Time series feature analysis method and system based on multi-dimensional data

Abstract

The application relates to the technical field of data processing, and discloses a time sequence feature analysis method and system based on multidimensional data. The method comprises the following steps: performing space-time feature extraction and dimension reduction processing on a multidimensional time sequence precipitation data set to obtain a precipitation space-time feature vector set; performing multi-scale space-time pattern recognition and clustering analysis on the precipitation space-time feature vector set to obtain a precipitation space-time pattern library, clustering results and an abnormal precipitation pattern list; constructing a deep learning precipitation prediction model to obtain an end-to-end global atmospheric reanalysis precipitation fusion prediction model; dynamically updating and online learning the global atmospheric reanalysis precipitation fusion prediction model to obtain a target precipitation space-time prediction model; and performing multidimensional space-time analysis on the target precipitation space-time prediction model to obtain a global atmospheric reanalysis precipitation fusion data analysis report and a weather processing pipeline scheduling strategy. The application improves the accuracy of time sequence feature analysis based on multidimensional data.

Description

Technical Field

[0001] This application relates to the field of data processing, and in particular to a method and system for time series feature analysis based on multidimensional data. Background Technology

[0002] In existing technologies, weather observation and forecasting systems typically rely on a single data source, such as ground-based observation stations or satellite remote sensing data. These systems forecast precipitation by analyzing this single data source; however, data from a single source has limitations and cannot fully reflect the complex changes in the atmosphere. Furthermore, existing precipitation forecasting models primarily rely on traditional statistical and numerical simulation methods, which struggle to capture complex nonlinear spatiotemporal characteristics, resulting in low forecast accuracy. Simultaneously, traditional precipitation forecasting methods lack dynamic updates and adaptive learning capabilities, failing to respond promptly to weather changes, thus impacting the accuracy and timeliness of forecasts.

[0003] The shortcomings of existing technologies lie in the lack of an effective multi-source data fusion method that can fully utilize multi-dimensional time-series data from weather stations, satellite remote sensing, and ground-based observation stations to improve the accuracy of precipitation forecasts. Furthermore, existing methods lack in-depth analysis and mining of the spatiotemporal characteristics of precipitation, failing to effectively identify anomalous precipitation patterns and multi-scale spatiotemporal patterns, resulting in limited predictive capabilities for extreme weather events. Additionally, existing precipitation forecasting models lack dynamic updating and online learning capabilities, making it difficult to adapt to rapidly changing weather conditions and providing high-precision, real-time precipitation forecasts. Summary of the Invention

[0004] This application provides a method and system for time series feature analysis based on multidimensional data, which can improve the accuracy of time series feature analysis based on multidimensional data.

[0005] Firstly, this application provides a method for time-series feature analysis based on multidimensional data. This method includes: collecting and preprocessing multi-source raw data from weather observation stations, satellite remote sensing, and ground observation stations to obtain a standardized multidimensional time-series precipitation dataset; extracting and reducing the spatiotemporal features of the multidimensional time-series precipitation dataset to obtain a precipitation spatiotemporal feature vector set; performing multi-scale spatiotemporal pattern recognition and cluster analysis on the precipitation spatiotemporal feature vector set to obtain a precipitation spatiotemporal pattern library, clustering results, and a list of anomalous precipitation patterns; constructing a deep learning precipitation prediction model based on the precipitation spatiotemporal pattern library, the clustering results, and the list of anomalous precipitation patterns to obtain an end-to-end global atmospheric reanalysis precipitation fusion prediction model; dynamically updating and learning the global atmospheric reanalysis precipitation fusion prediction model online to obtain a target precipitation spatiotemporal prediction model; and performing multidimensional spatiotemporal analysis on the target precipitation spatiotemporal prediction model to obtain a global atmospheric reanalysis precipitation fusion data analysis report and a weather processing pipeline scheduling strategy.

[0006] Secondly, this application provides a time series feature analysis system based on multidimensional data, the time series feature analysis system based on multidimensional data comprising:

[0007] The acquisition module is used to collect and preprocess raw data from multiple sources, including weather observation stations, satellite remote sensing, and ground observation stations, to obtain a standardized multidimensional time-series precipitation dataset.

[0008] The dimensionality reduction module is used to extract spatiotemporal features and perform dimensionality reduction on the multidimensional time-series precipitation dataset to obtain a set of precipitation spatiotemporal feature vectors.

[0009] The clustering module is used to perform multi-scale spatiotemporal pattern recognition and clustering analysis on the precipitation spatiotemporal feature vector set to obtain a precipitation spatiotemporal pattern library, clustering results, and a list of anomalous precipitation patterns.

[0010] The construction module is used to build a deep learning precipitation prediction model based on the precipitation spatiotemporal pattern library, the clustering results and the list of anomalous precipitation patterns, so as to obtain an end-to-end global atmospheric reanalysis precipitation fusion prediction model.

[0011] The update module is used to dynamically update and learn online the global atmospheric reanalysis precipitation fusion prediction model to obtain the target precipitation spatiotemporal prediction model.

[0012] The analysis module is used to perform multidimensional spatiotemporal analysis on the target precipitation spatiotemporal prediction model to obtain a global atmospheric reanalysis precipitation fusion data analysis report and a weather processing pipeline scheduling strategy.

[0013] The technical solution provided in this application obtains a standardized multidimensional time-series precipitation dataset by collecting and preprocessing raw data from multiple sources, including weather observation stations, satellite remote sensing, and ground observation stations. This dataset integrates the advantages of multiple data sources, avoiding the limitations of a single data source and making the data more comprehensive and accurate. Based on this, spatiotemporal feature extraction and dimensionality reduction are performed on the multidimensional time-series precipitation dataset to obtain a precipitation spatiotemporal feature vector set. This effectively captures the complex spatiotemporal characteristics of precipitation, improving data expressiveness and processing efficiency. Secondly, multi-scale spatiotemporal pattern recognition and clustering analysis are performed on the precipitation spatiotemporal feature vector set to construct a precipitation spatiotemporal pattern library, clustering results, and a list of anomalous precipitation patterns. This not only helps to deeply explore the spatiotemporal patterns of precipitation but also identifies anomalous precipitation patterns, improving the prediction ability for extreme weather events. Furthermore, based on the precipitation spatiotemporal pattern library, clustering results, and anomalous precipitation pattern list, a deep learning precipitation prediction model is constructed, resulting in an end-to-end global atmospheric reanalysis precipitation fusion prediction model. This deep learning model effectively utilizes large-scale data for training, capturing the complex nonlinear characteristics of precipitation and significantly improving the accuracy of precipitation forecasts. Through dynamic updates and online learning, the global atmospheric reanalysis precipitation fusion prediction model is optimized to obtain a target precipitation spatiotemporal prediction model. This allows the model to adaptively adjust parameters and respond promptly to weather changes, improving the real-time performance and accuracy of forecasts. Furthermore, multi-dimensional spatiotemporal analysis of the target precipitation spatiotemporal prediction model generates a global atmospheric reanalysis precipitation fusion data analysis report and a weather processing pipeline scheduling strategy. It enables comprehensive evaluation and analysis of precipitation forecast results. This invention, through the fusion of multi-source data, spatiotemporal feature extraction, multi-scale spatiotemporal pattern recognition, and the construction and optimization of deep learning models, not only significantly improves the accuracy and real-time performance of precipitation forecasts but also enhances the predictive ability for extreme weather events. Attached Figure Description

[0014] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0015] Figure 1 This is a schematic diagram of an embodiment of the time-series feature analysis method based on multidimensional data in this application.

[0016] Figure 2 This is a schematic diagram of an embodiment of the time series feature analysis system based on multidimensional data in this application. Detailed Implementation

[0017] This application provides a method and system for time-series feature analysis based on multidimensional data. The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in a sequence other than that illustrated or described herein. Furthermore, the terms "comprising" or "having" and any variations thereof are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0018] For ease of understanding, the specific process of the embodiments of this application is described below. Please refer to [link / reference]. Figure 1 One embodiment of the time-series feature analysis method based on multidimensional data in this application includes:

[0019] Step S101: Collect and preprocess the multi-source raw data from weather observation stations, satellite remote sensing and ground observation stations to obtain a standardized multidimensional time-series precipitation dataset;

[0020] Step S102: Extract spatiotemporal features and perform dimensionality reduction on the multidimensional time-series precipitation dataset to obtain a set of spatiotemporal feature vectors of precipitation.

[0021] Step S103: Perform multi-scale spatiotemporal pattern recognition and cluster analysis on the spatiotemporal feature vector set of precipitation to obtain a spatiotemporal pattern library of precipitation, clustering results and a list of anomalous precipitation patterns;

[0022] Step S104: Based on the precipitation spatiotemporal pattern library, clustering results and a list of anomalous precipitation patterns, construct a deep learning precipitation prediction model to obtain an end-to-end global atmospheric reanalysis precipitation fusion prediction model.

[0023] Step S105: Dynamically update and learn online the global atmospheric reanalysis precipitation fusion prediction model to obtain the target precipitation spatiotemporal prediction model;

[0024] Step S106: Perform multidimensional spatiotemporal analysis on the target precipitation spatiotemporal prediction model to obtain a global atmospheric reanalysis precipitation fusion data analysis report and a weather processing pipeline scheduling strategy.

[0025] It is understood that the executing entity of this application can be a time-series feature analysis system based on multi-dimensional data, or it can be a terminal or a server; the specific implementation is not limited here. This application's embodiments use a server as an example for illustration.

[0026] Specifically, multi-source raw data from weather observation stations, satellite remote sensing, and ground observation stations are collected and preprocessed to obtain a standardized multidimensional time-series precipitation dataset. The collected data includes various types of precipitation data from different times and locations, such as precipitation amount, precipitation intensity, and precipitation frequency. These data often have different timestamps and spatial coordinates, thus requiring timestamp calibration and spatial coordinate transformation. Timestamp calibration synchronizes the time of different data sources, ensuring data consistency; spatial coordinate transformation converts the data to a unified projection system, facilitating subsequent processing. After time synchronization and spatial transformation, the data undergoes quality assessment, identifying defective or abnormal data points, and multi-source data fusion is performed. During multi-source data fusion, the Kalman filter algorithm is used to eliminate noise, resulting in denoised and optimized data. The Kalman filter algorithm is an adaptive filtering algorithm that effectively reduces the impact of noise and improves data accuracy. Next, the denoised and optimized data undergoes spatiotemporal scale consistency adjustment to ensure spatiotemporal consistency, and then data discretization is performed to obtain discretized time-series data. Finally, multidimensional feature encoding is performed on the discretized time-series data to form a standardized multidimensional time-series precipitation dataset.

