Dynamic interpretable evaluation method, device and equipment for multi-source data fusion
By using multi-source data fusion and dynamic interpretable assessment methods, the problems of subjectivity and data integration in civil aviation safety assessments have been solved, enabling real-time and accurate risk assessments and improving the safety and efficiency of civil aviation operations.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- CIVIL AVIATION FLIGHT UNIV OF CHINA
- Filing Date
- 2025-12-17
- Publication Date
- 2026-05-08
AI Technical Summary
Existing civil aviation safety assessment methods rely heavily on expert experience, which is subjective and inconsistent. They are difficult to fully capture influencing factors, and multi-source heterogeneous data cannot be effectively integrated, resulting in insufficient accuracy and reliability of assessment results. They also lack environmental adaptation mechanisms and are unable to respond to dynamic changes in real time.
A dynamic and interpretable assessment method based on multi-source data fusion is adopted. Through multi-source data acquisition, federated feature alignment, dynamic assessment model and interpretability module, real-time radar surveillance data, ADS-B trajectory data, meteorological information and flight plan data are used. Combined with federated adversarial network and bidirectional gated cyclic unit BIGRU model, data feature alignment and dynamic weight adjustment are achieved to generate a visualized risk assessment map.
It has improved the utilization rate of multi-source data, captured spatiotemporal feature changes in real time, significantly shortened the early warning response time, enhanced the safety and timeliness of civil aviation operations, provided transparent and credible assessment results, and met the transparency requirements of the Civil Aviation Administration's safety audit.
Smart Images
Figure CN121998404A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of civil aviation operation safety factor assessment technology, and provides a dynamic interpretable assessment method, apparatus and equipment for multi-source data fusion. Background Technology
[0002] As is well known, accurate and reliable civil aviation safety assessment technology plays an irreplaceable and crucial role in preventing risks, ensuring flight safety, and improving overall operational efficiency. It permeates every aspect of airport operations, flight scheduling, and the flight process, serving as a vital cornerstone for ensuring the safe travel of hundreds of millions of passengers. However, existing civil aviation safety assessment methods have the following shortcomings: (1) It relies heavily on expert experience, and the allocation of factor weights is mainly achieved by manually constructing a judgment matrix. Different expert groups have a high degree of subjectivity and inconsistency when assessing key indicators such as "runway incursion risk". Moreover, due to the limitations of expert knowledge, it is difficult to fully capture all influencing factors.
[0003] (2) A large amount of data is generated during civil aviation operations, such as ACARS messages, QAR flight data, and weather radar information. However, many airports suffer from data silos, and these key multi-source heterogeneous data cannot be effectively integrated and shared. This leads to the dilemma of missing input dimensions in the construction of evaluation models, affecting data utilization, the comprehensiveness and accuracy of evaluation results.
[0004] (3) Existing civil aviation safety assessment systems mostly adopt a fixed-cycle update mechanism and lack an environmental adaptation mechanism, making it difficult to respond in a timely manner to dynamic changes in civil aviation operations, such as seasonal flight changes and extreme weather scenarios. In this case, as environmental changes and scenario complexity increase, the accuracy of the assessment model will gradually decrease, and the false alarm rate will increase accordingly, greatly reducing the effectiveness and reliability of the assessment.
[0005] Therefore, improving the effectiveness, accuracy, and reliability of civil aviation safety assessments has become an urgent problem to be solved. Summary of the Invention
[0006] This application provides a dynamic interpretable evaluation method, apparatus, and device for multi-source data fusion, which addresses the technical problems of poor effectiveness, accuracy, and reliability in existing technologies.
[0007] On the one hand, a dynamic and interpretable evaluation method for multi-source data fusion is provided, the method comprising: A multi-source data acquisition module is used to acquire data and obtain multi-source target data; wherein, the multi-source target data includes real-time radar surveillance data, ADS-B trajectory data, meteorological information, flight plan (FPL) and airport operation data; The target multi-source data is feature aligned using a federated feature alignment module to obtain multi-source aligned data; wherein, the federated feature alignment module is equipped with a feature alignment algorithm based on a federated adversarial network; The multi-source aligned data is input into the dynamic evaluation model module to obtain a dynamic weight vector; wherein, the dynamic evaluation model in the dynamic evaluation model module is a bidirectional gated recurrent unit (BIGRU) with an attention mechanism. Based on the dynamic weight vector, an interpretable module is used to perform spatiotemporal risk evolution and obtain a visualized risk assessment map.
[0008] Optionally, the step of acquiring target multi-source data using a multi-source data acquisition module includes: A multi-source data acquisition module is used to acquire initial multi-source data. The initial multi-source data is preprocessed to obtain the target multi-source data; wherein, the preprocessing includes outlier detection and missing value imputation.
[0009] Optionally, the step of using a federated feature alignment module to perform feature alignment on the target multi-source data to obtain multi-source aligned data includes: The target multi-source data is feature-aligned using the target optimal transfer function in the federated feature alignment module to obtain the multi-source aligned data; wherein the target optimal transfer function is expressed by the following formula:
[0010] in, To find the minimum value of the function G, ; To combat the losses; and These represent the feature distributions of the source and target domains, respectively. The Wasserstein distance between the source and target domains; This is a coefficient used to weigh adversarial loss against distributional difference loss.
