Unmanned aerial vehicle energy consumption safety-constrained navigation optimization method based on adaptive policy iteration

Abstract

The present application relates to the technical field of unmanned aerial vehicle navigation control, and discloses an unmanned aerial vehicle energy consumption safety constraint navigation optimization method based on adaptive policy iteration, which comprises the following steps: acquiring the flight state, battery state, environment state and task state of the unmanned aerial vehicle, and constructing a navigation state set; constructing a candidate navigation segment graph with action type labels, and calculating the predicted energy consumption of the candidate navigation segment; calculating the navigation energy debt after completing an actual navigation segment, and updating the energy consumption deviation coefficient of the corresponding action type; calculating the endurance safety shell of the candidate node to the safety point set based on the updated energy consumption deviation coefficient; screening a safety action set according to the current residual energy, predicted energy consumption of the candidate action and endurance safety shell of the arrival node; performing policy evaluation and policy improvement in the safety action set, and generating an unmanned aerial vehicle endurance navigation optimization instruction. The present application can avoid the problem of local path optimization but subsequent return or emergency landing inaccessibility, and improve the endurance navigation safety of the unmanned aerial vehicle.

Description

Technical Field

[0001] This invention relates to the field of unmanned aerial vehicle (UAV) navigation and control technology, specifically to a UAV energy consumption safety constraint navigation optimization method based on adaptive strategy iteration. Background Technology

[0002] In scenarios such as inspection, surveying, emergency reconnaissance, and logistics delivery, drones typically need to complete navigation tasks such as multi-point access, obstacle avoidance, return to home, or emergency landing within limited battery capacity. Existing drone navigation optimization methods usually combine flight distance, wind field, payload, remaining battery power, return-to-home battery power, and obstacle constraints for path planning, and select flight paths through path search algorithms, heuristic optimization algorithms, or strategy iteration algorithms to improve mission completion efficiency and reduce flight energy consumption.

[0003] However, most existing methods perform verification or path correction based on remaining battery power and return-to-home battery power after path planning or strategy selection. Their energy consumption constraints are usually used as result judgment conditions or penalty terms in the objective function. Because UAVs are affected by factors such as wind field changes, payload changes, battery degradation, frequent attitude adjustments, hovering operations, and obstacle avoidance maneuvers during actual flight, a continuous deviation will occur between predicted and actual energy consumption, accumulating in subsequent flight segments. Furthermore, the energy consumption deviation patterns differ for different types of flight maneuvers (such as level flight, climb, turn, hovering operations, and obstacle avoidance). Existing methods struggle to differentiate and specifically correct the energy consumption deviations of various maneuvers, leading to UAVs potentially choosing maneuvers with higher current mission gains but insufficient energy retention for return-to-home or diversion upon reaching the next node. This results in the problem of "local path optimization but unsafe subsequent flight endurance." Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a UAV energy consumption safety constraint navigation optimization method based on adaptive strategy iteration. This method solves the problem that existing UAV navigation optimization methods typically perform remaining battery power verification after path planning or strategy selection, making it difficult to prevent subsequent return-to-home or unreachable diversion actions from being selected in a timely manner. This leads to the risk of flight safety for UAVs under the condition of accumulated energy consumption prediction deviations under different action types.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a UAV energy consumption safety-constrained navigation optimization method based on adaptive strategy iteration, comprising: Acquire the drone's flight status, battery status, environmental status, and mission status to construct a navigation endurance status set; A candidate route map is constructed based on the endurance navigation status set. The candidate route map includes candidate nodes and candidate routes, and an action type label is configured for each candidate route. Based on the action type marker, flight status, battery status, and environmental status, calculate the predicted energy consumption for each candidate flight segment; After the UAV completes the actual flight segment, the flight segment energy debt is calculated based on the actual energy consumed and the corresponding predicted energy consumption, and the energy consumption deviation coefficient of the corresponding action type is updated based on the flight segment energy debt. Based on the updated energy consumption deviation coefficient, the endurance safety shell of the candidate node to the set of safe points is calculated. The endurance safety shell is determined based on the minimum predicted energy consumption, reserved safety energy and energy consumption prediction uncertainty of each safe point in the candidate node to the set of safe points. The energy consumption prediction uncertainty is determined based on the energy debt fluctuation value of each action type in the candidate path from the candidate node to the safe point. Based on the current remaining energy, the predicted energy consumption of candidate actions, and the endurance safety shell of the candidate action's arrival node, a set of safe actions is selected. Within the set of safety actions, strategy evaluation and strategy improvement are performed to generate UAV endurance and navigation optimization instructions.

[0006] Furthermore, the flight status includes the UAV's current position, flight altitude, flight speed, heading angle, attitude change frequency, and payload weight; the battery status includes remaining energy, output voltage, output current, battery temperature, discharge rate, and historical attenuation coefficient; the environmental status includes wind speed, wind direction, obstacle distribution, and no-fly zones; and the mission status includes pending mission points, completed mission points, mission point priority, and abandoned mission point markers.

[0007] Furthermore, the steps for constructing the candidate route map include: The current location of the drone, the mission point, the obstacle avoidance point, the return point, and the alternate landing point are selected as candidate nodes. Candidate flight segments are generated based on the spatial reachability relationships between candidate nodes; Action type markers are configured based on the flight maneuver characteristics of candidate flight segments. The action type markers include level flight segments, climb segments, descent segments, turning segments, hovering operation segments, obstacle avoidance segments, and return segments.

[0008] Furthermore, the steps for calculating the predicted energy consumption for each candidate flight segment include: The basic flight energy consumption is calculated based on the segment distance, altitude change, and maneuver type markings of the candidate segments; Based on the load weight, wind speed, wind direction, attitude change frequency, and battery status, determine the load correction, wind field correction, attitude control correction, and battery attenuation correction respectively. The predicted energy consumption is obtained by fusing the basic flight energy consumption, load correction, wind field correction, attitude control correction, battery attenuation correction, and energy consumption deviation coefficient for the corresponding action type.

