Unmanned aerial vehicle energy management optimization system based on dynamic demand

Abstract

The invention relates to the technical field of unmanned aerial vehicle cluster management, and discloses an unmanned aerial vehicle energy management optimization system based on a dynamic demand, which comprises a value potential field generation module, a broadcast module and a decision-making system arranged in an unmanned aerial vehicle, and is characterized in that the decision-making system generates a gravitation control instruction according to the gradient of a predictive value potential field; the method comprises the following steps that: a value gravity and density repulsive force real-time coupling instruction is established, the instruction can trigger an accompanying trend of unmanned aerial vehicle aggregation, meanwhile, a system generates a repulsive force adjustment parameter antagonistic to the gravity by sensing local unmanned aerial vehicle density, and finally, a net movement instruction is generated according to a counterbalance relationship between the repulsive force adjustment parameter and the gravity repulsive force adjustment parameter, and the value gravity and density repulsive force real-time coupling closed-loop self-adjustment mechanism is established; therefore, the problems of excessive resource aggregation and internal consumption inevitably caused by pure taxis guidance are solved, and an unmanned aerial vehicle cluster is changed from a system which is possibly unstable due to positive feedback into a steady-state ecological system which has macroscopic negative feedback regulation capability and can maintain low-entropy orderliness.

Description

Technical Field

[0001] This invention relates to an energy management optimization system for unmanned aerial vehicles (UAVs) based on dynamic demand, belonging to the field of UAV swarm management technology. Background Technology

[0002] Currently, in the operation of drone swarms, especially in large-scale, highly dynamic applications such as urban instant logistics, energy management typically follows a task-driven response approach. This means that after receiving a specific task instruction, the drone's management system passively calculates flight paths and power consumption to ensure the execution of a single task. This approach has the advantage of reliable and direct execution logic when handling independent, discrete tasks. However, as the scale of operations expands and task flows dynamically change, the limitations of this response approach under specific conditions become apparent. During periods of low task demand, a large number of drones that have completed their tasks are scattered in a standby state. Their remaining energy is considered a static and cost-free idle resource under this management approach. In reality, the effectiveness of their carried energy is reduced due to the mismatch between their spatial and temporal location and the probability of future tasks. That is, a drone located in an area with sparse future tasks faces the risk of continuous dissipation of its carrying capacity.

[0003] To improve this situation, a direct technical solution is to build a centralized dispatch system to predict and issue precise pre-deployment location instructions to each standby drone in real time. However, this centralized, high-frequency global computing and communication will generate concentrated high-load computing demands and communication overhead for large-scale drone clusters. The overall complexity and operational stability of the system also face challenges. Once the central system or communication link is interrupted, the coordinated operation of the entire cluster will be interrupted. This indicates that while pursuing the economic efficiency of cluster operation, the centralized management technical path has inherent constraints in terms of large-scale application and system reliability. Moreover, existing technologies also have fixed mindsets at the underlying control logic level of energy management, failing to improve the overall operational efficiency of the cluster. For example, the authorization announcement number CN11206098 Chinese invention patent 2B discloses a dynamic balancing energy management method for fuel cell drones. The core logic of this method is to monitor and regulate the difference between the hydrogen consumption rate of the fuel cell and the power consumption rate of the lithium battery in real time, in order to keep the consumption process of the two energy sources synchronized and avoid premature depletion of a single energy source. However, this strategy is essentially a management mode aimed at maintaining the steady state of its own energy system. Its decision-making is entirely limited to the energy state inside the drone, and completely ignores the spatiotemporal location of the drone as a transportation resource and its relationship with the dynamically changing external mission requirements. Therefore, even if this method can achieve a fine balance of internal energy, it cannot solve the problem of the overall transportation value dissipation caused by the drone's improper location during standby.

[0004] Specifically, existing technologies for managing the energy and location of drone swarms face the following technical challenges and bottlenecks: 1. Existing technologies generally adopt a task-driven, passive response model, treating standby drones as static resources. This results in the spatial distribution of drones merely representing the endpoints of historical tasks, leading to a spatiotemporal mismatch with the probability distribution of future task requirements, causing systemic idle capacity and value dissipation. 2. The conflict between centralized management and system robustness: While attempts to achieve global pre-deployment through centralized computing power can theoretically alleviate resource mismatch, large-scale swarm applications incur significant computational and communication overhead, creating a single point of failure. The system's complexity, latency, and vulnerability become bottlenecks for its large-scale application. 3. Existing energy management methods often focus on the internal energy efficiency balance of individual drones, neglecting the external impact of their behavior (such as location) on the overall macro-efficiency of the swarm as members of the swarm. Therefore, the technical problem to be solved by this invention is the need for a technology that can break free from the dependence on high-frequency centralized commands and guide standby drones to perform distributed, self-organized location optimization. Summary of the Invention

[0005] This invention provides a dynamic demand-based drone energy management optimization system. Its main purpose is to solve the problem that existing drone energy management methods treat passive standby as cost-free idleness and lack a decentralized and effective mechanism to guide drones to self-organize deployment to match future dynamic demands.