[0011] Optionally, the step of inputting the multi-source alignment data into the dynamic evaluation model module to obtain the dynamic weight vector includes: The BiGRU-Attention layer of the dynamic evaluation model is used to extract spatiotemporal features from the multi-source aligned data to obtain multiple spatiotemporal features; The multi-head attention unit of the preset dynamic evaluation model is used to process the multiple spatiotemporal features to obtain the original weight vector; The dynamic weight vector is determined by using the original weight vector, flow density, and a preset weight adjustment function.
[0012] Optionally, the preset weight adjustment function is expressed by the following formula:
[0013] in, This is a preset weight adjustment function; The original weights of the i-th feature; is the dynamic weight of the i-th feature; n is the total number of features.
[0014] Optionally, the step of obtaining a visualized risk assessment map by performing spatiotemporal risk evolution using an interpretability module based on a dynamic weight vector includes: An interpretability module is used to quantify the feature contribution of the multiple spatiotemporal features using game theory, thereby obtaining the contribution value of each spatiotemporal feature; Based on the dynamic weight vector, the contribution values of each spatiotemporal feature, and the preset risk value function, the risk value of the airspace region within the target time period is determined. Spatiotemporal risk evolution is performed on the risk values of airspace regions within the target time period to obtain the visualized risk assessment map.
[0015] Optionally, the preset risk value function is expressed by the following formula:
[0016] in, This is expressed as the risk value at position (x, y) at time t; Let i be the dynamic weight of the i-th spatiotemporal feature. Let be the contribution value of the i-th spatiotemporal feature at position (x, y) and time t.
[0017] On the one hand, a dynamic and interpretable evaluation device for multi-source data fusion is provided, the device comprising: The multi-source data acquisition unit is used to acquire data using the multi-source data acquisition module to obtain target multi-source data; wherein, the target multi-source data includes real-time radar surveillance data, ADS-B trajectory data, meteorological information, flight plan (FPL), and airport operation data; A federated feature alignment unit is used to perform feature alignment on the target multi-source data using a federated feature alignment module to obtain multi-source aligned data; wherein, the federated feature alignment module is equipped with a feature alignment algorithm based on a federated adversarial network; The dynamic evaluation unit is used to input the multi-source aligned data into the dynamic evaluation model module to obtain a dynamic weight vector; wherein, the dynamic evaluation model in the dynamic evaluation model module is a bidirectional gated recurrent unit (BIGRU) with an attention mechanism. The interpretability unit is used to perform spatiotemporal risk evolution based on the dynamic weight vector using interpretable modules to obtain a visualized risk assessment map.
[0018] On one hand, an electronic device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement any of the methods described above.
[0019] On the one hand, a storage medium is provided that stores computer program instructions thereon, which, when executed by a processor, implement any of the methods described above.
[0020] Compared with the prior art, the beneficial effects of this application are as follows: In this application, when conducting element assessment of civil aviation operation safety, firstly, a multi-source data acquisition module can be used to acquire target multi-source data; this target multi-source data includes real-time radar surveillance data, ADS-B trajectory data, meteorological information, flight plan (FPL), and airport operation data; then, a federated feature alignment module can be used to align the target multi-source data to obtain multi-source aligned data; this federated feature alignment module is equipped with a feature alignment algorithm based on a federated adversarial network; next, the multi-source aligned data can be input into a dynamic evaluation model module to obtain a dynamic weight vector; the dynamic evaluation model in the dynamic evaluation model module is a bidirectional gated recurrent unit (BIGRU) with an attention mechanism; finally, based on the dynamic weight vector, an interpretability module can be used to perform spatiotemporal risk evolution to obtain a visualized risk assessment map.
[0021] Based on this, this application employs a pre-defined feature alignment algorithm based on a federated adversarial network to align the features of target multi-source data. Therefore, this application effectively overcomes the limitations of existing technologies in parsing unstructured data such as image-based aircraft maintenance reports and time-series ADS-B data, significantly improving the utilization rate of multi-source data and providing a more comprehensive and accurate data foundation for subsequent evaluation and analysis. Furthermore, since the dynamic evaluation model in the dynamic evaluation model module is a bidirectional gated recurrent unit (BIGRU) with an attention mechanism, this application can capture spatiotemporal feature changes in real time. Especially in extreme weather scenarios, it can significantly shorten the early warning response time, enhance the safety and timeliness of civil aviation operations, and improve the effectiveness and reliability of the evaluation. Attached Figure Description
[0022] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the drawings described below are only embodiments of this application. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.
[0023] Figure 1 This is a schematic diagram of an application scenario provided by an embodiment of this application; Figure 2 A schematic diagram of a dynamic interpretable evaluation method for multi-source data fusion provided in an embodiment of this application; Figure 3 This is a schematic diagram of a dynamic interpretable evaluation device for multi-source data fusion provided in an embodiment of this application.
