Urban low-altitude unmanned aerial vehicle risk assessment method based on LES-observation data

Abstract

The application provides a city low-altitude unmanned aerial vehicle risk assessment method based on LES-observation data, and comprises the following steps: step 1, constructing a prior wind field database K g ; step 2, collecting observation data y in real time; step 3, based on the set of the prior wind field database K g matched with the current time t, the observation data y of the current time t, generating the set of posterior wind field X a by using a set Kalman filtering algorithm, and calculating the average posterior wind field X aa ; step 4, flight path risk evolution prediction; step 5, calculating the posterior wind field distribution diagram corresponding to each average posterior field X aa ; step 6, flight path risk assessment; and step 7, cyclically executing steps 2-6 to realize dynamic risk assessment. The application realizes dynamic risk assessment on the wind field on the flight path of the unmanned aerial vehicle.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of unmanned aerial vehicle flight safety, and particularly relates to a city low-altitude unmanned aerial vehicle risk assessment method based on LES-observation data. BACKGROUND

[0002] With the rapid development of urban air mobility (UAM) and unmanned aerial vehicle delivery business, the unmanned aerial vehicle flight safety problem in the city low-altitude environment is increasingly prominent. The buildings in the city environment can produce complex turbulent fields, including building wake, street canyon effect, heat island effect, etc., which can threaten the flight stability and safety of the unmanned aerial vehicle.

[0003] Defects and deficiencies of the prior art: 1. The traditional wind field prediction method is mostly based on a mesoscale meteorological model, and the spatial resolution is insufficient to capture the city microscale turbulent characteristics. 2. The real-time observation data is sparse, and it is difficult to accurately reflect the wind field distribution of the entire city space. 3. There is a lack of effective fusion method of high-precision simulation and real-time observation. 4. The risk assessment is mostly static analysis, and it is difficult to adapt to the dynamic changes of the city wind field. SUMMARY

[0004] The purpose of the application is to provide a city low-altitude unmanned aerial vehicle risk assessment method based on LES-observation data, which faces the application scenarios of unmanned aerial vehicle logistics and distribution, emergency rescue, etc. in the city complex building environment, and realizes the dynamic assessment of the wind field risk on the flight path of the unmanned aerial vehicle.

[0005] The application is based on massive large eddy simulation (LES) to generate a typical risk distribution of city airway, and combines discrete observation point data to infer the real-time wind field state, and then assesses the flight risk of the unmanned aerial vehicle.

[0006] A city low-altitude unmanned aerial vehicle risk assessment method based on LES-observation data, comprising the following steps: constructing a city wind field database based on large eddy simulation, collecting distributed observation point data in real time, generating real-time wind field estimation through data assimilation, calculating spatial risk distribution, assessing flight path risk and optimizing.

[0007] The technical scheme adopted is:

[0008] A city low-altitude unmanned aerial vehicle risk assessment method based on LES-observation data, comprising the following steps:

[0009] Step 1, constructing a prior wind field database K g ;

[0010] A three-dimensional grid city model and a plurality of LES simulation working conditions are constructed, and then the prior wind field database K g under each LES simulation working condition is calculated based on large eddy simulation LES simulation.

[0011] wherein g-LES simulation case number; K g a set of prior wind field data points;

[0012] The three-dimensional grid city model comprises a plurality of three-dimensional grids; each three-dimensional grid presents simulation data forming a prior wind field data point;

[0013] Step 2, collect observation data y at the current time t, t∈[T1, T2];

[0014] wherein T1- flight start time of the predetermined route; T1- flight end time of the predetermined route;

[0015] From the prior wind field database K g corresponding to the three-dimensional grid city model, select a plurality of grids, and take the actual region represented by the selected grids as observation points;

[0016] Arrange sensors on each observation point; the data collected by all sensors in real time form observation data y;

[0017] Step 3, based on the set of prior wind field databases K g matching the current time t, the observation data y of the current time t, and the data assimilation module to generate the set of posterior wind fields X a and calculate the average posterior wind field X aa .