[0006] To achieve the above objectives, the present invention provides a dynamic demand-based unmanned aerial vehicle (UAV) energy management optimization system, which includes: A value potential field generation module is configured to generate and dynamically update a predictive value potential field covering a preset operating area, representing the value density of future task opportunities, based on historical task data covering a preset time span and real-time task data collected at a preset update frequency. A broadcast module is configured to periodically broadcast a predictive value potential field to a swarm of drones within the operational area; Each drone in a drone swarm has a built-in decision-making system, which includes: A gravity command generation unit is configured to generate a gravity control command that drives the UAV to move to a higher region of the value potential field based on the gradient of the received predictive value potential field. The execution of the command will trigger a co-occurrence trend of the UAV gathering in a local region with a high value potential field. A density sensing module is configured to determine a density value in real time on the drone's local location, representing the density of the drone, by passively sensing other drones in the vicinity. A repulsion adjustment module is configured to, in response to the associated tendency of aggregation, convert the real-time determined density value into a repulsion adjustment parameter that is positively correlated with the density value through a preset nonlinear function. A closed-loop decision-making unit has its internal operating rules set as follows: the repulsive force adjustment parameter is used as a constraint against the gravity control command to generate a final net movement command and drive the UAV, so that the operation of the UAV in the region of higher value potential field is suppressed by the real-time change of local UAV density.

[0007] Preferably, the decision-making system is further configured to: when the UAV receives a mission invitation, determine whether to accept the mission based on the change in comprehensive potential energy caused by the execution of the mission, which is jointly determined by the UAV's remaining energy and the predictive value potential field value of the mission endpoint location, so as to give priority to accepting missions that can increase comprehensive potential energy.

[0008] Preferably, the repulsion adjustment module is configured to perform a transformation based on a preset nonlinear function with density value as input and repulsion adjustment parameter as output. The function curve of the nonlinear function is set such that: the output is zero when the density value is below a first threshold determined by a nonlinear growth inflection point based on the average task response time of the cluster; the output increases with the increase of density value when the density value is between the preset first threshold and a second threshold determined by the average task response time reaching saturation; and the output reaches a saturation upper limit value when the density value is above the preset second threshold.

[0009] Preferably, the closed-loop decision unit is configured to: generate net movement commands according to a rule for calculating net potential energy within a continuously iterative decision cycle; the rule for calculating net potential energy is defined as follows: ,in For net potential energy, The combined potential energy is determined by the drone's current remaining energy and the predictive value potential field. The repulsion term is determined by the repulsion adjustment parameter; the closed-loop decision unit is further configured to: take the gravity control command as a reference vector, and adjust the reference vector according to the repulsion term to generate the net movement command, wherein the adjustment operation is to seek the... The goal is to maximize the expected growth rate.

[0010] Preferably, the UAV decision-making system further includes: a peer-to-peer network communication module, wherein when the UAV receives a task with a value higher than a preset value threshold, the UAV, as the event source UAV, broadcasts a ripple signal containing the task location, value information, and event identifier to other UAVs within its physical perimeter via the peer-to-peer network communication module; and the decision-making system further includes a local potential field correction unit, configured to, when the UAV, as the receiver, receives the ripple signal, temporarily add an incremental value positively correlated with the value information contained in the ripple signal to the value at the task location contained in the predictive value potential field, so as to form a corrected value potential field locally on the UAV, and generate gravity control commands based on the gradient of the corrected value potential field.

[0011] Preferably, the decision-making system further includes: a ripple sequence caching unit configured to locally cache a series of consecutive ripple signals with the same event identifier received by the UAV, wherein the cached ripple signals contain their respective task position information and timestamp information; and a trajectory prediction unit configured to, based on the task position information and timestamp information in the series of consecutive ripple signals cached by the ripple sequence caching unit, calculate a moving velocity vector of the event locally and in real time on the UAV using a linear regression algorithm, and calculate an interception point based on the moving velocity vector and the UAV's own flight speed; and a gravity command generation unit configured to use the interception point as the target position for generating gravity control commands.

[0012] Preferably, the decision-making system further includes: an intrinsic cost perception module configured to obtain the actual motor output power of the UAV flight control system and compare it in real time with a reference power calculated based on the current flight state and a pre-stored aerodynamic model locally on the UAV to generate an energy cost index characterizing the current flight energy cost; and a value potential field distortion unit configured to use the energy cost index to correct the predictive value potential field before the gravity command generation unit generates gravity control commands, so as to generate a corrected value potential field reflecting the flight energy cost. The inherent operating rule of the correction process is set as follows: the value of any target point in the predictive value potential field is divided by a preset function calculated by taking the current energy cost index and the angle between the UAV heading and the target point direction as input to arrive at the target point.

[0013] Preferably, the value potential field generation module is configured to use a spatiotemporal Poisson process model to predict the value density of future task opportunities based on historical and real-time task data.

[0014] Preferably, the local potential field correction unit is configured to: set a preset time decay coefficient for the incremental value, so that the magnitude of the incremental value decreases over time; and, when the UAV receives a new round of predictive value potential field broadcast by the broadcast module, clear all corrections caused by the incremental value.

[0015] Preferably, the closed-loop decision unit is configured to: when driving the UAV to move autonomously, adopt a movement strategy, the movement strategy including controlling the flight speed of the UAV within a pre-set economic cruising speed range, and planning a smooth movement trajectory that can avoid known headwind areas, so as to maximize the growth rate of the UAV's net potential energy under the condition of minimum energy consumption.

[0016] Compared with the prior art, the beneficial effects of the present invention are: 1. This invention broadcasts a predictive value potential field, and the UAV moves autonomously according to the potential field gradient. This transforms the standby UAVs during mission intervals from static idle resources into dynamic resources that actively seek the optimal pre-deployment location. This changes the overall spatial distribution of the cluster from a passive response to the past to a pre-fitting of the probability of future demand. At the system level, it transforms the inherent energy and time dissipation into a pre-investment in the overall response efficiency of the cluster.

[0017] 2. This invention combines the attractive force of value representing opportunity with the repulsive force of density representing competition. This real-time antagonistic negative feedback mechanism forms a dynamic equilibrium at the individual drone level, while avoiding local congestion and resource consumption caused by purely profit-driven guidance at the macro level of the cluster, thus maintaining the low-entropy order and healthy steady state of the cluster distribution.