[0024] The diagram is labeled as follows: 10-Dynamic interpretable evaluation device for multi-source data fusion, 101-Processor, 102-Memory, 103-I / O interface, 104-Database, 30-Dynamic interpretable evaluation device for multi-source data fusion, 301-Multi-source data acquisition unit, 302-Federated feature alignment unit, 303-Dynamic evaluation unit, 304-Interpretability unit. Detailed Implementation
[0025] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application. Unless otherwise specified, the embodiments and features in the embodiments of this application can be arbitrarily combined with each other. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than that shown here.
[0026] As is well known, accurate and reliable civil aviation safety assessment technology plays an irreplaceable and crucial role in preventing risks, ensuring flight safety, and improving overall operational efficiency. It permeates every aspect of airport operations, flight scheduling, and the flight process, serving as a vital cornerstone for ensuring the safe travel of hundreds of millions of passengers. However, existing civil aviation safety assessment methods have the following shortcomings: (1) It relies heavily on expert experience, and the allocation of factor weights is mainly achieved by manually constructing a judgment matrix. Different expert groups have a high degree of subjectivity and inconsistency when assessing key indicators such as "runway incursion risk". Moreover, due to the limitations of expert knowledge, it is difficult to fully capture all influencing factors.
[0027] (3) A large amount of data is generated during civil aviation operations, such as ACARS messages, QAR flight data, and weather radar information. However, many airports have data silos, and these key multi-source heterogeneous data cannot be effectively integrated and shared, which leads to the dilemma of missing input dimensions in the construction of evaluation models, affecting data utilization, the comprehensiveness and accuracy of evaluation results.
[0028] (3) Existing civil aviation safety assessment systems mostly adopt a fixed-cycle update mechanism and lack an environmental adaptation mechanism, making it difficult to respond in a timely manner to dynamic changes in civil aviation operations, such as seasonal flight changes and extreme weather scenarios. In this case, as environmental changes and scenario complexity increase, the accuracy of the assessment model will gradually decrease, and the false alarm rate will increase accordingly, greatly reducing the effectiveness and reliability of the assessment.
[0029] Based on this, this application provides a dynamic and interpretable assessment method for multi-source data fusion. In this method, firstly, a multi-source data acquisition module is used to acquire target multi-source data, including real-time radar surveillance data, ADS-B trajectory data, meteorological information, flight plan (FPL), and airport operation data. Then, a federated feature alignment module is used to align the target multi-source data to obtain multi-source aligned data. This module incorporates a feature alignment algorithm based on a federated adversarial network. Next, the multi-source aligned data is input into a dynamic assessment model module to obtain a dynamic weight vector. The dynamic assessment model in this module is a bidirectional gated recurrent unit (BIGRU) with an attention mechanism. Finally, based on the dynamic weight vector, an interpretability module is used to perform spatiotemporal risk evolution to obtain a visualized risk assessment map. Based on this, this application employs a pre-defined feature alignment algorithm based on a federated adversarial network to align the features of target multi-source data. Therefore, this application effectively overcomes the limitations of existing technologies in parsing unstructured data such as image-based aircraft maintenance reports and time-series ADS-B data, significantly improving the utilization rate of multi-source data and providing a more comprehensive and accurate data foundation for subsequent evaluation and analysis. Furthermore, since the dynamic evaluation model in the dynamic evaluation model module is a bidirectional gated recurrent unit (BIGRU) with an attention mechanism, this application can capture spatiotemporal feature changes in real time. Especially in extreme weather scenarios, it can significantly shorten the early warning response time, enhance the safety and timeliness of civil aviation operations, and improve the effectiveness and reliability of the evaluation.
[0030] After introducing the design concept of the embodiments of this application, the following is a brief introduction to the application scenarios to which the technical solutions of the embodiments of this application can be applied. It should be noted that the application scenarios described below are only for illustrating the embodiments of this application and are not intended to limit the scope. In specific implementation, the technical solutions provided by the embodiments of this application can be flexibly applied according to actual needs.
[0031] like Figure 1 The diagram shown illustrates an application scenario provided by an embodiment of this application. This application scenario may include a dynamically interpretable evaluation device 10 for multi-source data fusion.
[0032] The dynamic interpretable evaluation device 10, which integrates multi-source data fusion, can perform dynamic interpretable evaluations of civil aviation safety based on multi-source data fusion. For example, it can be an in-vehicle computer, a personal computer (PC), a server, or a laptop. The dynamic interpretable evaluation device 10 may include one or more processors 101, memory 102, I / O interfaces 103, and a database 104. Specifically, the processor 101 can be a central processing unit (CPU) or a digital processing unit, etc. The memory 102 can be volatile memory, such as random-access memory (RAM); it can also be non-volatile memory, such as read-only memory, flash memory, hard disk drive (HDD), or solid-state drive (SSD); or it can be any other medium capable of carrying or storing desired program code in the form of instructions or data structures that can be accessed by a computer, but is not limited thereto. The memory 102 may be a combination of the aforementioned memories. The memory 102 may store some program instructions of the dynamic interpretable evaluation method for multi-source data fusion provided in this application embodiment. When these program instructions are executed by the processor 101, they can be used to implement the steps of the dynamic interpretable evaluation method for multi-source data fusion provided in this application embodiment, thereby solving the technical problem of poor effectiveness, accuracy, and reliability in the prior art. The database 104 may be used to store target multi-source data, preset feature alignment algorithms, multi-source alignment data, preset dynamic evaluation models, and visualized risk assessment maps involved in the solution provided in this application embodiment.
