A data analysis based unmanned aerial vehicle security screening system and method

Abstract

The application discloses a kind of unmanned plane safety screening system and method based on data analysis, comprising: by collecting unmanned plane space-time trajectory, equipment operation, airspace environment and communication interaction multidimensional data, key features are extracted and grouped by trajectory multi-feature clustering, utilize abnormal trajectory identification model to match benchmark library label abnormal fragment, realize the depth fusion of multiple data by multi-modal data collaborative analysis again, utilize security check data control and judge engine to complete abnormal correlation check and risk quantization, finally generate screening conclusion and control instruction.System includes six units of multi-source data acquisition transmission, trajectory feature extraction clustering, abnormal trajectory identification label, multi-modal data collaborative analysis, security risk judgment quantization and screening conclusion instruction output, form whole-process closed-loop processing mechanism, realize the comprehensive identification of unmanned plane security risk, accurate judgment and efficient control, improve the intelligent and standardized level of airspace safety management.

Description

Technical Field

[0001] This invention relates to the field of drone safety screening technology, and in particular to a drone safety screening system and method based on data analysis. Background Technology

[0002] With the rapid popularization of drone technology, its application in civilian, commercial, and special scenarios is becoming increasingly widespread. However, problems such as airspace safety conflicts, privacy leaks, and facility interference caused by disorderly drone flights are becoming increasingly prominent, posing a serious challenge to airspace management, public safety, and the protection of critical infrastructure. Currently, airspace control urgently needs real-time monitoring, risk identification, and rapid response to drone flight status. Traditional methods relying on manual inspections and single-device monitoring are no longer adequate for the flexibility and stealth characteristics of drone flights. There is an urgent need to build an intelligent, end-to-end safety screening system based on data analysis. By integrating multi-dimensional flight data with advanced algorithm models, this system can accurately identify and efficiently control drone flight safety risks, meeting the safety management requirements of different airspace scenarios.

[0003] Existing technologies for drone safety screening suffer from two significant drawbacks: First, traditional screening methods often rely on single-type data or single-algorithm models, focusing only on isolated analysis of trajectory data or equipment parameters. They fail to achieve deep collaborative integration of multimodal data, including spatiotemporal trajectory, equipment operation, airspace environment, and communication interaction, resulting in incomplete identification dimensions and insufficient accuracy in assessing safety risks in complex flight scenarios. Second, existing technologies lack a systematic data analysis engine and standardized risk assessment process. The extraction of trajectory anomaly features is not comprehensive enough, the quantification of risk levels lacks a unified parameter correlation logic, and a closed-loop processing mechanism covering the entire chain of "data acquisition - feature extraction - anomaly identification - risk assessment - command output" has not been formed, making it difficult to achieve rapid response and precise control of drone safety risks. Summary of the Invention

[0004] In order to overcome the shortcomings and deficiencies of existing technologies, this invention provides a data analysis-based unmanned aerial vehicle (UAV) safety screening system and method.

[0005] The technical solution adopted in this invention is a data analysis-based method for unmanned aerial vehicle (UAV) safety screening, comprising the following steps: S1, collecting spatiotemporal trajectory data, equipment operating parameter data, airspace environment data, and communication interaction data during UAV flight to establish a multi-dimensional raw data set; S2, calling a multi-feature clustering algorithm for UAV spatiotemporal trajectory to extract flight speed change features, heading angle features, altitude fluctuation features, and trajectory point density distribution features from the spatiotemporal trajectory data, and performing feature dimension mapping and clustering; S3, using an airspace abnormal trajectory identification model, performing feature matching and deviation calculation between the clustered trajectory features and a preset airspace safety trajectory benchmark library. S4. Calculate and mark potentially abnormal trajectory segments; S5. Activate the multimodal data collaborative intelligent analysis algorithm, integrating power system load data from equipment operating parameter data, navigation module error data, signal frequency band data from communication interaction data, and meteorological interference data from airspace environment data to construct a multimodal collaborative analysis matrix; S6. Utilize the UAV safety inspection data management and judgment engine to perform correlation verification on the marked abnormal trajectory segments and multimodal collaborative analysis results, and quantify and assign risk levels based on airspace safety level classification standards; S7. Based on the risk level quantification results, generate corresponding safety screening conclusions and management response instructions to complete the entire process of UAV safety screening.

[0006] Furthermore, the expression for the UAV spatiotemporal trajectory multi-feature clustering algorithm is as follows: ,in, The value of the cluster center for the k-th type of trajectory. Let be the velocity change at the i-th trajectory point. Let be the heading angle value of the i-th trajectory point. Let be the elevation fluctuation of the i-th trajectory point. Let be the neighborhood density value of the i-th trajectory point. These are the weight coefficients for each feature dimension. is the clustering convergence adjustment factor, and n is the total number of trajectory points.

[0007] Furthermore, the expression for the spatial anomaly trajectory recognition model is: Where A is the quantified value of trajectory anomaly. Let j be the value of the j-th feature parameter of the trajectory to be detected. Let j be the value of the standard feature parameter in the benchmark library. To correct the threshold for feature deviation, is the spatial safety factor, and m is the total number of characteristic parameters.

[0008] Furthermore, the expression for the multimodal data collaborative intelligent analysis algorithm is: Where M is the value of the multimodal collaborative analysis result, Let q be the parameter value of the p-th modal data. The weighting coefficients for the p-th modal data are... Let q be the confidence factor of the q-th parameter. The noise interference value of the p-th modal data. denoted as the calibration coefficient for the p-th modal data, and l is the number of parameters for each modal data.

[0009] Furthermore, the risk quantification expression of the drone security inspection data management and assessment engine is as follows: Where R is the quantified risk level, A is the quantified trajectory anomaly, and M is the result of the multimodal collaborative analysis. For the result weighting coefficient, This is the baseline threshold corresponding to the t-th level of airspace security. is the airspace level weight, k is the judgment correction coefficient, and s is the total number of airspace security levels.

[0010] Furthermore, the parameter correlation expression for the data analysis-based drone safety screening is as follows: Where S is the comprehensive evaluation value of safety screening, and R is the quantitative value of risk level. This is the risk impact coefficient. The cluster center value of the Kth type trajectory is... Z represents the association weight of the clustering results, and Z represents the total number of trajectory cluster categories.

[0011] Further, S3 includes the following sub-steps: S31, extracting various standard trajectory feature parameters from the preset airspace safety trajectory benchmark library, classifying and storing them according to airspace type, flight altitude, and flight purpose, and constructing a hierarchical benchmark feature set; S32, matching the clustered trajectory feature parameters to be detected one by one with the benchmark feature set according to the corresponding level, and calculating the absolute deviation value and relative deviation rate of each feature parameter; S33, setting a feature deviation threshold range, marking deviation parameters that exceed the threshold range, and counting the number of marked parameters and the cumulative deviation value; S34, based on the proportion of marked parameters and the cumulative deviation value, combined with the anomaly judgment criteria corresponding to the airspace safety level, determining the start point, end point, and duration of potential abnormal trajectory segments.

