics.

6. The method according to claim 5, characterized in that The calculating the rain reduction operation point location based on the moving direction and the coverage area includes: extracting a centroid coordinate sequence of the core area of ​​the convective cloud at consecutive time steps within a preset time window according to the spatial evolution path of the convective cloud, and calculating a moving direction vector and an average moving rate based on the centroid coordinate sequence; Performing a spatial projection operation on the boundary coordinates of the coverage area and the moving direction vector to determine the coordinates of the geometric center point of the downstream protection area, and generating a polar coordinate system with the geometric center point as the origin; Generating a radial detection path along the moving direction vector in the polar coordinate system, calculating the time evolution step length on each path based on the average moving rate, and generating a spatiotemporally correlated gridded detection point array; Performing radar echo intensity gradient field analysis on each detection point in the gridded detection point array to extract the vertical integrated liquid water content change rate and horizontal echo intensity gradient modulus of each detection point within a preset time step; The vertically integrated liquid water content change rate and the horizontal echo intensity gradient modulus are dimensionlessly normalized to generate a composite impact factor parameter. Detection points whose composite impact factor parameters exceed the target threshold are selected as candidate operation points. Generating a density clustering model based on the spatial distribution density of the candidate operation points, extracting cluster center coordinates through the density clustering model and performing position offset compensation according to the geometric topological characteristics of the downstream protection area to generate the location of the rain reduction operation point; The method further comprises: The projection component of the moving direction vector of each operation point in the rain reduction operation point position is checked for correlation with the coverage area parameter, and after eliminating redundant points in spatial distribution, an optimized operation point position coordinate sequence is output.

7. The method according to claim 5, characterized in that Generating the operation status assessment feature according to the maximum gradient change area and the preset interception point includes: Extracting the spatiotemporal evolution parameters of the area with the largest gradient change from the extrapolated results, including the echo intensity gradient modulus time series, the spatial integral value of the liquid water content, and the curvature radius of the core movement path; A three-dimensional spherical coordinate system is constructed within the spatial range of the preset interception point, and the cumulative value of the echo intensity flux within the preset distance upstream of the interception point and the dissipation rate attenuation coefficient within the preset distance downstream are calculated; Performing time phase matching processing on the echo intensity gradient modulus time series and the dissipation rate attenuation coefficient to generate a time-varying correlation matrix and extract the principal component eigenvector; A dimensionless fusion operation is performed on the spatial integral value of the liquid water content and the cumulative value of the echo intensity flux to obtain an operation energy intervention efficiency index, and a path deviation correction factor is calculated based on the curvature radius of the core moving path; Constructing a multidimensional state space based on the principal component eigenvector, mapping the operation energy intervention efficiency index and the path deviation correction factor into the state space to generate a state trajectory curve; Extract the time parameters and spatial coordinates corresponding to the extreme curvature points of the state trajectory curve, and generate a coverage matching evaluation matrix based on the operation time window of the preset interception point; Performing singular value decomposition on the coverage matching evaluation matrix, extracting the eigenvector corresponding to the maximum singular value as the operation coverage matching degree, and calculating the stability index of the eigenvector evolving over time as the operation effect sustainability parameter; The operation coverage matching degree and the operation effect persistence parameter are linearly weighted and synthesized to generate an operation status evaluation feature vector in a comprehensive quantitative form.

8. The method according to claim 1, characterized in that The method further comprises: Real-time acquisition of operation execution data fed back by the rain reduction operation service system, including the actual operation point location and radar echo monitoring data after the operation; Performing a matching analysis between the actual operation point position and the operation point position in the convective cloud rain reduction operation strategy to generate a position deviation correction feature; generating an effect deviation correction feature based on a difference between the radar echo monitoring data after the operation and the extrapolated result; Based on the position deviation correction feature and the effect deviation correction feature, the parameters of the spatiotemporal sequence prediction model are adjusted online.

9. The method according to claim 1, characterized in that The step of sending the convective cloud rain reduction operation strategy to the rain reduction operation service system for operation processing includes: generating a geographic coordinate instruction according to the location of the operation point in the convective cloud rain reduction operation strategy, wherein the geographic coordinate instruction includes latitude and longitude information and an operation altitude range; Converting the operation opportunity into a time control instruction for the operation equipment, wherein the time control instruction includes the operation start time, operation duration and operation interval period; generating an operation resource scheduling instruction according to the operation status evaluation feature, wherein the resource scheduling instruction includes a catalyst delivery amount and an equipment deployment priority; An operation instruction set is generated based on the geographic coordinate instruction, the time control instruction, and the resource scheduling instruction, and the operation instruction set is sent to a rain reduction operation service system to perform a multi-device collaborative operation.

10. A convective cloud rain reduction operation analysis system, characterized in that: The method comprises a processor and a memory, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor is enabled to perform the steps of any one of the methods of claims 1 to 9.

Images