A hydrogen energy unmanned aerial vehicle track energy consumption optimization system, method, device and medium

Abstract

The application discloses a hydrogen energy unmanned aerial vehicle track energy consumption optimization system, method, equipment and medium, and particularly relates to the field of unmanned aerial vehicle track planning and energy management, and comprises a data sensing module which acquires position, wind speed, power and the like data; a first processing unit generates a global energy consumption cost base map through three-level loop iteration and corrects the global energy consumption cost base map in a rolling window fine granularity, and obtains a local energy consumption correction base map; a second processing unit obtains a quality estimation strategy through three-level branch nested discrimination; a third processing unit calculates residual mass based on thrust-power coupling; a fourth processing unit obtains a resultant force vector through mass correction and parallel calculation; and an output module is converted into a desired track instruction. The application takes into account the global energy consumption optimization and local real-time adaptation, and performs adaptive correction on the variable mass characteristics caused by hydrogen fuel consumption, so that the energy efficiency of the whole flight range is effectively improved.

Description

Technical Field

[0001] This invention relates to the field of unmanned aerial vehicle (UAV) trajectory planning and energy management, and more specifically, to a hydrogen-powered UAV trajectory energy consumption optimization system, method, device, and medium. Background Technology

[0002] Hydrogen-powered drones are increasingly being used in long-endurance inspections and logistics transportation due to their advantages such as high energy density and long flight time. Their actual endurance depends not only on the performance of the fuel cell but also closely on flight path planning and energy management during flight. Achieving coordinated optimization of flight path and energy consumption in complex environments is a key issue in improving the operational efficiency of hydrogen-powered drones.

[0003] In the prior art, patent CN121608914A discloses a hydrogen-electric hybrid energy management method. This method uses a condition identification module to collect flight parameters to determine typical operating conditions, a condition monitoring module to collect energy parameters of the fuel cell and battery, and a decision module to construct a multi-dimensional decision matrix to dynamically match energy supply strategies, achieving coordinated energy management under various operating conditions. Patent CN121413883A discloses a multi-form hydrogen storage energy management optimization method. This method constructs physical models for three hydrogen storage forms: high-pressure gaseous state, cryogenic liquid state, and metal hydride. With the goal of minimizing hydrogen refueling costs and fuel cell lifespan loss, it uses the TD3 reinforcement learning algorithm to solve for the optimal power allocation strategy.

[0004] The aforementioned existing technologies have the following shortcomings: Decision matrices based on operational condition identification lack the ability to globally predict weather changes and terrain undulations throughout the flight, making them prone to getting trapped in local optima; global optimization based on reinforcement learning is highly dependent on model accuracy, consumes significant computational resources, and is difficult to deploy on air; neither approach considers the reverse impact of significant changes in overall aircraft mass due to hydrogen fuel consumption on the optimal flight path, meaning that the variable mass characteristics of heavy takeoff and light landing make it difficult for fixed-parameter planning strategies to maintain optimality throughout the entire flight. Therefore, a flight path resource scheduling scheme that balances global optimization and local adaptation while fully considering the variable mass flight characteristics is needed. Summary of the Invention

[0005] To overcome the aforementioned deficiencies of the prior art, the present invention provides a hydrogen-powered unmanned aerial vehicle (UAV) trajectory energy consumption optimization system, method, device, and medium, which solves the problems mentioned in the background art through the following solutions.

[0006] In a first aspect, the present invention provides a hydrogen-powered unmanned aerial vehicle (UAV) trajectory resource scheduling system based on energy consumption optimization, comprising: a data sensing module for acquiring signals and data, including: The pose sensing unit acquires position data, velocity data, and attitude angle data through inertial measurement devices and satellite positioning receivers deployed on the UAV body; The environmental perception unit acquires obstacle distribution data in the three-dimensional airspace through lidar and visual sensors deployed around the UAV. The meteorological sensing unit acquires wind speed vector data and atmospheric disturbance intensity data through pitot tubes and ultrasonic anemometers deployed on the nose and wings of the aircraft. The power sensing unit acquires motor speed data, tension output data, and power data through high-frequency current sensors and tension sensors deployed on the drive motor bus and fuel cell output end; The data processing module is used to perform multi-level loop processing on signals and data; The processing module includes: The first processing unit is used to generate a global energy consumption cost base map by combining location data, wind speed vector data, and power data through a three-level iterative cycle. The global energy consumption cost base map is then finely corrected online and updated in the background using a scrolling window to obtain a local energy consumption correction base map. The second processing unit is used to obtain the quality estimation strategy instruction by discriminating the position data, attitude angle data, atmospheric disturbance intensity data and obstacle distribution data through a three-level branch nesting process. The third processing unit is used to calculate the remaining mass data by combining the mass estimation strategy command, motor speed data, power data and thrust output data through the thrust-power response coupling relationship. The fourth processing unit is used to obtain the resultant force vector data by parallel calculation of the remaining mass data, local energy consumption correction base map, obstacle distribution data and wind speed vector data through mass correction and spatial region decomposition. The output module is used to convert the resultant force vector data into the desired trajectory command through the resultant force-track conversion and output it to the flight control system.

[0007] Preferably, the three-level loop iteration includes: The first-level loop iterates and generates a global energy consumption cost base map based on historical meteorological data and terrain elevation data with the first grid precision before flight. The second-level loop constructs a rolling window centered on the position data during flight, downscales the global energy cost base map to the second grid precision, and updates and corrects the energy consumption gradient within the rolling window online based on power data and wind speed vector data. The third-level loop receives measured energy consumption data from multiple machines via the communication link and periodically performs background Bayesian fusion updates on the global energy consumption cost map. The precision of the first grid is lower than that of the second grid.

