Urban low-altitude unmanned aerial vehicle operation risk assessment method and system based on bayes theory

Abstract

The application relates to a city low-altitude unmanned aerial vehicle operation risk assessment method and system based on Bayesian theory, relates to the unmanned aerial vehicle flight risk assessment technical field, and comprises the following steps: acquiring predicted attribute information of a task area, and determining a regional environment fluctuation coefficient; evaluating the sensitivity of evidence factors and unmanned aerial vehicle operation risks, and outputting factor sensitivity; setting evidence factor monitoring scalars, screening initial monitoring evidence factors to construct an initial factor group; periodically monitoring and acquiring an initial factor attribute information set according to the initial factor group, calling a risk assessment subnetwork to evaluate risks, and outputting a multi-dimensional operation risk probability; if a preset initial risk threshold set is not met, acquiring a global factor attribute information set according to evidence factor monitoring, and calling a global risk assessment network to evaluate risks. The application solves the technical problem that, in traditional unmanned aerial vehicle operation risk assessment, factor redundancy and high model complexity are caused by the fact that key influence factors are not screened, and it is difficult to balance monitoring efficiency and risk assessment accuracy.

Description

Technical Field

[0001] This application relates to the field of unmanned aerial vehicle (UAV) flight risk assessment, and in particular to a method and system for assessing the operational risks of urban low-altitude UAVs based on Bayesian theory. Background Technology

[0002] With the increasingly widespread application of drones in fields such as security patrol, the accurate and efficient assessment of the operational risks of low-altitude drones in urban areas has become a key technical requirement to ensure the safe implementation of patrol missions.

[0003] Currently, traditional risk assessment methods for urban low-altitude drone operations do not specifically screen key factors affecting risks. This results in both redundant factors and high model complexity due to the lack of adaptation and optimization of the assessment model. In actual inspections, either real-time performance is sacrificed due to over-reliance on monitoring of all dimensions, or the assessment accuracy is reduced due to simplification of monitoring dimensions. It is difficult to balance real-time performance and assessment accuracy, which not only affects the efficiency of inspection tasks but also increases the safety risks of drone operations. Summary of the Invention

[0004] This application provides a method and system for assessing the operational risks of urban low-altitude unmanned aerial vehicles (UAVs) based on Bayesian theory. It improves the problems of redundant factors, high model complexity, and difficulty in balancing real-time performance and accuracy in urban low-altitude UAV security patrols, thereby enhancing the efficiency and accuracy of UAV operational risk assessment.

[0005] The embodiments of this application disclose the following technical solutions:

[0006] In a first aspect, embodiments of this application provide a method for assessing the operational risks of urban low-altitude unmanned aerial vehicles (UAVs) based on Bayesian theory, the method comprising:

[0007] Before drones perform urban low-altitude security patrols, predictive attribute information of pre-defined evidence factors in the mission area within a pre-defined time zone is obtained, and the environmental fluctuation coefficient of the predicted area is assessed and determined.

[0008] Based on the predicted attribute information, the sensitivity of several evidence factors to the operational risks of drones is evaluated, and the sensitivity of several factors is output.

[0009] Based on the importance of the inspection task within the preset time zone and the regional environmental fluctuation coefficient, a monitoring scalar of evidence factors is set, and multiple initial monitoring evidence factors are selected based on the sensitivity of the factors to construct an initial factor group.

[0010] When the drone performs urban low-altitude security patrol and inspection tasks, the initial factor attribute information set is periodically monitored and obtained according to the initial factor group, and the adaptive risk assessment sub-network is called to conduct drone operation risk assessment and output multi-dimensional operation risk probability.

[0011] If the multidimensional operational risk probability does not meet the preset initial risk threshold set, the global factor attribute information set is obtained by monitoring the several evidence factors, and the global risk assessment network is invoked to accurately assess the operational risk of the UAV.

[0012] Secondly, embodiments of this application provide a risk assessment system for urban low-altitude unmanned aerial vehicle (UAV) operations based on Bayesian theory, the system comprising:

[0013] The prediction attribute and fluctuation assessment module is used to obtain the prediction attribute information of the preset evidence factor space in the preset time zone of the task area before the UAV performs urban low-altitude security patrol mission, and to assess and determine the environmental fluctuation coefficient of the prediction area.

[0014] The factor sensitivity assessment module is used to assess the sensitivity of several evidence factors to the operational risks of the UAV based on the predicted attribute information, and output several factor sensitivities.

[0015] The initial factor group construction module is used to set the monitoring scalar of evidence factors according to the importance of the inspection task in the preset time zone and the regional environmental fluctuation coefficient, and to screen multiple initial monitoring evidence factors according to the sensitivity of the several factors to construct an initial factor group.

[0016] The sub-network risk assessment module is used to periodically monitor and obtain the initial factor attribute information set according to the initial factor group when the UAV performs urban low-altitude security patrol and inspection tasks, and call the adaptive risk assessment sub-network to conduct UAV operation risk assessment and output multi-dimensional operation risk probability.

[0017] The global network risk assessment module is used to obtain a global factor attribute information set according to the several evidence factors if the multidimensional operational risk probability does not meet the preset initial risk threshold set, and call the global risk assessment network to accurately assess the operational risk of the UAV.

[0018] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0019] This application proposes a method and system for risk assessment of urban low-altitude unmanned aerial vehicle (UAV) operations based on Bayesian theory. By conducting pre-mission predictive assessment, factor sensitivity analysis, initial factor group construction, in-mission risk monitoring and assessment, and precise risk assessment in stages, it achieves dynamic and precise risk assessment throughout the entire process of urban low-altitude UAV security patrol missions. First, before the UAV performs an urban low-altitude security patrol mission, the predicted attribute information of preset evidence factors within a preset time zone of the mission area is obtained, and the environmental fluctuation coefficient of the predicted area is determined by combining the predicted environmental parameter sequence. Then, based on the predicted attribute information and historical UAV patrol records, the impact of each evidence factor on operational risk and its relevance to the mission are assessed, and the factor sensitivity is calculated and output. Subsequently, based on the importance of the patrol mission and the regional environmental fluctuation coefficient, a monitoring scalar of evidence factors is set, and core factors are selected according to sensitivity to construct an initial factor group. During mission execution, data is periodically monitored according to the initial factor group, and an adaptive risk assessment subnetwork is invoked to output multi-dimensional operational risk probabilities. If the risk probability does not meet a preset threshold, the global attribute information of all evidence factors is further monitored, and a global risk assessment network is invoked to conduct a precise assessment, outputting risk results that are more closely aligned with the actual scenario.

[0020] This application's technical solution addresses the problems in traditional UAV risk assessment, such as incomplete factor coverage leading to biased assessments, fixed models failing to adapt to dynamic environments, and a lack of flexible threshold standards for risk determination. It significantly improves the comprehensiveness, timeliness, and accuracy of risk assessment for urban low-altitude UAV operations, providing technical support for the safe and efficient execution of UAV-based urban security patrol missions. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 A flowchart illustrating the risk assessment method for urban low-altitude unmanned aerial vehicle (UAV) operations based on Bayesian theory provided in this application embodiment;

[0023] Figure 2 A schematic diagram of the structure of the urban low-altitude unmanned aerial vehicle (UAV) operation risk assessment system based on Bayesian theory provided in this application embodiment.

[0024] The components represented by each number in the attached diagram are explained below:

[0025] Predictive Attributes and Volatility Assessment Module 01, Factor Sensitivity Assessment Module 02, Initial Factor Group Construction Module 03, Sub-Network Risk Assessment Module 04, Global Network Risk Assessment Module 05. Detailed Implementation

[0026] This application provides a method and system for risk assessment of urban low-altitude drone operations based on Bayesian theory. It is used to solve the technical problems in the existing technology of urban low-altitude drone security inspection, which are caused by the redundancy of factors due to the lack of targeted screening of key influencing factors and the high model complexity due to the lack of optimization of the assessment model, thus making it difficult to balance real-time performance and assessment accuracy.

[0027] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. 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.

[0028] In the description of this application, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the stated features. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0029] In the description of this application, the term "for example" is used to mean "used as an example, illustration, or description." Any embodiment described as "for example" in this application is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is provided to enable any person skilled in the art to make and use the invention. Details are set forth in the following description for purposes of explanation. It should be understood that those skilled in the art will recognize that the invention can be made without using these specific details. In other instances, well-known structures and processes will not be described in detail to avoid obscuring the description of the invention with unnecessary detail. Therefore, the invention is not intended to be limited to the embodiments shown, but is consistent with the broadest scope of the principles and features disclosed in this application.

[0030] Example 1: The urban low-altitude UAV operation risk assessment method based on Bayesian theory provided in this application uses a multi-rotor UAV, which has the characteristics of vertical take-off and landing, strong low-altitude hovering ability, and high maneuverability, and can adapt to the security inspection needs in the complex low-altitude environment of the city.

[0031] As attached Figure 1 As shown, this application provides a method for risk assessment of urban low-altitude unmanned aerial vehicle (UAV) operations based on Bayesian theory. The method includes the following steps:

[0032] S110: Before the UAV performs urban low-altitude security patrol mission, it acquires the predicted attribute information of the space of preset evidence factors in the mission area within the preset time zone, and evaluates and determines the environmental fluctuation coefficient of the predicted area.