[0009] Furthermore, the steps for updating the energy consumption deviation coefficient for the corresponding action type include: Determine the action type marker corresponding to the actual flight segment; Calculate the deviation ratio between actual energy consumption and predicted energy consumption; The energy consumption deviation coefficient corresponding to the action type label is updated based on the deviation ratio and the adaptive update coefficient. Specifically, if the energy debt of the flight segment is positive, the energy consumption deviation coefficient of the corresponding action type is increased; if the energy debt of the flight segment is negative, the energy consumption deviation coefficient of the corresponding action type is decreased.

[0010] Furthermore, the steps for screening the set of safety actions include: Get the candidate actions in the current state; Calculate the node reached after executing each candidate action; Determine whether the remaining energy after deducting the predicted energy consumption of the candidate actions is greater than the endurance safety shell of the reached node; Candidate actions that meet the judgment results are added to the safe action set, while candidate actions that do not meet the judgment results are excluded.

[0011] Furthermore, the steps for implementing strategy evaluation and strategy improvement include: Within the set of safety actions, the strategic value of each safety action is calculated based on task benefits, predicted energy consumption, endurance safety margin, action switching disturbance cost, and task delay cost. Select the security action with the highest strategic value as the current action to be executed; Based on the changing trend of energy debt in consecutive flight segments, the energy consumption weight and mission benefit weight in the strategy evaluation are adjusted, and the next round of strategy improvement is executed based on the adjusted weights. Specifically, if the energy debt of multiple consecutive actual flight segments is positive, the energy consumption weight is increased and the mission benefit weight is decreased; if the energy debt of multiple consecutive actual flight segments is negative and the endurance safety margin is greater than a set threshold, the energy consumption weight is decreased and the mission benefit weight is increased.

[0012] Furthermore, it also includes a downgrade navigation step: If there are no normal task actions in the set of safety actions to continue executing task point expansion, stop executing task point expansion; The emergency action set is reconstructed based on the current remaining energy. The emergency action set includes diversion action, return to base action, hovering and holding action, and mission abandonment action. Based on the current remaining energy, the predicted energy consumption at the return point, and the predicted energy consumption at the alternate landing point, the drone's endurance navigation optimization instructions are determined from the set of emergency actions and output.

[0013] This invention also provides a UAV energy consumption safety constraint navigation optimization system based on adaptive strategy iteration, comprising: The state set construction module is used to acquire the drone's flight status, battery status, environmental status, and mission status to build a sustained navigation state set; A candidate segment map construction module is used to construct a candidate segment map based on the endurance navigation state set. The candidate segment map includes candidate nodes and candidate segments, and an action type marker is configured for each candidate segment. The predicted energy consumption calculation module is used to calculate the predicted energy consumption of each candidate flight segment based on the action type marker, flight status, battery status and environmental status. The energy consumption deviation update module is used to calculate the energy debt of the flight segment based on the actual energy consumed and the corresponding predicted energy consumption after the UAV completes the actual flight segment, and update the energy consumption deviation coefficient of the corresponding action type based on the energy debt of the flight segment. The endurance safety case calculation module is used to calculate the endurance safety case of the candidate node to the set of safe points based on the updated energy consumption deviation coefficient. The safety action filtering module is used to filter the set of safety actions based on the current remaining energy, the predicted energy consumption of candidate actions, and the endurance safety shell of the candidate action's arrival node. The strategy iteration module is used to perform strategy evaluation and strategy improvement within the set of safety actions, and generate UAV endurance and navigation optimization instructions.

[0014] This invention provides a method for optimizing UAV energy consumption and safety constraints navigation based on adaptive strategy iteration. It has the following beneficial effects: 1. This invention calculates the endurance safety shell from candidate nodes to the set of safe points based on the updated energy consumption deviation coefficient. The endurance safety shell is jointly determined based on the minimum predicted energy consumption, reserved safety energy, and the energy consumption prediction uncertainty determined by the energy debt fluctuation value of each action type. Furthermore, it filters the set of safe actions based on the current remaining energy, the predicted energy consumption of candidate actions, and the endurance safety shell of the node reached by the candidate actions. This ensures that strategy evaluation and strategy improvement are performed only within the range of actions that meet the endurance safety shell constraints. It can exclude high-risk actions that are unreachable due to subsequent return or diversion before action selection, avoiding the problem of the current path being locally optimal but subsequent endurance safety not being reachable.

[0015] 2. This invention calculates the segment energy debt after the UAV completes the actual flight segment and updates the energy consumption deviation coefficient of the corresponding action type based on the segment energy debt. This enables the predicted energy consumption of level flight segments, climb segments, descent segments, turning segments, hovering operation segments, obstacle avoidance segments, and return segments to be adaptively corrected according to the actual flight conditions, reducing the risk of prediction inaccuracies in static energy consumption models under conditions of wind field changes, load changes, battery degradation, and frequent attitude adjustments.

[0016] 3. This invention incorporates mission benefits, predicted energy consumption, endurance safety margin, action switching disturbance cost, and mission delay cost into the strategy evaluation, and adjusts the energy consumption weight and mission benefit weight in the strategy evaluation according to the changing trend of energy debt in continuous flight segments. This enables UAVs to achieve a dynamic balance between mission completion rate and endurance safety, reducing the situation of premature return or risky continuation of mission execution.

[0017] 4. By constructing candidate flight segment maps with action type labels, this invention enables energy consumption deviations to be attributed to specific action types, rather than path corrections based solely on total remaining energy. This allows for differentiated constraints on high-energy-consuming actions such as turning, climbing, hovering, and obstacle avoidance, thereby improving the adaptability of navigation strategies to complex flight maneuvers.

[0018] 5. When the set of safe actions is empty, the present invention performs a degraded navigation step. Based on the current remaining energy, the predicted energy consumption of the return point and the predicted energy consumption of the alternate landing point, it generates alternate landing actions, return actions, hovering waiting actions or mission abandonment actions. It can switch to emergency response strategies in a timely manner when the risk of endurance increases, thereby improving the flight safety and mission reliability of UAVs in the event of sudden strong winds, battery abnormalities, obstacle restrictions or changes in mission area. Attached Figure Description

[0019] Figure 1 A flowchart of a UAV energy consumption safety constraint navigation optimization method based on adaptive strategy iteration provided in an embodiment of the present invention.

