A transport risk prediction method based on big data analysis
Patent Information
- Application Number
- CN202611143048.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-07-30
- Publication Date
- 2026-10-09
AI Technical Summary
[0004]为了弥补以上不足,本发明提供了一种基于大数据分析的运输风险预测方法,旨在改善现有的运输风险预测方法大都采用单一运行数据或独立因素分析,容易造成旋翼式运输无人机在复杂环境下难以及时识别多因素共同作用引起的电池电压快速跌落风险的问题
1、本发明中,通过基于统一时空数据集合构建分别表征电池内阻变化、风场扰动及飞控电流响应的特征,并基于三者同步演化关系构建三源耦合状态表示,同时利用飞控电流响应特征与风场扰动特征同步关联并与电池内阻变化特征在统一时间窗口内联合映射,从而将电池内部状态变化、外部风场扰动以及飞控响应行为进行关联表征,改善了现有的运输风险预测方法大都采用单一运行数据或独立因素分析,由于无法描述环境扰动、控制响应与电池状态变化之间的耦合关系,从而造成旋翼式运输无人机在复杂环境下难以及时识别多因素共同作用引起的电池电压快速跌落风险的问题。
Smart Images

Figure CN122887190A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of unmanned aerial vehicle (UAV) logistics safety and intelligent decision-making technology, and in particular to a transportation risk prediction method based on big data analysis. Background Technology
[0002] With the development of the low-altitude economy and intelligent logistics, rotary-wing transport drones are increasingly being applied to urban delivery, emergency transport, and other scenarios. Rotary-wing transport drones typically operate autonomously using onboard batteries, combined with flight control, environmental perception, and path planning systems. To improve drone operational safety, existing transport risk prediction technologies are generally based on drone operational data acquisition, status monitoring, and risk analysis models, comprehensively analyzing battery status, environmental changes, control responses, and space environment conditions during flight. Specifically, onboard battery operational data reflects the power system status, flight control command data characterizes the drone's attitude adjustment process, environmental wind field prediction data describes airflow changes within the flight area, and ground space constraint data assists in flight path planning and safe zone selection.
[0003] Existing transportation risk prediction methods mostly use single operational data or independent factor analysis. Because they cannot describe the coupling relationship between environmental disturbances, control response and battery state changes, rotary-wing transport drones have difficulty in timely identifying the risk of rapid battery voltage drop caused by the combined effect of multiple factors in complex environments. Summary of the Invention
[0004] To overcome the above shortcomings, this invention provides a transportation risk prediction method based on big data analysis. It aims to improve the problem that most existing transportation risk prediction methods rely on single operational data or independent factor analysis, which can easily lead to the difficulty in timely identifying the risk of rapid battery voltage drop caused by the combined effects of multiple factors in complex environments for rotary-wing transport drones.
[0005] This invention provides the following technical solution: a transportation risk prediction method based on big data analysis, comprising the following steps: S1. Acquire onboard battery operation data, flight control command data, environmental wind field prediction data, and ground gridded safety constraint data during the operation of the rotary-wing transport drone, and perform time and space alignment processing on the onboard battery operation data, flight control command data, environmental wind field prediction data, and ground gridded safety constraint data to generate a unified spatiotemporal data set. S2. Based on a unified spatiotemporal data set, features representing battery internal resistance change, wind field disturbance, and flight control current response are constructed respectively, and a three-source coupled state representation is constructed based on the synchronous evolution relationship of the three. Among them, the flight control current response feature is synchronously associated with the wind field disturbance feature and jointly mapped with the battery internal resistance change feature within a unified time window to form a three-source coupled state representation. S3. Input the three-source coupling state representation into the risk prediction model, and physically constrain the risk prediction model based on the voltage balance relationship between the battery open-circuit voltage, discharge current, internal resistance and polarization voltage, and output the failure probability value of the battery cell voltage falling below the safety threshold and its upper bound of the confidence interval within the future preset time window. S4. Based on the ground pedestrian density, building shading degree and flat area reduction ratio in the ground gridded safety constraint data, generate a candidate safety grid set; S5. Based on the failure probability value and its upper bound of the confidence interval, select the candidate safety grid set grid by grid and adjust the UAV flight path to the spatial position corresponding to the selected candidate safety grid.
[0006] By adopting the above technical solution, features representing changes in battery internal resistance, wind field disturbances, and flight control current response are constructed based on a unified spatiotemporal data set. A three-source coupled state representation is constructed based on the synchronous evolution relationship of the three. At the same time, the flight control current response features are synchronously correlated with the wind field disturbance features and jointly mapped with the battery internal resistance change features within a unified time window. This allows for the correlation and characterization of changes in the battery's internal state, external wind field disturbances, and flight control response behavior. This improves upon existing transportation risk prediction methods, which mostly rely on single operational data or independent factor analysis. Because they cannot describe the coupling relationship between environmental disturbances, control response, and battery state changes, rotary-wing transport drones often struggle to identify the risk of rapid battery voltage drop caused by the combined effects of multiple factors in complex environments.