[0033] In this embodiment of the application, in order to accurately quantify the impact of environmental fluctuations on the risk assessment results of UAV operation, it is necessary to first determine a preset evidence factor space containing multiple key factors, then predict the attributes of evidence factors based on the space and in conjunction with mission-related information, and finally extract environmental information from the prediction results to conduct environmental volatility assessment in order to obtain the environmental volatility coefficient.

[0034] Specifically, the first step is to define a predefined space of evidence factors, which includes environmental factors, spatial factors, performance factors, and operational factors, with each factor set containing multiple evidence factors.

[0035] Furthermore, based on the obtained preset evidence factor space, combined with the path of the UAV mission, mission attribute information, and historical UAV inspection records of the mission area, attribute information prediction is performed on several evidence factors within the preset time zone to obtain several corresponding predicted attribute information.

[0036] Furthermore, environment type prediction attribute information is extracted from the above prediction attribute information. This environment type prediction attribute information contains multiple prediction environment parameter sequences.

[0037] Finally, environmental volatility is assessed based on multiple predicted environmental parameter sequences to obtain multiple environmental volatility levels. These environmental volatility levels are then weighted and fused to obtain the predicted regional environmental volatility coefficient.

[0038] This step involves first establishing a multi-dimensional, pre-defined space of evidence factors to build the basis for the assessment, then combining task and historical data to predict the attributes of evidence factors to form a prediction dataset, next extracting environmental prediction information, and finally conducting an environmental volatility assessment to obtain the environmental volatility coefficient. This provides a key basis for considering the impact of environmental volatility on operational risks when conducting UAV operational risk assessments based on Bayesian theory.

[0039] Step S110 in the method provided in this application embodiment includes:

[0040] Obtain a preset evidence factor space, wherein the preset evidence factor space includes an environmental factor set, a spatial factor set, a performance factor set, and an operational factor set, and each factor set includes multiple evidence factors;

[0041] Based on the preset evidence factor space, combined with the UAV execution mission path, mission attribute information and historical UAV inspection records of the mission area, attribute information prediction is performed on several evidence factors within the preset time zone to obtain several predicted attribute information of the several evidence factors.

[0042] Extract the environment type prediction attribute information from the plurality of prediction attribute information, wherein the environment type prediction attribute information includes multiple prediction environment parameter sequences;

[0043] Environmental volatility is assessed based on the multiple predicted environmental parameter sequences, multiple environmental volatility values ​​are obtained, and weighted fusion is performed to obtain the predicted regional environmental volatility coefficient.

[0044] In this embodiment of the application, in order to fully cover all kinds of influencing factors during the operation of urban low-altitude UAVs and capture the environmental change characteristics of the mission area within a preset time period, it is necessary to obtain a quantified regional environmental fluctuation coefficient through environmental fluctuation analysis to ensure that the subsequent risk assessment can fully adapt to the environmental uncertainty of the mission area.

[0045] First, a pre-defined evidence factor space is obtained. This space includes environmental factor sets, spatial factor sets, performance factor sets, and operational factor sets, with each set containing multiple evidence factors.

[0046] Specifically, the environmental factors set covers factors directly related to the UAV flight environment, such as wind speed, visibility, precipitation probability, and electromagnetic interference intensity within the preset time period of the mission area; the spatial factors set includes spatial characteristics such as the distance from buildings, high-voltage lines, and no-fly zones along the mission path, as well as ground population density and the probability of obstacles appearing dynamically; and the performance factors set includes equipment performance indicators such as the UAV's own battery endurance, GPS positioning accuracy, communication link stability, and sensor detection range and accuracy.

[0047] In addition, the operational factor set involves operational factors such as the take-off and landing point settings for inspection missions, mission complexity, whether beyond visual line of sight (BVLOS) flight mode is used, and the emergency response capabilities of operators. By constructing an evidence factor space that includes the above four categories of factors, omissions or biases in subsequent risk assessments due to incomplete factor coverage can be avoided.

[0048] Furthermore, after obtaining the preset evidence factor space, the attribute information of several evidence factors within the preset time zone is predicted by combining the drone's execution mission path, mission attribute information, and historical drone inspection records of the mission area.

[0049] Among them, the mission path information can identify the key spatial nodes that need to be focused on during the drone's flight; the mission attribute information includes the priority and objectives of the inspection mission, and different mission attributes have different levels of attention to various evidentiary factors. Historical drone inspection records contain the changing patterns and data characteristics of various evidentiary factors in the same or similar time periods and similar mission types in the mission area.

[0050] During the prediction process, specific numerical predictions should be given for quantifiable parameters. For example, the average wind speed in the task area during the preset time period is predicted to be 3-5 m / s, the maximum wind speed is not more than 8 m / s, and the drone battery life can support the task execution for 45-60 minutes.

[0051] Furthermore, for dynamic factors that are difficult to quantify directly, it is necessary to predict their probability of occurrence by combining the actual environmental characteristics of the task area.

[0052] For example, the probability of bird appearance needs to be set differently according to the type of area: In the core business district of a city with dense high-rise buildings, the space for bird activity is limited, and the probability of birds appearing on the task path within the preset time period is predicted to be ≤5%; In areas with high green coverage and near small rivers or urban parks, the frequency of bird habitat and activity increases, and the probability of bird appearance can be predicted to be 5%-20%; If the task area involves ecological areas such as large wetlands and lakes, the bird population is rich and the activity is frequent, and the probability of bird appearance needs to be predicted to be >20%.

[0053] Meanwhile, considering that densely built-up urban areas are prone to local airflow disturbances, it is necessary to supplement the prediction of the probability that sudden changes in building micro-meteorology will affect drone flight. Based on historical inspection data and building layout characteristics, this probability can be set at 3%-8%. As for the probability of sudden electromagnetic interference, affected by the stability of the urban electromagnetic environment, the overall prediction is below 5%.

[0054] Furthermore, after completing the prediction of the evidence factor attribute information, the environmental type prediction attribute information is extracted. This information is presented in the form of multiple predicted environmental parameter sequences, each of which corresponds to the temporal change characteristics of an environmental factor within a preset time period.

[0055] Finally, environmental volatility is assessed based on multiple predicted environmental parameter sequences to obtain multiple environmental volatility levels, which are then weighted and fused to obtain the predicted regional environmental volatility coefficient.

[0056] In the volatility assessment process, for each predicted environmental parameter series, the standard deviation statistical method is used to calculate its volatility, and then the corresponding environmental volatility is determined by the ratio of the standard deviation to the series mean.

[0057] Specifically, the mean of all parameter values ​​in the predicted environmental parameter sequence is first obtained, and then the sum of squared deviations of each parameter value from the mean is calculated based on the mean, thereby obtaining the standard deviation. Subsequently, the environmental volatility of the corresponding environmental parameter is obtained by dividing the standard deviation by the mean, and the environmental volatility of each predicted environmental parameter sequence is calculated.

[0058] Furthermore, after obtaining the volatility of each environmental factor, it is necessary to perform weighted fusion based on the impact weight of different environmental factors on the operational risks of UAVs to obtain the predicted regional environmental volatility coefficient, which provides a quantitative basis for setting the monitoring scalar of evidence factors in conjunction with the importance of the inspection task.

[0059] For example, if the task area is the core business district of a city, and the preset time period is the security patrol period from 8:00 AM to 11:00 AM on weekdays, after obtaining the preset evidence factor space, the attribute information of each evidence factor is predicted by combining the patrol route of the area, the task attributes of emergency security patrols, and historical patrol records of the same time period in the past 3 months.

[0060] Specifically, the predicted wind speed sequence is 8:00-2.5m / s, 8:30-3.1m / s, 9:00-3.5m / s, 9:30-3.8m / s, 10:00-3.3m / s, 10:30-2.8m / s, and 11:00-2.6m / s; the predicted visibility sequence is 8:00-12km, 8:30-10km, 9:00-9km, 9:30-8km, 10:00-9km, 10:30-11km, and 11:00-12km; and the predicted electromagnetic interference intensity sequence is 8:00-Level 1, 8:30-Level 1, 9:00-Level 2, 9:30-Level 2, 10:00-Level 1, 10:30-Level 1, and 11:00-Level 1.

[0061] The electromagnetic interference intensity levels are divided according to the degree of impact on the drone's communication link and data transmission: Level 1 represents slight electromagnetic interference, at which point the drone's communication link signal is stable, the data packet loss rate is <1%, and data such as images and positioning information can be transmitted in real time and completely during the inspection process; Level 2 represents moderate electromagnetic interference, at which point the drone's communication link signal experiences slight fluctuations, the data packet loss rate is 1%-3%, but core security inspection data can still be transmitted effectively, and only non-critical auxiliary data may experience a brief delay, which will not affect the accuracy of the risk assessment results.

[0062] Meanwhile, it is predicted that the minimum distance between the flight path and buildings will be 5-8 meters during this period, and the probability of birds appearing is 18%.

