Room-Bohai sea low-altitude meteorological safety engine artificial intelligence algorithm based on historical reanalysis and multi-mode data

Abstract

The invention belongs to the technical field of meteorological safety, and discloses a circum-Bohai sea low-altitude meteorological safety engine artificial intelligence algorithm based on historical reanalysis and multi-mode data. The low-altitude meteorological safety engine artificial intelligence algorithm comprises the following modules: a meteorological variable reconstruction and extraction module, an extreme event identification module, a model training module, a mode fusion module and a safety engine construction module. The low-altitude meteorological safety engine artificial intelligence algorithm of the Bohai Sea has the following advantages: (1) data fusion and timeliness improvement are realized; (2) the low-altitude small-scale extreme weather identification precision is improved; (3) performing multi-dimensional meteorological risk assessment system and standardized grading; and (4) performing dynamic adaptive optimization and real-time adjustment.

Description

Technical Field

[0001] The present invention belongs to the field of meteorological safety technology and relates to an artificial intelligence algorithm for a Bohai Rim low-altitude meteorological safety engine based on historical reanalysis and multi-mode data. Background Art

[0002] With the rapid development of emerging industries such as the low-altitude economy (such as drone transportation, short-distance aviation, and urban air traffic), smart city management, and coastal energy development, the importance of low-altitude (0-3 km) meteorological safety assurance has become increasingly prominent. Especially in the Bohai Rim region (Beijing, Tianjin, Hebei, Shandong, and Liaoning coastal areas), due to the complex terrain, significant land-sea interaction, and frequent monsoon activities, local severe convection, wind shear, low visibility, low-level jet streams, icing and other sudden meteorological disasters frequently occur, posing a great threat to aviation, port shipping, and urban operations. Traditional meteorological assurance methods mainly rely on numerical weather prediction (NWP) models (such as GFS, ECMWF), ground automatic station observations, wind profiler radars, satellite remote sensing data, etc. However, due to the limitations of:

[0003] (1) Insufficient model resolution;

[0004] (2) Incomplete spatial coverage of single observation data;

[0005] (3) Local small-scale sudden weather events are difficult to capture accurately;

[0006] (4) The real-time identification and accurate warning of severe low-altitude weather are not ideal.

[0007] With the rapid development of new-generation technologies such as artificial intelligence (especially deep learning), big data fusion analysis, and reinforcement learning, the meteorological field has gradually explored the application of AI methods in weather forecasting and risk assessment, aiming to break the bottleneck of traditional numerical forecasting and enhance the intelligent identification and dynamic warning capabilities of low-altitude meteorology.

[0008] In the Bohai Rim region, severe low-altitude weather has the following characteristics:

[0009] (1) Small spatial scale (within tens of kilometers) and fast temporal evolution (minutes to hours);

[0010] (2) Traditional ground stations and conventional weather radars have sparse coverage, making it difficult to form a continuous and detailed observation network;

[0011] (3) Although reanalysis data are stable, they lack the ability to depict sudden extreme events in real time;

[0012] (4) Single-model forecasts have large uncertainties and biases, especially in the forecasts of low-level wind fields, humidity fields, visibility, icing conditions and other factors.

[0013] Therefore, it is necessary to organically integrate the historical meteorological evolution experience (summarized through reanalysis data) with the current multi-model ensemble forecast (complementary information of different models), and combine artificial intelligence technology to conduct spatiotemporal feature mining and risk decision optimization, in order to truly realize the intelligent identification and efficient early warning of severe weather in the low-altitude environment around the Bohai Sea.

[0014] Currently, there are several technical solutions that attempt to use meteorological data and machine learning algorithms to predict meteorological events, including the following:

[0015] (1) Statistical research on meteorological characteristics based on reanalysis data. By using global reanalysis data such as ERA5 and MERRA-2, the evolution of historical meteorological elements is studied and statistical analysis of meteorological events is conducted. For example: calculating the multi-year trend of wind shear, turbulence, and visibility changes in a specific area; identifying extreme meteorological events (such as the number of days with low visibility and the frequency of severe gale); and using them for meteorological disaster risk assessment and planning guidance. Main problems: lack of real-time performance and inability to meet dynamic warning needs; only statistical inference can be performed and cannot be directly used for real-time decision-making; the resolution of reanalysis data is limited (ERA5 is 31 km or higher and 0.25° resolution), making it difficult to analyze fine-scale changes. (Hersbach, H. et al. (2020). The ERA5 globalreanalysis. Quarterly Journal of the Royal Meteorological Society, 146(730),1999–2049.)

