An artificial intelligence-based spatiotemporal sequence weather situation automatic typing method and system

Abstract

The application relates to the technical field of artificial intelligence, in particular to a spatiotemporal sequence weather situation automatic typing method and system based on artificial intelligence. The spatiotemporal sequence weather situation automatic typing method based on artificial intelligence comprises the following steps: acquiring at least one weather situation chart; inputting the at least one weather situation chart into a preset weather type classification model; determining a weather type corresponding to the at least one weather situation chart based on an output result of the weather type classification model; and the weather type classification model is obtained through the following steps: acquiring a historical weather situation chart data set; the historical weather situation chart data set comprises a plurality of historical weather situation charts and weather type labels and time sequence labels corresponding to the historical weather situation charts; taking the historical weather situation chart data set as a training sample, performing supervised learning training on an initial model, and obtaining the weather type classification model. The application provides a weather situation automatic typing method which is higher in objectivity, efficiency and convenience.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, specifically to an automatic classification method and system for spatiotemporal weather patterns based on artificial intelligence. Background Technology

[0002] Current methods for determining weather map types primarily rely on manual judgment by meteorological experts and operational personnel. This manual approach has significant drawbacks: First, due to differences in experience and understanding among experts and operational personnel, there may be considerable fluctuations in subjective judgment during transitional or shifting weather types, making it difficult to guarantee the consistency and accuracy of the results. Second, manual judgment is relatively inefficient, especially when processing large amounts of weather map data, and is time-consuming. For example, manually judging a set of weather maps requires access to a large amount of knowledge and experience, typically taking more than 3 seconds. With the increasing volume of meteorological data and the rising demands for timely weather forecasts, traditional manual judgment methods are no longer sufficient to meet the needs of modern meteorological operations. Therefore, there is an urgent need for a technical solution that can improve the efficiency and accuracy of weather type identification and reduce subjective differences. Summary of the Invention

[0003] This application provides an artificial intelligence-based method and system for automatic classification of spatiotemporal weather patterns.

[0004] The first aspect of this application provides an automatic classification method for spatiotemporal weather patterns based on artificial intelligence, including: Obtain at least one weather map; Input at least one weather map into the preset weather type classification model; The weather type corresponding to the at least one weather map is determined based on the output of the weather type classification model, wherein the weather type classification model is trained through the following steps: Obtain a historical weather map dataset; wherein: the historical weather map dataset includes multiple historical weather maps and weather type and time series labels corresponding to each historical weather map; The historical weather map dataset is used as training samples to train the initial model through supervised learning, resulting in the weather type classification model.

[0005] In one optional embodiment of this application, the weather type classification model is a multimodal architecture-based model, and the weather type classification model includes: A visual encoder is used to encode the at least one weather map to extract visual features; A language model is used to process textual information related to weather types.

[0006] In one optional embodiment of this application, the weather type classification model employs a multi-level attention mechanism to capture, in parallel, the macroscopic features of global atmospheric circulation and the detailed features of local weather systems in the weather situation map; and / or, The model employs a three-dimensional convolutional network to process and model temporal features of a set of continuous weather maps in time or space.

[0007] In an optional embodiment of this application, the step of using the historical weather map dataset as training samples to perform supervised learning training on the initial model includes: Select samples of preset mainstream weather types from the historical weather map dataset; The visual encoder and aligner of the initial model are fully adjusted based on the samples of the mainstream weather types. The language model portion of the initial model is then subjected to lightweight adjustments.

[0008] In an optional embodiment of this application, the step of using the historical weather map dataset as training samples to perform supervised learning training on the initial model further includes: Select preset long-tail weather type samples from the historical weather map dataset; With the parameters of the visual encoder and the aligner frozen, the language model portion of the initial model is lightly adjusted based on the long-tailed weather type samples.

[0009] In one optional embodiment of this application, obtaining at least one weather map includes: Obtain at least one set of weather situation maps; wherein each weather situation map in at least one set of weather situation maps corresponds to the same period and corresponds to multiple different preset height layers; the preset height layers are at least one of the preset ground layer, 850hPa height layer, 700hPa height layer and 500hPa height layer.

[0010] In an optional embodiment of this application, before inputting at least one weather situation map into a preset weather type classification model, the method further includes: In response to an instruction to acquire meteorological reanalysis data, at least one weather pattern map is generated based on the meteorological reanalysis data.

