Dynamic Early Warning Method for Weather Modification Operations Based on Multi-Source Data Fusion

Abstract

This invention discloses a dynamic early warning method for weather modification operations based on multi-source data fusion, specifically addressing the field of data risk. This includes multi-source meteorological monitoring data acquisition, data preprocessing and analysis, multi-source meteorological monitoring data fusion, centralized operational risk early warning and decision-making, and continuous monitoring and assessment. This invention significantly improves the timeliness of early warnings through dynamic multi-source data fusion. It utilizes timestamp alignment and spatial matching to synchronously integrate meteorological fields, equipment status, and environmental parameters, combined with real-time analysis using deep neural networks, achieving minute-level risk prediction and drastically shortening response time. Its multi-source collaborative mechanism breaks down data silos. Through a fusion strategy at the data layer, feature layer, and decision layer, it deeply correlates heterogeneous information such as wind field, temperature and humidity field data, and radar signals to construct a global risk assessment model, enhancing the reliability of complex meteorological decisions.

Description

Technical Field

[0001] This invention relates to the field of data risk technology, and more specifically, to a dynamic early warning method for weather modification operations based on multi-source data fusion. Background Technology

[0002] Artificial weather modification operations rely on precise early warning technologies to improve their safety and effectiveness. Current technologies typically involve collecting basic data through meteorological detection equipment, performing preliminary cleaning and analysis at regional nodes to generate early warning information, and finally having the command center execute operational decisions.

[0003] The process has three main shortcomings: First, traditional methods rely on a single data source and static models, making it difficult to capture sudden weather evolution and resulting in delayed warnings; second, the dispersed data processing nodes lead to insufficient fusion of multi-source meteorological information, and key parameters such as radar status and airspace restrictions are not analyzed in conjunction with meteorological field data; and finally, the visualization terminal has weak functionality, lacking dynamic warning area identification and interactive decision support, which affects command efficiency. Summary of the Invention

[0004] To overcome the aforementioned deficiencies in existing technologies, this invention provides a dynamic early warning method for weather modification operations based on multi-source data fusion. The method addresses the problems of poor early warning timeliness, insufficient data fusion leading to information silos, weak visualization capabilities affecting decision-making response delays, and overall limited ability to handle sudden weather events in traditional methods mentioned in the background.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a dynamic early warning method for weather modification operations based on multi-source data fusion, comprising:

[0006] S1: Multi-source meteorological monitoring data acquisition: Collect multi-source meteorological monitoring data through meteorological detection equipment. The multi-source meteorological monitoring data should cover multi-dimensional information, including meteorological field data, detection equipment status data, and operating environment parameters.

[0007] S2: Data Preprocessing and Analysis: The regional data processing node receives data from meteorological detection equipment, performs preprocessing operations on the data, conducts preliminary analysis on the preprocessed data, and filters out effective information for transmission;

[0008] S3: Multi-source meteorological monitoring data fusion: The multi-source data fusion center receives multi-source meteorological monitoring data from multiple regional data processing nodes and performs data fusion;

[0009] S4: Centralized Operational Risk Early Warning and Decision-Making: The early warning and command center receives the fused data, analyzes and reasons about the fused data through artificial intelligence models, and analyzes and locates potential operational risks based on the model output results;

[0010] S5: Continuous monitoring and evaluation: Continuously monitor the risk analysis results during the potential operational risk diagnosis process, evaluate the effectiveness of the potential operational risk diagnosis methods, and adjust the diagnosis methods based on the evaluation results.

[0011] Preferably, the multi-source meteorological monitoring data includes meteorological field data, detection equipment status data, and operational environment parameters. Specifically, the meteorological field data includes wind field, temperature and humidity field, air pressure field, and water vapor flux; the detection equipment status data includes radar status, radiosonde status, and ground station status; and the operational environment parameters include airspace status and terrain influence.

[0012] Preferably, the preprocessing operation specifically refers to cleaning and extracting features from the collected multi-source meteorological monitoring data through regional data processing nodes, and the preliminary analysis includes detecting local strong convection initiation signals and water vapor convergence characteristics using lightweight models, and uploading them to the multi-source data fusion center.