[0016] (2) Multi-model fusion methods based on the output of numerical forecast models, using ensemble forecast data from different forecast centers (such as ECMWF's ENS, GFS's GEFS) to fuse multi-model outputs, and improving the stability and reliability of forecasts through weighted averaging, Bayesian fusion, etc. For example: multi-model weighted fusion of wind field, humidity field, and temperature field; quantification of forecast uncertainty (probabilistic forecast); simple rules to judge dangerous weather conditions such as wind shear and turbulence. Main problems: Fusion methods are mostly linear or simple statistical rules, which cannot effectively extract complex nonlinear correlation features; lack of intelligent learning and dynamic optimization mechanisms based on actual observations or historical events; single risk assessment indicator system, which makes it difficult to quantify the comprehensive impact of multiple meteorological risks. (Leutbecher, M., & Palmer, TN (2007). Ensemble forecasting. Journal of Computational Physics, 227(7), 3515-3539.)

[0017] (3) Local weather event recognition models based on machine learning. In recent years, some studies have begun to use machine learning (such as support vector machines (SVM), random forests (RF), and convolutional neural networks (CNN)) to identify local weather phenomena such as wind shear, low visibility, and short-term heavy rainfall. For example: using observational data to train classifiers to determine whether extreme weather events have occurred; using meteorological characteristic variables (temperature, humidity, and wind speed change rate) as model inputs; and directly giving the probability or level of extreme weather. Main problems: Most models are limited to single-region or small-sample training, with weak generalization capabilities; lack a systematic design that integrates large-scale historical data and multi-model forecast data; and the warning results lack physical explanations, resulting in low user trust. (Lagerquist, R., McGovern, A. & Gagne, DJ, II. (2019) Deep learning for spatially explicit prediction of synoptic-scale fronts. Weather and Forecasting, 34, 1137-1160.)

[0018] Although the current technical system for low-altitude meteorological warning is constantly developing, the existing technology still has the following major problems:

[0019] (1) Insufficient data fusion capabilities and poor real-time performance;

[0020] The existing system mainly relies on a single source (such as a single numerical model output or a single observation data), lacks the unified fusion processing capability of historical data and multi-mode real-time data, resulting in a time lag in capturing local severe weather evolution and a slow response speed, which cannot meet the actual needs of low-altitude economy and emergency response for minute-level rapid warning.

[0021] (2) Low spatial resolution and insufficient accuracy in identifying extreme weather conditions;

[0022] Limited by the spatial resolution of traditional models (usually above 25 km) and simple statistical or rule-based methods, existing technologies find it difficult to identify extreme weather phenomena such as small-scale severe convection and wind shear within 5 km, resulting in a high recognition error rate and affecting the accuracy of early warnings.

[0023] (3) The risk assessment system is single and lacks standardized quantitative indicators;

[0024] Most existing low-altitude weather warning systems can only provide qualitative risk warnings and lack structured, multi-dimensional quantitative indicators (such as turbulence index, wind shear index, low visibility risk level, etc.), and cannot provide intuitive and graded decision-making basis for downstream application units (such as airports, ports, and drone control platforms).

[0025] (4) The early warning model lacks adaptive capabilities and is prone to false alarms and missed alarms;

[0026] Most existing systems use static rules to set thresholds and lack a mechanism to dynamically adjust warning parameters based on real-time weather changes. This results in high false alarm and missed alarm rates when the weather fluctuates drastically, reducing the reliability and practicality of the meteorological service system. Summary of the Invention

[0027] The present invention provides an artificial intelligence algorithm for the Bohai Rim low-altitude meteorological safety engine based on historical reanalysis and multi-modal data, aiming to achieve high-precision, real-time risk identification and early warning of low-altitude, small-scale severe weather.

[0028] The technical solution of the present invention:

[0029] (1) Meteorological variable reconstruction and extraction module:

[0030] Historical reanalysis data and real-time observation data are fused from multiple sources to ensure uniform resolution in spatial and temporal distribution. Historical reanalysis data include data from ERA5, JRA-55, and MERRA-2, while real-time observation data include ground station wind and temperature data, LIDAR-derived wind profiles, radar data, and microwave radiometer observations. Boundary layer height, wind shear, vorticity, inversion intensity, gradient, and disturbance are used as meteorological variables. Interpolation methods are used to improve the resolution and accuracy of meteorological variables in temporal and spatial distribution, followed by dimensionality reduction and compression.

[0031] (2) Extreme event identification module:

[0032] Based on the historical meteorological extreme events in the study area, a standardized labeling rule library was constructed. Historical meteorological extreme events include pollution accumulation events, wind shear events, and localized high wind events. Historical meteorological extreme events were collected to establish a sample event set. A long-short-term memory network was used to perform reconstruction error detection on the observation data sequence of meteorological variables for the identification of meteorological extreme events. The observation data sequence of meteorological variables is the observation value of one or more meteorological variables at consecutive time points.

[0033] Use deep autoencoders to automatically detect and label historical extreme meteorological events. Finally, construct composite indicators of multiple meteorological variables to improve the sensitivity of identifying extreme meteorological events.

[0034] (3) Model training module:

[0035] Based on the time series forecasting model, training is performed on time series data of historical meteorological variables and annotated historical meteorological extreme events. A combination of convolutional neural networks and recurrent neural networks is used to enhance the learning ability of local and long-term trends. To better consider the synergistic effects between modeling areas, a graph neural network is used to analyze the interaction of meteorological patterns between modeling areas. Nodes represent modeling areas, and edges represent the correlation between modeling areas, enhancing the ability to model spatiotemporal data. An integrated framework for multiple models is constructed to enhance adaptability and robustness to different meteorological patterns.