[0007] Furthermore, in S1, the steps of acquiring onboard battery operation data, flight control command data, environmental wind field prediction data, and ground gridded safety constraint data during the operation of the rotary-wing transport drone include: Acquire data on the voltage, current, and temperature of individual airborne battery cells; Acquire throttle command data and attitude control command data from the flight control command data; Acquire horizontal wind speed data, wind direction data, and vertical turbulence data from environmental wind field prediction data; Obtain ground pedestrian density data, building shading data, and flat area reduction ratio data from the ground gridded safety constraint data.
[0008] Furthermore, in S1, the step of performing time and space alignment processing on airborne battery operation data, flight control command data, environmental wind field prediction data, and ground gridded safety constraint data to generate a unified spatiotemporal data set includes: The timestamps of airborne battery operation data, flight control command data, environmental wind field prediction data, and ground gridded safety constraint data are uniformly mapped to the flight clock reference. The environmental wind field prediction data and the ground gridded safety constraint data are transformed into a unified spatial coordinate system. Data with different sampling frequencies are resampled according to time windows and time slice sequences are generated. The airborne battery operation data, flight control command data, environmental wind field prediction data, and ground gridded safety constraint data in the time-slice sequence are spliced together at corresponding positions to form a unified spatiotemporal data set.
[0009] Furthermore, in S2, the step of constructing features representing changes in battery internal resistance, wind field disturbances, and flight control current response based on a unified spatiotemporal data set includes: Extract airborne battery operation data from a unified spatiotemporal dataset and construct individual battery cell voltage change sequences and discharge current sequences according to time series. Calculate the battery internal resistance variation sequence based on the battery cell voltage variation sequence and discharge current sequence; Extract environmental wind field prediction data from a unified spatiotemporal dataset and construct wind speed and wind direction data sequences based on flight location mapping; Extract flight control command data from a unified spatiotemporal dataset and construct throttle command sequence and current response sequence.
[0010] Furthermore, in S2, the step of constructing the three-source coupled state representation based on the synchronous evolution relationship of the three includes: The battery internal resistance change sequence, wind speed data sequence, and current response sequence are time-aligned according to the flight time reference. The battery internal resistance change sequence, wind speed data sequence, and current response sequence are sliced according to a unified time window after time alignment. Within each time window, establish the correspondence between the battery internal resistance change sequence and the current response sequence, as well as the correspondence between the wind speed data sequence and the current response sequence; A three-source coupled state representation is generated based on the correspondence within each time window.
[0011] Furthermore, in S3, the step of physically constraining the risk prediction model based on the voltage balance relationship between the battery open-circuit voltage, discharge current, internal resistance, and polarization voltage includes: Extract the battery open-circuit voltage, discharge current, internal resistance, and polarization voltage from the three-source coupling state representation; The ohmic voltage drop calculation term is constructed based on the battery open-circuit voltage, discharge current, and internal resistance. A polarization voltage drop calculation term is constructed based on the battery open-circuit voltage, discharge current, and polarization voltage. The ohmic voltage drop calculation term and the polarization voltage drop calculation term are subjected to consistency constraints with the cell voltage output by the risk prediction model.
[0012] Furthermore, in S3, the step of outputting the failure probability value and the upper bound of the confidence interval of the battery cell voltage falling below the safety threshold within a future preset time window includes: Based on the three-source coupling state representation, time series data is extracted according to a preset time window; Input time series data into the risk prediction model and calculate the state of voltage of individual battery cells step by step; A voltage drop event sequence is generated based on the comparison relationship between the voltage state of individual battery cells and the safety threshold. Failure probability values are generated based on the voltage drop event sequence statistics, and the upper bound of the corresponding confidence interval is calculated.
[0013] Further, in S4, the step of generating the candidate security grid set includes: The ground area corresponding to the ground rasterized safety constraint data is divided into discrete raster cells and a raster index set is generated; Based on the raster index set, the ground pedestrian density, building shading degree and the proportion of flat and reducible area are mapped to raster cells; The safety score of each grid cell is calculated based on the ground pedestrian density, the building shielding degree, and the percentage of flat and retractable area. Based on the security score of each grid cell, grid cells are selected to form a candidate set of security grids.
[0014] Furthermore, in S5, the step of selecting the candidate safe grid set grid-by-grid based on the failure probability value and its upper bound confidence interval includes: Obtain the spatial coordinate information of each grid cell in the candidate safety grid set; The comprehensive selection weight of each grid cell is calculated based on the ground pedestrian density, building shading degree, and flat area reduction ratio corresponding to each grid cell, as well as the failure probability value and the upper limit of the confidence interval. Based on the comprehensive selection weights, each grid cell is filtered grid by grid. Sort the filtered raster cells and select the target raster cell.
[0015] Furthermore, in S5, the step of adjusting the UAV flight path to the spatial position corresponding to the selected candidate safety grid includes: Extract the spatial coordinates corresponding to the selected candidate safety grid and generate the target spatial location point; Connect the current spatial location of the drone with the target spatial location point by creating path nodes; Generate a continuous flight path sequence based on the path node connection relationship; The UAV flight path is updated and path adjustment is performed based on a continuous flight path sequence.
