Intelligent forecasting method and system for meteorological data based on satellite cloud image analysis
By constructing a dynamic evolution feature matrix and an adaptive fuzzy inference algorithm, the problems of lagging capture of the dynamic evolution law of cloud system and poor regional adaptability in satellite cloud image prediction were solved, and high-accuracy prediction of severe convective weather and timely disaster warning were achieved.
Patent Information
- Application Number
- CN202611285537.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-24
- Publication Date
- 2026-11-17
AI Technical Summary
Existing meteorological data prediction technologies based on satellite cloud images suffer from problems such as lagging capture of the dynamic evolution of cloud systems, bias in the analysis of the correlation between cloud image evolution and meteorological elements, and poor adaptability to regional meteorological forecasts.
By constructing a dynamic evolution feature matrix, combining an adaptive fuzzy inference algorithm and real-time cloud parameters, an initial prediction threshold is set, and the prediction threshold is classified, stored, and adjusted in conjunction with a hierarchical prediction threshold lookup table, thereby achieving dynamic adaptive adjustment and regional adaptation of the prediction threshold.
It improved the accuracy of severe convective weather forecasts, optimized the adaptability of regional meteorological forecasts, and enhanced the accuracy and timeliness of disaster early warnings.
Smart Images

Figure CN123113047A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of meteorological data analysis and processing technology, and in particular to a method and system for intelligent forecasting of meteorological data based on satellite cloud image analysis. Background Technology
[0002] Satellite cloud imagery, as a conventional means of acquiring information about atmospheric cloud systems, provides crucial data such as cloud morphology, cloud top temperature, and cloud movement trajectories, and has become a core data source for weather forecasting. However, existing meteorological data forecasting technologies based on satellite cloud imagery still suffer from the following core shortcomings: First, the use of fixed global prediction thresholds makes it impossible to accurately capture the entire process of cloud formation and dissipation, resulting in a serious lag in capturing the dynamic evolution of cloud systems and insufficient lead time for predicting short-term extreme weather such as severe convection. Second, when there is a deviation between cloud image evolution and the correlation analysis of meteorological elements, the existing methods cannot perform parameter calibration and deviation correction in a timely manner, resulting in a significant decrease in the confidence of the prediction results. Third, fixed global thresholds cannot adapt to the geographical features and meteorological characteristics of different regions, and cannot optimize thresholds for different terrains such as mountains, plains, and hills, resulting in extremely poor adaptability of regional weather forecasts.
[0003] In summary, existing technologies suffer from several technical problems, including the use of fixed global prediction thresholds leading to lags in capturing the dynamic evolution of cloud systems, the inability to promptly calibrate when deviations occur in the correlation analysis between cloud image evolution and meteorological elements, and low adaptability of regional meteorological forecasts. Summary of the Invention
[0004] The technical problem to be solved and the technical task proposed by this invention is to improve and refine existing technical solutions, and to provide an intelligent forecasting method and system for meteorological data based on satellite cloud image analysis, so as to improve the accuracy of severe convective weather forecasting, optimize the adaptability of regional meteorological forecasts, and improve the timeliness and reliability of disaster early warning. To this end, this invention adopts the following technical solution.
[0005] In a first aspect, the present invention provides an intelligent forecasting method for meteorological data based on satellite cloud image analysis, comprising the following steps: 1) Obtain a satellite cloud image feature dataset, and construct a dynamic evolution feature matrix based on the satellite cloud image feature dataset to characterize the dynamic change pattern of cloud clusters throughout their entire life cycle; 2) Set an initial prediction threshold that matches the meteorological characteristics of the target area by using historical meteorological data and geographical feature parameters of the target area; 3) An adaptive fuzzy inference algorithm is adopted, and real-time cloud morphology parameters, cloud top brightness temperature jump values and terrain uplift coefficients are input. Dynamic correction coefficients are calculated and configured, and the initial prediction threshold is synchronously corrected through the dynamic correction coefficients. 4) The initial forecast thresholds after synchronous correction are classified and stored according to meteorological element type, regional division and forecast lead time, and a graded forecast threshold comparison table is proposed. 5) Set up a cloud map dynamic evolution channel that has a mapping relationship with the dynamic evolution feature matrix, and use the hierarchical prediction threshold reference table as a benchmark to complete the linkage adjustment of disaster warning level and meteorological element data, and output the disaster warning level and the corresponding meteorological element prediction results.
[0006] This technical solution overcomes the problem that fixed thresholds cannot reflect the dynamic evolution of cloud systems by constructing a dynamic evolution feature matrix to capture the full life cycle patterns of cloud clusters. It employs adaptive fuzzy inference combined with real-time cloud system parameters, cloud top brightness-temperature jumps, and topographic lifting coefficients to dynamically correct the initial threshold, achieving real-time adaptive adjustment of the prediction threshold and improving the accuracy of severe convective weather prediction. A hierarchical prediction threshold lookup table enables refined classification, storage, and retrieval based on meteorological elements, regions, and timeframes, optimizing the adaptability of regional meteorological forecasts. Through the linkage and adjustment of the cloud image dynamic evolution channel and the hierarchical prediction threshold lookup table, the process barriers between meteorological forecasting and disaster early warning are broken down, improving the accuracy and timeliness of early warnings.
[0007] As a preferred technical means: In step 1), the satellite cloud image feature dataset includes spatiotemporal resolution data of thunderstorm cloud clusters, local precipitation cloud systems, and strong convective squall lines; the satellite cloud image feature dataset is collected through cloud top temperature monitoring layer, cloud system morphology monitoring layer, and atmospheric circulation monitoring layer, and transmitted to the ground receiving end via transmission network to complete the acquisition.
[0008] This technical solution, by setting up three monitoring layers—cloud top temperature, cloud morphology, and atmospheric circulation—and a dedicated transmission network, can acquire high spatiotemporal resolution data containing various strong convective cloud systems. This provides a rich and reliable data foundation for the subsequent construction of dynamic evolution feature matrices, improving the comprehensiveness and accuracy of predictions.
[0009] As a preferred technical means, the specific method for constructing the dynamic evolution feature matrix is as follows: A first feature index is set, which includes cloud image morphology features, motion vector features, and intensity change features. Based on the satellite cloud image feature dataset, each intensity evolution sub-matrix in the dynamic evolution feature matrix is determined by associating and mapping with the first feature index and weighting and fusing it. A second feature index is set, which includes cloud top temperature gradient, cloud expansion rate, and vertical airflow intensity. Based on the satellite cloud image feature dataset, each dynamic evolution sub-matrix in the dynamic evolution feature matrix is determined by associating and mapping with the second feature index and weighting fusion.
[0010] This technical solution sets first characteristic indicators (morphology, motion, intensity) and second characteristic indicators (temperature gradient, expansion rate, vertical airflow intensity) respectively, and uses correlation mapping and weighted fusion methods to extract the intensity evolution characteristics and dynamic evolution characteristics of cloud clusters respectively. This enables the dynamic evolution feature matrix to more comprehensively and accurately depict the entire life cycle change pattern of cloud clusters, providing high-quality feature inputs for subsequent threshold correction and early warning tuning.
[0011] As a preferred technical means: In step 5), the specific method for setting the cloud map dynamic evolution channel that has a mapping relationship with the dynamic evolution feature matrix is as follows: Connect to the historical meteorological event database and match the corresponding historical cloud map events using the dynamic evolution feature matrix; The duration of potential meteorological events is determined by combining the evolution cycle distribution of matched historical cloud map events with the current cloud system change rate. Using historical cloud map events that match the meteorological events predicted by the current cloud map features in terms of impact area, season of occurrence, and intensity level, the impact range of potential meteorological events is determined. Based on the duration and impact range of the potential meteorological events, the cloud map dynamic evolution channel is constructed and set.