[0063] Furthermore, the predicted attribute information of the above environmental types is extracted, and the environmental fluctuation is calculated for each environmental parameter sequence: the standard deviation of the wind speed sequence is 0.48 m / s and the environmental fluctuation is 0.16; the standard deviation of the visibility sequence is 1.6 km and the environmental fluctuation is 0.15; and the environmental fluctuation of the quantized electromagnetic interference intensity sequence is 0.25.

[0064] The environmental volatility is calculated as "standard deviation of environmental parameter series / mean of environmental parameter series". Taking wind speed series as an example, firstly, the mean wind speed from 8:00 to 11:00 is calculated ((2.5+3.1+3.5+3.8+3.3+2.8+2.6) / 7≈3.07m / s). Then, the standard deviation is obtained by summing the squares of the deviations of the wind speed from the mean at each time point (0.48m / s). Finally, the standard deviation is divided by the mean, i.e., 0.48 / 3.07≈0.16, to obtain the wind speed environmental volatility.

[0065] Secondly, the calculation methods for the fluctuation of visibility and electromagnetic interference intensity are the same. Electromagnetic interference intensity needs to be quantified first, with level 1 corresponding to 1 and level 2 corresponding to 2, and then the mean and standard deviation are calculated.

[0066] Furthermore, with wind speed weighting at 0.4, visibility weighting at 0.3, and electromagnetic interference weighting at 0.3, the predicted regional environmental fluctuation coefficient is calculated using the formula "Predicted regional environmental fluctuation coefficient = wind speed environmental fluctuation × wind speed weight + visibility environmental fluctuation × visibility weight + electromagnetic interference intensity environmental fluctuation × electromagnetic interference weight". The predicted regional environmental fluctuation coefficient is 0.16 × 0.4 + 0.15 × 0.3 + 0.25 × 0.3 = 0.184.

[0067] The weights of each parameter are determined by the frequency and severity of risk events caused by different environmental parameters in the historical UAV inspection records of the mission area. For example, wind speed has the highest historical proportion of flight safety risks, so it is given the highest weight of 0.4; visibility and electromagnetic interference have the next highest impact on risk, and are given weights of 0.3 respectively.

[0068] Through the above steps, a comprehensive system of evidence factors was constructed, ensuring the accuracy of attribute information prediction. Furthermore, a quantified regional environmental fluctuation coefficient was obtained through environmental volatility analysis, providing a scientific basis for setting monitoring scalars of evidence factors and screening initial monitoring evidence factors in conjunction with the importance of the task.

[0069] S120: Based on the predicted attribute information, evaluate the sensitivity of several evidence factors to the operational risk of the UAV, and output the sensitivity of several factors.

[0070] In this application embodiment, in the scenario of risk assessment of urban low-altitude UAV operation based on Bayesian theory, in order to accurately identify the key evidence factors affecting the risk of UAV operation, it is necessary to combine historical inspection records and task-related information to evaluate the evidence factors from the aspects of risk impact and task correlation, and then obtain the factor sensitivity.

[0071] Specifically, based on the historical UAV inspection records of the mission area, and according to the obtained predictive attribute information, the impact of these evidence factors on the operational risks of UAVs is assessed to output several risk impact levels.

[0072] Furthermore, based on several predictive attribute information, as well as the UAV's execution path and task attribute information, the correlation between several evidentiary factors and the inspection task is evaluated, and several task correlation results are output.

[0073] Finally, factor sensitivity is calculated based on several risk impacts and several task relevances, outputting several factor sensitivity values. The factor sensitivity is positively correlated with both risk impact and task relevance. In other words, the greater the impact of a particular evidentiary factor on operational risk and the closer its correlation with inspection tasks, the higher its factor sensitivity.

[0074] This step involves first assessing the impact of evidence factors on risk from a historical perspective, then combining task information to assess the relevance to the task, and finally integrating the two to obtain factor sensitivity. This process accurately identifies key evidence factors for assessing the operational risks of drones, laying the foundation for subsequent optimization of the Bayesian network structure and improvement of risk assessment efficiency.

[0075] Step S120 in the method provided in this application embodiment includes:

[0076] Based on the historical UAV inspection records of the mission area, and according to the aforementioned predictive attribute information, the impact of the aforementioned evidentiary factors on the operational risks of the UAV is assessed, and several risk impact degrees are output.

[0077] Based on the aforementioned predictive attribute information, the UAV's execution path, and the task attribute information, the correlation between the aforementioned evidentiary factors and the inspection task is evaluated, and several task correlation scores are output.

[0078] Factor sensitivity is calculated based on the aforementioned risk impact and task correlation, and several factor sensitivities are output. The factor sensitivity is positively correlated with the risk impact and task correlation.

[0079] In this embodiment of the application, in order to accurately identify the evidence factors that play a key role in the operational risks of drones, it is necessary to assess the risk impact and mission relevance of the evidence factors separately, and then calculate the sensitivity of the factors based on both, so as to ensure that core factors can be screened according to sensitivity in the future, thereby improving the pertinence and efficiency of risk assessment.

[0080] Specifically, based on the historical UAV inspection records of the mission area, combined with several predictive attribute information, the impact of several evidentiary factors on the operational risks of UAVs is assessed, and several risk impact values ​​are output.

[0081] Among them, the historical drone inspection records of the mission area include the types, frequency and severity of risk events caused by abnormal evidence factors under the same or similar environments and similar mission types in the past.

[0082] For example, historical drone inspection records show that when the predicted wind speed exceeds 6 m / s, the probability of the drone deviating from its flight path increases three times compared to normal wind speeds (2-4 m / s), and the proportion of mission interruptions due to deviation from the flight path reaches 40%. Based on this, it can be judged that wind speed, as an evidentiary factor, has a high degree of influence on operational risk. However, when the ground population density is below 100 people / square kilometer, even if a drone exhibits a slight abnormal attitude, the probability of causing a secondary accident is less than 1%. Therefore, the impact of ground population density on operational risk under this predictive attribute is relatively low.

[0083] During the evaluation process, for each evidence factor, the predictive attribute information needs to be matched with the risk data of the corresponding attribute range in the historical records. The risk impact of each evidence factor is obtained by weighting the probability of risk occurrence and the severity of consequences.

[0084] Furthermore, based on several predictive attribute information, the UAV's execution mission path, and mission attribute information, the correlation between several evidentiary factors and the inspection mission is evaluated to output several mission correlation scores.

[0085] In specific assessments, it is necessary to combine the predictive attribute information of the evidence factors with the key nodes of the task path and the core requirements of the task attributes. By judging whether the evidence factors directly support the achievement of the task objectives and whether they affect the execution of key aspects of the task, the task relevance can be quantified.

[0086] For example, the task relevance of "distance from buildings" is 0.9 in the inspection path of high-rise building clusters and 0.4 in the inspection path of suburban highways; the task relevance of "communication link stability" is 0.8 in emergency security inspections and 0.5 in routine inspections.

[0087] Finally, factor sensitivity is calculated based on several risk impacts and several task correlations, and several factor sensitivities are output.

[0088] Among them, since factor sensitivity is positively correlated with risk impact and task relevance, that is, the greater the impact of a certain evidence factor on risk and the closer its relevance to the task, the higher its sensitivity. In the calculation process, a linear weighted method can be used, substituting the risk impact and task relevance of each evidence factor into the formula "Factor Sensitivity = Risk Impact × Risk Impact Weight + Task Relevance × Task Relevance Weight" for calculation.

[0089] For example, if the risk impact weight is 0.5, the task relevance weight is 0.5, the risk impact of wind speed is 0.8, and the task relevance in the high-rise building cluster inspection task is 0.7, then the corresponding factor sensitivity = 0.8 × 0.5 + 0.7 × 0.5 = 0.75; the risk impact of "distance from the building" is 0.7, and the task relevance is 0.9, then the corresponding factor sensitivity = 0.7 × 0.5 + 0.9 × 0.5 = 0.8.

[0090] Furthermore, with a risk impact of ground population density of 0.3 and a task relevance of 0.4, the factor sensitivity is calculated as 0.3 × 0.5 + 0.4 × 0.5 = 0.35. This calculation method clearly distinguishes the importance of each evidentiary factor, providing a clear basis for subsequent selection of initial monitoring evidence factors based on sensitivity.

[0091] Through the above steps, the actual impact of each evidentiary factor on the operational risks of urban low-altitude drones was quantified, the close relationship between the evidentiary factors and the current task was clarified, and the resulting factor sensitivity can be used for subsequent screening of core factors to avoid interference from irrelevant factors, thus laying the foundation for improving the efficiency and accuracy of drone operation risk assessment.

[0092] S130: Set the monitoring scalar of evidence factors according to the importance of the inspection task in the preset time zone and the regional environmental fluctuation coefficient, and select multiple initial monitoring evidence factors according to the sensitivity of the several factors to construct an initial factor group;

[0093] In this embodiment of the application, in order to improve the efficiency of risk assessment while ensuring the accuracy of risk assessment, it is necessary to determine the appropriate number of evidence factors to monitor based on the importance of the inspection task and the regional environmental fluctuation coefficient, and then screen key evidence factors based on the sensitivity of the factors to construct an initial factor group.