[0016] The present invention has the following beneficial effects: 1. In this invention, features representing changes in battery internal resistance, wind field disturbances, and flight control current response are constructed based on a unified spatiotemporal data set. A three-source coupled state representation is constructed based on the synchronous evolution relationship of the three. Simultaneously, the flight control current response features are synchronously correlated with the wind field disturbance features and jointly mapped with the battery internal resistance change features within a unified time window. This allows for the correlated characterization of changes in battery internal state, external wind field disturbances, and flight control response behavior. This improves upon existing transportation risk prediction methods, which mostly rely on single operational data or independent factor analysis. Because these methods cannot describe the coupling relationship between environmental disturbances, control response, and battery state changes, rotorcraft transport drones often struggle to identify the risk of rapid battery voltage drop caused by the combined effects of multiple factors in complex environments.
[0017] 2. In this invention, the three-source coupling state is input into the risk prediction model, and the risk prediction model is physically constrained based on the voltage balance relationship between the battery open-circuit voltage, discharge current, internal resistance and polarization voltage. The failure probability value and the upper bound of the confidence interval of the battery cell voltage falling below the safety threshold within a preset time window are output. This realizes the risk probability prediction based on the coupling state and the battery voltage change law, which improves the problem that most existing transportation risk prediction methods use current state detection or fixed threshold judgment. Due to the lack of a prediction mechanism for the future battery cell voltage drop trend, risk identification usually depends on the state change after the failure occurs and it is difficult to conduct risk assessment in advance.
[0018] 3. In this invention, candidate safety grids are selected grid by grid based on the failure probability value and the upper bound of its confidence interval, and the UAV flight path is adjusted to the spatial position corresponding to the selected candidate safety grid. This enables the linkage decision between risk prediction results and ground space safety constraints, which improves the existing transportation risk handling methods, which mostly adopt fixed forced landing areas or passive response methods after failure. Since the predicted risk status is not analyzed in conjunction with ground safety conditions, it is difficult to take into account both flight safety and ground area safety in the case of UAV abnormality. Attached Figure Description
[0019] Figure 1 A schematic diagram of the overall process of a transportation risk prediction method based on big data analysis according to the present invention; Figure 2 This is a schematic diagram illustrating the process of constructing a unified spatiotemporal data set in a transportation risk prediction method based on big data analysis according to the present invention. Figure 3 This is a schematic diagram illustrating the process of constructing a three-source coupled state representation in a transportation risk prediction method based on big data analysis according to the present invention. Figure 4 This is a schematic diagram of the risk prediction process based on voltage balance relationship constraints in a transportation risk prediction method based on big data analysis according to the present invention. Figure 5 This is a schematic diagram of the candidate safety grid generation and route adjustment process in a transportation risk prediction method based on big data analysis according to the present invention. Figure 6 This is a schematic diagram of the module architecture of a transportation risk prediction system based on big data analysis according to the present invention. Detailed Implementation
[0020] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0021] Example 1: In the first embodiment of the present invention, a transportation risk prediction method based on big data analysis is provided, such as... Figure 1 As shown, the process includes the following steps: S1, acquiring onboard battery operation data, flight control command data, environmental wind field prediction data, and ground gridded safety constraint data during the operation of the rotary-wing transport drone, and performing time and space alignment processing on the onboard battery operation data, flight control command data, environmental wind field prediction data, and ground gridded safety constraint data to generate a unified spatiotemporal data set; Furthermore, in S1, the steps for acquiring onboard battery operation data, flight control command data, environmental wind field prediction data, and ground gridded safety constraint data during the operation of the rotary-wing transport drone include: Acquire data on the voltage, current, and temperature of individual airborne battery cells; Acquire throttle command data and attitude control command data from the flight control command data; Acquire horizontal wind speed data, wind direction data, and vertical turbulence data from environmental wind field prediction data; Obtain ground pedestrian density data, building shading data, and flat area reduction ratio data from the ground gridded safety constraint data.
[0022] Specifically, this method is applied to risk prediction scenarios for rotary-wing transport drones with batteries in a pseudo-fully charged state. A pseudo-fully charged state refers to a state where the battery's open-circuit voltage is not lower than 3.7V when it is stationary on the ground before takeoff, and the DC internal resistance calculated through a 1C constant current discharge pulse test reaches more than 1.5 times the factory initial value. The onboard battery operating data is acquired by the battery management system through individual cell voltage acquisition chips and current sensors, including the real-time voltage of each battery cell. Discharge current and battery cell temperature Meanwhile, the battery management system calculates the real-time state of charge based on the ampere-hour integral method. Simultaneously, the cloud retrieves the complete charge-discharge curves and internal resistance growth sequence of the battery pack's historical charge-discharge cycles, extracting the cumulative number of cycles and the initial internal resistance baseline value. This serves as the basis for assessing the aging state of internal resistance; flight control command data is extracted from the flight controller data bus, including throttle percentage commands. and attitude control commands Environmental wind field prediction data is accessed from the city meteorological cloud service API via a 5G link to extract the horizontal wind speed within a 1km grid ahead of the flight path. ,wind direction and vertical turbulence intensity Ground-based rasterized safety constraint data was obtained from the city's dynamic geographic information platform, including real-time pedestrian density data at a 5-meter resolution. Building shading data Data on the proportion of flat and retractable areas .