[0008] Preferably, the three-level branch nesting detection includes: The first-level branch outputs the cruise mode identifier when the rate of change of the attitude angle data is less than the first threshold; otherwise, it outputs the maneuver mode identifier. The second-level branch outputs a statically stable cruise sub-condition, a slightly turbulent sub-condition, or a strongly turbulent sub-condition based on the high-frequency component of the power spectrum of atmospheric disturbance intensity data under the cruise mode identifier; and outputs a coordinated turning sub-condition or a non-coordinated sideslip sub-condition based on the sideslip angle data under the maneuvering mode identifier. The third branch outputs an active excitation estimation strategy based on the static steady cruise sub-condition, a passive filtering observation strategy based on the mild turbulence sub-condition, and an open-loop maintenance strategy based on the strong turbulence sub-condition or the sub-condition under the maneuvering mode.

[0009] Preferably, the remaining mass data obtained by solving the thrust-power response coupling relationship specifically includes: In the static and stable cruise sub-condition, a throttle excitation signal of a preset frequency is actively injected, and motor speed and power data are collected in the corresponding time period. The amplitude offset of the power response is extracted by phase-locked amplification, and the current total mass is solved based on the rotor aerodynamic model. The remaining mass data is obtained by comparing it with the takeoff mass.

[0010] Preferably, the mass correction specifically involves scaling the repulsive force weight coefficient in the local potential field model proportionally based on the ratio of the remaining mass data to the total takeoff mass, and exponentially correcting the potential field step size parameter.

[0011] Preferably, the parallel computation of spatial region decomposition specifically includes: dividing the rolling window into multiple subspace blocks according to an octree structure, allocating them to different cores of the onboard multi-core processor to perform parallel computation of the obstacle repulsion field, target gravitational field and airflow utilization force field in each sub-block, and obtaining the resultant force vector data through reduction and summation.

[0012] Preferably, the method for resultant force-track conversion specifically includes: using the resultant force vector data as the desired acceleration vector for the next control cycle, integrating it to obtain the desired velocity and desired position sequence, generating attitude angle commands and throttle commands through attitude calculation, and forming the desired track command.

[0013] Secondly, the present invention provides a method for scheduling hydrogen-powered UAV flight paths based on energy consumption optimization, comprising the following steps: S1. Acquire signals and data through sensing modules deployed on the drone body, including: acquiring position data, speed data, and attitude angle data through the pose perception unit; acquiring obstacle distribution data through the environmental perception unit; acquiring wind speed vector data and atmospheric disturbance intensity data through the meteorological perception unit; and acquiring motor speed data, thrust output data, and power data through the power perception unit. S2. Input the location data, wind speed vector data and power data into the first processing unit, generate a global energy consumption cost base map through a three-level loop iteration, and perform fine-grained online correction and background fusion update on the global energy consumption cost base map in a scrolling window manner to obtain a local energy consumption correction base map. S3. Input the position data, attitude angle data, atmospheric disturbance intensity data and obstacle distribution data into the second processing unit, and obtain the quality estimation strategy instruction through three-level branch nesting discrimination; S4. Input the mass estimation strategy command, motor speed data, power data and thrust output data into the third processing unit, and calculate the current remaining mass data through the thrust-power response coupling relationship; S5. Input the current remaining mass data, local energy consumption correction base map, obstacle distribution data and wind speed vector data into the fourth processing unit, and obtain the resultant force vector data through parallel calculation of mass correction and spatial region decomposition. S6. Input the resultant force vector data into the output module, obtain the desired trajectory command through the resultant force-track conversion, and output it to the flight control system to achieve coordinated scheduling of trajectory and energy consumption.

[0014] The technical effects and advantages of this invention are as follows: 1. This invention generates a global energy consumption cost map through a three-level iterative cycle and corrects it online with a rolling window, enabling hydrogen-powered UAV trajectory planning to combine macroscopic optimal hydrogen consumption guidance with real-time adaptation to local wind fields. Compared with existing operating condition identification schemes, this invention can predict the impact of wind speed in the airspace ahead on the output power of hydrogen fuel cells, guiding the UAV to actively glide with the wind to reduce the hydrogen fuel consumption rate, effectively avoiding the problem of suboptimal local hydrogen consumption caused by limited visibility; 2. This invention employs a three-level branch nested discrimination and thrust-power coupled solution to actively excite and accurately estimate the remaining mass after hydrogen consumption during static steady cruise, passively filter and observe during turbulent conditions, and maintain an open-loop system during strong disturbances. This method does not rely on the electrochemical model of the hydrogen fuel cell or the pressure-capacity curve of the hydrogen storage tank; it directly solves the change in overall engine mass from the rotor aerodynamic characteristics to determine the consumed hydrogen mass, significantly reducing the onboard computational burden and adapting to the performance degradation of the hydrogen fuel cell. 3. This invention feeds back the remaining mass data (i.e., hydrogen consumption) to the potential field model, dynamically scaling the repulsive force weights and step size parameters. When hydrogen fuel consumption reduces the overall mass of the aircraft, the system automatically shrinks the range of the obstacle repulsive potential field, causing the flight path to automatically approach the obstacle as hydrogen fuel is consumed, shortening the remaining range. This fully leverages the low hydrogen consumption maneuverability advantage of hydrogen-powered UAVs during the light-load phase, solving the problems of conservative flight paths and high hydrogen consumption caused by neglecting the variable mass characteristics of hydrogen fuel consumption in existing technologies. Attached Figure Description

[0015] Figure 1This is a block diagram of the overall system architecture of the present invention; Figure 2 This is a flowchart of the three-level loop iteration of the present invention; Figure 3 This is a three-level branch nested discrimination tree diagram of the present invention; Figure 4 This is a schematic diagram of the thrust-power response coupling solution of the present invention; Figure 5 This is a diagram of the parallel computing architecture for spatial region decomposition of the present invention; Figure 6 This is an overall flowchart of the method of the present invention; Figure 7 This is a schematic diagram comparing the repulsive potential field before and after the mass correction of the present invention; Figure 8 This is a timing diagram of the three-level cyclic iterative data flow of the present invention. Detailed Implementation

[0016] The technical solutions of 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. Example 1

[0017] As attached Figure 1 As shown, this embodiment provides a hydrogen-powered UAV trajectory resource scheduling system based on energy consumption optimization, including a data sensing module, a data processing module, and an output module. The data sensing module acquires various signals and data during the UAV's flight, the data processing module performs multi-level cyclic processing on the signals and data, and the output module generates the desired trajectory command and outputs it to the flight control system.