[0027] After obtaining a standardized multidimensional time-series precipitation dataset, spatiotemporal feature extraction and dimensionality reduction are performed to obtain a set of spatiotemporal feature vectors for precipitation. First, fractal dimension calculation is performed on the multidimensional time-series precipitation dataset to obtain spatiotemporal complexity features. Fractal dimension is an indicator of data complexity, reflecting the spatiotemporal complexity of precipitation data. Next, symbolic dynamics encoding is performed to obtain symbolic sequence features. Symbolic dynamics encoding is a method that converts continuous time series into discrete symbolic sequences, capturing the dynamic characteristics of time series. Phase space is reconstructed and embedded into the multidimensional time-series precipitation dataset to obtain phase space trajectory features. Phase space reconstruction is a nonlinear time series analysis method that reveals the intrinsic structure and dynamic characteristics of time series. Then, Lyapunov exponents are calculated on the phase space trajectory features to obtain system stability features. Lyapunov exponents are an indicator of system stability, reflecting the stability of precipitation systems. Finally, recursive quantitative analysis is performed to obtain system periodicity features. Recursive quantitative analysis is a tool for analyzing the periodicity of time series, revealing the periodic characteristics of precipitation data. Multi-scale entropy analysis is performed to obtain temporal complexity characteristics. Multi-scale entropy analysis is an indicator used to measure the complexity of time series data, capturing the multi-scale complexity of precipitation data. Time-frequency analysis is then conducted on the multi-dimensional time-series precipitation dataset using the Hilbert-Huang transform to obtain nonlinear time-frequency characteristics. The Hilbert-Huang transform is a nonlinear method for time-frequency analysis that reveals the time-frequency properties of precipitation data. Finally, lag mutual information analysis is performed to obtain spatiotemporal dependency characteristics. Lag mutual information analysis is a tool used to measure dependencies between time series data, revealing the spatiotemporal dependencies of precipitation data. These spatiotemporal characteristics are then subjected to nonlinear dimensionality reduction to obtain a set of precipitation spatiotemporal feature vectors.

[0028] After obtaining the spatiotemporal feature vector set of precipitation, multi-scale spatiotemporal pattern recognition and clustering analysis are performed to obtain a spatiotemporal pattern library of precipitation, clustering results, and a list of anomalous precipitation patterns. First, the spatiotemporal feature vector set of precipitation is segmented by a spatiotemporal sliding window to obtain a set of multi-scale feature fragments. Spatiotemporal sliding window segmentation is a method for segmenting spatiotemporal data, capable of capturing the multi-scale features of precipitation data. Next, spatiotemporal pattern encoding is performed to obtain pattern encoding sequences. Spatiotemporal pattern encoding is a method for converting spatiotemporal features into discrete patterns, capable of capturing the spatiotemporal patterns of precipitation data. Then, frequent pattern mining is performed to obtain a candidate spatiotemporal pattern set. Frequent pattern mining is a tool for discovering frequently occurring patterns in data, capable of revealing common patterns in precipitation data. Next, spatiotemporal association rule analysis is performed to obtain spatiotemporal association patterns. Spatiotemporal association rule analysis is a method for discovering spatiotemporal association patterns in data, capable of revealing the spatiotemporal correlation of precipitation data. Finally, the spatiotemporal association patterns are clustered using a spectral clustering algorithm to obtain preliminary clustering results. Spectral clustering is a graph-based clustering method that effectively clusters spatiotemporal data. Then, fuzzy membership calculation is performed to obtain soft cluster assignments. Fuzzy membership calculation is a method for measuring the degree to which data points belong to different clusters, revealing fuzzy clustering relationships in precipitation data. Next, density peak detection is performed to obtain cluster centers. Density peak detection is a density-based clustering method that can discover cluster centers in the data. Based on the cluster centers, the spatiotemporal feature vector set of precipitation is redistributed to obtain the clustering result. The clustering result is an improved clustering result that better reveals the clustering structure of precipitation data. Then, outlier detection is performed to obtain a candidate set of anomalous precipitation patterns. Outlier detection is a method for discovering anomalous patterns in data, revealing anomalous patterns in precipitation data. The isolated forest algorithm is used to measure the anomalousness of the candidate set of anomalous precipitation patterns, resulting in a list of anomalous precipitation patterns. The isolated forest algorithm is a machine learning algorithm for anomaly detection that effectively measures the degree of data anomalousness. Finally, the spatiotemporal correlation patterns, clustering results, and a list of anomalous precipitation patterns are integrated to obtain a spatiotemporal precipitation pattern library.

[0029] Based on a spatiotemporal precipitation pattern library, clustering results, and a list of anomalous precipitation patterns, a deep learning precipitation prediction model is constructed, resulting in an end-to-end global atmospheric reanalysis precipitation fusion prediction model. First, pattern embedding is performed on the spatiotemporal precipitation pattern library to obtain a pattern embedding matrix. Pattern embedding is a method that converts patterns into vector representations, capturing the pattern features of precipitation data. Next, feature mapping is performed on the clustering results to obtain a clustering feature tensor. Feature mapping is a method that converts clustering results into feature representations, capturing the clustering features of precipitation data. Then, anomalous pattern encoding is performed on the list of anomalous precipitation patterns to obtain anomalous feature vectors. Anomalous pattern encoding is a method that converts anomalous patterns into vector representations, capturing the anomalous features of precipitation data. Next, the pattern embedding matrix, clustering feature tensor, and anomalous feature vectors are fused using multimodal features to obtain fused feature representation data. Multimodal feature fusion is a method that combines multiple feature representations to capture the comprehensive features of precipitation data. Finally, spatiotemporal dependency modeling is performed on the fused feature representation data using a graph convolutional network to obtain a spatiotemporal relationship graph. Graph convolutional networks (GCNNs) are deep learning algorithms based on graph structures that effectively model the spatiotemporal dependencies of data. Then, an attention mechanism is applied to the spatiotemporal graph to calculate weighted spatiotemporal features. The attention mechanism is a method for weighting the importance of features, which can improve the feature representation capability of precipitation data. A bidirectional long short-term memory (BSSM) network is used to perform sequence modeling on the weighted spatiotemporal features, yielding time-series prediction results. BSSM is a deep learning algorithm for processing sequential data, capable of capturing the temporal characteristics of precipitation data. Finally, deconvolution is performed on the time-series prediction results to obtain an end-to-end global atmospheric reanalysis precipitation fusion prediction model. Deconvolution is a method for restoring the spatial structure of data, which can improve the accuracy of precipitation prediction.

[0030] A global atmospheric reanalysis precipitation fusion prediction model is dynamically updated and learned online to obtain a spatiotemporal prediction model for the target precipitation. First, real-time precipitation observation data is segmented into time-series data segments. Data stream segmentation is a method of dividing continuous data into segments, which can capture the temporal changes in precipitation data. Next, concept drift detection is performed to obtain data distribution change indicators. Concept drift detection is a method used to detect changes in data distribution, which can reveal the changing characteristics of precipitation data. Based on the data distribution change indicators, parameter sensitivity analysis is performed on the global atmospheric reanalysis precipitation fusion prediction model to obtain a set of sensitive parameters. Parameter sensitivity analysis is a method used to evaluate the importance of model parameters, which can improve the adaptability of the precipitation prediction model. Then, incremental learning is performed on the sensitive parameter set to obtain an updated parameter vector. Incremental learning is a method of progressively updating the model, which can improve the real-time performance of the precipitation prediction model. Finally, the updated parameter vector is fused with the global atmospheric reanalysis precipitation fusion prediction model to obtain a provisional prediction model. Model fusion is a method of combining results from multiple models, which can improve the accuracy of precipitation prediction. Then, the temporary prediction model is compressed to obtain a lightweight prediction model. Model compression is a method to reduce model complexity and improve the computational efficiency of precipitation prediction. Through federated learning, the lightweight prediction model is trained in a distributed manner to obtain a multi-source collaborative optimization model. Federated learning is a distributed training method that improves the collaborative optimization capability of precipitation prediction models. Next, the multi-source collaborative optimization model is distilled to obtain a knowledge transfer model. Model distillation is a method to transfer knowledge from a large model to a small model, which improves the lightweight nature of precipitation prediction models. Then, the knowledge transfer model is trained with adversarial examples to obtain a robustness-enhanced model. Adversarial example training is a method to improve model robustness and enhance the anti-interference ability of precipitation prediction models. Finally, the robustness-enhanced model is subjected to multi-task learning to obtain a target precipitation spatiotemporal prediction model. Multi-task learning is a method to simultaneously optimize multiple tasks, which improves the overall performance of precipitation prediction models.

[0031] Multidimensional spatiotemporal analysis was performed on the target precipitation spatiotemporal prediction model to obtain a global atmospheric reanalysis precipitation fusion data analysis report and a weather processing pipeline scheduling strategy. First, the output of the target precipitation spatiotemporal prediction model was spatiotemporally decomposed to obtain multi-scale precipitation prediction results. Spatiotemporal decomposition is a method of breaking down data into different scales, which can capture the multi-scale characteristics of precipitation data. Next, spatiotemporal autocorrelation analysis was performed to obtain the evolution characteristics of the precipitation model. Spatiotemporal autocorrelation analysis is a method used to analyze the spatiotemporal correlation of data, which can reveal the evolution characteristics of precipitation data. Then, topological data analysis was performed to obtain the topological structure of the precipitation model. Topological data analysis is a method used to discover the topological structure of data, which can reveal the model structure of precipitation data. Based on the precipitation model topological structure, extreme event identification was performed to obtain an extreme precipitation risk assessment. Extreme event identification is a method used to discover extreme events, which can assess the risk of precipitation data. Finally, scenario simulation was performed to obtain multi-scenario precipitation impact predictions. Scenario simulation is a method used to simulate different scenarios, which can predict the impact of precipitation data. Through complex network analysis, a system vulnerability assessment of precipitation impact prediction under multiple scenarios is conducted to identify key influencing factors. Complex network analysis, a network-based analytical method, can assess the vulnerability of precipitation data. Then, causal inference is performed to obtain a precipitation system dynamic model. Causal inference is a method for discovering causal relationships, revealing the dynamic mechanisms of precipitation data. Based on the precipitation system dynamic model, a spatiotemporal knowledge graph is constructed to obtain a precipitation knowledge network. Spatiotemporal knowledge graph construction is a method for integrating spatiotemporal information to form a knowledge network of precipitation data. Finally, interpretability analysis is performed to obtain a global atmospheric reanalysis precipitation fusion data analysis report. Interpretability analysis is a method for interpreting model results, improving the interpretability of precipitation prediction. Based on the global atmospheric reanalysis precipitation fusion data analysis report, resource optimization allocation is performed to obtain a weather processing pipeline scheduling strategy. Resource optimization allocation is a method for optimizing resource allocation, improving the application effectiveness of precipitation prediction.