[0094] Specifically, the importance of the inspection task is first assessed based on the task attribute information to obtain the task importance coefficient. For example, if the task is to conduct security inspections in the core business district of the city, its importance is high, and the corresponding task importance coefficient is large; if it is just an inspection of ordinary suburban areas, the task importance coefficient is relatively small.

[0095] Furthermore, the ratio of the historical minimum importance coefficient of the inspection task within the preset time range to the importance coefficient of the inspection task is set as the first factor adjustment coefficient; at the same time, the ratio of the historical minimum regional environmental fluctuation coefficient within the preset time range to the regional environmental fluctuation coefficient is set as the second factor adjustment coefficient.

[0096] Further, the average of the first factor adjustment coefficient and the second factor adjustment coefficient is calculated to obtain the overall factor adjustment coefficient. This overall factor adjustment coefficient is then multiplied by the number of evidence factors in the preset evidence factor space and rounded to obtain the evidence factor monitoring scalar.

[0097] Finally, based on the obtained sensitivity of several factors, the evidence factors are arranged in descending order of sensitivity to generate an evidence factor sequence. The evidence factors with the highest monitoring scalar value in the evidence factor sequence are selected as the initial monitoring evidence factors, thereby obtaining multiple initial monitoring evidence factors to construct an initial factor group.

[0098] At this time, when the inspection task is of high importance and the regional environmental fluctuation coefficient is large, the number of monitoring indicators can be reduced to increase the monitoring frequency to cope with rapidly changing risks, while avoiding increasing the computational burden due to too many monitoring indicators, thus ensuring the efficiency of risk assessment. On the other hand, when the inspection task is of low importance or the regional environmental fluctuation coefficient is small, key factors can be reasonably selected to ensure the accuracy of risk assessment.

[0099] Step S130 in the method provided in this application embodiment includes:

[0100] The importance coefficient of the inspection task is obtained by assessing the importance of the inspection task based on the task attribute information.

[0101] The ratio of the historical minimum inspection task importance coefficient within the preset time range to the inspection task importance coefficient is set as the first factor adjustment coefficient;

[0102] The ratio of the historical minimum regional environmental fluctuation coefficient within a preset time range to the regional environmental fluctuation coefficient is set as the second factor adjustment coefficient.

[0103] The overall factor adjustment coefficient is calculated based on the first factor adjustment coefficient and the second factor adjustment coefficient, and multiplied by the number of evidence factors in the preset evidence factor space and rounded down to obtain the evidence factor monitoring scalar.

[0104] Based on the sensitivity of the aforementioned factors, the aforementioned evidence factors are arranged in descending order of sensitivity to generate an evidence factor sequence. The evidence factor with the highest monitoring scalar value in the evidence factor sequence is selected as the initial monitoring evidence factor, and multiple initial monitoring evidence factors are obtained to construct an initial factor group.

[0105] In this embodiment of the application, in order to ensure the accuracy of the risk assessment of UAV operation while taking into account the assessment efficiency, it is necessary to dynamically determine the number of evidence factors that need to be monitored in key areas by combining the importance of the inspection task and the regional environmental fluctuation coefficient, and then select the core initial monitoring evidence factors based on the sensitivity of the factors, so as to ensure that the subsequent risk assessment focuses on key factors and improves the efficiency and accuracy of the risk assessment.

[0106] Specifically, the importance of inspection tasks is first assessed based on task attribute information to obtain an importance coefficient. Task attribute information includes the priority of the inspection task, the importance of the core areas involved, and the scope of impact of the task's execution consequences.

[0107] For example, if the inspection task is "security inspection of major events in the core urban area", the task priority is the highest level. The core area is crucial to the city's image and safety. If risks occur during the execution of the task, it may cause serious social impact and safety accidents. After quantitative evaluation, the importance coefficient of this inspection task can be set to 0.9. On the other hand, the routine inspection task of crop growth in suburban farmland has a low priority and a small impact range. The importance coefficient of the inspection task can be set to 0.3.

[0108] Furthermore, the ratio of the historical minimum inspection task importance coefficient to the inspection task importance coefficient within the preset time range is set as the first factor adjustment coefficient.

[0109] The preset time range can be selected as the past year. During this period, if the historical minimum inspection task importance coefficient is 0.2, when the current inspection task importance coefficient is 0.9, the first factor adjustment coefficient = 0.2 / 0.9≈0.22; if the current inspection task importance coefficient is 0.3, the first factor adjustment coefficient = 0.2 / 0.3≈0.67.

[0110] At this point, the first factor adjustment coefficient is used to reflect the relative importance of the current task compared to the historical least important task. The more important the task, the smaller the first factor adjustment coefficient, and the adjustment range used to calculate the monitoring scalar of evidence factors will also change accordingly.

[0111] Furthermore, the ratio of the historical minimum regional environmental fluctuation coefficient to the regional environmental fluctuation coefficient within the preset time range is set as the second factor adjustment coefficient.

[0112] Similarly, taking the past year as the preset time range, assuming the historical minimum regional environmental fluctuation coefficient is 0.1, if the current regional environmental fluctuation coefficient is 0.184, then the second factor adjustment coefficient = 0.1 / 0.184≈0.54; if the current regional environmental fluctuation coefficient is 0.12, then the second factor adjustment coefficient = 0.1 / 0.12≈0.83.

[0113] At this point, the adjustment coefficient of the second factor reflects the relative degree of fluctuation of the current regional environment compared with the most stable historical environment. The greater the environmental fluctuation, the smaller the adjustment coefficient of the second factor.

[0114] Furthermore, the overall factor adjustment coefficient is calculated based on the obtained first factor adjustment coefficient and second factor adjustment coefficient, multiplied by the number of evidence factors in the preset evidence factor space and rounded down to obtain the evidence factor monitoring scalar.

[0115] The overall factor adjustment coefficient can be calculated by averaging the adjustment coefficients of the first factor and the second factor. For example, if the adjustment coefficient of the first factor is 0.22 and the adjustment coefficient of the second factor is 0.54, the overall factor adjustment coefficient = (0.22 + 0.54) / 2 = 0.38; if the adjustment coefficient of the first factor is 0.67 and the adjustment coefficient of the second factor is 0.83, the overall factor adjustment coefficient = (0.67 + 0.83) / 2 = 0.75.

[0116] Furthermore, assuming that the number of evidence factors in the preset evidence factor space is 20, when the overall factor adjustment coefficient is 0.38, the evidence factor monitoring scalar = 0.38 × 20 ≈ 7.6, which is 8 after rounding; when the overall factor adjustment coefficient is 0.75, the evidence factor monitoring scalar = 0.75 × 20 = 15, which is 15 after rounding.

[0117] Therefore, the more important the task and the greater the environmental fluctuation, the smaller the overall factor adjustment coefficient and the smaller the monitoring scalar of evidence factors, in order to meet the requirement that "when the task is of high importance and the regional environmental fluctuation is large, it is necessary to reduce the monitoring indicators to improve the efficiency of risk assessment".

[0118] Finally, based on the obtained sensitivity of several factors, the evidence factors are arranged in descending order of sensitivity to generate an evidence factor sequence. The evidence factors with the highest monitoring scalar value in the evidence factor sequence are selected as the initial monitoring evidence factors, thereby obtaining multiple initial monitoring evidence factors to construct an initial factor group.

[0119] For example, if the scalar of evidence factors to be monitored is 8, the top 8 evidence factors in terms of sensitivity will be selected as the initial monitoring evidence factors, and then an initial factor group will be constructed. Subsequent risk assessments will be based on these 8 core evidence factors, which not only ensures coverage of high-risk and highly relevant factors, but also improves assessment efficiency due to the reduction in the number of monitoring factors.

[0120] Through the above steps, the number of monitoring factors can be dynamically adjusted according to the importance of the task and environmental fluctuations, and the most critical initial monitoring evidence factors can be selected, laying an efficient and accurate foundation for subsequent risk assessment based on Bayesian theory.

[0121] S140: When the UAV performs urban low-altitude security patrol and inspection tasks, it periodically monitors and obtains the initial factor attribute information set according to the initial factor group, and calls the adaptive risk assessment sub-network to conduct UAV operation risk assessment and output multi-dimensional operation risk probability.

[0122] In this embodiment of the application, in order to quickly and accurately obtain the real-time operational risk status of the drone, it is necessary to rely on the initial factor group constructed in the early stage for regular monitoring, and at the same time call the risk assessment sub-network adapted to it to carry out assessment, so as to efficiently output the operational risk probability covering multiple dimensions and provide a scientific basis for security decision-making.

[0123] Specifically, the first step is to periodically monitor and obtain the initial factor attribute information set according to the initial factor set. The initial factor set is a collection of evidence factors that are critical to the impact of drone operation risks and are highly correlated with the mission, such as wind speed, distance from buildings, and communication link stability.

[0124] During the process of drones performing urban low-altitude security patrols, it is necessary to collect real-time attribute information of each factor in the initial factor group according to the preset monitoring frequency.

[0125] For example, the monitoring frequency is set to once every 2 minutes. For the "wind speed" factor, the wind speed value of the current flight area is obtained once every 2 minutes through the wind speed sensor carried by the drone.