[0033] In this embodiment, the dynamic interpretable evaluation device 10 for multi-source data fusion can obtain security evaluation instructions through the I / O interface 103. Then, the processor 101 of the dynamic interpretable evaluation device 10 for multi-source data fusion will solve the technical problems of poor effectiveness, accuracy, and reliability of the prior art according to the program instructions of the dynamic interpretable evaluation method for multi-source data fusion provided in this embodiment of the application stored in the memory 102. In addition, target multi-source data, preset feature alignment algorithms, multi-source aligned data, preset dynamic evaluation models, and visualized risk assessment maps can be stored in the database 104.
[0034] Of course, the methods provided in the embodiments of this application are not limited to... Figure 1 The application scenarios shown can also be used in other possible scenarios, and this application embodiment does not impose any limitations. Figure 1 The functions that the various devices in the application scenarios shown can achieve will be described in subsequent method embodiments, and will not be elaborated on here. Below, the methods of the embodiments of this application will be described in conjunction with the accompanying drawings.
[0035] like Figure 2 The diagram shown is a flowchart illustrating a dynamic and interpretable evaluation method for multi-source data fusion provided in this application. This method can... Figure 1 The dynamic interpretable evaluation device 10 for multi-source data fusion is used to perform this process. Specifically, the process flow of this method is described below.
[0036] Step 201: Use a multi-source data acquisition module to acquire data and obtain target multi-source data.
[0037] The target multi-source data includes real-time radar surveillance data, ADS-B trajectory data, meteorological information, flight plans (FPL), airspace status and airport operation data, etc.
[0038] As can be seen, the multi-source data acquisition module of this application can be used to integrate heterogeneous data sources in civil aviation operations, thereby laying a data foundation for comprehensive and accurate assessment of civil aviation operation elements. Furthermore, multi-source data provides a full-dimensional data base for operational safety assessment, covering the four major elements of "human-machine-environment-control." Among these, radar surveillance data and ADS-B trajectory data can accurately reflect the real-time position and flight status of aircraft in the air, aiding in the monitoring of the "machine" element; meteorological intelligence can provide crucial evidence for flight environment assessment, linking to the "environment" element; flight plans (FPL) can clearly outline the expected flight trajectory and mission arrangements, involving the "control" level; and airport operation data covers information such as ground support and resource allocation, closely linked to the multiple elements of "human-machine-environment-control," providing comprehensive data support for collaborative decision-making, risk warning, and safety assessment by air traffic control departments, airlines, and airport operators, effectively improving the safety and efficiency of civil aviation operations.
[0039] In one possible implementation, in order to ensure that different types of data sources can be stably and efficiently accessed by the system, a distributed data interface adapter can be used in this application to solve the interface compatibility problem between different data sources. It supports seamless access of ICAO standard data formats (e.g., ASTERIX, AIXM, etc.) and real-time streaming data (e.g., Kafka).
[0040] In another possible implementation, in order to ensure the integrity and reliability of the input data and meet the high-quality data requirements of civil aviation operation element assessment, a data quality verification layer can also be embedded in the multi-source data acquisition module in this application.
[0041] Specifically, firstly, a multi-source data acquisition module can be used to collect initial multi-source data. Then, the initial multi-source data can be preprocessed to obtain the target multi-source data. Preprocessing includes techniques such as outlier detection (e.g., the Isolation Forest algorithm) and missing value imputation (e.g., spatiotemporal kriging interpolation). Based on this, strict quality control can be performed on the collected initial multi-source data, removing outliers and appropriately supplementing missing data.
[0042] Step 202: Use the federated feature alignment module to perform feature alignment on the target multi-source data to obtain multi-source aligned data.
[0043] The federated feature alignment module includes a feature alignment algorithm based on a federated adversarial network.
[0044] In this application, the federal feature alignment module mainly addresses the problem of data silos across regions and departments, realizes feature space alignment of multi-entity data under privacy protection, ensures that data from all parties can be effectively integrated without leaking privacy, and provides a unified and coordinated data feature foundation for subsequent civil aviation operation element assessment.
[0045] Specifically, this application proposes an improved Optimal Transport (OT) algorithm, which minimizes inter-domain distribution differences by introducing a federated adversarial network with Wasserstein distance constraints to achieve feature alignment. That is, the target optimal transport function in the federated feature alignment module can be used to align the target multi-source data to obtain multi-source aligned data; the target optimal transport function can be expressed by the following formula:
[0046] in, To find the minimum value of the function G, ; To combat the losses; and These represent the feature distributions of the source and target domains, respectively. The Wasserstein distance between the source and target domains; These are the coefficients used to weigh adversarial loss and distributional difference loss. Furthermore, based on this formula, it is possible to minimize the adversarial loss... Wasserstein distance between the source domain and the target domain This makes the feature distributions of different domains as close as possible, thereby achieving alignment of the feature space.