[0032] For example, in a practical application, multi-source raw precipitation data for a certain region is collected, including ground observation data from weather stations, satellite remote sensing data, and data from automatic weather stations. These data have different acquisition frequencies in time and different spatial distribution ranges. Through timestamp calibration and spatial coordinate transformation, these data are synchronized to the same time and spatial scales, forming unified multidimensional time-series data. Then, noise in the data is eliminated using the Kalman filter algorithm to obtain denoised optimized data. Next, the denoised optimized data undergoes spatiotemporal scale consistency adjustment and discretization processing to form a standardized multidimensional time-series precipitation dataset. Through fractal dimension calculation, symbolic dynamics encoding, phase space reconstruction, Lyapunov exponent calculation, recursive quantitative analysis, multi-scale entropy analysis, Hilbert-Huang transform, and hysteresis mutual information analysis, spatiotemporal complexity features, symbolic sequence features, phase space trajectory features, system stability features, system periodicity features, time-series complexity features, nonlinear time-frequency features, and spatiotemporal dependence features of the precipitation data are extracted. These features are then subjected to nonlinear dimensionality reduction processing to obtain a set of spatiotemporal feature vectors for precipitation. This study employs spatiotemporal sliding window segmentation, spatiotemporal pattern encoding, frequent pattern mining, spatiotemporal association rule analysis, spectral clustering algorithm, fuzzy membership calculation, density peak detection, outlier detection, and isolated forest algorithm to perform multi-scale spatiotemporal pattern recognition and clustering analysis on the spatiotemporal feature vector set of precipitation, resulting in a precipitation spatiotemporal pattern library, clustering results, and a list of anomalous precipitation patterns. Based on the precipitation spatiotemporal pattern library, clustering results, and anomalous precipitation pattern list, a deep learning precipitation prediction model is constructed. Through pattern embedding, feature mapping, anomalous pattern encoding, multimodal feature fusion, graph convolutional networks, attention mechanisms, bidirectional long short-term memory networks, and deconvolution operations, an end-to-end global atmospheric reanalysis precipitation fusion prediction model is obtained. This model is dynamically updated and learned online through data stream segmentation, concept drift detection, parameter sensitivity analysis, incremental learning, model fusion, model compression, federated learning, model distillation, adversarial example training, and multi-task learning to form a target precipitation spatiotemporal prediction model. Finally, through spatiotemporal decomposition, spatiotemporal autocorrelation analysis, topological data analysis, extreme event identification, scenario simulation, complex network analysis, causal inference, spatiotemporal knowledge graph construction, and interpretability analysis, a multidimensional spatiotemporal analysis of the target precipitation prediction model is conducted. This results in a global atmospheric reanalysis precipitation fusion data analysis report, which is used to optimize resource allocation and formulate weather processing pipeline scheduling strategies. In practical applications, these methods can be used to accurately predict future precipitation in a region, identify potential extreme precipitation events, and provide optimized resource scheduling strategies to effectively address the impacts of precipitation.

[0033] In this embodiment, a standardized multidimensional time-series precipitation dataset is obtained by collecting and preprocessing raw data from multiple sources, including weather observation stations, satellite remote sensing, and ground observation stations. This dataset integrates the advantages of multiple data sources, avoiding the limitations of a single data source, and making the data more comprehensive and accurate. Based on this, spatiotemporal feature extraction and dimensionality reduction are performed on the multidimensional time-series precipitation dataset to obtain a precipitation spatiotemporal feature vector set. This effectively captures the complex spatiotemporal characteristics of precipitation, improving data expressiveness and processing efficiency. Secondly, multi-scale spatiotemporal pattern recognition and clustering analysis are performed on the precipitation spatiotemporal feature vector set to construct a precipitation spatiotemporal pattern library, clustering results, and a list of anomalous precipitation patterns. This not only helps to deeply explore the spatiotemporal patterns of precipitation but also identifies anomalous precipitation patterns, improving the prediction ability for extreme weather events. Furthermore, based on the precipitation spatiotemporal pattern library, clustering results, and anomalous precipitation pattern list, a deep learning precipitation prediction model is constructed, resulting in an end-to-end global atmospheric reanalysis precipitation fusion prediction model. This deep learning model effectively utilizes large-scale data for training, capturing the complex nonlinear characteristics of precipitation and significantly improving the accuracy of precipitation forecasts. Through dynamic updates and online learning, the global atmospheric reanalysis precipitation fusion prediction model is optimized to obtain a target precipitation spatiotemporal prediction model. This allows the model to adaptively adjust parameters and respond promptly to weather changes, improving the real-time performance and accuracy of forecasts. Furthermore, multi-dimensional spatiotemporal analysis of the target precipitation spatiotemporal prediction model generates a global atmospheric reanalysis precipitation fusion data analysis report and a weather processing pipeline scheduling strategy. It enables comprehensive evaluation and analysis of precipitation forecast results. This invention, through the fusion of multi-source data, spatiotemporal feature extraction, multi-scale spatiotemporal pattern recognition, and the construction and optimization of deep learning models, not only significantly improves the accuracy and real-time performance of precipitation forecasts but also enhances the predictive ability for extreme weather events.

[0034] In one specific embodiment, the process of performing step S101 may specifically include the following steps:

[0035] (1) Timestamp calibration is performed on the multi-source raw data to obtain time-synchronized data, and spatial coordinate transformation is performed on the time-synchronized data to obtain unified projection data;

[0036] (2) Perform data quality assessment on the unified projection data to obtain quality-labeled data;

[0037] (3) Multi-source data fusion is performed on the quality label data to obtain preliminary fused data;

[0038] (4) Noise is eliminated from the initial fused data using the Kalman filter algorithm to obtain noise-reduced optimized data;

[0039] (5) Adjust the spatiotemporal scale consistency of the noise reduction optimization data to obtain scale-uniform data;

[0040] (6) The scale-uniform data is discretized to obtain discretized time series data, and the discretized time series data is encoded with multidimensional features to obtain a standardized multidimensional time series precipitation dataset.

[0041] Specifically, timestamp calibration is performed on multi-source raw data to obtain time-synchronized data. The purpose of timestamp calibration is to ensure that data from different sources are compared and analyzed under the same time base. This involves converting the timestamps of all data to a unified time format, typically UTC (Coordinated Universal Time). During calibration, time deviations caused by clock errors from different devices need to be addressed using interpolation methods or time alignment algorithms to ensure data consistency in the time dimension. After obtaining the time-synchronized data, spatial coordinate transformation is performed to convert the data to a unified spatial reference system. The purpose of spatial coordinate transformation is to spatially align data from different sources, typically using a geographic coordinate system (such as WGS84) as a unified projection reference system. During the transformation, the transformation relationships between different coordinate systems need to be considered. Coordinate transformation is performed using mathematical transformation formulas or GIS (Geographic Information System) tools to ensure data consistency in the spatial dimension, ultimately resulting in unified projected data.

[0042] Based on uniformly projected data, data quality assessment is performed to identify outliers or missing values, resulting in quality-labeled data. The purpose of data quality assessment is to ensure the reliability and accuracy of the data, typically using statistical analysis methods or machine learning algorithms. Statistical analysis methods include calculating basic statistics (such as mean, variance, and extreme values) to identify obvious outliers; machine learning algorithms include anomaly detection algorithms (such as Isolation Forest and Support Vector Machines) to identify abnormal patterns in the data through model training. After assessment, outliers or missing values ​​are marked in the quality-labeled data for correction or removal in subsequent processing.

[0043] Multi-source data fusion is performed on quality-labeled data to obtain preliminary fused data. The purpose of multi-source data fusion is to comprehensively utilize data from different sources to improve the comprehensiveness and accuracy of the data. During the fusion process, a weighted average method or a Bayesian fusion method is used to weight the multi-source data according to their weights, resulting in the fused data. The weighted average method assigns weights to data from different sources and sums the data according to these weights; the Bayesian fusion method is based on a Bayesian probability model, fusing data while considering prior probabilities and likelihood functions to obtain the probability distribution of the fused data. After fusion, noise removal is performed on the preliminary fused data to improve its quality. The Kalman filter algorithm is used to remove noise from the preliminary fused data, resulting in denoised optimized data. The Kalman filter algorithm is a recursive estimation algorithm based on a state-space model, which can effectively filter out noise from data while retaining its effective information. The specific operation includes constructing a state-space model, setting state transition equations and observation equations, and using the Kalman filter algorithm to recursively filter the data to obtain optimized data. The advantage of the Kalman filter algorithm is that it can adaptively adjust the filter parameters, process non-stationary data, and has a good filtering effect.

[0044] After obtaining the denoising-optimized data, spatiotemporal scale consistency adjustment is performed to obtain scale-uniform data. The purpose of spatiotemporal scale consistency adjustment is to standardize the data in both time and space, ensuring consistency across different scales. Specific operations include temporal scale adjustment (such as sampling rate conversion and time window smoothing) and spatial scale adjustment (such as spatial interpolation and spatial resampling). Temporal scale adjustment achieves consistency in the data over time by changing the sampling frequency or performing time window smoothing; spatial scale adjustment achieves consistency in the data over the spatial scale by performing spatial interpolation or resampling.

[0045] Data discretization is performed on scale-uniform data to obtain discretized time-series data. The purpose of data discretization is to convert continuous data into discrete data to facilitate subsequent feature extraction and analysis. Specific operations include equidistant discretization or equal-frequency discretization, dividing the data into several discrete intervals. Equidistant discretization divides the data at fixed intervals to obtain discrete intervals; equal-frequency discretization divides the data according to frequency, ensuring that each interval contains approximately the same amount of data.

[0046] Finally, multidimensional feature encoding is performed on the discretized time-series data to obtain a standardized multidimensional time-series precipitation dataset. The purpose of multidimensional feature encoding is to transform discrete data into multidimensional feature vectors, facilitating subsequent spatiotemporal feature analysis. Specific operations include feature extraction and encoding of the data, converting the spatiotemporal features of the data into multidimensional feature vectors. Feature extraction methods include spatiotemporal sliding window, symbolic dynamics encoding, and phase space reconstruction; encoding methods include one-hot encoding and embedded vector encoding. Through these methods, the spatiotemporal features of the data are encoded to form multidimensional feature vectors, resulting in a standardized multidimensional time-series precipitation dataset.

[0047] For example, when collecting multi-source raw precipitation data for a region, including observations from ground weather stations, satellite remote sensing data, and data from automatic weather stations, these data have different timestamps and spatial coordinates. Through timestamp calibration, all data are converted to a unified UTC time format. Then, through spatial coordinate transformation, the data is converted to a unified WGS84 coordinate system. Quality assessment is performed on the unified projection data, using the Isolation Forest algorithm to identify outliers and missing values, resulting in quality-labeled data. Next, multi-source data fusion is performed on the quality-labeled data. A weighted average method is used to sum the data from different sources according to their weights, resulting in preliminary fused data. Noise is eliminated from the preliminary fused data using the Kalman filter algorithm. A state-space model is constructed, and recursive filtering is applied to obtain denoised optimized data. Spatiotemporal scale consistency adjustment is performed on the denoised optimized data. Through time window smoothing and spatial interpolation methods, the data achieves consistency in both time and spatial scales, resulting in scale-uniformed data. Finally, the scale-uniformed data is discretized at equal intervals, dividing the data into several discrete intervals, resulting in discretized time-series data. Finally, multidimensional feature encoding was performed on the discretized time-series data. Symbolic dynamics encoding and phase space reconstruction methods were used to transform the spatiotemporal characteristics of the data into multidimensional feature vectors, forming a standardized multidimensional time-series precipitation dataset. In this process, through steps such as timestamp calibration, spatial coordinate transformation, data quality assessment, multi-source data fusion, Kalman filtering, spatiotemporal scale consistency adjustment, data discretization, and multidimensional feature encoding, the multi-source raw data was successfully processed into a standardized multidimensional time-series precipitation dataset, ensuring data quality and consistency.

