Optimization method for data acquisition of unmanned aerial vehicle in wireless power supply internet of things

Abstract

According to the optimization method for data acquisition of the unmanned aerial vehicle in the wireless power supply Internet of Things, a nonlinear energy collection model is adopted, an optimization target is modeled as a mixed integer nonlinear programming problem, and the optimization problem is decomposed into an equipment service sequence generation module (a main module) and an unmanned aerial vehicle trajectory optimization module (a sub module). In the main module, the service priority of equipment is dynamically adjusted on the basis of equipment data volume and historical service times, meanwhile, in view of mutual conflict of optimization targets, a multi-objective evolutionary strategy is adopted to generate a Pareto frontier solution set, and an optimal equipment access sequence of each round is screened from the Pareto frontier solution set, and in the sub-module, for the continuous action parameter adjustment problem of the unmanned aerial vehicle, the optimal equipment access sequence of the unmanned aerial vehicle is obtained. The input dimension of the neural network is reduced by constructing a lightweight state space, and a multi-target reward function including data collection amount and energy consumption is designed, so that collaborative optimization and long-term balance of equipment service fairness, the data collection amount and the energy consumption of the unmanned aerial vehicle are realized.

Description

Technical Field

[0001] This invention relates to the field of intelligent control methods for unmanned aerial vehicles (UAVs) in wirelessly powered Internet of Things (WPH) networks, and more particularly to an optimized method for UAVs to collect data in WPH networks. Background Technology

[0002] With the rapid development of IoT technology, massive numbers of terminal devices are being deployed in remote and harsh environments to perform tasks. Due to the special nature of these environments, such as toxic gases, radiation, and remote locations, manually replacing batteries or installing wiring is costly. Therefore, a data acquisition solution with efficient power supply capabilities is urgently needed. The development of 5G and even 6G technologies provides strong support for the widespread deployment and data collection of IoT. Among these solutions, wireless power stations based on radio frequency signals use wireless power transmission technology to charge low-power devices, avoiding the problems of battery replacement and wiring for IoT devices. This is a crucial means of solving the power supply problem for large-scale IoT devices. Another effective method is for IoT devices to obtain energy from renewable energy sources. Compared to the latter, WPT (Wireless Power Transmission) technology offers stability and continuity, providing better service to devices and extending their lifespan. However, WPS systems have limitations such as high construction costs and low charging efficiency over long distances or in environments with obstacles.

[0003] Drones, with their high mobility, low deployment costs, and low-altitude coverage, are considered ideal carriers for power replenishment and data backhaul of wide-area IoT devices. In particular, drones equipped with WPT technology can effectively overcome terrain limitations and provide efficient services to ground equipment. Drones have become an important component of IoT networks, not only expanding the application scenarios of IoT but also providing strong technical support for its development.

[0004] Existing IoT devices use harvest-transfer protocols based on RF technology for energy harvesting and data transmission. Specifically, IoT nodes collect energy via the downlink and then use that energy to transmit information via the uplink. Drones act as mobile full-duplex hybrid access points, simultaneously performing energy transmission and data reception. The optimization objectives of drone IoT systems vary significantly, closely related to the characteristics of the research object and its task requirements. Some optimization objectives aim to maximize the uplink throughput of IoT devices; some analyze the impact of different time allocation schemes on system throughput, latency, and energy consumption; others maximize the minimum throughput of the device by jointly optimizing link scheduling, power and time allocation, and the drone's hovering position.

[0005] However, the aforementioned optimization studies of UAV IoT systems are all based on the idealized assumption of a single fading channel, which fails to reflect the complexity of real-world communication environments. Given the limited battery capacity of UAVs, energy consumption optimization becomes a key research area. For example, using system energy efficiency as the optimization objective, maximizing energy efficiency indicators can achieve synergistic optimization of system throughput and UAV energy consumption. However, when a UAV is constrained by its service range and cannot reach the target node within a single time slot, its data collection volume degrades to zero, making it difficult for a fractional objective function to simultaneously optimize both. Therefore, for scenarios with service range constraints, a multi-objective optimization model is constructed by introducing a tradeoff coefficient to consider both throughput and energy consumption indicators. It is known that UAV flight energy consumption is related to its speed, while for a given charging power, UAV charging energy consumption is related to its hovering position, because hovering position affects channel gain, which in turn affects the charging rate. Furthermore, for UAV-assisted wireless power supply networks with multiple IoT devices, the issue of service fairness among devices also needs to be considered.

