Intelligent power distribution method and system for unmanned aerial vehicle power supply system

Abstract

The application discloses an intelligent power distribution method and system of a UAV power supply system, relates to the technical field of UAV power supply, and comprises the following steps: acquiring first and second running state data of a current flight cycle, and an initial action set and an initial expected energy efficiency value pre-generated for a candidate power supply platform; extracting a first feature set based on the first running state data; performing frequency domain transformation on the second running state data to extract a second feature set; generating a target aerodynamic drag penalty coefficient by using the first feature set, and calculating a predicted duty cycle and a transient undervoltage safety margin of a candidate action in combination with the second feature set and other parameters; performing rigid action shielding processing to obtain a legal action subset; and finally performing dynamic remodeling processing on the initial expected energy efficiency value based on a saturation margin and a penalty coefficient to output a target power distribution control instruction; and the application improves the energy efficiency and safety of power distribution under a complex aerodynamic environment.

Description

Technical Field

[0001] This invention relates to the field of unmanned aerial vehicle (UAV) power supply technology, specifically to an intelligent power distribution method and system for UAV power supply systems. Background Technology

[0002] With the rapid development of aviation logistics technology, long-endurance fixed-wing cargo drones have been widely used in long-distance delivery and cruise missions. In order to pursue the best flight economy, these drones usually need to precisely adjust their flight speed according to real-time weather conditions and mission requirements to maintain ideal energy utilization efficiency. In the existing power control scheme, the power supply voltage of the drone propulsion system is often directly provided by the battery pack, and its voltage platform usually remains fixed during flight.

[0003] However, in actual complex missions, when a UAV encounters strong headwinds or mission changes that cause its energy reserves to approach the safety threshold, the flight control system typically adopts a strategy of reducing flight speed to extend its range by switching to an economical speed with a higher lift-to-drag ratio. If the power bus voltage remains at a high-voltage platform at this time, the motor controller will be forced to perform pulse width modulation at an extremely small duty cycle. This not only significantly increases the switching losses of power devices but also causes the overall efficiency of the propulsion system to drop sharply in the low-speed range. This nonlinear fluctuation in efficiency not only offsets the energy-saving effect of deceleration but may even induce power output saturation or response lag under extreme weather interference due to the mismatch between the power bus and the motor operating point, thus leading to the risk of the flight platform stalling and crashing.

[0004] Therefore, how to solve the efficiency mismatch and transient stability problem between the fixed power supply voltage and the dynamically fluctuating motor speed and aerodynamic load during the variable airspeed cruise of UAVs has become a core technical challenge to further improve the energy utilization rate of long-endurance aircraft and ensure flight safety under all operating conditions. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides an intelligent power distribution method and system for unmanned aerial vehicle (UAV) power supply systems.

[0006] To achieve the above objectives, the technical solution of the present invention is as follows:

[0007] In a first aspect, the present invention discloses an intelligent power distribution method for a UAV power supply system, comprising the following steps:

[0008] Acquire the first and second operational status data of the UAV in the current flight cycle, as well as the pre-generated set of initial actions and corresponding initial expected energy efficiency values ​​for at least one candidate power supply platform;

[0009] Among them, the first operating status data represents the spatial motion attitude and relative airflow state, and the second operating status data represents the electrical output state of the power system; the candidate power supply platform is associated with preset platform reference electrical parameters;

[0010] Based on the first operating state data, a first set of features is extracted to characterize the degree of deviation of the aerodynamic efficiency of the thruster and the degree of wake unsteadiness.

[0011] The second operating state data is subjected to frequency domain transformation processing to extract the second feature set, which characterizes the periodic ripple amplitude caused by the asymmetric load of the thruster in the corresponding electrical circuit.

[0012] The target aerodynamic drag penalty coefficient is generated using the first feature set. Combined with the second feature set, the second operating state data, and the platform reference electrical parameters of the corresponding candidate power supply platform, the predicted duty cycle and transient undervoltage safety margin corresponding to each candidate action in the initial action set are calculated respectively.

[0013] For the initial set of actions, a rigid action shielding process is performed based on the transient undervoltage safety margin and the predicted duty cycle to obtain a subset of legal actions;

[0014] Based on the saturation margin of the predicted duty cycle and the target aerodynamic drag penalty coefficient for each action in the legal action subset, the initial expected energy efficiency value corresponding to the action is dynamically reshaped, and the target power distribution control command is output based on the reshaped result.

[0015] Secondly, this invention discloses an intelligent power distribution system for a drone power supply system, comprising:

[0016] The data acquisition module is used to acquire the first operational status data and the second operational status data of the UAV in the current flight cycle, as well as the initial action set and the corresponding initial expected energy efficiency value pre-generated for at least one candidate power supply platform; wherein, the first operational status data represents the spatial motion attitude and relative airflow state, and the second operational status data represents the electrical output state of the power system; the candidate power supply platform is associated with preset platform reference electrical parameters;

[0017] The first feature extraction module is used to extract a first set of features characterizing the aerodynamic efficiency deviation and wake unsteadiness of the propeller based on the first operating state data.

[0018] The second feature extraction module is used to perform frequency domain transformation processing on the second operating state data and extract a second feature set. The second feature set characterizes the periodic ripple amplitude caused by the asymmetric load of the thruster in the corresponding electrical circuit.

[0019] The action parameter prediction module is used to generate the target aerodynamic drag penalty coefficient using the first feature set; and to calculate the predicted duty cycle and transient undervoltage safety margin for each candidate action in the initial action set by combining the second feature set, the second operating state data and the platform reference electrical parameters of the corresponding candidate power supply platform.

[0020] The action shielding module is used to perform rigid action shielding processing on the initial action set based on the transient undervoltage safety margin and the predicted duty cycle to obtain a legal action subset.

[0021] The dynamic reshaping and output module is used to perform dynamic reshaping processing on the initial expected energy efficiency value corresponding to each action based on the saturation margin of the predicted duty cycle and the target aerodynamic drag penalty coefficient for each action in the legal action subset, and output the target power distribution control command based on the reshaping result.