[0018] The specific implementation method is as follows:

[0019] The data sensing module consists of four parts: pose sensing unit, environment sensing unit, meteorological sensing unit, and dynamic sensing unit.

[0020] The attitude perception unit acquires the UAV's position, velocity, and attitude angle data in real time through an inertial measurement unit (IMU) and a satellite positioning receiver deployed on the UAV's fuselage. The IMU preferably uses a nine-axis IMU, including a three-axis accelerometer, a three-axis gyroscope, and a three-axis magnetometer, providing complete attitude calculation information. The satellite positioning receiver preferably supports GPS and BeiDou dual-mode positioning, with a positioning accuracy better than 1.5 meters and a data update frequency ≥10Hz.

[0021] The environmental perception unit acquires obstacle distribution data in the three-dimensional airspace through lidar and visual sensors deployed around the UAV. The lidar is preferably a mechanically rotating multi-line lidar with a horizontal field of view of 360°, a vertical field of view of ≥30°, and a detection range of ≥100 meters, capable of constructing a three-dimensional point cloud map of the surrounding environment in real time. The visual sensor is preferably a binocular stereo camera, which serves as a supplement to the lidar and is used to identify obstacles with weak textures or strong reflectivity.

[0022] The meteorological sensing unit acquires wind speed vector data and atmospheric disturbance intensity data through pitot tubes, ultrasonic anemometers, and sideslip angle sensors deployed on the nose and wings. The pitot tube measures the UAV's flight speed relative to the air; the ultrasonic anemometer calculates the three-dimensional wind speed vector by measuring the change in the propagation time of ultrasonic waves in the air, with a measurement range of 0 to 30 m / s, an accuracy better than 0.3 m / s, and an update frequency ≥20 Hz; the sideslip angle sensor directly measures the angle between the aircraft's longitudinal axis and the direction of the relative airflow, i.e., the sideslip angle data.

[0023] The power sensing unit acquires motor speed data, tension output data, and power data through high-frequency current sensors and tension sensors deployed on the drive motor bus and fuel cell output. The high-frequency current sensor uses the Hall effect principle, with a sampling frequency ≥1kHz, and can capture transient changes in motor current; the tension sensor is installed at the connection between the motor and the arm to directly measure the tension value generated by the rotor; a power meter is installed at the fuel cell output to monitor the fuel cell's output voltage, current, and power in real time.

[0024] The data processing module includes a first processing unit, a second processing unit, a third processing unit, and a fourth processing unit. The four processing units exchange data through an onboard data bus to form a complete data processing pipeline.

[0025] As attached Figure 1 and attached Figure 2 As shown, the first processing unit in this embodiment is used to generate a global energy consumption cost base map by generating the location data, wind speed vector data and power data through a three-level iterative cycle, and to perform fine-grained online correction and background fusion update on the global energy consumption cost base map in a scrolling window manner to obtain a local energy consumption correction base map.

[0026] The specific implementation method is as follows:

[0027] The first-level loop is executed offline before flight. On the ground station computer, historical meteorological data and terrain elevation data of the mission airspace are first acquired. Historical meteorological data can be obtained from regional wind field statistics released by meteorological departments, including average wind speed, wind direction and their probability distribution in different seasons and time periods; terrain elevation data can be obtained from publicly available DEM digital elevation models, with a spatial resolution preferably of 30 meters. Then, the mission airspace is divided into three-dimensional grids, with the first grid having a preferred accuracy of 100m×100m×20m. For each grid node, the equivalent energy consumption value of that node is calculated based on the hovering power, level flight power and additional power required to overcome wind resistance of the UAV at that node. The specific calculation model is as follows: hovering power is determined by the current total mass of the UAV and rotor efficiency; level flight power is determined by flight speed and fuselage aerodynamic drag coefficient; wind resistance additional power is determined by the synthesis relationship between the wind speed vector and the UAV ground speed vector. An iterative relaxation algorithm is used to propagate backward from the target point, gradually updating the minimum cumulative energy consumption value of each grid node until the maximum change in the entire map is less than 1% of the initial maximum value, at which point the iteration terminates. After the iteration is completed, a global energy consumption cost base map is obtained. The global energy consumption cost base map means that each grid node is assigned an energy consumption cost value. The cost value reflects the estimated minimum cumulative energy consumption from the node to the target point. The lower the cost value, the less hydrogen fuel energy is required to fly from the node to the target point.

[0028] The second-level loop is executed online during flight. After the UAV takes off, the first processing unit constructs a scrolling window centered on the real-time position data. The size of the scrolling window is preferably 500m × 500m × 100m, and it moves synchronously with the UAV. The global energy consumption cost base map within the scrolling window is downscaled to a second grid precision, preferably 10m × 10m × 5m, significantly higher than the first grid precision. The downscaling process uses bilinear interpolation or Kriging interpolation to ensure the continuity of the energy consumption gradient. Subsequently, based on the real-time power data provided by the dynamic sensing unit and the wind speed vector data provided by the meteorological sensing unit, the energy consumption gradient of each fine grid node within the scrolling window is updated and corrected online. The correction algorithm uses Gaussian process regression, using real-time observed power-position-wind speed correlation data as training samples to locally adjust the prior energy consumption base map. The update frequency is preferably 20Hz, consistent with the main loop frequency of the flight control system. The corrected fine grid energy consumption map is the local energy consumption correction base map, which is directly used for subsequent potential field calculations.

