A two-stage optimization method and device for an end-to-end crowd sensing system

Abstract

This application discloses a two-stage optimization method and apparatus for end-to-end swarm intelligence sensing systems. The method includes: first, establishing time-slot models, computational models, communication models, and energy consumption models that represent the complete task execution process between sensing devices, UAVs, and execution devices in the swarm intelligence sensing system. Based on these mathematical models, the association factors between the UAV and ground equipment, communication resource allocation, computational resource allocation, and the UAV's flight trajectory are collaboratively optimized. First, the global search capability of a genetic algorithm is used to provide a high-quality initial exploration region. Second, a deep deterministic policy gradient is used to further train and explore this region to obtain the optimal solution that reduces system communication latency and saves energy consumption, thereby ensuring the high efficiency and low energy consumption of the "sensing-communication-computation-execution" task closed loop and minimizing overall energy consumption.

Description

Technical Field

[0001] This application relates to the field of swarm intelligence sensing technology, and in particular to a two-stage optimization method and apparatus for end-to-end swarm intelligence sensing systems. Background Technology

[0002] With the rapid development of the Internet of Things (IoT), crowdsensing technology has become an important means to achieve large-scale data collection and intelligent decision-making. Against this backdrop, end-to-end crowdsensing systems face numerous challenges, such as a surge in data volume and limited communication resources. However, the processing capabilities and transmission conditions of existing sensing devices are insufficient to meet users' demands for latency and reliability, leading to poor coordination among sensing, communication, computing, and execution stages, thus hindering overall system performance and user experience.

[0003] In traditional communication models, sensing devices typically send the collected data directly to the execution end for processing, solving the data transmission problem in simple scenarios. However, when the environment is complex, the data volume is large, or communication resources are limited, sensing devices are constrained by their own computing power and struggle to complete large-scale or highly complex tasks. At the same time, ground links are easily affected by obstructions, resulting in non-line-of-sight transmission, which leads to increased communication latency, higher energy consumption, and decreased reliability. These factors affect the rapid response link from "sensing" to "execution," failing to adequately meet the closed-loop requirements of low latency and high reliability.

[0004] As auxiliary communication and computing nodes, drones possess air-to-ground line-of-sight links, flexible deployment, and edge computing capabilities, effectively enhancing the performance of the entire "perception-communication-computation-execution" chain. By associating sensing devices and execution terminals with suitable drones, the system can leverage the advantages of aerial nodes to achieve localized data processing and efficient forwarding, connecting the closed loop from data acquisition to decision execution. Therefore, closed-loop optimization of end-to-end swarm intelligence sensing systems based on drone assistance has become a research hotspot in both academia and industry.

[0005] Drone-assisted swarm intelligence sensing technology can improve communication conditions and increase transmission efficiency. In existing technologies, drones are typically deployed in fixed locations, and optimization of drone-assisted swarm intelligence sensing technology usually only considers optimization at a single dimension level, such as optimizing the fixed deployment location of drones or optimizing the use of communication resources in the entire swarm intelligence sensing system.

[0006] The existing technologies for optimizing drone-assisted swarm intelligence sensing technology, as mentioned above, typically only consider optimization at a single dimension level, and no effective solution has yet been proposed. Summary of the Invention

[0007] Embodiments of this disclosure provide a way to at least address the technical problem in the prior art where optimization of UAV-assisted swarm intelligence perception technology typically only considers optimization at a single-dimensional level.