[0047] As can be seen, this application, through the federal feature alignment module, breaks down regional and departmental barriers to civil aviation data, enabling feature fusion and collaboration among multiple entities while protecting data privacy. This provides more comprehensive and accurate data support for the assessment of civil aviation operational elements. This helps improve the accuracy and reliability of assessments, optimize flight path planning and airspace resource allocation, thereby enhancing the operational efficiency and safety of the entire civil aviation system and promoting collaborative cooperation among different regions and departments in civil aviation safety management.
[0048] In one possible implementation, to improve the quality of feature alignment and the accuracy of civil aviation operation element assessment, this application can construct a domain knowledge graph covering key civil aviation knowledge such as navigation rules and airspace topology. Graph Attention (GAT) is then used to weight and fuse the features of nodes and their neighboring nodes during feature mapping, highlighting key features and making the feature mapping results more consistent with the actual logic of civil aviation operations. For example, based on the airspace topology, attention weights for different airspace nodes can be reasonably allocated to ensure that the feature mapping accurately reflects the correlation between airspaces and the logic of flight paths, thereby improving the quality of feature alignment and the accuracy of civil aviation operation element assessment.
[0049] Step 203: Input the multi-source aligned data into the dynamic evaluation model module to obtain the dynamic weight vector.
[0050] Among them, the dynamic evaluation model in the dynamic evaluation model module is a bidirectional gated recurrent unit (BIGRU) with an attention mechanism.
[0051] In this application, the dynamic assessment model module can provide accurate and dynamic quantitative basis for air traffic control decision-making by assessing the airspace operation risk level, capacity utilization rate and control load index in real time, thereby ensuring the safety and efficiency of airspace operation.
[0052] Specifically, firstly, a BiGRU-Attention layer in a dynamically evaluated model can be used to extract spatiotemporal features from multi-source aligned data to obtain multiple spatiotemporal features. The BiGRU-Attention layer can capture the temporal dependencies of control commands and the spatial correlation of the airspace situation. The BiGRU structure includes a forward GRU and a backward GRU; the forward GRU processes the input data from front to back according to the time series. Capture the state information at the current time t The reverse GRU processes input data from back to front according to the time sequence. Capture the state information at the current time t Finally, by concatenating the forward and reverse state information, the hidden state of the BiGRU is obtained as shown in the following formula. (i.e., spatiotemporal characteristics):
[0053] Thus, through this BiGRU structure, the temporal dependencies of control commands and the spatial correlations of airspace situation can be fully explored, and the spatiotemporal characteristics of airspace operation status can be accurately grasped.
[0054] Then, a multi-head attention mechanism can be used to focus on key conflict points (e.g., convergence waypoints, weather disturbance areas, etc.). That is, a pre-defined dynamic evaluation model's multi-head attention unit can be used to process multiple spatiotemporal features to obtain the original weight vector. In this application, the output of BiGRU can be... The query matrix Q, key matrix K, and value matrix V are generated through linear transformation. The attention mechanism can then be implemented by calculating the relationship between these three matrices, which can be represented by the following formula:
[0055] in, is the dimension of the key vector.
[0056] Based on this, a multi-head attention mechanism can be used to calculate multiple spatiotemporal features separately through multiple attention heads. Then, the calculation results are concatenated and linearly transformed to obtain the final output (i.e., the original weight vector) as shown in the following formula:
[0057]
[0058] Among them, the i-th attention head , , and Let be the parameter matrix of the i-th attention head. Therefore, this application can capture features of different dimensions through this multi-head attention mechanism, making the model more accurate in identifying key conflict points, avoiding information omissions, and thus effectively improving the accuracy of the evaluation.
[0059] Finally, to make the assessment results more consistent with actual airspace operation, this application designs a "dynamic weighting mechanism." This mechanism dynamically allocates the weights of the assessment indicators based on changes in real-time traffic density by designing a reasonable weight adjustment function. Specifically, the dynamic weighting vector can be determined using the original weight vector, traffic density, and a preset weight adjustment function.
[0060] Specifically, we can assume the flow density is The original weight vector of the evaluation index is The preset weight adjustment function is expressed by the following formula:
[0061] in, This is a preset weight adjustment function; The original weights of the i-th feature; Let be the dynamic weight of the i-th feature; n be the total number of features. This formula can be based on flow density. Changes to the original weights Adjustments are made to allow the weights of assessment indicators to change adaptively when traffic density varies, making the assessment results more consistent with the actual airspace operation and providing more targeted and timely support for control decisions.
[0062] Therefore, the dynamic weight vector can be calculated based on the aforementioned preset weight adjustment function formula. .
[0063] As can be seen, the dynamic assessment model module of this application, with its advanced spatiotemporal feature extraction technology and flexible dynamic weighting mechanism, can accurately assess airspace operational risk levels, capacity utilization, and control load indices in real time. This provides controllers with real-time and accurate quantitative assessment results, enabling them to quickly grasp the airspace operational status, identify potential risks in advance, formulate scientific and reasonable control strategies, optimize airspace resource allocation, improve airspace capacity utilization, and reduce control load. This effectively enhances the safety and efficiency of civil aviation operations, ensures smooth flight operations, and improves the passenger travel experience, playing a significant role in promoting the optimization and upgrading of the entire civil aviation operation system.