[0022] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0023] 1. By constructing a fusion vector containing aerodynamic and electrical features, and using the initial action set generated by reinforcement learning to dynamically adjust the power supply platform, the problem of drastic efficiency decline in the low-speed range of the motor under fixed voltage is solved. Through dynamic reshaping logic based on the predicted duty cycle saturation margin, the system can find the bus voltage with optimal electrical efficiency in real time. This voltage-speed dynamic following mechanism maximizes the energy-saving potential of the power system in the full speed range, effectively extending the cruise time and economic range of cargo drones.

[0024] 2. An aerodynamic drag penalty coefficient was constructed by using slip deviation, macroscopic flight angle of attack and flow field jitter variance. This coefficient can sensitively detect the stall cliff effect under slow flight conditions at large angles of attack. Combined with the predicted duty cycle, rigid action shielding is performed to block low-pressure topology actions that would lead to power saturation in advance. This physical boundary driven decision mechanism ensures that the propulsion system always retains sufficient torque compensation space, thus guaranteeing the aerodynamic stability of the UAV during variable airspeed.

[0025] 3. A dual-track decision logic based on aerodynamic impedance threshold is introduced, enabling the UAV to automatically switch decision paradigms according to environmental risks. Under stable airflow conditions, the system executes an efficiency-driven strategy that prioritizes the reshaped target reward function value to optimize energy consumption. When encountering severe aerodynamic disturbances or impedance bursts, the system automatically switches to a torque-driven strategy based on saturation margin. This nonlinear intelligent transition mechanism ensures that the UAV can unconditionally prioritize control margin under extreme weather conditions. Attached Figure Description

[0026] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0027] Figure 1 This is an overall block diagram of the method in Embodiment 1 of the present invention;

[0028] Figure 2 This is a flowchart illustrating the overall execution process of the method in Embodiment 1 of the present invention.

[0029] Figure 3 This is an overall block diagram of the system in Embodiment 2 of the present invention. Detailed Implementation

[0030] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. 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.

[0031] In the field of modern long-endurance fixed-wing cargo drones, the efficiency of the propulsion system and energy management strategies are considered key indicators for improving mission economy. This optimal energy efficiency is essentially a process of dynamically matching electrical energy with aerodynamic work at the physical level. Specifically, it involves reconfiguring the power supply topology through the battery management system and dynamically switching the voltage platform of the power bus to ensure that the motor and ESC always operate within their most efficient speed-voltage combined range.

[0032] However, existing technologies lack a verification mechanism for the coupling relationship between the input aerodynamic load state and the output bus electrical transients, making it impossible to accurately identify nonlinear aerodynamic slip and high-frequency ripple voltage drop issues during UAV cruise at varying airspeeds. Nonlinear aerodynamic slip manifests as the thruster advance ratio deviating from the design point during low-speed cruise or high angle-of-attack attitudes, causing torque demand to rapidly approach the power saturation region under specific low-voltage platforms. High-frequency ripple voltage drop manifests as periodic torque pulsations caused by asymmetric slip current coupling into the electrical circuit, amplifying current ripple under low-voltage topology and causing transient voltage drops in the bus. Consequently, a strict physical mapping relationship cannot be established between aerodynamic load fluctuations and the steady-state response of the electric drive system. This makes the system highly susceptible to decision-making errors or triggering hardware protection shutdowns when switching voltage platforms, thus affecting the safety of power distribution commands and the certainty of energy efficiency improvements.

[0033] For example, on long-haul cargo routes encountering strong headwinds, drones reduce their cruising speed from optimal endurance speed to a long-range economical speed to conserve energy. In this situation, traditional power distribution systems, using reinforcement learning algorithms, can only detect the decrease in rotational speed and instruct the battery pack to switch to a low-voltage platform to pursue theoretical electrical efficiency. Furthermore, because the system fails to detect the nonlinear slippage of the propeller advance ratio at high angles of attack and the AC ripple generated by asymmetrical loads, the motor controller's duty cycle instantaneously reaches the 100% modulation limit after the voltage platform switch. Specifically, the system misinterprets this insufficient thrust margin as normal low-speed operation or fails to anticipate the impact of current ripple on the weak voltage platform, causing the bus voltage to momentarily drop below the undervoltage lockout protection threshold, resulting in a sudden power system shutdown or flight platform stall.

[0034] If the above problems are not resolved, the power distribution system will continuously lose its ability to objectively determine the power supply status under complex aerodynamic environments. Specifically, unrecognized aerodynamic slip will cause the UAV to fall into a low-voltage power saturation trap, resulting in the power system losing its ability to compensate for sudden disturbances. Simultaneously, uncorrected high-frequency ripple voltage drop will cause a continuous deterioration of the bus electrical environment, making it impossible to quantitatively monitor transient safety margins, ultimately leading to irreversible hardware shutdown. Therefore, inaccurate power distribution feedback will systematically hinder long-endurance UAVs from mastering the core skill of electro-aerodynamic cross-domain coordinated control, making it impossible to achieve optimal energy utilization while ensuring flight safety under all operating conditions.