[0018] Step 4, flight path risk evolution prediction:

[0019] Based on the average posterior field X aa at the current time t, the average posterior field X aa at the continuous M time points before the current time t, the prediction model obtains the average posterior field X aa corresponding to P time points in time t+1~t+P.

[0020] Step 5, calculate the posterior wind field distribution map corresponding to each average posterior field X aa .

[0021] wherein the posterior risk distribution map is a set of comprehensive risk indexes RI corresponding to each three-dimensional grid;

[0022] Step 6, flight path risk assessment:

[0023] Along the predetermined route, extract the posterior wind field distribution map corresponding to the time falling within T1~T2.

[0024] Step 7, cyclically execute steps 2-6 to realize dynamic risk assessment.

[0025] Preferably, step 3 specifically comprises the following steps:

[0026] Step 3A, obtain the wind field condition where the observation point is located at the current time t, select N LES simulation working conditions matched with the wind field condition where the observation point is located, and obtain the prior wind field database K g as one prior wind field X b ;

[0027] Step 3B, apply the set Kalman filtering algorithm to any prior wind field X b , observation data y, and obtain a posterior wind field X a , that is, obtain the set of posterior wind fields X a ;

[0028] Step 3C, obtain the average value of all posterior wind fields X a , and obtain the average posterior wind field X aa .

[0029] Preferably, step 5 specifically comprises the following steps:

[0030] Step 5A, calculate the risk index of each three-dimensional grid based on the average posterior field X aa , and weight the respective risk indexes to form a comprehensive risk index RI;

[0031] Step 5B, render the RI value to the three-dimensional grid city model to form a posterior risk distribution map.

[0032] Preferably, in step 5A, the risk index comprises:

[0033] instantaneous wind speed over-limit probability index , turbulence intensity index , wind shear intensity index , vertical airflow intensity index .

[0034] Preferably, step 6 further comprises the steps of cumulative risk calculation and high-risk identification:

[0035] Calculate the sum of the comprehensive risk indexes RI of the flight segments; the flight segments are a set in [T1, T2];

[0036] If the comprehensive risk index of a flight segment is continuously higher than a preset safety threshold, it is identified as a high-risk flight segment.

[0037] Preferably, the step 1 further comprises the following steps:

[0038] According to the prior wind field database K g , divide the risk levels;

[0039] Mark the corresponding real area on the map with the risk level of the three-dimensional grid city model to form a prior risk distribution map;

[0040] Select an observation point according to the prior risk distribution map.

[0041] Preferably, the prior wind field data point comprises:

[0042] Average wind speed, average wind direction, turbulent kinetic energy TKE, Reynolds stress, vorticity.

[0043] Preferably, the LES simulation working condition comprises average wind direction and average wind speed.

[0044] Preferably, the data assimilation module adopts an ensemble Kalman filter or a variational assimilation method, and the sensor comprises a three-dimensional ultrasonic wind speed and direction instrument.

[0045] Compared with the prior art, the advantages of the present application are:

[0046] 1. High precision: combining high-resolution LES simulation and real-time observation, a risk assessment with a precision of meters is provided.

[0047] The prior art relies mainly on macro-weather forecasts or sparse ground observation stations, and the resolution is usually on the order of kilometers, which cannot resolve the complex microscale turbulence (such as alley wind, downwash, corner vortex, etc.) caused by urban building groups. The present application uses large eddy simulation (LES) technology to generate a three-dimensional wind field database with a resolution of 1-5 meters in the offline stage, which can finely resolve the micro-vortex structure that poses a serious threat to the attitude of the unmanned aerial vehicle. Combined with real-time data assimilation correction, the final risk assessment atlas also reaches a precision of meters, enabling the unmanned aerial vehicle to identify and avoid previously unpredictable local extreme airflow in advance, greatly improving the safety of flight.

[0048] 2. Real-time: based on an efficient data assimilation algorithm, the update is realized in seconds.