[0012] This technical solution connects to a historical meteorological event database and combines it with a dynamic evolution feature matrix to match historical cloud map events. It uses historical evolution cycles and current rates of change to predict event duration, and uses historical impact areas, seasons, and intensity levels to predict the scope of event impact. This enables the dynamic evolution channel of cloud maps to make predictions based on the patterns of similar historical events, thus improving the reliability and interpretability of the predictions.
[0013] As a preferred technical means, the method further includes: Based on the dynamic evolution feature matrix, and combined with the meteorological element association tree and the current atmospheric stability parameters, the influence weights of potential meteorological elements are determined. Based on the matching degree between the triggering conditions of element mutations in similar historical meteorological events and the geographical feature parameters associated with the meteorological characteristics of the target area, and combined with the influence weight of the potential meteorological elements, the transmission path of the potential meteorological elements is determined. Based on the influence weights of the potential meteorological elements and the transmission paths of the potential meteorological elements, the evolution parameters of the cloud map dynamic evolution channel are optimized.
[0014] This technical solution determines the transmission path of meteorological elements by introducing the matching degree analysis of the triggering conditions of element mutations and geographical feature parameters, and optimizes the evolution parameters of the cloud image dynamic evolution channel accordingly. This enables the channel to more accurately simulate the evolution process of cloud clusters in specific geographical environments, thereby improving the spatial precision and physical consistency of the prediction.
[0015] As a preferred technical means, the method for determining the influence weight of the potential meteorological elements is as follows: Based on each intensity evolution sub-matrix and each dynamic evolution sub-matrix in the dynamic evolution feature matrix, the coupling characteristics of meteorological elements are determined by combining the meteorological element association tree. The influence weights of the potential meteorological elements are determined by using the correlation coefficients of the coupling characteristics of the meteorological elements and combining them with the current atmospheric stability parameters.
[0016] This technical solution extracts the coupling characteristics of meteorological elements by combining the intensity evolution submatrix and the dynamic evolution submatrix with the meteorological element association tree, and then determines the influence weight by combining the correlation coefficient and atmospheric stability parameters. It can quantitatively assess the importance of different meteorological elements under the current atmospheric conditions, and provide a scientific basis for determining the transmission path and optimizing the evolution parameters.
[0017] As a preferred technical means: before completing the linkage adjustment between disaster warning level and meteorological element data in step 5), a step of cross-validation of the cloud image dynamic evolution channel and meteorological element association tree is also included, specifically: When the deviation between the prediction confidence level output by the cloud image dynamic evolution channel and the prediction confidence level output by the meteorological element association tree meets the preset deviation tolerance, it is determined that the two prediction confidence levels are consistent, and the subsequent disaster warning level and meteorological element data linkage adjustment steps are executed.
[0018] This technical solution introduces a dual-model cross-validation mechanism using a dynamic cloud image evolution channel and a meteorological element association tree. Linked tuning is only performed when the prediction confidence levels of the two independent models are consistent, effectively avoiding the bias that may occur with a single model and improving the credibility of prediction results and the reliability of early warning decisions.
[0019] As a preferred technical means, the cross-validation step further includes: When the deviation between the prediction confidence level output by the cloud image dynamic evolution channel and the prediction confidence level output by the meteorological element association tree exceeds the preset deviation tolerance, the SHAP value of the Fast connection layer of the cloud image dynamic evolution channel is sent to the Slow connection layer through a lateral connection. Using the SHAP value as the attention weight, feature response modulation is applied to the Slow connection layer. After modulation, the prediction confidence of the cloud image dynamic evolution channel and the meteorological element association tree is recalculated. The above lateral connection and feature response modulation steps are repeated until the confidence deviation between the two meets the preset deviation tolerance.
[0020] When the prediction confidence of the two models is inconsistent, this technical solution uses the SHAP value as the attention weight to modulate the feature response of the Slow connection layer, realizing the adaptive tuning of the model. Through iterative optimization, the prediction results of the two models gradually converge to consistency, significantly improving the robustness and prediction consistency of the model under complex weather conditions.
[0021] Secondly, the present invention provides a meteorological data intelligent prediction system based on satellite cloud image analysis, used to implement the aforementioned meteorological data intelligent prediction method based on satellite cloud image analysis, the system comprising: Dataset acquisition module: used to acquire satellite cloud image feature datasets; Dynamic evolution feature matrix construction module: connected to the dataset acquisition module, used to construct a dynamic evolution feature matrix that characterizes the dynamic change law of the cloud cluster throughout its entire life cycle based on the satellite cloud image feature dataset; Initial prediction threshold setting module: used to set an initial prediction threshold that matches the meteorological characteristics of the target area based on historical meteorological data and geographical feature parameters of the target area; Synchronous correction module: connected to the initial prediction threshold setting module, used to use an adaptive fuzzy inference algorithm, input real-time cloud morphology parameters, cloud top brightness temperature jump value and terrain lifting coefficient, calculate and configure dynamic correction coefficient, and synchronously correct the initial prediction threshold through the dynamic correction coefficient; Classification and storage module: connected to the synchronization correction module, used to classify and store the initial prediction thresholds after synchronization correction according to meteorological element type, regional division and prediction time, and to formulate a graded prediction threshold comparison table; Linkage Adjustment Module: Connected to the dynamic evolution feature matrix construction module and the classification storage module, it is used to set up cloud map dynamic evolution channels that have a mapping relationship with the dynamic evolution feature matrix, and to complete the linkage adjustment of disaster warning level and meteorological element data based on the hierarchical prediction threshold reference table.
[0022] This technical solution utilizes a dynamic evolution feature matrix construction module to systematically integrate and quantify the features of satellite cloud image datasets. This allows for the accurate capture of the dynamic changes throughout the entire lifecycle of cloud formation and dissipation, addressing the shortcomings of existing technologies such as delayed capture of cloud evolution patterns and fragmented feature representation. It provides ample lead time and a reliable analytical foundation for predicting short-term extreme weather events like severe convection. Through the coordinated operation of an initial prediction threshold setting module and a synchronous correction module, an initial prediction threshold is first set based on historical meteorological data and geographical features of the target area, aligning with the region's background meteorological characteristics. Then, an adaptive fuzzy inference algorithm, combined with three core real-time parameters—real-time cloud morphology, cloud top brightness-temperature transitions, and topographic lifting coefficient—achieves synchronous dynamic correction of the prediction threshold. This completely overcomes the limitations of existing technologies that rely on fixed global prediction thresholds, effectively solving the problem of poor regional prediction adaptability under different terrains and meteorological scenarios, and significantly reducing the false alarm and missed alarm rates in extreme weather predictions. Through a categorized storage module, dynamically corrected prediction thresholds are categorized and managed in a multi-dimensional manner according to meteorological element type, regional division, and prediction timeliness. A standardized hierarchical prediction threshold lookup table is established, enabling accurate and rapid retrieval of thresholds under different prediction scenarios. This provides a unified and standardized quantitative benchmark for meteorological forecasting and warning level determination, solving the problems of chaotic threshold retrieval and inconsistent warning judgment standards in existing technologies. Through a linkage adjustment module, a cloud map dynamic evolution channel with a direct mapping relationship to the dynamic evolution feature matrix is constructed. Combined with the standardized hierarchical prediction threshold lookup table, the linkage adjustment between disaster warning levels and meteorological element data is completed, breaking down the process barriers between meteorological element prediction and disaster warning issuance. This achieves real-time and standardized adaptation of prediction results and warning levels, significantly improving the timeliness and accuracy of meteorological disaster warnings, and directly linking warning results with core meteorological element data, ensuring the traceability and operational practicality of warning information.