[0029] The third-level loop executes in the background in multi-drone collaborative scenarios. When multiple hydrogen-powered drones of the same model are performing missions in the same airspace, each drone transmits its measured energy consumption data, location data, and meteorological data back to the ground station or nearby drones via a communication link. The communication link preferably uses a 4G or 5G mobile communication network with a data transmission frequency of 1Hz. The ground station or a designated lead drone periodically initiates a background batch processing loop, preferably every 10 minutes. The collected multi-drone measured energy consumption data is then updated using Bayesian fusion with the existing global energy cost base map. The fusion algorithm employs a Bayesian inference framework, using the existing base map as the prior distribution and the new observation data as the likelihood function to calculate the posterior distribution and update the energy consumption values ​​and confidence intervals of each node in the base map. The updated base map is broadcast to all drones via the communication link, enabling continuous evolution and accuracy improvement of the base map.

[0030] As attached Figure 8 The diagram shows the data flow time sequence of the three-level iterative loop in this embodiment. The first-level loop is executed offline once at time T0 before flight, generating an initial global energy cost base map based on historical meteorological and terrain data. After the UAV takes off, at time T1, the second-level loop starts, running online at a high frequency of 20Hz. Within each control cycle, a rolling window is built centered on the current position to perform fine-grained corrections on the base map and output a local energy consumption correction base map, which is directly used for potential field calculation. The third-level loop is executed asynchronously in the background during flight, with an update cycle of 10 minutes. It collects measured energy consumption data from multiple UAVs, performs Bayesian fusion, and feeds the update results back to the global energy cost base map and broadcasts them to each UAV for reference by the second-level loop in subsequent flights. The three-level loop forms a closed loop of "offline generation - online correction - background evolution," enabling the energy consumption base map to be continuously optimized as the flight mission is executed, becoming more accurate with use.

[0031] As attached Figure 1 and attached Figure 3 As shown, the second processing unit in this embodiment is used to obtain a quality estimation strategy instruction by using the position data, attitude angle data, atmospheric disturbance intensity data, and obstacle distribution data through a three-level branch nesting discrimination.

[0032] The specific implementation method is as follows:

[0033] The second processing unit first receives real-time position and attitude angle data from the pose sensing unit, atmospheric disturbance intensity data from the meteorological sensing unit, and obstacle distribution data from the environmental sensing unit. The specific logic of the three-level branch nested discrimination is as follows.

[0034] The first-level branch discrimination is based on the rate of change of attitude angle data. Attitude angles include pitch, roll, and yaw, and the rate of change is obtained by calculating the difference between adjacent sampling periods. A first threshold of 5° / s is set. This threshold was determined through statistical analysis of a large amount of flight test data and can effectively distinguish between the UAV's stable cruise state and active maneuver state. When the rate of change of attitude angles is <5° / s, the second processing unit outputs a cruise mode identifier; when the rate of change of attitude angles is ≥5° / s, it outputs a maneuver mode identifier.

[0035] The second-level branch discrimination executes different discrimination logic under the cruise mode identifier and the maneuver mode identifier respectively. Under the cruise mode identifier, the second processing unit reads the high-frequency component of the power spectrum of the atmospheric disturbance intensity data. The high-frequency component of the power spectrum is obtained by performing a Fast Fourier Transform (FFT) on the atmospheric disturbance intensity time series and integrating the energy in the frequency band >1Hz, denoted as […]. The second threshold is set to 0.5m. 2 / s 3 The third threshold is 2.0m. 2 / s 3 .when <0.5m 2 / s 3 When this occurs, it indicates the presence of strong atmospheric disturbance, and the output indicates a strong turbulent subcondition. Under the maneuvering mode identifier, the second processing unit processes the data based on the sideslip angle. The sideslip angle is determined by direct measurement from sideslip angle sensors (or weather vanes) deployed on the nose or wing. When the angle is less than 3°, the UAV is considered to be in a coordinated turn state, and a coordinated turn sub-condition is output; when... When the angle is ≥3°, the UAV is considered to be in an uncoordinated sideslip state, and the uncoordinated sideslip sub-condition is output.

[0036] The third-level branch discrimination outputs corresponding mass estimation strategy instructions based on the aforementioned sub-conditions. When the sub-condition is static stable cruise, an active excitation estimation strategy instruction is output, instructing the third processing unit to perform high-precision mass estimation by actively injecting throttle excitation signals. When the sub-condition is mild turbulence, a passive filtering observation strategy instruction is output, instructing the third processing unit to perform disturbance-free passive observation estimation using an extended Kalman filter. When the sub-condition is strong turbulence, coordinated turning, or uncoordinated sideslip, an open-loop maintenance strategy instruction is output, instructing the third processing unit to pause mass estimation updates and maintain the most recent valid estimate. The discrimination logic of the third-level branch is based on the following principles: in the static stable cruise sub-condition with the most stable flight state and lowest environmental noise, the most accurate but somewhat disturbed active excitation method is used; in the mild turbulence sub-condition with moderate noise, a disturbance-free but slightly less accurate passive filtering method is used; in high noise or high maneuvering conditions, to avoid erroneous estimation, updates are directly frozen, relying on the system's own robustness to maintain operation.