[0012] Further, S4 includes the following sub-steps: S41, unifying the data format of equipment operating parameter data, navigation module error data, communication signal frequency band data, and meteorological interference data, converting them into standardized numerical data; S42, constructing four data sub-matrices according to data type: power system, navigation system, communication system, and environmental interference, specifying that the row vectors of each sub-matrix represent parameter types and the column vectors represent time nodes; S43, setting a multimodal data collaborative weight matrix, allocating weight coefficients according to the degree of influence of various types of data on safety screening, and performing collaborative fusion of sub-matrices through matrix multiplication; S44, performing dimensional reduction processing on the fused collaborative analysis matrix, retaining the calibrated influencing parameters, eliminating redundant information, and generating a simplified multimodal collaborative analysis result.

[0013] Further, step S5 includes the following sub-steps: S51, associating and mapping the feature parameters corresponding to the marked abnormal trajectory segments with the relevant parameters in the multimodal collaborative analysis results to establish a parameter association comparison table; S52, performing logical verification on the associated parameters according to the UAV security inspection data management and judgment rules to determine whether there is a causal relationship between the abnormal trajectory and the multimodal data anomaly; S53, calling the risk quantification indicators in the airspace security level classification standard, substituting the verified abnormal information into the risk level assessment formula, and calculating the risk score corresponding to each anomaly; S54, summing up the various risk scores, performing weighted summation according to preset weights to obtain the final risk level quantification assignment, and marking the risk points.

[0014] A data-driven drone safety screening system, applied to a data-driven drone safety screening method, includes: a multi-source data acquisition and transmission unit for acquiring drone flight spatiotemporal trajectory data, equipment operating parameter data, airspace environment data, and communication interaction data, and transmitting the data to a data processing unit in real time; a trajectory feature extraction and clustering unit, connected to the multi-source data acquisition and transmission unit, for extracting multi-dimensional trajectory features and performing clustering using a multi-feature clustering algorithm for drone spatiotemporal trajectories; and an abnormal trajectory identification and marking unit, connected to the trajectory feature extraction and clustering unit, for identifying abnormal trajectories using an airspace abnormal trajectory identification model. The system includes: a detection trajectory matching and baseline trajectory marking; a multimodal data collaborative analysis unit, connected to the multi-source data acquisition and transmission unit and the abnormal trajectory identification and marking unit, which performs fusion analysis of multiple types of data through a multimodal data collaborative intelligent analysis algorithm; a security risk assessment and quantification unit, connected to the multimodal data collaborative analysis unit, which uses the UAV security inspection data management and assessment engine to perform abnormal information correlation verification and risk level quantification; and a screening conclusion generation and command output unit, connected to the security risk assessment and quantification unit, which generates security screening conclusions and management response commands based on the risk level quantification results, enabling real-time output and execution of commands.

[0015] Beneficial Effects: This invention proposes a data analysis-based drone safety screening system and method. Through a comprehensive design encompassing multi-dimensional data acquisition, trajectory feature clustering, anomaly identification, multi-modal data collaborative analysis, risk assessment quantification, and conclusion command output, it achieves intelligent and precise drone safety screening. Addressing the limitations of traditional methods such as single data dimensions and isolated analysis, this invention integrates multiple types of data, including spatiotemporal trajectory, equipment operation, airspace environment, and communication interaction. Through multi-modal collaborative analysis, it achieves deep data fusion, significantly broadening the dimensions of risk identification and improving the comprehensiveness of anomaly detection in complex scenarios. Furthermore, it addresses the shortcomings of existing technologies, such as the lack of a systematic assessment mechanism and inconsistent risk quantification. To address shortcomings, this paper utilizes a dedicated clustering algorithm, anomaly identification model, and security inspection data management and analysis engine to construct a closed-loop processing flow of "feature extraction - anomaly labeling - correlation verification - level quantification." This enables precise correlation between trajectory anomalies and multi-dimensional data anomalies. Standardized risk quantification rules ensure the consistency and reliability of the analysis results, effectively improving the accuracy and efficiency of security screening. Simultaneously, the fully automated process reduces manual intervention and rapidly generates control response instructions, meeting the real-time and accuracy requirements of airspace security management. This comprehensively solves the problems of one-sided identification, non-standardized analysis, and delayed response in traditional screening methods, providing efficient and feasible technical support for UAV airspace safety management. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating the overall process of the method of the present invention. Figure 2 This is a flowchart of method step S3 of the present invention; Figure 3 This is a flowchart of method step S4 of the present invention; Figure 4 This is a flowchart of step S5 of the method of the present invention; Figure 5 This is a diagram showing the system unit composition of the present invention. Detailed Implementation

[0017] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. The application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0018] like Figure 1As shown, a data analysis-based method for drone safety screening includes the following steps: S1, collecting spatiotemporal trajectory data, equipment operating parameter data, airspace environment data, and communication interaction data during drone flight to establish a multi-dimensional raw data set; S2, calling a multi-feature clustering algorithm for drone spatiotemporal trajectory to extract flight speed change features, heading angle features, altitude fluctuation features, and trajectory point density distribution features from the spatiotemporal trajectory data, and performing feature dimension mapping and clustering; S3, using an airspace abnormal trajectory identification model, matching and calculating the deviation between the clustered trajectory features and a preset airspace safety trajectory benchmark library to mark potential abnormal trajectories. In the abnormal trajectory segment; S4, the multimodal data collaborative intelligent analysis algorithm is activated, integrating power system load data from equipment operating parameter data, navigation module error data, signal frequency band data from communication interaction data, and meteorological interference data from airspace environment data to construct a multimodal collaborative analysis matrix; S5, using the UAV safety inspection data management and judgment engine, the marked abnormal trajectory segment and the multimodal collaborative analysis results are correlated and verified, and the risk level is quantified and assigned in combination with the airspace safety level classification standard; S6, based on the risk level quantification results, the corresponding safety screening conclusion and management response instructions are generated to complete the full-process safety screening of the UAV.