[0006] Extensive research has been conducted on optimizing performance metrics such as throughput, energy efficiency, and fairness in single-round data collection for UAV-assisted wireless power networks. However, insufficient attention has been paid to optimizing performance metrics in multi-round tasks. While significant progress has been made in UAV trajectory design and energy transmission efficiency, the historical dependence of device services in multi-round tasks has not been considered, making it difficult to guarantee fairness. Alternatively, although fairness metrics have been introduced, their optimization framework is limited to the fairness of device services in a single round and does not consider the constraints of UAV service distance. Therefore, it is necessary to comprehensively address the issues of random distribution of IoT devices based on demand, limited UAV energy and service cycles, the inability of UAVs to access all devices in a single data collection round, and unfair device services in multiple rounds in practical applications. This requires maximizing data collection volume and minimizing UAV energy consumption while considering long-term device service fairness. Summary of the Invention

[0007] To address the challenges posed by the random distribution of demand among IoT devices, the limited energy and service life of drones, the inability of drones to access all devices in a single data collection round, and the unfairness of device service across multiple rounds, this invention provides an optimized method for drones to collect data in wirelessly powered IoT networks.

[0008] The technical solution adopted by this invention to solve the above-mentioned technical problems is: an optimization method for data collection by a drone in a wirelessly powered Internet of Things (IoT) network. The IoT network includes K devices. A single drone flies N rounds at a constant altitude H. During each round, it interacts with fewer than K devices. Each interaction involves the drone hovering and charging the device, followed by the device transmitting data back to the drone. After collecting data from the device, the drone continues flying to the hovering position of the next device. This method is used to obtain the optimal device interaction sequence for the drone to fly to multiple devices in each round, as well as the corresponding optimal flight speed and optimal flight trajectory. The method includes the following steps: I. Establishing the optimization problem by constructing a system model S101, Channel Model A probabilistic path loss modeling method is adopted to establish an air-to-ground channel model between the UAV and the equipment. A coordinate system containing K devices is established on the same horizontal plane, and the horizontal position of the UAV at any time t is defined as: The position of device k is: ; The distance between the drone and device k is: (1); The probability that the drone and device k establish a Loss of Speed ​​(LoS) communication link at time t is: (2); In equation (2), η1 and η2 are the propagation environment type coefficients; The probability that the drone and device k establish an NLoS communication link at time t is: (3); The line-of-sight channel gain is: (4); In equation (4), Let δ be the channel power gain at the reference distance, which is 1m, and δ be the path loss exponent, where 2 < δ < 2.5; The non-line-of-sight channel gain is: (5); In equation (5), σ is the additional attenuation coefficient for non-line-of-sight links; Downlink channel gain between the drone and device k and uplink channel gain Taking approximately equal values, the channel gain between the UAV and device k is expressed as: (6); S102, Data Transmission Model The data transmission distance constraint between the drone and device k is: (7); In equation (7), R max This represents the maximum data transmission coverage radius of the drone. Let k be a binary variable representing whether the drone selects device k as the target device. When device k is the target device of the drone, =1, otherwise =0; The data upload rate between device k and the drone is: (8); In equation (8), B is the wireless communication bandwidth, and P d τ represents the data transmission power of the device. 2 Noise power; S103, Data Collection Volume and Energy Consumption Model When the flight speed is v, the flight energy consumption of the drone is: (9); In equation (9), P0 is the airfoil power in hovering state, P1 is the induced power in hovering state, and U tip denoted as the tip velocity of the UAV blade, v0 as the average induced velocity of the rotor in hovering state, d0 as the fuselage drag ratio, ρ as the air density, s0 as the rotor solidity, and A as the rotor disk surface area. When device k uploads data to the drone, the data volume D is reached. k The required time is: (10); Drones with power P u When charging device k, the power that reaches device k after attenuation is: (11); The rate at which device k harvests energy through the RF-EH circuit is: (12); In equation (12), P max For the maximum DC output power, ω1 and ω2 are the characteristic constants of the EH system circuit; The time required for the drone to charge device k is: (13); The total amount of data collected by the drone in one round is: (14); The total charging energy consumption of the drone in one round is: (15); The total hovering energy consumption of the drone in one round is: (16); The total flight energy consumption of the drone in one round is: (17); The total energy consumption of the drone in one round is: (18); S104, Fairness Model An integrated recorder is used to describe the number of interactions of device k and the total number of interactions of the drone in a round, and the fairness index of device k is obtained: (19); In equation (19), Let k be the number of interactions. ; S105. Establish an optimization problem. By improving fairness, increasing data collection volume, and reducing energy consumption, we establish an optimization problem for device interaction sequences, drone flight speed, and flight trajectory: (20); In equation (20), C1 constrains the flight speed v of the UAV; C2 constrains the drone's flight angle θ; C3 restricts the drone to selecting only one device at any given time; C4 represents the data transmission distance constraint between the drone and the equipment; C5 constrains the drone's energy consumption to not exceed the maximum battery capacity E. max ; C6 constrains the total time consumed by the drone in each round to not exceed the round duration T; II. The optimization model is decomposed into a main module and sub-modules. The main module is used to obtain the optimal device interaction sequence. (twenty one); The submodule is used to obtain the optimal flight speed and optimal flight trajectory: (twenty two); The multi-objective master-slave collaborative optimization algorithms used by the main module and sub-modules include: S201. Optimize and train the Actor-Critic network model of the submodule. For each time t, the Actor network Based on the current state Generate corresponding actions Critic Network Evaluate the value of state-action pairs and apply them to the reward function. Receive the corresponding reward, and then check the state at the next moment. Forming a quadruple And store the quadruple into the experience replay buffer; state Including the drone's location, target device location, remaining energy, and boundary constraint indicator variables, represented as: (twenty three); In equation (23), The relative distance between the drone and the target device. For the remaining energy of the drone, This is a boundary indicator variable used to indicate whether the drone has deviated from the operating area; action include and flight angle , represented as: (twenty four); reward function for: (25); In equation (25), To maximize rewards for data collection, For drones, the reward is to minimize energy consumption. The flight range assistance rewards are represented as follows: (26); Multiple quadruplets are randomly selected from the experience replay buffer as inputs to the Actor loss function and the Critic loss function. The Actor network is updated by maximizing the Actor loss function, and the Critic network is updated by minimizing the Critic loss function. The Actor loss function is: (27); The Critic loss function is: (28); Where y t For network target values: (29); In equation (29), γ is the discount factor. Represents the target Actor network. Represents the target Critic network; The Actor-Critic network model is continuously updated using a soft update strategy until it has completed training. (30); In equation (30), ε is the soft update coefficient; S202. Optimize the initial rounds using the main module. First, initialize the location sets of K devices. Data volume set Historical interaction count set ; Set the number of devices selected in the initial round, w, and randomly generate multiple populations. Each population represents a device interaction sequence formed by arranging w devices in sequence. A dynamic priority adjustment mechanism is used to reduce the priority of high-frequency interactive devices, and offspring populations are generated through crossover and mutation. The dynamic priority adjustment mechanism is as follows: (31); In equation (31), This is the device priority weighting coefficient. To prevent division by zero of small constants; Calculate the total data volume C, fairness index J, and path length L for each path corresponding to each offspring population. Calculate the fitness of each offspring population using a multi-objective fitness function. Then, perform non-dominated sorting on the fitness of all offspring populations and generate a Pareto front solution set. For each solution, perform a linear weighted summation and then select the maximum value: (32); In equation (32), For data volume weight vector, For the fairness weight vector, This is the path length weight vector; The offspring population corresponding to the maximum value is used as the device interaction sequence for the initial round. The device interaction sequence is input into the submodule, and the flight speed and flight trajectory corresponding to the device interaction sequence are obtained through the trained Actor-Critic network model. Then, it is verified whether the constraints C1-C6 of the optimization problem are satisfied. S203. When all constraints are satisfied, change the number of devices selected to w+1, repeat S202, and verify again whether the newly obtained device interaction sequence, flight speed and flight trajectory satisfy the constraints. Repeat until they are not satisfied. Keep the number of devices selected when the last constraint is satisfied as the optimal number of selections w0, and use the device interaction sequence, flight speed and flight trajectory when the last constraint is satisfied as the optimal device interaction sequence, optimal flight speed and optimal flight trajectory for the initial round. S204. In each subsequent round, first, the set of historical interaction counts is... Update: (33); In equation (33), Let k be the number of historical interactions of device k in round n-1. Let be a binary variable representing whether the drone selects device k in the nth round. When device k is the target device of the drone in the nth round, =1, otherwise =0; Then, using the optimal selection quantity w0, S202 and S203 are performed to obtain the optimal device interaction sequence, optimal flight speed, and optimal flight trajectory for all rounds.