[0035] Example 1:

[0036] like Figures 1-2 As shown, the intelligent power distribution method for the UAV power supply system includes the following steps:

[0037] Step S1: Obtain the first operational status data and the second operational status data of the UAV in the current flight cycle, as well as the initial action set and corresponding initial expected energy efficiency value pre-generated for at least one candidate power supply platform; wherein, the first operational status data represents the spatial motion attitude and relative airflow state, and the second operational status data represents the electrical output state of the power system; the candidate power supply platform is associated with preset platform reference electrical parameters;

[0038] In this embodiment, step S1 serves as the perception and decision-making preparation stage of the entire power distribution process. Its core lies in constructing a multi-dimensional state space and mapping it to a pre-trained decision logic to obtain an initial solution. To achieve this, this embodiment pre-constructs a power policy generation network based on a deep reinforcement learning architecture to generate an initial action set and corresponding initial expected energy efficiency values ​​for candidate power supply platforms. This policy generation network employs a multi-layer fully connected neural network structure. Its input layer receives a fusion vector containing flight environment features and power electrical features. The intermediate hidden layer captures nonlinear energy conversion relationships through linear transformations and activation functions. The output layer provides the probability distribution of switching actions for different voltage platforms through a Softmax function, and defines the state-action value (Q-value) corresponding to each action as the initial expected energy efficiency value. The reinforcement learning model is used because UAVs face complex aerodynamic disturbances during variable airspeed cruise. Traditional lookup table methods are insufficient to cover all operating conditions, while this model can establish a deep correlation between the current state and the optimal theoretical energy efficiency through offline learning from historical flight data. Specifically, the training process of a policy generation network includes: constructing a training environment containing a state space, an action space, and a reward function; wherein, the reward function... Defined as the efficiency of the power system With energy margin The weighted sum, i.e. ,in , These are the training weight coefficients. During training iterations, historical flight samples are used as state inputs, and the network weights are optimized by maximizing the cumulative reward value, enabling the network to learn control laws that balance energy consumption and safety at different airspeeds. Furthermore, to eliminate dimensional differences, the flight environment features and power electronics features received by the input layer are normalized before input, scaling them to the [0,1] interval.

[0039] Specifically, during the real-time data acquisition process in step S1, the system first acquires the first operational state data of the UAV in the current flight cycle. This dataset focuses on characterizing the spatial motion attitude and relative airflow state. The system first obtains a flight sensor state sequence containing initial airspeed parameters from the Pitot tube and atmospheric data computer, and then calculates the first and second derivatives of these airspeed parameters with respect to time. It is worth noting that, in order to identify spurious signals caused by airflow turbulence or sensor jitter, the system compares the calculated second derivative with a preset physical maneuvering acceleration limit threshold (e.g., set to...). The system compares the data. If the absolute value of the second derivative of the airspeed at a certain sampling point exceeds the threshold, the system determines that the data is contaminated by noise and then activates a preset Kalman filter. This filter updates the state using the predicted value from the previous moment and the current measurement residual, and outputs the smoothed optimal estimate as the final true airspeed. In addition, the first operating state data also synchronously collects the mechanical speed of the motor, the pitch angle of the fuselage, and the track angle, which represents the slope of the actual track. These physical quantities together constitute the basic input for aerodynamic load analysis.

[0040] Simultaneously, the system monitors the electrical output status of the power system in real time, i.e., the second operating status data. This includes the current voltage of the power bus, the current pulse-width modulation (PWM) duty cycle output by the motor controller, and the three-phase current sequence obtained through high-frequency sampling (e.g., a sampling frequency of 10kHz). These electrical parameters reflect the current power consumption level and electrical stability of the power system. Based on this, the system retrieves the platform reference electrical parameters associated with at least one candidate power supply platform. These parameters are stored in the system's battery management knowledge base, specifically representing the theoretical steady-state voltage of the platform at the current battery state of charge (SOC) and the estimated AC internal resistance reflecting the battery power characteristics.

[0041] Based on the fused vectors collected above, the system inputs them into the aforementioned strategy-generated network. Assume the current UAV cruising airspeed is... The motor speed is The drone is currently in a tilting-up attitude, increasing its angle of attack. The policy network, through a non-linear mapping of its internal weight matrix, outputs a set of actions containing multiple candidate actions (e.g., "maintain current 400V platform", "switch to 300V platform", "switch to 200V platform"). Specifically, the output layer dimension N of the policy generation network strictly corresponds to the number of discrete voltage platforms supported by the drone's battery management system. Each neuron in the output layer is pre-bound with a specific physical electrical label; for example, the first neuron is bound to a fully series high-voltage platform (400V), and the second neuron is bound to a semi-series medium-voltage platform (200V). During operation, the weight matrix maps the high-dimensional hidden features to an original vector of length N, where each dimension of the vector represents the theoretical benefit of the corresponding physical platform under the current aerodynamic conditions. The system normalizes this original vector using the Softmax function to obtain the probability distribution of each action; simultaneously, it extracts the original values ​​before normalization as the state-action value (Q-value). Since the Q-value represents the expectation of long-term energy gain in reinforcement learning algorithms, it is directly mapped to the initial expected energy efficiency value in this embodiment. It is worth noting that the action set is not randomly generated by the network, but is the complete set of physical platforms corresponding to all neurons in the output layer, or a subset of actions with probability values ​​greater than a certain threshold (such as 0.05) selected through a preset threshold, thereby ensuring that the output candidate solutions are executable.

[0042] For each action, the network will provide a corresponding initial expected energy efficiency value. For example, if switching to a 300V platform, the network predicts an initial expected energy efficiency value of 0.85, which represents the theoretical efficiency score that the system is expected to achieve without considering transient physical constraints. This process realizes the transformation from raw physical quantities to preliminary decision-making schemes, laying the data foundation for subsequent safety reshaping by combining aerodynamic slip characteristics and frequency domain ripple characteristics.

[0043] Step S2: Extract a first set of features based on the first operating state data to characterize the degree of deviation of the aerodynamic efficiency of the thruster and the degree of wake unsteadiness;

[0044] In this embodiment, step S2 achieves a quantitative assessment of the propeller's operational quality by constructing an aerodynamic characteristic analysis engine. Before executing the specific calculation process, this embodiment first configures a propeller nominal characteristic database in the system background. This database is not a simple numerical record, but a mathematical model established based on the theoretical efficiency curves of the propeller obtained from ground wind tunnel experiments. The system pre-determines the optimal design advance ratio of this type of propeller under standard atmospheric pressure through polynomial fitting. The reason for introducing this benchmark is that the advance ratio is a core dimensionless parameter that measures the efficiency of a thruster in converting mechanical energy into kinetic energy. By measuring the deviation between the actual state and the theoretical sweet spot, the power loss state of the thruster at high angles of attack or low airspeeds can be effectively characterized.