[0049] The urban low-altitude wind field changes rapidly, and the traditional numerical simulation method requires a huge amount of calculation and cannot realize real-time prediction. The present application uses the mode of "offline high-precision calculation + online data assimilation" to complete the LES calculation in advance, forming a prior database. In the online stage, only an efficient data assimilation algorithm (such as ensemble Kalman filter) needs to be run to quickly correct the prior field using real-time observation data. The update cycle of the entire assimilation and evaluation can be shortened to seconds (for example, 10 seconds), fully meeting the needs of the unmanned aerial vehicle to make immediate risk judgments and dynamic adjustments of the flight path when performing tasks such as logistics distribution and emergency rescue.

[0050] 3. Reliability: quantify uncertainty through the Bayesian framework to provide probabilistic risk assessment.

[0051] Traditional risk assessment usually gives a certain risk level, but does not tell the reliability of the prediction.

[0052] The present application uses the ensemble Kalman filter data assimilation method to generate posterior wind field while synchronously outputting the uncertainty distribution of wind field estimation (such as the standard deviation of wind speed at each point). This means that the system can not only predict "the wind speed at a certain place may be 5 m / s", but also give "the prediction has a 95% confidence interval between 4.5 m / s to 5.5 m / s". This probabilistic risk assessment provides a key "confidence" reference for flight decision, so that the UAV can adopt a more refined obstacle avoidance strategy, for example, choosing to detour or increasing the safety margin in high uncertainty areas, thereby realizing more robust and reliable autonomous flight.

[0053] 4. Adaptability: able to adapt to the needs of different types of UAVs and flight tasks.

[0054] Different types of UAVs (such as multi-rotor, fixed-wing), different load states (empty, full load) and different task types (such as aerial photography, transportation, rescue) have different wind field tolerance and sensitivity.

[0055] The risk assessment module of the present application is highly parameterized, and the calculation of its risk index integrates multiple risk factors, and the weight coefficients of each factor can be dynamically configured according to the specific performance parameters of the UAV (such as maximum wind resistance level, maneuverability, structural strength) and task priority. This flexibility ensures that the risk assessment results are highly targeted, providing optimal flight decision support for different application scenarios.

[0056] 5. Scalability: modular design, easy to extend to different cities and scenarios.

[0057] The system uses modular design, and each module (offline simulation, real-time observation, data assimilation, risk assessment, trajectory optimization) has independent functions and clear interfaces. This design brings excellent scalability. BRIEF DESCRIPTION OF DRAWINGS

[0058] Figure 1 Flowchart of the urban low-altitude UAV risk assessment method based on LES-observation data

[0059] Figure 2 Prior wind field database K simulated by large eddy simulation with grid resolution of 1 m g . DETAILED DESCRIPTION

[0060] The LES-observation data based urban low-altitude UAV risk assessment method of the present application will be described in more detail below in conjunction with the schematic diagram, which represents the preferred embodiment of the present application, it should be understood that the present application described herein can be modified by those skilled in the art, while still achieving the advantageous effects of the present application. Therefore, the following description should be understood as a broad knowledge for those skilled in the art, and not as a limitation of the present application.

[0061] As Figures 1-2 , a LES-observation data based urban low-altitude UAV risk assessment method, comprising the following steps:

[0062] Step 1, constructing a prior wind field database K g :

[0063] Constructing a three-dimensional grid city model and several LES simulation conditions, and then calculating the prior wind field database K under each LES simulation condition based on large eddy simulation LES simulation g ;

[0064] Wherein, g-LES simulation condition number; K g -prior wind field data point set;

[0065] The three-dimensional grid city model comprises several three-dimensional grids; the simulation data presented by each three-dimensional grid forms a prior wind field data point; the grid resolution reaches 1-5 meters, which can analyze the urban micro-scale turbulent structure.

[0066] Figure 2 In the figure, gray represents buildings, red points represent prior wind field data points, and red lines represent flow lines connecting the red points.

[0067] The prior wind field data includes:

[0068] Average wind speed, average wind direction, turbulent kinetic energy TKE, Reynolds stress, and vorticity.

[0069] Step 1A specifically includes the following steps:

[0070] 1. Data preparation: obtaining high-precision three-dimensional building model data and digital elevation model (DEM) terrain data of the target city (such as a certain science and technology park, core business district) and the like.