[0009] Preferably, in S201, the experience playback buffer is initialized before storing the quadruple into the experience playback buffer.

[0010] Preferably, in S202, the number of devices selected in the initial round is set to w=K / 2.

[0011] According to the above technical solution, the beneficial effects of the present invention are: This invention addresses the issue of unfair device services in multi-round data collection processes within a wirelessly powered IoT network composed of multiple IoT devices and a single drone. Considering the differences in data buffer capacity among various device types, and taking into account the service range and battery energy limitations of the drone, a nonlinear energy harvesting model is adopted. The optimization objective is modeled as a mixed-integer nonlinear programming problem, which is then decomposed into two modules: device service sequence generation (main module) and drone trajectory optimization (sub-module). Through dynamic priority adjustment and sub-module calling mechanisms, the collaborative optimization and long-term equilibrium of device service fairness, data collection volume, and drone energy consumption are achieved.

[0012] The main module dynamically adjusts the device service priority based on the device data volume and historical service frequency. At the same time, given the conflicting optimization objectives, a multi-objective evolutionary strategy is adopted to generate Pareto front solutions, and the optimal device access sequence for each round is selected from these solutions to improve the long-term service fairness of the system.

[0013] In this submodule, to address the problem of adjusting continuous motion parameters for UAVs, a lightweight state space is constructed to reduce the input dimensionality of the neural network. A multi-objective reward function, incorporating both data collection volume and energy consumption, is designed to guide the UAV in efficiently learning trajectory strategies, thereby achieving synergistic optimization of data collection volume and energy consumption. Furthermore, an adaptive Gaussian noise exploration mechanism is introduced to effectively enhance strategy diversity. Attached Figure Description

[0014] Figure 1 This is a system model diagram of the present invention; Figure 2 This is the algorithm framework diagram for the submodule; Figure 3 A graph showing the change in the fairness index across multiple data collection rounds; Figure 4 A graph showing the change in the fairness index across multiple data collection rounds; Figure 5 Comparison results under different strategies: (a) total data volume, (b) total energy consumption; Figure 6 Comparison results under different strategies: (a) energy efficiency, (b) average energy consumption; Figure 7 Comparison results under different strategies: (a) total flight energy consumption, (b) total hovering energy consumption; Figure 8 Comparison results under different strategies: (a) hovering time, (b) flight time. Detailed Implementation

[0015] An optimized method for data collection by drones in wirelessly powered Internet of Things (IoT) networks, as described above... Figure 1 As shown, the wireless power supply IoT network includes K devices. A single drone flies N rounds at a constant altitude H. During each round of flight, it selects and interacts with multiple devices, fewer than K in number. The interaction process for each device involves the drone hovering and charging the device, and then the device transmitting data back to the drone. After the drone completes the data collection from the device, it continues to fly to the hovering position of the next device.

[0016] This method is used to obtain the optimal device interaction sequence in each round where the drone flies sequentially to multiple devices, as well as the corresponding optimal flight speed and optimal flight trajectory, and includes the following steps: I. Establishing the optimization problem by constructing a system model S101, Channel Model A probabilistic path loss modeling method is adopted to establish an air-to-ground channel model between the UAV and the equipment. A coordinate system containing K devices is established on the same horizontal plane, and the horizontal position of the UAV at any time t is defined as: The position of device k is: ; The distance between the drone and device k is: (1).

[0017] The probability that the drone and device k establish a Loss of Speed ​​(LoS) communication link at time t is: (2); In equation (2), η1 and η2 are the propagation environment type coefficients.

[0018] The probability that the drone and device k establish an NLoS communication link at time t is: (3).

[0019] The line-of-sight channel gain is: (4); In equation (4), Let δ be the channel power gain at the reference distance, which is 1m, and let δ be the path loss exponent, where 2 < δ < 2.5.

[0020] The non-line-of-sight channel gain is: (5); In equation (5), σ is the additional attenuation coefficient of the non-line-of-sight link.

[0021] Downlink channel gain between the drone and device k and uplink channel gain Taking approximately equal values, the channel gain between the UAV and device k is expressed as: (6).

[0022] S102, Data Transmission Model The data transmission distance constraint between the drone and device k is: (7); In equation (7), R max This represents the maximum data transmission coverage radius of the drone. Let k be a binary variable representing whether the drone selects device k as the target device. When device k is the target device of the drone, =1, otherwise =0.

[0023] The data upload rate between device k and the drone is: (8); In equation (8), B is the wireless communication bandwidth, and P d τ represents the data transmission power of the device. 2 This represents noise power.

[0024] S103, Data Collection Volume and Energy Consumption Model When the flight speed is v, the flight energy consumption of the drone is: (9); In equation (9), P0 is the airfoil power in hovering state, P1 is the induced power in hovering state, and U tip denoted as the tip velocity of the UAV blade, v0 as the average induced velocity of the rotor in hovering state, d0 as the fuselage drag ratio, ρ as the air density, s0 as the rotor solidity, and A as the rotor disk surface area.

[0025] When device k uploads data to the drone, the data volume D is reached. k The required time is: (10).

[0026] Drones with power P u When charging device k, the power that reaches device k after attenuation is: (11).

[0027] The rate at which device k harvests energy through the RF-EH circuit is: (12); In equation (12), P max ω1 and ω2 are the characteristic constants of the EH system circuit, where ω1 represents the maximum DC output power.

[0028] The time required for the drone to charge device k is: (13). The total amount of data collected by the drone in one round is: (14). The total charging energy consumption of the drone in one round is: (15). The total hovering energy consumption of the drone in one round is: (16). The total flight energy consumption of the drone in one round is: (17). The total energy consumption of the drone in one round is: (18).

[0029] S104, Fairness Model An integrated recorder is used to describe the number of interactions of device k and the total number of interactions of the drone in a round, and the fairness index of device k is obtained: (19); In equation (19), Let k be the number of interactions. .

[0030] S105. Establish an optimization problem. By improving fairness, increasing data collection volume, and reducing energy consumption, we establish an optimization problem for device interaction sequences, drone flight speed, and flight trajectory: (20).