[0023] like Figure 2 As shown, further, in S1, the steps of performing time and space alignment processing on airborne battery operation data, flight control command data, environmental wind field prediction data, and ground gridded safety constraint data to generate a unified spatiotemporal data set include: The timestamps of airborne battery operation data, flight control command data, environmental wind field prediction data, and ground gridded safety constraint data are uniformly mapped to the flight clock reference. The environmental wind field prediction data and the ground gridded safety constraint data are transformed into a unified spatial coordinate system. Data with different sampling frequencies are resampled according to time windows and time slice sequences are generated. The airborne battery operation data, flight control command data, environmental wind field prediction data, and ground gridded safety constraint data in the time-slice sequence are spliced together at corresponding positions to form a unified spatiotemporal data set.
[0024] Specifically, the GPS timestamp of the UAV flight control system is used as the flight clock reference, and the timestamps of the four types of data are uniformly mapped to this reference. The latitude and longitude coordinates of the environmental wind field prediction data and the ground gridded safety constraint data are transformed to the Northeast ENU coordinate system with the UAV takeoff point as the origin. Data with different sampling frequencies are downsampled or interpolated and resampled with a time window of 30 seconds and a reference frequency of 100Hz to generate time slice sequences of equal length. The 30-second window takes into account both the amount of data required for coupled feature extraction and the real-time requirements of edge computing. In each time slice, the airborne battery operation data, flight control command data, environmental wind field prediction data, and ground gridded safety constraint data are aligned according to the timestamps and then stitched together into a unified spatiotemporal data set.
[0025] S2. Based on a unified spatiotemporal data set, features representing battery internal resistance change, wind field disturbance, and flight control current response are constructed respectively, and a three-source coupled state representation is constructed based on the synchronous evolution relationship of the three. Among them, the flight control current response feature is synchronously associated with the wind field disturbance feature and jointly mapped with the battery internal resistance change feature within a unified time window to form a three-source coupled state representation. Furthermore, in S2, the steps of constructing characteristics representing battery internal resistance changes, wind field disturbances, and flight control current responses based on a unified spatiotemporal data set include: Extract airborne battery operation data from a unified spatiotemporal dataset and construct individual battery cell voltage change sequences and discharge current sequences according to time series. Calculate the battery internal resistance variation sequence based on the battery cell voltage variation sequence and discharge current sequence; Extract environmental wind field prediction data from a unified spatiotemporal dataset and construct wind speed and wind direction data sequences based on flight location mapping; Extract flight control command data from a unified spatiotemporal dataset and construct throttle command sequence and current response sequence.
[0026] Specifically, the voltage sequence of each battery cell is extracted from a unified spatiotemporal dataset. and discharge current sequence The DC internal resistance of a battery was identified online using a recursive least squares method with a forgetting factor of 0.95, resulting in a sequence of DC internal resistance changes. Environmental wind field prediction data is extracted from a unified spatiotemporal dataset, and combined with the real-time yaw angle of the UAV, the horizontal wind speed and direction in the geographic coordinate system are converted into a forward wind speed component sequence in the body coordinate system. and lateral wind speed component sequence The amplitude of the horizontal combined wind speed in the volumetric coordinate system based on the forward and lateral wind speed components. Extract flight control throttle command sequences from a unified spatiotemporal dataset. Based on the linear proportional relationship between throttle and current, superimposed with a 50ms first-order inertial delay element, the current response hysteresis characteristics of the electronic speed governor are simulated and mapped to a current response sequence. .
[0027] like Figure 3 As shown, further, in S2, the steps for constructing the three-source coupled state representation based on the synchronous evolution relationship of the three include: The battery internal resistance change sequence, wind speed data sequence, and current response sequence are time-aligned according to the flight time reference. The battery internal resistance change sequence, wind speed data sequence, and current response sequence are sliced according to a unified time window after time alignment. Within each time window, establish the correspondence between the battery internal resistance change sequence and the current response sequence, as well as the correspondence between the wind speed data sequence and the current response sequence; A three-source coupled state representation is generated based on the correspondence within each time window.
[0028] Specifically, the battery internal resistance change sequence Water platform wind speed sequence With current response sequence Time alignment was performed based on the flight clock reference. The three aligned sequences were sliced using a sliding time window with a width of 2 seconds and a step size of 0.1 seconds; the 2-second window corresponds to the typical response cycle of the UAV flight control system. Within each time window, the correlation coefficient between the internal resistance change sequence and the current response sequence was calculated. And the correlation coefficient between the wind speed sequence and the current response sequence. ; to each window mean mean Mean and correlation coefficient and Combined into a three-source coupled state vector ; Calculate the coupling strength index based on the three-source coupling state vector ;in The initial internal resistance reference value, The maximum wind speed that the drone can withstand is set at 15 m / s. The rated discharge current is taken as 22A, with weighting. , , Principal component analysis of the contributions of various factors to the pseudo-fully charged events in 500,000 historical flight hours determined them to be 0.5, 0.3, and 0.2, respectively. When the system enters the pseudo-fully charged risk warning zone, this warning state corresponds to a pseudo-fully charged dynamic risk state, satisfying the combined characteristics of the following: the online estimated internal resistance is continuously higher than the initial value by 150%, the current state of charge is not lower than 30%, and the deviation between the load voltage and the expected voltage from the lookup table is not lower than 0.3V; all windows and corresponding The values constitute a three-source coupled state representation.