[0071] 2. Calculation domain and grid division: based on the above data, using computational fluid dynamics (CFD) software (such as ANSYSICEM CFD, snappyHexMesh) to construct the calculation domain of large eddy simulation (LES), and the grid resolution is set to 1-5 meters to ensure that the key turbulent structure can be analyzed.

[0072] 3. Design of working condition matrix: Based on the historical meteorological data of the target city for at least 5 years, the characteristics such as dominant wind direction, wind speed distribution, and atmospheric stability are statistically analyzed to design a multi-dimensional typical working condition matrix. For example: (1) Wind direction: 8 main directions; (2) Wind speed: 3-5 levels (3) Atmospheric stability: stable, neutral, unstable (4) Seasonal and time period changes.

[0073] 4. Perform batch LES calculation: On a high-performance computing (HPC) cluster, use mature CFD software (such as OpenFOAM, Ansys Fluent, etc.) to perform unsteady calculation on all the above working conditions. The simulation time of each working condition needs to ensure that the flow field is fully developed and reaches a quasi-steady state in a statistical sense, usually with a physical simulation time of not less than 1 hour, and continue to calculate for a period of time to collect sufficient statistical samples.

[0074] 5. Data post-processing: Post-process the simulation results of each working condition, extract and store the following three-dimensional field data: average wind speed, average wind direction, turbulent kinetic energy (TKE), Reynolds stress, vorticity, etc. These data constitute the prior wind field database of the system.

[0075] Step 1B, according to K g Divide the risk level;

[0076] Step 1C, mark the risk level on the map for the actual area corresponding to the three-dimensional grid city model to form a prior risk map.

[0077] Specifically: Based on the above statistical quantities, a preliminary risk level is constructed.

[0078] For example, areas with average turbulent kinetic energy exceeding a certain threshold or areas with strong downdrafts throughout the year are marked as static high-risk areas on the map. This atlas is a preliminary depiction of the inherent wind environment risk of the city and can be used to guide the placement of sensors (anemometers, temperature sensors, humidity sensors).

[0079] Step 2, collect observation data y at the current time t, t ∈ [T1, T2];

[0080] Where T1 is the start time of the scheduled flight; T1 is the end time of the scheduled flight;

[0081] From the prior wind field database K g Select several three-dimensional grids in the three-dimensional grid city model, and take the actual area represented by the selected grid as the observation point;

[0082] Arrange anemometers at each observation point; the data collected by all anemometers in real time constitute the observation data y.

[0083] Step 2 specifically includes the following steps:

[0084] 1. Sensor Network Deployment: Based on the prior risk map, deploy fixed 3D ultrasonic anemometers at identified key locations (such as rooftops, street canyon entrances, and open areas). Simultaneously, vehicles or drones moving within the area can be used as mobile observation platforms to form a dynamic and static integrated three-dimensional observation network.

[0085] 2. Data Acquisition and Transmission: Establish a low-latency, high-reliability data transmission network to collect data such as wind speed, wind direction, temperature, and humidity from various sensors in a standard format (such as JSON) and via protocols such as MQTT, and aggregate them to the central data server at a frequency of once every second (e.g., once every 1-5 seconds).

[0086] 3. Data Quality Control: A rigorous quality control process is implemented for the received raw data.

[0087] (1) Range check: Remove values ​​that are obviously beyond the physical possibility range (such as negative wind speed or more than 100 meters per second).

[0088] (2) Time consistency check: If a sensor reading changes drastically and unreasonably within a short period of time, it is marked as suspicious data.

[0089] (3) Spatial consistency check: The reading of a sensor is compared with the readings of multiple neighboring sensors. If the value deviates significantly from the surrounding trend, it is marked as suspicious data.

[0090] 4. Outlier marking and interpolation: Marked outliers are removed, and short-term missing data are interpolated using methods such as Kriging interpolation or time series interpolation to ensure the integrity of the data stream.