[0011] In one optional embodiment of this application, the meteorological reanalysis data includes meteorological elements; When the preset altitude layer is the ground layer, the meteorological elements include at least one of sea level pressure, air temperature, and wind field elements; When the preset height layer is 850hPa, 700hPa or 500hPa, the meteorological elements include at least one of geopotential height, air temperature and wind field.

[0012] In one optional embodiment of this application, determining the weather type corresponding to the at least one weather map based on the output of the weather type classification model includes: Receive multiple candidate weather types and their corresponding confidence scores output by the weather type classification model for the at least one weather situation map; The candidate weather types are sorted according to the confidence scores, and the top N ranked candidate weather types are selected as the target weather types, where N is a preset integer greater than 1.

[0013] A second aspect of this application provides an artificial intelligence-based automatic classification system for spatiotemporal weather patterns, comprising at least: The acquisition module is used to acquire at least one weather map. The determination module is used to input the at least one weather situation map into a preset weather type classification model; Based on the output of the weather type classification model, the weather type corresponding to the at least one weather map is determined. The training module is used to train the weather type classification model through the following steps: obtaining a historical weather map dataset; wherein: the historical weather map dataset includes multiple historical weather maps and weather type labels and time series labels corresponding to each historical weather map; using the historical weather map dataset as training samples, supervising the initial model to obtain the weather type classification model.

[0014] This application embodiment achieves rapid, accurate, and objective type identification of weather maps by introducing an intelligent weather type classification model based on a multimodal architecture and combining it with historical weather map drawing, storage, and retrieval functions. Specifically, this application embodiment achieves significant beneficial effects through the following technical means: 1. Improved efficiency and objectivity in weather type identification: The embodiments of this application use an artificial intelligence model to replace traditional manual judgment. The model is trained on a large amount of historical weather map data, which is equivalent to the brain of an experienced meteorological expert. Its judgment speed is far superior to that of the human brain (for example, it only takes 1-3 seconds to judge a set of weather maps), and it eliminates the subjective differences and fluctuations that may be caused by manual judgment, making the judgment results more stable and objective.

[0015] 2. Provides convenient weather map drawing function: This application embodiment integrates a weather map drawing program. Users only need to provide ERA5 reanalysis data that meets the requirements to quickly complete the drawing of weather maps. Due to the ease of obtaining ERA5 data and the standardized drawing style, this function has wide applicability and simplifies the data preparation process.

[0016] 3. A comprehensive historical weather map sample database and retrieval function have been established: This application embodiment includes historical weather maps from 30 years (January 1, 1976 to December 31, 2005) as a sample database, and has developed a function to support retrieval by date, altitude layer, and weather type. This provides meteorological researchers and operational personnel with an efficient tool for historical data analysis and weather process review, greatly improving the value of data utilization and work efficiency. Attached Figure Description

[0017] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1 An interactive diagram illustrating an AI-based automatic weather pattern classification method for spatiotemporal sequences, provided in one embodiment of this application. Figure 2 A flowchart illustrating an embodiment of this application of an automatic classification method for spatiotemporal weather patterns based on artificial intelligence; Figure 3 This is a schematic diagram of an artificial intelligence-based spatiotemporal sequence weather situation automatic classification system provided in one embodiment of this application. Detailed Implementation

[0018] In the process of developing this application, the inventors discovered that there is an urgent need for a technical solution that can improve the efficiency and accuracy of weather type identification and reduce subjective differences.

[0019] To address the aforementioned issues, this application provides an artificial intelligence-based method and system for automatically classifying spatiotemporal weather patterns.

[0020] The solutions in this application embodiment can be implemented using various computer languages, such as the object-oriented programming language Java and the interpreted scripting language JavaScript.

[0021] To make the technical solutions and advantages of the embodiments of this application clearer, the exemplary embodiments of this application will be described in further detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not an exhaustive list of all embodiments. It should be noted that, unless otherwise specified, the embodiments and features in the embodiments of this application can be combined with each other.

[0022] Please see Figure 1 and Figure 2 The automatic classification method for spatiotemporal weather patterns based on artificial intelligence provided in this application includes the following steps 201-203: Step 201: Obtain at least one weather map; The weather map can be image data from any source used to describe weather conditions in a specific region and time. For example, the weather map can be an image drawn from meteorological observation data, numerical weather prediction model output data, or meteorological reanalysis data. The weather map can be a single image or a collection of multiple images. For example, a surface weather map showing the surface pressure field and wind field at a certain moment can be acquired, or a 500hPa upper-air weather map showing the 500hPa geopotential height field and temperature field at a certain moment can be acquired. The acquisition method can be achieved through various means such as reading from local storage devices, downloading from a network server, or real-time acquisition by sensors. This provides raw input data for subsequent intelligent weather type identification.