[0048] In one specific embodiment, the process of performing step S102 may specifically include the following steps:

[0049] (1) The fractal dimension of the multidimensional time-series precipitation dataset is calculated to obtain the spatiotemporal complexity characteristics;

[0050] (2) Symbolic dynamics encoding is performed on the multidimensional time-series precipitation dataset to obtain symbolic sequence features;

[0051] (3) By reconstructing the phase space, the multidimensional time-series precipitation dataset is embedded to obtain phase space trajectory features;

[0052] (4) Calculate the Lyapunov exponent on the phase space trajectory characteristics to obtain the system stability characteristics;

[0053] (5) Recursive quantitative analysis was performed on the multidimensional time-series precipitation dataset to obtain the periodic characteristics of the system;

[0054] (6) Perform multi-scale entropy analysis on the multi-dimensional time-series precipitation dataset to obtain the time-series complexity characteristics;

[0055] (7) Time-frequency analysis of the multidimensional time-series precipitation dataset was performed using Hilbert-Huang transform to obtain nonlinear time-frequency characteristics;

[0056] (8) Perform hysteresis mutual information analysis on the multidimensional time-series precipitation dataset to obtain spatiotemporal dependence characteristics;

[0057] (9) Nonlinear dimensionality reduction is performed on the spatiotemporal complexity features, symbol sequence features, system stability features, system periodicity features, time sequence complexity features, nonlinear time-frequency features and spatiotemporal dependence features to obtain the precipitation spatiotemporal feature vector set.

[0058] Specifically, fractal dimension calculation of a multidimensional time-series precipitation dataset is performed to obtain spatiotemporal complexity characteristics. Fractal dimension is a method for measuring complexity, reflecting the complex structure of precipitation data in time and space. In the calculation process, the precipitation data is first segmented, and then the box-counting method is used to calculate the fractal dimension of each segment. The box-counting method calculates the fractal dimension by embedding data into boxes of different sizes and counting the number of data points in each box. The larger the fractal dimension, the higher the spatiotemporal complexity of the data. Symbolic dynamics encoding of the multidimensional time-series precipitation dataset is performed to obtain symbolic sequence characteristics. Symbolic dynamics encoding is a method for converting continuous data sequences into discrete symbolic sequences, simplifying data representation while preserving its main features. In the encoding process, the data is first standardized, and then converted into a symbolic sequence according to a predefined threshold. For example, setting the threshold as high, medium, and low precipitation levels, high precipitation is encoded as A, medium precipitation as B, and low precipitation as C, ultimately resulting in a symbolic sequence composed of A, B, and C.

[0059] Phase space trajectory features can be obtained by embedding multidimensional time-series precipitation datasets into reconstructed phase space. Phase space reconstruction is a nonlinear time series analysis method that can reveal the intrinsic dynamic characteristics of the data. In the reconstruction process, the delay time and embedding dimension are first determined, and then the original time series is delayed and embedded into a high-dimensional phase space to form a phase space trajectory. The delay time and embedding dimension can be determined using an autocorrelation function or a pseudo nearest neighbor algorithm. The phase space trajectory reflects the dynamic behavior of the data, providing a foundation for subsequent feature extraction. The Lyapunov exponent is calculated on the phase space trajectory features to obtain system stability characteristics. The Lyapunov exponent is an index that measures the stability of a dynamic system and reflects the system's sensitivity to initial conditions. In the calculation process, several initial points are first selected on the phase space trajectory, and then the evolution trajectories of these points in the phase space are tracked, calculating the exponential separation velocity between the trajectories. When the Lyapunov exponent is positive, the system is highly sensitive to initial conditions and exhibits chaotic characteristics; when it is negative, the system tends to be stable.

[0060] Recursive quantitative analysis of multidimensional time-series precipitation datasets aims to reveal the periodic characteristics of the system. Recursive quantitative analysis is a tool used to analyze the periodicity of time series data, revealing periodic behavior within the data. The analysis first constructs a recurrence plot, mapping the time series onto a two-dimensional matrix. By calculating the Euclidean distance between different time points, it is determined whether the data is recursive. The periodic structure within the recurrence plot reflects the periodic characteristics of the data, such as period length and period intensity.

[0061] Multi-scale entropy analysis of multidimensional time-series precipitation datasets is performed to obtain temporal complexity characteristics. Multi-scale entropy analysis is an indicator of time series complexity, capturing the complexity of data at different scales. The analysis first decomposes the time series at multiple scales, for example, through wavelet transform or empirical mode decomposition, dividing the data into subsequences at different time scales, and then calculates the entropy value of each subsequence. A higher entropy value indicates higher data complexity and a greater amount of information contained within the data. Time-frequency analysis of multidimensional time-series precipitation datasets using Hilbert-Huang transform is performed to obtain nonlinear time-frequency characteristics. Hilbert-Huang transform is a nonlinear and non-stationary time series analysis method that reveals the time-frequency properties of data. The analysis first decomposes the time series into several intrinsic mode functions (IMFs) through empirical mode decomposition, and then performs a Hilbert transform on each IMF to obtain the instantaneous frequency and instantaneous amplitude. By plotting time-frequency graphs, the distribution of data at different times and frequencies can be visually observed.

[0062] Lag mutual information analysis (LMI) is performed on multidimensional time-series precipitation datasets to obtain spatiotemporal dependency characteristics. LMI is a tool for measuring the dependency between time series, revealing the spatiotemporal correlation between data. The analysis first involves lag embedding the time series to form lagged sequences, and then calculating the mutual information at different lag times to reflect the information transfer and dependency between the two time series. A higher mutual information content indicates a stronger dependency between the two sequences.

[0063] Finally, nonlinear dimensionality reduction is performed on spatiotemporal complexity features, symbol sequence features, system stability features, system periodicity features, time-series complexity features, nonlinear time-frequency features, and spatiotemporal dependence features to obtain a set of precipitation spatiotemporal feature vectors. The purpose of nonlinear dimensionality reduction is to reduce the dimensionality of features, retain key information, and simplify data representation. Commonly used nonlinear dimensionality reduction methods include Principal Component Analysis (PCA), t-distributed neighborhood embedding (t-SNE), and autoencoders. During dimensionality reduction, high-dimensional feature vectors are first input into the dimensionality reduction algorithm. The algorithm is then optimized to reduce the dimensionality of the features, resulting in low-dimensional feature representations. These low-dimensional feature vectors retain most of the information from the original data while reducing computational complexity and improving the efficiency of subsequent analysis.

[0064] For example, a multidimensional time-series precipitation dataset of a certain region is collected. The fractal dimension of the data is calculated using box counting to determine the fractal dimension of each time period, yielding spatiotemporal complexity characteristics. Next, the data is encoded using symbolic dynamics, defining high, medium, and low precipitation levels, converting the data into symbolic sequences. Then, by reconstructing the phase space, the time series is delayed and embedded into a three-dimensional phase space, forming a phase space trajectory. The Lyapunov exponent is calculated on the phase space trajectory, and an initial point is selected to track the trajectory. The exponential separation velocity is calculated to obtain the system stability characteristics. Next, recursive quantitative analysis is performed on the data, constructing a recursion graph. The recursive relationship between time points is determined by calculating Euclidean distance, yielding the system's periodicity characteristics. Finally, multi-scale entropy analysis is performed on the data. Wavelet transform is used to decompose the data into subsequences at different time scales, and the entropy value of each subsequence is calculated to obtain the time-series complexity characteristics. Empirical Mode Decomposition (EMD) was performed on the data using the Hilbert-Huang Transform (HMT) to obtain several Intrinsic Mode Functions (IMFs). A Hilbert Transform was then applied to each IMF to obtain the instantaneous frequency and amplitude, forming a time-frequency plot and revealing nonlinear time-frequency characteristics. Finally, lag mutual information analysis was performed on the data. Lag sequences were formed through delayed embedding, and the mutual information at different lag times was calculated to obtain spatiotemporal dependence characteristics. These characteristics were then subjected to nonlinear dimensionality reduction using principal component analysis (PCA) to reduce the high-dimensional feature vectors to three-dimensional feature vectors, forming a spatiotemporal feature vector set for precipitation. In this process, through a series of feature extraction and analysis methods, the multidimensional time-series precipitation dataset was successfully transformed into a spatiotemporal feature vector set for precipitation.

[0065] In one specific embodiment, the process of executing step S103 may specifically include the following steps:

[0066] (1) Perform spatiotemporal sliding window segmentation on the spatiotemporal feature vector set of precipitation to obtain a multi-scale feature fragment set;

[0067] (2) Spatiotemporal pattern coding is performed on the multi-scale feature fragment set to obtain the pattern coding sequence;

[0068] (3) Perform frequent pattern mining on the pattern coding sequence to obtain a candidate spatiotemporal pattern set;

[0069] (4) Perform spatiotemporal correlation rule analysis on the candidate spatiotemporal pattern set to obtain spatiotemporal correlation patterns;

[0070] (5) Cluster the spatiotemporal correlation patterns using the spectral clustering algorithm to obtain preliminary clustering results;

[0071] (6) Perform fuzzy membership calculation on the preliminary clustering results to obtain soft clustering assignments;

[0072] (7) Density peak detection is performed on the soft clustering assignment to obtain the cluster centers;

[0073] (8) The spatiotemporal feature vector set of precipitation is redistributed based on the cluster centers to obtain the clustering results;

[0074] (9) Outlier detection is performed on the clustering results to obtain a candidate set of abnormal precipitation patterns;

[0075] (10) The abnormal precipitation pattern candidate set is quantified by the isolated forest algorithm to obtain the abnormal precipitation pattern list;

[0076] (11) Integrate the spatiotemporal correlation patterns, clustering results and a list of abnormal precipitation patterns to obtain a precipitation spatiotemporal pattern library.

[0077] Specifically, a crucial step is to perform spatiotemporal sliding window segmentation on the precipitation spatiotemporal feature vector set to obtain a multi-scale feature fragment set. Spatiotemporal sliding window segmentation divides the original spatiotemporal feature vector set into fixed-size fragments, facilitating the capture of multi-scale spatiotemporal features. Specifically, a fixed time window length and spatial window size are set, and then the window slides across the entire spatiotemporal dataset with a certain step size, extracting feature data within the window. For example, setting the time window length to 10 days and the spatial window to 10 square kilometers, sliding with a step size of 1 day and 1 square kilometer, extracting data within the window with each slide, forming a multi-scale feature fragment set. Spatiotemporal pattern encoding of the multi-scale feature fragment set is to transform complex spatiotemporal features into discrete patterns that are easy to analyze. Spatiotemporal pattern encoding encodes each fragment as a pattern label by defining several feature patterns. For example, precipitation intensity can be divided into several levels (such as high, medium, and low), and then encoded according to the precipitation intensity of each fragment, resulting in a pattern encoding sequence. This encoding process simplifies continuous spatiotemporal feature data into discrete pattern labels, facilitating subsequent pattern mining and analysis.