[0031] In equation (20), C1 constrains the flight speed v of the UAV; C2 constrains the drone's flight angle θ; C3 restricts the drone to selecting only one device at any given time; C4 represents the data transmission distance constraint between the drone and the equipment; C5 constrains the drone's energy consumption to not exceed the maximum battery capacity E. max ; C6 constrains the total time consumed by the drone in each round to not exceed the round duration T.

[0032] II. The optimization model is decomposed into a main module and sub-modules. The main module is used to obtain the optimal device interaction sequence. (twenty one).

[0033] The submodule is used to obtain the optimal flight speed and optimal flight trajectory: (twenty two).

[0034] The multi-objective master-slave collaborative optimization algorithms used by the main module and sub-modules include: S201. Optimize and train the Actor-Critic network model of the submodule. like Figure 2 As shown, for each time t, the Actor network Based on the current state Generate corresponding actions Critic Network Evaluate the value of state-action pairs and apply them to the reward function. Receive the corresponding reward, and then check the state at the next moment. Forming a quadruple The quadruple is then stored in the experience replay buffer, which is initialized before storage begins.

[0035] state Including the drone's location, target device location, remaining energy, and boundary constraint indicator variables, represented as: (twenty three); In equation (23), The relative distance between the drone and the target device. For the remaining energy of the drone, This is a boundary indicator variable used to indicate whether the drone has deviated from the operating area.

[0036] action include and flight angle , represented as: (twenty four).

[0037] reward function for: (25); In equation (25), To maximize rewards for data collection, For drones, the reward is to minimize energy consumption. The flight range assistance rewards are represented as follows: (26).

[0038] Multiple quadruplets are randomly selected from the experience replay buffer as inputs to the Actor loss function and the Critic loss function. The Actor network is updated by maximizing the Actor loss function, and the Critic network is updated by minimizing the Critic loss function.

[0039] The Actor loss function is: (27).

[0040] The Critic loss function is: (28).

[0041] Where y t For network target values: (29); In equation (29), γ is the discount factor. Represents the target Actor network. The target network is the Critic network.

[0042] The Actor-Critic network model is continuously updated using a soft update strategy until it has completed training. (30); In equation (30), ε is the soft update coefficient.

[0043] S202. Optimize the initial rounds using the main module. First, initialize the location sets of K devices. Data volume set Historical interaction count set .

[0044] Set the number of devices to be selected in the initial round to w = K / 2 and randomly generate multiple populations. Each population represents a device interaction sequence formed by arranging w devices in sequence.

[0045] A dynamic priority adjustment mechanism is used to reduce the priority of high-frequency interactive devices, and offspring populations are generated through crossover and mutation. The dynamic priority adjustment mechanism is as follows: (31); In equation (31), This is the device priority weighting coefficient. To prevent division by zero of small constants.

[0046] Calculate the total data volume C, fairness index J, and path length L for each path corresponding to each offspring population. Calculate the fitness of each offspring population using a multi-objective fitness function. Then, perform non-dominated sorting on the fitness of all offspring populations and generate a Pareto front solution set.

[0047] For each solution, perform a linear weighted summation and then select the maximum value: (32); In equation (32), For data volume weight vector, For the fairness weight vector, This is the path length weight vector.

[0048] The offspring population corresponding to the maximum value is used as the device interaction sequence for the initial round. This device interaction sequence is input into the submodule, and the flight speed and flight trajectory corresponding to the device interaction sequence are obtained through the trained Actor-Critic network model. Then, it is verified whether the constraints C1-C6 of the optimization problem are satisfied.

[0049] S203. When all constraints are satisfied, change the number of devices selected to w+1, repeat S202, and verify again whether the newly obtained device interaction sequence, flight speed, and flight trajectory satisfy the constraints. Repeat until they are not satisfied. Then, retain the number of devices selected when the constraints were last satisfied, i.e., the number when S202 was last performed, as the optimal number of selections w0. Use the device interaction sequence, flight speed, and flight trajectory when the constraints were last satisfied as the optimal device interaction sequence, optimal flight speed, and optimal flight trajectory for the initial round.

[0050] S204. In each subsequent round, first, the set of historical interaction counts is... Update: (33); In equation (33), Let k be the number of historical interactions of device k in round n-1. Let be a binary variable representing whether the drone selects device k in the nth round. When device k is the target device of the drone in the nth round, =1, otherwise =0.

[0051] Then, using the optimal selection quantity w0, S202 and S203 are performed to obtain the optimal device interaction sequence, optimal flight speed, and optimal flight trajectory for all rounds.

[0052] Experimental setup (1) Parameter settings: 100 IoT devices are randomly distributed within a 1000m×1000m square area. The number of data collection rounds is set to 100. At the beginning of each data collection round, the UAV starts its task from a designated location and returns to that location before the end of the task cycle. When the distance constraint between the target device and the UAV is met, the UAV hovers to provide services to the device. The charging power of the UAV is set to 15W, and the transmission power of the IoT devices is set to 0.01mW. Three types of devices with different data buffer capacities are selected, with data buffer capacities of {1.2, 1, 0.8} MB, and the corresponding number of devices is {30, 40, 30}. The battery capacity of the UAV is set to 100000J. The UAV must return to the initial point while meeting the energy consumption constraint to ensure the safety of the UAV. The weight vector of the multi-objective dynamic balance is set to 0.3, 0.5, 0.2, and the weights between the data volume and the number of historical services are set to 0.4 and 0.6. The specific system parameters are shown in Table 1.