[0036] The modeling area is a spatial unit divided within the scope of the study or a uniformly divided high-resolution grid, and each modeling area serves as a node in the graph neural network.

[0037] Meteorological patterns are the spatiotemporal evolution patterns of meteorological variables within a modeling region and their interactions. In graph neural networks, edges are used to represent the similarities or physical transmission relationships between modeling regions.

[0038] (4) Mode fusion module:

[0039] Perform unified spatiotemporal gridding on data from multiple sources;

[0040] Perform Bayesian model averaging on the processed output and fuse the results of multiple models in the model training module through weighted averaging method;

[0041] Use error correction algorithm to perform dynamic error correction on the results;

[0042] (5) Security engine building modules:

[0043] The security engine building module supports real-time rolling forecasts, updating inference results every 10 minutes to ensure rapid response to sudden extreme weather events. It also updates extreme weather event warnings every 30 minutes, providing the latest event identification results, level assessments, and risk maps.

[0044] The safety engine building module outputs risk level maps, probability heat maps, and event time sequence diagrams for extreme weather events, helping relevant departments monitor weather changes in real time;

[0045] The safety engine construction module provides a decision support interface, connects data with port dispatching, airport meteorological platforms and environmental monitoring platforms, and pushes meteorological risk warnings and preventive measures in a timely manner.

[0046] Beneficial effects of the present invention:

[0047] (1) Data fusion and timeliness improvement;

[0048] Existing technologies often rely on a single model or data source, failing to fully utilize the combination of historical reanalysis data and multi-model real-time forecast data, resulting in poor warning timeliness. This invention, by combining historical ERA5 and other reanalysis data with multi-model ensemble forecasts, fully utilizes multi-source heterogeneous data to provide high-frequency, real-time dynamic monitoring of meteorological conditions, enabling rapid response and timely warnings, significantly improving the response speed to sudden extreme weather events in low-altitude environments.

[0049] (2) Improve the accuracy of identifying low-altitude, small-scale extreme weather;

[0050] Existing technologies often rely on numerical weather prediction (NWP) models, which have low spatial resolution and struggle to accurately capture small-scale, low-altitude severe weather events (such as wind shear, turbulence, and low visibility). This invention combines deep learning with physical reasoning methods, specifically by combining CNN and Transformer models. This method can extract localized, small-scale weather features from high-resolution data, improving the ability to precisely identify low-altitude weather events and reducing false positives and missed alerts.

[0051] (3) Multidimensional meteorological risk assessment system and standardized classification;

[0052] Traditional weather warning systems often rely on qualitative descriptions or simple rules, lacking standardized quantitative risk assessments. This invention, by constructing a multi-dimensional risk index system that includes pollution accumulation index, wind shear index, and low visibility risk level, provides users with more scientific and standardized risk assessment results, facilitating specific operational decisions in practical applications.

[0053] (4) Dynamic adaptive optimization and real-time adjustment;

[0054] Most existing technologies use fixed rules or thresholds for early warning, failing to dynamically adjust to real-time weather changes, leading to significant problems with false alarms and missed alerts. This invention, through a reinforcement learning mechanism, adjusts the early warning threshold based on real-time weather data, achieving adaptive optimization of the early warning strategy. This significantly reduces the probability of false alarms and missed alerts, further improving the robustness and reliability of the system. DETAILED DESCRIPTION

[0055] The specific implementation of the present invention is further described below in conjunction with the technical solution.

[0056] An artificial intelligence algorithm for the Bohai Sea low-altitude meteorological safety engine based on historical reanalysis and multi-modal data, including the following modules:

[0057] (1) Meteorological variable reconstruction and extraction module:

[0058] Historical reanalysis data and real-time observation data are multi-source fused to ensure uniform resolution in spatial and temporal distribution. The historical reanalysis data include data obtained from ERA5, JRA-55, and MERRA-2, while the real-time observation data include ground station wind temperature data, LIDAR inverted wind profiles, radar data, and microwave radiometer observation data. Boundary layer height, wind shear, vorticity, inversion intensity, gradient, and disturbance are used as meteorological variables. Interpolation methods are used to improve the resolution and expression accuracy of meteorological variables in temporal and spatial distribution, and then dimensionality reduction and compression are performed as follows: First, data integrity is achieved through multiple interpolation methods (such as Kriging and IDW). Second, principal component analysis (PCA) and empirical orthogonal function (EOF) are combined in the dimensionality reduction step to extract the main modal features. Finally, deep autoencoders are used for compression.