[0013] Preferably, the fusion of multi-source meteorological monitoring data includes the following steps:

[0014] A1: Receives data uploaded from data processing nodes in various regions and summarizes and organizes multi-source meteorological monitoring data of the same type;

[0015] A2: Perform timestamp alignment and spatial location matching on multi-source meteorological monitoring data. By calibrating and synchronizing the timestamps, the multi-source meteorological monitoring data can be made consistent in both time and space.

[0016] A3: Synchronize data from different sampling frequencies to ensure that data from different modalities are consistent in time resolution;

[0017] A4: Select an appropriate fusion method to fuse the processed data.

[0018] Preferably, operational risk warning and decision-making includes the following steps:

[0019] B1: The early warning and command center receives the fused data transmitted from the multi-source data fusion center and temporarily stores it in the cache area;

[0020] B2: Select and load the artificial intelligence model. After loading the model, adjust the model according to the actual situation of the current network.

[0021] B3: Input the temporarily cached data into the selected adapted artificial intelligence model, then perform model analysis and reasoning and output the analysis results.

[0022] Preferably, the evaluation of the effectiveness of potential operational risk diagnosis methods specifically includes the following steps:

[0023] C1: Set a time interval, monitor and collect the true positive example X1, false positive example X2, true negative example X3, and false negative example X4 within the time interval, and collect n consecutive time intervals, labeled as 1, 2, ..., i, ..., n respectively. Calculate the average true positive example, average false positive example, average true negative example, and average false negative example using the true positive example X1i, false positive example X2i, true negative example X3i, and false negative example X4i of each time interval.

[0024] C2: Divide the sum of the average true examples and the average true negative examples by the sum of the average true examples, average false positive examples, average true negative examples, and average false negative examples using the formula. To obtain the early warning accuracy ηA, divide the average true positive rate by the sum of the average true positive rate and the average false positive rate using the formula... To obtain the early warning accuracy ηP, divide the average true cases by the sum of the average true cases and the average false cases using the formula... The warning recall rate ηR is obtained, and the product of the warning precision rate and the warning recall rate is divided by the sum of the warning precision rate and the warning recall rate and multiplied by 2. The F1 value is obtained by using the formula F1=2×ηP×ηR / (ηP+ηR).

[0025] C3: The risk detection accuracy, early warning precision, early warning recall, and F1 score are combined using the formula Score=ω1×ηA+ω2×ηP+ω3×ηR+ω4×F1 to obtain the comprehensive risk detection score, Q, where ω1, ω2, ω3, and ω4 represent the weighting coefficients of risk detection accuracy, early warning precision, early warning recall, and F1 score, respectively, and their sum is 1.

[0026] Preferably, different sampling frequencies are used, such as meteorological field data being collected multiple times per second and detection equipment status data being collected every few minutes. The multi-source data fusion center synchronously processes these data with different sampling rates. By using interpolation or downsampling techniques, data with higher sampling rates can be downsampled to match the data with lower sampling rates in terms of time intervals. For data with lower sampling rates, more data points can be generated using interpolation methods to align it with the high-sampling-rate data in terms of time scale, thus avoiding information loss or incorrect fusion due to differences in sampling rates.

[0027] Preferably, the data fusion method includes data-layer fusion, feature-layer fusion, and decision-layer fusion. Specifically, data-layer fusion includes direct concatenation and weighted fusion. Direct concatenation involves directly concatenating raw or pre-processed data from different modalities. Weighted fusion assigns different weights to each modality based on its importance or reliability, and then performs a weighted sum. Feature-layer fusion includes feature selection and combination, and feature transformation and fusion. Feature selection and combination involves selecting representative features from different modalities and combining them into a new feature set. Feature transformation and fusion involves transforming the features of different modalities to achieve the same feature space or distribution before fusion. Decision-layer fusion includes independent classification and result fusion, and probabilistic fusion. Independent classification and result fusion involves using independent classifiers or diagnostic models to perform operational risk warnings on data from different modalities, obtaining their respective diagnostic results, and then fusing these diagnostic results. Probabilistic fusion involves fusing the risk probabilities of each independent diagnostic model to obtain a comprehensive risk probability.

[0028] Preferably, true positives represent the number of high-risk weather events that were correctly warned, false positives represent the number of false warnings, true negatives represent the number of events that were correctly determined to be without operational conditions, and false negatives represent the number of events that should have been warned but were not.