[0023] As a preferred technical means, the system further includes: Cross-validation module: It is connected to the cloud image dynamic evolution channel and the preset meteorological element association tree in the linkage tuning module respectively. It is used to calculate the deviation between the prediction confidence level output by the cloud image dynamic evolution channel and the prediction confidence level output by the meteorological element association tree, and to determine whether the preset deviation tolerance is met. Feature modulation module: connected to the cross-validation module, used to send the SHAP value of the Fast connection layer of the cloud map dynamic evolution channel to the Slow connection layer through a lateral connection when the deviation exceeds the preset deviation tolerance, use the SHAP value as the attention weight to perform feature response modulation on the Slow connection layer, and feed the modulation result back to the cross-validation module to recalculate the prediction confidence until the deviation meets the preset deviation tolerance.
[0024] This technical solution adds a cross-validation module and a feature modulation module to realize real-time comparison and adaptive tuning of the prediction results of the two models within the system. When the prediction confidence is inconsistent, the SHAP attention mechanism is used to modulate the feature response of the Slow connection layer and iteratively optimize it, so that the system has self-correction capability and significantly improves the stability and reliability of the prediction results under complex meteorological conditions.
[0025] Beneficial effects: 1. By constructing a two-dimensional dynamic evolution feature matrix, it is possible to capture the entire life cycle change pattern of cloud clusters from formation to dissipation, overcoming the shortcomings of fixed threshold methods in capturing the dynamic evolution pattern of cloud systems, and is conducive to extending the forecast lead time for short-term extreme weather such as severe convection.
[0026] 2. An adaptive fuzzy inference algorithm is adopted, which combines real-time cloud morphology parameters, cloud top brightness temperature jump values and topographic lifting coefficients to configure dynamic correction coefficients. The initial prediction threshold is synchronously corrected, realizing dynamic adaptive adjustment of the prediction threshold. This solves the problem of poor regional adaptability of fixed global thresholds and improves the accuracy of severe convective weather prediction.
[0027] 3. By formulating a graded forecast threshold comparison table, and classifying and storing the forecasts according to meteorological element type, regional division and forecast time, refined and regionalized meteorological forecast threshold management is achieved, optimizing the adaptability of regional meteorological forecasts under different terrains and time periods, and providing a standardized quantitative benchmark for warning level determination.
[0028] 4. By setting up a dynamic evolution channel for cloud maps and linking it with a graded prediction threshold comparison table, the coordinated adaptation and adjustment of disaster warning levels and meteorological element data were achieved, breaking down the process barriers between meteorological forecasting and disaster warning, and improving the accuracy and timeliness of disaster warning.
[0029] 4. By introducing dual-model cross-validation and a feature response modulation mechanism based on SHAP values, adaptive tuning and closed-loop optimization can be performed when the prediction confidence is inconsistent. This solves the problem of not being able to calibrate in time when there are deviations in the correlation analysis between cloud image evolution and meteorological elements, and further improves the reliability and robustness of the prediction results. Attached Figure Description
[0030] Figure 1 This is a flowchart illustrating the intelligent meteorological data prediction method based on satellite cloud image analysis according to the present invention.
[0031] Figure 2 This is a schematic diagram of the intelligent meteorological data prediction system based on satellite cloud image analysis according to the present invention. Detailed Implementation
[0032] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings.
[0033] Example 1 A method for intelligent forecasting of meteorological data based on satellite cloud image analysis is provided, which includes the following steps.
[0034] S1: Obtain satellite cloud image feature dataset, and construct a dynamic evolution feature matrix based on the satellite cloud image feature dataset to characterize the dynamic change pattern of cloud clusters throughout their entire life cycle; Specifically, the dynamic evolution feature matrix is used to integrate and analyze multi-dimensional dynamic features in satellite cloud images. Through a matrix structure, it quantifies and correlates features such as cloud morphology, motion vectors, and temperature changes, thereby capturing the entire lifecycle changes of clouds from formation to dissipation. Constructing the dynamic evolution feature matrix requires preprocessing the satellite cloud image data, extracting key features, and then weighting and fusing these key features to form a mathematical structure that reflects the dynamic changes of clouds. This mathematical structure reflecting the dynamic changes of clouds is defined as the dynamic evolution feature matrix. The specific implementation process includes: S1.1: Acquisition of Satellite Cloud Image Feature Dataset The satellite cloud image feature dataset contains spatiotemporal resolution features of thunderstorm cloud clusters, local precipitation cloud systems, and strong convective squall lines. Based on the cloud top temperature monitoring layer, cloud system morphology monitoring layer, and atmospheric circulation monitoring layer, it is connected to the ground receiving end through a transmission network to receive the satellite cloud image feature dataset.
[0035] The satellite cloud image feature dataset is a database containing various cloud system features to support weather forecasting. It covers different types of cloud systems, including thunderstorm clouds, localized precipitation clouds, and strong convective squall lines, and records the spatiotemporal resolution characteristics of these cloud systems, including cloud top temperature, cloud morphology, and movement trajectories. This feature data is acquired through satellite monitoring and transmitted in real-time to ground receivers via a transmission network. Specifically, the cloud top temperature monitoring layer monitors temperature changes at cloud tops, the cloud morphology monitoring layer monitors changes in cloud morphology, and the atmospheric circulation monitoring layer monitors dynamic changes in atmospheric circulation.
[0036] The satellite cloud imagery dataset is obtained through a three-layered monitoring system: cloud top temperature monitoring, cloud morphology monitoring, and atmospheric circulation monitoring. This data includes key parameters such as cloud top temperature, cloud morphology, and movement trajectory. The dataset is then transmitted in real-time to ground-based receivers via a transmission network to determine the satellite cloud imagery dataset. The cloud top temperature monitoring layer detects rapid changes in cloud top temperature, often an early indicator of thunderstorm cloud development. The cloud morphology monitoring layer captures changes in cloud shape, including expansion and contraction. The atmospheric circulation monitoring layer monitors dynamic changes in atmospheric circulation, including rising and descending air currents. The real-time transmission and reception of this satellite cloud imagery dataset provides high-resolution dynamic data support for subsequent weather forecasting.
[0037] S1.2: Construction of the dynamic evolution feature matrix.
[0038] Based on the aforementioned satellite cloud image feature dataset, a dynamic evolution feature matrix is constructed by setting dual-dimensional feature indicators, associating and mapping them, and weighted fusion. The specific implementation steps are as follows.
[0039] S1.2.1: Set the first feature index and construct the intensity evolution sub-matrix.
[0040] A first feature index is set, which includes cloud image morphology features, motion vector features, and intensity change features. Based on the satellite cloud image feature dataset, the intensity evolution sub-matrices in the dynamic evolution feature matrix are determined by using correlation mapping and weighted fusion with the first feature index.
[0041] The first feature index is a set of key parameters used to describe cloud image features, including cloud morphology features, motion vector features, and intensity change features. These are used to extract data related to the dynamic evolution of cloud clusters from the satellite cloud image feature dataset. Motion vector features include the cloud system's movement speed, and intensity change features include cloud cover. Association mapping refers to matching specific data in the satellite cloud image feature dataset with the first feature index to ensure data accuracy and relevance. Weighted fusion combines multiple feature indices by assigning different weights to each feature. The intensity evolution submatrix is part of the dynamic evolution feature matrix and is used to describe the characteristics of cloud cluster intensity changes.