[0029] S3. Input the three-source coupling state representation into the risk prediction model, and physically constrain the risk prediction model based on the voltage balance relationship between the battery open-circuit voltage, discharge current, internal resistance and polarization voltage, and output the failure probability value of the battery cell voltage falling below the safety threshold and its upper bound of the confidence interval within the future preset time window. like Figure 4 As shown, further, in S3, the steps for applying physical constraints to the risk prediction model based on the voltage balance relationship between the battery open-circuit voltage, discharge current, internal resistance, and polarization voltage include: Extract the battery open-circuit voltage, discharge current, internal resistance, and polarization voltage from the three-source coupling state representation; The ohmic voltage drop calculation term is constructed based on the battery open-circuit voltage, discharge current, and internal resistance. A polarization voltage drop calculation term is constructed based on the battery open-circuit voltage, discharge current, and polarization voltage. The ohmic voltage drop calculation term and the polarization voltage drop calculation term are subjected to consistency constraints with the cell voltage output by the risk prediction model.
[0030] Specifically, the risk prediction model needs to be trained offline before deployment. The rules for constructing training samples are as follows: positive samples are labeled with all event points where the voltage of each battery cell drops sharply from above 3.5V to below 3.0V within 5 seconds, triggering landing or parachute deployment. Feature sequence segments from the 5 minutes prior to this point are extracted and labeled as high-risk. Negative samples are randomly extracted from normal flight records and labeled as normal. At the same time, an adversarial negative sample is included, which is a segment where the battery internal resistance exceeds 1.5 times the initial value and the ambient wind speed exceeds 80% of the wind resistance limit, but the flight controller does not issue a large throttle command. This segment is still marked as normal to reduce false alarms. Time series SMOTE is used to enhance the minority class of high-risk samples. During model training, the battery open-circuit voltage is extracted from the three-source coupled state representation. Discharge current DC internal resistance and polarization voltage ;in Based on the current state of charge The information was obtained from the OCV-SOC table. The calculation is simplified by the high polarization approximation of the Butler-Volmer equation. In the formula The ideal gas constant is taken as 8.314 J / (mol·K). The absolute temperature of a battery cell is expressed in Kelvin. The charge transfer coefficient is set to 0.5. The Faraday constant is taken as 96485 C / mol. The equivalent exchange current of a single battery cell is taken as 0.1A, which is a typical value for a 22Ah rated capacity battery cell at 25°C; an ohmic voltage drop calculation term is constructed. Constructing polarization voltage drop calculation terms The risk prediction model uses a Transformer temporal encoder to extract features, which are then processed by a 6-layer multi-head self-attention encoder to extract the CLS marker bits. These are then compressed into a 64-dimensional risk feature vector by a fully connected layer. A parallel binary classification auxiliary branch is then added after the encoder to predict whether the remaining available battery power meets the vertical landing requirements and outputs a Boolean flag indicating power sufficiency. Following this, two layers of deep Gaussian process regression heads are connected to predict the terminal voltage. The total loss function during training is ;in This is the sum of the negative log-likelihood of the deep Gaussian process and the cross-entropy loss of the auxiliary branch binary classification. For model prediction of terminal voltage The terminal voltage calculated from the physical equations The mean square error between them This is a boundary constraint penalty term. The constraint charge state ranges from 0 to 1, and the DC internal resistance is always greater than or equal to 0. A square penalty is applied when the boundary is exceeded. This is a hyperparameter. and In this scheme, 0.1 and 0.01 are preferred.
[0031] Furthermore, in S3, the step of outputting the failure probability value and the upper bound of the confidence interval of the battery cell voltage falling below the safety threshold within a future preset time window includes: Based on the three-source coupling state representation, time series data is extracted according to a preset time window; Input time series data into the risk prediction model and calculate the state of voltage of individual battery cells step by step; A voltage drop event sequence is generated based on the comparison relationship between the voltage state of individual battery cells and the safety threshold. Failure probability values are generated based on the voltage drop event sequence statistics, and the upper bound of the corresponding confidence interval is calculated.
[0032] Specifically, the time-series data corresponding to the three-source coupling state is extracted as the input sequence according to a preset time window of 5 minutes. The 5-minute window is set based on the statistical time distribution of 327 historical pseudo-full charge events from the time the CSI entered the warning zone to the actual voltage drop below 3.0V, with the 90th percentile value being 4.2 minutes. The input sequence is then fed into a pre-trained risk prediction model, which outputs the predicted values of the voltage of each battery cell step by step. ;Compare With safety threshold The threshold is the safe discharge cutoff voltage of the lithium polymer battery; a voltage drop event sequence is generated; the number of events occurring in the drop event sequence is counted, and the ratio of the occurrence frequency to the total number of time steps is used as the failure probability value. Simultaneously, the uncertainty prediction head of the model outputs the upper bound of the confidence interval for this probability. , The probability value is between 0 and 1, reflecting the uncertainty of the prediction result.
[0033] S4. Based on the ground pedestrian density, building shading degree and flat area reduction ratio in the ground gridded safety constraint data, generate a candidate safety grid set; like Figure 5 As shown, further, in S4, the step of generating the candidate security grid set includes: The ground area corresponding to the ground rasterized safety constraint data is divided into discrete raster cells and a raster index set is generated; Based on the raster index set, the ground pedestrian density, building shading degree and the proportion of flat and reducible area are mapped to raster cells; The safety score of each grid unit is calculated based on ground pedestrian density, building shielding degree, and the proportion of flat and reducible area. Based on the security score of each grid cell, grid cells are selected to form a candidate set of security grids.