[0019] Step S1 is the data foundation building stage for drone safety screening, comprehensively collecting multi-dimensional raw data to support subsequent analysis. During implementation, the drone's onboard GPS positioning module, inertial measurement unit, flight control system, meteorological sensors, and wireless signal receiving module collect four types of key data at set frequencies: spatiotemporal trajectory data is acquired 10 times per second, including parameters such as longitude, latitude, altitude, flight speed, and heading angle, ensuring the continuity of trajectory recording; equipment operating parameter data is collected 20 times per second, including motor speed, remaining battery power, number of satellites received by the navigation module, and sensor error values, with motor speeds ranging from 1000 to 8000 RPM. The system operates at a speed of [speed value missing] revolutions per minute, with battery power accuracy reaching 1%, and receives data from 4 to 20 satellites. Airspace environmental data is collected once per minute, including wind speed, wind direction, and visibility. Wind speed measurement ranges from 0 to 25 m / s (accuracy 0.1 m / s), wind direction from 0 to 360 degrees, and visibility from 10 to 10,000 meters. Communication interaction data is simultaneously captured, including signal frequency band, signal strength, and data transmission rate. The frequency band covers 2.4 GHz to 5.8 GHz, signal strength from -120 dBm to -30 dBm, and transmission rate from 1 Mbps to 100 Mbps. After data collection, the four types of data are aligned and correlated according to timestamps. Blank and invalid data caused by signal interruptions are removed, constructing a structured, multi-dimensional raw data set. Each data record includes time information, data type identifier, and specific parameter values, ensuring data integrity, consistency, and traceability, providing high-quality data support for subsequent algorithm analysis.

[0020] Step S2 involves in-depth mining of trajectory data. Feature extraction and clustering provide a categorized feature base for anomaly identification, extracting and classifying key motion features from massive trajectory data. During implementation, a multi-feature clustering algorithm for UAV spatiotemporal trajectories is first invoked to extract four core features from the spatiotemporal trajectory data collected in S1: flight speed variation features are obtained by calculating the speed difference between adjacent trajectory points (0.1-second time intervals), ranging from -5 to 5 m / s, and the fluctuation pattern is analyzed through a sequence of 50 consecutive intervals of difference; heading angle features are calculated by the heading angle deviation between adjacent trajectory points, ranging from -180 to 180 degrees, reflecting the magnitude of flight direction changes; altitude fluctuation features are determined by the difference between the altitude of each trajectory point and the average altitude calculated within a sliding window (window size 10 trajectory points), with a fluctuation range of -100 to 100 meters; and trajectory point density distribution features are obtained by counting the number of trajectory points within a 10m × 10m grid, ranging from 1 to 50. After feature extraction, dimensional mapping is performed, linearly transforming the four types of features to a standardized range of 0 to 1 according to preset rules to eliminate dimensional differences. Then, clustering is initiated, with the number of cluster categories set to 5 to 10 (to adapt to different spatial complexities). The classification is completed by calculating the similarity between each trajectory point and the category center. The process is iteratively optimized until the similarity between trajectory points of the same category is higher than 85% and the similarity between different categories is lower than 30%. Finally, a set of trajectory features with category labels is output, providing a classification basis for anomaly identification in S3.

[0021] Step S3 is the abnormal trajectory identification stage. Potential risks are located by comparing the trajectory with a benchmark database, and trajectory segments deviating from safety regulations are accurately marked. During implementation, a pre-set airspace safety trajectory benchmark database is first invoked. This database is built based on a large amount of legal flight data and stores standard trajectory feature parameters categorized by airspace type (airport clear zone, urban core area, etc.), flight altitude range (0 to 120 meters low altitude, 120 to 500 meters medium altitude), and flight purpose (aerial photography, surveying, etc.). These parameters include speed variation range, heading angle threshold, altitude fluctuation upper limit, trajectory point density range, and other indicators. The database is also categorized by safety level (Level 1 to Level 5, with Level 1 being the highest). The trajectory features clustered in S2 are matched with standard features of the same level in the benchmark database according to the corresponding airspace type, altitude, and estimated purpose. The absolute deviation (difference between the detected value and the standard value) and relative deviation (absolute deviation / standard value) are calculated for each parameter. Deviation judgment thresholds are set as follows: absolute deviation threshold of 0.5 m / s for speed change, 15 degrees for heading angle, 10 meters for altitude fluctuation, and 5 trajectory point densities, with a uniform relative deviation of 20%. For airspaces of levels one and two, the relative deviation threshold is lowered to 15%, increasing the stringency of detection in high-safety-level airspace. Parameters exceeding the thresholds are marked, and the number of consecutive abnormal trajectory points is counted. When the number reaches 10 or more, it is judged as a potential abnormal trajectory segment. Information such as the segment's start time, end time, number of trajectory points, and cumulative deviation value are recorded. Through this hierarchical matching and deviation dual judgment mechanism, the accurate location of abnormal trajectories is achieved, clarifying the objects for subsequent multimodal analysis.

[0022] Step S4 achieves deep fusion of multimodal data, overcoming the limitations of single-dimensional data analysis and comprehensively integrating equipment, communication, and environmental data to support risk assessment. During implementation, key parameters are first extracted from the data collected in S1: equipment operating parameters focus on power system load (motor load rate 0% to 100%, 5 times per second), battery discharge current 0A to 30A (accuracy 0.1A), navigation module error (GPS positioning error 0.1 to 10 meters, 10 times per second), and gyroscope drift error 0.01 to 0.5 degrees / hour; communication interaction data extracts signal frequency bands (covering mainstream 2.4GHz and 5.8GHz bands and special 1.2GHz and 1.5GHz bands), modulation methods, and interference intensity 0dB to 60dB; airspace environmental data extracts gust intensity 0 to 15 m / s, temperature -20℃ to 60℃ (accuracy 0.5℃), and humidity 10% to 95% (accuracy 1%). The extracted data was formatted and converted into standardized numerical data. Four sub-matrices were constructed according to type: dynamic system, navigation system, communication system, and environmental interference. The row vectors represent parameter types, and the column vectors represent time nodes (consistent with the trajectory data timestamps, with an interval of 0.1 seconds). The matrix elements are the parameter values ​​at the corresponding time nodes. A collaborative weight matrix was set, with a weight of 0.3 for the dynamic system, 0.25 for the navigation system, 0.25 for the communication system, and 0.2 for environmental interference (total weight sum 1). The four sub-matrices were fused through matrix multiplication to obtain a preliminary collaborative analysis matrix. Finally, principal component analysis was used for dimensionality reduction, retaining principal components with a cumulative contribution rate of over 90%, and removing redundant and noisy data to generate a simplified multimodal collaborative analysis result, including collaborative correlation information and key influencing factors for various data types, providing comprehensive data support for S5 risk assessment.