[0023] Step 202: Input at least one weather map into the preset weather type classification model; The weather type classification model is a trained artificial intelligence model whose function is to identify and output the corresponding weather type based on the input weather map. The model can be built based on deep learning techniques, such as Convolutional Neural Networks (CNN), Recurrent Neural Networks (RNN), or Transformer architectures. The term "preset" means that the model has been trained and optimized before deployment, possessing the ability to classify and discriminate weather maps. Inputting the weather map into the model means using the image data of the weather map as input features for the model to process and analyze. For example, the pixel data of the weather map can be converted into a tensor format acceptable to the model and fed into the model's input layer.

[0024] Step 203: Determine the weather type corresponding to the at least one weather map based on the output of the weather type classification model; After receiving the weather map as input, the weather type classification model generates one or more output results based on its internally learned features and patterns. These output results typically include one or more candidate weather types and confidence scores or probability values ​​associated with these types. For example, the model might output a confidence score of 0.85 for "cold front weather," 0.10 for "warm front weather," and 0.05 for "high-pressure ridge weather." The goal of this step is to ultimately determine the weather type represented by the weather map based on the model's output. Typically, the candidate weather type with the highest confidence or probability is selected as the final classification result. For example, in the above example, "cold front weather" would be determined as the weather type corresponding to the weather map. This step automates and intelligently classifies weather map types, significantly improving classification efficiency and reducing subjective differences caused by manual judgment.

[0025] The weather type classification model is a trained model that receives weather maps as input and outputs corresponding weather type classification results. During training, the model learns the mapping relationship between the visual features of weather maps and weather types. The weather type classification model is trained through the following steps: Obtain a historical weather map dataset; wherein: the historical weather map dataset includes multiple historical weather maps and weather type and time series labels corresponding to each historical weather map; The historical weather map dataset is used as training samples to train the initial model through supervised learning, resulting in the weather type classification model.

[0026] First, a historical weather map dataset is acquired. This dataset includes multiple historical weather maps and corresponding weather type labels for each map. For example, the dataset might contain daily weather maps at the surface, 850hPa, 700hPa, and 500hPa altitudes at UTC 00 from January 1, 1976 to December 31, 2005. Each historical weather map has been pre-identified by meteorological experts and labeled with its corresponding weather type (e.g., 14 categories, coded with two English letters). This labeling information serves as the model's "ground truth" or "actual" data.

[0027] Secondly, the historical weather map dataset is used as training samples to supervise the training of an initial model to obtain the weather type classification model. During training, the initial model receives historical weather maps as input and attempts to predict their weather types. By comparing the model's predictions with pre-labeled real weather types, and based on the difference between the two (i.e., the loss), the model's internal parameters are adjusted using backpropagation algorithms and optimizers (such as Adam, SGD, etc.). This iterative process continues until the model's performance on the training set reaches a preset standard, or its performance on the validation set no longer improves. For example, training with a large amount of labeled historical weather map data optimizes the model parameters, ensuring its accuracy and robustness. After training, the model will be able to automatically identify weather types, supporting more efficient weather analysis.

[0028] Through the aforementioned training process, the weather type classification model can learn image feature patterns corresponding to different weather types from a large amount of historical data, thereby possessing the ability to accurately classify new and unseen weather maps. This supervised learning-based training method enables the model to simulate or even surpass the judgment ability of human experts, significantly improving the efficiency and objectivity of weather type identification. For example, in this application, using 30 years of historical image data as training samples, an artificial intelligence model was autonomously trained. Based on weather maps, it can identify the top three weather types that best match the characteristics, with an accuracy rate exceeding 65%. The model training effect has passed the review of industry experts. This indicates that the model can provide reliable discrimination results in practical applications and effectively assist meteorological operations and research.

[0029] This application embodiment achieves rapid, accurate, and objective type identification of weather maps by introducing an intelligent weather type classification model based on a multimodal architecture and combining it with historical weather map drawing, storage, and retrieval functions. Specifically, this application embodiment achieves significant beneficial effects through the following technical means: 1. Improved efficiency and objectivity in weather type identification: The embodiments of this application use an artificial intelligence model to replace traditional manual judgment. The model is trained on a large amount of historical weather map data, which is equivalent to the brain of an experienced meteorological expert. Its judgment speed is far superior to that of the human brain (for example, it only takes 1-3 seconds to judge a set of weather maps), and it eliminates the subjective differences and fluctuations that may be caused by manual judgment, making the judgment results more stable and objective.