[0091] Step 3: Based on the prior wind field database K matched with the current time t g Given a set of data and the observation data y at the current time t, the posterior wind field X is generated using the ensemble Kalman filter algorithm. a The set of values, and calculate the mean posterior wind field X. aa Specifically, it includes the following steps:

[0092] Step 3A: Obtain the wind field conditions at the observation point at the current time t, select N LES simulation conditions that match the wind field conditions at the observation point, and store the prior wind field database K under each LES simulation condition. g As a priori wind field X b .

[0093] Specifically, based on macro-meteorological forecasts, the system obtains the background wind field conditions (such as prevailing wind direction, wind speed, and atmospheric stability) at the current time t. The system then automatically retrieves and selects each prior wind field X from the offline database.b .

[0094] To embody the prior uncertainty, we can select the closest N prior wind fields from the database K g (or add perturbation to a single case) to form an ensemble, i.e. a set of prior wind fields X b .

[0095] Step 3B, apply the data assimilation algorithm of ensemble Kalman filter to any prior wind field X b , observation data y and observation operator H, to get a posterior wind field X a , i.e. obtain a set of posterior wind fields X a .

[0096] 1. Compute the Kalman gain (K): The Kalman gain K determines how much the prior field is corrected by the innovation. Its computation is based on the error covariance matrix (Pb) of the prior wind field X b and the covariance matrix (R) of the observation error.

[0097] 2. State update: Use the Kalman gain to weight the innovation and feed it back to the entire three-dimensional wind field to generate the optimal posterior wind field (X a ), i.e. the wind field that has incorporated the observation information.

[0098] X a = X b + K(y - H(x b ))

[0099] The observation operator H is a mapping matrix or interpolation function, whose function is to interpolate the state variables (wind speed, wind direction) on the high-resolution three-dimensional model grid points to the positions of the spatially discrete sensor observation points. That is, H(x) represents the simulated value (data in the prior wind field database) corresponding to the observation point when the model state is x.

[0100] Compute the innovation: Compare the real-time observation data (y) with the predicted value of the model at the observation point (H(xb)), and calculate the difference, i.e. the innovation vector d=y−H(x b ). The innovation reflects the deviation of the current prior field from the true situation.

[0101] where X a is the posterior wind field, X b is the prior wind field, K is the Kalman gain, y is the observation data, and H is the observation operator. This step will update each member in the prior ensemble.

[0102] 3. Error Covariance Matrix Update: The error covariance matrix of the analysis field is updated synchronously to prepare for the next assimilation window. The updated ensemble sample then reflects the posterior uncertainty distribution.

[0103] The entire step 3B is performed cyclically at a fixed assimilation period (e.g., 10 seconds).

[0104] Step 3C: Calculate all posterior wind fields X a The average value, i.e., the average posterior wind field X. aa .

[0105] Step 4: Predict the evolution of flight path risks.

[0106] Based on the mean posterior field X at the current time t aa The average posterior field X of the M consecutive times preceding the current time t aa The forecast model (short-term nowcasting model) obtains the average posterior field X corresponding to P times from t+1 to t+P. aa .

[0107] The short-term nowcasting model is a spatiotemporal sequence prediction model based on deep learning. Its network structure, input, and output are as follows:

[0108] 1. Network Structure:

[0109] An encoder-decoder architecture based on Convolutional Long Short-Term Memory (ConvLSTM) is adopted, combined with the Skip Connections concept from U-Net, to accurately capture the complex spatiotemporal evolution characteristics of wind fields.

[0110] 2. The model input is a five-dimensional tensor representing the average posterior wind field Xaa over the past M consecutive time steps. The specific format of this tensor is as follows:

[0111]

[0112] in:

[0113] t represents the current time;

[0114] Specify the length of the input time series (for example, if you want to take data from the past 5 minutes, with each time step being 30 seconds, then M=10).

[0115] yes The mean posterior wind field at time t is itself a three-dimensional data volume, each... A plurality of physical quantity channels contained in each grid point of the three-dimensional grid city model, including at least: three-dimensional wind speed components (u, v, w), turbulent kinetic energy (TKE), etc.