[0064] Step 204: Based on the dynamic weight vector, use the interpretability module to perform spatiotemporal risk evolution and obtain a visualized risk assessment map.
[0065] In this application, the interpretability module can reveal the decision-making basis of the assessment results, meet the transparency requirements of the Civil Aviation Administration of China (CAAC) safety audit, provide air traffic control departments with traceable and understandable assessment logic, and enhance the credibility and authority of the assessment results.
[0066] Specifically, firstly, the interpretability module can be used to quantify the feature contributions of multiple spatiotemporal features using game theory to obtain the contribution value of each spatiotemporal feature. That is, the interpretability module can quantify feature contributions based on game theory and use the SHAP value (derived from the Shapley value in game theory) to measure the contribution of each feature to the prediction result.
[0067] In this application, for a prediction result Its feature contribution metric can be represented by the following function:
[0068] in, As the base value, Let be the contribution value of the i-th feature. This formula represents the prediction result as the sum of the base value (the prediction value when all features are absent) and the contribution values of each feature.
[0069] As can be seen, by calculating the SHAP value of each feature, this application can quantify the specific impact of spatiotemporal features (i.e., key risk factors, which may be meteorological conditions, flight traffic, airspace complexity, etc.) on the assessment results such as airspace operation risk level, capacity utilization rate, and control load index, thereby identifying key risk factors and providing accurate risk insights for air traffic control decisions.
[0070] Then, the risk value of the airspace region within the target time period can be determined based on the dynamic weight vector, the contribution values of each spatiotemporal feature, and the preset risk value function. In this application, the preset risk value function can be expressed by the following formula:
[0071] in, This is expressed as the risk value at position (x, y) at time t; Let i be the dynamic weight of the i-th spatiotemporal feature. Let be the contribution value of the i-th spatiotemporal feature at position (x, y) and time t.
[0072] Based on this, a visualized risk assessment map (3D heatmap) can be obtained by analyzing the spatiotemporal risk evolution of risk values in airspace areas within a target time period. That is, the spatiotemporal risk evolution path can be displayed using a 3D heatmap, supporting retrospective analysis of control plans. In this application, the 3D heatmap uses time as the z-axis and airspace area as the xy-plane, with color intensity representing the degree of risk.
[0073] As can be seen, the interpretability module of this application provides a transparent and credible basis for decision-making in the assessment of civil aviation operational elements through aviation compliance SHAP and traceability visualization technology. On the one hand, it meets the Civil Aviation Administration's requirements for transparency in safety audits, ensuring that the assessment process and results are verifiable and traceable. On the other hand, it helps air traffic control departments to deeply understand the logic and risk factors behind the assessment results, improving the scientific nature and accuracy of air traffic control decisions. Simultaneously, through visualization, it facilitates communication and collaboration among different professionals (such as controllers, safety managers, and technicians), enhancing the safety and efficiency of airspace operations, promoting the overall optimization of the civil aviation operation system, and is of great significance for improving the safety level and management efficiency of civil aviation operations.
[0074] In one possible implementation, in order to improve the scientific nature and timeliness of control decisions and ensure the safety and smooth operation of airspace, this application also includes a "real-time feedback mechanism module". This module can transform the assessment results into actionable control recommendations, forming a "assessment-decision-execution" closed loop, thereby improving the scientific nature and timeliness of control decisions and ensuring the safety and smooth operation of airspace.
[0075] Specifically, a tiered alert system can be used to trigger differentiated responses based on risk levels. That is, the airspace operational risk level (i.e., the predicted risk level) can be configured. The risk level is divided into three levels: low, medium, and high. For low risk, only routine monitoring and periodic updates to the risk status are required; for medium risk, controllers should be notified promptly, and preventative measures should be recommended; for high risk, an emergency alarm should be triggered immediately, and specific emergency control strategies should be provided. The core formula is as follows:
[0076] in, For high-level risk thresholds, This is the threshold for the rate of change of risk at a high level. The formula indicates that "when the assessed risk value R exceeds a set threshold T, and the rate of change of risk ΔR reaches a certain level, an alarm of the corresponding level is triggered." Furthermore, this tiered mechanism ensures that controllers can respond quickly to different risk conditions, improving the safety of airspace operations.
[0077] In addition, to ensure flight safety and efficiency, this application can use a digital twin sandbox to build an airspace operation mirror based on Unity3D to simulate the flow evolution effect after the adjustment of control strategies.
[0078] Furthermore, by constructing a digital twin model of the airspace, data such as the geographical information, airspace structure, and flight traffic of the actual airspace are input into Unity3D to create a virtual environment highly consistent with the real airspace. Within this environment, flight traffic is simulated and predicted based on current control policies and assessment results. The traffic prediction formula is as follows:
[0079] in, For predicted future traffic, For current traffic, The flow rate variation coefficient, This is the estimated rate of change in traffic based on historical data and current trends.