[0030] 2. Provides convenient weather map drawing function: This application embodiment integrates a weather map drawing program. Users only need to provide ERA5 reanalysis data that meets the requirements to quickly complete the drawing of weather maps. Due to the ease of obtaining ERA5 data and the standardized drawing style, this function has wide applicability and simplifies the data preparation process.

[0031] 3. A comprehensive historical weather map sample database and retrieval function have been established: This application embodiment includes historical weather maps from 30 years (January 1, 1976 to December 31, 2005) as a sample database, and has developed a function to support retrieval by date, altitude layer, and weather type. This provides meteorological researchers and operational personnel with an efficient tool for historical data analysis and weather process review, greatly improving the value of data utilization and work efficiency.

[0032] In one optional embodiment of this application, the weather type classification model is a multimodal architecture-based model, and the weather type classification model includes: A visual encoder is used to encode the at least one weather map to extract visual features; A language model is used to process textual information related to weather types.

[0033] In one optional embodiment of this application, the weather type classification model employs a multi-level attention mechanism to capture in parallel the macroscopic features of global atmospheric circulation and the detailed features of local weather systems in the weather situation map; In one optional embodiment of this application, the model employs a three-dimensional convolutional network to process and model temporal features of a set of weather situation maps that are continuous in time or space.

[0034] The multimodal architecture-based model also employs a three-dimensional convolutional network to process a set of continuous weather maps in time or space, thereby modeling temporal features. For example, when given a set of continuous weather maps at different time points or altitudes, the three-dimensional convolutional network can capture dynamic changes in these image sequences, such as the movement, development, and evolution trends of weather systems. This is crucial for understanding the dynamic characteristics of weather processes; for instance, by analyzing continuous weather maps, the model can identify the formation, development, and dissipation of cyclones, or the movement path and intensity changes of fronts. By modeling temporal features, the method in this embodiment can more accurately identify the types of complex weather situations, especially during the evolution of weather systems, providing more refined discrimination results.

[0035] This embodiment employs a multimodal architecture-based model, combining a visual encoder and a language model, to simultaneously process image and text information, achieving a deep understanding of weather maps. A multi-layered attention mechanism allows the model to focus on both macroscopic atmospheric circulation and key details such as troughs and ridges, improving the comprehensiveness of feature extraction. The 3D convolutional network effectively models temporal features, offering significant advantages in analyzing continuously changing weather processes, thereby enhancing the accuracy and precision of complex weather pattern identification. For example, when predicting typhoon paths, traditional models may only identify the typhoon's current location, while the model in this embodiment, through a 3D convolutional network analyzing continuous time-series weather maps, can better capture the typhoon's speed and direction of movement, thus improving the accuracy of typhoon type identification and providing a more reliable basis for subsequent path prediction.

[0036] In an optional embodiment of this application, the above-mentioned use of the historical weather map dataset as training samples to perform supervised learning training on the initial model includes the following steps: Select samples of preset mainstream weather types from the historical weather map dataset; The visual encoder and aligner of the initial model are fully adjusted based on the samples of the mainstream weather types. The language model portion of the initial model is then subjected to lightweight adjustments.

[0037] Select preset long-tail weather type samples from the historical weather map dataset; With the parameters of the visual encoder and the aligner frozen, the language model portion of the initial model is lightly adjusted based on the long-tailed weather type samples.

[0038] The first training phase focuses on optimizing the model's ability to identify mainstream weather types. Specifically, samples of mainstream weather types are selected from historical weather map datasets. These samples are those that appear frequently and constitute a large proportion of the historical dataset; for example, in northern my country, cold high-pressure systems in winter and subtropical high-pressure systems in summer. Using these mainstream weather type samples, the initial model's visual encoder and aligner are fully fine-tuned. The visual encoder extracts visual features from the weather maps, such as identifying macroscopic and local features in the image information, including pressure fields, wind fields, and temperature fields. The aligner aligns the visual features with text descriptions, ensuring the model understands the relationship between image content and weather type labels. Full parameter fine-tuning involves updating all trainable parameters of the visual encoder and aligner to ensure they fully learn the feature representations of mainstream weather types. Simultaneously, a first lightweight fine-tuning is performed on the language model. This language model handles textual information related to weather types, such as the name and description of the weather type. Lightweight fine-tuning (such as LoRA, or low-rank adaptation) aims to efficiently adjust language models with a small number of parameters and computational cost, enabling them to better understand and generate text descriptions related to mainstream weather types. Through the first training phase, the model can develop a stable and accurate ability to recognize common weather types, laying the foundation for subsequent processing of rare weather types.