[0018] 3. This invention integrates the strategic guidance of the globally slowly changing value potential field with the tactical opportunities of local instantaneous ripple signals on the UAV, and further introduces endogenous flight cost perception for decision correction. This makes the UAV's final decision a comprehensive result of the combined effects of economic opportunities, physical costs and group competition, ensuring the dual rationality of each autonomous movement in terms of energy economy and system integrity. Attached Figure Description

[0019] Figure 1 This is a system flowchart illustrating the coupling of value attraction and density repulsion in this invention. Figure 2 This is a performance comparison chart of key operational indicators between the system of this invention and the control group; Figure 3 This is a system information interaction use case diagram for the autonomous decision-making of the UAV in this invention. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be described in further detail below. Obviously, the described embodiments are only some embodiments of this invention, not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0021] This invention provides a dynamic demand-based drone energy management optimization system, applicable to large-scale, highly dynamic urban real-time logistics operation networks. It addresses the technical problem of systemic capacity mismatch and efficiency dissipation caused by the mismatch between the spatiotemporal location and future mission probability of standby drones, which are statically idle resources. The system architecture primarily includes a value potential field generation module deployed on a ground server, a broadcast module, and a decision-making system integrated into each drone in the drone swarm. The value potential field generation module generates and updates global guidance information representing future operational value. The broadcast module distributes this information to all drones within the operational area. The drone's built-in decision-making system, based on this global guidance information and its own perceived local state information, performs self-organized standby position optimization and mission decision-making, thus forming a closed-loop, self-regulating energy management optimization system.

[0022] In one specific implementation, to address the risk of energy dissipation during mission lulls due to mismatches between the location and future mission opportunities, the system is equipped with a value potential field generation module. This module is implemented as a prediction engine deployed on a cloud server based on a spatiotemporal Poisson process model. The initial state definition procedure of this model is as follows: the key technical characteristics of its input object are defined as a structured dataset containing records of all historical missions in the operating area over the past 24 months. Each record includes the 3D geographic coordinates of the mission, the timestamp of the event, and the final settlement value of the mission. The functional specifications of the enabling environment are defined as having... A computing server with at least 16 CPU cores and 64GB of memory is sufficient to support statistical modeling calculations for millions of data points. Within an operational cycle, for example, every 5 minutes, this module performs a calculation. Its internal operating rules are set as follows: the entire preset operational area, such as a 10km x 10km urban area, is divided into 100m x 100m grid cells on a horizontal plane. Based on the aforementioned historical and real-time task data, it performs probability density prediction of the total value of tasks that may occur in each grid cell within the next 30 minutes. The output of this prediction process is a two-dimensional matrix covering the entire operational area, where the value of each element is the predictive value potential field value of the corresponding geographical location. For example, one hour before the evening rush hour, a grid unit located in a core business district ( ) corresponding The value might be calculated as 95.8, while the grid cell located in a certain country park ( ) corresponding The value is 3.2. This predictive value potential field is broadcast periodically to the drone swarm in the operating area in the form of data packets through the broadcast module, thereby providing each drone with global guidance information on where there is higher operating value in the future, giving its autonomous movement during standby an economic basis.

[0023] To enable drones to autonomously respond to the guidance of the aforementioned value potential field, the drone's built-in decision-making system includes a gravity command generation unit. In urban instant logistics applications, when a drone completes a delivery task and enters a standby state, it faces the decision of where to wait for the next task. Therefore, the gravity command generation unit is configured to, upon receiving the latest value potential field data from the broadcast module, first identify the drone's current grid cell and its eight adjacent grid cells. The value is then calculated, and then a gravity control command is generated to drive the UAV to move to a higher region of the value potential field by calculating the gradient between these nine values, i.e. the direction of the largest rate of change. This command is a two-dimensional velocity vector in the data structure, which points to the neighboring grid with the highest value. Its magnitude is proportional to the value gradient, but is constrained by a preset upper limit of economic cruising speed, such as 5 m / s. Through this operation mode, the UAV is no longer silent and idle in the standby state, but is transformed into a dynamic resource that continuously seeks the optimal pre-deployment position, so that the spatial distribution of the UAV actively tends to the high probability area of ​​future missions.

[0024] Purely profit-driven guidance can lead to excessive concentration of resources in a few high-value areas, resulting in the risk of localized congestion and resource depletion. To address this issue, the system introduces a constraint mechanism that counteracts gravity through the built-in density sensing module and repulsion adjustment module of the UAV. The density sensing module utilizes the communication link for status broadcasting between UAVs to passively monitor the number of beacon signals broadcast by other nearby UAVs, thereby determining a density value in real time that represents the local UAV density at the current location. This density value is a floating-point number, with units of UAVs per square kilometer. The repulsion adjustment module, based on a preset nonlinear function, transforms this real-time determined density value into a repulsion adjustment parameter that counteracts gravity control commands. The calibration procedure for this nonlinear function is as follows: In a simulation environment, a value potential field with a single value peak is first established, and then the number of UAVs deployed is gradually increased, with data recorded. The average task response time of the human-machine swarm is determined by a process quantification procedure as follows: the drone density at which the average response time begins to show a non-linear growth inflection point, such as an increase exceeding 5%, is set as the first threshold, for example, 20 drones / km²; the drone density at which the average response time reaches a saturation plateau, such as an increase exceeding 50%, is set as the second threshold, for example, 80 drones / km². Accordingly, the non-linear function is set as follows: when the density value is below the first threshold, the repulsion adjustment parameter output is zero, producing no suppression effect; when the density value is between the first and second thresholds, the output increases linearly with the increase in density; when the density value is above the second threshold, the output reaches a saturation upper limit value to generate the strongest constraint. This repulsion adjustment parameter is vector-operated with the gravity control command in the closed-loop decision unit to generate a net movement command, thereby suppressing profit-seeking behavior through local density and avoiding excessive aggregation of transportation resources.