[0037] As attached Figure 1 and attached Figure 4 As shown, the third processing unit in this embodiment is used to calculate the remaining mass data (i.e., the current total mass of the UAV) by solving the mass estimation strategy command, motor speed data, power data and thrust output data through the thrust-power response coupling relationship.

[0038] The specific implementation method is as follows:

[0039] The third processing unit receives the quality estimation strategy instruction output by the second processing unit and executes the corresponding solution program according to the instruction type.

[0040] When the mass estimation strategy command is an active excitation estimation strategy, the third processing unit first confirms that the UAV is currently in a static steady cruise sub-condition, and then sends a throttle excitation command to the flight control system. The throttle excitation signal is preferably in the form of a sine wave with a frequency set to 2Hz. This frequency is much higher than the response bandwidth of the UAV flight control system, so it will not have a significant impact on the flight attitude, while being much lower than the cutoff frequency of the motor current loop, ensuring effective execution. The amplitude of the excitation signal is set to 3% of the current throttle value to ensure that the excitation amplitude is sufficient to produce an observable power response without causing significant flight speed fluctuations. Simultaneously with the injection of the excitation signal, the third processing unit synchronously acquires motor speed and power data at a sampling rate of 1kHz, with an acquisition duration preferably of 2 seconds, including at least four complete excitation signal cycles.

[0041] After data acquisition, the third processing unit performs phase-locked amplification (PLA) on the power data. Using the injected 2Hz sinusoidal signal as a reference, the PLA performs phase-sensitive detection and low-pass filtering on the power signal to accurately extract the amplitude and phase information of the 2Hz frequency component from the noise background. This amplitude is the amplitude offset of the power response, denoted as . This reflects the motor's sensitivity to throttle input.

[0042] According to rotor aerodynamics theory, in hovering or low-speed level flight, the rotor thrust T is proportional to the square of the motor speed n, and the motor input power P is proportional to the cube of the speed n. When the total mass of the UAV decreases due to hydrogen fuel consumption, the thrust required to maintain the same flight state decreases, and the corresponding motor speed and power also decrease. Under the condition of injecting the same amplitude throttle excitation signal, the power response amplitude offset of a lightly loaded UAV will be greater than that of a heavily loaded UAV. The third processing unit has a built-in pre-calibrated rotor aerodynamic model, which was established through ground bench experiments and recorded the mapping relationship of power response amplitude offset under different total masses and different throttle excitation amplitudes. In one embodiment, the model is a monotonically decreasing function in the calibration interval, for example, the product of coefficients a and b is negative. The model can be expressed by the fitting formula as: , Where m is the current total mass of the drone. As the baseline throttle value, The extracted 2Hz power response amplitude offset is given by a, b, c, and d, which are constant coefficients determined through ground bench experiments under different mass counterweights and throttle conditions. These coefficients are calculated by referring to tables or substituting into formulas, derived from the measured values. The current total mass m of the drone is obtained by inverse solving. This current total mass is the remaining mass after the hydrogen fuel is consumed.

[0043] In a specific calibration embodiment: the UAV is fixed to a ground test bench, and counterweights are sequentially loaded at 100%, 90%, 80%, and 70% of the total takeoff mass. A baseline throttle is set at each mass. At 50%, inject a 2Hz sinusoidal excitation signal with an amplitude of 3% of the current throttle, and record the corresponding power response amplitude offset. The multiple sets obtained By fitting the data points to a curve using the least squares method, the specific values ​​of the constants a, b, c, and d can be determined.

[0044] When the mass estimation strategy command is a passive filtering observation strategy, the third processing unit does not actively inject excitation signals. Instead, it uses an extended Kalman filter to perform sliding window fusion estimation on the motor speed data, power data, and thrust output data. The state vector of the extended Kalman filter includes the total mass of the UAV and motor efficiency parameters, while the observation vector includes motor speed, power, and thrust. The sliding window width is preferably 10 seconds, and the filter update frequency is 20Hz. Through recursive estimation over multiple consecutive cycles, the estimation gradually converges to the current remaining mass data.

[0045] When the mass estimation strategy instruction is an open-loop maintenance strategy, the third processing unit freezes the mass estimation update operation, directly outputs the most recent valid remaining mass estimate as the current remaining mass data, and simultaneously enables open-loop compensation based on the motor temperature sensor to cope with the small deviations that may be caused by drastic changes in flight status.

[0046] As attached Figure 1 and attached Figure 5 As shown, the fourth processing unit in this embodiment is used to obtain the resultant force vector data by parallel calculation of the remaining mass data, local energy consumption correction base map, obstacle distribution data and wind speed vector data through mass correction and spatial region decomposition.

[0047] The specific implementation method is as follows:

[0048] The fourth processing unit first performs quality correction. This is based on the remaining quality data output by the third processing unit. Calculate the current total mass and the total mass at takeoff. The ratio of the mass decay rate to the mass degradation rate is called the mass decay rate. : .

[0049] With the continued consumption of hydrogen fuel, The mass typically varies between 0.7 and 1.0. The lighter the mass, the lower the inertia of the drone, and the stronger its maneuverability. Based on this, the fourth processing unit performs adaptive mass correction on the repulsive force weight coefficient and step size parameter in the local potential field model according to the mass decay ratio. The local potential field model adopts the basic framework of the artificial potential field method, where the formula for calculating the obstacle repulsive force field is: , in, This is the repulsive force weighting coefficient. Let i be the minimum distance between the drone and the i-th obstacle. The distance decay function is preferably... , Let be the unit vector pointing from the obstacle to the drone.

[0050] The fourth processing unit will use the repulsion weighting coefficient. Ratio of mass decay Perform scaling: , Meanwhile, the potential field step size parameter Perform index correction: , This means that lightly loaded drones experience less virtual repulsion and have a larger planning step size, allowing them to plan shorter flight paths that are closer to obstacles, thereby further shortening flight distances and saving hydrogen fuel consumption.