[0023] Step S5 is the core of risk quantification. It achieves accurate risk assessment through correlation verification and level assignment, providing a quantitative basis for subsequent control instruction generation. During implementation, the characteristic parameters of potential abnormal trajectory segments marked in S3 (abnormal features such as speed changes, heading angles, duration, and cumulative deviation) are first correlated and mapped with relevant parameters from the multimodal collaborative analysis results in S4. A parameter correlation comparison table is established to clarify the correspondence between abnormal trajectories and parameters such as motor load rate, positioning error, signal interference, and gust intensity. Based on the preset UAV security inspection data control and judgment rules (based on airspace safety standards and UAV operation specifications), logical verification is performed on the correlated parameters to determine whether the abnormal trajectory has a causal relationship with equipment failure, communication interference, or sudden environmental changes, eliminating false anomalies caused by data acquisition errors or normal environmental fluctuations. The risk quantification indicators in the airspace safety level classification standard are called, and the anomaly information after correlation verification is substituted into the risk assessment logic. Corresponding scores are assigned according to the severity of the anomaly (minor, moderate, severe), with trajectory anomalies having a weight of 0.6 and multimodal data anomalies having a weight of 0.4. By combining the airspace safety level benchmark thresholds (different thresholds for levels one to five) with weighting coefficients, the scores of each anomaly are weighted and summed to obtain a quantitative risk level (range 0 to 100). At the same time, key risk points (such as abnormal signal frequency bands, excessive motor load rate, etc.) are marked to ensure the accuracy and standardization of risk assessment and provide quantitative support for the screening conclusions of S6.

[0024] Step S6 is the final stage of the safety screening process. It generates conclusions and instructions to complete the entire process control, transforming the quantified risk results into actionable safety management actions. During implementation, the risk level classification standards are first clarified. Based on the quantitative values ​​(0 to 100) assigned in S5, risks are divided into four levels: 0 to 25 for low risk, 26 to 50 for medium risk, 51 to 75 for high risk, and 76 to 100 for extremely high risk. Corresponding safety screening conclusions are generated for each risk level. A low-risk conclusion indicates that the drone's flight complies with airspace safety regulations; a medium-risk conclusion indicates potential safety hazards but does not constitute a serious threat; a high-risk conclusion indicates obvious violations of flight regulations; and an extremely high-risk conclusion indicates a significant safety risk that may endanger public safety. Based on the findings, control and response instructions are generated: For low-risk cases, a continuous monitoring instruction is issued, maintaining the existing data collection and analysis frequency; for medium-risk cases, a warning instruction is issued, sending safety alerts to the drone operator via communication link, while simultaneously increasing the monitoring frequency to 1.5 times the original level; for high-risk cases, a flight restriction instruction is issued, instructing the drone to reduce its altitude, adjust its course to a safe area, or limit its speed; for extremely high-risk cases, a mandatory control instruction is issued, disabling some drone functions (such as hovering and climbing), or guiding the drone to make an emergency landing in a pre-set safe area. After generation, the instructions are transmitted in real-time to the drone control terminal and airspace management platform via a wireless communication module, simultaneously recording the screening results, instruction content, execution time, drone identification, and other information to form a complete screening file, completing the entire drone safety screening process and achieving closed-loop risk control.

[0025] Preferably, the expression for the UAV spatiotemporal trajectory multi-feature clustering algorithm is: ,in, The cluster center value of the Kth type trajectory is... Let be the velocity change at the i-th trajectory point. Let be the heading angle value of the i-th trajectory point. Let be the elevation fluctuation of the i-th trajectory point. Let be the neighborhood density value of the i-th trajectory point. These are the weight coefficients for each feature dimension. is the clustering convergence adjustment factor, and n is the total number of trajectory points.

[0026] Specifically, the UAV spatiotemporal trajectory multi-feature clustering algorithm is based on the core requirements of trajectory multi-feature fusion and similarity measurement. It integrates four key features—speed change, heading angle, altitude fluctuation, and trajectory point density—to construct a clustering model that balances the importance of each feature with data balance. First, the comprehensive feature contribution of a single trajectory point needs to be quantified. Therefore, the four feature parameters are multiplied by their corresponding weight coefficients and then summed. The weight coefficients are determined based on the degree of influence of each feature on trajectory clustering: speed change feature weight is 0.3, heading angle feature weight is 0.25, altitude fluctuation feature weight is 0.25, and trajectory point density feature weight is 0.2, ensuring the total weight sum is 1. To eliminate the influence of differences in feature parameter values, a normalization process using Euclidean distance in the denominator is introduced to prevent a single feature value from dominating the clustering result. Simultaneously, a clustering convergence adjustment factor is added, ranging from 0.8 to 1.2, to optimize the convergence speed and stability of the clustering process. The total number of trajectory points is determined based on the actual flight time, with the number of points per trajectory segment typically ranging from 500 to 2000. The implementation of this formula is as follows: First, extract the specific values ​​of four types of feature parameters for each trajectory point from the spatiotemporal trajectory data, substitute them into the formula to calculate the cluster association value corresponding to each trajectory point, then determine the cluster center value of each category through iterative calculation, complete the grouping based on the similarity between the trajectory point and the cluster center value, and improve the accuracy and rationality of trajectory clustering through multi-feature weighted fusion and normalization processing to ensure that trajectories of the same type have similar motion characteristics, providing a reliable classification basis for subsequent abnormal trajectory identification.

[0027] Preferably, the expression for the spatial anomaly trajectory recognition model is: Where A is the quantified value of trajectory anomaly. Let j be the value of the j-th feature parameter of the trajectory to be detected. Let j be the value of the standard feature parameter in the benchmark library. To correct the threshold for feature deviation, is the spatial safety factor, and m is the total number of characteristic parameters.

[0028] Specifically, the airspace anomaly trajectory recognition model is based on the deviation quantification logic between the trajectory to be detected and the standard trajectory. Through feature-by-feature deviation calculation and comprehensive correction, it achieves accurate assessment of the degree of trajectory anomaly. The determination of anomaly trajectories needs to reflect the degree of deviation of each feature parameter from the benchmark value. Therefore, the deviation coefficient of each feature is integrated in a product form. The deviation coefficient of each feature is calculated by "1 minus the relative deviation", which ensures that the larger the deviation, the smaller the coefficient, and also reflects the synergistic effect of each feature. A feature deviation correction threshold is introduced, with a value range of 0.01 to 0.1, to avoid anomalies in the relative deviation calculation when the standard feature parameter value is too small. The airspace safety factor is set according to different airspace safety levels. The coefficients for airspace levels one to five are 1.5, 1.3, 1.0, 0.8, and 0.6, respectively, to ensure that the anomaly judgment in high-safety-level airspaces is more stringent. The total number of feature parameters is fixed at 4, corresponding one-to-one with the four types of features extracted by trajectory clustering. The formula is implemented as follows: Standard feature parameter values ​​for the corresponding level are retrieved from a pre-defined airspace safety trajectory benchmark library. The actual values ​​of the four types of feature parameters for the trajectory to be detected are obtained. The absolute and relative deviations are calculated for each parameter, and then substituted into the formula to obtain a quantified value for trajectory anomaly. The quantified value ranges from 0 to 1; the closer it is to 0, the more severe the trajectory anomaly. Through multi-feature deviation collaborative calculation and safety level coefficient correction, accurate quantification of abnormal trajectories under different airspace scenarios is achieved, providing core data support for risk level assessment.