[0059] (1.1) Multi-source data fusion:

[0060] Historical reanalysis data (ERA5: 0.25°, JRA-55: 1.25°, MERRA-2: 0.5°) are uniformly resampled to a high-resolution grid (e.g., 5 km);

[0061] Real-time observation data is quality-controlled and then integrated into a high-resolution grid;

[0062] Quality control is the quality control processing of real-time observation data, including integrity check, physical rationality test, time and space consistency check and outlier removal, to ensure the accuracy and reliability of real-time observation data;

[0063] (1.2) Processing of meteorological variables:

[0064] Boundary layer height (BLH): The mixing layer height diagnostic formula (such as Richardson number method, Holzworth method) is integrated with the ERA5 diagnostic value;

[0065] Wind shear and vorticity: Calculate wind shear and vorticity index (VWS, wind shear magnitude) based on the 925-850hPa layer wind speed profile to identify low-altitude unstable conditions;

[0066] Inversion intensity: Calculate the temperature lapse rate in the near-surface layer and use strong inversion events as triggering conditions for pollution accumulation events or extreme low visibility events;

[0067] Gradient and disturbance: Use the central difference method to construct the temperature, humidity, and wind speed gradient fields and extract the local disturbance characteristics within the boundary layer;

[0068] (1.3) Feature Dimensionality Reduction and Deep Compressed Modeling:

[0069] First, use a variety of interpolation algorithms (such as Kriging, IDW, etc.) to perform spatial interpolation on missing or irregular data points to ensure data integrity;

[0070] Then, principal component analysis (PCA) and empirical orthogonal function (EOF) analysis were used to reduce the dimensionality of meteorological variables and extract the most representative features;

[0071] Finally, a deep autoencoder is used to compress high-dimensional meteorological data, extract potential high-order features, and reduce computational complexity. High-dimensional meteorological data is a three-dimensional meteorological data tensor composed of multiple meteorological variables, multiple vertical height layers, and multiple time steps.

[0072] (2) Extreme event identification module:

[0073] Based on the historical meteorological extreme events in the study area, a standardized labeling rule library was constructed. Historical meteorological extreme events include pollution accumulation events, wind shear events, and localized high wind events. Historical meteorological extreme events were collected to establish a sample event set. A long short-term memory (LSTM) network was used to perform reconstruction error detection on the observation data sequence of meteorological variables for the identification of meteorological extreme events. The observation data sequence of meteorological variables is the observation value of one or more meteorological variables at consecutive time points.

[0074] Use deep autoencoders to automatically detect and label historical extreme meteorological events. Finally, construct composite indicators of multiple meteorological variables, such as a combined indicator of temperature gradient and wind speed, to improve the sensitivity of identifying extreme meteorological events. The specific steps are as follows:

[0075] (2.1) Construction of standardized label rule library:

[0076] Based on the historical meteorological extreme events in the sample event set, the criteria are defined as follows:

[0077] Pollution accumulation event: BLH < 300m + wind speed < 2m / s + PM2.5 > 100μg / m³ + inversion layer;

[0078] Wind shear event: ΔV / ΔZ>0.015s⁻¹; where ΔV is the vertical height difference and ΔZ is the change in wind speed between two altitude levels;

[0079] Local gale events: 10-minute wind speed > 12 m / s + turbulent kinetic energy > threshold;

[0080] Label historical meteorological extreme events in the sample event set and construct training and validation sets;

[0081] (2.2) Anomaly detection method integration:

[0082] LSTM reconstruction error method: A long short-term memory (LSTM) network is trained using data from normal periods. When the LSTM is fed data containing extreme meteorological events, the error between the LSTM's actual observations and predictions increases significantly. Normal period data refers to time series data collected during periods without extreme meteorological events that reflect the normal variation patterns of meteorological variables.

[0083] Deep autoencoder identification: Deep autoencoders are trained to compress and reconstruct meteorological variables. A significant increase in reconstruction error indicates a candidate extreme meteorological event.

[0084] Local Outlier Factor (Isolation Forest & LOF): used for outlier detection in a low-dimensional compressed space; the low-dimensional compressed space is the space after dimensionality reduction by a deep autoencoder or PCA;

[0085] (2.3) Multivariate composite indicator design:

[0086] Construct mathematical models of the Polluted Weather Potential Index (PAI) and the Wind Shear Index (WSI);

[0087] The mathematical model of the Polluted Weather Potential Index is as follows: PAI=w1*BLH −1 +w2*windspeed −1 +w3*inversion intensity+w4*humidity stability;

[0088] Among them, w1-w4 are the weights of the corresponding meteorological variables;

[0089] The mathematical model of the wind shear index WSI is as follows: ;

[0090] Fuzzy logic rules or decision tree method (CART) are used to classify the levels of meteorological extreme events into four categories: mild, moderate, severe and extremely severe.

[0091] (3) Model training module:

[0092] Based on time series prediction models (such as LSTM, Transformer, and Temporal Fusion Transformer), training is performed on time series data of historical meteorological variables and annotated historical meteorological extreme events. A combination of convolutional neural networks (CNN) and recurrent neural networks (RNN) is used to enhance learning capabilities for local and long-term trends. To better account for the synergistic effects between modeling regions, a graph neural network (GCN) is used to analyze the interaction of meteorological patterns between modeling regions, with nodes representing modeling regions and edges representing the correlation between modeling regions, enhancing spatiotemporal data modeling capabilities. An integrated framework for multiple models (such as bagging, boosting, and stacking) is constructed to enhance adaptability and robustness to different meteorological patterns.