[0020] Figure 2 The diagram shows the structure of the UAV energy consumption safety constraint navigation optimization system based on adaptive strategy iteration, as provided in this embodiment of the invention.

[0021] Figure 3 This is a schematic diagram illustrating the linkage between flight segment energy debt and endurance safety shell, provided for an embodiment of the present invention. Detailed Implementation

[0022] The technical solution of the present invention will now be clearly and completely described 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.

[0023] This application provides a method for optimizing UAV energy consumption safety constraints navigation based on adaptive strategy iteration. This method is applicable to multi-rotor UAVs, fixed-wing UAVs, vertical take-off and landing UAVs, and UAVs carrying inspection, mapping, reconnaissance, or delivery payloads. The UAV is equipped with a flight control unit, a positioning unit, a battery management unit, an environmental perception unit, and a mission management unit. The flight control unit acquires the UAV's flight altitude, flight speed, heading angle, and attitude change frequency; the positioning unit acquires the UAV's current position; the battery management unit acquires remaining energy, output voltage, output current, battery temperature, discharge rate, and historical attenuation coefficient; the environmental perception unit acquires wind speed, wind direction, obstacle distribution, and no-fly zones; and the mission management unit acquires pending mission points, completed mission points, mission point priorities, and abandoned mission point markers.

[0024] In this embodiment, the UAV does not determine whether the remaining energy is sufficient for return after the complete path planning is completed. Instead, before each strategy iteration, it first filters a set of safe actions based on the endurance safety shell of the candidate action's arrival node, and performs strategy evaluation and improvement only within this set of safe actions. Simultaneously, after each actual flight segment, the UAV calculates the segment energy debt between the actual energy consumed and the predicted energy consumption, and updates the energy consumption deviation coefficient of the corresponding action type based on this debt. This allows the predicted energy consumption of subsequent candidate flight segments of the same type to be adaptively corrected according to actual flight conditions.

[0025] Example 1

[0026] like Figure 1 As shown, this embodiment provides a method for optimizing UAV energy consumption and safety constraints navigation based on adaptive policy iteration, including the following steps: S1. Obtain the drone's flight status, battery status, environmental status, and mission status to construct a navigation endurance status set.

[0027] Flight status includes the UAV's current position, altitude, speed, heading angle, attitude change frequency, and payload weight. The UAV's current position is obtained from an RTK positioning module, GPS positioning module, visual positioning module, or integrated navigation module. Altitude is obtained from a barometer, laser rangefinder, or flight control altitude estimation. The attitude change frequency is statistically derived from the rate of change of the UAV's roll, pitch, and yaw angles per unit time, characterizing the intensity of the UAV's attitude adjustment in the current environment. The payload weight is pre-input from the mission configuration file or estimated from the motor output, motor speed, and hovering thrust during takeoff.

[0028] Battery status includes remaining energy, output voltage, output current, battery temperature, discharge rate, and historical degradation coefficient. Remaining energy is expressed in watt-hours or joules and can be calculated by the battery management unit based on the state of charge, rated capacity, and current voltage. The historical degradation coefficient is determined by the number of battery cycles, changes in battery internal resistance, battery capacity degradation records, and historical discharge curves.

[0029] Environmental conditions include wind speed, wind direction, obstacle distribution, and no-fly zones. Wind speed and direction are obtained from airborne wind speed sensors or calculated based on the UAV's airspeed, ground speed, and attitude. Obstacle distribution is obtained from LiDAR, visual depth cameras, millimeter-wave radar, or pre-stored maps. No-fly zones are obtained from electronic fence data, mission area boundaries, or airspace control information.

[0030] Task status includes pending task points, completed task points, task point priorities, and abandonable task point markers. Task point priorities are used to determine the order in which tasks continue execution when battery life risks increase. Abandonable task point markers indicate whether the corresponding task point can be skipped or postponed when battery life is insufficient.

[0031] The system encodes flight status, battery status, environmental status, and mission status into a unified endurance navigation status set. Each status node in the endurance navigation status set includes at least spatial location, remaining energy, mission completion progress, and a safe reachability marker. The safe reachability marker indicates whether the UAV can still reach at least one safe point in the set of safe points from that status node.

[0032] S2. Construct a candidate route map based on the endurance navigation state set. The candidate route map includes candidate nodes and candidate routes, and configure an action type label for each candidate route.

[0033] Candidate nodes include the UAV's current location, mission point, obstacle avoidance point, return-to-home point, and alternate landing point. Obstacle avoidance points are generated based on obstacle boundaries, no-fly zone boundaries, and safe distances. Return-to-home points can be takeoff points, mission base stations, or mobile recovery points. Alternate landing points are preset safe landing areas, or temporary safe landing areas dynamically selected based on real-time terrain, obstacle distribution, and no-fly zones.

[0034] The system generates candidate flight segments based on the spatial reachability between candidate nodes. If there is an obstacle crossing, no-fly zone crossing, exceeding the maximum flight distance, exceeding the maximum climb angle limit, or exceeding the communication coverage range between two candidate nodes, the corresponding candidate flight segment is deleted. For the retained candidate flight segments, the system configures action type tags based on the flight maneuver characteristics of the candidate flight segments. Action type tags include level flight segments, climb segments, descent segments, turning segments, hovering operation segments, obstacle avoidance segments, and return segments.

[0035] Specifically, if the altitude change of a candidate segment is less than a set altitude threshold and the heading angle change is less than a set heading threshold, it is configured as a level flight segment; if the altitude increase of a candidate segment is greater than a set altitude threshold, it is configured as a climb segment; if the altitude decrease of a candidate segment is greater than a set altitude threshold, it is configured as a descent segment; if the heading angle change of a candidate segment is greater than a set heading threshold, it is configured as a turn segment; if a candidate segment includes a task point stay operation, it is configured as a hovering operation segment; if a candidate segment is used to bypass obstacles or no-fly zones, it is configured as an obstacle avoidance segment; if the arrival node of a candidate segment is a return point, it is configured as a return segment.

[0036] In one implementation, the same candidate flight segment has a primary action type label and an additional action type label. For example, when a candidate flight segment contains both obstacle avoidance and climb maneuvers, the primary action type label for the candidate flight segment is obstacle avoidance, and the additional action type label is climb. The system assigns weights to each action type label according to a preset action proportion or flight segment feature proportion, and then weights and fuses the energy consumption deviation coefficients corresponding to each action type to obtain the comprehensive energy consumption deviation coefficient for the candidate flight segment.