[0116] Therefore, the dimension of the input tensor can be represented as (M, H, W, D, C),

[0117] Where: M - input time series length;

[0118] H, W, D - height, width and depth of the three-dimensional grid city model;

[0119] C - number of feature channels for each grid point (such as wind speed components u, v, w and TKE, then C = 4).

[0120] 3. Model output (Model Output):

[0121] The output of the model is also a five-dimensional tensor, representing the predicted wind field at future consecutive P time points. The specific format of this tensor is:

[0122]

[0123] Where: is the prediction result of the wind field at time t+1, and the dimension of the output tensor is (P, H, W, D, C), which is consistent with the input tensor in spatial dimensions (H, W, D) and feature channel dimensions (C).

[0124] Where, the unit of time is hour.

[0125] Based on the future wind field sequence of this model output, the system can calculate the risk indicators and comprehensive risk index RI for each future time (t+1,..., t+P) according to the method in step 4A, and generate the evolution trend of the future risk distribution map, providing forward-looking decision support for the path planning and risk warning of the unmanned aerial vehicle.

[0126] Step 5, calculate each average posterior field X aa corresponding posterior wind field distribution map.

[0127] In this implementation, M+P average posterior fields X aa corresponding posterior wind field distribution map.

[0128] The posterior risk distribution map is a set of comprehensive risk indexes RI corresponding to each three-dimensional grid.

[0129] Step 5A, based on the average posterior field X aa calculate the risk indicators for each three-dimensional grid, and weight the various risk indicators to form the comprehensive risk index RI;

[0130] Instantaneous wind speed over-limit probability index , turbulence intensity index , wind shear intensity index , vertical air current intensity index .

[0131] The calculation processes of the above four indexes all belong to the prior art.

[0132] RI=

[0133] , are weight coefficients.

[0134] Step 5B, rendering the RI value to the three-dimensional grid city model to form a risk distribution map.

[0135] Using different colors (such as green-yellow-red) or transparencies to represent the risk levels, forming an intuitive three-dimensional dynamic risk "cloud map" for the UAV flight control system or ground station operator to use.

[0136] Step 6, flight path risk assessment and high-risk identification.

[0137] Flight path risk assessment: along the predetermined route, extract the corresponding posterior wind field distribution map at the time falling within T1~T2.

[0138] The predetermined route corresponds to a series of route time points, and two route time points constitute the upper and lower limits of a flight leg.

[0139] Cumulative risk calculation and high-risk identification:

[0140] Calculate the sum of the comprehensive risk indexes RI of the flight legs; the flight leg is a set in [T1, T2];

[0141] If the comprehensive risk index of a certain flight leg is continuously higher than the preset safety threshold, it is identified as a high-risk flight leg.

[0142] Dynamic path optimization: once a high-risk leg is identified, the system will start the path optimization algorithm. This algorithm automatically searches and plans one or more alternative paths that bypass the high-risk area and have the lowest cumulative risk in the three-dimensional space, taking the risk index and uncertainty as the pathfinding cost, and sends it to the UAV flight control system as a rerouting suggestion.

[0143] Step 7, loop execution of steps 2~6 to achieve dynamic risk assessment.

[0144] In the next cycle, the current time t in step 2 becomes t+1.

[0145] For example, in the first cycle, the current time t is 8:00 AM, and the next cycle of the current time t is 9:00 AM, and so on.

[0146] The application provides a city low-altitude unmanned aerial vehicle risk assessment system based on large eddy simulation and observation data fusion, comprising the following five modules:

[0147] 1. Offline simulation module

[0148] Objective: Based on the city three-dimensional building model and large eddy simulation (LES) technology, generate city wind field database and typical risk distribution atlas covering typical meteorological conditions, as the prior information source of the system.