[0078] Frequent pattern mining of pattern encoding sequences aims to discover recurring spatiotemporal patterns in the data. Frequent pattern mining utilizes algorithms such as Apriori or FP-growth to identify sets of patterns that frequently appear in the dataset. Specifically, it involves scanning the encoding sequence, counting the frequency of each pattern and its combinations, setting a support threshold, and filtering out patterns with frequencies exceeding the threshold to form a candidate spatiotemporal pattern set. For example, during analysis, if a precipitation pattern (such as three consecutive days of high precipitation) is found to occur frequently in the dataset, such a pattern becomes a candidate spatiotemporal pattern.

[0079] Spatiotemporal association rule analysis of candidate spatiotemporal pattern sets aims to reveal the relationships between patterns. This analysis identifies significantly correlated pattern pairs by calculating support, confidence, and lift. Specifically, it statistically analyzes each candidate pattern pair, calculating their frequency of co-occurrence (support), the probability of the subsequent pattern appearing when the former appears (confidence), and the strength of the association (lift). For example, if analysis reveals a high-confidence and high-lift association between a high-precipitation pattern and a low-precipitation pattern, they can be identified as spatiotemporally correlated patterns.

[0080] Clustering spatiotemporal correlation patterns using spectral clustering aims to group similar patterns together. Spectral clustering is a graph-based clustering method that constructs a similarity matrix between correlated patterns, performs eigenvalue decomposition on the matrix, projects the patterns into a low-dimensional space, and then performs K-means clustering in that low-dimensional space. Specifically, it first calculates the similarity between each pair of spatiotemporally correlated patterns, constructs a similarity matrix, then performs eigenvalue decomposition on the matrix, selects the top k eigenvectors as a new feature space, and performs K-means clustering in this feature space to obtain preliminary clustering results. For example, spectral clustering can group several high precipitation patterns with similar spatiotemporal characteristics into one cluster, forming a preliminary clustering result.

[0081] Fuzzy membership calculation is performed on the preliminary clustering results to further refine the clustering outcomes. Fuzzy membership calculation uses the fuzzy C-means algorithm to calculate the probability (membership) of each data point belonging to each cluster. Specifically, the center of each cluster is first initialized, then the membership is calculated based on the distance between the data point and the cluster center, the position of the cluster center is updated, and this process is iterated until convergence. This method yields a soft clustering assignment for each data point. For example, in the preliminary clustering results, a data point may belong to multiple clusters simultaneously; fuzzy membership calculation provides the probability distribution of its belonging to each cluster.

[0082] Density peak detection is used in soft clustering assignments to find cluster centers. Density peak detection identifies cluster centers by calculating the local density of each data point and its distance to the nearest high-density point. Specifically, it first calculates the local density of each point, typically using a Gaussian kernel function, then calculates the distance from each point to the nearest high-density point, plots a density-distance map, and identifies cluster centers by observing density peaks. For example, density peak detection can identify several high-density points that are far apart from each other in a soft clustering assignment, serving as the final cluster centers.

[0083] The purpose of redistributing the spatiotemporal feature vector set of precipitation based on cluster centers is to improve the clustering results. The redistribution process assigns each data point to the cluster containing the nearest cluster center, resulting in a new clustering outcome. Specifically, it calculates the distance from each data point to all cluster centers and assigns the data point to the cluster containing the nearest center. For example, during the redistribution process, a data point that originally belonged to multiple clusters might be assigned to the cluster containing the nearest cluster center, thus creating an optimized clustering result.

[0084] Outlier detection in clustering results is used to identify anomalous precipitation patterns. Outlier detection calculates the distance between each data point and its cluster center, identifying points that are farther away as outliers. Specifically, the distance from each data point to the cluster center is first calculated, and then a threshold is set. Points exceeding this threshold are considered outliers, forming a candidate set of anomalous precipitation patterns. For example, in the clustering results, a data point far from the cluster center is identified as an outlier and included as a candidate for an anomalous precipitation pattern.

[0085] The purpose of using the Isolation Forest algorithm to quantify the anomalies in the candidate set of abnormal precipitation patterns is to ultimately determine the list of abnormal precipitation patterns. The Isolation Forest algorithm is an anomaly detection algorithm based on decision trees. It constructs multiple random trees, randomly partitions data points, and calculates the isolation path length of each point within the trees. Points with shorter isolation paths are more likely to be anomalies. Specifically, it constructs multiple random trees, calculates the isolation path length of each candidate point for an abnormal precipitation pattern, sets a threshold, and identifies points with isolation path lengths shorter than the threshold as anomalies, thus forming the list of abnormal precipitation patterns. For example, by using the Isolation Forest algorithm to calculate the isolation path length of points in the candidate set of abnormal precipitation patterns, several abnormal precipitation patterns are ultimately identified, forming the list of abnormal precipitation patterns.

[0086] Integrating spatiotemporal correlation patterns, clustering results, and a list of anomalous precipitation patterns aims to construct a comprehensive spatiotemporal precipitation pattern library. Spatiotemporal correlation patterns reveal the relationships within precipitation data, clustering results provide clustering information for precipitation patterns, and the list of anomalous precipitation patterns identifies abnormal patterns within the precipitation data. By integrating this information, a comprehensive spatiotemporal precipitation pattern library is formed, providing a foundation for subsequent precipitation forecasting and analysis. For example, during the integration process, discovered spatiotemporal correlation patterns, optimized clustering results, and identified anomalous precipitation patterns are combined to form a precipitation spatiotemporal pattern library containing a variety of information, providing comprehensive support for further analysis and applications.

[0087] For example, to collect a multidimensional time-series precipitation dataset for a certain region, the data is first segmented into a spatiotemporal sliding window. The time window length is set to 10 days, and the spatial window to 10 square kilometers. The window slides at 1-day and 1-square-kilometer steps to extract data within each window, forming a multi-scale feature fragment set. Then, spatiotemporal pattern encoding is performed on this multi-scale feature fragment set, dividing precipitation intensity into high, medium, and low levels. Encoding is then performed based on the precipitation intensity of each fragment, resulting in a pattern encoding sequence. Next, frequent pattern mining is performed on the pattern encoding sequence. The Apriori algorithm is used to mine frequently occurring precipitation patterns, forming a candidate spatiotemporal pattern set. Spatiotemporal association rule analysis is then performed on the candidate spatiotemporal pattern set, calculating the support, confidence, and lift between patterns to identify significantly correlated pattern pairs, thus obtaining spatiotemporally associated patterns. Spectral clustering is used to cluster the spatiotemporally associated patterns, constructing a similarity matrix, performing eigenvalue decomposition and K-means clustering to obtain preliminary clustering results. Finally, fuzzy membership calculation is performed on the preliminary clustering results, using the fuzzy C-means algorithm to calculate the soft clustering assignment for each data point. Density peak detection is performed on soft clustering assignments to calculate local density and distance, identifying cluster centers. Based on these cluster centers, the spatiotemporal feature vector set of precipitation is redistributed, calculating the distance from each data point to the cluster center to form the clustering results. Outlier detection is performed on the clustering results, calculating the distance from each data point to the cluster center to identify outliers and form a candidate set of anomalous precipitation patterns. Anomalies in the candidate set of anomalous precipitation patterns are quantified using the isolated forest algorithm, calculating the isolation path length, identifying outliers, and forming a list of anomalous precipitation patterns. Finally, the spatiotemporal correlation patterns, clustering results, and anomalous precipitation pattern list are integrated to construct a complete spatiotemporal precipitation pattern library.

[0088] In one specific embodiment, the process of executing step S104 may specifically include the following steps:

[0089] (1) Perform model embedding on the precipitation spatiotemporal model library to obtain the model embedding matrix;

[0090] (2) Perform feature mapping on the clustering results to obtain the clustering feature tensor;

[0091] (3) Encode the abnormal precipitation pattern list to obtain the abnormal feature vector;

[0092] (4) Multimodal feature fusion is performed on the pattern embedding matrix, clustering feature tensor and anomaly feature vector to obtain fused feature representation data;

[0093] (5) Spatiotemporal dependency modeling of fused feature representation data is performed using graph convolutional networks to obtain a spatiotemporal relationship graph;

[0094] (6) Perform attention mechanism calculation on the spatiotemporal relationship graph to obtain weighted spatiotemporal features;

[0095] (7) Sequence modeling of weighted spatiotemporal features is performed using a bidirectional long short-term memory network to obtain time series prediction results;

[0096] (8) Perform deconvolution operation on the time series prediction results to obtain an end-to-end global atmospheric reanalysis precipitation fusion prediction model.

[0097] Specifically, pattern embedding is performed on the precipitation spatiotemporal pattern library to obtain a pattern embedding matrix. Pattern embedding is a method to convert discrete spatiotemporal patterns into continuous vector representations. The specific operation involves representing each spatiotemporal pattern as a high-dimensional vector, which is typically trained using word embedding algorithms (such as Word2Vec or GloVe). These algorithms construct a large spatiotemporal pattern dictionary and train vector representations based on the co-occurrence relationships of spatiotemporal patterns. For example, in Word2Vec, each spatiotemporal pattern is embedded into a continuous vector space, making similar spatiotemporal patterns closer in the vector space, thus forming a pattern embedding matrix. Feature mapping is then performed on the clustering results to obtain a clustering feature tensor. Feature mapping is the process of representing each cluster in the clustering results as a feature vector. Specifically, the centroid of each cluster is first determined, then the distance from each data point to its cluster centroid is calculated, and these distances are used as components of the feature vector. In this way, the clustering results can be converted into a high-dimensional feature tensor, where each element represents the distance from the data point to the cluster centroid. This feature representation method can retain important information in the clustering results, which facilitates subsequent analysis and processing.

[0098] Anomaly pattern encoding is performed on the list of anomalous precipitation patterns to obtain anomalous feature vectors. Anomaly pattern encoding is the process of representing each anomalous pattern in the list as a feature vector. Specifically, each anomalous pattern is first standardized, and then the standardized feature values ​​are used as components of the feature vector. In this way, the list of anomalous precipitation patterns can be converted into a high-dimensional feature vector, with each feature vector representing an anomalous pattern. This encoding method can retain important information in the anomalous patterns, facilitating subsequent analysis and processing. Multimodal feature fusion is performed on the pattern embedding matrix, clustering feature tensor, and anomalous feature vectors to obtain fused feature representation data. Multimodal feature fusion is a method of combining feature representations from different sources to form a comprehensive feature representation. Specifically, the pattern embedding matrix, clustering feature tensor, and anomalous feature vectors are concatenated, and then fused using a fully connected neural network to obtain a new feature representation. This process can retain important information from different modal features while eliminating redundant information, forming a comprehensive feature representation data that facilitates subsequent spatiotemporal dependency modeling.