[0037] S3. Calculate the predicted energy consumption for each candidate flight segment based on the action type marker, flight status, battery status, and environmental status.

[0038] Predicted energy consumption includes base flight energy consumption, payload correction, wind field correction, attitude control correction, battery degradation correction, and energy consumption deviation coefficients for corresponding maneuvers. Base flight energy consumption is determined by the candidate flight segment's distance, altitude change, set flight speed, and the UAV's base power. Payload correction is determined by the difference between the payload weight and the UAV's nominal payload. Wind field correction is determined by the angle between wind speed, wind direction, and the candidate flight segment's direction. Attitude control correction is determined by the attitude change frequency and maneuver type markings. Battery degradation correction is determined by battery temperature, discharge rate, and historical degradation coefficients.

[0039] In one exemplary embodiment, the attitude control correction amount It can be determined in the following way:

[0040] in, The frequency of attitude change. For action type tagging The corresponding attitude energy consumption coefficient. Attitude energy consumption coefficients marked for different action types. This can be obtained through pre-calibration, for example, the hovering operation segment and the obstacle avoidance segment corresponding to... The range is greater than the level flight segment. Those skilled in the art can also establish a mapping relationship between attitude change frequency and additional energy consumption based on the specific UAV model through polynomial fitting or table lookup methods; these are all equivalent implementations of this application.

[0041] In a preferred embodiment, the predicted energy consumption of the candidate flight segment is calculated as follows:

[0042] in, Indicates from candidate nodes Fly to candidate node Predicted energy consumption; Indicates basic flight energy consumption; This represents the overall energy consumption deviation coefficient corresponding to the candidate flight segment; Indicates the load correction amount; Indicates wind field correction amount; This indicates the attitude control correction amount; This indicates the amount of battery degradation correction.

[0043] For candidate segments with only one action type label, the comprehensive energy consumption deviation coefficient is the energy consumption deviation coefficient corresponding to that action type label. For candidate segments with multiple action type labels, the comprehensive energy consumption deviation coefficient is obtained by weighted fusion of the energy consumption deviation coefficients corresponding to multiple action type labels. The weight of each action type label is determined by the segment distance percentage, altitude change percentage, heading change percentage, or hovering operation duration percentage.

[0044] The base flight energy consumption is obtained by multiplying the UAV's base power by the estimated flight time of the candidate segment. The estimated flight time of the candidate segment is determined by the segment distance and the recommended flight speed. For hovering operation segments, the base flight energy consumption is obtained by multiplying the hovering power by the operation dwell time. For climb segments, the system adds the climb energy consumption caused by altitude changes to the base flight energy consumption. For descent segments, the system sets a descent energy consumption correction amount based on the UAV's motor control characteristics.

[0045] The energy consumption deviation coefficient is a crucial adaptive parameter in this application. The system maintains corresponding energy consumption deviation coefficients for different action types. For example, the level flight segment corresponds to the level flight energy consumption deviation coefficient, the climb segment to the climb energy consumption deviation coefficient, the descent segment to the descent energy consumption deviation coefficient, the turn segment to the turn energy consumption deviation coefficient, the hovering operation segment to the hovering energy consumption deviation coefficient, the obstacle avoidance segment to the obstacle avoidance energy consumption deviation coefficient, and the return segment to home to the return segment. The initial values ​​of each energy consumption deviation coefficient are set based on historical flight data, UAV model parameters, or ground calibration data.

[0046] like Figure 3As shown in Figure S4, after the UAV completes the actual flight segment, calculate the segment energy debt based on the actual energy consumed and the corresponding predicted energy consumption, and update the energy consumption deviation coefficient of the corresponding action type based on the segment energy debt.

[0047] The UAV executes the actual flight segment according to the current UAV endurance navigation optimization instructions. After the actual flight segment is completed, the system reads the remaining energy at the start and end times of the actual flight segment and calculates the actual energy consumed. The actual energy consumed can also be obtained by integrating the product of the output voltage and the output current over time.

[0048] Segment energy debt represents the deviation between actual energy consumption and predicted energy consumption, and it is calculated as follows:

[0049] in, Indicates the first Segment energy debt for each actual flight segment; Indicates the first Actual energy consumption for each actual flight segment; Indicates the first The predicted energy consumption is calculated before each actual flight segment is executed.

[0050] If the energy debt of a flight segment is positive, it means that the actual energy consumption of the actual flight segment is greater than the predicted energy consumption, and the system determines that there is a risk of underestimation of energy consumption for the corresponding action type of the actual flight segment. If the energy debt of a flight segment is negative, it means that the actual energy consumption of the actual flight segment is less than the predicted energy consumption, and the system determines that the endurance margin for the corresponding action type of the actual flight segment can be appropriately released.

[0051] In a preferred embodiment, the energy consumption deviation coefficient for the corresponding action type is updated in the following manner:

[0052] in, Indicates the updated action type tag The corresponding energy consumption deviation coefficient; Indicates the action type tag before the update. The corresponding energy consumption deviation coefficient; This represents the adaptive update coefficient. The adaptive update coefficient is dynamically adjusted based on the degree of wind field fluctuation, the degree of battery temperature change, and the mission risk level.

[0053] When an actual flight segment has multiple action type markers, the system allocates the segment's energy debt to the corresponding action type markers according to their weights, and updates the energy consumption deviation coefficients for each action type. For example, if an actual flight segment includes both obstacle avoidance and climb, the system allocates the segment's energy debt based on the proportion of obstacle avoidance distance and the proportion of climb altitude, thus updating the energy consumption deviation coefficients for the obstacle avoidance segment and the climb segment, respectively.

[0054] To avoid sudden changes in the energy consumption deviation coefficient due to occasional measurement errors, the system performs a limit on the energy debt of flight segments. When the deviation ratio of actual energy consumption to predicted energy consumption exceeds a set upper limit, it is limited to the set upper limit; when the deviation ratio of actual energy consumption to predicted energy consumption is lower than a set lower limit, it is limited to the set lower limit. The system can also use a sliding window to calculate the average deviation ratio of multiple similar actual flight segments before updating the energy consumption deviation coefficient for the corresponding action type.