[0051] As attached Figure 7The diagram shows a comparison of the repulsive potential field range before and after mass correction in this embodiment. During takeoff, with a full hydrogen fuel load, the UAV is relatively heavy, and the mass decay ratio η ≈ 1.0. The repulsive weight coefficient remains at the baseline value, resulting in a large repulsive potential field range around obstacles, allowing the planned flight path to maintain a safe distance from obstacles. In the later stages of flight, as the hydrogen fuel cell continues to consume hydrogen, the overall mass decreases significantly, and the mass decay ratio η drops to approximately 0.7 to 0.8. The repulsive weight coefficient scales down proportionally, and the repulsive potential field range around the same obstacle shrinks accordingly. At this point, the mass-corrected potential field model allows the UAV to fly closer to obstacles, effectively shortening the remaining flight distance and further reducing the hydrogen fuel consumption rate. This adaptive mechanism fully utilizes the physical characteristics of variable-mass flight in hydrogen-powered UAVs, ensuring that the flight path planning always matches the current remaining hydrogen quantity throughout the entire flight.

[0052] After quality correction is completed, the fourth processing unit performs parallel computation of spatial region decomposition. Centered on the current location and bounded by the effective range of the local energy consumption correction base map, a computation window of the same size as the scrolling window is constructed. This computation window is recursively divided into multiple sub-blocks using an octree spatial data structure. The recursive partitioning condition of the octree is: if the number of obstacle point clouds contained in a sub-block exceeds a preset threshold, or the energy consumption gradient change rate of the sub-block exceeds a preset threshold, it is further divided into eight smaller sub-blocks; otherwise, partitioning stops. Finally, an octree with a depth ≤ 5 layers is formed, and the leaf nodes are the sub-blocks to be computed in parallel.

[0053] The aforementioned subspace blocks are allocated to different cores of the onboard multi-core processor for parallel computation. The onboard multi-core processor preferably employs a heterogeneous architecture, comprising four high-performance ARM Cortex-A53 cores and two real-time ARM Cortex-R5 cores. The A53 cores handle the parallel computation of the potential field, while the R5 cores handle real-time communication with the flight control system. The computational task of each sub-block consists of three parts: 1. Obstacle repulsive force field calculation: Based on the distance and direction of obstacles within the sub-block, and using the mass-corrected repulsive force weighting coefficient... Calculate the repulsion vector .

[0054] 2. Target gravitational field calculation: Calculate the gravitational vector based on the relative position of the target point and the center of the sub-block. .

[0055] 3. Airflow Utilization Force Field Calculation: Based on wind speed vector data, when a downwind component with an angle <45° to the target direction is detected in the local airspace, an additional tangential gravitational vector is generated. This guides drones to actively utilize wind power for gliding to reduce energy consumption.

[0056] After each core completes its sub-block calculations, the results are aggregated in the main core for reduction and summation. The reduction and summation uses a pairwise reduction method, merging the force vectors calculated by each sub-block step by step, ultimately obtaining the resultant force vector data acting on the current center of mass of the UAV. : , This resultant force vector comprehensively reflects the guidance information of the globally optimal energy consumption path, the need to avoid local obstacles, and the optimization strategy of using natural wind power for energy saving. It is the final decision output of the coordinated optimization of trajectory planning and energy consumption.

[0057] The output module is connected to the fourth processing unit of the data processing module to receive the resultant force vector data. The desired trajectory command is obtained through the resultant-track conversion and output to the flight control system.

[0058] The specific implementation method is as follows:

[0059] The output module uses the resultant force vector data as the desired acceleration vector for the UAV's next control cycle. : ; For the desired acceleration vector Time integration is performed to obtain the desired velocity vector; time integration is then performed again on the desired velocity vector to obtain the desired position sequence. The integration process employs a numerical integration method, with the integration step size consistent with the control cycle. From the desired velocity vector and the desired position sequence, desired attitude angle commands for the UAV are generated through attitude calculation, including pitch, roll, and yaw commands. Simultaneously, based on the magnitude and direction of the desired acceleration vector, the throttle commands for each rotor motor are calculated. The attitude angle commands and throttle commands together constitute the desired trajectory commands, which are transmitted to the flight control system for execution via the onboard data bus.

[0060] The flight control system, based on the received desired flight path command, drives each rotor motor to output throttle at a specified throttle and adjusts the UAV's attitude via servo control surfaces, enabling the UAV to fly along the planned trajectory. Throughout the flight, the data sensing module continuously collects data, the data processing module continuously processes the data in a loop, and the output module continuously generates updated desired flight path commands, forming a complete closed-loop control system.

[0061] As attached Figure 6 As shown, this embodiment also provides a hydrogen-powered UAV trajectory resource scheduling method based on energy consumption optimization. This method is applied to the above system and specifically includes the following steps: Step S1: Acquire signals and data through the sensing modules deployed on the drone body. Specifically: acquire position data, velocity data, and attitude angle data through the pose perception unit; acquire obstacle distribution data through the environment perception unit; acquire wind speed vector data and atmospheric disturbance intensity data through the weather perception unit; and acquire motor speed data, thrust output data, and power data through the power perception unit.

[0062] Step S2: Input the location data, wind speed vector data, and power data into the first processing unit, generate a global energy consumption cost base map through a three-level iterative cycle, and perform fine-grained online correction and background fusion update on the global energy consumption cost base map in a scrolling window manner to obtain a local energy consumption correction base map.

[0063] Step S3: Input the position data, attitude angle data, atmospheric disturbance intensity data and obstacle distribution data into the second processing unit, and obtain the quality estimation strategy instruction through three-level branch nesting discrimination.