[0053] (2) Network structure: The dimensions of the input layer, hidden layer and output layer of the Actor network are respectively [s dim ], [400, 300], [a dim The dimensions of the input layer, hidden layer, and output layer of the Critic network are [s]. dim +a dim [400, 300, 1]. The learning rates for both the Actor network and the Critic network are set to [400, 300, 1]. The experience buffer size is 8000, the batch size is 64, the reward decay factor is 0.9, and the soft target update parameters are... The hidden layer is fully connected and uses the ReLU function as the activation function. In the output layer, the velocity uses the sigmoid function as its activation function, and the angle uses the tanh function as its activation function to satisfy the constraints.

[0054] Experimental Results Analysis from Figure 3 As can be seen, Jain's fairness value gradually increases with optimization and then stabilizes at a relatively high level. There is a slight, short-term decrease in fairness value during the first 20 rounds. This is because the difference in the number of times each device serves is not significant at this stage, and the algorithm tends to serve devices with larger data volumes and closer proximity, ignoring some devices, thus causing a short-term decrease in fairness value. However, since the device priority in the algorithm is dynamically adjusted, it will gradually stabilize at a higher level later on.

[0055] from Figure 4 It can be seen that during the training of the drone's speed control and flight trajectory, the cumulative reward continuously decreases for approximately the first twenty training rounds. This is because the drone lacks experience learning at this stage and is in an autonomous exploration phase. After the experience buffer stores enough data, the drone randomly selects data tuples of batch size to train the network. The first 250 training rounds constitute the drone's main exploration and learning phase, during which both the reward and loss values ​​fluctuate significantly before stabilizing and converging.

[0056] The flight strategy selection for UAVs follows the following optimization criteria: (1) Using the maximum flight speed can effectively shorten the mission time and quickly reach the target equipment to complete the service; (2) Using the maximum endurance speed can significantly reduce energy consumption and achieve energy-saving operation of equipment service. Furthermore, theoretical analysis shows that when the UAV hovers directly above the target equipment, the wireless channel gain reaches its theoretical maximum value. At this position, the system's data transmission rate is at its maximum, the time and energy consumption for the UAV to charge the equipment are at their minimum, and the hovering time and hovering energy consumption required for the service are also at their minimum.

[0057] To verify the algorithm's performance, the strategies generated by the algorithm in the example were compared with the following two benchmark strategies: 1. Maximum Speed ​​Hovering Data Acquisition Strategy (MVHS): The drone flies at maximum speed and hovers directly above the device to collect data. 2. Maximum Endurance Hovering Data Acquisition Strategy (MEHS): The drone flies at maximum endurance speed and hovers directly above the device to collect data. Based on experimental requirements, 30 rounds were randomly selected from the multi-round optimal device service sequence for verification experiments. All experimental data were the average of 100 evaluation results.

[0058] Figure 5 The results in (a) show that the implementation strategy achieves a higher total data collection volume in a single round of data collection, and the implementation strategy outperforms the two comparative strategies in terms of data collection volume. Figure 5(b) In the comparison of total energy consumption of the three strategies, the MVHS strategy has the highest energy consumption due to its use of the maximum allowed speed, while the MEHS strategy flies at the maximum endurance speed, resulting in the lowest flight energy consumption. Both strategies hover directly above the device to provide service, leading to lower charging and hovering energy consumption. Therefore, the MEHS strategy has the lowest total energy consumption. The embodiment employs an adaptive speed control strategy. Although the flight energy consumption is relatively higher than the MEHS strategy in order to serve as many devices as possible, the drone hovers to provide service to the target device when the distance constraint between the target device and the drone is met. This reduces the flight distance to some extent, thereby lowering flight energy consumption. Although hovering and charging energy consumption increase, the total energy consumption is still lower than that of the MVHS strategy.

[0059] Figure 6 (a) It can be seen that, due to the synergistic effect of the high data collection volume and controllable energy consumption of the embodiment strategy, the UAV energy efficiency of the embodiment strategy is significantly better than the other two comparative strategies. Figure 6 The experimental results in (a) verify the effectiveness of the algorithm in energy efficiency optimization. Figure 6 (b) is a graph showing the average energy consumption variation of the three strategies. It can be seen that the average energy consumption of the example strategy falls between the two comparative strategies. Since the MVHS strategy achieves the best channel gain by hovering above the device to collect data, and the number of devices served by the MVHS strategy per round is greater than that of the MEHS strategy, the hovering energy consumption of the MVHS strategy is the lowest, followed by the MEHS strategy. Figure 7 (b) This verifies the above conclusion. Figure 7 (a) shows the comparison of total flight energy consumption for the three strategies. It can be seen that since the speeds of the comparative strategies are the maximum allowable value and the maximum range, their corresponding flight energy consumption also reaches the highest and lowest values, respectively. The speed of the implementation strategy is adjustable. In order to serve as many devices as possible, the flight energy consumption is relatively large, but the overall energy consumption is basically below that of the MVHS strategy. It is superior to the comparative strategies in terms of maximizing data collection, minimizing total system energy consumption, and maximizing system energy efficiency.