[0039] The second training phase aims to enhance the model's ability to classify long-tail weather types. Specifically, long-tail weather type samples are selected from the historical weather map dataset. Long-tail weather type samples refer to weather types that occur less frequently and in smaller quantities in the historical dataset, such as rare extreme weather events or specific regional weather phenomena. With the parameters of the visual encoder and aligner frozen, the language model is fine-tuned only using these long-tail weather type samples. Freezing the parameters of the visual encoder and aligner avoids interference or forgetting of learned mainstream weather type feature representations when processing rare samples. Fine-tuning only the language model efficiently combines the textual information of long-tail weather types with the model's existing visual understanding capabilities, thereby improving the model's recognition accuracy for these rare weather types while avoiding overfitting. For example, oversampling the long-tail samples or using a class-weighted cross-entropy loss function can further balance the impact of different categories of samples on model training.

[0040] Through the phased training strategy described above, the method in this embodiment can effectively solve the problem of imbalanced weather type data distribution. The first phase ensures the stability of the model's identification of mainstream weather types, while the second phase specifically enhances the model's classification ability for rare, long-tail weather types. This training method makes the final weather type classification model more robust and generalizable, achieving high-precision discrimination for various weather types. For example, in practical applications, the TOP3 accuracy can exceed 65%, significantly outperforming models trained in a single phase.

[0041] In an optional embodiment of this application, step 201, obtaining at least one weather map, includes: Obtain at least one set of weather maps; Among them, each weather pattern map in at least one set of weather pattern maps corresponds to the same period and corresponds to multiple different preset height layers; the preset height layer is at least one of the preset ground layer, 850hPa height layer, 700hPa height layer and 500hPa height layer.

[0042] Acquire weather maps to be classified. These weather maps can be existing image files, such as those retrieved from a historical database, or real-time generated maps. For example, a set of weather maps can be received from a data storage device or network interface. Each weather map in the set corresponds to the same date, for example, October 26, 2023. Furthermore, each weather map in the set corresponds to multiple different preset altitude layers, such as the surface layer, the 850 hPa altitude layer, the 700 hPa altitude layer, and the 500 hPa altitude layer. By acquiring multiple weather maps at different altitude layers for the same date, more comprehensive three-dimensional meteorological information can be provided for subsequent weather type classification models. For example, surface weather maps can reflect weather systems near the Earth's surface, while upper-level weather maps can reveal the characteristics of upper-level circulation. Combining information from these different altitude layers helps the model more accurately understand and classify complex weather situations, such as determining whether an upper-level trough is combined with a surface cyclone, thereby improving classification accuracy.

[0043] Through the above steps, the method of this embodiment can obtain weather situation maps at different altitudes on the same date, providing more comprehensive three-dimensional meteorological information for the weather type classification model, which helps the model to more accurately understand and judge complex weather situations, thereby improving the judgment accuracy.

[0044] In an optional embodiment of this application, before step 202, which involves inputting at least one weather situation map into a preset weather type classification model, the aforementioned AI-based automatic classification method for spatiotemporal weather patterns further includes: In response to an instruction to acquire meteorological reanalysis data, at least one weather pattern map is generated based on the meteorological reanalysis data.

[0045] Before inputting the at least one weather map into the weather type classification model, this embodiment can generate weather maps as needed by acquiring meteorological reanalysis data. Specifically, users can input meteorological reanalysis data into the software, such as ERA5 reanalysis data from the European Centre for Medium-Range Weather Forecasts (ECMWF). The meteorological reanalysis data contains meteorological element data for specific dates, times, and altitudes. After receiving this data, the software will visualize it into one or more weather maps according to preset meteorological industry drawing standards and specifications. For example, weather maps at the surface, 850 hPa, 700 hPa, or 500 hPa altitudes can be generated. This drawing function allows users to generate the weather maps to be analyzed directly within the software without relying on external drawing tools, greatly improving operational convenience and data processing integration. The weather maps generated in this way maintain the same format and content as the historical weather maps used for model training, ensuring data standardization for input into the model, thereby improving the accuracy and reliability of subsequent judgment results.