[0099] Spatiotemporal dependencies in the fused feature representation data are modeled using Graph Convolutional Networks (GCNs) to obtain a spatiotemporal relationship graph. A GCN is a graph-based neural network that effectively captures spatiotemporal dependencies in data. Specifically, the fused feature representation data is first constructed as a graph, where nodes represent spatiotemporal features and edges represent spatiotemporal dependencies between nodes. Then, a GCN is used to perform convolution operations on the graph to extract features from each node. The convolution operation involves calculating a weighted sum of features from each node and its neighbors, followed by a non-linear activation function to obtain a new feature representation for the node. By stacking multiple GCNs, deeper features in the spatiotemporal relationship graph can be extracted layer by layer, forming a comprehensive spatiotemporal relationship graph. An attention mechanism is then applied to the spatiotemporal relationship graph to obtain weighted spatiotemporal features. The attention mechanism is a method that assigns weights to different features, improving the effectiveness of feature representation. Specifically, the attention weights of each node's neighbors are calculated for each node in the spatiotemporal relationship graph; these weights can be calculated using an attention function (such as dot product attention or multi-head attention). Then, a weighted average is applied to the features of each node, with the weights calculated using an attention function, thus forming weighted spatiotemporal features. This method can highlight important features in the spatiotemporal relationship graph and improve the effectiveness of feature representation.

[0100] A time-series prediction result is obtained by using a Bidirectional Long Short-Term Memory (BiLSTM) network to model the weighted spatiotemporal features. BiLSTM is a neural network capable of processing time-series data and capturing the dependencies between sequences. Specifically, the weighted spatiotemporal features are input into the BiLSTM network, which consists of forward and backward LSTM layers. The forward LSTM layer handles the forward dependencies in the sequence data, while the backward LSTM layer handles the inverse dependencies. The combination of forward and backward LSTM layers captures the global dependencies in the sequence data and outputs the time-series prediction result. The time-series prediction result is then deconvolved to obtain an end-to-end global atmospheric reanalysis precipitation fusion prediction model. Deconvolution is a method for restoring the spatial structure of data, mapping low-dimensional features back to the original high-dimensional space. Specifically, the time-series prediction result is input into the deconvolution layer, which, through reverse convolution, maps the low-dimensional features back to the original high-dimensional space, thus forming an end-to-end precipitation prediction model. This method can effectively restore the spatial structure of the original data and improve the performance of the prediction model.

[0101] For example, when collecting spatiotemporal precipitation data for a certain region, the data is first embedded using a pattern embedding algorithm, and a pattern embedding matrix is ​​obtained by training the algorithm using Word2Vec. Next, feature mapping is performed on the clustering results, calculating the distance from each data point to the cluster center to form a clustering feature tensor. Then, the list of anomalous precipitation patterns is encoded as anomalous patterns, converting each anomalous pattern into a feature vector. The pattern embedding matrix, clustering feature tensor, and anomalous feature vectors are concatenated, and multimodal feature fusion is performed using a fully connected neural network to obtain fused feature representation data. A graph convolutional network is used to model the spatiotemporal dependencies of the fused feature representation data, constructing a spatiotemporal relationship graph, and deep features are extracted from the spatiotemporal relationship graph using a multi-layer GCN. An attention mechanism is applied to the spatiotemporal relationship graph to assign importance weights to features, forming weighted spatiotemporal features. The weighted spatiotemporal features are input into a bidirectional long short-term memory network, and forward and backward LSTM layers capture the global dependencies of the sequence data, outputting time series prediction results. Finally, a deconvolution operation is performed on the time-series prediction results to map the low-dimensional features back to the high-dimensional space, resulting in an end-to-end global atmospheric reanalysis precipitation fusion prediction model.

[0102] In one specific embodiment, the process of executing step S105 may specifically include the following steps:

[0103] (1) Data stream segmentation is performed on real-time precipitation observation data to obtain time-series data segments;

[0104] (2) Perform concept drift detection on time series data segments to obtain data distribution change indicators;

[0105] (3) Based on the data distribution change index, parameter sensitivity analysis was performed on the global atmospheric reanalysis precipitation fusion prediction model to obtain the sensitive parameter set;

[0106] (4) Incremental learning is performed on the sensitive parameter set to obtain the updated parameter vector;

[0107] (5) The updated parameter vector is fused with the global atmospheric reanalysis precipitation fusion prediction model to obtain a temporary prediction model;

[0108] (6) Compress the temporary prediction model to obtain a lightweight prediction model;

[0109] (7) The lightweight prediction model is trained in a distributed manner through federated learning to obtain a multi-source collaborative optimization model;

[0110] (8) Perform model distillation on the multi-source collaborative optimization model to obtain the knowledge transfer model;

[0111] (9) The knowledge transfer model is trained with adversarial examples to obtain a robust enhancement model;

[0112] (10) Multi-task learning is performed on the robustness enhancement model to obtain the spatiotemporal prediction model of the target precipitation.

[0113] Specifically, data stream segmentation of real-time precipitation observation data to obtain time-series data segments is the first step in achieving dynamic updates and online learning. Data stream segmentation divides continuous real-time precipitation observation data into several time-series data segments according to fixed time windows. These segments can be time windows of fixed length, such as hourly, daily, or weekly data segments. By processing these data segments, dynamic changes in precipitation data can be captured. For example, segmenting real-time precipitation observation data into hourly time-series data segments yields a continuous set of time-series data segments. Concept drift detection of these time-series data segments to obtain data distribution change indicators is a crucial step in detecting changes in precipitation data distribution. Concept drift detection is a method for identifying changes in data distribution, capable of detecting trend changes, seasonal variations, or sudden changes in data. Commonly used concept drift detection algorithms include the Page-Hinkley detection method, the ADWIN algorithm, and the DDM algorithm. The specific operation involves statistical analysis of each time-series data segment, calculating statistics such as the mean and variance of the data, and then comparing it with historical data to identify changes in data distribution. For example, the Page-Hinkley test can be used to calculate the mean and variance of each time series data segment, compare it with historical data, identify the trend of data distribution change, and obtain data distribution change indicators.

[0114] Based on data distribution change indicators, parameter sensitivity analysis is performed on the global atmospheric reanalysis precipitation fusion prediction model. Obtaining a set of sensitive parameters is a crucial step in optimizing the prediction model. Parameter sensitivity analysis is a method to evaluate the impact of model parameters on prediction results. By analyzing the influence of different parameters on the model output, sensitive parameters with significant impact on model performance are identified. Specifically, each parameter in the model is perturbed, the magnitude of the change in model output before and after the perturbation is calculated, and then the sensitivity of the parameter is evaluated based on the magnitude of the change. For example, by applying small perturbations to each parameter in the precipitation prediction model and calculating the magnitude of the change in model output, sensitive parameters with significant impact on model performance can be identified, forming a set of sensitive parameters.

[0115] Incremental learning of the sensitive parameter set to obtain an updated parameter vector is a crucial step in adaptive model learning. Incremental learning is a method for gradually updating model parameters by continuously learning from new data and progressively optimizing them. Specifically, it involves using the parameters in the sensitive parameter set as learning objects, making small adjustments to each parameter, calculating the error between the adjusted model output and the actual observed data, adjusting the parameter values ​​based on the error, and iterating until the error is minimized. For example, by making small adjustments to the sensitive parameters, calculating the error between the model output and the actual observed data, and adjusting the parameter values ​​based on the error, an updated parameter vector is formed.

[0116] The key step in achieving dynamic model updates is fusing the updated parameter vector with the global atmospheric reanalysis precipitation fusion prediction model to obtain a provisional prediction model. Model fusion is a method of combining parameters from old and new models. It involves merging the updated parameter vector with the original model parameters to form a new model. Specifically, the updated parameter vector and the original model parameters are weighted and averaged to obtain a new parameter vector. This new parameter vector is then applied to the model to form a provisional prediction model. For example, by weighting the updated parameter vector and the original model parameters with a weight of 0.5, a new parameter vector is obtained and applied to the precipitation prediction model to form a provisional prediction model.

[0117] Model compression, specifically lightweight model compression, is a key step in improving computational efficiency. Model compression is a method to reduce model complexity by employing techniques such as pruning, quantization, or distillation to lower computational complexity and storage requirements. Specifically, this involves pruning the temporary prediction model by removing unimportant parameters and connections, quantizing model parameters to convert floating-point representations to lower-precision representations, or distilling knowledge from complex models to simpler ones. For example, pruning a temporary prediction model by removing unimportant parameters and connections reduces computational complexity, resulting in a lightweight prediction model.

[0118] Distributed training of a lightweight prediction model using federated learning to obtain a multi-source collaborative optimization model is a key step in achieving collaborative optimization of multi-source data. Federated learning is a distributed machine learning method that achieves collaborative optimization of data by training the model on multiple nodes. Specifically, the lightweight prediction model is distributed across multiple nodes, each using local data for training. The training results are then aggregated at a central server, where the model parameters are weighted and averaged to form a globally optimized model. For example, by distributing the lightweight prediction model across multiple meteorological observation stations, each station can train using local data, and the training results can be aggregated at a central server for weighted averaging of the model parameters, thus forming a multi-source collaborative optimization model.

[0119] Model distillation, performed on multi-source co-optimization models to obtain knowledge transfer models, is a key step in improving model lightweighting and performance. Model distillation is a method that transfers knowledge from a complex model to a simpler model, improving the performance of the simpler model by having it learn the output of the complex model. Specifically, the multi-source co-optimization model is first used to predict training data, yielding prediction results. Then, the knowledge transfer model learns these predictions, optimizing its parameters by minimizing the error between the predictions and the actual observed data. For example, by using a multi-source co-optimization model to predict training data, obtaining prediction results, and then having the knowledge transfer model learn these predictions, the error between the predictions and the actual observed data is minimized to form the knowledge transfer model.

[0120] Training a knowledge transfer model with adversarial examples to obtain a robust enhancement model is a key step in improving the model's resistance to interference. Adversarial example training is a method to improve model robustness by incorporating adversarial examples during training, thereby enhancing the model's resistance to noise and attacks. Specifically, adversarial examples are first generated by making minor perturbations to the original data. These adversarial examples are then added to the training data for model training. By minimizing the prediction error between the adversarial examples and the original data, the model's robustness is improved. For example, by making minor perturbations to the original precipitation data to generate adversarial examples, and then adding these adversarial examples to the training data for model training, a robust enhancement model is formed.

[0121] Multi-task learning of the robustness enhancement model to obtain the target precipitation spatiotemporal prediction model is a key step in improving the overall performance of the model. Multi-task learning is a method that optimizes multiple tasks simultaneously by sharing model parameters to achieve collaborative optimization of multiple tasks. Specifically, tasks such as precipitation prediction, precipitation anomaly detection, and precipitation pattern recognition are simultaneously incorporated into the model training. By sharing model parameters and performing joint optimization, the overall performance of the model is improved. For example, by simultaneously incorporating precipitation prediction, precipitation anomaly detection, and precipitation pattern recognition tasks into the model training and performing joint optimization by sharing model parameters, a target precipitation spatiotemporal prediction model can be formed.