[0045] Specifically, step S2 first performs a quantitative calculation of the deviation of the thruster aerodynamic efficiency based on the basic flight state data output in step S1. The system then uses the real-time acquired airspeed... The mechanical speed n of the motor and the preset physical diameter of the propeller Calculate the actual forward ratio To ensure dimensional consistency, the system first converts the motor's mechanical speed n, which is in revolutions per minute (rpm), to revolutions per second (RPS). The calculation formula is as follows: ;

[0046] After obtaining the actual advance ratio, the system compares it with the preset optimal design advance ratio. The difference is calculated, and the absolute value is extracted to obtain the slippage deviation. ;

[0047] ;

[0048] This slip deviation directly reflects the degree of deviation between the actual inflow state of the propeller blade and the ideal design state.

[0049] Subsequently, the system extracts the macroscopic flight angle of attack through geometric calculation. This characterizes the axial shift of the airflow caused by fuselage attitude adjustments. The system reads the pitch angle from the first operating state data. with track angle And perform the subtraction operation: This indicator is used to determine whether the drone is in an asymmetric force condition of slow, pitch-up flight. When the pitch angle is significantly higher than the track angle, the propeller disk is not perpendicular to the inflow direction, which will cause violent fluctuations in the advance ratio.

[0050] To further quantify the unsteady-state nature of the aerodynamic load wake, the system initiated dynamic variance analysis logic based on a time-sliding window. The system pre-sets a step size of... A sampling window of (e.g., 2 seconds) is used to continuously extract the slippage deviation sequence calculated from each sampling point within that window. Based on this sequence, the system calculates its statistical variance, thereby obtaining the variance of the flow field jitter. :

[0051] ;

[0052] in This is the arithmetic mean of the slip deviation within the window. This flow field jitter variance characterizes the instantaneous stability of the propeller wake flow field and aerodynamic loads. If... Increase, even if the current average slip deviation Even if the airflow is still within a safe range, the system will determine that it has entered an unstable, unsteady state. Finally, the system will calculate the slip deviation. Macroscopic angle of attack of flight and flow field jitter variance They are collectively encapsulated and identified as the first feature set, serving as the core criterion for subsequent calculation of the power distribution penalty coefficient.

[0053] Step S3: Perform frequency domain transformation processing on the second operating state data to extract the second feature set. The second feature set characterizes the periodic ripple amplitude caused by the asymmetric load of the thruster in the corresponding electrical circuit.

[0054] In this embodiment, step S3 achieves quantitative capture of microscopic electrical fluctuations in the power system by constructing a frequency domain feature extraction engine. This embodiment pre-configures a high-frequency current analysis module in the airborne processing terminal. The core architecture of this module is based on the Fast Fourier Transform (FFT) algorithm and spectral feature matching. Frequency domain analysis is used because when the UAV flies at high angles of attack, uneven forces on the propeller disk generate a P-factor effect. This physical asymmetric aerodynamic load is directly coupled to the motor spindle, forming periodic torque pulsations. These torque pulsations are reflected in the electrical circuit as specific harmonic ripples highly correlated with the motor speed frequency. Frequency domain analysis can accurately extract this microscopic electrical characteristic representing aerodynamic instability from complex phase current noise.

[0055] Specifically, step S3 first acquires the second operating state data collected in step S1, focusing on extracting the motor mechanical speed n and the three-phase current sequence within a set timeframe. To determine the target frequency for analysis, the system first calculates the rotor mechanical rotation fundamental frequency. The system reads the motor mechanical speed n in revolutions per minute and obtains the reference frequency representing the time required for one revolution of the rotor mechanical speed through time constant conversion. The calculation formula is as follows: ;

[0056] For example, if the current stepper motor mechanical speed n is 3000 RPM, then the corresponding rotor mechanical rotation base frequency is... The frequency is 50Hz. This frequency point is the core index for finding the asymmetric load signature in subsequent spectrum analysis.

[0057] Subsequently, the system performs frequency domain transformation on the three-phase current sequences within a set timeframe (e.g., a 200-millisecond sampling window). To improve the accuracy of the spectral analysis and suppress spectral leakage, the system first applies a Hanning window to the original current time-domain signal for weighted preprocessing, and then calls the Fast Fourier Transform algorithm to transform the three-phase current sequences from the time domain to the frequency domain. Assuming the current sampling frequency... The frequency is 10kHz, and the transformed spectral resolution is... The system locates and extracts the fundamental frequency of the rotor's mechanical rotation from the generated frequency response curve. The current amplitude corresponding to the matched frequency point. To eliminate quantization errors caused by signal truncation, the system uses the centroid method to... Nearby spectral lines are interpolated and corrected to obtain a high-precision reference AC ripple amplitude. The specific application of the center of gravity method The weighted average of the maximum amplitude point in the neighborhood and its two adjacent points is used to correct for spectral leakage bias caused by non-integer sampling periods and signal truncation. By calculating the energy distribution between adjacent spectral lines, the amplitude center corresponding to the actual rotating fundamental frequency is corrected, thereby ensuring the extracted reference AC ripple amplitude. It can accurately reflect the intensity of current fluctuations caused by asymmetric loads, without being affected by the FFT fence effect.

[0058] It is worth noting that this reference AC ripple amplitude Essentially, this quantifies the intensity of electrical load fluctuations generated per revolution of the thruster. In practical applications, if the UAV is in level flight, the ripple amplitude typically remains at a low baseline level; however, once it enters a high angle of attack or encounters asymmetric airflow, the amplitude at this frequency point will significantly increase. The system ultimately determines this extracted baseline AC ripple amplitude as the second feature set. This feature set provides the necessary microscopic current margin criteria for assessing the transient undervoltage risk under the low-voltage platform in subsequent step S4, ensuring that power distribution decisions can perceive the nonlinear mapping of aerodynamic loads at the electrical level.