[0034] Specifically, a circular ground area with a radius of 200 meters, centered on the drone's current location, is divided into discrete grid cells with a resolution of 5 meters. The 200-meter radius corresponds to the maximum horizontal range that the drone can reach in gliding mode within a 5-minute prediction window, and the row and column numbers are used as the grid index set; ground pedestrian density data is also included. Building shading data Data on the proportion of flat and retractable areas Mapping spatial coordinates to each grid cell, where the pedestrian density is... This indicates the number of people per 100 square meters, and the building's shade ratio. The value ranges from 0 to 1. A higher value indicates more severe building obstruction, and the flat area can reduce the proportion of the area. The value ranges from 0 to 1; the ground safety index is calculated for each grid cell. ;in The normalized reference value for human population density is taken as 10 people per 100 square meters in this plan; the ground safety index is selected. A set of candidate safety grids is formed by grid cells with a minimum size of 0.6.
[0035] S5. Based on the failure probability value and the upper bound of its confidence interval, select the candidate safety grid set grid by grid and adjust the UAV flight path to the spatial position corresponding to the selected candidate safety grid. Furthermore, in S5, the step of selecting the candidate safe grid set grid-by-grid based on the failure probability value and its upper bound confidence interval includes: Obtain the spatial coordinate information of each grid cell in the candidate safety grid set; The comprehensive selection weight of each grid cell is calculated based on the ground pedestrian density, building shading, and flat area reduction ratio corresponding to each grid cell, as well as the failure probability value and the upper limit of the confidence interval. Each grid cell is filtered grid by grid based on a comprehensive selection weight. Sort the filtered raster cells and select the target raster cell.
[0036] Specifically, based on the power sufficiency flag output by the model's auxiliary branch. Determine if the remaining battery energy supports a vertical descent; when and When deemed low risk, maintain the current route and reduce cruising speed by 10% to increase safety margin; when ,or but When the risk level is determined to be medium, the spatial coordinate information of each grid cell in the candidate safety grid set is obtained and the ground safety index is calculated. Select The highest grid cell adjusts its course and moves at a constant speed towards that grid cell, while simultaneously sending a warning message to the control center via the 5G link; when and If the result is true, it is considered high-risk, and a safety grid is selected from the candidate safety grid set. The highest grid unit descends vertically at the maximum safe descent rate, simultaneously broadcasting emergency landing coordinates and obstacle avoidance warnings, and activating visual obstacle avoidance; when and If the result is false, it is judged as an emergency risk, and the attitude is automatically adjusted. The aircraft glides over the highest grid cell, then activates its parachute and sends the highest priority distress coordinates. The probability thresholds of 0.3 and 0.7 were determined by maximizing the Youden exponent under the constraint of ensuring a recall rate of no less than 95% and a false alarm rate of no more than 5% through ROC analysis of 327 false full-charge events from 500,000 flight hours. The confidence threshold of 0.6 was set based on the 90th percentile of the prediction variance. During grid selection, the ground safety index... The pedestrian density coefficient is dynamically adjusted based on the drone's current flight altitude. When the flight altitude is above 100 meters, the pedestrian density coefficient is scaled down from 0.5 to 0.35, and when the flight altitude is below 30 meters, the pedestrian density coefficient is increased from 0.5 to 0.65.
[0037] Furthermore, in S5, the step of adjusting the drone's flight path to the spatial location corresponding to the selected candidate safety grid includes: Extract the spatial coordinates corresponding to the selected candidate safety grid and generate the target spatial location point; Connect the current spatial location of the drone with the target spatial location point by creating path nodes; Generate a continuous flight path sequence based on the path node connection relationship; The UAV flight path is updated and path adjustment is performed based on a continuous flight path sequence.
[0038] Specifically, the system obtains the center latitude and longitude coordinates of the selected candidate safety grid, and generates the target spatial location point by combining the current ground altitude and the preset hovering height of 10 meters. The current spatial location of the UAV and the target spatial location point are used as the starting and ending points of the path planning. The initial path nodes are generated by connecting them with straight lines. If the straight path crosses the 3D building model, the 3D A* algorithm is used to insert intermediate detour nodes. The path nodes are connected in sequence to generate a continuous flight path sequence containing longitude, latitude, and altitude. The continuous flight path sequence is written into the waypoint buffer of the flight control system and path switching is performed.
[0039] After the flight, the high-frequency airborne data, environmental data, and risk probability sequences output by the model during the flight are packaged and uploaded to the cloud. The complete data chain that triggers events of medium risk or above is automatically marked as a high-value sample and stored in the training library. The cloud regularly uses newly added high-value samples and negative samples to incrementally train the risk prediction model. After verifying the performance improvement through A / B testing, the model is pushed to all drones in the fleet to achieve self-evolution and update. The airborne inference model has a parameter size of 30MB after INT8 quantization. On the NVIDIA Jetson Orin platform, the single-frame inference latency is less than 15ms and the power consumption of a single inference is less than 5W. The airborne end only needs to reserve 30% of the GPU computing power to meet the real-time processing requirements of 200Hz.