[0029] Preferably, the expression for the multimodal data collaborative intelligent analysis algorithm is: Where M is the value of the multimodal collaborative analysis result, Let q be the parameter value of the p-th modal data. The weighting coefficients for the p-th modal data are... Let q be the confidence factor of the q-th parameter. The noise interference value of the p-th modal data. denoted as the calibration coefficient for the p-th modal data, and l is the number of parameters for each modal data.

[0030] Specifically, the multimodal data collaborative intelligent analysis algorithm, based on the weighted fusion and noise correction requirements of multiple data types, achieves deep collaborative analysis of multimodal data through a two-layer summation and error compensation mechanism. Multimodal data fusion needs to reflect the differences in importance of different modal data and their internal parameters; therefore, a two-layer summation structure is adopted. The first layer sums by modal type, and the second layer sums by parameters within each modality. Weight coefficients are assigned to the four types of modal data: 0.3 for power system data, 0.25 for navigation system data, 0.25 for communication system data, and 0.2 for environmental interference data. The confidence factor for each parameter is determined based on the parameter measurement accuracy, ranging from 0.7 to 1.0; the higher the measurement accuracy, the closer the factor is to 1.0. Noise interference exists during multimodal data acquisition; therefore, an error compensation term is introduced, with the noise interference value determined based on sensor performance, ranging from 0.01 to 0.1. The calibration coefficient is set from 0.9 to 1.1 to correct the impact of noise on the fusion results. Each type of modal data has 3 to 5 parameters to ensure the comprehensiveness of the data fusion. The implementation of this formula involves standardizing and unifying the format of four types of modal data, extracting the specific values ​​of parameters within each modality, and substituting these values ​​into the formula to calculate the multimodal collaborative analysis result. The result value ranges from 0 to 10, with larger values ​​indicating more significant data collaboration anomalies. Through hierarchical weighted fusion and noise correction, the correlation information between different types of data is fully explored, overcoming the limitations of single-data-dimensional analysis and improving the comprehensiveness and accuracy of security risk identification.

[0031] Preferably, the risk quantification expression of the drone security inspection data management and analysis engine is: Where R is the quantified risk level, A is the quantified trajectory anomaly, and M is the result of the multimodal collaborative analysis. For the result weighting coefficient, This is the baseline threshold corresponding to the t-th level of airspace security. is the airspace level weight, k is the judgment correction coefficient, and s is the total number of airspace security levels.

[0032] Specifically, the risk quantification of the drone security inspection data management and analysis engine is based on the correlation verification logic between abnormal trajectories and multimodal data. Through weighted summation and benchmark threshold normalization, it achieves accurate quantification of risk levels. Risk quantification needs to comprehensively consider the impact of trajectory anomalies and multimodal data anomalies. Therefore, the quantified values ​​of both are multiplied by their respective weighting coefficients and then summed. The trajectory anomaly weighting coefficient is 0.6, and the multimodal collaborative analysis result weighting coefficient is 0.4, highlighting the core impact of trajectory anomalies. The denominator uses the weighted sum of airspace safety level benchmark thresholds. The benchmark thresholds for levels one through five are 8.0, 6.0, 4.0, 2.0, and 1.0, respectively, ensuring that risk quantification matches airspace safety levels. The analysis correction coefficient is determined based on the statistical results of the misjudgment rate in actual application scenarios, ranging from 0.9 to 1.1, to optimize the accuracy of risk quantification. The total number of airspace safety levels is fixed at five. The implementation of this formula is as follows: Obtain the quantified value of trajectory anomalies and the result value of multimodal collaborative analysis; retrieve the benchmark threshold and level weight of the corresponding airspace safety level; substitute them into the formula to calculate the quantified value of the risk level, with the quantified value ranging from 0 to 100. Through two-factor weighted fusion and airspace benchmark normalization, the standardized quantification of risk levels is achieved, ensuring the consistency and comparability of risk assessment results under different airspace scenarios, and providing a quantitative basis for the generation of control and response instructions.

[0033] Preferably, the parameter correlation expression for the data analysis-based drone safety screening is: Where S is the comprehensive evaluation value of safety screening, and R is the quantitative value of risk level. This is the risk impact coefficient. The cluster center value of the Kth type trajectory is... Z represents the association weight of the clustering results, and Z represents the total number of trajectory cluster categories.

[0034] Specifically, the parameter correlation expression for drone safety screening based on data analysis, based on the deep correlation requirement between risk quantification results and trajectory clustering results, achieves a comprehensive safety screening evaluation through linear combination. The comprehensive safety screening evaluation needs to consider the impact of risk level and trajectory clustering results; therefore, the risk level quantification value and cluster center value are linearly combined. The risk impact coefficient is determined according to the strictness of control, ranging from 1.0 to 1.2. The correlation weight of the clustering results is determined according to the correlation between the cluster category and the abnormal trajectory, ranging from 0.1 to 0.3, with higher correlation resulting in greater weight. The total number of trajectory cluster categories is 5 to 10 to ensure the consistency of parameter correlation. The implementation of this formula is as follows: obtain the risk level quantification value and the cluster center value of each category of trajectory, substitute them into the formula to calculate the comprehensive safety screening evaluation value, with the evaluation value ranging from 0 to 120. Based on the evaluation value, four risk levels are divided: low, medium, high, and extremely high, corresponding to evaluation value ranges of 0 to 25, 26 to 50, 51 to 75, and 76 to 120, respectively. Its significance lies in achieving a comprehensive evaluation of safety screening by integrating the results of risk quantification and trajectory clustering, avoiding the one-sidedness of a single evaluation indicator, and providing a comprehensive and accurate decision-making basis for the generation of final safety screening conclusions and control response instructions.

[0035] Preferred, such as Figure 2 As shown, step S3 includes the following sub-steps: S31, extracting various standard trajectory feature parameters from the preset airspace safety trajectory benchmark library, classifying and storing them according to airspace type, flight altitude, and flight purpose, and constructing a hierarchical benchmark feature set; S32, matching the clustered trajectory feature parameters to be detected one by one with the benchmark feature set according to the corresponding level, and calculating the absolute deviation value and relative deviation rate of each feature parameter; S33, setting a feature deviation threshold range, marking deviation parameters that exceed the threshold range, and counting the number of marked parameters and the cumulative deviation value; S34, based on the proportion of marked parameters and the cumulative deviation value, combined with the anomaly judgment standard corresponding to the airspace safety level, determining the start point, end point, and duration of potential abnormal trajectory segments.