[0059] Step S4: Generate the target aerodynamic drag penalty coefficient using the first feature set, and combine it with the second feature set, the second operating state data and the platform reference electrical parameters of the corresponding candidate power supply platform to calculate the predicted duty cycle and transient undervoltage safety margin for each candidate action in the initial action set.

[0060] Before executing the specific calculation process, this embodiment pre-configures multi-dimensional mapping logic in the airborne processing terminal, aiming to transform the aerodynamic characteristics in the time domain and the electrical characteristics in the frequency domain into quantitative indicators usable for power distribution decisions. This cross-domain coupled modeling is employed because, during variable airspeed cruise, the efficiency loss of the thruster not only manifests as steady-state energy consumption but also implies the risk of reduced operating margin due to aerodynamic slippage. By establishing a multi-dimensional evaluation system that includes penalty coefficients, predicted duty cycles, and safety margins, the system can quantitatively predict whether the power system can simultaneously meet aerodynamic thrust requirements and electrical transient stability after switching to the target voltage platform.

[0061] Specifically, step S4 first uses the first feature set output from step S2 to generate the target aerodynamic drag penalty coefficient. The system reads the slip deviation. Macroscopic angle of attack of flight and flow field jitter variance The fusion operation is performed using a preset nonlinear exponential function:

[0062] ;

[0063] In this formula, These are the preset first, second, and third weighting constants. Their physical meaning is that when the angle of attack and the slip deviation increase simultaneously, aerodynamic drag does not increase linearly, but rather exhibits an exponential increase characteristic similar to a stall cliff. For example, setting... ,like from Increase to ,and Increasing from 0.1 to 0.3, the calculated... This will surge from approximately 2.06 to approximately 7.85, providing a strong risk signal for subsequent decision-making. It is worth noting that this is due to the potential for macroscopic angles of attack during drone descent. In the case of negative values, to ensure the monotonically increasing nature of the penalty logic, this embodiment modifies the formula... The item undergoes non-negative truncation. Specifically, if the calculated result is less than zero, it is forced to take the value as zero or a preset minimum positive value to ensure the aerodynamic drag penalty coefficient. It is always used as a positive risk constraint factor in subsequent calculations to avoid the logical failure of the reward function due to the reversal of the positive and negative values ​​of physical quantities.

[0064] Subsequently, the system performs a prediction duty cycle for each candidate action in the initial action set. Calculation. To ensure that the UAV can maintain its current thrust requirements after the voltage platform switch, the system performs proportional conversion processing based on the principles of power conservation and linear mapping of thrust:

[0065] ;

[0066] in, This represents the current pulse width modulation duty cycle. This is the current bus voltage. This represents the theoretical steady-state voltage of the candidate power supply platform.

[0067] Assuming we are currently on a high-voltage platform ( The following Cruise, if the candidate action involves switching to a low-pressure platform Then predict the duty cycle. Will be adjusted to This indicator is used to predict whether the motor controller will lose thrust adjustment margin due to excessive duty cycle after a voltage drop.

[0068] Finally, the system combines the second feature set output in step S3 to calculate the transient undervoltage safety margin of the corresponding candidate power supply platform. The system first sets the reference AC ripple amplitude in the phase current domain. Mapped to the bus domain, the predicted ripple voltage drop depth is calculated and combined with the estimated AC internal resistance of the candidate power supply platform. and the hardware-preset undervoltage lockout protection threshold Perform the following calculations:

[0069] ;

[0070] This formula determines whether the hardware will trigger power-off protection by simulating the transient impact of amplified current ripple on the bus voltage under a low-voltage topology. If the calculation result... The closer a value is to or less than 0, the higher the risk of electrical failure for that candidate action under aerodynamic asymmetric loads. The system ultimately outputs a value for each candidate action. The indicators provide comprehensive data support for the rigid shielding process in step S5.

[0071] Step S5: For the initial action set, perform rigid action shielding based on the transient undervoltage safety margin and the predicted duty cycle to obtain a subset of legal actions;

[0072] In this embodiment, step S5 establishes a low-level safety boundary for candidate topology schemes by constructing an action shielding engine. This embodiment pre-configures a rigid admission criterion library based on hardware limit constraints in the processing terminal. The rigid shielding mechanism is employed because, in the complex aerodynamic environment of variable airspeed cruise, power distribution decisions not only need to consider long-term energy efficiency but must also prioritize ensuring the transient electrical safety and control continuity of the power system. This engine essentially sets a physical dead zone for the decision-making system, preventing intelligent algorithms from falling into fatal traps of power-off protection or thrust saturation in subsequent optimizations due to the pursuit of local efficiency gains by pre-blocking action paths that do not meet the underlying hardware operating conditions.

[0073] Specifically, step S5 first retrieves the transient undervoltage safety margin for each candidate power supply platform calculated in step S4. Compared with the predicted duty cycle The system then iterates through and logically verifies each candidate action in the initial action set.

[0074] For each candidate action, the system executes the following two-dimensional hard constraint determination logic:

[0075] The first dimension is the determination of electrical safety, which is based on the transient undervoltage safety margin. The system will make a judgment. It will check whether the indicator meets the following conditions: ;

[0076] If this condition is met, it means that after considering the ripple voltage drop caused by aerodynamic asymmetric load and the influence of battery AC internal resistance, the target candidate platform's bus voltage will very likely drop to the preset undervoltage lockout protection threshold during switching or operation. The following is an explanation: This transient undervoltage will trigger a forced shutdown protection at the hardware level, causing the drone to lose power in the air. Therefore, the system determines that this action is electrically infeasible.