[0080] It is evident that by simulating traffic flow evolution under different control strategies, controllers can assess the effectiveness of these strategies in advance, optimize control decisions, improve airspace capacity utilization, and reduce potential conflict risks. This simulation not only provides controllers with intuitive decision support but also helps them make more accurate judgments in complex airspace environments, ensuring flight safety and efficiency.
[0081] In summary, this application has the following advantages: (1) A cross-domain data fusion mechanism based on federated feature alignment is proposed. With the help of the collaborative optimization of the domain adaptation network, the traditional system can effectively overcome the parsing bottleneck of unstructured data such as image-type aircraft maintenance reports and time-series ADS-B data, and achieve a significant improvement in the utilization rate of multi-source data, providing a more comprehensive and accurate data foundation for subsequent evaluation and analysis.
[0082] (2) A dynamic evaluation model is proposed, which innovatively adopts a hybrid architecture of bidirectional gated recurrent unit (BiGRU) and attention mechanism. This architecture can capture spatiotemporal feature changes in real time, and can significantly shorten the early warning response time, especially in extreme weather scenarios, thereby enhancing the safety and timeliness of civil aviation operations.
[0083] (3) In the interpretability module, traceability visualization technology that meets airworthiness standards is introduced. This technology can output a high-confidence assessment report, effectively solving the problem of subjective bias in the traditional expert scoring method, making the assessment results more objective, credible and persuasive.
[0084] Based on the same inventive concept, embodiments of this application provide a dynamic and interpretable evaluation device 30 for multi-source data fusion, such as... Figure 3 As shown, the dynamic interpretable evaluation device 30 for multi-source data fusion includes: The multi-source data acquisition unit 301 is used to acquire data using the multi-source data acquisition module to obtain target multi-source data; wherein, the target multi-source data includes real-time radar surveillance data, ADS-B trajectory data, meteorological information, flight plan (FPL) and airport operation data; The federated feature alignment unit 302 is used to perform feature alignment on the target multi-source data using the federated feature alignment module to obtain multi-source aligned data; wherein, the federated feature alignment module is equipped with a feature alignment algorithm based on a federated adversarial network; The dynamic evaluation unit 303 is used to input multi-source aligned data into the dynamic evaluation model module to obtain a dynamic weight vector; wherein, the dynamic evaluation model in the dynamic evaluation model module is a bidirectional gated recurrent unit (BIGRU) with an attention mechanism. Interpretability unit 304 is used to perform spatiotemporal risk evolution based on dynamic weight vectors using interpretable modules to obtain a visualized risk assessment map.
[0085] Optionally, the multi-source data acquisition unit 301 is also used for: A multi-source data acquisition module is used to acquire initial multi-source data. The initial multi-source data is preprocessed to obtain the target multi-source data; the preprocessing includes outlier detection and missing value imputation.
[0086] Optionally, the federated feature alignment unit 302 is also used for: The target optimal transfer function in the federated feature alignment module is used to perform feature alignment on the target multi-source data to obtain multi-source aligned data; the target optimal transfer function is expressed by the following formula:
[0087] in, To combat the losses; and These represent the feature distributions of the source and target domains, respectively. The Wasserstein distance between the source and target domains; This is a coefficient used to weigh adversarial loss against distributional difference loss.
[0088] Optionally, the dynamic evaluation unit 303 is also used for: A BiGRU-Attention layer of a dynamic evaluation model is used to extract spatiotemporal features from multi-source aligned data, thereby obtaining multiple spatiotemporal features. A multi-head attention unit of a pre-defined dynamic evaluation model is used to process multiple spatiotemporal features to obtain the original weight vector; The dynamic weight vector is determined by using the original weight vector, flow density, and a preset weight adjustment function.
[0089] Optionally, interpretability unit 304 is also used for: An interpretability module is used to measure the feature contribution of multiple spatiotemporal features using game theory, and the contribution value of each spatiotemporal feature is obtained. Based on the dynamic weight vector, the contribution values of various spatiotemporal features, and the preset risk value function, the risk value of the airspace region within the target time period is determined. Spatiotemporal risk evolution of risk values in airspace regions within the target time period is performed to obtain a visualized risk assessment map.
[0090] The dynamic interpretable evaluation device 30 for multi-source data fusion can be used to perform... Figure 2 The method performed by the dynamic interpretable evaluation apparatus for multi-source data fusion in the illustrated embodiment can be referenced for understanding the functions that each functional module of the dynamic interpretable evaluation apparatus 30 for multi-source data fusion can achieve. Figure 2 The embodiments shown are described in detail below.
[0091] In some possible implementations, various aspects of the methods provided in this application can also be implemented as a program product comprising program code that, when run on a computer device, causes the computer device to perform the steps of the methods according to the various exemplary embodiments of this application described above. For example, the computer device may perform actions such as... Figure 2 The method performed by the dynamic interpretable evaluation device for multi-source data fusion in the illustrated embodiment.