[0126] In addition, for the factor of "distance from buildings", the drone's lidar or visual sensors are used to continuously monitor and record the real-time distance to surrounding buildings; for the factor of "communication link stability", the attribute information is reflected by parameters such as signal strength and packet loss rate fed back in real time by the communication platform.

[0127] Furthermore, the attribute information of each initial factor obtained from each monitoring is summarized to form an initial factor attribute information set, providing real-time data input for subsequent risk assessment.

[0128] Furthermore, the adaptive risk assessment subnetwork is invoked to conduct drone operation risk assessment, so as to quickly perform inference calculations on real-time data in the initial factor attribute information set.

[0129] The method provided in this application embodiment, which involves "constructing a global risk assessment network and a risk assessment sub-network matching channel", includes:

[0130] Based on the historical UAV inspection records of the task area, sample data is collected under the constraints of the aforementioned evidence factors to obtain multiple global factor sets of samples, and the historical multidimensional operational risk probabilities under different global factor sets of samples are statistically analyzed to obtain multiple operational risk probability sets of samples.

[0131] Using the global factor set of multiple samples and the operational risk probability set of multiple samples as training data, a Bayesian network is trained to generate a global risk assessment network.

[0132] The global risk assessment network is pruned to obtain multiple risk assessment sub-networks, which are then combined to obtain a risk assessment sub-network matching channel.

[0133] In this embodiment of the application, in order to build a basic network architecture that can comprehensively and accurately assess the operational risks of drones, it is necessary to first build a global risk assessment network that covers various evidence factors and risk correlations, and then generate multiple targeted risk assessment sub-networks through edge trimming and form matching channels to meet the needs of rapid and accurate risk assessment under different combinations of monitoring factors.

[0134] Specifically, firstly, based on the historical UAV inspection records of the mission area, sample data is collected under several evidentiary factors to obtain multiple global factor sets of samples.

[0135] At the same time, the historical multidimensional operational risk probabilities under different global factor sets of samples are statistically analyzed to obtain multiple sample operational risk probability sets.

[0136] Among them, using evidence factors as constraints means selecting historical drone inspection records that contain complete evidence factor attribute information, and combining the attribute values ​​of all evidence factors in each record into a global sample factor set.

[0137] Simultaneously, the historical multidimensional operational risk probabilities corresponding to the global factor set of this sample are statistically analyzed, i.e., the probability of various risks occurring, thus forming a sample operational risk probability set. These multidimensional operational risks include flight safety risks, environmental adaptation risks, mission completion risks, system health risks, and safety incident propagation risks.

[0138] For example, in a historical drone inspection record, the wind speed is 4 m / s, the visibility is 10 km, the distance to a building is 50 meters, and the remaining battery life is 2 hours. These correspond to a flight safety risk probability of 0.1, an environmental adaptation risk probability of 0.05, and a mission execution risk probability of 0.08. This set of evidence factors and the sample operational risk probability set constitute a training data sample. By collecting a large number of similar samples, sufficient and realistic data support is provided for subsequent Bayesian network training.

[0139] Furthermore, a Bayesian network is trained using the obtained global factor set of multiple samples and the operational risk probability set of multiple samples as training data to generate a global risk assessment network.

[0140] Bayesian networks are a type of probabilistic graph model, composed of directed acyclic graphs (DAGs). Nodes represent random variables (i.e., evidence factors or operational risk events), and directed edges represent conditional dependencies. Each node is equipped with a conditional probability table (CPT). It can establish causal / dependency relationships between evidence factors and operational risks in uncertain environments, thereby enabling reasoning, prediction, and decision-making processes.

[0141] Specifically, when training a Bayesian network, the first steps are structure learning and parameter learning to clarify the network structure and the probability distribution of each node. Structure learning aims to determine the directed dependencies between evidentiary factors and operational risk events, such as determining whether the "wind speed" node points to the "flight safety risk" node to indicate whether changes in wind speed affect flight safety risks, or whether there are other directions of dependency. Parameter learning, on the other hand, calculates the probability distribution of each node given its parent node, generating a conditional probability table for each node.

[0142] Furthermore, when using sample data for training, taking the "wind speed" node and the "flight safety risk" node as examples, the wind speed data in the sample is first divided into intervals, such as low wind speed (≤3m / s), medium wind speed (3-6m / s), and high wind speed (>6m / s) intervals.

[0143] Simultaneously, the probability of flight safety risks occurring in each wind speed range is calculated. It is assumed that the probability of flight safety risk occurring is 0.05 in the low wind speed range, 0.15 in the medium wind speed range, and 0.3 in the high wind speed range. Through similar statistical analysis, the conditional probability distribution of "wind speed" on "flight safety risk" is determined, thus refining the conditional probability table for the "flight safety risk" node.

[0144] Furthermore, for cases involving multiple parent nodes, such as the "task execution risk" node, whose possible parent nodes include "communication link stability" and "battery endurance," it is necessary to statistically analyze the probability of task execution risk occurring under different combinations of communication link stability states and battery endurance ranges, and then construct a conditional probability table for this node.

[0145] For example, if the stability of the communication link is divided into three states: "stable", "slightly fluctuating" and "severely fluctuating", the battery life is divided into three ranges: "sufficient (remaining battery life ≥ 90 minutes)", "medium (remaining battery life 60-90 minutes)" and "insufficient (remaining battery life < 60 minutes)".

[0146] Statistical data shows that the probability of task execution risk is 0.08 when the communication link is stable and the battery life is sufficient; 0.15 when the communication link is stable but the battery life is moderate; 0.25 when the communication link is stable but the battery life is insufficient; 0.12 when the communication link fluctuates slightly and the battery life is sufficient; 0.2 when the communication link fluctuates slightly and the battery life is moderate; 0.3 when the communication link fluctuates slightly and the battery life is insufficient; 0.2 when the communication link fluctuates severely and the battery life is sufficient; 0.35 when the communication link fluctuates severely and the battery life is moderate; and 0.5 when the communication link fluctuates severely and the battery life is insufficient. Based on these statistical results, a conditional probability table for the "task execution risk" node can be constructed.

[0147] Ultimately, through structural and parameter learning of all sample data, the Bayesian network can fully capture the probabilistic correlation between evidence factors and various operational risks. The resulting global risk assessment network can cover the complex correlation between all evidence factors and multidimensional operational risks, and can infer the corresponding multidimensional operational risk probability based on any combination of evidence factors.

[0148] Furthermore, the global risk assessment network is pruned to obtain multiple risk assessment subnetworks, which are then combined to obtain a risk assessment subnetwork matching channel.

[0149] In the method provided in this application embodiment, "performing edge pruning on the global risk assessment network to obtain multiple risk assessment sub-networks, and combining them to obtain a risk assessment sub-network matching channel" includes:

[0150] Based on the aforementioned evidence factors, the factors are arranged and combined, and the power set of the aforementioned evidence factors is subtracted from the empty set to obtain multiple sample factor groups, wherein each sample factor group includes at least one evidence factor.

[0151] Randomly select the first sample factor group, and prune the edges of other evidence factors in the global risk assessment network except for the first sample factor group to obtain the first risk assessment sub-network, and then perform the operation in sequence to obtain several risk assessment sub-networks.

[0152] The risk assessment subnetwork matching channel is obtained by combining the multiple sample factor groups and the several risk assessment subnetwork mappings.

[0153] In this embodiment of the application, in order to further improve the real-time performance and efficiency of drone operation risk assessment in different scenarios, the global risk assessment network needs to be pruned to generate multiple targeted risk assessment sub-networks, which are then combined to form a matching channel to adapt to the rapid risk assessment needs under different combinations of evidence factors.

[0154] Specifically, firstly, several evidentiary factors are arranged and combined, and then the power set of several evidentiary factors is subtracted from the empty set to obtain multiple sample factor groups, and each sample factor group includes at least one evidentiary factor.

[0155] For example, if there are three evidentiary factors: A (wind speed), B (visibility), and C (electromagnetic interference intensity), then the sample factor set obtained by subtracting the empty set from their power set is {A}, {B}, {C}, {A, B}, {A, C}, {B, C}, {A, B, C}. Through such permutations and combinations, all possible combinations of evidentiary factors can be covered.

[0156] Furthermore, a first sample factor group is randomly selected, and the edges of other evidence factors in the global risk assessment network, excluding the first sample factor group, are pruned to obtain a first risk assessment subnetwork. This process is repeated to obtain several risk assessment subnetworks.

[0157] For example, when a sample factor group {A, B} is selected, the edges related to evidence factor C will be pruned in the global risk assessment network, and only the nodes related to evidence factors A and B and the edges between these nodes will be retained, thereby forming a first risk assessment sub-network that focuses on the relationship between factors A and B and risk.

[0158] Furthermore, the sample factor group {A, C} is selected, and the edges related to evidence factor B in the global network are pruned to obtain the second risk assessment subnetwork.

[0159] This process of edge pruning is repeated for each sample factor group, resulting in several risk assessment subnetworks. Each subnetwork focuses only on the key correlations between certain evidentiary factors and risk events, making the network structure simpler than the global network.