[0029] Preferably, the operational risk warning accuracy reflects the proportion of correctly diagnosed samples to the total number of samples, the warning precision reflects the proportion of samples diagnosed as risks that are actually risks, the warning recall reflects the proportion of actual risk samples that are correctly diagnosed, and the F1 value reflects the difference between the warning precision and the warning recall.

[0030] Preferably, the weighting coefficients are determined based on the operational objectives, such as rainfall enhancement focusing on hit rate ηR, and hail prevention focusing on accuracy ηP and lead time. Cross-validation and other methods are used to conduct experiments on different weight combinations. By comparing the model's performance on the validation set under different weights, the weight combination that optimizes the comprehensive evaluation index is selected.

[0031] Preferably, adjusting the diagnostic method specifically includes setting a scoring standard value, comparing the comprehensive risk detection score with the scoring standard value, and if the comprehensive risk detection score is greater than the scoring standard value, it indicates that the potential operational risk detection effect is qualified and no adjustment is required. If the comprehensive risk detection score is less than the scoring standard value, it indicates that the potential operational risk detection effect is unqualified. The potential operational risk detection method is dynamically adjusted in combination with the specific risk detection accuracy, early warning precision rate, early warning recall rate and F1 value.

[0032] The technical effects and advantages of this invention are as follows:

[0033] This invention significantly improves the timeliness of early warnings through dynamic fusion of multi-source data. By utilizing timestamp alignment and spatial matching technology, it synchronously integrates meteorological fields, equipment status, and environmental parameters, and combines them with a deep neural network model for real-time analysis, enabling minute-level risk prediction for sudden weather events such as severe convection, thus greatly shortening emergency response time.

[0034] The multi-source collaboration mechanism effectively breaks down data silos. Through the fusion strategy of data layer, feature layer and decision layer, it deeply correlates wind field, temperature and humidity field data with heterogeneous information such as radar signals and airspace status to form a global operation risk assessment model and enhance the reliability of decision-making under complex meteorological conditions.

[0035] The upgraded visual interactive system optimizes command efficiency, dynamically renders warning area levels and overlays forecast field information on the terminal, and provides interactive functions such as avoidance route planning. It helps personnel to intuitively grasp the weather situation and quickly adjust operation plans, forming a closed loop of "monitoring-early warning-feedback" and improving the accuracy of the whole process control. Attached Figure Description

[0036] Figure 1 This is a schematic diagram of the overall structure of the present invention. Detailed Implementation

[0037] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0038] refer to Figure 1 The dynamic early warning method for weather modification operations based on multi-source data fusion, as shown, includes:

[0039] S1: Multi-source meteorological monitoring data acquisition: Collect multi-source meteorological monitoring data through meteorological detection equipment. The multi-source meteorological monitoring data should cover multi-dimensional information, including meteorological field data, detection equipment status data, and operating environment parameters.

[0040] Furthermore, multi-source meteorological monitoring data includes meteorological field data, detection equipment status data, and operational environment parameters. Specifically, meteorological field data includes wind field, temperature and humidity field, air pressure field, and water vapor flux; detection equipment status data includes radar status, radiosonde status, and ground station status; and operational environment parameters include airspace status and terrain influence.

[0041] S2: Data Preprocessing and Analysis: The regional data processing node receives data from meteorological detection equipment, performs preprocessing operations on the data, conducts preliminary analysis on the preprocessed data, and filters out effective information for transmission;

[0042] Furthermore, the preprocessing operation specifically refers to cleaning and extracting features from the collected multi-source meteorological monitoring data through regional data processing nodes. The preliminary analysis includes detecting local strong convection initiation signals and water vapor convergence characteristics using lightweight models and uploading them to the multi-source data fusion center.

[0043] S3: Multi-source meteorological monitoring data fusion: The multi-source data fusion center receives multi-source meteorological monitoring data from multiple regional data processing nodes and performs data fusion;

[0044] Furthermore, the fusion of multi-source meteorological monitoring data specifically includes the following steps:

[0045] A1: Receives data uploaded from data processing nodes in various regions and summarizes and organizes multi-source meteorological monitoring data of the same type;

[0046] This embodiment specifically illustrates the process of summarizing and organizing multi-source meteorological monitoring data. For example, it summarizes the network characteristic data sent by all regional data processing nodes into one data set and the equipment status characteristic data into another set. This facilitates the unified processing and analysis of data of the same modality in the future and prepares for data fusion.