[0042] The first feature index is defined, including cloud image morphology features, motion vector features, and intensity change features. Data related to these features is extracted from the satellite cloud image feature dataset, and the data is matched with the first feature index through association mapping. Cloud image morphology features can be extracted from the shape and area data of cloud images, and motion vector features can be extracted from the cloud cluster's movement trajectory data. These feature data are combined through weighted fusion to generate an intensity evolution sub-matrix, which can accurately reflect the changing trend of cloud cluster intensity. Through the above steps, key features are extracted and fused to generate a dynamic data structure that can accurately describe the changes in cloud cluster intensity.
[0043] S1.2.2: Set the second characteristic index and construct the dynamic evolution sub-matrix. A second feature index is set, including cloud top temperature gradient, cloud expansion rate, and vertical airflow intensity. Based on the satellite cloud image feature dataset, the dynamic evolution sub-matrices in the dynamic evolution feature matrix are determined by using correlation mapping and weighted fusion with the second feature index.
[0044] The second feature index is a set of key parameters used to describe the dynamic characteristics of cloud clusters, including cloud top temperature gradient, cloud cluster expansion rate, and vertical airflow intensity, which are used to extract data related to the dynamic evolution of cloud clusters from satellite cloud image feature datasets; the dynamic evolution sub-matrix is a part of the dynamic evolution feature matrix, used to describe the characteristics of cloud cluster dynamic changes, including cloud cluster expansion, contraction, rising, and sinking.
[0045] A second set of feature indicators is defined, including cloud top temperature gradient, cloud expansion rate, and vertical airflow intensity. The cloud top temperature gradient reflects the vertical variation of cloud top temperature, the cloud expansion rate reflects the rate of change of cloud volume, and the vertical airflow intensity reflects the rising or falling speed of airflow within the cloud. Data related to these features are extracted from the satellite cloud image feature dataset, and the data is matched with the second set of feature indicators through association mapping. The cloud top temperature gradient can be extracted from the vertical distribution data of cloud top temperature, and the cloud expansion rate can be extracted from the data on cloud volume changes. These feature data are combined through weighted fusion to generate a dynamic evolution sub-matrix, which accurately reflects the changing trend of cloud dynamics. Through the above steps, key dynamic features are extracted and fused to generate a dynamic data structure that accurately describes the dynamic changes of cloud clusters.
[0046] S1.2.3: Integrate to form a complete dynamic evolution feature matrix.
[0047] By aligning and integrating the aforementioned intensity evolution sub-matrices and dynamic evolution sub-matrices along their spatiotemporal dimensions, a complete dynamic evolution feature matrix is formed. This matrix perfectly maps to the spatiotemporal evolution characteristics of cloud clusters, enabling it to fully capture the dynamic changes throughout the entire lifecycle of cloud clusters, from formation to dissipation. This solves the problems of lagging capture of cloud system dynamic evolution patterns and fragmented feature representation in existing technologies.
[0048] S2: Set an initial forecast threshold that conforms to the regional meteorological characteristics by using historical meteorological data and geographical feature parameters.
[0049] An initial forecast threshold serves as a baseline for weather forecasting. Set based on historical meteorological data and geographic characteristic parameters, it determines whether a meteorological event will occur and its intensity. Geographic characteristic parameters include topographic altitude, latitude, and longitude, reflecting the climatic characteristics of different regions. By combining historical meteorological data and geographic characteristic parameters, an initial forecast threshold matching the regional meteorological characteristics is set, thereby improving forecast accuracy. The specific steps include: S2.1: Acquisition and organization of basic data.
[0050] Collect historical meteorological data for the target area, including cloud image observation data, measured meteorological element data, and records of disastrous meteorological events from previous years. At the same time, collect geographical feature parameters of the target area, including all parameters that can reflect the geographical characteristics of the region and affect its meteorological characteristics, such as terrain altitude, latitude, longitude, terrain slope, and surface type.
[0051] S2.2: Setting the initial prediction threshold.
[0052] Based on compiled historical meteorological data, statistical analysis is used to determine baseline characteristic thresholds for various meteorological events within the target area. Simultaneously, considering the influence of geographical features on regional meteorology, the baseline characteristic thresholds are adjusted for regional adaptability, ultimately setting initial prediction thresholds that perfectly match the meteorological characteristics of the target area. These initial prediction thresholds serve as benchmark values for determining whether a meteorological event has occurred and for classifying its intensity. Their deep adaptation to the geographical characteristics and historical meteorological patterns of the target area ensures the basic accuracy of meteorological forecasts. For example, for mountainous areas with significant topographic uplift, given their higher susceptibility to severe convective weather, the corresponding initial prediction threshold is set to a lower value to more sensitively capture precipitation, severe convection, and other meteorological events. For plains areas, the initial threshold is adjusted accordingly based on historical meteorological patterns to ensure complete adaptation to regional characteristics.
[0053] S3: Adaptive fuzzy inference algorithm is adopted. Input real-time cloud morphology parameters, cloud top brightness temperature jump value and terrain uplift coefficient, calculate and configure dynamic correction coefficient, and synchronously correct the initial prediction threshold through dynamic correction coefficient.
[0054] This step, based on the real-time evolution of cloud clusters and the real-time influence of regional terrain, dynamically calibrates the initial prediction threshold in real time. This addresses the problem of fixed and rigid thresholds in existing technologies, which cannot adapt to real-time changes in cloud clusters. The specific steps are as follows: S3.1: Construction of an adaptive fuzzy inference system.
[0055] This step employs an adaptive fuzzy inference system capable of achieving accurate reasoning and decision-making in meteorological forecasting scenarios with high uncertainty and fuzziness. The adaptive fuzzy inference system uses real-time cloud morphology parameters, cloud top brightness-temperature jump values, and topographic uplift coefficients as input variables, and dynamic correction coefficients as output variables. It incorporates a fuzzy inference rule base built upon professional experience in the meteorological field and statistical patterns of historical meteorological events.
[0056] Among them, the real-time cloud morphology parameters are parameters that reflect the real-time morphology of cloud clusters, such as changes in cloud shape, area, and contour. The cloud top brightness temperature jump value is the rapid change in cloud top brightness temperature per unit time, which can directly reflect the intensity of cloud cluster convection development. The topographic lifting coefficient is the intensity of the lifting effect of the topography on the airflow in the target area, which can reflect the real-time influence of the topography on cloud cluster development. The dynamic correction coefficient is a core parameter used to adjust the initial prediction threshold, which can realize the real-time synchronous correction of the initial prediction threshold.
[0057] S3.2: Calculation and configuration of dynamic correction coefficients.
[0058] The real-time cloud morphology parameters, cloud top brightness temperature jump values, and terrain uplift coefficients are input into the adaptive fuzzy inference system. The system then performs fuzzification processing, fuzzy rule inference, and defuzzification calculation to finally output accurate dynamic correction coefficients, thus completing the real-time configuration of the dynamic correction coefficients.
[0059] S3.3: Synchronous real-time correction of the initial prediction threshold.
[0060] Based on the aforementioned configured dynamic correction coefficients, the initial prediction threshold set in step S2 is synchronously and in real-time corrected using preset correction calculation rules. For example, when a large jump in cloud top brightness temperature is detected, it indicates that the cloud cluster is in a rapid development phase, significantly increasing the probability of convective outbreaks. In this case, the adaptive fuzzy inference system will output a correspondingly increased dynamic correction coefficient, thereby lowering the initial prediction threshold, improving prediction sensitivity, and capturing the development trend of severe convective weather in advance. When the cloud cluster development slows down, the system will adjust the dynamic correction coefficients accordingly and optimize the threshold to reduce the false alarm rate.
[0061] By implementing this step, the limitations of the existing technology of fixing the global threshold are completely overcome, and the prediction threshold is dynamically adapted and adjusted according to the real-time evolution of cloud clusters and the influence of regional terrain. This solves the problem of rigid threshold control and significantly improves the accuracy of prediction.