[0036] Specifically, step S3 constructs a hierarchical, multi-dimensional abnormal trajectory identification process through four sub-steps, improving the accuracy and adaptability of abnormal trajectory marking. During implementation, step S31 is executed first, extracting various standard trajectory feature parameters from a pre-set airspace safety trajectory benchmark library. This benchmark library includes six airspace types such as airport clear zone, urban core area, and general airspace; three flight altitude ranges: 0 to 120 meters, 120 to 500 meters, and 500 to 1000 meters; and standard data corresponding to five flight purposes such as aerial photography, surveying, and patrol. These data are structured and stored according to the above classification dimensions, constructing a benchmark feature set comprising 18 levels. Each level includes eight core parameters such as speed variation range and heading angle threshold. Next, step S32 is executed. The clustered trajectory feature parameters from step S2 are matched with standard parameters of the corresponding level in the baseline feature set according to the actual airspace type, altitude range, and estimated flight purpose. The absolute deviation value and relative deviation rate are calculated using a step-by-step comparison method. The absolute deviation value is the difference between the parameter to be detected and the standard parameter, and the relative deviation rate is the ratio of the absolute deviation value to the standard parameter. The calculation accuracy is retained to three decimal places. Then, step S33 is executed to set the feature deviation threshold range. The absolute deviation threshold for speed change features is 0.5 m / s, and the relative deviation threshold is 20%. The absolute deviation threshold for heading angle features is 15 degrees, and the relative deviation threshold is 25%. The absolute deviation threshold for altitude fluctuation features is 10 meters, and the relative deviation threshold is 30%. The absolute deviation threshold for trajectory point density distribution features is 5 points, and the relative deviation threshold is 35%. Parameters exceeding any threshold range are marked, and the number of marked parameters and the cumulative value of each deviation are counted. Finally, S34 is executed. Based on the proportion of the number of marked parameters to the total number of parameters and the cumulative deviation value, combined with the anomaly judgment criteria corresponding to the airspace safety level, if the proportion of marked parameters in Level 1 to Level 2 airspace exceeds 30% or the cumulative deviation value exceeds 50, it is judged as an anomaly. If the proportion of marked parameters in Level 3 to Level 5 airspace exceeds 40% or the cumulative deviation value exceeds 60, it is judged as an anomaly. The starting point, ending point and duration of potential abnormal trajectory segments are clearly defined to ensure that the identification of abnormal trajectories is both comprehensive and meets the safety management requirements of different airspaces.

[0037] Preferred, such as Figure 3As shown, step S4 includes the following sub-steps: S41, unifying the data format of equipment operating parameter data, navigation module error data, communication signal frequency band data, and meteorological interference data, converting them into standardized numerical data; S42, constructing four data sub-matrices according to data type: power system, navigation system, communication system, and environmental interference, specifying that the row vectors of each sub-matrix represent parameter types and the column vectors represent time nodes; S43, setting a multimodal data collaborative weight matrix, allocating weight coefficients according to the degree of influence of various types of data on safety screening, and performing collaborative fusion of sub-matrices through matrix multiplication; S44, performing dimensional reduction processing on the fused collaborative analysis matrix, retaining the calibrated influencing parameters, eliminating redundant information, and generating a simplified multimodal collaborative analysis result.

[0038] Specifically, step S4 achieves standardized processing, structured construction, collaborative fusion, and simplification optimization of multimodal data through four sub-steps, fully exploring the correlation value of multiple data types and providing comprehensive support for risk assessment. During implementation, step S41 is performed first, standardizing the formats of equipment operating parameter data, navigation module error data, communication signal frequency band data, and meteorological interference data. Non-numerical data such as analog signal data in equipment operating parameters and frequency range data in communication signal frequency bands are converted into standardized numerical data from 0 to 100 through encoding, ensuring the fusionability of different data types, with data errors controlled within ±1% during the conversion process. Next, step S42 is executed to construct four data sub-matrices based on data type. The power system sub-matrix includes five parameters such as motor load rate and battery discharge current; the navigation system sub-matrix includes four parameters such as GPS positioning error and gyroscope drift error; the communication system sub-matrix includes three parameters such as signal frequency band and interference intensity; and the environmental interference sub-matrix includes four parameters such as gust intensity and temperature. The row vectors of each sub-matrix correspond to specific parameter types, and the column vectors correspond to time nodes. The time node interval is consistent with the trajectory data acquisition frequency of 0.1 seconds, and the matrix elements are the standardized parameter values ​​of the corresponding time nodes. Then, step S43 is executed to set a multimodal data collaborative weight matrix. Based on the degree of influence of various types of data on safety screening, the power system data is assigned a weight of 0.3, the navigation system data 0.25, the communication system data 0.25, and the environmental interference data 0.2, with a total weight of 1. The four data sub-matrices are collaboratively fused with the weight matrix through matrix multiplication to obtain a preliminary collaborative analysis matrix with a dimension of 16 rows (total number of parameters) × N columns (number of time nodes). Finally, S44 is executed, and principal component analysis is used to reduce the dimensionality of the fused collaborative analysis matrix. The cumulative contribution rate threshold is set to 90%, the core influence parameters corresponding to the first 6 principal components are retained, redundant information and noisy data are removed, and a simplified multimodal collaborative analysis result is generated. This result is presented in the form of a parameter correlation matrix, which clearly reflects the collaborative relationship between different types of data and provides structured data for subsequent anomaly correlation verification.

[0039] Preferred, such as Figure 4 As shown, step S5 includes the following sub-steps: S51, associating and mapping the feature parameters corresponding to the marked abnormal trajectory segments with the relevant parameters in the multimodal collaborative analysis results to establish a parameter association comparison table; S52, performing logical verification on the associated parameters according to the UAV security inspection data management and judgment rules to determine whether there is a causal relationship between the abnormal trajectory and the multimodal data anomaly; S53, calling the risk quantification indicators in the airspace security level classification standard, substituting the abnormal information after association verification into the risk level assessment formula, and calculating the risk score corresponding to each anomaly; S54, summing up the various risk scores, performing weighted summation according to preset weights to obtain the final risk level quantification assignment, and marking the risk points.