[0160] Finally, a risk assessment subnetwork matching channel is obtained based on the mapping combination of multiple sample factor groups and several risk assessment subnetworks. This means establishing a one-to-one mapping relationship between each sample factor group and its corresponding risk assessment subnetwork, and equipping each evidence factor combination with a dedicated rapid assessment tool.

[0161] Furthermore, when monitoring data for a combination of certain evidence factors is obtained, the corresponding risk assessment subnetwork can be quickly found through the matching channel. The risk assessment can be carried out using the subnetwork, which not only ensures the relevance of the assessment but also greatly improves the reasoning speed and meets the real-time requirements.

[0162] For example, suppose the global risk assessment network includes four evidence factor nodes: wind speed, visibility, electromagnetic interference intensity, and battery endurance, as well as risk nodes such as flight safety risk and mission execution risk. The nodes are connected by directed edges, reflecting complex probabilistic dependencies.

[0163] Following the above method, multiple sample factor groups are obtained by arranging and combining the evidence factors, such as {wind speed, visibility} and {electromagnetic interference intensity, battery endurance}. Selecting the sample factor group {wind speed, visibility}, edges related to electromagnetic interference intensity and battery endurance in the global network are pruned, resulting in a risk assessment subnetwork that retains only the key edges between wind speed / visibility nodes and flight safety risk / mission execution risk nodes.

[0164] Furthermore, a sample factor group {electromagnetic interference intensity, battery life} is selected, and edges related to wind speed and visibility in the global network are pruned to obtain another risk assessment subnetwork. These sample factor groups are then mapped to their corresponding subnetworks to form a risk assessment subnetwork matching channel.

[0165] Through the above steps, multiple simplified risk assessment sub-networks were extracted from the complex global risk assessment network, and a matching channel was built, providing strong support for rapid assessment of drone operation risks in different scenarios.

[0166] Furthermore, based on the obtained initial factor attribute information set, the combinations of evidentiary factors contained therein, such as {wind speed, visibility}, are analyzed. Simultaneously, using the risk assessment subnetwork matching channel, the risk assessment subnetwork mapping to the sample factor group {wind speed, visibility} is quickly located. After invoking this adapted risk assessment subnetwork, real-time data on wind speed and visibility from the initial factor attribute information set are input into the risk assessment subnetwork.

[0167] At this point, the risk assessment subnetwork uses the key edges between the wind speed and visibility nodes and the flight safety risk and mission execution risk nodes that it retains, as well as the probabilistic dependencies obtained through training, to infer and calculate the operational risks of the drone.

[0168] Ultimately, the system outputs multi-dimensional operational risk probabilities. For example, the flight safety risk probability might be 0.12, meaning that the drone has a 12% chance of experiencing a flight safety issue; the mission execution risk probability might be 0.09, meaning that the probability of a problem occurring during mission execution is 9%, and so on.

[0169] The multidimensional operational risk probabilities obtained through the above steps comprehensively present the current operational risk status of the drone, providing accurate basis for risk decisions during drone operation, such as whether to adjust the flight path or take emergency measures.

[0170] S150: If the multidimensional operational risk probability does not meet the preset initial risk threshold set, the global factor attribute information set is obtained by monitoring the several evidence factors, and the global risk assessment network is called to accurately assess the operational risk of the UAV.

[0171] In this embodiment of the application, in order to avoid misjudgment of risks due to simplified assessment and to ensure that the specific types and severity of current operational risks can be fully and accurately identified, it is necessary to further expand the monitoring scope to obtain more complete factor data, and to carry out in-depth assessment with the help of a global risk assessment network covering all relationships, so as to provide accurate and reliable risk basis for subsequent emergency decision-making.

[0172] The urban low-altitude UAV operation risk assessment method based on Bayesian theory provided in this application also includes:

[0173] Configure a preset initial risk threshold set, wherein the preset initial risk threshold set includes several initial risk thresholds, and each initial risk threshold corresponds one-to-one with the probability of operational risk;

[0174] The risk indicator proportion is set according to the importance coefficient of the inspection task and the regional environmental fluctuation coefficient, wherein the risk indicator proportion is negatively correlated with the importance coefficient of the inspection task and the regional environmental fluctuation coefficient.

[0175] The multidimensional operational risk probability is mapped and judged according to the preset initial risk threshold set. If the operational risk probability exceeds the corresponding initial risk threshold, an indicator overflow is recorded. If the indicator overflow ratio exceeds the risk indicator ratio, it is determined that the multidimensional operational risk probability does not meet the preset initial risk threshold set.

[0176] Specifically, when determining whether the multidimensional operational risk probability meets safety requirements, a preset initial risk threshold set must first be configured. This preset initial risk threshold set contains several initial risk thresholds, and each initial risk threshold corresponds one-to-one with the corresponding operational risk probability dimension, ensuring that different types of risks have clear safety judgment criteria.

[0177] Among them, the setting of the preset initial risk threshold set should be closely combined with the mission safety requirements and historical risk data. That is, it should refer to the probability thresholds of safety accidents caused by different risk dimensions in the past drone urban low-altitude security patrols, as well as the specific requirements of the current mission for safety performance, so that the threshold of each risk dimension has practical guiding significance.

[0178] For example, initial risk thresholds are configured for the three core risk dimensions of flight safety risk, mission execution risk, and equipment failure risk: the initial risk threshold for flight safety risk is set to 0.15 based on historical accident data; the initial risk threshold for mission execution risk is set to 0.12 considering the requirements of security patrol for mission completion; and the initial risk threshold for equipment failure risk is set to 0.1 based on the equipment reliability test results.

[0179] Furthermore, if the flight safety risk probability obtained through the risk assessment sub-network is 0.2, the mission execution risk probability is 0.1, and the equipment failure risk probability is 0.08, subsequent judgments can be made based on the aforementioned initial risk thresholds.

[0180] Furthermore, the risk indicator ratio is set according to the importance coefficient of the inspection task and the regional environmental fluctuation coefficient. This risk indicator ratio is negatively correlated with the importance coefficient of the inspection task and the regional environmental fluctuation coefficient. That is, when the inspection task is of high importance and the regional environment fluctuates greatly, the risk indicator ratio needs to be set lower, which means that the tolerance for risk spillover is more stringent in order to match the high safety requirements. Conversely, the risk indicator ratio of routine inspection tasks and environmentally stable areas can be appropriately increased to avoid overreaction affecting inspection efficiency.

[0181] Furthermore, the multidimensional operational risk probabilities are mapped and judged one by one according to the preset initial risk threshold set: the flight safety risk probability of 0.2 exceeds the corresponding initial risk threshold of 0.15, and an index overflow needs to be recorded; the task execution risk probability of 0.1 does not exceed 0.12 and the equipment failure risk probability of 0.08 does not exceed 0.1, so there is no index overflow in these two dimensions.

[0182] At this point, the overflow ratio of the indicators is calculated using the formula "Indicator overflow ratio = Number of overflow dimensions / Total number of risk dimensions", i.e., 1 (flight safety risk overflow) / 3 (total risk dimensions) ≈ 33%. If the risk indicator ratio previously set based on the mission importance coefficient and the regional environmental fluctuation coefficient was 25%, since 33% > 25%, it is determined that the current multi-dimensional operational risk probability does not meet the preset initial risk threshold set, and further global factor monitoring and global risk assessment network calls are needed to achieve accurate assessment.

[0183] Furthermore, a global factor attribute information set is obtained by monitoring several evidence factors. These evidence factors refer to all evidence factors within a pre-defined evidence factor space, including the initial monitoring evidence factors selected in the early stage, as well as other factors not included in the initial factor group.

[0184] During the monitoring process, real-time attribute information of all evidence factors should be collected synchronously at the same monitoring frequency as the initial factor group. For example, precipitation probability data (currently 10%) can be supplemented by meteorological sensors, and the real-time distance between temporary ground obstacles and drones (currently 35 meters) can be obtained by millimeter-wave radar.

[0185] Furthermore, the real-time attribute information of all the obtained evidence factors is aggregated to form a global factor attribute information set to ensure that all key data that may affect the risk assessment are covered without omission.

[0186] Finally, this set of global factor attribute information is input into the global risk assessment network that was previously trained based on historical inspection records of the task area.

[0187] Specifically, the global risk assessment network is based on Bayesian inference algorithms, which fully utilize the prior probabilities of each node (including all evidence factor nodes and multidimensional risk nodes) and the conditional probabilities between nodes. Finally, it outputs more accurate multidimensional operational risk probabilities that are more in line with the current actual operation scenario. For example, the specific probability values ​​of each dimension such as flight safety risk, mission execution risk, and equipment failure risk provide data support for the subsequent formulation of targeted risk response strategies.

[0188] For example, in the scenario of security patrol in the core business district of a city from 8:00 to 11:00 am, assuming that through monitoring and sub-network evaluation of the initial factor set (eight factors such as wind speed, distance from buildings, and communication link stability), the multi-dimensional operational risk probabilities are obtained: flight safety risk 0.18 (initial risk threshold 0.15), mission execution risk 0.09 (initial risk threshold 0.12), and equipment failure risk 0.07 (initial risk threshold 0.1). The overflow ratio of the indicators is 1 / 3 ≈ 33%, while the current risk indicator ratio is set at 25%, so it is determined that the initial risk threshold set is not met.