[0062] S4: The initial forecast thresholds after synchronous correction are classified and stored according to meteorological element type, regional division and forecast lead time, and a graded forecast threshold comparison table is proposed.
[0063] This step is used to standardize and systematize the dynamically corrected forecast thresholds, providing a standardized quantitative benchmark for subsequent meteorological forecasting and early warning linkage. The specific steps are as follows: S4.1: Categorized storage of the corrected threshold.
[0064] Based on the initial prediction threshold after real-time correction in step S3, three core classification dimensions are set to complete multi-dimensional classification storage. The three classification dimensions are as follows: Meteorological element type dimension: The corrected thresholds are classified according to the different types of meteorological elements such as precipitation, wind speed, temperature, thunderstorms, and hail. Regional classification dimensions: The corrected thresholds are classified according to different terrain regions such as mountainous areas, plains, hills, and coastal areas, as well as administrative divisions and gridded spatial divisions. Forecast timeliness dimension: The corrected thresholds are classified according to different forecast timeliness, such as short-term forecast, medium-term forecast, and long-term forecast.
[0065] After classifying according to the above three dimensions, the corrected predicted thresholds are stored in the corresponding database to achieve classified storage and ensure rapid retrieval and accurate use of subsequent thresholds.
[0066] Specifically, the corrected thresholds are classified according to the types of meteorological elements associated with precipitation, wind speed, and temperature; further subdivided according to the association between mountainous areas, plains, and coastal areas; and classified according to the forecast lead time of short-term, medium-term, and long-term associations. For example, for precipitation forecasts in mountainous areas, the short-term forecast threshold is set to a lower value to improve the timeliness of the warning. The graded forecast threshold comparison table can provide a standardized basis for accurate forecasts in different scenarios, improving the flexibility and accuracy of forecasts.
[0067] S4.2: Formulation of a hierarchical prediction threshold comparison table The tiered forecast threshold lookup table is a systematically organized table that categorizes forecast thresholds corresponding to different meteorological element types, regional divisions, and forecast lead times. This allows for the rapid retrieval and application of appropriate thresholds based on different meteorological forecast scenarios. Categorized storage involves classifying the corrected forecast thresholds according to meteorological element types associated with precipitation, wind speed, and temperature; regional divisions associated with mountainous areas, plains, and coastal areas; and forecast lead times associated with short-term, medium-term, and long-term forecasts, and storing these categories in a database for quick retrieval. Based on the categorized storage of corrected forecast thresholds, and categorized by combination of meteorological element types, regional divisions, and forecast lead times, the corresponding forecast thresholds for different scenarios are systematically organized to form a standardized tiered forecast threshold lookup table. This tiered forecast threshold lookup table comprehensively covers forecast thresholds for different meteorological forecast scenarios, providing a standardized and unified basis for judgment in different scenarios. It enables rapid and accurate retrieval of thresholds for different scenarios, significantly improving the flexibility and accuracy of forecasts.
[0068] S5: Set up a cloud map dynamic evolution channel that has a mapping relationship with the dynamic evolution feature matrix, and use the hierarchical prediction threshold reference table as a benchmark to complete the linkage adjustment between disaster warning level and meteorological element data, and output the disaster warning level and the corresponding meteorological element prediction results.
[0069] The cloud map dynamic evolution channel is a model mapped to the dynamic evolution feature matrix. It is used to simulate and predict the dynamic change process of cloud maps. By analyzing historical meteorological events and current cloud map characteristics, it predicts the duration and impact range of potential meteorological events. Linkage adjustment refers to linking disaster warning levels with meteorological element data to ensure the accuracy and timeliness of warning information.
[0070] The cloud image dynamic evolution channel is set up to effectively predict the development trend of potential meteorological events. The disaster warning level and meteorological element data are linked and adjusted by a graded prediction threshold comparison table. According to the graded prediction threshold comparison table and the prediction results of the cloud image dynamic evolution channel, the disaster warning level is determined. Specifically, when heavy precipitation is predicted in a certain area, the warning level is raised according to the graded prediction threshold comparison table, and the warning information is issued in a timely manner to improve the timeliness and pertinence of the warning.
[0071] The goal of this step is to achieve accurate prediction of cloud evolution trends, confidence verification of prediction results, and standardized linkage between meteorological forecasting and disaster early warning. Specifically, it includes the following steps.
[0072] S5.1: Set up a cloud map dynamic evolution channel that has a mapping relationship with the dynamic evolution feature matrix.
[0073] Connect to the historical meteorological event database and combine it with the dynamic evolution feature matrix to determine matching historical cloud map events; combine the evolution cycle distribution of matching historical cloud map events with the current cloud system change rate to determine the duration of potential meteorological events; determine the impact range of potential meteorological events by matching historical cloud map events that affect the same area, the same season, and the same intensity level; set up cloud map dynamic evolution channels based on the duration and impact range of potential meteorological events.
[0074] The historical meteorological event database is a database that stores data on past meteorological events, including information such as event type, time, location, and intensity. Matching historical cloud map events involves comparing the data in the database with a dynamic evolution feature matrix to find historical events with similar characteristics to the current cloud map. The duration of a potential meteorological event is a prediction of the possible duration of the current meteorological event based on the evolution cycle of the matched historical cloud map events and the current rate of cloud system change. The impact range of a potential meteorological event is a prediction of the potential impact range of the current meteorological event based on the impact area, season, and intensity level of the matched historical cloud map events. The cloud map dynamic evolution channel is used to simulate and predict the dynamic changes in cloud maps, combining the duration and impact range of potential meteorological events to support disaster early warning.
[0075] The current dynamic evolution feature matrix is connected to a historical meteorological event database via a data interface to obtain historical data. Using feature data from the dynamic evolution feature matrix, including cloud morphology, motion vectors, and intensity changes, the database is searched for historical events with similar features to the current cloud image. If the current cloud image shows a rapidly developing thunderstorm cloud, similar thunderstorm events are searched in the database. The evolution cycle distribution of matching historical cloud image events is analyzed, and combined with the current cloud system's rate of change, including the cloud expansion rate and movement speed, the possible duration of the current meteorological event is predicted.
[0076] Based on the impact area, season, and intensity level of historical cloud image events, the potential impact range of the current meteorological event is predicted. If historical data shows that similar events mainly affect mountainous and plain areas in summer, and the current event also occurs in summer with a similar intensity level, then the current event is predicted to likely affect the same areas. Based on the duration and impact range of potential meteorological events, a dynamic cloud image evolution channel is constructed to simulate the dynamic changes of cloud clusters and predict future changes in cloud morphology, location, and intensity. By combining historical data and current cloud image characteristics, the duration and impact range of potential meteorological events are predicted.
[0077] S5.2: Optimization of evolution parameters for the dynamic evolution channel of cloud map.
[0078] The transmission path of potential meteorological elements is determined by the matching degree between the triggering conditions of abrupt changes in historical similar meteorological events in the influence weight of potential meteorological elements and the geographical feature parameters associated with regional meteorological characteristics. Based on the influence weight of potential meteorological elements and the transmission path of potential meteorological elements, the evolution parameters of the dynamic evolution channel of cloud map are optimized.
[0079] The impact weights of potential meteorological elements are used to assess the importance of different meteorological elements in a specific meteorological event. Meteorological elements include temperature, humidity, and wind speed. The triggering conditions for element mutations refer to the conditions that cause significant changes in meteorological elements in similar historical meteorological events, including sudden drops in temperature and sudden increases in wind speed. Geographical characteristic parameters refer to geographical parameters related to regional meteorological characteristics, including topographic altitude, latitude, and longitude. The transmission path of potential meteorological elements refers to the path of meteorological elements propagation in different regions and under different conditions, which is determined by analyzing historical data and geographical characteristic parameters. Evolutionary parameters refer to parameters used to describe changes in cloud clusters in the dynamic evolution channel of cloud images, including cloud cluster movement speed and expansion rate.