[0093] The modeling area is a spatial unit divided within the scope of the study or a uniformly divided high-resolution grid, and each modeling area serves as a node in the graph neural network.

[0094] Meteorological patterns are the spatiotemporal evolution patterns of meteorological variables within a modeling region and their interactions. In graph neural networks, edges are used to represent the similarities or physical transmission relationships between modeling regions.

[0095] The specific steps are as follows:

[0096] (3.1) Integration framework of multiple models:

[0097] The integration framework of multiple models is a multi-model collaborative spatiotemporal feature learning system. The multi-model collaborative spatiotemporal feature learning system includes a local time series modeling module, a key variable driven modeling module, a local spatial structure extraction module, a regional collaborative modeling module, and a convolution + recurrent hybrid time series modeling module. The five modules are run in parallel. The details are as follows:

[0098] (a) Local time series modeling module (LSTM or GRU):

[0099] Input: Time series data consisting of meteorological variables (temperature, humidity, wind speed, wind direction, BLH, etc.);

[0100] Task: Learn the historical evolution of meteorological variables and predict the continuous value changes in the next N hours;

[0101] Output: probability score of extreme meteorological events;

[0102] (b) Key variable driven modeling module (Transformer or TFT):

[0103] Input: Time series data consisting of meteorological variables (temperature, humidity, wind speed, wind direction, BLH, etc.) and the weighting mechanism of meteorological variables;

[0104] Task: Use the attention mechanism to capture which meteorological variables are more likely to drive the occurrence of extreme meteorological events in the current period;

[0105] Output: driving factor attention weight;

[0106] (c) Local spatial structure extraction module (CNN-LSTM):

[0107] Input: An image (2D matrix) of meteorological variables in a high-resolution grid.

[0108] Task: Use CNN to extract spatial gradient information and shape features (such as fronts and wind shear areas); LSTM then processes their temporal evolution trends;

[0109] Output: spatial gradient recognition score;

[0110] (d) Regional collaborative modeling module, select GCN or GAT:

[0111] Input: Construct a regional graph structure. This structure abstracts each modeling region within the study area into a node in a graph neural network, and uses edges between nodes to represent the relationships between modeling regions. The modeling region is a divided spatial unit or a uniformly divided high-resolution grid. Each modeling region node contains boundary layer height, wind speed, temperature and humidity, inversion intensity, and vorticity feature vectors.

[0112] Task: Use graph convolutional networks or graph attention networks to mine and model the risk propagation relationship and mutual influence mechanism between regions on the regional graph structure;

[0113] Output: Modeling regional linkage transmission risk score, which is used to describe the spatial diffusion trend and risk linkage intensity of meteorological extreme events between different modeling regions;

[0114] (e) Convolution + recurrent mixed temporal modeling module, select CNN-RNN (such as CNN-BiLSTM):

[0115] Input: historical meteorological variable time series, single meteorological variable or combination of multiple meteorological variables;

[0116] Task: Use CNN to extract local pattern features (such as short-term mutations and local fluctuations) of meteorological variable time series, and then use RNN (LSTM or GRU) to capture long-term dependencies and trends;

[0117] Output: Fusion time series forecast results, used to identify precursory features of extreme meteorological events;

[0118] (3.2) Multi-model integration method:

[0119] (a) Multi-model integration structure:

[0120] First-level model (Base Learners): Combine the outputs of the five modules in step (3.1) at the feature level to form an integrated feature vector and perform normalization.

[0121] Secondary fusion model (Meta Learner): uses lightweight integrated models such as logistic regression, random forest, and XGBoost to fuse the output results of the primary model; or uses a deep stacking neural network to self-learn the combined weights of the output of the primary model;

[0122] (b) Output content:

[0123] Probability of occurrence of extreme meteorological events: output probability of pollution accumulation, probability of local strong winds, and probability of wind shear;

[0124] Uncertainty estimation: Calculate the standard deviation and consistency between outputs through multi-model integration methods (such as stacking and Bayesian model averaging), thereby providing a credibility range;

[0125] Event level prediction results: Based on the integrated comprehensive score, combined with rules and fuzzy logic, the four levels of event levels (mild, moderate, severe, and extremely severe) are output;

[0126] Key factor explanation information: Backtracking the weights and contributions of important features in multi-model fusion to form a traceable decision path.

[0127] (c) Deployment form:

[0128] Based on multi-model deployment as a scalable model API interface, it accepts multi-variable inputs at any time period and automatically completes reasoning, judgment and interpretation output.