[0046] After acquiring or drawing a weather map, it is used as input data and sent to a pre-trained weather type classification model. This weather type classification model is the core component of this method; its function is to extract features and perform pattern recognition on the input weather map to determine its weather type. Upon receiving the weather map, the model processes it and outputs a classification result. This result typically includes one or more candidate weather types, along with a confidence score for each type. Based on these outputs, this method determines the final weather type. For example, the type with the highest confidence score can be selected as the final classification result, or multiple high-confidence candidate types can be provided for user reference, depending on the specific application requirements. This automated classification method significantly improves the efficiency of weather type classification, reducing the manual classification process, which originally took several minutes or even longer, to 1-3 seconds. It also effectively avoids the subjective differences that may arise from manual classification, making the classification results more objective and stable.

[0047] In one optional embodiment of this application, the meteorological reanalysis data includes meteorological elements; When the preset altitude layer is the ground layer, the meteorological elements include at least one of sea level pressure, air temperature, and wind field elements; When the preset height layer is 850hPa, 700hPa or 500hPa, the meteorological elements include at least one of geopotential height, air temperature and wind field.

[0048] In this embodiment, the meteorological reanalysis data includes data for at least one preset height layer, which is selected from the group consisting of the surface layer, the 850 hPa height layer, the 700 hPa height layer, and the 500 hPa height layer. Furthermore, the data includes meteorological elements corresponding to the preset height layer, wherein the meteorological elements include: when the preset height layer is the surface layer, sea level pressure, air temperature, and wind field elements; and when the preset height layer is the 850 hPa, 700 hPa, or 500 hPa height layer, geopotential height, air temperature, and wind field elements.

[0049] Specifically, the meteorological reanalysis data used in this embodiment is preferably ERA5 reanalysis data from the European Centre for Medium-Range Weather Forecasts (ECMWF), which has become a standard data source in the meteorological field due to its high accuracy and wide applicability. By downloading reanalysis data for a specified date (e.g., UTC 00) and altitude (e.g., surface, 850 hPa, 700 hPa, 500 hPa) from the ECMWF official website, the generated weather map can be ensured to have a high degree of authority and accuracy.

[0050] For the surface layer, the required meteorological elements include sea level pressure, 2-meter air temperature, and the u-component and v-component winds (constituting wind field elements). These elements comprehensively reflect near-surface atmospheric conditions and are crucial for analyzing surface weather systems (such as high pressure, low pressure, and fronts). For the upper atmosphere, such as the 850hPa, 700hPa, and 500hPa levels, the required meteorological elements include geopotential height, air temperature, and the u-component and v-component winds (constituting wind field elements). The geopotential height map visually displays the location and intensity of upper-level troughs, ridges, high-pressure centers, and low-pressure centers; the air temperature map reflects the upper-level temperature distribution; and the wind field elements reveal the characteristics of upper-level atmospheric circulation. By explicitly specifying the source, altitude layer, and meteorological elements of the meteorological reanalysis data, this embodiment ensures that the generated weather map conforms to meteorological industry standards, providing high-quality, standardized, and physically meaningful input data for subsequent intelligent weather type identification. This standardized data input effectively reduces model identification errors caused by data quality issues, thereby ensuring the reliability and accuracy of the identification results.

[0051] In an optional embodiment of this application, step 203 above, determining the weather type corresponding to the at least one weather map based on the output result of the weather type classification model, includes the following steps: Receive multiple candidate weather types and their corresponding confidence scores output by the weather type classification model for the at least one weather situation map; After at least one weather map is input into a pre-defined weather type classification model, the model analyzes the input image based on its internal training results and discrimination logic, and outputs multiple possible (i.e., "candidate") weather types. Each candidate weather type is accompanied by a confidence score, which indicates the degree or probability that the model believes the weather type matches the input weather map. For example, for an input weather map, the model might output candidate weather types and their confidence scores such as "high-pressure ridge, confidence 0.85", "low-pressure trough, confidence 0.10", and "shear line, confidence 0.05". These confidence scores are typically values ​​between 0 and 1, with higher scores indicating greater confidence in the model's classification of the weather type.

[0052] The candidate weather types are sorted according to the confidence scores, and the top N ranked candidate weather types are selected as the target weather types, where N is a preset integer greater than 1.