[0047] A2: Perform timestamp alignment and spatial location matching on multi-source meteorological monitoring data. By calibrating and synchronizing the timestamps, the multi-source meteorological monitoring data can be made consistent in both time and space.

[0048] This embodiment requires a specific explanation: because the data collection times of data processing nodes in different regions may vary slightly, timestamp alignment and spatial location matching are necessary to ensure the accuracy of data fusion. Timestamps are the time information recorded during data collection; inconsistencies in timestamps will lead to incorrect associations when analyzing risks. Timestamp alignment and spatial location matching ensure that different modalities collected at the same time can be correctly matched.

[0049] A3: Synchronize data from different sampling frequencies to ensure that data from different modalities are consistent in time resolution;

[0050] This embodiment requires specific explanation regarding different sampling frequencies. For example, meteorological field data is collected multiple times per second, while detection equipment status data is collected every few minutes. The multi-source data fusion center synchronously processes these data with different sampling rates. This can be achieved by using interpolation or downsampling techniques. For data with a higher sampling rate, downsampling can be used to match the data with a lower sampling rate in terms of time interval. For data with a lower sampling rate, interpolation methods can be used to generate more data points to align it with the high sampling rate data in terms of time scale, thus avoiding information loss or incorrect fusion due to differences in sampling rates.

[0051] A4: Select an appropriate fusion method to fuse the processed data.

[0052] This embodiment specifically describes the data fusion method, which includes data-layer fusion, feature-layer fusion, and decision-layer fusion. Data-layer fusion specifically includes direct concatenation and weighted fusion. Direct concatenation involves directly concatenating raw or pre-processed data from different modalities. Weighted fusion assigns different weights to each modality based on its importance or reliability, and then performs a weighted sum. Feature-layer fusion specifically includes feature selection and combination, and feature transformation and fusion. Feature selection and combination involves selecting representative features from different modalities and combining them into a new feature set. Feature transformation and fusion involves transforming the features of different modalities to achieve the same feature space or distribution before fusion. Decision-layer fusion specifically includes independent classification and result fusion, and probabilistic fusion. Independent classification and result fusion involves using independent classifiers or diagnostic models to perform operational risk warnings on data from different modalities, obtaining their respective diagnostic results, and then fusing these diagnostic results. Probabilistic fusion involves fusing the risk probabilities of each independent diagnostic model to obtain a comprehensive risk occurrence probability.

[0053] S4: Centralized Operational Risk Early Warning and Decision-Making: The early warning and command center receives the fused data, analyzes and reasons about the fused data through artificial intelligence models, and analyzes and locates potential operational risks based on the model output results;

[0054] Furthermore, operational risk warning and decision-making include the following steps:

[0055] B1: The early warning and command center receives the fused data transmitted from the multi-source data fusion center and temporarily stores it in the cache area;

[0056] B2: Select and load the artificial intelligence model. After loading the model, adjust the model according to the actual situation of the current network.

[0057] In this embodiment, it should be specifically explained that the early warning command center stores a variety of artificial intelligence models suitable for diagnosing potential operational risks, including deep neural networks, convolutional neural networks, recurrent neural networks, random forests, and support vector machines. Based on the characteristics of the network, the objectives of operational risk early warning, and the characteristics of historical data, the most suitable model is selected and loaded. Small-scale pre-training is performed using historical risk data to make the model better adapt to the operational risk early warning task.

[0058] B3: Input the temporarily cached data into the selected adapted artificial intelligence model, then perform model analysis and reasoning and output the analysis results.

[0059] This embodiment requires specific explanation regarding the input process. It is crucial to strictly ensure that the order and structure of the data match the model's expectations to avoid errors in the model's inference results due to incorrect data format. Model analysis and inference include forward propagation and feature extraction, and model inference and output. Forward propagation and feature extraction refer to the process of forward propagation after data enters the model. As the data propagates through the network, these features are gradually abstracted from the original meteorological field data features into key feature representations related to potential operational risks. Model inference and output refer to the final output of the inference results after a series of feature extractions and processing. In risk classification, the model outputs the probability value of each possible risk type; the category with the highest probability is the risk type predicted by the model. In risk location, the model may output the probability of risk occurring at each node or link in the network. By sorting these probability values, the location most likely to experience risk can be determined.