[0055] When a flight segment experiences communication interruption, emergency obstacle avoidance, sensor malfunction, missing battery management unit data, or forced takeover by a user, the system marks that flight segment as an abnormal update segment. The segment energy debt of an abnormal update segment is only used for log recording and does not participate in the energy consumption deviation coefficient update for the corresponding action type.

[0056] This step allows the system to distinguish the sources of energy consumption anomalies for different maneuver types. For example, in headwind conditions, the energy consumption deviation coefficient increases during level flight segments; in environments with frequent obstacle avoidance, the energy consumption deviation coefficient increases during turning and obstacle avoidance segments; and under heavy loads or low battery temperatures, the energy consumption deviation coefficient increases during climb and hovering maneuvers. In subsequent strategy iterations, the system recalculates predicted energy consumption based on the updated energy consumption deviation coefficients, thereby reducing the probability of high-energy-consumption-risk maneuvers being selected.

[0057] S5. Based on the updated energy consumption deviation coefficient, calculate the endurance safety shell from the candidate node to the set of safe points.

[0058] The set of safe landing points includes return points, alternate landing points, and mission termination points. Return points are the takeoff point, mission base station, or mobile recovery point. Alternate landing points are pre-defined safe landing areas or temporary safe landing areas selected in real-time. Mission termination points are areas where the mission can be safely completed and await recovery.

[0059] For each candidate node in the candidate flight segment diagram, the system calculates the minimum predicted energy consumption from that candidate node to each safe point in the set of safe points. The minimum predicted energy consumption is obtained in the candidate flight segment diagram using Dijkstra's algorithm, A* algorithm, or dynamic programming algorithm. When calculating the minimum predicted energy consumption, the system recalculates the predicted energy consumption of the candidate flight segment using the updated energy consumption deviation coefficient, so that the endurance safety shell can reflect the current actual flight conditions.

[0060] In a preferred embodiment, the endurance containment of the candidate node is determined in the following manner:

[0061] in, Indicates candidate nodes Corresponding battery life protection case; Represents the set of safe points; Indicates from candidate nodes Reach a safe point The lowest predicted energy consumption; This indicates that reserve energy for safety. Indicates candidate nodes The corresponding energy consumption prediction uncertainty; This represents the uncertainty amplification factor.

[0062] The uncertainty of energy consumption prediction The energy debt fluctuation value of each action type in the candidate path from the candidate node to the safe point is determined. The system calculates the standard deviation of segment energy debt for each action type in the most recent several actual flight segments. When calculating the candidate safe path from the candidate node to the safe point, the system obtains the action types and the number of flight segments included in the candidate safe path, and accumulates or weights the standard deviation of segment energy debt for each action type according to the number of flight segments to obtain the energy consumption prediction uncertainty corresponding to the candidate node. If the historical sample size for a certain action type is insufficient, the system uses the preset uncertainty of that action type or the uncertainty of similar action types as a substitute.

[0063] Reserved safety energy is used to withstand sudden wind fields, positioning drift, communication delays, and energy consumption during landing. The uncertainty amplification factor is set according to the mission risk level. The higher the mission risk level, the larger the uncertainty amplification factor, the thicker the endurance safety shell, and the fewer high-risk candidate actions are allowed.

[0064] When wind conditions change drastically, the energy consumption deviation coefficient of the action type fluctuates significantly, battery temperature is abnormal, or the candidate safe path includes many climbing, obstacle avoidance, and hovering operations, the uncertainty of energy consumption prediction increases, and the corresponding range safety housing needs to be thickened. Therefore, the system can automatically shrink the space of subsequent selectable actions when the reliability of energy consumption prediction decreases.

[0065] S6. Based on the current remaining energy, the predicted energy consumption of candidate actions, and the endurance safety shell of the candidate actions reaching the node, filter the set of safe actions.

[0066] In the current state, the system extracts candidate actions that the UAV can perform from the candidate flight segment map. Candidate actions include flying to the next mission point, flying to the obstacle avoidance point, flying to the return point, flying to the alternate landing point, reducing speed and continuing to fly, adjusting altitude and continuing to fly, hovering and waiting, and abandoning the mission point.

[0067] For each candidate action, the system calculates the predicted energy consumption required to execute the candidate action and determines the node to which the candidate action will arrive. Subsequently, the system determines whether the current remaining energy, after deducting the predicted energy consumption of the candidate action, is greater than the endurance safety shell required to reach the node.

[0068] In a preferred embodiment, the selection criteria for the set of safety actions are as follows:

[0069] in, Indicates the current remaining energy; Indicates candidate actions Predicted energy consumption; Indicates candidate actions The arrival node; Indicates arrival at node The corresponding battery life protection case.

[0070] If a candidate action meets the above conditions, the system adds it to the safe action set; if a candidate action does not meet the above conditions, the system excludes it. Therefore, even if a candidate action has high mission benefits, it will not enter the strategy evaluation stage if its execution would prevent the drone from retaining sufficient energy for return or diversion.

[0071] In this embodiment, the set of safe actions serves as a prerequisite constraint for policy iteration. Policy evaluation and improvement are performed only within the set of safe actions, preventing situations where actions with high policy value but unattainable in terms of endurance safety are selected.

[0072] S7. Execute strategy evaluation within the set of safety actions.

[0073] The system calculates the strategy value for each safety action in the set of safety actions. The strategy value comprehensively considers task benefits, predicted energy consumption, endurance safety margin, action switching disturbance cost, and task delay cost.

[0074] In a preferred embodiment, the strategy value is calculated as follows:

[0075] in, Representing state Perform safety actions Strategic value; Indicates the rewards for completing the task; Indicates a safety action Predicted energy consumption; Indicates the safety margin of battery life; Indicates the cost of action switching disturbance; Indicates the cost of task delay; , , , These represent the weighting coefficients corresponding to predicted energy consumption, range safety margin, action switching disturbance cost, and mission delay cost, respectively.