[0053] After receiving multiple candidate weather types and their confidence scores, the system sorts these candidate weather types in descending order based on their confidence scores. Then, based on a preset integer N (for example, N can be set to 2, 3, or 5), the top N candidate weather types are selected as the final judgment result. For example, if N is set to 3, and the candidate weather types and their confidence scores output by the model are: "High-pressure ridge (0.85)", "Low-pressure trough (0.10)", "Shear line (0.05)", and "Typhoon (0.005)", then the final determined weather type will be "High-pressure ridge", "Low-pressure trough", and "Shear line".

[0054] By outputting multiple candidate weather types and their confidence scores, and selecting the top N results, this embodiment provides users with richer discrimination information. Especially when weather types are ambiguous or in transitional periods, a single discrimination result may not fully reflect the actual situation. Providing multiple high-confidence candidate types offers meteorological experts a variety of possibilities for reference and decision-making, thereby improving the practicality and decision support capabilities of the discrimination results. For example, when the model outputs relatively high confidence scores for both "high-pressure ridge" and "low-pressure trough," it may mean that the region is in a transitional phase from a high-pressure ridge to a low-pressure trough, providing weather forecasters with more refined judgment criteria.

[0055] It should be understood that although the steps in the flowchart are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order constraint on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the diagram may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the sub-steps or stages of other steps.

[0056] Please see Figure 3 One embodiment of this application provides an artificial intelligence-based spatiotemporal sequence weather situation automatic classification system 300, which includes at least: Module 310 is used to acquire at least one weather map; The determination module 320 is used to input the at least one weather situation map into a preset weather type classification model; Based on the output of the weather type classification model, the weather type corresponding to the at least one weather map is determined. Training module 330 is used to train the weather type classification model through the following steps: obtaining a historical weather map dataset; wherein: the historical weather map dataset includes multiple historical weather maps and weather type labels and time series labels corresponding to each historical weather map; using the historical weather map dataset as training samples, supervising the initial model to obtain the weather type classification model.

[0057] In one optional embodiment of this application, the weather type classification model is a multimodal architecture-based model, and the weather type classification model includes: A visual encoder is used to encode the at least one weather map to extract visual features; A language model is used to process textual information related to weather types.

[0058] In one optional embodiment of this application, the weather type classification model employs a multi-level attention mechanism to capture, in parallel, the macroscopic features of global atmospheric circulation and the detailed features of local weather systems in the weather situation map; and / or, The model employs a three-dimensional convolutional network to process and model temporal features of a set of continuous weather maps in time or space.

[0059] In an optional embodiment of this application, the step of using the historical weather map dataset as training samples to perform supervised learning training on the initial model includes: Select samples of preset mainstream weather types from the historical weather map dataset; The visual encoder and aligner of the initial model are fully adjusted based on the samples of the mainstream weather types. The language model portion of the initial model is then subjected to lightweight adjustments.

[0060] In an optional embodiment of this application, the step of using the historical weather map dataset as training samples to perform supervised learning training on the initial model further includes: Select preset long-tail weather type samples from the historical weather map dataset; With the parameters of the visual encoder and the aligner frozen, the language model portion of the initial model is lightly adjusted based on the long-tailed weather type samples.

[0061] In one optional embodiment of this application, obtaining at least one weather map includes: Obtain at least one set of weather situation maps; wherein each weather situation map in at least one set of weather situation maps corresponds to the same period and corresponds to multiple different preset height layers; the preset height layers are at least one of the preset ground layer, 850hPa height layer, 700hPa height layer and 500hPa height layer.

[0062] In an optional embodiment of this application, before inputting at least one weather situation map into a preset weather type classification model, the method further includes: In response to an instruction to acquire meteorological reanalysis data, at least one weather pattern map is generated based on the meteorological reanalysis data.

[0063] In one optional embodiment of this application, the meteorological reanalysis data includes meteorological elements; When the preset altitude layer is the ground layer, the meteorological elements include at least one of sea level pressure, air temperature, and wind field elements; When the preset height layer is 850hPa, 700hPa or 500hPa, the meteorological elements include at least one of geopotential height, air temperature and wind field.

[0064] In one optional embodiment of this application, determining the weather type corresponding to the at least one weather map based on the output of the weather type classification model includes: Receive multiple candidate weather types and their corresponding confidence scores output by the weather type classification model for the at least one weather situation map; The candidate weather types are sorted according to the confidence scores, and the top N ranked candidate weather types are selected as the target weather types, where N is a preset integer greater than 1.

[0065] Specific limitations regarding the aforementioned AI-based automatic spatiotemporal weather pattern classification system 300 can be found in the limitations of the AI-based automatic spatiotemporal weather pattern classification method described above, and will not be repeated here. Each module in the aforementioned AI-based automatic spatiotemporal weather pattern classification system 300 can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module.