[0189] Furthermore, real-time data of all evidence factors were added to the monitoring: precipitation probability 8%, ground population density 190 people / square kilometer, drone battery temperature 38°C, operator concentration score 92 points, etc., to form a global factor attribute information set.

[0190] Simultaneously, the global risk assessment network was invoked. Through comprehensive analysis, it was found that "wind speed of 3.5 m / s + 8% probability of precipitation" resulted in a 5% decrease in drone lift, and "drone battery temperature of 38℃" reduced the power system output efficiency by 3%. The combination of these two factors further increased the flight safety risk. The final output was a corrected multi-dimensional operational risk probability: flight safety risk 0.21, mission execution risk 0.10, and equipment failure risk 0.08. The core influencing factors of flight safety risk were identified as "wind speed + probability of precipitation" (contribution of 70%), providing operators with a precise basis for formulating emergency measures such as "reducing the flight altitude by 10 meters to reduce the impact of airflow and activating the battery cooling mode".

[0191] Through the above steps, a rapid, comprehensive, and accurate risk assessment can be conducted when simplified assessments cannot meet safety requirements. This ensures the efficiency of routine inspections and provides in-depth risk analysis when risks are imminent, thus guaranteeing the safe execution of drone-based urban low-altitude security inspection missions.

[0192] The embodiments of this application, through the specific implementation methods described above, achieve the following technical effects:

[0193] This application proposes a Bayesian-based method for assessing the operational risks of urban low-altitude unmanned aerial vehicles (UAVs). First, before the UAV performs urban low-altitude security patrols, a pre-defined evidence factor space is constructed, encompassing environmental, spatial, performance, and operational factors. Combining task path information, task attribute information, and historical UAV patrol records, the attributes of the evidence factors are predicted, and the regional environmental fluctuation coefficient is assessed. Then, based on the predicted attribute information, the sensitivity of the factors is calculated from both risk impact and task relevance perspectives. Combining task importance and the environmental fluctuation coefficient, the monitoring scalar of the evidence factors is determined, and core factors are selected to construct an initial factor group. During task execution, the initial factor group is periodically monitored, and an adapted risk assessment subnetwork is invoked to output multi-dimensional operational risk probabilities. If the probabilities do not meet a pre-defined threshold, global factors are further monitored, and a global risk assessment network is invoked to conduct a precise assessment.

[0194] The Bayesian theory-based risk assessment method for urban low-altitude drone operations provided in this application solves the problems of incomplete factor coverage, poor model adaptability, and rigid risk judgment in traditional drone risk assessment through a hierarchical and progressive technical solution. It provides a scientific, efficient, and accurate risk assessment solution for urban low-altitude drone security patrol tasks, effectively reducing the probability of risk misjudgment and omission, and providing strong technical support for the safe operation of drones. It can be applied to various low-altitude drone operation scenarios such as urban security and emergency patrol.

[0195] Example 2, as shown in the appendix Figure 2As shown, based on the inventive concept of the Bayesian theory-based urban low-altitude UAV operation risk assessment method provided in Embodiment 1, this application also provides a Bayesian theory-based urban low-altitude UAV operation risk assessment system, specifically including:

[0196] The Predictive Attributes and Fluctuation Assessment Module 01 is used to obtain the predictive attribute information of the preset evidence factors space in the preset time zone of the task area before the UAV performs urban low-altitude security patrol mission, and to assess and determine the environmental fluctuation coefficient of the predicted area.

[0197] The factor sensitivity assessment module 02 is used to assess the sensitivity of several evidence factors to the operational risk of the UAV based on the predicted attribute information, and output the sensitivity of several factors.

[0198] The initial factor group construction module 03 is used to set the evidence factor monitoring scalar based on the importance of the inspection task in the preset time zone and the regional environmental fluctuation coefficient, and to screen multiple initial monitoring evidence factors based on the sensitivity of the several factors to construct an initial factor group.

[0199] The sub-network risk assessment module 04 is used to periodically monitor and obtain the initial factor attribute information set according to the initial factor group when the UAV performs urban low-altitude security patrol and inspection tasks, and call the adaptive risk assessment sub-network to conduct UAV operation risk assessment and output multi-dimensional operation risk probability.

[0200] The global network risk assessment module 05 is used to obtain a global factor attribute information set according to the monitoring of the several evidence factors if the multidimensional operational risk probability does not meet the preset initial risk threshold set, and to call the global risk assessment network to conduct a precise assessment of the UAV operational risk.

[0201] In one embodiment, the predictive attribute and volatility assessment module 01 is further used for:

[0202] A preset evidence factor space is obtained, comprising an environmental factor set, a spatial factor set, a performance factor set, and an operational factor set, with each factor set including multiple evidence factors. Based on the preset evidence factor space, and combining the UAV's execution mission path, mission attribute information, and historical UAV inspection records of the mission area, attribute information prediction is performed on several evidence factors within a preset time zone to obtain several predicted attribute information for the several evidence factors. Environmental type predicted attribute information is extracted from the several predicted attribute information, wherein the environmental type predicted attribute information includes multiple predicted environmental parameter sequences. Environmental volatility is assessed based on the multiple predicted environmental parameter sequences to obtain multiple environmental volatility degrees, which are then weighted and fused to obtain the predicted regional environmental volatility coefficient.

[0203] In one embodiment, the factor sensitivity assessment module 02 is further used for:

[0204] Based on historical UAV inspection records of the task area, the impact of several predictive attribute information on the operational risks of the UAV is evaluated, and several risk impact degrees are output. Based on the several predictive attribute information, the UAV's execution task path, and task attribute information, the correlation between the several evidentiary factors and the inspection task is evaluated, and several task correlation degrees are output. Based on the several risk impact degrees and several task correlation degrees, factor sensitivity is calculated, and several factor sensitivity degrees are output. Among them, the factor sensitivity is positively correlated with the risk impact degree and the task correlation degree.

[0205] In one embodiment, the initial factor group construction module 03 is further configured to:

[0206] The importance coefficient of the inspection task is obtained by assessing the importance of the inspection task according to the task attribute information; the ratio of the historical minimum importance coefficient of the inspection task within a preset time range to the inspection task importance coefficient is set as the first factor adjustment coefficient; the ratio of the historical minimum regional environmental fluctuation coefficient within a preset time range to the regional environmental fluctuation coefficient is set as the second factor adjustment coefficient; the overall factor adjustment coefficient is calculated based on the first factor adjustment coefficient and the second factor adjustment coefficient, and multiplied by the number of evidence factors in the preset evidence factor space and rounded to obtain the evidence factor monitoring scalar; based on the sensitivity of the several factors, the several evidence factors are arranged in descending order of factor sensitivity to generate an evidence factor sequence, and the evidence factor with the highest evidence factor monitoring scalar in the evidence factor sequence is selected as the initial monitoring evidence factor, thus obtaining multiple initial monitoring evidence factors to construct an initial factor group.

[0207] In one embodiment, the sub-network risk assessment module 04 further includes:

[0208] Based on historical UAV inspection records of the task area, sample data is collected under the constraints of the aforementioned evidence factors to obtain multiple global factor sets. The historical multidimensional operational risk probabilities under different global factor sets are statistically analyzed to obtain multiple operational risk probability sets. The multiple global factor sets and multiple operational risk probability sets are used as training data to train a Bayesian network and generate a global risk assessment network. The global risk assessment network is then pruned to obtain multiple risk assessment subnetworks, which are combined to obtain a risk assessment subnetwork matching channel.

[0209] Furthermore, the sub-network risk assessment module 04 also includes:

[0210] Based on the aforementioned evidence factors, factors are arranged and combined. The power set of the aforementioned evidence factors is subtracted from the empty set to obtain multiple sample factor groups, wherein each sample factor group includes at least one evidence factor. A first sample factor group is randomly selected, and the edges of other evidence factors in the global risk assessment network, excluding the first sample factor group, are pruned to obtain a first risk assessment sub-network. This process is repeated to obtain multiple risk assessment sub-networks. A risk assessment sub-network matching channel is obtained by mapping and combining the multiple sample factor groups and the multiple risk assessment sub-networks.

[0211] In one embodiment, the urban low-altitude drone operation risk assessment system based on Bayesian theory further includes:

[0212] Configure a preset initial risk threshold set, wherein the preset initial risk threshold set includes several initial risk thresholds, and each initial risk threshold corresponds one-to-one with the operational risk probability; set the risk indicator proportion according to the inspection task importance coefficient and the regional environmental fluctuation coefficient, wherein the risk indicator proportion is negatively correlated with the inspection task importance coefficient and the regional environmental fluctuation coefficient; map and judge the multidimensional operational risk probability according to the preset initial risk threshold set; if the operational risk probability exceeds the corresponding initial risk threshold, record an indicator overflow; if the indicator overflow ratio exceeds the risk indicator proportion, determine that the multidimensional operational risk probability does not meet the preset initial risk threshold set.

[0213] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this specification. Additionally, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some implementations, multitasking and parallel processing are possible or may be advantageous.

[0214] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

[0215] This specification and accompanying drawings are merely illustrative examples of this application and are intended to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from its scope. Therefore, if such modifications and variations fall within the scope of this application and its equivalents, this application intends to include such modifications and variations.