[0025] To integrate global, slowly changing guidance information with local, real-time, and sudden information to address high-value but short-duration sudden mission opportunities, the UAV's decision-making system also includes a peer-to-peer network communication module and a local potential field correction unit. When a UAV in the cluster receives a mission with a value exceeding a pre-set value threshold, such as an emergency medical delivery mission (valued more than five times that of a normal mission), this UAV, acting as the event source UAV, broadcasts a ripple signal containing the mission's location, value information, and a unique event identifier to other UAVs within a 500-meter physical radius via the peer-to-peer network communication module. Upon receiving this ripple signal, the local potential field correction unit in the decision-making system of the other UAVs temporarily adds an increment positively correlated with the value information contained in the ripple signal to the value at the mission location contained in the ripple signal in their locally cached predictive value potential field matrix. The calculation of this increment is set to... ,in, The valuable information in the ripple signal. The increment is set to a preset amplification factor, such as 1.5. At the same time, the increment is configured with a time decay factor, such as decaying by 20% every 10 seconds, until it is cleared when a new round of global predictive value potential field is received by the broadcast module. Thus, a corrected value potential field that integrates global prediction and local burst information is formed locally on the UAV. Gravity control commands are generated based on the gradient of the corrected value potential field, so that nearby standby UAVs autonomously move toward the high-value mission location to form a response echelon.

[0026] In dealing with dynamically evolving events, For example, when tracking a moving emergency target, a swarm of drones chasing only one reported location of the event would result in a response lag. To achieve a shift from reactive pursuit to predictive interception, the drone's decision-making system also includes a ripple sequence buffering unit and a trajectory prediction unit. The ripple sequence buffering unit is configured to locally buffer a series of consecutive ripple signals with the same event identifier received by the drone. The buffer queue length is set to 5, and each buffer entry contains {task position coordinates, timestamp}. When the number of buffered ripple signals reaches a minimum threshold, such as 3, the trajectory prediction unit is activated. Its internal operating rule is set to a linear regression algorithm based on least squares. This algorithm takes a series of position and timestamp data points in the buffer queue as input and calculates a velocity vector of the event, including direction and speed, locally on the drone in real time. For example, if the positions of three consecutive ripple signals are... (100, 100), (110, 112), (121, 123), timestamps are respectively =0s, =5s, =10s, the algorithm will calculate that the event is moving in a specific direction at a speed of about 2.1 m / s; then, the trajectory prediction unit calculates an optimal interception point based on the moving speed vector and the UAV's own economic cruise speed. This interception point is then used by the gravity command generation unit as the target position for generating gravity control commands, thereby updating the UAV's moving target point from the last known position of the event to an interception point.

[0027] In actual operating environments, factors such as wind fields cause the energy cost of UAV flight to vary in different directions. A theoretically high-value target point may be a real energy trap if it is located in a strong headwind. Therefore, the decision-making system also includes a cost perception module and a value potential field distortion unit. The cost perception module obtains the actual motor output power of the UAV flight control system and compares it in real time with a reference power calculated based on the current flight state and a pre-stored aerodynamic model locally on the UAV. This generates a dimensionless energy cost index (ECI) characterizing the current flight energy cost. For example, if the reference power for cruising at 5 m / s in standard still air is 100 watts, and the current actual output power is 150 watts, then the ECI is calculated to be 1.5. The value potential field distortion unit uses this ECI to adjust the energy cost of the UAV before the gravity command generation unit generates the gravity control command. The received predictive value potential field undergoes correction processing. Its inherent operating rule is set as follows: the value of any target point in the predictive value potential field is divided by a preset function calculated using the angle between the current ECI, the UAV's heading, and the target point's direction as input. This results in a decrease in the value of target points in the upwind direction and a relative increase in the value in the downwind direction. Therefore, the UAV's movement command is the result of both economic opportunity and physical feasibility, ensuring the rationality of its profit-driven autonomous movement in terms of energy economy. At the mission acceptance decision level, when the UAV receives a mission invitation, its decision system does not simply accept it. Instead, it judges whether to accept the mission based on the change in comprehensive potential energy caused by executing the mission, determined by the UAV's remaining energy and the predictive value potential field value at the mission endpoint. This comprehensive potential energy is defined as... ,in, This represents the estimated remaining battery power after the mission is completed. The decision-making system prioritizes tasks that increase overall potential energy, based on the value potential field of the mission endpoint at the expected arrival time. Thus, the acceptance or rejection of a task becomes a factor in optimizing the long-term operational value of the drone.