[0080] Analyze the triggering conditions for abrupt changes in meteorological elements in similar historical meteorological events. For example, in thunderstorm events, a sudden drop in temperature and a sudden increase in wind speed may be key triggering conditions. Assess the degree of match between these triggering conditions and the geographic characteristic parameters associated with the current regional meteorological characteristics. If the current region is mountainous, orographic lifting may enhance the triggering conditions for sudden increases in wind speed; based on the degree of match, determine the potential transmission paths of meteorological elements. If historical data shows that sudden increases in wind speed during thunderstorm events typically propagate from mountainous areas to plains, and the geographic characteristics of the current region match historical data, then the transmission path of sudden increases in wind speed can be determined to be from mountainous areas to plains.
[0081] Based on the influence weights and transmission paths of potential meteorological elements, the evolution parameters in the cloud image dynamic evolution channel are adjusted. If the transmission path of a sudden increase in wind speed is from mountainous areas to plains, and the influence weight of wind speed is high, the weight of wind speed changes in the cloud image dynamic evolution channel can be increased. By optimizing the evolution parameters, the prediction accuracy of the cloud image dynamic evolution channel for current meteorological events can be improved. If the optimized evolution parameters can more accurately simulate the movement and expansion process of cloud clusters, the prediction capability for severe convective weather can be improved. By analyzing historical data and geographical feature parameters, the transmission paths of meteorological elements are determined, and the evolution parameters of the cloud image dynamic evolution channel are optimized, thereby improving the accuracy and reliability of meteorological forecasts.
[0082] Based on the intensity evolution sub-matrices and dynamic evolution sub-matrices in the dynamic evolution feature matrix, the coupling characteristics of meteorological elements are determined by combining the meteorological element association tree; the influence weight of potential meteorological elements is determined by combining the correlation coefficient of the coupling characteristics of meteorological elements with the current atmospheric stability parameters.
[0083] Meteorological element association tree is a data structure used to represent the relationships between different meteorological elements. It displays the causal relationships and interactions between meteorological elements through a tree structure. Meteorological element coupling characteristics refer to the characteristics of the interactions between different meteorological elements. By analyzing these characteristics, we can understand the synergistic change patterns between meteorological elements. Correlation coefficient is used to measure the correlation strength between two or more meteorological elements. Atmospheric stability parameters are parameters used to describe atmospheric stability, including temperature stratification, humidity stratification, etc., reflecting the stable state of the atmosphere.
[0084] Based on the intensity evolution sub-matrix and dynamic evolution sub-matrix in the dynamic evolution feature matrix, feature data related to meteorological elements are extracted. The intensity evolution sub-matrix can provide features such as cloud top temperature change and cloud expansion rate, while the dynamic evolution sub-matrix can provide features such as vertical airflow intensity and cloud movement speed. Combined with the meteorological element correlation tree, the correlations between these features are analyzed. Cloud top temperature change may be strongly correlated with vertical airflow intensity, while cloud expansion rate may be strongly correlated with cloud movement speed. By analyzing these correlations, the coupling characteristics of meteorological elements are determined. If the correlation coefficient between cloud top temperature change and vertical airflow intensity is high, it can be considered that there is a strong coupling relationship between these two elements.
[0085] The correlation coefficients of the coupling characteristics of meteorological elements are calculated, such as the correlation coefficient between cloud top temperature change and vertical airflow intensity using statistical analysis methods. The influence weights of meteorological elements are then assessed in conjunction with current atmospheric stability parameters. If the current atmosphere is unstable, the influence weight of cloud top temperature change may increase. Based on the correlation coefficients and atmospheric stability parameters, the influence weights of potential meteorological elements are determined. If the correlation coefficient between cloud top temperature change and vertical airflow intensity is 0.8, and the current atmosphere is unstable, these two elements can be assigned higher influence weights. By analyzing the coupling relationships between meteorological elements and the current atmospheric state, the influence weights of each meteorological element are determined, thereby providing more accurate parameter support for weather forecasting.
[0086] S5.3: Cross-validation of cloud map dynamic evolution channel and meteorological element association tree.
[0087] Before completing the linkage adjustment between disaster warning levels and meteorological element data, the prediction results are cross-validated to ensure their reliability. The specific implementation steps are as follows: S5.3.1 Consistency Judgment The prediction confidence level P1 output by the cloud image dynamic evolution channel and the prediction confidence level P2 output by the meteorological element association tree are calculated respectively, with a preset deviation tolerance of ±5%. When |P1-P2|≤5%, the prediction confidence levels of the two are determined to be consistent, the prediction results are confirmed to be reliable, and the subsequent steps of linking disaster warning level and meteorological element data are executed.
[0088] Cross-validation is a method to verify the accuracy of a model by comparing the prediction results of two independent models to confirm the reliability of the prediction; prediction confidence is an indicator to measure the credibility of the prediction results, usually expressed as probability or confidence interval; linkage adjustment refers to linking disaster warning levels with meteorological element data to ensure the accuracy and timeliness of warning information.
[0089] The prediction results of the cloud image dynamic evolution channel and the meteorological element association tree are compared. These two models predict from the perspectives of cloud image dynamic change and meteorological element association, respectively, and the prediction confidence scores of the cloud image dynamic evolution channel and the meteorological element association tree are calculated. When the prediction confidence scores of the two models are consistent (e.g., the difference is within the allowable error range), the prediction results are considered reliable.
[0090] By combining a tiered prediction threshold comparison table and considering the prediction results and confidence levels of two models—the cloud image dynamic evolution channel and the meteorological element association tree—the disaster warning level is determined. If both models predict heavy rainfall in a certain area with high confidence, the warning level is raised to a higher level according to the tiered prediction threshold comparison table. The disaster warning level is then linked to meteorological element data for adjustment. If the warning level is high, the monitoring frequency of meteorological element data is automatically adjusted, and more data collection points are added to provide more detailed meteorological information to support the warning. Cross-validation ensures the reliability of the prediction results, and linked adjustment ensures the accuracy and timeliness of the disaster warning information.
[0091] S5.3.2 Deviation Closed-Loop Calibration When the prediction confidence under the dynamic evolution channel of cloud imagery and the prediction confidence under the meteorological element association tree are inconsistent (i.e., |P1-P2|>5%), and the prediction confidence is inconsistent and the deviation exceeds the preset deviation tolerance, the following closed-loop calibration steps are performed: the lateral connection of the SHAP value of the Fast connection layer of the dynamic evolution channel of cloud imagery is sent to the Slow connection layer; the SHAP value is used as the attention weight to perform feature response modulation on the Slow connection layer; then, the prediction confidence under the dynamic evolution channel of cloud imagery and the prediction confidence under the meteorological element association tree are re-determined until the confidence deviation meets the preset deviation tolerance.
[0092] Fast and Slow connection layers are two different types of connection layers in neural networks, used to handle rapidly changing and slowly changing features, respectively; SHAP (SHapley Additive exPlanations) values evaluate the importance of a feature by calculating the contribution of each feature to the model output; lateral connections are a neural network structure that allows information exchange between different layers; attention weights are used to adjust the model's attention to different features; feature response modulation is a process of adjusting the influence of features on the model output, optimizing the model's predictive performance by adjusting weights; preset bias tolerance is used to determine whether the deviation between the prediction confidence scores of two models is within an acceptable range.
[0093] By adjusting the feature weights of the models, the problem of inconsistent prediction confidence between the two models can be solved, thereby improving the reliability and consistency of the prediction results.