[0077] The second dimension is the control safety judgment, which is based on the predicted duty cycle. The system will make a judgment. It will check whether the indicator meets the following conditions: ;

[0078] If this condition is met, it indicates that, in order to maintain the thrust required for the current flight cycle, the control duty cycle required by the motor controller (ESC) has reached or exceeded its 100% modulation limit at a lower candidate voltage platform. Physically, this means that the power system has entered a power saturation state, and the flight control system will completely lose its ability to adjust the motor upwards, making it unable to cope with sudden gusts of wind or perform maneuvers. Therefore, the system determines that the maneuver is uncontrollable.

[0079] In the actual processing flow of this embodiment, the system performs a forced elimination operation on the initial action set. As long as the target candidate action meets any of the above-mentioned judgment conditions (i.e., either electrical safety or control safety is not qualified), the system permanently deletes the candidate action from the initial action set, so that it will no longer participate in the subsequent reward value reshaping and optimization decision.

[0080] For example, suppose the initial action set contains three candidate actions: action A (switch to 350V platform), action B (switch to 300V platform), and action C (maintain 400V platform). After calculation in step S4, action A's... Action B Action C In this step of the process, action B is due to... It was blocked due to a breach of electrical safety limits. If, at this time, the system experiences a sharp increase in the predicted duty cycle due to severe aerodynamic slippage, such as in action A... If the value changes to 1.02, action A will also be eliminated. Finally, the system determines the remaining set of candidate actions after the elimination process as a subset of legal actions and passes it to the subsequent step S6 for refined energy efficiency evaluation. Through this rigid filtering, the scheme logically forms a protective net built by physical laws, ensuring that the output target power distribution control commands always operate within a safe physical envelope.

[0081] Step S6: Based on the saturation margin of the predicted duty cycle and the target aerodynamic drag penalty coefficient of each action in the legal action subset, perform dynamic reshaping processing on the initial expected energy efficiency value corresponding to the action, and output the target power distribution control command based on the reshaping result.

[0082] In this embodiment, step S6 achieves the final screening and instruction output of each candidate power distribution scheme in the subset of legal actions by constructing an energy efficiency-power joint optimization engine. This embodiment is configured with a strategy reshaping criterion library based on a nonlinear penalty mechanism. The core architecture logic of this engine lies in establishing a dynamic game model between energy consumption and control safety. The reason for adopting a penalty function with the reciprocal of the saturation margin as the core is that when the power system is cruising at variable airspeed, its risk stems not only from steady-state efficiency loss, but also from the loss of thrust adjustment capability. When the predicted duty cycle is close to 100%, the system must forcibly reduce the priority of this action with an exponentially increasing penalty term, thereby constructing a virtual potential energy wall to ensure that while pursuing ultimate economy, the system always retains sufficient torque compensation space to cope with aerodynamic uncertainties.

[0083] Specifically, step S6 first obtains the subset of legal actions determined in step S5, and for each legal action in the subset, extracts the predicted duty cycle calculated in step S4. and the target aerodynamic drag penalty coefficient The system then uses a preset zero-prevention constant. The saturation margin characterizing the current power system's adjustment potential is calculated, and this saturation margin is defined as a numerical value of 1 minus the predicted duty cycle. The surplus portion afterward. Based on this, the system constructs a dynamic penalty term. Its calculation logic is to apply the target aerodynamic drag penalty coefficient. Dividing by this saturation margin, the corresponding mathematical expression is as follows:

[0084] ;

[0085] In this formula, This characterizes the external drag risk caused by airflow slippage and changes in angle of attack, while the denominator term... This characterizes the hardware constraint boundaries within the system. This means that even if the aerodynamic drag is small, if the target voltage platform is too low, causing the duty cycle to approach saturation, the dynamic penalty term will still increase sharply as the denominator approaches zero.

[0086] Subsequently, the system utilizes a preset penalty weight constant. The dynamic penalty term is weighted and the initial expected energy efficiency value is obtained from step S1. Subtracting the weighted dynamic penalty term from the result yields the reshaped result. This refers to the reshaped target reward function value. The specific reshaping formula is:

[0087] ;

[0088] in, This represents the theoretical maximum electrical efficiency predicted by the reinforcement learning algorithm based on historical data, while This represents the safety costs that must be sacrificed to address the current risks of transient aerodynamic slip and duty cycle saturation.

[0089] It is worth noting that, in order to cope with different levels of meteorological threats, the system also introduces a dual-track decision logic based on aerodynamic impedance thresholds when outputting target power distribution control commands. The system incorporates the target aerodynamic drag penalty coefficient... It is compared in real time with a preset aerodynamic impedance threshold. If determined... If the value is below this threshold, it indicates that the current atmospheric environment is stable, and the system's execution efficiency dominates the decision-making process, meaning that the reshaped result is selected from the subset of legal actions. The action with the highest value is taken as the final instruction. If a decision is made... If the value is greater than or equal to this threshold, it indicates that the UAV is experiencing severe nonlinear slip or intense aerodynamic disturbance, and the system is forced to switch to torque-driven decision-making. At this point, the system no longer relies on energy efficiency ranking, but directly obtains the saturation margin corresponding to each action. The action with the largest corresponding saturation margin value is determined as the target power distribution control command.

[0090] For example, suppose there are two schemes in the subset of legal actions: Scheme 1 has an initial expected efficiency of 0.95, but due to a low predicted voltage, the duty cycle reaches 0.92; Scheme 2 has an initial expected efficiency of 0.88 and a duty cycle of 0.75. Under normal operating conditions, if Smaller and The settings are reasonable; Option 1 may be due to... It is more likely to be selected. However, if a gust of wind occurs at this time, it may cause... If the voltage spikes and exceeds the threshold, the system will automatically disable the energy efficiency advantage of Option 1 and instead prioritize Option 2 (which maintains a higher voltage platform) to ensure that the UAV has sufficient transient torque to overcome aerodynamic load disturbances. Finally, the system controls the power system to execute the selected action, completing intelligent power distribution scheduling for variable airspeed cruise scenarios.