[0028] Example 1: On Tuesday afternoon at 3 PM, in a large urban instant logistics drone network, the operation system enters a period of low workload. Dozens of drones that have just completed the previous round of delivery tasks are scattered throughout the city in a standby state. At this time, the city's meteorological system issues a warning that a short-term severe convective weather event will hit Area A in the western part of the city one hour later. This warning triggers an immediate demand for emergency supplies from a large number of users in Area A. At the same time, a high-value cold chain transportation task for biological products, which needs to urgently cross the city and whose path happens to pass through the periphery of Area A, is generated. The system faces the dual operational pressure of responding to a localized surge in demand in a short period of time and ensuring that the high-value task can traverse the harsh environment. In the initial stage of this scenario, after receiving the meteorological warning data and the real-time data of the order flow in Area A, the system's value potential field generation module, through its spatiotemporal Poisson process model, generates the predictive value potential field value of the grid cells covering Area A within a 5-minute calculation cycle. The average value increased from 15 during the plateau period to over 80, while other areas of the city... The value remains at a low level. A potential field map with a sharp spatial gradient, representing the value density of future mission opportunities, is generated and distributed through the broadcast module. At this time, the drones scattered in the eastern and central parts of the city, in a standby state, autonomously generate gravity control commands based on the received value potential field gradient, driving these drones to redeploy to the periphery of Area A at an economical cruising speed of 5 meters per second, without waiting for any point-to-point scheduling commands from the central system. This operation mode transforms the energy consumed by passively waiting in the traditional way into deployment investment to improve future response efficiency. As the standby drones gradually gather to the periphery of Area A, when the local drone density at the edge of Area A exceeds the set first threshold, namely 20 drones / square kilometer, the repulsive force adjustment module in the drone decision-making system begins to take effect. This mechanism, along with the attractive force provided by the gravity command generation unit, forms a dynamic balance, causing subsequent drones to no longer continuously rush towards the highest point of the value potential field. Instead, they form a distributed standby cluster in the surrounding secondary value areas. This synergistic effect of attraction and repulsion allows the drone cluster to maintain a spatial distribution that avoids excessive concentration of transport capacity while moving towards high-value areas, thus balancing the operational requirements of hotspot response and balanced layout.

[0029] Meanwhile, the event source drone, having accepted the emergency biological product transport mission, immediately broadcast a ripple signal containing mission location and value information to its surrounding area via a peer-to-peer network communication module upon startup. Upon receiving this signal, drones near the flight path temporarily increased the value potential field value of the corresponding area through their local potential field correction units. This allowed these drones to proactively create a priority path and also enabled them to act as potential support units, moderately following the high-value mission's flight path. When the event source drone reached the edge of Area A and was about to enter a strong crosswind area, its cost perception module detected that the ratio of the motor's actual output power to the reference power, i.e., the Energy Cost Index (ECI), increased from the usual 1.1 to 1.8. The value potential field distortion unit then used this ECI to correct the drone's local value potential field map. A target point with the theoretically shortest straight-line distance but high crosswind energy consumption had its value reduced in the corrected value potential field, while a target point that was slightly farther away but located in... The value of waypoints in tailwind areas is relatively enhanced, leading the UAV's closed-loop decision-making unit to choose a more energy-efficient detour. Here, the global predictive value potential field provides operational value guidance, while the UAV's local ECI provides immediate physical cost constraints. The integration of these two in the UAV's local decision-making system transforms the UAV's flight path decision-making from a simple task scheduling problem into a comprehensive optimization process that integrates operational economy and flight physical cost. Ultimately, when the emergency material delivery task in Area A reaches its peak, a large number of UAVs have already formed a high-density standby cluster around Area A through autonomous pre-deployment, reducing the task response time by 60% compared to historical records of similar emergencies. The emergency biological product transportation task also avoids the high energy consumption risk brought by severe convective weather, completing the delivery with more remaining power than expected. The spatial distribution of the entire UAV cluster changes from an initial random and dispersed state to a pre-deployed state corresponding to the distribution of sudden and concentrated demand.

[0030] Example 2: To quantify the effect of the closed-loop self-regulating mechanism combining value attraction and density repulsion in the technical solution of this invention on suppressing excessive aggregation of UAV swarms, this example conducted a comparative simulation experiment. The experiment aimed to objectively verify the mechanism's role in improving the operational efficiency of UAV swarms through data. The experiment was conducted on a discrete event simulation platform based on a multi-agent queuing network model. This platform simulated a 10 km x 10 km urban operating area. The task generation logic within this area followed a spatiotemporal Poisson process, which was set to continuously generate high-value tasks within a 1 km² area near the area's center (5 km, 5 km), thereby reproducing a stable, continuously attractive high-value area for UAVs in the simulation environment. To balance the statistical significance of the simulation results with... The simulation resource overhead was calculated, with a total operation period of 8 hours and a drone swarm size of 100 drones. This scale was used to induce congestion in high-value areas. Two groups were set up: a control group and the prototype group of this invention. Both groups used the same simulation platform and task generation model. The control group's drone decision-making system only activated the gravity command generation unit, meaning the drones only moved based on the gradient of the predicted value potential field. The prototype group's drone decision-making system fully activated all the aforementioned modules, especially including a repulsive force adjustment module that performs negative adjustment based on local density. During the experiment, the simulation platform iterated at 1-second time steps and recorded various performance indicators. After the 8-hour simulation period, the data from both groups were statistically averaged, resulting in the comparison results shown in Table 1.

[0031] Table 1: A comparison of performance indicators between the control group and the sample group of the present invention.

[0032] Experimental grouping Average drone density in hotspot areas (number of drones / year) Average task waiting time (min) Average energy consumption rate (W) of drones Task completion rate (%) control group 78.5 12.3 145.2 91.3 Sample of the present invention 25.1 2.8 110.7 99.2 Referring to Table 1, in the control group, due to the lack of a suppression mechanism, a large number of drones were continuously attracted and concentrated in a 1 square kilometer hotspot area, with an average drone density of 78.5 drones / square kilometer. This high density resulted in an average waiting time of 12.3 minutes when new tasks were generated in this area. Simultaneously, the prolonged queuing and hovering also increased the average energy consumption rate of drones to 145.2 watts, and some drones returned to base due to energy depletion, resulting in an 8.7% task failure rate. In the sample group of this invention, the repulsion adjustment module took effect after the local drone density exceeded a preset threshold, suppressing the influx of subsequent drones and ensuring a more even distribution of drones in the hotspot area. The average density was maintained at 25.1 drones per square kilometer. At this density level, the average mission waiting time was 2.8 minutes. The orderly allocation of resources reduced ineffective hovering and maneuvering, keeping the average energy consumption rate at 110.7 watts and ensuring a mission completion rate of 99.2%. Experimental data showed that relying solely on the profit-seeking guidance mechanism of the value potential field can lead to excessive aggregation of transport resources under certain conditions, resulting in a deterioration of the overall system's operational efficiency. By introducing a repulsive force adjustment mechanism that is coupled in real time with the local drone density, it is possible to effectively suppress resource consumption caused by positive feedback while using value gravity for resource deployment, thereby improving mission response efficiency and energy utilization efficiency.