[0060] S5: Continuous monitoring and evaluation: Continuously monitor the risk analysis results during the potential operational risk diagnosis process, evaluate the effectiveness of the potential operational risk diagnosis methods, and adjust the diagnosis methods based on the evaluation results;

[0061] Furthermore, evaluating the effectiveness of potential operational risk diagnostic methods specifically includes the following steps:

[0062] C1: Set a time interval, monitor and collect the true positive example X1, false positive example X2, true negative example X3, and false negative example X4 within the time interval, and collect n consecutive time intervals, labeled as 1, 2, ..., i, ..., n respectively. Calculate the average true positive example, average false positive example, average true negative example, and average false negative example using the true positive example X1i, false positive example X2i, true negative example X3i, and false negative example X4i of each time interval.

[0063] In this embodiment, it should be specifically explained that true positives represent the number of high-risk weather events that are correctly warned, false positives represent the number of false warnings, true negatives represent the number of events that are correctly determined to be without operational conditions, and false negatives represent the number of events that should have been warned but were not.

[0064] C2: Divide the sum of the average true examples and the average true negative examples by the sum of the average true examples, average false positive examples, average true negative examples, and average false negative examples using the formula. To obtain the early warning accuracy ηA, divide the average true positive rate by the sum of the average true positive rate and the average false positive rate using the formula... To obtain the early warning accuracy ηP, divide the average true cases by the sum of the average true cases and the average false cases using the formula... The warning recall rate ηR is obtained, and the product of the warning precision rate and the warning recall rate is divided by the sum of the warning precision rate and the warning recall rate and multiplied by 2. The F1 value is obtained by using the formula F1=2×ηP×ηR / (ηP+ηR).

[0065] In this embodiment, it should be specifically explained that the accuracy rate of the operation risk warning reflects the proportion of correctly diagnosed samples to the total number of samples; the precision rate of the warning reflects the proportion of samples diagnosed as risk that are actually risky; the recall rate of the warning reflects the proportion of actual risk samples that are correctly diagnosed; and the F1 value reflects the difference between the precision rate and the recall rate of the warning.

[0066] C3: The risk detection accuracy, early warning precision, early warning recall, and F1 score are combined using the formula Score=ω1×ηA+ω2×ηP+ω3×ηR+ω4×F1 to obtain the comprehensive risk detection score, Q, where ω1, ω2, ω3, and ω4 represent the weighting coefficients of risk detection accuracy, early warning precision, early warning recall, and F1 score, respectively, and their sum is 1.

[0067] In this embodiment, it should be specifically noted that the determination of the weighting coefficients must be based on the operational objectives. For example, rain enhancement focuses on the hit rate ηR, while hail prevention focuses on the accuracy rate ηP and lead time. Cross-validation and other methods are used to conduct experiments on different weight combinations. By comparing the model's performance on the validation set under different weights, the weight combination that optimizes the comprehensive evaluation index is selected.

[0068] Furthermore, adjusting the diagnostic method specifically includes setting a scoring standard value, comparing the comprehensive risk detection score with the scoring standard value, and if the comprehensive risk detection score is greater than the scoring standard value, it indicates that the potential operational risk detection effect is qualified and no adjustment is needed. If the comprehensive risk detection score is less than the scoring standard value, it indicates that the potential operational risk detection effect is unqualified. The potential operational risk detection method is dynamically adjusted in combination with the specific risk detection accuracy, early warning precision rate, early warning recall rate and F1 score.

[0069] This embodiment requires specific explanation of the adjustment methods, including adjusting model parameters, optimizing feature selection, improving data collection, and integrating multiple diagnostic methods. Adjusting model parameters refers to adjusting the parameters of the operational risk warning model based on the problems reflected in the scoring results. For example, when the recall rate is low, the model's complexity can be increased to better capture risk features, thereby improving the ability to identify risk samples. Improving data collection means that if the scoring results indicate data problems, such as imbalanced data leading to insufficient samples for certain risk types, affecting the model's ability to detect these risks, the data collection method needs to be improved. Integrating multiple diagnostic methods means considering the fusion of different operational risk warning methods based on the scoring results.