[0091] In summary, through the deep extraction and fusion processing of aerodynamic time-domain and electrical frequency-domain characteristics, this embodiment achieves a leap from traditional static efficiency mapping to cross-domain dynamic safety assessment. Specifically, by utilizing a penalty mechanism constructed from slip deviation and flow field jitter variance, the nonlinear drag burst of the thruster under high angle-of-attack cruise is effectively identified, logically blocking the path for the power system to fall into the power saturation region. Simultaneously, the transient undervoltage safety margin captured through phase current frequency domain analysis provides a micro-level hardware safety boundary for power distribution decisions, physically mitigating the risk of undervoltage lockout shutdown caused by topology switching. Based on this, dynamic reshaping logic based on predicted duty cycle saturation margin maximizes the energy-saving potential brought by power platform reconfiguration while ensuring transient torque compensation capability. Ultimately, this dual-track decision-making combined with airflow impedance thresholds enables the UAV to spontaneously switch energy efficiency management priorities according to real-time weather conditions, significantly improving the economy of long-endurance missions while establishing a full-condition flight safety barrier composed of physical boundaries and control margins.

[0092] Example 2:

[0093] like Figure 3 As shown, the intelligent power distribution system of the UAV power supply system includes:

[0094] The data acquisition module is used to acquire the first operational status data and the second operational status data of the UAV in the current flight cycle, as well as the initial action set and the corresponding initial expected energy efficiency value pre-generated for at least one candidate power supply platform; wherein, the first operational status data represents the spatial motion attitude and relative airflow state, and the second operational status data represents the electrical output state of the power system; the candidate power supply platform is associated with preset platform reference electrical parameters;

[0095] The first feature extraction module is used to extract a first set of features characterizing the aerodynamic efficiency deviation and wake unsteadiness of the propeller based on the first operating state data.

[0096] The second feature extraction module is used to perform frequency domain transformation processing on the second operating state data and extract a second feature set. The second feature set characterizes the periodic ripple amplitude caused by the asymmetric load of the thruster in the corresponding electrical circuit.

[0097] The action parameter prediction module is used to generate the target aerodynamic drag penalty coefficient using the first feature set; and to calculate the predicted duty cycle and transient undervoltage safety margin for each candidate action in the initial action set by combining the second feature set, the second operating state data and the platform reference electrical parameters of the corresponding candidate power supply platform.

[0098] The action shielding module is used to perform rigid action shielding processing on the initial action set based on the transient undervoltage safety margin and the predicted duty cycle to obtain a legal action subset.

[0099] The dynamic reshaping and output module is used to perform dynamic reshaping processing on the initial expected energy efficiency value corresponding to each action based on the saturation margin of the predicted duty cycle and the target aerodynamic drag penalty coefficient for each action in the legal action subset, and output the target power distribution control command based on the reshaping result.

[0100] The above description is merely an example and illustration of the structure of the present invention. Those skilled in the art can make various modifications or additions to the specific embodiments described, or use similar methods to replace them, as long as they do not deviate from the structure of the invention or exceed the scope defined in the claims, all of which should fall within the protection scope of the present invention.

[0101] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0102] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to any specific implementation. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. An intelligent power distribution method for a UAV power supply system, characterized in that, Includes the following steps: Acquire the first and second operational status data of the UAV in the current flight cycle, as well as the pre-generated set of initial actions and corresponding initial expected energy efficiency values ​​for at least one candidate power supply platform; The first operating status data represents the spatial motion attitude and relative airflow state, and the second operating status data represents the electrical output state of the power system; the candidate power supply platform is associated with preset platform reference electrical parameters. Based on the first operating state data, a first set of features is extracted to characterize the degree of deviation of the aerodynamic efficiency of the thruster and the degree of wake unsteadiness. The second operating state data is subjected to frequency domain transformation processing to extract a second feature set, which characterizes the periodic ripple amplitude caused by the asymmetric load of the thruster in the corresponding electrical circuit. The target aerodynamic drag penalty coefficient is generated using the first feature set, and the predicted duty cycle and transient undervoltage safety margin corresponding to each candidate action in the initial action set are calculated by combining the second feature set, the second operating state data and the platform reference electrical parameters of the corresponding candidate power supply platform. For the initial set of actions, a rigid action shielding process is performed based on the transient undervoltage safety margin and the predicted duty cycle to obtain a subset of legal actions; Based on the saturation margin of the predicted duty cycle corresponding to each action in the subset of legal actions and the target aerodynamic drag penalty coefficient, dynamic reshaping processing is performed on the initial expected energy efficiency value corresponding to the action, and the target power distribution control command is output based on the reshaping result.

2. The intelligent power distribution method for the UAV power supply system according to claim 1, characterized in that: The first operating status data includes the actual airspeed, motor mechanical speed, pitch angle, and track angle; the second operating status data includes the current bus voltage, current pulse width modulation duty cycle, and three-phase current sequence; the platform reference electrical parameters include the theoretical steady-state voltage and AC internal resistance estimate corresponding to the candidate power supply platform.

3. The intelligent power distribution method for the UAV power supply system according to claim 2, characterized in that: The extraction process of the first feature set includes: calculating the actual advance ratio based on the actual airspeed, the motor mechanical speed, and the preset propeller physical diameter; calculating the absolute difference between the actual advance ratio and the preset optimal design advance ratio to obtain the slip deviation; calculating the difference between the pitch angle and the track angle to obtain the macroscopic flight angle of attack; extracting multiple consecutive slip deviations within a preset time sliding window and calculating the statistical variance to obtain the flow field jitter variance; and jointly determining the slip deviation, the macroscopic flight angle of attack, and the flow field jitter variance as the first feature set. The extraction process of the second feature set includes: calculating the rotor mechanical rotation fundamental frequency based on the motor mechanical speed; performing a fast Fourier transform on the three-phase current sequence within a set time width to convert the three-phase current sequence from the time domain to the frequency domain; extracting the current amplitude corresponding to the frequency point that matches the rotor mechanical rotation fundamental frequency in the frequency domain to obtain the reference AC ripple amplitude; and determining the reference AC ripple amplitude as the second feature set.