[0129] (4) Mode fusion module:

[0130] Data from multiple sources (reanalysis, observations, and future scenarios) are uniformly processed in space and time to ensure comparability. The processed outputs are subjected to Bayesian model averaging (BMA), fusing the results of different models through weighted averaging to reduce the error of a single model. Error correction algorithms (such as Kalman filtering) are applied to dynamically correct the results to improve model accuracy and predictive capabilities. Based on short-term warnings of 0–7 days and combined with future climate scenarios for the 2030s, long-term trends and the likelihood of extreme events are assessed to provide assessment and decision support for future meteorological risks. The specific implementation steps are as follows:

[0131] (4.1) Spatial and temporal standardization:

[0132] Establish a unified spatial grid (e.g., 0.05°×0.05°) and temporal resolution (10 minutes to 1 hour) and resample multi-source data (ERA5 / observation / CMIP6 data).

[0133] (4.2) Bayesian Model Averaging (BMA) Ensemble:

[0134] The outputs of multiple models are weighted averaged, and the weights are adaptively updated based on the performance of the models during the historical validation period; providing a fused probability density function of extreme weather events.

[0135] (4.3) Dynamic error correction:

[0136] Short-term forecast biases are corrected using dynamic linear models (DLM) or Kalman filtering methods; long-term future climate (CMIP6 SSP scenarios) are aligned with current observations through bias correction functions (such as Quantile Mapping).

[0137] (5) Security engine building modules:

[0138] The Safety Engine Building Module supports real-time rolling forecasts, updating inference results every 10 minutes to ensure rapid response to sudden extreme weather events. Extreme weather event warnings are updated every 30 minutes, providing the latest event identification results, level assessments, and risk maps. The Safety Engine Building Module outputs extreme weather event risk level maps, probability heat maps, and event time series diagrams, helping relevant departments monitor weather changes in real time. The Safety Engine Building Module provides a decision-making support interface, connecting data with port dispatchers, airport meteorological platforms, and environmental monitoring platforms to promptly deliver meteorological risk warnings and preventative measures.

[0139] The specific implementation steps are as follows:

[0140] (5.1) Real-time reasoning and scrolling mechanism:

[0141] The system accesses real-time meteorological data streams, triggering inference of the security engine building module every 10 minutes; and aggregates and updates extreme weather event identification results, level assessments, and risk maps every 30 minutes.

[0142] (5.2) Output content design:

[0143] Risk Level Map: displays the spatial distribution of event levels (mild to very severe);

[0144] Probabilistic Map: Displays the probability of an event occurring using contour lines or heat maps;

[0145] Timeline Plot: Shows the changes in extreme event index within the next 24 hours.

[0146] (5.3) System docking and deployment:

[0147] Develop interfaces with port dispatching, airport meteorology, and environmental monitoring platforms; export API interface services, provide machine-to-machine (M2M) calling capabilities, and achieve docking with city and regional meteorological safety operation platforms.

[0148] Through the organic combination of the above modules, this system can achieve high-precision, real-time risk identification and early warning of low-altitude, small-scale severe weather in the Bohai Rim region, providing effective technical support for regional meteorological safety.

Claims

1. An artificial intelligence algorithm for the Bohai Sea low-altitude meteorological safety engine based on historical reanalysis and multi-modal data, characterized by: Includes the following modules: (1) Meteorological variable reconstruction and extraction module: Historical reanalysis data and real-time observation data are fused from multiple sources to ensure uniform resolution in spatial and temporal distribution. Historical reanalysis data include data from ERA5, JRA-55, and MERRA-2, while real-time observation data include ground station wind and temperature data, LIDAR-derived wind profiles, radar data, and microwave radiometer observations. Boundary layer height, wind shear, vorticity, inversion intensity, gradient, and disturbance are used as meteorological variables. Interpolation methods are used to improve the resolution and accuracy of meteorological variables in temporal and spatial distribution, followed by dimensionality reduction and compression. (2) Extreme event identification module: Based on the historical meteorological extreme events in the study area, a standardized labeling rule library was constructed. Historical meteorological extreme events include pollution accumulation events, wind shear events, and localized high wind events. Historical meteorological extreme events were collected to establish a sample event set. A long-short-term memory network was used to perform reconstruction error detection on the observation data sequence of meteorological variables for the identification of meteorological extreme events. The observation data sequence of meteorological variables is the observation value of one or more meteorological variables at consecutive time points. Use deep autoencoders to automatically detect and label historical extreme meteorological events. Finally, construct composite indicators of multiple meteorological variables to improve the sensitivity of identifying extreme meteorological events. (3) Model training module: Based on the time series forecasting model, training is performed on time series data of historical meteorological variables and annotated historical meteorological extreme events. A combination of convolutional neural networks and recurrent neural networks is used to enhance the learning ability of local and long-term trends. To better consider the synergistic effects between modeling areas, a graph neural network is used to analyze the interaction of meteorological patterns between modeling areas. Nodes represent modeling areas, and edges represent the correlation between modeling areas, enhancing the ability to model spatiotemporal data. An integrated framework for multiple models is constructed to enhance adaptability and robustness to different meteorological patterns. The modeling area is a spatial unit divided within the scope of the study or a uniformly divided high-resolution grid, and each modeling area serves as a node in the graph neural network. Meteorological patterns are the spatiotemporal evolution patterns of meteorological variables within a modeling region and their interactions. In graph neural networks, edges are used to represent the similarities or physical transmission relationships between modeling regions. (4) Mode fusion module: Perform unified spatiotemporal gridding on data from multiple sources; Perform Bayesian model averaging on the processed output and fuse the results of multiple models in the model training module through weighted averaging method; Use error correction algorithm to perform dynamic error correction on the results; (5) Security engine building modules: The security engine building module supports real-time rolling forecasts, updating inference results every 10 minutes to ensure rapid response to sudden extreme weather events. It also updates extreme weather event warnings every 30 minutes, providing the latest event identification results, level assessments, and risk maps. The safety engine construction module outputs risk level maps, probability heat maps, and event time sequence diagrams for extreme weather events, helping relevant departments monitor weather changes in real time; The safety engine construction module provides a decision support interface, connects data with port dispatching, airport meteorological platforms and environmental monitoring platforms, and pushes meteorological risk warnings and preventive measures in a timely manner.