[0064] Step S4: Input the mass estimation strategy command, motor speed data, power data and thrust output data into the third processing unit, and calculate the current remaining mass data (i.e. the current total mass of the UAV) through the thrust-power response coupling relationship.

[0065] Step S5: Input the current remaining mass data, local energy consumption correction base map, obstacle distribution data and wind speed vector data into the fourth processing unit, and obtain the resultant force vector data through parallel calculation of mass correction and spatial region decomposition.

[0066] Step S6: Input the resultant force vector data into the output module, obtain the desired trajectory command through resultant force-track conversion, and output it to the flight control system to achieve coordinated scheduling of trajectory and energy consumption.

[0067] The above steps S1 to S6 are executed sequentially in a loop. After each execution of S6, the system returns to S1 to start a new control cycle until the UAV completes its flight mission and lands.

[0068] This embodiment achieves deep coupling between trajectory planning and energy management for hydrogen-powered UAVs by constructing a technical route of "global energy consumption cost base map - local rolling correction - mass adaptive potential field - parallel resultant force calculation". On the one hand, the three-level iterative cycle transforms the energy consumption base map from static to dynamic self-evolving, solving the problem of excessive reliance on offline models in traditional methods. On the other hand, the combination of three-level branch nested discrimination and thrust-power coupled calculation achieves real-time accurate estimation of remaining mass without relying on high-precision electrochemical models. Furthermore, mass correction dynamically adjusts the potential field parameters with fuel consumption, fully adapting to the physical characteristics of variable-mass flight of hydrogen-powered UAVs. Finally, spatial region decomposition and parallel computing ensure the real-time executability of complex algorithms on an airborne embedded platform. Example 2

[0069] Building upon Example 1, this example further optimizes the execution strategy for background Bayesian fusion updates in the third-level loop. When one of the participating drones detects a significant deviation between its sensor data and the current global energy cost base map (e.g., the relative error between measured power and base map predicted power > 15%), the drone proactively initiates a base map update request to the ground station. Upon receiving the request, the ground station immediately initiates a priority-expedited background fusion update, instead of waiting for a fixed 10-minute cycle. This mechanism enables the global energy cost base map to respond quickly to sudden weather changes, further improving the system's adaptability in complex dynamic environments.

[0070] Secondly, when the passive filtering observation strategy is executed in the third processing unit, this embodiment adopts an adaptive sliding window width mechanism. By calculating the stability index of the flight state (such as the sliding standard deviation of the attitude angle change rate), when the index is lower than a preset threshold for 5 consecutive seconds, the flight state is determined to be stable, and the sliding window width is automatically shortened to 5 seconds to improve the estimation response speed; when the stability index is higher than the preset threshold, a disturbance is determined to exist, and the sliding window width is automatically extended to 15 seconds to enhance the stability and noise resistance of the estimation.

[0071] Furthermore, during the quality correction performed by the fourth processing unit, this embodiment further introduces compensation for the fuel cell performance degradation factor. This is achieved by recording the cumulative operating time of the fuel cell. The attenuation factor is calculated based on the preset attenuation curve. The value of this factor ranges from 1.0 to 1.2. The correction amount for the repulsive force weight coefficient is fine-tuned, i.e.: , This is to compensate for the impact of changes in power output characteristics caused by fuel cell aging on the accuracy of mass estimation and trajectory planning.

[0072] Secondly: The accompanying drawings of the embodiments disclosed in this invention only involve the structures involved in the embodiments disclosed in this invention. Other structures can refer to the general design. In the absence of conflict, the same embodiment and different embodiments of this invention can be combined with each other. In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. 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 hydrogen energy unmanned aerial vehicle track energy consumption optimization system, characterized in that, include: The data sensing module, used to acquire signals and data, includes: The pose sensing unit acquires position data, velocity data, and attitude angle data through inertial measurement devices and satellite positioning receivers deployed on the UAV body; The environmental perception unit acquires obstacle distribution data in the three-dimensional airspace through lidar and visual sensors deployed around the UAV. The meteorological sensing unit acquires wind speed vector data and atmospheric disturbance intensity data through pitot tubes and ultrasonic anemometers deployed on the nose and wings of the aircraft. The power sensing unit acquires motor speed data, tension output data, and power data through high-frequency current sensors and tension sensors deployed on the drive motor bus and fuel cell output end; The data processing module is used to perform multi-level cyclic processing on the signals and data; The processing module includes: The first processing unit is used to generate a global energy consumption cost base map by taking the location data, wind speed vector data and power data through a three-level loop iteration, and to perform fine-grained online correction and background fusion update on the global energy consumption cost base map in a scrolling window manner to obtain a local energy consumption correction base map. The second processing unit is used to obtain a quality estimation strategy instruction by using the position data, attitude angle data, atmospheric disturbance intensity data and obstacle distribution data through a three-level branch nesting discrimination. The third processing unit is used to calculate the remaining mass data by solving the mass estimation strategy command, motor speed data, power data and thrust output data through the thrust-power response coupling relationship. The fourth processing unit is used to obtain the resultant force vector data by parallel calculation of the remaining mass data, local energy consumption correction base map, obstacle distribution data and wind speed vector data through mass correction and spatial region decomposition. The output module is used to convert the resultant force vector data into a desired trajectory command through resultant force-track conversion and output it to the flight control system. 2.The hydrogen energy unmanned aerial vehicle route resource scheduling system based on energy consumption optimization of claim 1, wherein, The three-level loop iteration includes: The first-level loop iteratively generates the global energy consumption cost base map based on historical meteorological data and terrain elevation data with the first grid precision before flight. In the second-level loop, a rolling window is constructed centered on the position data during flight. The global energy consumption cost base map is scaled down to the second grid precision, and the energy consumption gradient within the rolling window is updated and corrected online based on the power data and the wind speed vector data. The third-level loop receives measured energy consumption data transmitted back from multiple machines through the communication link and periodically performs background Bayesian fusion updates on the global energy consumption cost base map. The precision of the first grid is lower than that of the second grid. 3.The hydrogen energy unmanned aerial vehicle route resource scheduling system based on energy consumption optimization of claim 1, wherein, The three-level branch nesting detection includes: The first-level branch outputs a cruise mode identifier when the rate of change of the attitude angle data is less than a first threshold; otherwise, it outputs a maneuver mode identifier. The second-level branch outputs a statically stable cruise sub-condition, a slightly turbulent sub-condition, or a strongly turbulent sub-condition based on the high-frequency component of the power spectrum of the atmospheric disturbance intensity data under the cruise mode identifier; and outputs a coordinated turning sub-condition or a non-coordinated sideslip sub-condition based on the sideslip angle data under the maneuvering mode identifier. The third-level branch outputs an active excitation estimation strategy based on the statically stable cruise sub-condition, a passive filtering observation strategy based on the mildly turbulent sub-condition, and an open-loop maintenance strategy based on the sub-condition under the strong turbulent sub-condition or maneuvering mode.