[0070] S6: Visual Interaction: Intuitively displays real-time weather conditions and forecast fields, warning area levels, operational suggestions and related data, and provides interactive functions for command personnel and operational personnel.

[0071] This embodiment specifically explains that the visual interactive terminal obtains the operational risk warning results and adjustment methods from the early warning command center, and presents them to the user in an intuitive and visual way, such as radar mosaic, satellite cloud image animation, distribution map of key physical quantities (water vapor, vertical velocity, unstable energy), numerical forecast product comparison, and dynamic map of the warning area, clearly presenting the risks. In addition, users can interact with the system through the visual interactive terminal.

[0072] This invention collects multi-source meteorological monitoring data through meteorological detection equipment, covering meteorological field data such as wind field and temperature and humidity field, detection equipment status data such as radar status, and operational environment parameters such as airspace status. Next, the data is preprocessed and analyzed at the regional data processing node, including data cleaning and feature extraction, and a lightweight model is used to detect local strong convective signals and water vapor convergence characteristics to filter effective information. Then, the multi-source data fusion center receives this data and performs data fusion processing, involving timestamp alignment, spatial location matching, and synchronization of different sampling frequencies. Afterwards, fusion methods such as data layer, feature layer, or decision layer fusion are applied to integrate the information. Subsequently, the early warning and command center receives the fused data, loads and adjusts artificial intelligence models such as deep neural networks or random forests, and outputs operational risk early warning results through model analysis and inference. The diagnostic process is continuously monitored to evaluate the effectiveness of the method. By setting time intervals, indicators such as early warning accuracy, precision, recall, and F1 score are calculated to obtain a comprehensive risk detection score. Based on the score, the diagnostic method is adjusted, such as optimizing model parameters or improving data collection. Finally, the real-time weather conditions, forecast field, warning area level, and operational suggestions are displayed intuitively through a visual interactive terminal, which also provides interactive functions for command personnel and operational personnel. The whole process forms a closed loop to dynamically improve the accuracy of the warning.

[0073] Secondly: The accompanying drawings of the embodiments disclosed in this invention only involve the structures involved in the embodiments disclosed in this invention. Other structures can refer to the general design. In the absence of conflict, the same embodiment and different embodiments of this invention can be combined with each other.