[0033] Example 3: This example combines Figures 1 to 3 This describes a dynamic demand-based drone energy management optimization system, such as... Figure 1 As shown, the process begins by taking historical and real-time task data as input data and processing it through the value potential field generation module. This module generates and updates a predictive value potential field based on this data. The predictive value potential field is then transmitted to the broadcasting module, which periodically broadcasts it in the form of value potential field guidance information. Upon receiving this guidance information, the UAV's built-in gravity command generation unit generates a beneficial gravity control command based on the gradient of the value potential field. Simultaneously, the density sensing module in the system senses the local UAV density in real time and transmits this density information to the repulsion adjustment module. The latter converts the density value into a repulsion adjustment parameter as an antagonistic constraint. Finally, the gravity control command and the repulsion adjustment parameter are input to the closed-loop decision unit. This unit generates the final output command, i.e., the net movement command, by balancing gravity and repulsion, thereby driving the UAV to perform movement.

[0034] like Figure 2 As shown, two bar charts are used to represent the performance data of the control group and the sample group of the present invention, respectively. This comparison covers four key performance indicators, namely, hotspot area density (unit: racks / ...). The system sample group, compared to the control group, showed a significant reduction in hotspot density, task waiting time (in minutes), energy consumption rate (in W), and task completion rate (in %). The graphical comparison of these data demonstrates that the system sample group showed a significant improvement in task completion rate compared to the control group. This verifies the beneficial effects of the present invention's technical solution in improving cluster operation efficiency and resource utilization.

[0035] like Figure 3 As shown, system administrators and external data systems, such as those containing weather and order information, act as external participants, providing input for the system's value potential field generation function. The UAV, as the core executor, mainly revolves around two core functions: autonomous position optimization and accepting or rejecting tasks. The autonomous position optimization function is related to the broadcast guidance information function and drives the UAV's standby position adjustment through a series of internal calculations, such as the perceived local density, perceived flight cost, and the final calculated net movement command, as shown in the dashed ellipse box in the figure.

[0036] Example 4: To ensure stable performance of the system in different operating environments, a standardized deployment and calibration procedure is required for several core algorithm modules built into the system. This example aims to disclose the specific combination algorithm of gravity and repulsion in the UAV closed-loop decision unit, the parameterization implementation of the nonlinear function in the repulsion adjustment module, and the lifecycle management method of the prediction model in the value potential field generation module, so as to eliminate uncertainties in the implementation process. In a specific engineering deployment scenario, the challenge lies in how to effectively use the scalar information of local density sensed by the UAV to adjust the vector information of gravity control command generated by the value potential field gradient. To this end, the closed-loop decision unit adopts an algorithm based on vector modulus adjustment. The specific process is set as follows: First, the gravity command generation unit calculates the original gravity control command pointing to the region with higher value, denoted as vector. The second step involves the repulsion adjustment module adjusting the force based on the current local drone density. With a pre-defined nonlinear function Calculate a repulsion adjustment parameter in the range [0, 1]. ,in, The third step is for the closed-loop decision-making unit to use the formula... To calculate the final net movement instruction The internal operating logic of this algorithm is that when the local density is low, Approaching zero, the net movement command is essentially equal to the original gravitational command. When local density increases, As the value approaches 1, the magnitude of the net movement command is reduced to near zero, thereby producing a suppressive effect at the motion level. This process achieves effective fusion between information of different dimensions.

[0037] It should be noted that the above nonlinear function A preferred implementation is to use a parameterized sigmoid logic stethoscope, the specific form of which is: The parameter calibration procedure for this function is as follows: First, based on the simulation calibration method in Example 2, obtain the first threshold at which the suppression effect begins to appear. And the second threshold where the inhibitory effect tends to saturate Secondly, the center point parameter of the function... Set as the arithmetic mean of two thresholds, i.e. Next, set the growth rate parameter of the function. Its value is determined by a constraint condition, namely, when the density equals When, the function value It should be a small starting value, such as 0.1, from which the result can be calculated. The value; for example, if obtained through simulation , ,but It was determined to be 50, and will Set to approximately 0.15 to meet the requirements. The boundary conditions; through this procedure, an abstract nonlinear function is transformed into a set of deterministic parameters with clear physical meaning that can be directly configured by engineers; furthermore, to ensure the long-term prediction accuracy of the spatiotemporal Poisson process model in the value potential field generation module, its deployment adopts a lifecycle management approach combining offline initialization and online updates; in the initial deployment phase of the system, offline initialization training is performed, the input of which is historical task data covering at least one year of operation, and the intensity function of the Poisson process is estimated using a maximum likelihood estimation algorithm. A fitting process is performed to obtain a basic model that reflects the correlation between the probability of task occurrence and variables such as day of the week, specific time period, and geographical location. After the system is put into actual operation, it switches to online incremental update mode. The inherent operating rules of this mode are set as follows: the system continuously caches real-time task data from the past 7 days locally and updates the data based on a sliding time window. Every day at dawn, the base model is incrementally updated using the latest data. The update algorithm uses gradient descent with a time decay factor, giving higher weights to more recent data. In this way, the model can learn autonomously and adapt to changes in task distribution caused by emerging business hotspots or urban planning changes, so that its predictive ability will not decline over time.