[0149] Specific scheme:

[0150] (1) Based on the target city high-precision three-dimensional building model and terrain data, using large eddy simulation (LES) technology, constructing a three-dimensional wind field database for micro-scale turbulence analysis;

[0151] (2) Design a multi-dimensional working condition matrix for typical seasons, time periods and meteorological conditions to generate a large amount of prior wind field data covering different environmental conditions;

[0152] (3) Generate city wind field database, simulation results include average wind speed, wind direction, turbulence intensity, Reynolds stress, vortex structure and other parameters;

[0153] (4) Based on statistical analysis, construct a typical risk distribution atlas, and identify typical high-risk areas (such as building corners, street canyons, etc.).

[0154] 2. Real-time observation module

[0155] Objective: Obtain real-time wind field data through distributed sensor network to make up for the limitations of LES database in local and instantaneous state estimation.

[0156] Specific scheme:

[0157] (1) Fixed wind speed sensors and mobile observation platforms are arranged in key areas of the city (together forming a wind speed sensor network) to collect real-time meteorological observation data including wind speed, wind direction, temperature, humidity, etc.

[0158] (2) Establish a data acquisition and transmission system to realize second-level update;

[0159] (3) Implement data quality control, including range check, time consistency and space consistency check, eliminate outliers and use interpolation method to correct missing data;

[0160] (4) Real-time observation data as input of data assimilation module, providing observation constraint for LES prior wind field.

[0161] 3. Data assimilation module

[0162] Objective: To generate a dynamically updated three-dimensional posterior wind field estimate by fusing LES prior wind field and real-time observation data using a Bayesian inference framework.

[0163] Specific scheme:

[0164] (1) Optimal fusion of offline LES prior wind field and real-time observation data using Ensemble Kalman Filter (EnKF) algorithm;

[0165] (2) Dynamic update of LES prior wind field according to observation data to generate posterior wind field state estimate;

[0166] (3) Synchronous output of wind field uncertainty quantification results, i.e. in the data assimilation process, combine error covariance matrix to dynamically calculate the uncertainty distribution of the current wind field estimate, output in the form of local standard deviation field or confidence interval, used to describe the credibility of wind speed / wind direction prediction. Uncertainty information as an important input parameter for risk assessment, used to reflect the reliability of wind field prediction in a certain area, to assist flight decision.

[0167] (4) The module is iteratively updated according to a fixed assimilation period (e.g. 10 seconds) to realize second-level dynamic wind field reconstruction and credibility estimation.

[0168] 4. Risk assessment module

[0169] Objective: To realize dynamic flight risk assessment of each location in urban space by combining UAV performance parameters and posterior wind field.

[0170] Specific scheme:

[0171] (1) Based on dynamically updated posterior wind field and uncertainty distribution, combine UAV maximum wind resistance capability, attitude control capability and other performance parameters to calculate flight risk index of each location in space;

[0172] (2) Risk factors include: instantaneous wind speed, turbulence intensity, wind shear intensity, vertical airflow, etc.

[0173] (3) Risk results are three-dimensional risk distribution map, supporting dynamic risk avoidance and flight strategy optimization of UAV system.

[0174] 5. Trajectory prediction and optimization module

[0175] Objective: To provide risk prediction and dynamic path optimization suggestions on future flight route according to UAV task and risk distribution.

[0176] Specific scheme:

[0177] (1) Based on the current UAV position information and the predetermined route, combined with the three-dimensional risk distribution map, the wind field state and risk evolution on the future path are predicted;

[0178] (2) The cumulative risk of the flight path is calculated, and multiple alternative low-risk path suggestions are provided;

[0179] (3) The flight path is dynamically adjusted in the high-risk or high-uncertainty area, and the robust optimization of the flight path is realized.

[0180] The above are only preferred embodiments of the present application, and do not have any limiting effect on the present application. Any person skilled in the art, without departing from the scope of the technical solutions of the present application, makes any form of equivalent replacement or modification of the technical solutions and technical contents disclosed by the present application, and still belongs to the protection scope of the present application.