[0076] The range safety margin is determined in the following way:

[0077] in, Indicates the execution of safety actions The remaining energy exceeds the capacity of the endurance safety shell. The larger the endurance safety margin, the higher the safety margin after the drone performs this safety action.

[0078] Mission rewards are determined based on mission point priority, mission point completion status, and the number of remaining mission points. Higher rewards are achieved if a safe action allows the drone to approach a high-priority mission point. Lower rewards are achieved if a safe action points abort mission points and the current flight risk is high.

[0079] The cost of maneuver switching disturbances is used to suppress frequent sharp turns, climbs, and descents. If the change in heading angle, altitude, or speed of a safe maneuver is significant relative to the previous actual flight segment, the cost of maneuver switching disturbances is high. The cost of mission delays measures the extent to which safe maneuvers increase mission completion time.

[0080] S8. Improve execution strategy and generate drone endurance and navigation optimization instructions.

[0081] In each round of strategy iteration, the system uses the currently executed action selected in the previous round as the initial strategy and recalculates the strategy value of each safety action based on the updated set of safety actions. If the recalculated optimal safety action is different from the currently executed action in the previous round, the recalculated optimal safety action replaces the currently executed action. If the optimal safety actions obtained in two consecutive rounds are the same, or if the preset number of iterations is reached, the strategy improvement for the current round is deemed complete, and the corresponding UAV endurance and navigation optimization command is output.

[0082] The UAV endurance navigation optimization instructions include the target point for the next flight segment, recommended speed, recommended altitude, mission continuation marker, mission abort marker, and return-to-home or alternate-landing point selection results. After the strategy improvement is completed, the UAV executes the corresponding actual flight segment according to the UAV endurance navigation optimization instructions. After the actual flight segment is completed, the system re-executes S1 to S8, forming an adaptive strategy iteration process based on flight segment energy debt feedback.

[0083] In one implementation, the system adjusts the weights in the strategy evaluation based on the changing trend of energy debt across consecutive flight segments. If the energy debt of multiple consecutive actual flight segments is positive, it indicates that the actual energy consumption of the UAV is consistently higher than the predicted energy consumption, and the system increases the weight coefficient corresponding to the predicted energy consumption. At the same time, increase the weighting coefficient corresponding to the range safety margin. Reduce task rewards The relative impact in strategy evaluation leads drones to prioritize actions with low energy consumption and higher safety margins. If the energy debt of multiple consecutive actual flight segments is negative and the endurance safety margin is greater than the set margin threshold, it indicates that the current prediction is too conservative, and the system reduces the weight coefficient corresponding to the predicted energy consumption. Increase task rewards The relative impact allows the drone to continue performing high-priority tasks.

[0084] In another implementation, the system employs value iteration, policy iteration, near-end policy optimization, or constrained reinforcement learning algorithms to perform policy improvement. Regardless of the policy optimization algorithm used, the action space is defined by safe actions, and the endurance safety shell is used as the action admission condition.

[0085] S9. When the set of safety actions is empty, execute the downgrade navigation step.

[0086] In this embodiment, an empty safety action set means that the set of ordinary mission actions used to continue mission point expansion is empty. The downgrade navigation step no longer uses the ordinary mission action set, but instead reconstructs the emergency action set based on the current remaining energy. The emergency action set only includes diversion actions, return-to-base actions, hovering and waiting actions, and mission abort actions.

[0087] When the set of safety actions is empty, the system stops expanding the mission point and no longer selects ordinary mission actions. The system generates alternate landing actions, return-to-home actions, hovering waiting actions, or mission abort actions based on the current remaining energy, the predicted energy consumption of the return point, and the predicted energy consumption of the alternate landing point.

[0088] Specifically, if the current remaining energy is sufficient to reach at least one alternate landing point, the system will prioritize generating an alternate landing action; if the current remaining energy is sufficient to reach the return point, the system will generate a return action; if the communication link is briefly interrupted and the current remaining energy is sufficient to maintain a short-term hover, the system will generate a hovering waiting action; if the mission point is an abortable mission point and continuing to execute would increase the risk of endurance failure, the system will generate a mission abort action.

[0089] The system outputs alternate landing, return to home, hovering, or mission abandonment actions as UAV endurance navigation optimization commands, and simultaneously records the reason for triggering downgraded navigation, the current remaining energy at the time of triggering, the endurance safety shell, the segment energy debt, and the energy consumption deviation coefficient.

[0090] Example 2

[0091] like Figure 2 As shown, this embodiment provides a UAV energy consumption safety constraint navigation optimization system based on adaptive strategy iteration, used to execute the method in Embodiment 1. The system includes a state set construction module, a candidate flight segment map construction module, a predicted energy consumption calculation module, an energy consumption deviation update module, an endurance safety shell calculation module, a safety action screening module, and a strategy iteration module.

[0092] The state set construction module is used to acquire the UAV's flight status, battery status, environmental status, and mission status to build a sustained navigation state set. The module reads flight status through the flight control communication interface, battery status through the battery management interface, environmental status through the environmental perception interface, and mission status through the mission management interface. The flight control communication interface uses CAN bus, serial interface, MAVLink communication protocol, or other flight control data interfaces.

[0093] The candidate flight segment map construction module is used to construct a candidate flight segment map based on the endurance navigation state set. The module uses the UAV's current position, mission point, obstacle avoidance point, return point, and alternate landing point as candidate nodes, generates candidate flight segments based on the spatial reachability relationships between candidate nodes, and assigns an action type label to each candidate flight segment.

[0094] The predicted energy consumption calculation module calculates the predicted energy consumption for each candidate flight segment based on action type markers, flight status, battery status, and environmental conditions. Internally, the module stores energy consumption deviation coefficients corresponding to different action type markers and incorporates these coefficients into the predicted energy consumption calculation. Each time the predicted energy consumption calculation module receives an update from the energy consumption deviation update module, it recalculates the predicted energy consumption for the affected candidate flight segments.

[0095] The energy consumption deviation update module calculates the energy debt of a flight segment after the UAV completes the actual flight segment based on the actual energy consumed and the corresponding predicted energy consumption, and updates the energy consumption deviation coefficient for the corresponding action type based on the energy debt. The module maintains energy consumption deviation coefficients for different action types and adjusts the corresponding action type's energy consumption deviation coefficient according to the positive or negative direction and deviation ratio of the flight segment's energy debt. For abnormally updated flight segments, the module only records the energy debt and does not update the energy consumption deviation coefficient.