2. The artificial intelligence algorithm for the Bohai Sea low-altitude meteorological safety engine based on historical reanalysis and multi-modal data according to claim 1 is characterized in that: The specific implementation process of the meteorological variable reconstruction and extraction module is as follows: (1.1) Multi-source data fusion: Historical reanalysis data are uniformly resampled to a high-resolution grid; Real-time observation data is quality-controlled and then integrated into a high-resolution grid; (1.2) Processing of meteorological variables: Boundary layer height: The mixed layer height diagnostic formula is integrated with the ERA5 diagnostic value; Wind shear and vorticity: Calculate wind shear and vorticity indices based on the 925-850hPa wind speed profile to identify low-altitude unstable conditions; Inversion intensity: Calculate the temperature lapse rate in the near-surface layer and use strong inversion events as triggering conditions for pollution accumulation events or extreme low visibility events; Gradient and disturbance: Use the central difference method to construct the temperature, humidity, and wind speed gradient fields and extract the local disturbance characteristics within the boundary layer; (1.3) Feature Dimensionality Reduction and Deep Compressed Modeling: First, a variety of interpolation algorithms are used to spatially interpolate missing or irregular data points to ensure the integrity of meteorological variable data; Then, principal component analysis and empirical orthogonal function analysis are used to reduce the dimension of meteorological variables and extract the most representative features; Finally, a deep autoencoder is used to compress high-dimensional meteorological data, extract potential high-order features, and reduce computational complexity; the high-dimensional meteorological data is a three-dimensional meteorological data tensor composed of multiple meteorological variables, multiple vertical height layers, and multiple time steps.

3. The Bohai Rim Low-Altitude Meteorological Safety Engine artificial intelligence algorithm based on historical reanalysis and multi-modal data according to claim 1 is characterized in that: The specific implementation process of the extreme event identification module is as follows: (2.1) Construction of standardized label rule library: Based on the historical meteorological extreme events in the sample event set, the criteria are defined as follows: Pollution accumulation event: BLH < 300m + wind speed < 2m / s + PM2.5 > 100μg / m³ + inversion layer; Wind shear event: ΔV / ΔZ>0.015s⁻¹; where ΔV is the vertical height difference and ΔZ is the change in wind speed between two altitude levels; Local gale events: 10-minute wind speed > 12 m / s + turbulent kinetic energy > threshold; Label historical meteorological extreme events in the sample event set and construct training and validation sets; (2.2) Anomaly detection method integration: LSTM reconstruction error method: Long-short-term memory (LSTM) networks are trained using data from normal periods. When the LSTM network is fed data containing extreme meteorological events, the error between the LSTM network's actual observations and predictions increases significantly. Normal period data refers to time series data collected during periods without extreme meteorological events that reflect the normal variation patterns of meteorological variables. Deep autoencoder identification: Deep autoencoders are trained to compress and reconstruct meteorological variables. A significant increase in reconstruction error indicates a candidate extreme meteorological event. Local outlier factor: used to detect outliers in a low-dimensional compressed space; the low-dimensional compressed space is the space after dimensionality reduction by a deep autoencoder or PCA; (2.3) Multivariate composite indicator design: Construct mathematical models of the Polluted Weather Potential Index (PAI) and the Wind Shear Index (WSI); The mathematical model of the Polluted Weather Potential Index is as follows: PAI=w1*BLH −1 +w2*windspeed −1 +w3*inversion intensity+w4*humidity stability; Among them, w1-w4 are the weights of the corresponding meteorological variables; The mathematical model of the wind shear index WSI is as follows: ; Fuzzy logic rules or decision tree methods are used to classify meteorological extreme events into four categories: light, moderate, severe, and extremely severe.