Claims

1. A method for risk assessment of urban low-altitude unmanned aerial vehicle based on LES-observation data, characterized in that, The method comprises the following steps: Step 1, constructing a prior wind field database K g ; A three-dimensional grid city model and several LES simulation conditions are constructed, and then a priori wind field database K under each LES simulation condition is calculated based on large eddy simulation LES simulation g ; where g - LES simulation case number; K g - a set of prior wind field data points; The three-dimensional grid city model comprises a plurality of three-dimensional grids; each three-dimensional grid presents simulation data forming a prior wind field data point; Step 2, collect observation data y at the current time t, t [T1, T2]; Wherein, T1 is the flight start time of the predetermined route; T1 is the flight end time of the predetermined route; from a prior wind field database K g corresponding to a selected number of grids in the three-dimensional grid city model, and taking a real region represented by the selected grids as an observation point; Arranging sensors on each observation point; the data collected by all sensors in real time form observation data y; Step 3, generating a set of posterior wind fields X based on the prior wind field database K matching the current time t, the observation data y of the current time t, and the data assimilation module g Step 4, calculating the average posterior wind field X a Step 5, calculating the average posterior wind field X aa ; Step 4, flight path risk evolution prediction: Based on the average posterior field X at the current time t aa , the average posterior field X of the previous M time points before the current time t aa , the average posterior field X corresponding to P time points in time t+1~t+P is obtained by the prediction model aa ; Step 5, Compute each mean posterior field X aa Corresponding posterior wind field distribution plot: Wherein, the posterior risk distribution map is a set of comprehensive risk indexes RI corresponding to each three-dimensional grid; Step 6, flight path risk assessment: Along the predetermined route, extract the posterior wind field distribution map corresponding to the time falling within T1~T2; Further comprising the steps of cumulative risk calculation and high risk identification: Calculate the sum of the comprehensive risk indexes RI of the flight segments; the flight segment is a set in [T1, T2]; If the comprehensive risk index of a flight segment is continuously higher than the preset safety threshold, it is identified as a high-risk flight segment; Step 7, cyclically execute steps 2-6 to realize dynamic risk assessment.

2. The LES-observation data based urban low-altitude UAV risk assessment method according to claim 1, characterized in that, Step 3 specifically comprises the following steps: Step 3A, obtain the wind field condition where the observation point is located at the current time t, select N LES simulation working conditions matched with the wind field condition where the observation point is located, and obtain the prior wind field database K under each LES simulation working condition g As a prior wind field X b ; Step 3B, apply the ensemble Kalman filter algorithm to the prior wind field X b and the observation data y, each resulting in a posterior wind field X a , i.e. obtaining an ensemble of posterior wind fields X a . Step 3C, compute the average of all posterior wind fields X a to obtain the average posterior wind field X aa . 3.The LES-observation data based urban low-altitude UAV risk assessment method according to claim 1, characterized in that, Step 5 specifically comprises the following steps: Step 5A, computing the average posterior field X aa The risk indicators for each three-dimensional grid are calculated and the individual risk indicators are weighted to form a composite risk index, RI. Step 5B, render the RI value to the three-dimensional grid city model to form a posterior risk distribution map.

4. The LES-observation data based urban low-altitude UAV risk assessment method according to claim 3, characterized in that, In step 5A, the risk index includes: Instantaneous wind speed over-limit probability index , turbulence intensity index , wind shear intensity index , vertical air current intensity index .

5. The LES-observation data based urban low-altitude UAV risk assessment method according to claim 1, characterized in that, The step 1 further comprises the following steps: According to the prior wind field database K g Risk level is divided; Mark the corresponding real area of the three-dimensional grid city model on the map to form a prior risk distribution map; Select observation points according to the prior risk distribution map.

6. The LES-observation data based urban low-altitude UAV risk assessment method according to claim 1, characterized in that, The prior wind field data point includes: Average wind speed, average wind direction, turbulent kinetic energy TKE, Reynolds stress, vorticity.

7. The LES-observation data based urban low-altitude UAV risk assessment method according to claim 1, characterized in that, The LES simulation working condition includes average wind direction and average wind speed. 8.The LES-observation data based urban low-altitude UAV risk assessment method according to claim 1, characterized in that, The data assimilation module adopts ensemble Kalman filtering or variational assimilation method, and the sensor includes a three-dimensional ultrasonic wind speed and direction instrument.

Images