[0122] For example, firstly, real-time precipitation observation data for a certain region is segmented into hourly time-series data segments. Then, concept drift detection is performed on these time-series data segments. The Page-Hinkley test is used to calculate the mean and variance of each segment, and compared with historical data to identify changes in data distribution, obtaining data distribution change indicators. Based on these change indicators, parameter sensitivity analysis is performed on the global atmospheric reanalysis precipitation fusion prediction model. Small perturbations are made to the model parameters, and changes in the model output are calculated to identify a set of sensitive parameters. Incremental learning is performed on the sensitive parameter set. By adjusting parameter values, the error between the model output and the observed data is minimized, resulting in an updated parameter vector. The updated parameter vector is then weighted and averaged with the original model parameters to form a temporary prediction model. The temporary prediction model is pruned, removing unimportant parameters to form a lightweight prediction model. The lightweight model is distributed to multiple observation stations using federated learning, trained using local data, and the training results are aggregated to form a multi-source collaborative optimization model. The output of the multi-source collaborative optimization model is used to train a knowledge transfer model, minimizing the error to form the knowledge transfer model. Adversarial examples are used to train the knowledge transfer model, generating adversarial examples to improve the model's robustness and form a robust enhancement model. Finally, through multi-task learning, tasks such as precipitation prediction, anomaly detection, and pattern recognition are simultaneously added to the training to form a spatiotemporal prediction model for the target precipitation.

[0123] In one specific embodiment, the process of executing step S106 may specifically include the following steps:

[0124] (1) Spatiotemporal decomposition of the output of the target precipitation spatiotemporal prediction model is performed to obtain multi-scale precipitation prediction results;

[0125] (2) Spatiotemporal autocorrelation analysis was performed on the multi-scale precipitation prediction results to obtain the evolution characteristics of precipitation patterns;

[0126] (3) Perform topological data analysis on the evolution characteristics of precipitation patterns to obtain the topological structure of precipitation patterns;

[0127] (4) Extreme events are identified based on the topology of precipitation patterns to obtain an assessment of extreme precipitation risk;

[0128] (5) Scenario simulations were conducted for extreme precipitation risk assessment to obtain multi-scenario precipitation impact predictions;

[0129] (6) Through complex network analysis, the system vulnerability of precipitation impact prediction under multiple scenarios is assessed, and key impact factors are obtained;

[0130] (7) Causal inferences were made about key influencing factors to obtain a dynamic model of the precipitation system;

[0131] (8) Based on the dynamic model of the precipitation system, a spatiotemporal knowledge graph was constructed to obtain a precipitation knowledge network;

[0132] (9) Conduct interpretability analysis on the precipitation knowledge network to obtain a global atmospheric reanalysis precipitation fusion data analysis report;

[0133] (10) Based on the global atmospheric reanalysis precipitation fusion data analysis report, resource optimization allocation is carried out to obtain the weather processing pipeline scheduling strategy.

[0134] Specifically, the spatiotemporal decomposition of the output of the target precipitation spatiotemporal prediction model is achieved through multi-scale wavelet transform. This process decomposes the precipitation prediction results into components at different spatiotemporal scales, including long-term trends, seasonal variations, and short-term fluctuations. For example, for an annual precipitation prediction of a region, we might obtain a 10-year growth trend, 3-5 year periodic fluctuations, and intra-annual seasonal variations. Spatiotemporal autocorrelation analysis is then performed on these multi-scale precipitation prediction results using Moran's I index and the time series autocorrelation function. Moran's I index is used to quantify spatial autocorrelation, while the time series autocorrelation function is used to identify temporal dependencies. This step can reveal the evolutionary characteristics of precipitation patterns in time and space, such as the movement trend of precipitation bands or the periodic changes in local precipitation intensity.

[0135] Topological data analysis is performed on the identified precipitation pattern evolution characteristics. This step uses a persistent cohomology algorithm to represent the precipitation pattern as a high-dimensional topological structure. By calculating the Betti number and persistence graph, topological features of the precipitation pattern, such as connectivity and voids, can be captured. These topological features reflect the structural stability and complexity of the precipitation system. Based on the obtained precipitation pattern topology, extreme event identification is performed using extreme value theory. This process employs the generalized extreme value distribution (GEV) and peak exceedance (POT) models to model precipitation events exceeding a specific threshold. In this way, the probability and intensity of extreme precipitation events can be assessed, thus obtaining extreme precipitation risk assessment results.

[0136] Scenario simulations were performed based on the results of the extreme precipitation risk assessment, and multiple possible precipitation scenarios were generated using the Monte Carlo method. These scenarios considered different climate change pathways, land use changes, and human activity impacts, thus yielding multi-scenario precipitation impact predictions.

[0137] System vulnerability assessment for multi-scenario precipitation impact prediction is conducted using complex network analysis. This step constructs a precipitation impact network, where nodes represent different geographical regions or system components, and edges represent their interactions. Key influencing factors in the system are identified by calculating network centrality, clustering coefficients, and other indicators. Causal inferences are then performed on these key influencing factors using structural equation modeling (SEM) or Bayesian networks. This process establishes a causal relationship graph between factors, thereby obtaining a dynamic model of the precipitation system. This model describes how different factors interact and influence the dynamic changes of the precipitation system.

[0138] Based on the dynamic model of the precipitation system, a spatiotemporal knowledge graph is constructed. This step maps the entities and relationships in the dynamic model to a graph database, forming a knowledge network containing temporal and spatial information. This network includes not only precipitation-related entities (such as climate factors and geographical features) but also the spatiotemporal relationships between them.

[0139] Interpretability analysis is performed on the constructed precipitation knowledge network using techniques such as SHAP (SHapley Additive ex Planations) values. This process interprets the model's predictions, revealing the contribution of different factors to precipitation forecasts, thus generating a comprehensive global atmospheric reanalysis precipitation fusion data analysis report. Finally, based on this data analysis report, resource optimization is performed. This step uses multi-objective optimization algorithms, such as genetic algorithms or particle swarm optimization, to balance the meteorological observation needs, forecast accuracy requirements, and resource constraints of different regions, ultimately yielding a weather processing pipeline scheduling strategy.

[0140] For example, in a large watershed spanning multiple climate zones, the above methods were used to analyze 50 years of historical precipitation data and 30 years of projected precipitation. Spatiotemporal decomposition revealed long-term trends in annual precipitation across different regions, showing an increasing trend in upstream areas and a decreasing trend in downstream areas. Spatiotemporal autocorrelation analysis revealed the seasonal migration characteristics of precipitation patterns, while topological data analysis identified several key precipitation "hotspot" areas. Extreme event analysis predicted a 30% increase in the risk of once-in-a-century rainfall events in certain areas of the watershed over the next 30 years. Scenario simulation and complex network analysis further indicated that this increased risk is mainly influenced by regional land-use change and global warming. Causal inference models quantified the relative importance of these factors, providing a basis for developing targeted adaptation strategies. Finally, based on these analytical results, the layout of meteorological observation stations and the resource allocation of early warning systems were optimized, improving the forecasting and response capabilities for extreme precipitation events.

[0141] The above describes the time series feature analysis method based on multidimensional data in the embodiments of this application. The following describes the time series feature analysis system based on multidimensional data in the embodiments of this application. Please refer to [link / reference]. Figure 2 One embodiment of the time-series feature analysis system based on multidimensional data in this application includes:

[0142] The acquisition module 201 is used to collect and preprocess multi-source raw data from weather observation stations, satellite remote sensing and ground observation stations to obtain a standardized multi-dimensional time-series precipitation dataset.

[0143] Dimensionality reduction module 202 is used to extract spatiotemporal features and perform dimensionality reduction processing on the multidimensional time-series precipitation dataset to obtain a set of precipitation spatiotemporal feature vectors;

[0144] Clustering module 203 is used to perform multi-scale spatiotemporal pattern recognition and clustering analysis on the precipitation spatiotemporal feature vector set to obtain a precipitation spatiotemporal pattern library, clustering results and a list of anomalous precipitation patterns;

[0145] The construction module 204 is used to construct a deep learning precipitation prediction model based on the precipitation spatiotemporal pattern library, the clustering results and the list of anomalous precipitation patterns, to obtain an end-to-end global atmospheric reanalysis precipitation fusion prediction model.

[0146] The update module 205 is used to dynamically update and learn online the global atmospheric reanalysis precipitation fusion prediction model to obtain the target precipitation spatiotemporal prediction model.

[0147] Analysis module 206 is used to perform multidimensional spatiotemporal analysis on the target precipitation spatiotemporal prediction model to obtain a global atmospheric reanalysis precipitation fusion data analysis report and a weather processing pipeline scheduling strategy.

[0148] Through the collaborative efforts of the aforementioned components, a standardized multidimensional time-series precipitation dataset can be obtained by collecting and preprocessing raw data from multiple sources, including weather observation stations, satellite remote sensing, and ground-based observation stations. This dataset integrates the advantages of multiple data sources, avoiding the limitations of a single data source and making the data more comprehensive and accurate. Based on this, spatiotemporal feature extraction and dimensionality reduction are performed on the multidimensional time-series precipitation dataset to obtain a precipitation spatiotemporal feature vector set. This method effectively captures the complex spatiotemporal characteristics of precipitation, improving data expressiveness and processing efficiency. Secondly, multi-scale spatiotemporal pattern recognition and clustering analysis are performed on the precipitation spatiotemporal feature vector set to construct a precipitation spatiotemporal pattern library, clustering results, and a list of anomalous precipitation patterns. This not only helps to deeply explore the spatiotemporal patterns of precipitation but also identifies anomalous precipitation patterns, improving the prediction capability for extreme weather events. Furthermore, based on the precipitation spatiotemporal pattern library, clustering results, and anomalous precipitation pattern list, a deep learning precipitation prediction model is constructed, resulting in an end-to-end global atmospheric reanalysis precipitation fusion prediction model. This deep learning model effectively utilizes large-scale data for training, capturing the complex nonlinear characteristics of precipitation and significantly improving the accuracy of precipitation forecasts. Through dynamic updates and online learning, the global atmospheric reanalysis precipitation fusion prediction model is optimized to obtain a target precipitation spatiotemporal prediction model. This allows the model to adaptively adjust parameters and respond promptly to weather changes, improving the real-time performance and accuracy of forecasts. Furthermore, multi-dimensional spatiotemporal analysis of the target precipitation spatiotemporal prediction model generates a global atmospheric reanalysis precipitation fusion data analysis report and a weather processing pipeline scheduling strategy. It enables comprehensive evaluation and analysis of precipitation forecast results. This invention, through the fusion of multi-source data, spatiotemporal feature extraction, multi-scale spatiotemporal pattern recognition, and the construction and optimization of deep learning models, not only significantly improves the accuracy and real-time performance of precipitation forecasts but also enhances the predictive ability for extreme weather events.