[0074] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A dynamic early warning method for weather modification operations based on multi-source data fusion, characterized in that, include: S1: Multi-source meteorological monitoring data acquisition: Collect multi-source meteorological monitoring data through meteorological detection equipment. The multi-source meteorological monitoring data should cover multi-dimensional information, including meteorological field data, detection equipment status data, and operating environment parameters. S2: Data Preprocessing and Analysis: The regional data processing node receives data from meteorological detection equipment, performs preprocessing operations on the data, conducts preliminary analysis on the preprocessed data, and filters out effective information for transmission; S3: Multi-source meteorological monitoring data fusion: The multi-source data fusion center receives multi-source meteorological monitoring data from multiple regional data processing nodes and performs data fusion; The fusion of multi-source meteorological monitoring data specifically includes the following steps: A1: Receives data uploaded from data processing nodes in various regions and summarizes and organizes multi-source meteorological monitoring data of the same type; A2: Perform timestamp alignment and spatial location matching on multi-source meteorological monitoring data. By calibrating and synchronizing the timestamps, the multi-source meteorological monitoring data can be made consistent in both time and space. A3: Synchronize data from different sampling frequencies to ensure that data from different modalities are consistent in time resolution; A4: Select a fusion method to fuse the processed data; Data fusion methods include data-layer fusion, feature-layer fusion, and decision-layer fusion. Data-layer fusion specifically includes direct concatenation and weighted fusion. Direct concatenation involves directly concatenating raw or pre-processed data from different modalities. Weighted fusion assigns different weights to each modality based on its importance or reliability, then performs a weighted sum. Feature-layer fusion specifically includes feature selection and combination, and feature transformation and fusion. Feature selection and combination involves selecting representative features from different modalities and combining them into a new feature set. Feature transformation and fusion involves transforming the features of different modalities to achieve a common feature space or distribution before fusion. Decision-layer fusion specifically includes independent classification and result fusion, and probabilistic fusion. Independent classification and result fusion involves using independent classifiers or diagnostic models to perform operational risk warnings for different modalities, obtaining their respective diagnostic results, and then fusing these results. Probabilistic fusion involves fusing the risk probabilities of each independent diagnostic model to obtain a comprehensive risk probability. S4: Centralized Operational Risk Early Warning and Decision-Making: The early warning and command center receives the fused data, analyzes and reasons about the fused data through artificial intelligence models, and analyzes and locates potential operational risks based on the model output results; Operational risk warning and decision-making includes the following steps: B1: The early warning and command center receives the fused data transmitted from the multi-source data fusion center and temporarily stores it in the cache area; B2: Select and load the artificial intelligence model. After loading the model, adjust the model by pre-training it using historical risk data, based on the current operational risk warning objectives. B3: Input the temporarily cached data into the selected adapted artificial intelligence model, then perform model analysis and reasoning and output the analysis results; S5: Continuous monitoring and evaluation: Continuously monitor the risk analysis results during the potential operational risk diagnosis process, evaluate the effectiveness of the potential operational risk diagnosis methods, and adjust the diagnosis methods based on the evaluation results; The specific steps for evaluating the effectiveness of potential operational risk diagnostic methods include: C1: Set a time interval, monitor and collect the true positive example X1, false positive example X2, true negative example X3, and false negative example X4 within the time interval, and collect n consecutive time intervals, labeled as 1, 2, ..., i, ..., n respectively. Calculate the average true positive example, average false positive example, average true negative example, and average false negative example using the true positive example X1i, false positive example X2i, true negative example X3i, and false negative example X4i of each time interval. True positive example X1 represents the number of high-risk weather events that were correctly warned, false positive example X2 represents the number of false warnings, true negative example X3 represents the number of events that were correctly determined to be unsuitable for operation, and false negative example X4 represents the number of events that should have been warned but were not. C2: Divide the sum of the average true positives and average true negatives by the sum of the average true positives, average false positives, average true negatives, and average false negatives using the formula. To obtain the early warning accuracy ηA, divide the average true positive rate by the sum of the average true positive rate and the average false positive rate using the formula... To obtain the early warning accuracy ηP, divide the average true cases by the sum of the average true cases and the average false cases using the formula... The warning recall rate ηR is obtained, and the product of the warning precision rate and the warning recall rate is divided by the sum of the warning precision rate and the warning recall rate and multiplied by 2. The F1 value is obtained by using the formula F1=2×ηP×ηR / (ηP+ηR). C3: The risk detection accuracy, early warning precision, early warning recall, and F1 score are combined using the formula Score=ω1×ηA+ω2×ηP+ω3×ηR+ω4×F1 to obtain the comprehensive risk detection score, Q, where ω1, ω2, ω3, and ω4 represent the weighting coefficients of risk detection accuracy, early warning precision, early warning recall, and F1 score, respectively, and their sum is 1.

2. The method for dynamic early warning of weather modification operations based on multi-source data fusion according to claim 1, characterized in that: Multi-source meteorological monitoring data includes meteorological field data, detection equipment status data, and operational environment parameters. Specifically, meteorological field data includes wind field, temperature and humidity field, air pressure field, and water vapor flux. Detection equipment status data includes radar status, radiosonde status, and ground station status. Operational environment parameters include airspace status and terrain influence.

3. The method for dynamic early warning of weather modification operations based on multi-source data fusion according to claim 1, characterized in that: Preprocessing specifically refers to cleaning and extracting features from the collected multi-source meteorological monitoring data through regional data processing nodes. Preliminary analysis includes detecting local strong convection initiation signals and water vapor convergence characteristics using lightweight models and uploading them to the multi-source data fusion center.

4. The method for dynamic early warning of weather modification operations based on multi-source data fusion according to claim 1, characterized in that: The accuracy rate of risk warning reflects the proportion of correctly diagnosed samples out of the total number of samples; the precision rate reflects the proportion of samples diagnosed as risk that are actually risky; the recall rate reflects the proportion of actual risk samples that are correctly diagnosed; and the F1 score reflects the difference between the precision rate and the recall rate.

Images