4. The artificial intelligence algorithm for the Bohai Sea low-altitude meteorological safety engine based on historical reanalysis and multi-modal data according to claim 1 is characterized in that: The specific implementation process of the model training module is as follows: (3.1) Integration framework of multiple models: The integration framework of multiple models is a multi-model collaborative spatiotemporal feature learning system; the multi-model collaborative spatiotemporal feature learning system includes a local time series modeling module, a key variable driven modeling module, a local spatial structure extraction module, a regional collaborative modeling module, and a convolution + recurrent hybrid time series modeling module, all of which are run in parallel. The details are as follows: (a) Local time series modeling module, select LSTM or GRU: Input: time series data consisting of meteorological variables; Task: Learn the historical evolution of meteorological variables and predict the continuous value changes in the next N hours; Output: probability score of extreme meteorological events; (b) Key variable driven modeling module, select Transformer or TFT: Input: time series data composed of meteorological variables and the weight mechanism of meteorological variables; Task: Use the attention mechanism to capture which meteorological variables are more likely to drive the occurrence of extreme meteorological events in the current period; Output: driving factor attention weight; (c) Local spatial structure extraction module, CNN-LSTM: Input: An image composed of meteorological variables in a high-resolution grid; Task: Use CNN to extract spatial gradient information and shape features; LSTM then processes their temporal evolution trend; Output: spatial gradient recognition score; (d) Regional collaborative modeling module, select GCN or GAT: Input: Construct a regional graph structure. This structure abstracts each modeling region within the study area into a node in a graph neural network, and uses edges between nodes to represent the relationships between modeling regions. The modeling region is a divided spatial unit or a uniformly divided high-resolution grid. Each modeling region node contains boundary layer height, wind speed, temperature and humidity, inversion intensity, and vorticity feature vectors. Task: Use graph convolutional networks or graph attention networks to mine and model the risk propagation relationship and mutual influence mechanism between regions on the regional graph structure; Output: Modeling regional linkage transmission risk score, which is used to describe the spatial diffusion trend and risk linkage intensity of meteorological extreme events between different modeling regions; (e) Convolution + recurrent mixed temporal modeling module, select CNN-RNN: Input: Time series data of historical meteorological variables, which can be a single meteorological variable or a combination of multiple meteorological variables; Task: Use CNN to extract local pattern features of time series data of meteorological variables, and then use RNN to capture long-term dependencies and trends; Output: Fusion time series forecast results, used to identify precursory features of extreme meteorological events; (3.2) Multi-model integration method: (a) Multi-model integration structure: First-level model: The outputs of the five modules in step (3.1) are combined at the feature level to form an integrated feature vector, which is then standardized. Secondary fusion model: Use logistic regression, random forest, or XGBoost to fuse the output of the primary model; or use a deep stacking neural network to self-learn the combined weights of the output of the primary model; (b) Output content: Probability of occurrence of extreme meteorological events: output probability of pollution accumulation, probability of local strong winds, and probability of wind shear; Uncertainty estimation: Calculate the standard deviation and consistency between outputs through multi-model integration methods to provide a confidence range; Event level prediction results: Based on the comprehensive score after multi-model fusion, combined with rules and fuzzy logic, the four levels of event levels (mild, moderate, severe, and extremely severe) are output; Key factor explanation information: Backtracking the weights and contributions of important features in multi-model fusion to form a traceable decision path; (c) Deployment form: Based on multi-model deployment as a scalable model API interface, it accepts multi-variable inputs at any time period and automatically completes reasoning, judgment and interpretation output.

5. The artificial intelligence algorithm for the Bohai Sea low-altitude meteorological safety engine based on historical reanalysis and multi-modal data according to claim 1 is characterized in that: The specific implementation process of the mode fusion module is as follows: (4.1) Spatial and temporal standardization: Establish a unified spatial grid and temporal resolution to resample multi-source data; (4.2) Bayesian model averaging fusion: Perform a weighted average of the outputs of multiple models, with the weights adaptively updated based on their performance during historical validation periods; providing a fused probability density function for extreme weather events; (4.3) Dynamic error correction: Short-term forecast biases are corrected using dynamic linear models or Kalman filtering methods; long-term future climate is aligned with current observations through a bias correction function.

6. The artificial intelligence algorithm for the Bohai Sea low-altitude meteorological safety engine based on historical reanalysis and multi-modal data according to claim 1 is characterized in that: The specific implementation process of the security engine building module is as follows: (5.1) Real-time reasoning and scrolling mechanism: Access to real-time meteorological data streams triggers inference of the security engine building module every 10 minutes; and aggregates and updates the identification results, level assessments, and risk maps of extreme weather events every 30 minutes. (5.2) Output content design: Risk level map: displays the spatial distribution of event levels; Probability heat map: displays the probability of an event occurring using contour lines or heat maps; Event time series diagram: shows the changes in extreme event index in the next 24 hours; (5.3) System docking and deployment: Develop interfaces with port dispatching, airport meteorology, and environmental monitoring platforms; export API interface services, provide machine-to-machine calling capabilities, and achieve docking with city and regional meteorological safety operation platforms.

Images