4. The intelligent power distribution method for the UAV power supply system according to claim 3, characterized in that: The calculation process for the target aerodynamic drag penalty coefficient includes: Based on a preset exponential function, the slip deviation, macroscopic flight angle of attack, and flow field jitter variance in the first feature set are nonlinearly fused to obtain the target aerodynamic drag penalty coefficient; the calculation formula is: ； in, The target aerodynamic drag penalty coefficient, For macroscopic flight angle of attack, This is the slip deviation amount. The variance of the flow field jitter; These are the first preset weight constant, the second preset weight constant, and the third preset weight constant, respectively.

5. The intelligent power distribution method for the UAV power supply system according to claim 4, characterized in that: The calculation process for the predicted duty cycle includes: Under the constraint of maintaining the current thrust demand unchanged, the predicted duty cycle is obtained by performing proportional scaling based on the current bus voltage, the current pulse width modulation duty cycle, and the theoretical steady-state voltage of the corresponding candidate power supply platform; the calculation formula is: ； in, To predict the duty cycle, It is the product of the current pulse width modulation duty cycle and the current bus voltage. This is the theoretical steady-state voltage.

6. The intelligent power distribution method for the UAV power supply system according to claim 5, characterized in that: The calculation process for the transient undervoltage safety margin includes: The ripple voltage drop depth is calculated based on the predicted duty cycle, the reference AC ripple amplitude, and the estimated AC internal resistance of the corresponding candidate power supply platform. Based on the theoretical steady-state voltage, the ripple voltage drop depth, and the preset undervoltage lockout protection threshold, the transient undervoltage safety margin is calculated; the calculation formula is: ； in, For transient undervoltage safety margin, For the theoretical steady-state voltage, The ripple voltage drop depth is calculated by multiplying the baseline AC ripple amplitude, the predicted duty cycle, and the estimated AC internal resistance. This is the undervoltage lockout protection threshold.

7. The intelligent power distribution method for the UAV power supply system according to claim 6, characterized in that: The process of determining the subset of legal actions includes: Iterate through each candidate action in the initial action set; For a target candidate action, if it is determined that the transient undervoltage safety margin corresponding to the target candidate action is less than 0, or if it is determined that the predicted duty cycle corresponding to the target candidate action is greater than or equal to 1, then the target candidate action is forcibly removed from the initial action set. The set of candidate actions remaining after the elimination process is determined as the subset of legal actions.

8. The intelligent power distribution method for the UAV power supply system according to claim 7, characterized in that: The initial expected energy efficiency value is dynamically reshaped, and the target power distribution control command is output based on the reshaped result, including: For each legal action in the subset of legal actions, the saturation margin is calculated based on the corresponding predicted duty cycle. The dynamic penalty term is calculated by dividing the target aerodynamic drag penalty coefficient by the saturation margin. The dynamic penalty term is weighted using a preset penalty weight constant, and the weighted dynamic penalty term is subtracted from the initial expected energy efficiency value corresponding to the legal action to obtain the reshaped result; the calculation formula is: ； ； in, For dynamic penalty items, The target aerodynamic drag penalty coefficient, This is the saturation margin, which is calculated by subtracting the predicted duty cycle from 1. In addition to zero constant protection get; The result after reshaping This is the initial expected energy efficiency value. For the penalty weight constant Weighted dynamic penalty item; Control the power system to execute and maximize values The corresponding legal actions are identified and designated as the target power distribution control commands.

9. The intelligent power distribution method for the UAV power supply system according to claim 8, characterized in that: The process of outputting the target power distribution control command based on the reshaped result also includes: The target aerodynamic drag penalty coefficient is compared with a preset aerodynamic impedance threshold. If the target aerodynamic drag penalty coefficient is less than the aerodynamic impedance threshold, then the efficiency-driven decision is: in the subset of legal actions, the candidate action with the largest corresponding reshaped result value is determined as the target power distribution control command; If the target aerodynamic drag penalty coefficient is greater than or equal to the aerodynamic impedance threshold, then a torque-dominated decision is executed: obtain the saturation margin corresponding to each candidate action in the legal action subset, and determine the candidate action with the largest corresponding saturation margin value in the legal action subset as the target power distribution control command.

10. An intelligent power distribution system for a UAV power supply system, characterized in that: The intelligent power distribution method using the UAV power supply system as described in any one of claims 1-9 includes: The data acquisition module is used to acquire the first operational status data and the second operational status data of the UAV in the current flight cycle, as well as the initial action set and the corresponding initial expected energy efficiency value pre-generated for at least one candidate power supply platform; wherein, the first operational status data represents the spatial motion attitude and relative airflow state, and the second operational status data represents the electrical output state of the power system; the candidate power supply platform is associated with preset platform reference electrical parameters; The first feature extraction module is used to extract a first set of features characterizing the aerodynamic efficiency deviation and wake unsteadiness of the thruster based on the first operating state data. The second feature extraction module is used to perform frequency domain transformation processing on the second operating state data and extract a second feature set, which characterizes the periodic ripple amplitude caused by the asymmetric load of the thruster in the corresponding electrical circuit. The action parameter prediction module is used to generate a target aerodynamic drag penalty coefficient using the first feature set; and to calculate the predicted duty cycle and transient undervoltage safety margin for each candidate action in the initial action set by combining the second feature set, the second operating state data and the platform reference electrical parameters of the corresponding candidate power supply platform. The action shielding module is used to perform rigid action shielding processing on the initial action set based on the transient undervoltage safety margin and the predicted duty cycle to obtain a subset of legal actions; The dynamic reshaping and output module is used to perform dynamic reshaping processing on the initial expected energy efficiency value corresponding to each action in the subset of legal actions, based on the saturation margin of the predicted duty cycle and the target aerodynamic drag penalty coefficient, and output the target power distribution control command based on the reshaping result.

Images