[0038] Example 5: To improve the accuracy of the Energy Cost Index (ECI) calculated by the cost-aware module in the system, each UAV undergoes a standardized offline reference power calibration procedure before its initial deployment or after any maintenance affecting its aerodynamic characteristics. This procedure requires the UAV to perform a series of preset flight maneuvers in an open area with no wind or low wind speed, including vertical hovering under different loads and horizontal straight flight with a 10% increment gradient within the range of 0% to 100% economic cruise speed. In each stable flight state, the flight control system records the actual average motor output power required to maintain that state and associates this power value with the corresponding flight state, including speed and load, thereby generating and storing an aerodynamic model specific to that UAV in the form of a lookup table. This model is then used by the cost-aware module to calculate the reference power for any flight state.

[0039] To balance the transmission efficiency of high-value tactical information with the occupancy rate of communication channels in the ripple signal mechanism, the value threshold used to trigger ripple signal broadcasting in the system adopts a dynamic and adaptive setting method. Specifically, the system's background management module continuously calculates the statistical distribution of the value of completed tasks within each grid cell, including its average value, based on a sliding time window, such as the past hour, using grid cells as units. with standard deviation Accordingly, within this grid cell, the value threshold for triggering a ripple signal broadcast when a drone receives a new task. It is then dynamically set to This allows the triggering conditions for broadcasting behavior to be adaptively adjusted based on the real-time task value distribution in different regions and time periods.

[0040] Example 6: To provide a reproducible parameter benchmark that balances energy consumption and potential energy growth efficiency for the autonomous movement strategy adopted by a UAV in standby mode, this example discloses a standardized offline calibration procedure for determining the economic cruise speed range of a UAV. The objective function of this procedure is set to maximize the comprehensive potential energy growth per unit energy consumption. The procedure requires placing a standard-load UAV in a windless environment. First, it is instructed to fly horizontally in a straight line at multiple discrete speed points (e.g., with a gradient of 1 m / s) from its lowest stable flight speed to its highest cruise speed. Onboard sensors are used to accurately record the energy consumption per unit distance at each speed point. Secondly, in the simulation environment, a value potential field with a standard linear gradient is set, and the operation of the UAV at different speeds is calculated. The time growth rate of its overall potential energy during flight Finally, the energy efficiency ratio at each speed point was calculated. The speed range corresponding to the peak region of this ratio, such as the speed range covered by more than 90% of the peak, is calibrated and stored as the economic cruise speed range of the UAV. When the UAV's decision module performs autonomous movement, its flight speed will be preferentially constrained within this range.

[0041] To further verify the key role of the repulsive force adjustment mechanism introduced in this invention, which is coupled with the local UAV density in real time, in solving the problem of excessive resource aggregation, the following comparative examples are provided.

[0042] Comparative Example 1: To objectively evaluate the actual effect of the technical solution of the present invention, this comparative example uses the same simulation platform, environmental parameters, and task generation logic as Example 2, including a 10 km x 10 km urban operation area, a hotspot area located in the center of the area that continuously generates high-value tasks, and a cluster of 100 drones. The only difference from the embodiment of the present invention is that the drone decision-making system in this comparative example is set to a simplified configuration with only profit-seeking guidance capabilities. Specifically, the density sensing module and repulsion adjustment module are removed from the system, and the operating rules of its closed-loop decision-making unit are modified to: the gravity control command generated by the gravity command generation unit based on the gradient of the predictive value potential field is directly executed as the final net movement command, without introducing any antagonistic constraints based on local drone density. This configuration aims to simulate a purely opportunity-value-based guidance technical path that is most likely to be adopted by those skilled in the art when solving the problem of pre-deployment of standby drones. The experimental process is the same as in Example 2. After an 8-hour simulation operation cycle, the key operation indicators are statistically averaged, and the results are recorded in Table 2.

[0043] Table 2: Comparison of key performance indicators between Comparative Example 1 and the technical solution of the present invention.

[0044] Experimental grouping Average drone density in hotspot areas (number of drones / year) Average task waiting time (min) Average energy consumption rate (W) of drones Task completion rate (%) Comparative Example 1 78.5 12.3 145.2 91.3 Test group of the method of the present invention 25.1 2.8 110.7 99.2 During the simulation, it was observed that, due to the lack of a suppression mechanism, the drone swarm in Comparative Example 1, under the continuous attraction of the value potential field, concentrated in a large number of hotspot areas, resulting in an average density of 78.5 drones / square kilometer. The simulation log frequently recorded path replanning events caused by local airspace congestion. This high-density aggregation extended the average waiting time for new tasks to 12.3 minutes. At the same time, due to the long queuing, hovering, and evasive maneuvers, the average energy consumption rate of drones increased significantly to 145.2 watts. Ultimately, some drones returned to base early because they ran out of energy while waiting, resulting in a 91.3% task completion rate and 8.7% of tasks failing due to lack of transport capacity. The experimental results show that a technical solution that relies solely on the gradient of the value potential field for profit-seeking guidance, although it can guide drones to high-opportunity areas, inevitably leads to excessive aggregation of transport resources. This aggregation directly causes a series of problems such as deterioration of task response efficiency, increased energy consumption, and decreased task success rate, proving that this technical approach cannot fundamentally solve the problem of resource consumption in swarm operations.