[0040] Example 2: In the second embodiment of the present invention, the present invention provides a transportation risk prediction system based on big data analysis, such as... Figure 6 As shown, it includes the following modules: The data acquisition module is used to collect onboard battery operation data, flight control command data, environmental wind field prediction data, and ground gridded safety constraint data during the operation of the rotary-wing transport drone. It also performs spatiotemporal alignment processing on the onboard battery operation data, flight control command data, environmental wind field prediction data, and ground gridded safety constraint data to generate a unified spatiotemporal data set. The coupling state construction module is used to extract battery internal resistance variation characteristics, wind field disturbance characteristics, and flight control current response characteristics based on the unified spatiotemporal data set, and to construct a three-source coupling state representation based on the synchronous evolution relationship of the three. The risk prediction module is used to receive the three-source coupling state representation, and constrain the risk prediction model based on the voltage balance relationship between the battery open-circuit voltage, discharge current, internal resistance and polarization voltage, and output the failure probability value and the upper bound of the confidence interval of the battery cell voltage falling below the safety threshold within a future preset time window. The safety grid generation module is used to generate a set of candidate safety grids based on ground pedestrian density, building shading and flat area reduction ratio in ground gridded safety constraint data. The path adjustment module is used to select a target grid from the candidate safety grid set based on the failure probability value and its upper bound of the confidence interval, and adjust the UAV flight path to the spatial position corresponding to the target grid.
[0041] In urban low-altitude logistics delivery scenarios, rotary-wing transport drones flying in densely built-up areas may experience problems such as sudden current surges, increased internal resistance voltage drops, and rapid voltage drops in individual battery cells due to complex wind field disturbances and battery aging. Furthermore, existing risk avoidance methods struggle to determine safe landing areas in advance based on ground environmental conditions. To address these issues, this invention employs a transportation risk prediction system based on big data analysis, such as... Figure 6 As shown. The specific implementation process of this system is as follows: First, the data acquisition module collects onboard battery operation data, flight control command data, environmental wind field prediction data, and ground gridded safety constraint data during the operation of the rotary-wing transport drone. It then performs spatiotemporal alignment processing on the above data to generate a unified spatiotemporal data set, thereby achieving a unified correlation between battery status, flight control response, environmental disturbances, and ground safety conditions.
[0042] Then, the coupled state construction module extracts the battery internal resistance change characteristics, wind field disturbance characteristics, and flight control current response characteristics based on a unified spatiotemporal data set, and constructs a three-source coupled state representation based on the synchronous evolution relationship of the three, so that the correlation process between wind field change, flight control adjustment and battery voltage drop is uniformly described, and the ability to characterize the transportation risk state under complex disturbance conditions is improved.
[0043] Then, the risk prediction module receives the three-source coupling state representation and constrains the risk prediction model based on the voltage balance relationship between the battery open-circuit voltage, discharge current, internal resistance and polarization voltage. It outputs the failure probability value of the battery cell voltage falling below the safety threshold within a preset time window and its upper bound of the confidence interval, so as to realize the early prediction of the risk of sudden battery failure.
[0044] Then, the safety grid generation module generates a set of candidate safety grids based on the ground pedestrian density, building shading and flat landable area ratio in the ground gridded safety constraint data, so that the candidate landing areas simultaneously meet the ground safety conditions and the space landable conditions.
[0045] Finally, the path adjustment module selects the target grid from the candidate safety grid set based on the failure probability value and its upper bound confidence interval, and adjusts the UAV flight path to the spatial position corresponding to the target grid, thereby realizing the linkage between risk prediction results and risk avoidance path adjustment and improving the safety of UAVs in low-altitude transportation.
[0046] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
Claims
1. A transportation risk prediction method based on big data analysis, characterized in that, Includes the following steps: S1. Acquire onboard battery operation data, flight control command data, environmental wind field prediction data, and ground gridded safety constraint data during the operation of the rotary-wing transport drone, and perform time and space alignment processing on the onboard battery operation data, flight control command data, environmental wind field prediction data, and ground gridded safety constraint data to generate a unified spatiotemporal data set. S2. Based on a unified spatiotemporal data set, features representing battery internal resistance change, wind field disturbance, and flight control current response are constructed respectively, and a three-source coupled state representation is constructed based on the synchronous evolution relationship of the three. Among them, the flight control current response feature is synchronously associated with the wind field disturbance feature and jointly mapped with the battery internal resistance change feature within a unified time window to form a three-source coupled state representation. S3. Input the three-source coupling state representation into the risk prediction model, and physically constrain the risk prediction model based on the voltage balance relationship between the battery open-circuit voltage, discharge current, internal resistance and polarization voltage, and output the failure probability value of the battery cell voltage falling below the safety threshold and its upper bound of the confidence interval within the future preset time window. S4. Based on the ground pedestrian density, building shading degree and flat area reduction ratio in the ground gridded safety constraint data, generate a candidate safety grid set; S5. Based on the failure probability value and its upper bound of the confidence interval, select candidate safety grids grid by grid and adjust the UAV flight path to the spatial position corresponding to the selected candidate safety grid.
2. The transportation risk prediction method based on big data analysis according to claim 1, characterized in that, In S1, the steps of acquiring onboard battery operation data, flight control command data, environmental wind field prediction data, and ground gridded safety constraint data during the operation of the rotary-wing transport drone include: Acquire data on the voltage, current, and temperature of individual airborne battery cells; Acquire throttle command data and attitude control command data from the flight control command data; Acquire horizontal wind speed data, wind direction data, and vertical turbulence data from environmental wind field prediction data; Obtain ground pedestrian density data, building shading data, and flat area reduction ratio data from the ground gridded safety constraint data.