[0092] Those skilled in the art will understand that all or part of the steps of the above method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it performs the steps of the above method embodiments. The aforementioned storage medium includes various media capable of storing program code, such as mobile storage devices, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks. Alternatively, if the integrated units of this application are implemented as software functional modules and sold or used as independent products, they can also be stored in a computer-readable storage medium. Based on this understanding, the technical solutions of the embodiments of this application, or the parts that contribute to the prior art, can be embodied in the form of software products. These computer software products are stored in a storage medium and include several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as mobile storage devices, ROM, RAM, magnetic disks, or optical disks.
[0093] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.
[0094] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.
Claims
1. A dynamic and interpretable evaluation method for multi-source data fusion, characterized in that, The method includes: A multi-source data acquisition module is used to acquire data and obtain multi-source target data; wherein, the multi-source target data includes real-time radar surveillance data, ADS-B trajectory data, meteorological information, flight plan (FPL) and airport operation data; The target multi-source data is feature aligned using a federated feature alignment module to obtain multi-source aligned data; wherein, the federated feature alignment module is equipped with a feature alignment algorithm based on a federated adversarial network; The multi-source aligned data is input into the dynamic evaluation model module to obtain a dynamic weight vector; wherein, the dynamic evaluation model in the dynamic evaluation model module is a bidirectional gated recurrent unit (BIGRU) with an attention mechanism. Based on the dynamic weight vector, an interpretable module is used to perform spatiotemporal risk evolution and obtain a visualized risk assessment map.
2. The method as described in claim 1, characterized in that, The step of acquiring target multi-source data using a multi-source data acquisition module includes: A multi-source data acquisition module is used to acquire initial multi-source data. The initial multi-source data is preprocessed to obtain the target multi-source data; wherein, the preprocessing includes outlier detection and missing value imputation.
3. The method as described in claim 1, characterized in that, The step of using a federated feature alignment module to perform feature alignment on the target multi-source data to obtain multi-source aligned data includes: The target multi-source data is feature-aligned using the target optimal transfer function in the federated feature alignment module to obtain the multi-source aligned data; wherein the target optimal transfer function is expressed by the following formula: in, To find the minimum value of the function G, ; To combat the losses; and These represent the feature distributions of the source and target domains, respectively. The Wasserstein distance between the source and target domains; This is a coefficient used to weigh adversarial loss against distributional difference loss.
4. The method as described in claim 1, characterized in that, The step of inputting the multi-source aligned data into the dynamic evaluation model module to obtain the dynamic weight vector includes: The BiGRU-Attention layer of the dynamic evaluation model is used to extract spatiotemporal features from the multi-source aligned data to obtain multiple spatiotemporal features; The multi-head attention unit of the preset dynamic evaluation model is used to process the multiple spatiotemporal features to obtain the original weight vector; The dynamic weight vector is determined by using the original weight vector, flow density, and a preset weight adjustment function.
5. The method as described in claim 4, characterized in that, The preset weight adjustment function is expressed by the following formula: in, This is a preset weight adjustment function; The original weights of the i-th feature; is the dynamic weight of the i-th feature; n is the total number of features.
6. The method as described in claim 4, characterized in that, The step of obtaining a visualized risk assessment map by performing spatiotemporal risk evolution using an interpretable module based on a dynamic weight vector includes: An interpretability module is used to quantify the feature contribution of the multiple spatiotemporal features using game theory, thereby obtaining the contribution value of each spatiotemporal feature; Based on the dynamic weight vector, the contribution values of each spatiotemporal feature, and the preset risk value function, the risk value of the airspace region within the target time period is determined. Spatiotemporal risk evolution is performed on the risk values of airspace regions within the target time period to obtain the visualized risk assessment map.
7. The method as described in claim 1, characterized in that, The preset risk value function is expressed by the following formula: in, This is expressed as the risk value at position (x, y) at time t; Let i be the dynamic weight of the i-th spatiotemporal feature. Let be the contribution value of the i-th spatiotemporal feature at position (x, y) and time t.
8. A dynamic and interpretable evaluation device for multi-source data fusion, characterized in that, The device includes: The multi-source data acquisition unit is used to acquire data using the multi-source data acquisition module to obtain target multi-source data; wherein, the target multi-source data includes real-time radar surveillance data, ADS-B trajectory data, meteorological information, flight plan (FPL), and airport operation data; A federated feature alignment unit is used to perform feature alignment on the target multi-source data using a federated feature alignment module to obtain multi-source aligned data; wherein, the federated feature alignment module is equipped with a feature alignment algorithm based on a federated adversarial network; The dynamic evaluation unit is used to input the multi-source aligned data into the dynamic evaluation model module to obtain a dynamic weight vector; wherein, the dynamic evaluation model in the dynamic evaluation model module is a bidirectional gated recurrent unit (BIGRU) with an attention mechanism. The interpretability unit is used to perform spatiotemporal risk evolution based on the dynamic weight vector using interpretable modules to obtain a visualized risk assessment map.
9. An electronic device, characterized in that, The device includes: Memory, used to store program instructions; A processor is configured to invoke program instructions stored in the memory and execute the method described in any one of claims 1-7 according to the obtained program instructions.
10. A storage medium, characterized in that, The storage medium stores computer-executable instructions for causing a computer to perform the method described in any one of claims 1-7.