[0149] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0150] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for time-series feature analysis based on multidimensional data, characterized in that, The time-series feature analysis method based on multidimensional data includes: Multi-source raw data from weather observation stations, satellite remote sensing, and ground observation stations are collected and preprocessed to obtain a standardized multidimensional time-series precipitation dataset. Spatiotemporal feature extraction and dimensionality reduction are performed on the multidimensional time-series precipitation dataset to obtain a set of precipitation spatiotemporal feature vectors; Multi-scale spatiotemporal pattern recognition and cluster analysis are performed on the precipitation spatiotemporal feature vector set to obtain a precipitation spatiotemporal pattern library, cluster results, and a list of anomalous precipitation patterns; Based on the precipitation spatiotemporal pattern library, the clustering results, and the list of anomalous precipitation patterns, a deep learning precipitation prediction model is constructed to obtain an end-to-end global atmospheric reanalysis precipitation fusion prediction model. The global atmospheric reanalysis precipitation fusion prediction model is dynamically updated and learned online to obtain the spatiotemporal prediction model for the target precipitation. Multidimensional spatiotemporal analysis was performed on the target precipitation spatiotemporal prediction model to obtain a global atmospheric reanalysis precipitation fusion data analysis report and a weather processing pipeline scheduling strategy; The process involves performing multi-scale spatiotemporal pattern recognition and clustering analysis on the precipitation spatiotemporal feature vector set to obtain a precipitation spatiotemporal pattern library, clustering results, and a list of anomalous precipitation patterns, including: The spatiotemporal feature vector set of precipitation is segmented by a spatiotemporal sliding window to obtain a multi-scale feature fragment set; Spatiotemporal pattern coding is performed on the multi-scale feature fragment set to obtain a pattern coding sequence; Frequent pattern mining is performed on the pattern-encoded sequence to obtain a candidate spatiotemporal pattern set; Spatiotemporal correlation rule analysis is performed on the candidate spatiotemporal pattern set to obtain spatiotemporal correlation patterns; The spatiotemporal correlation patterns were clustered using a spectral clustering algorithm to obtain preliminary clustering results. Fuzzy membership calculation is performed on the preliminary clustering results to obtain soft clustering assignments; Density peak detection is performed on the soft clustering assignment to obtain the cluster centers; Based on the cluster centers, the precipitation spatiotemporal feature vector set is redistributed to obtain the clustering results; Outlier detection is performed on the clustering results to obtain a candidate set of abnormal precipitation patterns; The abnormal precipitation pattern candidate set is quantified using the isolated forest algorithm to obtain the abnormal precipitation pattern list. The spatiotemporal correlation patterns, the clustering results, and the list of abnormal precipitation patterns are integrated to obtain the precipitation spatiotemporal pattern library.

2. The time-series feature analysis method based on multidimensional data according to claim 1, characterized in that, The process involves collecting and preprocessing raw data from multiple sources, including weather observation stations, satellite remote sensing, and ground observation stations, to obtain a standardized multidimensional time-series precipitation dataset, which includes: The multi-source raw data is timestamped to obtain time-synchronized data, and the time-synchronized data is spatially transformed to obtain unified projection data. The unified projection data is subjected to data quality assessment to obtain quality-labeled data; The quality-labeled data is fused from multiple sources to obtain preliminary fused data; The noise is removed from the preliminary fused data using the Kalman filter algorithm to obtain noise-reduced optimized data. The noise reduction optimization data is adjusted for spatiotemporal scale consistency to obtain scale-uniform data; The scale-uniform data is discretized to obtain discretized time-series data, and the discretized time-series data is then encoded with multidimensional features to obtain the standardized multidimensional time-series precipitation dataset.

3. The time-series feature analysis method based on multidimensional data according to claim 1, characterized in that, The process of extracting and reducing the spatiotemporal features of the multidimensional time-series precipitation dataset to obtain a set of precipitation spatiotemporal feature vectors includes: The spatiotemporal complexity characteristics are obtained by calculating the fractal dimension of the multidimensional time-series precipitation dataset. Symbolic dynamics encoding is performed on the multidimensional time-series precipitation dataset to obtain symbolic sequence features; Phase space trajectory features are obtained by embedding the multidimensional time-series precipitation dataset into the reconstructed phase space. The Lyapunov exponent is calculated on the phase space trajectory characteristics to obtain the system stability characteristics; Recursive quantitative analysis was performed on the multidimensional time-series precipitation dataset to obtain the system's periodic characteristics; Multi-scale entropy analysis was performed on the multi-dimensional time-series precipitation dataset to obtain the time-series complexity characteristics; The nonlinear time-frequency characteristics of the multidimensional time-series precipitation dataset were obtained by performing time-frequency analysis on the Hilbert-Huang transform. Lag mutual information analysis was performed on the multidimensional time-series precipitation dataset to obtain spatiotemporal dependency features; The spatiotemporal complexity feature, the symbol sequence feature, the system stability feature, the system periodicity feature, the temporal complexity feature, the nonlinear time-frequency feature, and the spatiotemporal dependence feature are subjected to nonlinear dimensionality reduction to obtain the precipitation spatiotemporal feature vector set.

4. The time-series feature analysis method based on multidimensional data according to claim 1, characterized in that, Based on the precipitation spatiotemporal model library, the clustering results, and the list of anomalous precipitation patterns, a deep learning precipitation prediction model is constructed to obtain an end-to-end global atmospheric reanalysis precipitation fusion prediction model, including: The precipitation spatiotemporal model library is embedded to obtain a model embedding matrix; The clustering results are subjected to feature mapping to obtain a clustering feature tensor; The list of abnormal precipitation patterns is encoded to obtain an abnormal feature vector; The pattern embedding matrix, the clustering feature tensor, and the anomaly feature vector are fused using multimodal feature fusion to obtain fused feature representation data; The spatiotemporal dependency model of the fused feature representation data is performed using a graph convolutional network to obtain a spatiotemporal relationship graph; The spatiotemporal relationship graph is subjected to attention mechanism calculation to obtain weighted spatiotemporal features; The weighted spatiotemporal features are sequence-modeled using a bidirectional long short-term memory network to obtain time-series prediction results; The time-series prediction results are deconvolved to obtain the end-to-end global atmospheric reanalysis precipitation fusion prediction model.

5. The time-series feature analysis method based on multidimensional data according to claim 1, characterized in that, The process of dynamically updating and learning the global atmospheric reanalysis precipitation fusion prediction model to obtain the target precipitation spatiotemporal prediction model includes: Real-time precipitation observation data is segmented into time-series data segments; Concept drift detection is performed on the time-series data segments to obtain data distribution change indicators; Based on the data distribution change index, a parameter sensitivity analysis was performed on the global atmospheric reanalysis precipitation fusion prediction model to obtain a set of sensitive parameters. Incremental learning is performed on the sensitive parameter set to obtain the updated parameter vector; The updated parameter vector is fused with the global atmospheric reanalysis precipitation fusion prediction model to obtain a temporary prediction model. The temporary prediction model is compressed to obtain a lightweight prediction model; The lightweight prediction model is trained in a distributed manner using federated learning to obtain a multi-source collaborative optimization model. The multi-source collaborative optimization model is distilled to obtain a knowledge transfer model; The knowledge transfer model is trained with adversarial examples to obtain a robust enhancement model; The robustness enhancement model is subjected to multi-task learning to obtain the spatiotemporal prediction model for the target precipitation.

6. The time-series feature analysis method based on multidimensional data according to claim 1, characterized in that, The multidimensional spatiotemporal analysis of the target precipitation spatiotemporal prediction model yields a global atmospheric reanalysis precipitation fusion data analysis report and a weather processing pipeline scheduling strategy, including: Spatiotemporal decomposition is performed on the output of the target precipitation spatiotemporal prediction model to obtain multi-scale precipitation prediction results; Spatiotemporal autocorrelation analysis was performed on the multi-scale precipitation prediction results to obtain the evolution characteristics of the precipitation pattern; Topological data analysis was performed on the evolution characteristics of the precipitation pattern to obtain the topological structure of the precipitation pattern; Extreme events are identified based on the precipitation pattern topology to obtain an extreme precipitation risk assessment. Scenario simulations were performed on the extreme precipitation risk assessment to obtain multi-scenario precipitation impact predictions; The system vulnerability of the multi-scenario precipitation impact prediction was assessed by complex network analysis to obtain key impact factors; Causal inferences were performed on the key influencing factors to obtain a dynamic model of the precipitation system; A spatiotemporal knowledge graph was constructed based on the aforementioned precipitation system dynamics model to obtain a precipitation knowledge network. An interpretability analysis was performed on the precipitation knowledge network to obtain the global atmospheric reanalysis precipitation fusion data analysis report; Based on the global atmospheric reanalysis precipitation fusion data analysis report, resource optimization allocation is performed to obtain the weather processing pipeline scheduling strategy.

7. A time-series feature analysis system based on multidimensional data, used to execute the time-series feature analysis method based on multidimensional data as described in any one of claims 1-6, characterized in that, The time-series feature analysis system based on multidimensional data includes: The acquisition module is used to collect and preprocess raw data from multiple sources, including weather observation stations, satellite remote sensing, and ground observation stations, to obtain a standardized multidimensional time-series precipitation dataset. The dimensionality reduction module is used to extract spatiotemporal features and perform dimensionality reduction on the multidimensional time-series precipitation dataset to obtain a set of precipitation spatiotemporal feature vectors. The clustering module is used to perform multi-scale spatiotemporal pattern recognition and clustering analysis on the precipitation spatiotemporal feature vector set to obtain a precipitation spatiotemporal pattern library, clustering results, and a list of anomalous precipitation patterns. The construction module is used to build a deep learning precipitation prediction model based on the precipitation spatiotemporal pattern library, the clustering results and the list of anomalous precipitation patterns, so as to obtain an end-to-end global atmospheric reanalysis precipitation fusion prediction model. The update module is used to dynamically update and learn online the global atmospheric reanalysis precipitation fusion prediction model to obtain the target precipitation spatiotemporal prediction model. The analysis module is used to perform multi-dimensional spatiotemporal analysis on the target precipitation spatiotemporal prediction model to obtain a global atmospheric reanalysis precipitation fusion data analysis report and a weather processing pipeline scheduling strategy. The process involves performing multi-scale spatiotemporal pattern recognition and clustering analysis on the precipitation spatiotemporal feature vector set to obtain a precipitation spatiotemporal pattern library, clustering results, and a list of anomalous precipitation patterns, including: The spatiotemporal feature vector set of precipitation is segmented by a spatiotemporal sliding window to obtain a multi-scale feature fragment set; Spatiotemporal pattern coding is performed on the multi-scale feature fragment set to obtain a pattern coding sequence; Frequent pattern mining is performed on the pattern-encoded sequence to obtain a candidate spatiotemporal pattern set; Spatiotemporal correlation rule analysis is performed on the candidate spatiotemporal pattern set to obtain spatiotemporal correlation patterns; The spatiotemporal correlation patterns were clustered using a spectral clustering algorithm to obtain preliminary clustering results. Fuzzy membership calculation is performed on the preliminary clustering results to obtain soft clustering assignments; Density peak detection is performed on the soft clustering assignment to obtain the cluster centers; Based on the cluster centers, the precipitation spatiotemporal feature vector set is redistributed to obtain the clustering results; Outlier detection is performed on the clustering results to obtain a candidate set of abnormal precipitation patterns; The abnormal precipitation pattern candidate set is quantified using the isolated forest algorithm to obtain the abnormal precipitation pattern list. The spatiotemporal correlation patterns, the clustering results, and the list of abnormal precipitation patterns are integrated to obtain the precipitation spatiotemporal pattern library.

Images