3. The transportation risk prediction method based on big data analysis according to claim 1, characterized in that, In S1, the step of performing time and space alignment processing on airborne battery operation data, flight control command data, environmental wind field prediction data, and ground gridded safety constraint data to generate a unified spatiotemporal data set includes: The timestamps of airborne battery operation data, flight control command data, environmental wind field prediction data, and ground gridded safety constraint data are uniformly mapped to the flight clock reference. The environmental wind field prediction data and the ground gridded safety constraint data are transformed into a unified spatial coordinate system. Data with different sampling frequencies are resampled according to time windows and time slice sequences are generated. The airborne battery operation data, flight control command data, environmental wind field prediction data, and ground gridded safety constraint data in the time-slice sequence are spliced together at corresponding positions to form a unified spatiotemporal data set.
4. The transportation risk prediction method based on big data analysis according to claim 1, characterized in that, In S2, the step of constructing features representing battery internal resistance changes, wind field disturbances, and flight control current responses based on a unified spatiotemporal data set includes: Extract airborne battery operation data from a unified spatiotemporal dataset and construct individual battery cell voltage change sequences and discharge current sequences according to time series. Calculate the battery internal resistance variation sequence based on the battery cell voltage variation sequence and discharge current sequence; Extract environmental wind field prediction data from a unified spatiotemporal dataset and construct wind speed and wind direction data sequences based on flight location mapping; Extract flight control command data from a unified spatiotemporal dataset and construct throttle command sequence and current response sequence.
5. The transportation risk prediction method based on big data analysis according to claim 1, characterized in that, In S2, the step of constructing a three-source coupled state representation based on the synchronous evolution relationship of the three includes: The battery internal resistance change sequence, wind speed data sequence, and current response sequence are time-aligned according to the flight time reference. The battery internal resistance change sequence, wind speed data sequence, and current response sequence are sliced according to a unified time window after time alignment. Within each time window, establish the correspondence between the battery internal resistance change sequence and the current response sequence, as well as the correspondence between the wind speed data sequence and the current response sequence; A three-source coupled state representation is generated based on the correspondence within each time window.
6. The transportation risk prediction method based on big data analysis according to claim 1, characterized in that, In S3, the step of applying physical constraints to the risk prediction model based on the voltage balance relationship between battery open-circuit voltage, discharge current, internal resistance, and polarization voltage includes: Extract the battery open-circuit voltage, discharge current, internal resistance, and polarization voltage from the three-source coupling state representation; The ohmic voltage drop calculation term is constructed based on the battery open-circuit voltage, discharge current, and internal resistance. A polarization voltage drop calculation term is constructed based on the battery open-circuit voltage, discharge current, and polarization voltage. The ohmic voltage drop calculation term and the polarization voltage drop calculation term are subjected to consistency constraints with the cell voltage output by the risk prediction model.
7. The transportation risk prediction method based on big data analysis according to claim 1, characterized in that, In S3, the step of outputting the failure probability value of the battery cell voltage falling below the safety threshold and the upper bound of its confidence interval within a future preset time window includes: Based on the three-source coupling state representation, time series data is extracted according to a preset time window; Input time series data into the risk prediction model and calculate the state of voltage of individual battery cells step by step; A voltage drop event sequence is generated based on the comparison relationship between the voltage state of individual battery cells and the safety threshold. Failure probability values are generated based on the voltage drop event sequence statistics, and the upper bound of the corresponding confidence interval is calculated.
8. The transportation risk prediction method based on big data analysis according to claim 1, characterized in that, In S4, the step of generating the candidate security grid set includes: The ground area corresponding to the ground rasterized safety constraint data is divided into discrete raster cells and a raster index set is generated; Based on the raster index set, the ground pedestrian density, building shading degree and the proportion of flat and reducible area are mapped to raster cells; The safety score of each grid cell is calculated based on the ground pedestrian density, the building shielding degree, and the percentage of flat and retractable area. Based on the security score of each grid cell, grid cells are selected to form a candidate set of security grids.
9. The transportation risk prediction method based on big data analysis according to claim 1, characterized in that, In S5, the step of selecting the candidate safe grid set grid by grid based on the failure probability value and its upper bound confidence interval includes: Obtain the spatial coordinate information of each grid cell in the candidate safety grid set; The comprehensive selection weight of each grid cell is calculated based on the ground pedestrian density, building shading degree, and flat area reduction ratio corresponding to each grid cell, as well as the failure probability value and the upper limit of the confidence interval. Based on the comprehensive selection weights, each grid cell is filtered grid by grid. Sort the filtered raster cells and select the target raster cell.
10. The transportation risk prediction method based on big data analysis according to claim 1, characterized in that, In S5, the step of adjusting the UAV flight path to the spatial position corresponding to the selected candidate safety grid includes: Extract the spatial coordinates corresponding to the selected candidate safety grid and generate the target spatial location point; Connect the current spatial location of the drone with the target spatial location point by creating path nodes; Generate a continuous flight path sequence based on the path node connection relationship; The UAV flight path is updated and path adjustment is performed based on a continuous flight path sequence.