4. The hydrogen energy unmanned aerial vehicle route resource scheduling system based on energy consumption optimization according to claim 1, characterized in that, The method of obtaining the remaining mass data through the thrust-power response coupling relationship specifically includes: In the static and stable cruise sub-condition, a throttle excitation signal of a preset frequency is actively injected, and motor speed and power data are collected in the corresponding time period. The amplitude offset of the power response is extracted by phase-locked amplification, and the current total mass is solved based on the rotor aerodynamic model. The remaining mass data is obtained by comparing it with the takeoff mass.

5. The energy consumption optimization-based hydrogen energy unmanned aerial vehicle route resource scheduling system according to claim 1, characterized in that, The mass correction specifically involves scaling the repulsive force weight coefficient in the local potential field model proportionally based on the ratio of the remaining mass data to the total takeoff mass, and exponentially correcting the potential field step size parameter. 6.The hydrogen energy unmanned aerial vehicle route resource scheduling system based on energy consumption optimization of claim 2, wherein, The parallel computation of spatial region decomposition specifically includes: dividing the scrolling window into multiple subspace blocks according to an octree structure, allocating them to different cores of the onboard multi-core processor to perform parallel computation of the obstacle repulsion field, target gravitational field, and airflow utilization force field within each sub-block, and obtaining the resultant force vector data through reduction and summation.

7. The energy consumption optimization-based hydrogen energy unmanned aerial vehicle route resource scheduling system according to claim 1, characterized in that, The method for converting the resultant force to the trajectory specifically includes: using the resultant force vector data as the desired acceleration vector for the next control cycle, integrating it to obtain the desired velocity and desired position sequence, and generating attitude angle and throttle commands through attitude calculation to form the desired trajectory command.

8. A method for optimizing the energy consumption of hydrogen-powered unmanned aerial vehicle (UAV) flight paths, applied to the energy consumption optimization-based hydrogen-powered UAV flight path resource scheduling system described in any one of claims 1-7, characterized in that, Includes the following steps: S1. Acquire signals and data through sensing modules deployed on the drone body, including: acquiring position data, speed data, and attitude angle data through the pose perception unit; acquiring obstacle distribution data through the environmental perception unit; acquiring wind speed vector data and atmospheric disturbance intensity data through the meteorological perception unit; and acquiring motor speed data, thrust output data, and power data through the power perception unit. S2. Input the location data, wind speed vector data and power data into the first processing unit, generate a global energy consumption cost base map through a three-level loop iteration, and perform fine-grained online correction and background fusion update on the global energy consumption cost base map in a scrolling window manner to obtain a local energy consumption correction base map. S3. Input the position data, attitude angle data, atmospheric disturbance intensity data and obstacle distribution data into the second processing unit, and obtain the mass estimation strategy instruction through three-level branch nesting discrimination. S4. Input the mass estimation strategy command, motor speed data, power data and thrust output data into the third processing unit, and calculate the current remaining mass data through the thrust-power response coupling relationship; S5. Input the current remaining mass data, local energy consumption correction base map, obstacle distribution data and wind speed vector data into the fourth processing unit, and obtain the resultant force vector data through parallel calculation of mass correction and spatial region decomposition. S6. The resultant force vector data input / output module obtains the desired trajectory command through resultant force-track conversion and outputs it to the flight control system to achieve coordinated scheduling of trajectory and energy consumption.

9. The method for optimizing the flight path energy consumption of a hydrogen-powered unmanned aerial vehicle according to claim 8, characterized in that, When the quality estimation strategy instruction is a passive filtering observation strategy, an adaptive sliding window width mechanism is adopted: the sliding standard deviation of the attitude angle change rate is calculated as a stationarity index. When the stationarity index is continuously lower than a preset threshold, the sliding window width is shortened to 5 seconds; when the stationarity index is higher than the preset threshold, the sliding window width is extended to 15 seconds.

10. A method for optimizing the flight path energy consumption of a hydrogen-powered unmanned aerial vehicle according to claim 8, characterized in that, The quality correction also includes fuel cell performance degradation compensation: recording the cumulative operating time of the fuel cell, calculating the degradation factor according to a preset degradation curve, wherein the degradation factor ranges from 1.0 to 1.2; The attenuation factor is multiplied by the correction amount of the repulsion weight coefficient to compensate for the changes in power output characteristics caused by fuel cell aging.

11. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the method according to any one of claims 8-10.

12. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the program implements the steps of the method according to any one of claims 8-10.

Images