[0066] In one embodiment, a computer device is provided, comprising a processor, a memory, a network interface, and a database connected via a system bus. The processor provides computational and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The database stores data. The network interface communicates with external terminals via a network connection. When the computer program is executed by the processor, it implements the above-described AI-based automatic spatiotemporal weather pattern classification method. The method includes a memory and a processor; the memory stores a computer program; and the processor, when executing the computer program, implements any step of the above-described AI-based automatic spatiotemporal weather pattern classification method.

[0067] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, can perform any of the steps in the above-described AI-based method for automatic classification of spatiotemporal weather patterns.

[0068] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0069] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0070] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0071] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0072] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

[0073] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. An automatic classification method for spatiotemporal weather patterns based on artificial intelligence, characterized in that, include: Obtain at least one weather map; Input at least one weather map into the preset weather type classification model; The weather type corresponding to the at least one weather map is determined based on the output of the weather type classification model, wherein the weather type classification model is trained through the following steps: Obtain a historical weather map dataset; wherein: the historical weather map dataset includes multiple historical weather maps and weather type and time series labels corresponding to each historical weather map; The historical weather map dataset is used as training samples to train the initial model through supervised learning, resulting in the weather type classification model.

2. The method according to claim 1, characterized in that, The weather type classification model is a multimodal architecture-based model, which includes: A visual encoder is used to encode the at least one weather map to extract visual features; A language model is used to process textual information related to weather types.

3. The method according to claim 2, characterized in that, The weather type classification model employs a multi-level attention mechanism to capture, in parallel, the macroscopic features of global atmospheric circulation and the detailed features of local weather systems in the weather situation map; and / or, The model employs a three-dimensional convolutional network to process and model temporal features of a set of continuous weather maps in time or space.

4. The method according to claim 3, characterized in that, The step of using the historical weather map dataset as training samples to perform supervised learning training on the initial model includes: Select samples of preset mainstream weather types from the historical weather map dataset; The visual encoder and aligner of the initial model are fully adjusted based on the samples of the mainstream weather types. The language model portion of the initial model is then subjected to lightweight adjustments.

5. The method according to claim 4, characterized in that, The step of using the historical weather map dataset as training samples to perform supervised learning training on the initial model further includes: Select preset long-tail weather type samples from the historical weather map dataset; With the parameters of the visual encoder and the aligner frozen, the language model portion of the initial model is lightly adjusted based on the long-tailed weather type samples.

6. The method according to claim 1, characterized in that, The acquisition of at least one weather map includes: Obtain at least one set of weather situation maps; wherein each weather situation map in at least one set of weather situation maps corresponds to the same period and corresponds to multiple different preset height layers; the preset height layers are at least one of the preset ground layer, 850hPa height layer, 700hPa height layer and 500hPa height layer.

7. The method according to claim 1, characterized in that, Before inputting at least one weather map into the preset weather type classification model, the method further includes: In response to an instruction to acquire meteorological reanalysis data, at least one weather pattern map is generated based on the meteorological reanalysis data.

8. The method according to claim 7, characterized in that, The meteorological reanalysis data includes meteorological elements; When the preset altitude layer is the ground layer, the meteorological elements include at least one of sea level pressure, air temperature, and wind field elements; When the preset height layer is 850hPa, 700hPa or 500hPa, the meteorological elements include at least one of geopotential height, air temperature and wind field.

9. The method according to claim 1, characterized in that, The determination of the weather type corresponding to the at least one weather map based on the output of the weather type classification model includes: Receive multiple candidate weather types and their corresponding confidence scores output by the weather type classification model for the at least one weather situation map; The candidate weather types are sorted according to the confidence scores, and the top N ranked candidate weather types are selected as the target weather types, where N is a preset integer greater than 1.

10. An automatic weather pattern classification system based on artificial intelligence in spatiotemporal sequences, characterized in that, At least including: The acquisition module is used to acquire at least one weather map. The determination module is used to input the at least one weather situation map into a preset weather type classification model; Based on the output of the weather type classification model, the weather type corresponding to the at least one weather map is determined. The training module is used to train the weather type classification model through the following steps: obtaining a historical weather map dataset; wherein: the historical weather map dataset includes multiple historical weather maps and weather type labels and time series labels corresponding to each historical weather map; using the historical weather map dataset as training samples, supervising the initial model to obtain the weather type classification model.

Images