[0040] Specifically, step S5 constructs a risk assessment process of "association-verification-quantification-marking" through four sub-steps to achieve accurate association of abnormal information and scientific quantification of risk levels. During implementation, S51 is executed first, mapping the characteristic parameters corresponding to the potential abnormal trajectory segments marked in step S3, including six core parameters such as abnormal speed changes and heading angle deviations, to the relevant parameters in the multimodal collaborative analysis results obtained in step S4. A parameter association comparison table is established to clarify the correspondence between abnormal trajectory speed mutations and parameters such as motor load rate changes and battery discharge current fluctuations, forming 12 sets of key association items to ensure the comprehensiveness of the association mapping. Next, S52 is executed, based on preset UAV security inspection data management and assessment rules. These rules include 20 core logical judgment clauses, including three types of causal relationship judgment standards: equipment failure, communication interference, and environmental mutations. Logical verification of the associated parameters is performed, comparing the temporal synchronicity and numerical correlation of parameter changes to determine whether there is a direct or indirect causal relationship between the abnormal trajectory and multimodal data anomalies. The verification accuracy rate is no less than 95%, eliminating false anomalies caused by data acquisition errors or normal environmental fluctuations. Then, step S53 is executed, which calls the risk quantification indicators in the airspace safety level classification standard. Levels one through five of the airspace correspond to different quantification benchmarks. The verified anomaly information is substituted into the risk level assessment logic, and different scores are assigned according to the severity of the anomalies: 5 points for minor anomalies, 10 points for moderate anomalies, and 20 points for severe anomalies. The risk score for each anomaly is calculated, rounded to the nearest integer. Finally, step S54 is executed, which summarizes all risk scores and performs a weighted sum according to preset weights. The weight for trajectory anomalies is 0.6, and the weight for multimodal data anomalies is 0.4, resulting in the final risk level quantification value, ranging from 0 to 100. Key risk points are also marked, including eight high-impact anomalies such as abnormal signal frequencies and excessive motor load rates. A risk quantification report is generated, providing accurate and comprehensive risk assessment basis for the generation of the safety screening conclusion in step S6.

[0041] like Figure 5 As shown, a data analysis-based drone safety screening system is applied to a data analysis-based drone safety screening method. The system includes: a multi-source data acquisition and transmission unit for collecting drone flight spatiotemporal trajectory data, equipment operating parameter data, airspace environment data, and communication interaction data, and transmitting the data to a data processing unit in real time; a trajectory feature extraction and clustering unit connected to the multi-source data acquisition and transmission unit, which extracts multi-dimensional trajectory features and performs clustering using a drone spatiotemporal trajectory multi-feature clustering algorithm; and an abnormal trajectory identification and labeling unit connected to the trajectory feature extraction and clustering unit, which uses an airspace abnormal trajectory identification model to achieve... The system includes: matching the target trajectory with the baseline trajectory and marking abnormal segments; a multimodal data collaborative analysis unit, connected to the multi-source data acquisition and transmission unit and the abnormal trajectory identification and marking unit, which performs fusion analysis of multiple types of data through a multimodal data collaborative intelligent analysis algorithm; a security risk assessment and quantification unit, connected to the multimodal data collaborative analysis unit, which uses the UAV security inspection data management and assessment engine to perform abnormal information correlation verification and risk level quantification; and a screening conclusion generation and command output unit, connected to the security risk assessment and quantification unit, which generates security screening conclusions and management response commands based on the risk level quantification results, enabling real-time output and execution of commands.

[0042] The formula of this invention enables unified calculation of different scalar and vector parameters. It eliminates dimensional differences and clarifies parameter correlation logic through standardization, weight allocation, and dimensional adaptation mechanisms. Taking the multi-feature clustering algorithm for UAV spatiotemporal trajectories as an example, vector parameters such as velocity change and heading angle, and scalar parameters such as neighborhood density, are first converted into standardized data within the same numerical range through feature dimension mapping. Then, preset weight coefficients are used to assign importance to each parameter in relation to safety screening, avoiding interference between the directional characteristics of vector parameters and the magnitude characteristics of scalar parameters. In the airspace abnormal trajectory identification model, the deviation calculation between the feature parameters (vectors) of the trajectory to be detected and the standard parameters (scalars) in the benchmark library is performed. A feature deviation correction threshold is introduced to eliminate numerical magnitude differences, and the airspace safety factor further calibrates the influence weights of different types of parameters. The multimodal data collaborative intelligent analysis algorithm integrates the noise interference value of scalars and the modal parameter values ​​of vectors through a two-layer summation structure. Parameter correlation is established using weight coefficients and confidence factors, while the error compensation term under the square root balances the calculation deviations of parameters in different dimensions. These formulas are designed with "standardized and unified dimensions + clear correlation of weight allocation + balanced characteristics of adapting terms", which transforms the direction and magnitude information of vector parameters and the numerical information of scalar parameters into quantitative indicators that can be calculated collaboratively. This not only preserves the core characteristics of each parameter, but also realizes the fusion analysis of cross-type parameters, adapting to the multi-dimensional data processing needs of UAV safety screening.

[0043] A data-driven unmanned aerial vehicle (UAV) safety screening system and method are proposed. This system constructs a multi-dimensional, end-to-end intelligent safety screening framework, achieving accurate identification and efficient control of UAV safety risks through multi-source data integration and dedicated algorithm models. At the methodological level, it is based on multi-type data collection, comprehensively covering key information such as flight trajectory, equipment operation, airspace environment, and communication interaction, avoiding blind spots caused by single data dimensions. Through progressive processing via trajectory multi-feature clustering, anomaly identification, multi-modal collaborative analysis, and a judgment engine, a complete technical chain from data extraction to risk quantification is formed, ensuring the systematic and logical nature of the screening process. At the system level, modular unit design enables seamless data flow and functional connection between each stage, ensuring efficient progress of the screening process. Simultaneously, standardized parameter processing and risk quantification rules improve the consistency and reliability of screening results.

[0044] This invention addresses the problems of single data dimensions and isolated analysis in traditional technologies. By integrating multimodal data and employing a collaborative analysis mechanism, it achieves deep fusion and correlation verification of different data types, significantly broadening the coverage of risk identification and improving the comprehensiveness of anomaly detection in complex scenarios. Addressing the shortcomings of existing technologies, such as the lack of a systematic assessment mechanism and non-standardized risk quantification, this invention utilizes a dedicated algorithm model and assessment engine to construct a closed-loop processing flow. It accurately correlates trajectory anomaly features with multi-dimensional data anomalies, ensuring the accuracy and standardization of risk assessment results through standardized quantification rules and level classification. Simultaneously, the fully automated processing reduces human intervention, significantly improving screening response speed. This effectively solves the problems of one-sided identification, chaotic assessment, and delayed response in traditional methods, providing more efficient and reliable technical support for UAV airspace safety management.