[0060] Figure 8 The hovering time and flight time distributions of the three strategies were compared. Figure 6 The experimental results show that the MVHS strategy serves more devices per round than the MEHS strategy. However, since the charging time and data acquisition time on the same device are the same for both strategies, the MEHS strategy has the shortest total hovering time, followed by the MVHS strategy. In contrast, the example strategy optimizes the drone's hovering position, sometimes at the cost of slightly reduced transmission efficiency, in exchange for shorter flight distance and flight time. Therefore, its hovering time is longer than the two comparison strategies, while its flight time is the shortest among all strategies.

[0061] In summary, this embodiment achieves a long-term balance between equipment service fairness, data collection volume, and UAV energy consumption. The main module combines a multi-objective optimization framework with a dynamic feedback mechanism for equipment service fairness, obtaining the optimal equipment service sequence through weighted summation. The sub-modules use a lightweight state space and a multi-objective reward function to optimize the UAV flight strategy and trajectory, achieving a synergistic improvement in UAV energy consumption and data collection volume.

Claims

1. An optimized method for data acquisition by a drone in a wirelessly powered Internet of Things (IoT) network, wherein the IoT network comprises K devices, and a single drone flies N rounds at a constant altitude H. During each round, fewer than K devices are selected for interaction. Each interaction involves the drone hovering and charging the device, followed by the device transmitting data back to the drone. After completing data acquisition from the device, the drone continues flying to the hovering position of the next device. This method is used to obtain the optimal device interaction sequence for the drone sequentially flying to multiple devices in each round, as well as the corresponding optimal flight speed and optimal flight trajectory. The method is characterized by… Includes the following steps: I. Establishing the optimization problem by constructing a system model S101, Channel Model A probabilistic path loss modeling method is adopted to establish an air-to-ground channel model between the UAV and the equipment. A coordinate system containing K devices is established on the same horizontal plane, and the horizontal position of the UAV at any time t is defined as: The position of device k is: ; The distance between the drone and device k is: （1）； The probability that the drone and device k establish a Loss of Speed ​​(LoS) communication link at time t is: （2）； In equation (2), η1 and η2 are the propagation environment type coefficients; The probability that the drone and device k establish an NLoS communication link at time t is: （3）； The line-of-sight channel gain is: （4）； In equation (4), Let δ be the channel power gain at the reference distance, which is 1m, and δ be the path loss exponent, where 2 < δ < 2.5; The non-line-of-sight channel gain is: （5）； In equation (5), σ is the additional attenuation coefficient for non-line-of-sight links; Downlink channel gain between the drone and device k and uplink channel gain Taking approximately equal values, the channel gain between the UAV and device k is expressed as: （6）； S102, Data Transmission Model The data transmission distance constraint between the drone and device k is: （7）； In equation (7), R max This represents the maximum data transmission coverage radius of the drone. Let k be a binary variable representing whether the drone selects device k as the target device. When device k is the target device of the drone, =1, otherwise =0; The data upload rate between device k and the drone is: （8）； In equation (8), B is the wireless communication bandwidth, and P d τ represents the data transmission power of the device. 2 Noise power; S103, Data Collection Volume and Energy Consumption Model When the flight speed is v, the flight energy consumption of the drone is: （9）； In equation (9), P0 is the airfoil power in hovering state, P1 is the induced power in hovering state, and U tip denoted as the tip velocity of the UAV blade, v0 as the average induced velocity of the rotor in hovering state, d0 as the fuselage drag ratio, ρ as the air density, s0 as the rotor solidity, and A as the rotor disk surface area. When device k uploads data to the drone, the data volume D is reached. k The required time is: （10）； Drones with power P u When charging device k, the power that reaches device k after attenuation is: （11）； The rate at which device k harvests energy through the RF-EH circuit is: （12）； In equation (12), P max For the maximum DC output power, ω1 and ω2 are the characteristic constants of the EH system circuit; The time required for the drone to charge device k is: （13）； The total amount of data collected by the drone in one round is: （14）； The total charging energy consumption of the drone in one round is: （15）； The total hovering energy consumption of the drone in one round is: （16）； The total flight energy consumption of the drone in one round is: （17）； The total energy consumption of the drone in one round is: （18）； S104, Fairness Model An integrated recorder is used to describe the number of interactions of device k and the total number of interactions of the drone in a round, and the fairness index of device k is obtained: （19）； In equation (19), Let k be the number of interactions. ; S105. Establish an optimization problem. By improving fairness, increasing data collection volume, and reducing energy consumption, we establish an optimization problem for device interaction sequences, drone flight speed, and flight trajectory: （20）； In equation (20), C1 constrains the flight speed v of the UAV; C2 constrains the