[0094] S5.4: Linkage adjustment between disaster early warning level and meteorological element data.
[0095] After completing cross-validation and confirming the reliability of the prediction results, the linkage adjustment between disaster warning levels and meteorological element data is completed based on the graded prediction threshold comparison table formulated in step S4. The specific implementation process is as follows: The cross-validated meteorological element prediction results are matched one by one with the thresholds of each level in the graded prediction threshold comparison table. Based on the matching results, the disaster warning level for the corresponding region and the corresponding prediction time is determined. At the same time, the disaster warning level is linked and bound with the corresponding meteorological element data to complete the linkage adjustment between the two. Finally, the disaster warning level and the corresponding meteorological element prediction results are output.
[0096] By implementing this step, the process barriers between meteorological element forecasting and disaster early warning issuance have been broken down, and the standardized and real-time linkage between disaster early warning levels and meteorological element data has been achieved, ensuring the accuracy, timeliness and traceability of early warning information.
[0097] Example 2 Based on the same inventive concept as the intelligent meteorological data prediction method based on satellite cloud image analysis in Embodiment 1 above, such as Figure 2 As shown, this embodiment provides an intelligent meteorological data forecasting system based on satellite cloud image analysis. The system includes: I. Dataset Acquisition Module: This module is used to acquire satellite cloud image feature datasets, which include the spatiotemporal resolution features of thunderstorm cloud clusters, local precipitation cloud systems, and strong convective squall lines. Specifically, it is used to complete the acquisition, reception, and preprocessing of satellite cloud image data through cloud top temperature monitoring layer, cloud system morphology monitoring layer, and atmospheric circulation monitoring layer, and output standardized satellite cloud image feature datasets to the dynamic evolution feature matrix construction module.
[0098] II. Dynamic Evolution Feature Matrix Construction Module: This module communicates with the dataset acquisition module and is used to construct a dynamic evolution feature matrix based on the satellite cloud image feature dataset.
[0099] A first feature index is set, which includes cloud image morphology features, motion vector features, and intensity change features. Based on the satellite cloud image feature dataset, the intensity evolution sub-matrices in the dynamic evolution feature matrix are determined by using correlation mapping and weighted fusion with the first feature index.
[0100] A second feature index is set, including cloud top temperature gradient, cloud expansion rate, and vertical airflow intensity. Based on the satellite cloud image feature dataset, the dynamic evolution sub-matrices in the dynamic evolution feature matrix are determined by using correlation mapping and weighted fusion with the second feature index.
[0101] III. Initial Prediction Threshold Setting Module: This module uses historical meteorological data and geographic feature parameters to set initial prediction thresholds that conform to the meteorological characteristics of the region. Specifically, it retrieves and statistically analyzes historical meteorological data and geographic feature parameters to set initial prediction thresholds that are adapted to the meteorological characteristics of the target region, and then outputs the results to the synchronization correction module.
[0102] IV. Synchronous Correction Module: This module communicates with the initial prediction threshold setting module. It employs adaptive fuzzy inference, inputting real-time cloud morphology parameters, cloud top brightness-temperature jump values, and terrain uplift coefficients. Dynamic correction coefficients are configured to synchronously correct the initial prediction threshold. Specifically, it is used for the operation of the adaptive fuzzy inference system, the calculation and configuration of dynamic correction coefficients, completing the real-time synchronous correction of the initial prediction threshold, and outputting the corrected prediction threshold to the classification storage module.
[0103] V. Classification and Storage Module: Communicating with the synchronization correction module, after synchronously correcting the initial prediction thresholds, it classifies and stores the data according to meteorological element type, regional division, and prediction lead time, and drafts a graded prediction threshold comparison table. Specifically, it is used for multi-dimensional classification and storage of corrected thresholds, drafting and updating the graded prediction threshold comparison table, and outputting the graded prediction threshold comparison table to the linkage tuning module.
[0104] VI. Linkage Tuning Module: This module communicates with both the dynamic evolution feature matrix construction module and the classification storage module. It sets up cloud image dynamic evolution channels that have a mapping relationship with the dynamic evolution feature matrix, and uses a graded prediction threshold lookup table to link disaster warning levels with meteorological element data for linkage tuning. Specifically, it is used for constructing and optimizing parameters of cloud image dynamic evolution channels, outputting meteorological element prediction results, linking disaster warning levels with meteorological element data for linkage tuning, and finally outputting warning and prediction results.
[0105] Furthermore, the linkage tuning module is used to execute the following methods: Connect to the historical meteorological event database and combine it with the dynamic evolution feature matrix to determine matching historical cloud map events; combine the evolution cycle distribution of matching historical cloud map events with the current cloud system change rate to determine the duration of potential meteorological events; determine the impact range of potential meteorological events by matching historical cloud map events that affect the same area, the same season, and the same intensity level; set up cloud map dynamic evolution channels based on the duration and impact range of potential meteorological events.
[0106] Furthermore, the linkage tuning module is also used to execute the following methods: The transmission path of potential meteorological elements is determined by the matching degree between the triggering conditions of abrupt changes in historical similar meteorological events in the influence weight of potential meteorological elements and the geographical feature parameters associated with regional meteorological characteristics. Based on the influence weight of potential meteorological elements and the transmission path of potential meteorological elements, the evolution parameters of the dynamic evolution channel of cloud map are optimized.
[0107] Furthermore, the linkage tuning module is also used to execute the following methods: Based on the intensity evolution sub-matrices and dynamic evolution sub-matrices in the dynamic evolution feature matrix, the coupling characteristics of meteorological elements are determined by combining the meteorological element association tree; the influence weight of potential meteorological elements is determined by combining the correlation coefficient of the coupling characteristics of meteorological elements with the current atmospheric stability parameters.
[0108] Furthermore, the linkage tuning module is also used to execute the following methods: Cross-validation of cloud image dynamic evolution channel and meteorological element association tree: When the prediction confidence under cloud image dynamic evolution channel and the prediction confidence under meteorological element association tree are consistent, disaster warning level and meteorological element data are linked and adjusted in conjunction with the graded prediction threshold comparison table.
[0109] Furthermore, the linkage tuning module is also used to execute the following methods: When the prediction confidence under the dynamic evolution channel of cloud imagery and the prediction confidence under the meteorological element association tree are inconsistent, the lateral connection of the SHAP value of the Fast connection layer of the dynamic evolution channel of cloud imagery is sent to the Slow connection layer. The SHAP value is used as the attention weight to modulate the feature response of the Slow connection layer. Then, the prediction confidence under the dynamic evolution channel of cloud imagery and the prediction confidence under the meteorological element association tree are re-determined until the confidence deviation meets the preset deviation tolerance.
[0110] The intelligent meteorological data prediction method and specific implementation logic based on satellite cloud image analysis in Example 1 are also fully applicable to the intelligent meteorological data prediction system based on satellite cloud image analysis in this embodiment. Through the foregoing detailed description of the intelligent meteorological data prediction method based on satellite cloud image analysis, those skilled in the art can clearly understand the intelligent meteorological data prediction system based on satellite cloud image analysis in this embodiment. Therefore, for the sake of brevity, it will not be described in detail here.
[0111] The above are specific embodiments of the present invention, which demonstrate the substantial features and progress of the present invention. Equivalent modifications can be made to them according to actual usage needs, under the guidance of the present invention, and all such modifications are within the scope of protection of this solution.