[0096] The endurance safety case calculation module is used to calculate the endurance safety case from candidate nodes to the set of safe points based on the updated energy consumption deviation coefficient. The module determines the endurance safety case corresponding to each candidate node based on the lowest predicted energy consumption, reserved safety energy, and energy consumption prediction uncertainty of each safe point in the set of safe points.

[0097] The safety action filtering module filters the set of safety actions based on the current remaining energy, the predicted energy consumption of candidate actions, and the endurance safety shell of the node to which the candidate actions arrive. The module only allows candidate actions that satisfy the endurance safety shell constraint to enter the safety action set.

[0098] The strategy iteration module is used to perform strategy evaluation and improvement within the set of safe actions, generating optimized UAV endurance and navigation commands. The module calculates the strategy value based on mission benefits, predicted energy consumption, endurance safety margin, action switching disturbance cost, and mission delay cost, and converts the safest action with the highest strategy value into flight control executable navigation commands.

[0099] In one implementation, the strategy iteration module further includes a degraded navigation unit. The degraded navigation unit is used to reconstruct the emergency action set when the safety action set is empty, generating alternative landing actions, return-to-home actions, hovering and waiting actions, or mission abort actions, and outputting corresponding UAV endurance navigation optimization instructions.

[0100] Example 3

[0101] This embodiment uses a power line inspection drone as an example to illustrate the specific operation process of this application.

[0102] The drone takes off from the base station, planning to inspect multiple pole and tower task points sequentially. The drone initially has full remaining energy. The system sets the base station as the return-to-home point and the open areas along the route as alternate landing points. The system constructs a candidate flight segment map based on the pole and tower task points, obstacle avoidance points, base stations, and alternate landing points, and assigns an action type label to each candidate flight segment.

[0103] After takeoff, the system calculates the predicted energy consumption for each candidate flight segment based on the current flight status, battery status, environmental status, and mission status. Initially, preset energy consumption deviation coefficients are used for level flight, climb, turn, and hovering operation segments. After the UAV performs the first inspection flight segment, the system calculates the actual energy consumption based on the output voltage and current recorded by the battery management unit and compares it with the predicted energy consumption before execution to obtain the segment's energy debt.

[0104] If a persistent headwind occurs during the first inspection flight segment, and the actual energy consumption exceeds the predicted energy consumption, then the energy debt for that segment is positive. Since the actual flight segment is marked as a level flight segment, the system increases the energy consumption deviation coefficient corresponding to the level flight segment. Subsequently, the system recalculates the predicted energy consumption for subsequent level flight candidate segments and recalculates the endurance safety shell for each candidate node.

[0105] As the drone approaches the second pole-mounted task point, the task management unit indicates that this task point has a higher priority. The strategy iteration module evaluates safety actions such as flying to the second pole-mounted task point, detouring to an obstacle avoidance point, and returning to base early within the set of safe actions. If the endurance safety shell constraint is still met after flying to the second pole-mounted task point, the system selects to fly to the second pole-mounted task point. If continuous headwinds result in insufficient endurance safety margin after flying to the second pole-mounted task point, this candidate action will not be included in the set of safe actions, and even if the task point has a higher priority, it will not be selected by the strategy iteration module.

[0106] During the inspection process, if multiple hovering photography operations by the drone result in a continuous positive energy debt for the hovering operation segment, the system increases the energy consumption deviation coefficient corresponding to the hovering operation segment, and also increases the weight of the predicted energy consumption and the weight of the endurance safety margin. Subsequently, the strategy iteration module reduces unnecessary hovering operations or shortens the dwell time at abandonable task points.

[0107] If, at any given moment, the UAV detects that its remaining energy is insufficient to continue flying to any mission point while retaining energy for return or diversion, and the only remaining safe action set is the return or diversion action, then the strategy iteration module directly outputs the return point or diversion point navigation command. If the set of normal mission actions used to continue mission point expansion is empty, then the degradation navigation unit stops mission point expansion, constructs an emergency action set based on the current remaining energy, selects the nearest diversion or return point, and outputs the corresponding UAV endurance navigation optimization command.

[0108] Through the above implementation methods, this application realizes the transformation from path planning under a fixed energy consumption model to an iterative strategy driven by segment energy debt and a safety shell constraint for endurance. The UAV can continuously correct the energy consumption prediction of different action types based on the energy consumption deviation during actual flight, and perform a set of safe actions before each round of strategy iteration, thereby avoiding the problem of high current local path benefits but subsequent inability to safely return or divert to base.

[0109] The above are merely preferred embodiments of this application and are not intended to limit this application. Those skilled in the art can make equivalent substitutions or adjustments to the data acquisition method, strategy iteration algorithm, candidate flight segment map construction method, safety point set setting method, and endurance safety shell calculation method without departing from the concept of this application; all such substitutions or adjustments should fall within the protection scope of this application.

Claims

1. A method for optimizing UAV energy consumption and safety constraints navigation based on adaptive strategy iteration, characterized in that, include: Acquire the drone's flight status, battery status, environmental status, and mission status to construct a navigation endurance status set; A candidate route map is constructed based on the endurance navigation status set. The candidate route map includes candidate nodes and candidate routes, and an action type label is configured for each candidate route. Based on the action type marker, flight status, battery status, and environmental status, calculate the predicted energy consumption for each candidate flight segment; After the UAV completes the actual flight segment, the flight segment energy debt is calculated based on the actual energy consumed and the corresponding predicted energy consumption, and the energy consumption deviation coefficient of the corresponding action type is updated based on the flight segment energy debt. Based on the updated energy consumption deviation coefficient, the endurance safety shell of the candidate node to the set of safe points is calculated. The endurance safety shell is determined based on the minimum predicted energy consumption, reserved safety energy and energy consumption prediction uncertainty of each safe point in the candidate node to the set of safe points. The energy consumption prediction uncertainty is determined based on the energy debt fluctuation value of each action type in the candidate path from the candidate node to the safe point. Based on the current remaining energy, the predicted energy consumption of candidate actions, and the endurance safety shell of the candidate action's arrival node, a set of safe actions is selected. Within the set of safety actions, strategy evaluation and strategy improvement are performed to generate UAV endurance and navigation optimization instructions.