[0045] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "set," "install," "connect," "link," and "fix" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal communication between two components. Those skilled in the art will understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0046] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various equivalent changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A data analysis-based method for unmanned aerial vehicle (UAV) safety screening, characterized in that, Includes the following steps: S1 collects spatiotemporal trajectory data, equipment operation parameter data, airspace environment data, and communication interaction data during the flight of the UAV, and establishes a multi-dimensional raw data set; S2 calls the UAV spatiotemporal trajectory multi-feature clustering algorithm to extract flight speed change features, heading angle features, altitude fluctuation features and trajectory point density distribution features from the spatiotemporal trajectory data, and performs feature dimension mapping and clustering. S3, through the airspace abnormal trajectory identification model, performs feature matching and deviation calculation between the clustered trajectory features and the preset airspace safe trajectory benchmark library, and marks potential abnormal trajectory segments. S4, activate the multimodal data collaborative intelligent analysis algorithm, integrate the power system load data, navigation module error data, signal frequency band data in the equipment operation parameter data, and meteorological interference data in the airspace environment data to construct a multimodal collaborative analysis matrix; S5 utilizes the drone security inspection data management and analysis engine to perform correlation verification on marked abnormal trajectory segments and multimodal collaborative analysis results, and quantifies and assigns risk level values ​​in conjunction with airspace security level classification standards. S6 generates corresponding safety screening conclusions and control response instructions based on the risk level quantification results, completing the entire process of drone safety screening.

2. The data analysis-based drone safety screening method according to claim 1, characterized in that, The expression for the UAV spatiotemporal trajectory multi-feature clustering algorithm is as follows: ， in, The value of the cluster center for the k-th type of trajectory. Let be the velocity change at the i-th trajectory point. Let be the heading angle value of the i-th trajectory point. Let be the elevation fluctuation of the i-th trajectory point. Let be the neighborhood density value of the i-th trajectory point. These are the weight coefficients for each feature dimension. is the clustering convergence adjustment factor, and n is the total number of trajectory points.

3. The data analysis-based drone safety screening method according to claim 1, characterized in that, The expression for the spatial anomaly trajectory recognition model is: ， Where A is the quantified value of trajectory anomaly. Let j be the value of the j-th feature parameter of the trajectory to be detected. Let j be the value of the standard feature parameter in the benchmark library. To correct the threshold for feature deviation, is the spatial safety factor, and m is the total number of characteristic parameters.

4. The data analysis-based drone safety screening method according to claim 1, characterized in that, The expression for the multimodal data collaborative intelligent analysis algorithm is: ， Where M is the value of the multimodal collaborative analysis result, Let q be the parameter value of the p-th modal data. The weighting coefficients are for the p-th modal data. Let q be the confidence factor of the q-th parameter. The noise interference value of the p-th modal data. denoted as the calibration coefficient for the p-th modal data, and l is the number of parameters for each modal data.

5. The data analysis-based drone safety screening method according to claim 1, characterized in that, The risk quantification expression of the drone security inspection data management and analysis engine is as follows: ， Where R is the quantified risk level, A is the quantified trajectory anomaly, and M is the result of the multimodal collaborative analysis. For the result weighting coefficient, This is the baseline threshold corresponding to the t-th level of airspace security. The weights are categorized into levels, k is the assessment correction coefficient, and s is the total number of airspace security levels.

6. The data analysis-based drone safety screening method according to claim 1, characterized in that, The parameter correlation expression for the data analysis-based drone safety screening is as follows: Where S is the comprehensive evaluation value of safety screening, and R is the quantitative value of risk level. This is the risk impact coefficient. The value of the cluster center for the k-th type of trajectory. is the association weight of the clustering results, and z is the total number of trajectory cluster categories.

7. The data analysis-based drone safety screening method according to claim 1, characterized in that, S3 includes the following steps: S31. Extract various standard trajectory feature parameters from the preset airspace safety trajectory benchmark library, classify and store them according to airspace type, flight altitude and flight purpose, and construct a hierarchical benchmark feature set; S32, the clustered trajectory feature parameters to be detected are matched one by one with the benchmark feature set according to the corresponding level, and the absolute deviation value and relative deviation rate of each feature parameter are calculated. S33, set the characteristic deviation threshold range, mark the deviation parameters that exceed the threshold range, and count the number of marked parameters and the cumulative deviation value; S34. Based on the proportion of the number of marked parameters and the cumulative value of the deviation, combined with the anomaly judgment criteria corresponding to the airspace safety level, determine the starting point, ending point and duration of potential abnormal trajectory segments.

8. The data analysis-based drone safety screening method according to claim 1, characterized in that, S4 includes the following steps: S41 performs data format unification processing on equipment operating parameter data, navigation module error data, communication signal frequency band data and meteorological interference data, converting them into standardized numerical data. S42, construct four data sub-matrices according to data type: power system, navigation system, communication system, and environmental interference, and specify that the row vectors of each sub-matrix are parameter types and the column vectors are time nodes; S43, set a multimodal data collaborative weight matrix, allocate weight coefficients according to the degree of influence of various types of data on security screening, and perform collaborative fusion of sub-matrices through matrix multiplication; S44 performs dimensionality reduction on the fused collaborative analysis matrix, retains the calibration influence parameters, removes redundant information, and generates a simplified multimodal collaborative analysis result.

9. The data analysis-based drone safety screening method according to claim 1, characterized in that, S5 includes the following steps: S51, associate and map the feature parameters corresponding to the marked abnormal trajectory segments with the relevant parameters in the multimodal collaborative analysis results to establish a parameter association comparison table; S52, based on the data management and analysis rules for drone security inspection, performs logical verification on the associated parameters to determine whether there is a causal relationship between abnormal trajectories and multimodal data anomalies; S53, call the risk quantification indicators in the airspace safety level classification standard, substitute the verified abnormal information into the risk level assessment formula, and calculate the risk score corresponding to each abnormality. S54: Summarize the various risk scores, perform a weighted summation according to preset weights to obtain the final quantitative risk level, and mark the risk points.

10. A data analysis-based unmanned aerial vehicle (UAV) safety screening system, characterized in that, The system is applied to the data analysis-based drone safety screening method described in claim 1, comprising: The multi-source data acquisition and transmission unit is used to collect UAV flight spatiotemporal trajectory data, equipment operating parameter data, airspace environment data and communication interaction data, and transmit the data to the data processing unit in real time. The trajectory feature extraction and clustering unit is connected to the multi-source data acquisition and transmission unit. It extracts multi-dimensional features of the trajectory and completes clustering grouping through the UAV spatiotemporal trajectory multi-feature clustering algorithm. An abnormal trajectory recognition and marking unit is connected to the trajectory feature extraction and clustering unit. It uses a spatial abnormal trajectory recognition model to match the trajectory to be detected with the baseline trajectory and to mark abnormal segments. The multimodal data collaborative analysis unit is connected to the multi-source data acquisition and transmission unit and the abnormal trajectory identification and marking unit, respectively, and completes the fusion analysis of multiple types of data through the multimodal data collaborative intelligent analysis algorithm; The security risk assessment and quantification unit is connected to the multimodal data collaborative analysis unit, and uses the drone security inspection data management and assessment engine to perform abnormal information correlation verification and risk level quantification. The screening conclusion generation and instruction output unit is connected to the safety risk assessment and quantification unit. Based on the risk level quantification results, it generates safety screening conclusions and control response instructions, enabling real-time output and execution of instructions.

Images