Claims

1. A risk assessment method for urban low-altitude unmanned aerial vehicle (UAV) operations based on Bayesian theory, characterized in that... The methods include: Before drones perform urban low-altitude security patrols, predictive attribute information of pre-defined evidence factors in the mission area within a pre-defined time zone is obtained, and the environmental fluctuation coefficient of the predicted area is assessed and determined. Based on the predicted attribute information, the sensitivity of several evidence factors to the operational risks of drones is evaluated, and the sensitivity of several factors is output. Based on the importance of the inspection task within the preset time zone and the regional environmental fluctuation coefficient, a monitoring scalar of evidence factors is set, and multiple initial monitoring evidence factors are selected based on the sensitivity of the factors to construct an initial factor group. When the drone performs urban low-altitude security patrol and inspection tasks, the initial factor attribute information set is periodically monitored and obtained according to the initial factor group, and the adaptive risk assessment sub-network is called to conduct drone operation risk assessment and output multi-dimensional operation risk probability. If the multidimensional operational risk probability does not meet the preset initial risk threshold set, the global factor attribute information set is obtained by monitoring the several evidence factors, and the global risk assessment network is called to accurately assess the operational risk of the UAV. The methods for constructing the matching channel between the global risk assessment network and the risk assessment sub-network include: Based on the historical UAV inspection records of the task area, sample data is collected under the constraints of the aforementioned evidence factors to obtain multiple global factor sets of samples, and the historical multidimensional operational risk probabilities under different global factor sets of samples are statistically analyzed to obtain multiple operational risk probability sets of samples. Using the global factor set of multiple samples and the operational risk probability set of multiple samples as training data, a Bayesian network is trained to generate a global risk assessment network. The global risk assessment network is pruned to obtain multiple risk assessment sub-networks, which are then combined to form a risk assessment sub-network matching channel, including: Based on the aforementioned evidence factors, the factors are arranged and combined, and the power set of the aforementioned evidence factors is subtracted from the empty set to obtain multiple sample factor groups, wherein each sample factor group includes at least one evidence factor. Randomly select the first sample factor group, and prune the edges of other evidence factors in the global risk assessment network except for the first sample factor group to obtain the first risk assessment sub-network, and then perform the operation in sequence to obtain several risk assessment sub-networks. The risk assessment subnetwork matching channel is obtained by combining the multiple sample factor groups and the several risk assessment subnetwork mappings.

2. The urban low-altitude unmanned aerial vehicle (UAV) operation risk assessment method based on Bayesian theory according to claim 1, characterized in that, Acquire the predictive attribute information of the task area within a preset time zone and a preset evidence factor space, and evaluate and determine the environmental fluctuation coefficient of the prediction area, including: Obtain a preset evidence factor space, wherein the preset evidence factor space includes an environmental factor set, a spatial factor set, a performance factor set, and an operational factor set, and each factor set includes multiple evidence factors; Based on the preset evidence factor space, combined with the UAV execution mission path, mission attribute information and historical UAV inspection records of the mission area, attribute information prediction is performed on several evidence factors within the preset time zone to obtain several predicted attribute information of the several evidence factors. Extract the environment type prediction attribute information from the plurality of prediction attribute information, wherein the environment type prediction attribute information includes multiple prediction environment parameter sequences; Environmental volatility is assessed based on the multiple predicted environmental parameter sequences, multiple environmental volatility values ​​are obtained, and weighted fusion is performed to obtain the predicted regional environmental volatility coefficient.

3. The urban low-altitude unmanned aerial vehicle (UAV) operation risk assessment method based on Bayesian theory according to claim 2, characterized in that, Based on the predicted attribute information, the sensitivity of several evidentiary factors to the operational risks of drones is evaluated, and the sensitivity of several factors is output, including: Based on the historical UAV inspection records of the mission area, and according to the aforementioned predictive attribute information, the impact of the aforementioned evidentiary factors on the operational risks of the UAV is assessed, and several risk impact degrees are output. Based on the aforementioned predictive attribute information, the UAV's execution path, and the task attribute information, the correlation between the aforementioned evidentiary factors and the inspection task is evaluated, and several task correlation scores are output. Factor sensitivity is calculated based on the aforementioned risk impact and task correlation, and several factor sensitivities are output. The factor sensitivity is positively correlated with the risk impact and task correlation.

4. The urban low-altitude unmanned aerial vehicle (UAV) operation risk assessment method based on Bayesian theory according to claim 3, characterized in that, Based on the importance of the inspection task within the preset time zone and the regional environmental fluctuation coefficient, a monitoring scalar of evidence factors is set. Multiple initial monitoring evidence factors are selected based on the sensitivity of these factors to construct an initial factor group, including: The importance coefficient of the inspection task is obtained by assessing the importance of the inspection task based on the task attribute information. The ratio of the historical minimum inspection task importance coefficient within the preset time range to the inspection task importance coefficient is set as the first factor adjustment coefficient; The ratio of the historical minimum regional environmental fluctuation coefficient within a preset time range to the regional environmental fluctuation coefficient is set as the second factor adjustment coefficient. The overall factor adjustment coefficient is calculated based on the first factor adjustment coefficient and the second factor adjustment coefficient, and multiplied by the number of evidence factors in the preset evidence factor space and rounded down to obtain the evidence factor monitoring scalar. Based on the sensitivity of the aforementioned factors, the aforementioned evidence factors are arranged in descending order of sensitivity to generate an evidence factor sequence. The evidence factor with the highest monitoring scalar value in the evidence factor sequence is selected as the initial monitoring evidence factor, and multiple initial monitoring evidence factors are obtained to construct an initial factor group.

5. The urban low-altitude unmanned aerial vehicle (UAV) operation risk assessment method based on Bayesian theory according to claim 1, characterized in that, The method further includes: Configure a preset initial risk threshold set, wherein the preset initial risk threshold set includes several initial risk thresholds, and each initial risk threshold corresponds one-to-one with the probability of operational risk; The risk indicator proportion is set according to the importance coefficient of the inspection task and the regional environmental fluctuation coefficient, wherein the risk indicator proportion is negatively correlated with the importance coefficient of the inspection task and the regional environmental fluctuation coefficient. The multidimensional operational risk probability is mapped and judged according to the preset initial risk threshold set. If the operational risk probability exceeds the corresponding initial risk threshold, an indicator overflow is recorded. If the indicator overflow ratio exceeds the risk indicator ratio, it is determined that the multidimensional operational risk probability does not meet the preset initial risk threshold set.

6. A risk assessment system for urban low-altitude unmanned aerial vehicle (UAV) operations based on Bayesian theory, characterized in that, The system is used to execute the urban low-altitude unmanned aerial vehicle (UAV) operation risk assessment method based on Bayesian theory as described in any one of claims 1-5, and the system includes: The prediction attribute and fluctuation assessment module is used to obtain the prediction attribute information of the preset evidence factor space in the preset time zone of the task area before the UAV performs urban low-altitude security patrol mission, and to assess and determine the environmental fluctuation coefficient of the prediction area. The factor sensitivity assessment module is used to assess the sensitivity of several evidence factors to the operational risks of the UAV based on the predicted attribute information, and output several factor sensitivities. The initial factor group construction module is used to set the monitoring scalar of evidence factors according to the importance of the inspection task in the preset time zone and the regional environmental fluctuation coefficient, and to screen multiple initial monitoring evidence factors according to the sensitivity of the several factors to construct an initial factor group. The sub-network risk assessment module is used to periodically monitor and obtain the initial factor attribute information set according to the initial factor group when the UAV performs urban low-altitude security patrol and inspection tasks, and call the adaptive risk assessment sub-network to conduct UAV operation risk assessment and output multi-dimensional operation risk probability. The global network risk assessment module is used to obtain a global factor attribute information set according to the several evidence factors if the multidimensional operational risk probability does not meet the preset initial risk threshold set, and call the global risk assessment network to accurately assess the operational risk of the UAV. The methods for constructing the matching channel between the global risk assessment network and the risk assessment sub-network include: Based on the historical UAV inspection records of the task area, sample data is collected under the constraints of the aforementioned evidence factors to obtain multiple global factor sets of samples, and the historical multidimensional operational risk probabilities under different global factor sets of samples are statistically analyzed to obtain multiple operational risk probability sets of samples. Using the global factor set of multiple samples and the operational risk probability set of multiple samples as training data, a Bayesian network is trained to generate a global risk assessment network. The global risk assessment network is pruned to obtain multiple risk assessment sub-networks, which are then combined to form a risk assessment sub-network matching channel, including: Based on the aforementioned evidence factors, the factors are arranged and combined, and the power set of the aforementioned evidence factors is subtracted from the empty set to obtain multiple sample factor groups, wherein each sample factor group includes at least one evidence factor. Randomly select the first sample factor group, and prune the edges of other evidence factors in the global risk assessment network except for the first sample factor group to obtain the first risk assessment sub-network, and then perform the operation in sequence to obtain several risk assessment sub-networks. The risk assessment subnetwork matching channel is obtained by combining the multiple sample factor groups and the several risk assessment subnetwork mappings.

Images