[0045] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0046] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A dynamic demand based unmanned aerial vehicle energy management optimization system, comprising: The system comprises: a value potential field generation module configured to generate and dynamically update a predictive value potential field representing the value density of future task opportunities in a preset operation area based on historical task data covering a preset time span and real-time task data collected at a preset update frequency; a broadcasting module configured to periodically broadcast the predictive value potential field to a drone cluster in the operation area; a drone in the drone cluster, which is built-in with a decision system, the decision system comprising: a gravity instruction generation unit configured to generate a gravity control instruction driving the drone to move to a higher value potential field area according to the gradient of the received predictive value potential field, the execution of the instruction causing an accompanying tendency of aggregation of the drones in the local area with higher value potential field; a density perception module configured to determine a density value representing the local drone density at the position of the drone in real time through passive perception of other drones in the surrounding; a repulsive force adjustment module configured to convert the real-time determined density value into a repulsive force adjustment parameter positively correlated with the density value through a preset nonlinear function in view of the accompanying tendency of aggregation; a closed-loop decision unit, whose internal operation rules are set to take the repulsive force adjustment parameter as a restraint quantity antagonistic to the gravity control instruction to generate a final net movement instruction and drive the drone, so that the movement of the drone towards the higher value potential field area is inhibited by the real-time change of the local drone density at the position of the drone.

2. The dynamic demand based UAV energy management optimization system of claim 1, wherein, The decision system is further configured to judge whether to accept a task invitation based on the change of a comprehensive potential energy determined by the residual energy of the drone and the predictive value potential field value of the position where the task ends caused by the execution of the task, and preferentially accept the task that can increase the comprehensive potential energy.

3. The dynamic demand based UAV energy management optimization system of claim 1, wherein, The repulsive force adjustment module is configured to perform conversion according to a preset nonlinear function with the density value as the input and the repulsive force adjustment parameter as the output, and the function curve of the nonlinear function is set to output zero when the density value is lower than a first threshold value determined based on the nonlinear growth inflection point of the average task response time of the cluster, increase with the increase of the density value when the density value is between the first threshold value and a second threshold value determined based on the saturation of the average task response time, and reach a saturated upper limit value when the density value is higher than the second threshold value.

4. The dynamic demand based UAV energy management optimization system of claim 1, wherein, The closed-loop decision unit is configured to generate a net movement instruction in a decision cycle of a continuous iteration according to a rule for calculating a net potential energy, the calculation rule of the net potential energy is defined as wherein, is the net potential energy, is a comprehensive potential energy determined by the current residual energy of the unmanned aerial vehicle and the predictive value potential field, is a repulsive force term determined by a repulsive force adjustment parameter; the closed-loop decision unit is further configured to take the attractive force control instruction as a reference vector, and adjust the reference vector according to the repulsive force term to generate the net movement instruction, wherein the operation of the adjustment aims to maximize the expected growth rate.

5. The dynamic demand based UAV energy management optimization system of claim 1, wherein, The decision system of the UAV further comprises a peer-to-peer network communication module, the UAV is configured to, when it accepts a task with a value higher than a preset value threshold, broadcast a ripple signal containing the task location, value information and event identifier as an event source UAV to other UAVs within its physical surrounding range through the peer-to-peer network communication module; and the decision system further comprises a local potential field correction unit configured to, when the UAV receives the ripple signal as a receiver, temporarily increase the value at the task location contained in the ripple signal in the predictive value potential field by an increment value positively correlated with the value information contained in the ripple signal, to form a corrected value potential field locally at the UAV, and generate the gravity control instruction based on the gradient of the corrected value potential field.

6. The dynamic demand based UAV energy management optimization system of claim 5, wherein, The decision system further comprises a ripple sequence cache unit configured to locally cache a series of continuous ripple signals with the same event identifier received by the UAV, the cached ripple signals containing respective task location information and timestamp information; and a trajectory prediction unit configured to calculate an event moving speed vector locally and instantaneously at the UAV based on the task location information and timestamp information in the series of continuous ripple signals cached by the ripple sequence cache unit through a linear regression algorithm, and calculate an interception point based on the moving speed vector and the flight speed of the UAV itself; and the gravity instruction generation unit is configured to take the interception point as the target position for generating the gravity control instruction.

7. The dynamic demand based UAV energy management optimization system of claim 1, wherein, The decision system further comprises an internally generated cost perception module configured to generate an energy cost index representing the current flight energy cost by acquiring the actual motor output power of the flight control system of the UAV and comparing it with a baseline power calculated based on the current flight state and a pre-stored aerodynamic model at the UAV; and a value potential field distortion unit configured to perform correction processing on the predictive value potential field using the energy cost index before the gravity instruction generation unit generates the gravity control instruction, to generate a corrected value potential field reflecting the flight energy cost, the internal operation rule of the correction processing being set as: dividing the value of any target point in the predictive value potential field by an estimated energy cost index to the target point calculated by taking the current energy cost index and the included angle between the heading of the UAV and the direction of the target point as inputs of a preset function.

8. The dynamic demand based UAV energy management optimization system of claim 1, wherein, The value potential field generation module is configured to adopt a spatiotemporal Poisson process model to predict the value density of future task opportunities based on historical and real-time task data.

9. The dynamic demand based UAV energy management optimization system of claim 5, wherein, The local potential field correction unit is configured to set a preset time decay coefficient for the increment value, so that the increment value decreases over time; and when the UAV receives a new round of predictive value potential field broadcast by the broadcast module, all corrections brought by the increment value are removed.

10. The dynamic demand based UAV energy management optimization system of claim 1, wherein, The closed-loop decision unit is configured to adopt a movement strategy when driving the UAV to move autonomously, the movement strategy comprising controlling a flight speed of the UAV within a pre-set economic cruise speed range and planning a smooth movement trajectory capable of avoiding a known headwind region.

Images