2. The UAV energy consumption safety constraint navigation optimization method based on adaptive strategy iteration according to claim 1, characterized in that, The flight status includes the UAV's current position, flight altitude, flight speed, heading angle, attitude change frequency, and payload weight; the battery status includes remaining energy, output voltage, output current, battery temperature, discharge rate, and historical attenuation coefficient; the environmental status includes wind speed, wind direction, obstacle distribution, and no-fly zones; and the mission status includes mission points to be visited, completed mission points, mission point priority, and abandoned mission point markers.

3. The UAV energy consumption safety constraint navigation optimization method based on adaptive strategy iteration according to claim 1, characterized in that, The steps for constructing a candidate flight segment map include: The current location of the drone, the mission point, the obstacle avoidance point, the return point, and the alternate landing point are selected as candidate nodes. Candidate flight segments are generated based on the spatial reachability relationships between candidate nodes; Action type markers are configured based on the flight maneuver characteristics of candidate flight segments. The action type markers include level flight segments, climb segments, descent segments, turning segments, hovering operation segments, obstacle avoidance segments, and return segments.

4. The UAV energy consumption safety constraint navigation optimization method based on adaptive strategy iteration according to claim 1, characterized in that, The steps for calculating the predicted energy consumption for each candidate flight segment include: The basic flight energy consumption is calculated based on the segment distance, altitude change, and maneuver type markings of the candidate segments; Based on the load weight, wind speed, wind direction, attitude change frequency, and battery status, determine the load correction, wind field correction, attitude control correction, and battery attenuation correction respectively. The predicted energy consumption is obtained by fusing the basic flight energy consumption, load correction, wind field correction, attitude control correction, battery attenuation correction, and energy consumption deviation coefficient for the corresponding action type.

5. The UAV energy consumption safety constraint navigation optimization method based on adaptive strategy iteration according to claim 1, characterized in that, The steps for updating the energy consumption deviation coefficient for the corresponding action type include: Determine the action type marker corresponding to the actual flight segment; Calculate the deviation ratio between actual energy consumption and predicted energy consumption; The energy consumption deviation coefficient corresponding to the action type label is updated based on the deviation ratio and the adaptive update coefficient. Specifically, if the energy debt of the flight segment is positive, the energy consumption deviation coefficient of the corresponding action type is increased; if the energy debt of the flight segment is negative, the energy consumption deviation coefficient of the corresponding action type is decreased.

6. The UAV energy consumption safety constraint navigation optimization method based on adaptive strategy iteration according to claim 1, characterized in that, The steps for filtering the set of safety actions include: Get the candidate actions in the current state; Calculate the node reached after executing each candidate action; Determine whether the remaining energy after deducting the predicted energy consumption of the candidate actions is greater than the endurance safety shell of the reached node; Candidate actions that meet the judgment results are added to the safe action set, while candidate actions that do not meet the judgment results are excluded.

7. The UAV energy consumption safety constraint navigation optimization method based on adaptive strategy iteration according to claim 5, characterized in that, The steps for evaluating and improving a strategy include: Within the set of safety actions, the strategic value of each safety action is calculated based on task benefits, predicted energy consumption, endurance safety margin, action switching disturbance cost, and task delay cost. Select the security action with the highest strategic value as the current action to be executed; Based on the changing trend of energy debt in consecutive flight segments, the energy consumption weight and mission benefit weight in the strategy evaluation are adjusted, and the next round of strategy improvement is executed based on the adjusted weights. Specifically, if the energy debt of multiple consecutive actual flight segments is positive, the energy consumption weight is increased and the mission benefit weight is decreased; if the energy debt of multiple consecutive actual flight segments is negative and the endurance safety margin is greater than a set threshold, the energy consumption weight is decreased and the mission benefit weight is increased.

8. The UAV energy consumption safety constraint navigation optimization method based on adaptive strategy iteration according to claim 1, characterized in that, It also includes the downgrade navigation step: If there are no normal task actions in the set of safety actions to continue executing task point expansion, stop executing task point expansion; The emergency action set is reconstructed based on the current remaining energy. The emergency action set includes diversion action, return to base action, hovering and holding action, and mission abandonment action. Based on the current remaining energy, the predicted energy consumption at the return point, and the predicted energy consumption at the alternate landing point, the drone's endurance navigation optimization instructions are determined from the set of emergency actions and output.

9. A UAV energy consumption safety-constrained navigation optimization system based on adaptive strategy iteration, characterized in that, The UAV energy consumption safety constraint navigation optimization method based on adaptive policy iteration as described in any one of claims 1-8 includes: The state set construction module is used to acquire the drone's flight status, battery status, environmental status, and mission status to build a sustained navigation state set; A candidate segment map construction module is used to construct a candidate segment map based on the endurance navigation state set. The candidate segment map includes candidate nodes and candidate segments, and an action type marker is configured for each candidate segment. The predicted energy consumption calculation module is used to calculate the predicted energy consumption of each candidate flight segment based on the action type marker, flight status, battery status and environmental status. The energy consumption deviation update module is used to calculate the energy debt of the flight segment based on the actual energy consumed and the corresponding predicted energy consumption after the UAV completes the actual flight segment, and update the energy consumption deviation coefficient of the corresponding action type based on the energy debt of the flight segment. The endurance safety shell calculation module is used to calculate the endurance safety shell of the candidate node to the set of safe points based on the updated energy consumption deviation coefficient. The endurance safety shell is determined according to the minimum predicted energy consumption, reserved safety energy and energy consumption prediction uncertainty of each safe point in the candidate node to the set of safe points. The energy consumption prediction uncertainty is determined according to the energy debt fluctuation value of each action type in the candidate path from the candidate node to the safe point. The safety action filtering module is used to filter the set of safety actions based on the current remaining energy, the predicted energy consumption of candidate actions, and the endurance safety shell of the candidate action's arrival node. The strategy iteration module is used to perform strategy evaluation and strategy improvement within the set of safety actions, and generate UAV endurance and navigation optimization instructions.

Images