drone's flight angle θ; C3 restricts the drone to selecting only one device at any given time; C4 represents the data transmission distance constraint between the drone and the equipment; C5 constrains the drone's energy consumption to not exceed the maximum battery capacity E. max ; C6 constrains the total time consumed by the drone in each round to not exceed the round duration T; II. The optimization model is decomposed into a main module and sub-modules. The main module is used to obtain the optimal device interaction sequence. （21）； The submodule is used to obtain the optimal flight speed and optimal flight trajectory: （22）； The multi-objective master-slave collaborative optimization algorithms used by the main module and sub-modules include: S201. Optimize and train the Actor-Critic network model of the submodule. For each time t, the Actor network Based on the current state Generate corresponding actions Critic Network Evaluate the value of state-action pairs and apply them to the reward function. Receive the corresponding reward, and then check the state at the next moment. Forming a quadruple And store the quadruple into the experience replay buffer; state Including the drone's location, target device location, remaining energy, and boundary constraint indicator variables, represented as: （23）； In equation (23), The relative distance between the drone and the target device. For the remaining energy of the drone, This is a boundary indicator variable used to indicate whether the drone has deviated from the operating area; action include and flight angle , represented as: （24）； reward function for: （25）； In equation (25), To maximize rewards for data collection, For drones, the reward is to minimize energy consumption. The flight range assistance rewards are represented as follows: （26）； Multiple quadruplets are randomly selected from the experience replay buffer as inputs to the Actor loss function and the Critic loss function. The Actor network is updated by maximizing the Actor loss function, and the Critic network is updated by minimizing the Critic loss function. The Actor loss function is: （27）； The Critic loss function is: （28）； Where y t For network target values: （29）； In equation (29), γ is the discount factor. Represents the target Actor network. Represents the target Critic network; The Actor-Critic network model is continuously updated using a soft update strategy until it has completed training. （30）； In equation (30), ε is the soft update coefficient; S202. Optimize the initial rounds using the main module. First, initialize the location sets of K devices. Data volume set Historical interaction count set ; Set the number of devices selected in the initial round, w, and randomly generate multiple populations. Each population represents a device interaction sequence formed by arranging w devices in sequence. A dynamic priority adjustment mechanism is used to reduce the priority of high-frequency interactive devices, and offspring populations are generated through crossover and mutation. The dynamic priority adjustment mechanism is as follows: （31）； In equation (31), This is the device priority weighting coefficient. To prevent division by zero of small constants; Calculate the total data volume C, fairness index J, and path length L for each path corresponding to each offspring population. Calculate the fitness of each offspring population using a multi-objective fitness function. Then, perform non-dominated sorting on the fitness of all offspring populations and generate a Pareto front solution set. For each solution, perform a linear weighted summation and then select the maximum value: （32）； In equation (32), For data volume weight vector, For the fairness weight vector, This is the path length weight vector; The offspring population corresponding to the maximum value is used as the device interaction sequence for the initial round. The device interaction sequence is input into the submodule, and the flight speed and flight trajectory corresponding to the device interaction sequence are obtained through the trained Actor-Critic network model. Then, it is verified whether the constraints C1-C6 of the optimization problem are satisfied. S203. When all constraints are satisfied, change the number of devices selected to w+1, repeat S202, and verify again whether the newly obtained device interaction sequence, flight speed and flight trajectory satisfy the constraints. Repeat until they are not satisfied. Keep the number of devices selected when the last constraint is satisfied as the optimal number of selections w0, and use the device interaction sequence, flight speed and flight trajectory when the last constraint is satisfied as the optimal device interaction sequence, optimal flight speed and optimal flight trajectory for the initial round. S204. In each subsequent round, first, the set of historical interaction counts is... Update: （33）； In equation (33), Let k be the number of historical interactions of device k in round n-1. Let be a binary variable representing whether the drone selects device k in the nth round. When device k is the target device of the drone in the nth round, =1, otherwise =0; Then, using the optimal selection quantity w0, S202 and S203 are performed to obtain the optimal device interaction sequence, optimal flight speed, and optimal flight trajectory for all rounds.

2. The optimized method for data acquisition by UAVs in a wirelessly powered Internet of Things network according to claim 1, characterized in that: In S201, the experience playback buffer is initialized before storing the quadruple into the experience playback buffer.

3. The optimized method for data acquisition by UAVs in a wirelessly powered Internet of Things network according to claim 1, characterized in that: In S202, the number of devices selected in the initial round is set to w=K / 2.

Images