Claims
1. A method for intelligent forecasting of meteorological data based on satellite cloud image analysis, characterized in that, Includes the following steps: 1) Obtain a satellite cloud image feature dataset, and construct a dynamic evolution feature matrix based on the satellite cloud image feature dataset to characterize the dynamic change pattern of cloud clusters throughout their entire life cycle; 2) Set an initial prediction threshold that matches the meteorological characteristics of the target area by using historical meteorological data and geographical feature parameters of the target area; 3) An adaptive fuzzy inference algorithm is adopted, and real-time cloud morphology parameters, cloud top brightness temperature jump values and terrain uplift coefficients are input. Dynamic correction coefficients are calculated and configured, and the initial prediction threshold is synchronously corrected through the dynamic correction coefficients. 4) The initial forecast thresholds after synchronous correction are classified and stored according to meteorological element type, regional division and forecast lead time, and a graded forecast threshold comparison table is proposed. 5) Set up a cloud map dynamic evolution channel that has a mapping relationship with the dynamic evolution feature matrix, and use the hierarchical prediction threshold reference table as a benchmark to complete the linkage adjustment of disaster warning level and meteorological element data, and output the disaster warning level and the corresponding meteorological element prediction results.
2. The intelligent meteorological data forecasting method based on satellite cloud image analysis according to claim 1, characterized in that, In step 1), the satellite cloud image feature dataset includes spatiotemporal resolution data of thunderstorm cloud clusters, local precipitation cloud systems, and strong convective squall lines; the satellite cloud image feature dataset is collected through cloud top temperature monitoring layer, cloud system morphology monitoring layer, and atmospheric circulation monitoring layer, and transmitted to the ground receiving end via transmission network to complete the acquisition.
3. The intelligent meteorological data forecasting method based on satellite cloud image analysis according to claim 1, characterized in that, The specific method for constructing the dynamic evolution feature matrix is as follows: A first feature index is set, which includes cloud image morphology features, motion vector features, and intensity change features. Based on the satellite cloud image feature dataset, each intensity evolution sub-matrix in the dynamic evolution feature matrix is determined by associating and mapping with the first feature index and weighting and fusing it. A second feature index is set, which includes cloud top temperature gradient, cloud expansion rate, and vertical airflow intensity. Based on the satellite cloud image feature dataset, each dynamic evolution sub-matrix in the dynamic evolution feature matrix is determined by associating and mapping with the second feature index and weighting fusion.
4. The intelligent meteorological data forecasting method based on satellite cloud image analysis according to claim 1, characterized in that, In step 5), the specific method for setting the cloud map dynamic evolution channel that has a mapping relationship with the dynamic evolution feature matrix is as follows: Connect to the historical meteorological event database and match the corresponding historical cloud map events using the dynamic evolution feature matrix; The duration of potential meteorological events is determined by combining the evolution cycle distribution of matched historical cloud map events with the current cloud system change rate. Using historical cloud map events that match the meteorological events predicted by the current cloud map features in terms of impact area, season of occurrence, and intensity level, the impact range of potential meteorological events is determined. Based on the duration and impact range of the potential meteorological events, the cloud map dynamic evolution channel is constructed and set.
5. The intelligent meteorological data forecasting method based on satellite cloud image analysis according to claim 4, characterized in that, The method further includes: Based on the dynamic evolution feature matrix, and combined with the meteorological element association tree and the current atmospheric stability parameters, the influence weights of potential meteorological elements are determined. Based on the matching degree between the triggering conditions of element mutations in similar historical meteorological events and the geographical feature parameters associated with the meteorological characteristics of the target area, and combined with the influence weight of the potential meteorological elements, the transmission path of the potential meteorological elements is determined. Based on the influence weights of the potential meteorological elements and the transmission paths of the potential meteorological elements, the evolution parameters of the cloud map dynamic evolution channel are optimized.
6. The intelligent meteorological data forecasting method based on satellite cloud image analysis according to claim 5, characterized in that, The method for determining the influence weights of the potential meteorological elements is as follows: Based on each intensity evolution sub-matrix and each dynamic evolution sub-matrix in the dynamic evolution feature matrix, the coupling characteristics of meteorological elements are determined by combining the meteorological element association tree. The influence weights of the potential meteorological elements are determined by using the correlation coefficients of the coupling characteristics of the meteorological elements and combining them with the current atmospheric stability parameters.
7. The intelligent meteorological data forecasting method based on satellite cloud image analysis according to claim 1, characterized in that, Before completing the linkage adjustment between disaster warning level and meteorological element data in step 5), the process also includes a step of cross-validating the cloud image dynamic evolution channel and meteorological element association tree, specifically: When the deviation between the prediction confidence level output by the cloud image dynamic evolution channel and the prediction confidence level output by the meteorological element association tree meets the preset deviation tolerance, it is determined that the two prediction confidence levels are consistent, and the subsequent disaster warning level and meteorological element data linkage adjustment steps are executed.
8. The intelligent forecasting method for meteorological data based on satellite cloud image analysis according to claim 7, characterized in that, The cross-validation step also includes: When the deviation between the prediction confidence level output by the cloud image dynamic evolution channel and the prediction confidence level output by the meteorological element association tree exceeds the preset deviation tolerance, the SHAP value of the Fast connection layer of the cloud image dynamic evolution channel is sent to the Slow connection layer through a lateral connection. Using the SHAP value as the attention weight, feature response modulation is applied to the Slow connection layer. After modulation, the prediction confidence of the cloud image dynamic evolution channel and the meteorological element association tree is recalculated. The above lateral connection and feature response modulation steps are repeated until the confidence deviation between the two meets the preset deviation tolerance.
9. A meteorological data intelligent forecasting system based on satellite cloud image analysis, characterized in that, The system for implementing the method of any one of claims 1 to 8 comprises: Dataset acquisition module: used to acquire satellite cloud image feature datasets; Dynamic evolution feature matrix construction module: connected to the dataset acquisition module, used to construct a dynamic evolution feature matrix that characterizes the dynamic change law of the cloud cluster throughout its entire life cycle based on the satellite cloud image feature dataset; Initial prediction threshold setting module: used to set an initial prediction threshold that matches the meteorological characteristics of the target area based on historical meteorological data and geographical feature parameters of the target area; Synchronous correction module: connected to the initial prediction threshold setting module, used to use an adaptive fuzzy inference algorithm, input real-time cloud morphology parameters, cloud top brightness temperature jump value and terrain lifting coefficient, calculate and configure dynamic correction coefficient, and synchronously correct the initial prediction threshold through the dynamic correction coefficient; Classification and storage module: connected to the synchronization correction module, used to classify and store the initial prediction thresholds after synchronization correction according to meteorological element type, regional division and prediction time, and to formulate a graded prediction threshold comparison table; Linkage Adjustment Module: Connected to the dynamic evolution feature matrix construction module and the classification storage module, it is used to set up cloud map dynamic evolution channels that have a mapping relationship with the dynamic evolution feature matrix, and to complete the linkage adjustment of disaster warning level and meteorological element data based on the hierarchical prediction threshold reference table.
10. The intelligent meteorological data forecasting system based on satellite cloud image analysis according to claim 9, characterized in that, Also includes: Cross-validation module: It is connected to the cloud image dynamic evolution channel and the preset meteorological element association tree in the linkage tuning module respectively. It is used to calculate the deviation between the prediction confidence level output by the cloud image dynamic evolution channel and the prediction confidence level output by the meteorological element association tree, and to determine whether the deviation meets the preset deviation tolerance. Feature modulation module: connected to the cross-validation module, used to send the SHAP value of the Fast connection layer of the cloud map dynamic evolution channel to the Slow connection layer through a lateral connection when the deviation exceeds the preset deviation tolerance, use the SHAP value as the attention weight to perform feature response modulation on the Slow connection layer, and feed the modulation result back to the cross-validation module to recalculate the prediction confidence until the deviation meets the preset deviation tolerance.