[0008] According to one aspect of the present disclosure, a two-stage optimization method for an end-to-end swarm intelligence sensing system is provided. The swarm intelligence sensing system includes several unmanned aerial vehicles (UAVs), several sensing devices, and several execution devices. The method includes: initializing a population, where each individual in the population represents a solution vector, the solution vector including association state information between the UAVs, sensing devices, and execution devices, flight trajectory information of the UAVs, communication parameters, and computing resource parameters; and determining the total energy consumption corresponding to the individual as the fitness corresponding to the individual based on a pre-constructed time-slot model, communication model, computing model, and energy consumption model, wherein the time-slot model is used to represent one round in the swarm intelligence sensing system. In different stages of the task, the communication model is used to characterize the communication resources used by the sensing devices to upload sensing information to the UAVs and by the UAVs to transmit data to the execution devices. The computation model is used to characterize the computational resources used by the UAVs. The energy consumption model is used to quantify the uplink and downlink communication energy consumption, UAV computational energy consumption, and flight energy consumption during the operation of the swarm intelligence sensing system. Using a genetic algorithm, the optimal individual is searched with the goal of minimizing the fitness value of the individual. The optimal individual is used as the initial state, and the optimal solution vector is determined by the deep deterministic policy gradient algorithm. The optimal solution vector is used by several UAVs, several sensing devices, and several execution devices to perform tasks in the swarm intelligence sensing system.

[0009] According to another aspect of the present disclosure, a storage medium is also provided, the storage medium including a stored program, wherein the methods described above are executed by a processor when the program is running.

[0010] According to another aspect of the embodiments of this disclosure, a two-stage optimization device for an end-to-end swarm intelligence sensing system is also provided. The swarm intelligence sensing system includes several drones, several sensing devices, and several execution devices. The device includes: an initialization module for initializing a population, where each individual in the population represents a solution vector, the solution vector including association state information between the drones, sensing devices, and execution devices, the drones' flight trajectory information, communication parameters, and computing resource parameters; and an energy consumption determination module for determining the total energy consumption corresponding to an individual based on a pre-constructed time-slot model, communication model, computing model, and energy consumption model, as the fitness corresponding to the individual, wherein the time-slot model represents a round of tasks in the swarm intelligence sensing system. At different stages of the task, the communication model is used to characterize the communication resources used by the sensing devices to upload sensing information to the drones and by the drones to transmit data to the execution devices. The computation model is used to characterize the computational resources used by the drones. The energy consumption model is used to quantify the uplink and downlink communication energy consumption, drone computational energy consumption, and flight energy consumption during the operation of the swarm intelligence sensing system. The first optimization module is used to search for the optimal individual by using a genetic algorithm to minimize the fitness value of the individual. The second optimization module is used to determine the optimal solution vector by taking the optimal individual as the initial state and using a deep deterministic policy gradient algorithm. The optimal solution vector is used by several drones, several sensing devices, and several execution devices to perform tasks in the swarm intelligence sensing system.

[0011] According to another aspect of the present disclosure, a two-stage optimization apparatus for an end-to-end swarm intelligence sensing system is also provided. The swarm intelligence sensing system includes several unmanned aerial vehicles (UAVs), several sensing devices, and several execution devices. The apparatus includes a processor and a memory connected to the processor, used to provide instructions to the processor to perform the following processing steps: initializing a population, where each individual in the population represents a solution vector, the solution vector including association state information between the UAVs, sensing devices, and execution devices, UAV flight trajectory information, communication parameters, and computational resource parameters; and determining the total energy consumption corresponding to an individual based on a pre-constructed time-slot model, communication model, computation model, and energy consumption model, as the fitness corresponding to that individual. The system employs a time-slot model to represent different stages in a single task within the swarm intelligence sensing system, a communication model to characterize the communication resources used by sensing devices to upload sensing information to drones and by drones to transmit data to execution devices, a computation model to characterize the computational resources used by drones, and an energy consumption model to quantify the uplink and downlink communication energy consumption, drone computational energy consumption, and flight energy consumption during the operation of the swarm intelligence sensing system. A genetic algorithm is used to search for the optimal individual with the objective of minimizing its fitness value. Using the optimal individual as the initial state, a deep deterministic policy gradient algorithm is employed to determine the optimal solution vector. This optimal solution vector is then used by several drones, sensing devices, and execution devices to perform tasks within the swarm intelligence sensing system.

[0012] In this embodiment, according to the technical solution, a time-slot model, a computational model, a communication model, and an energy consumption model are established to represent the complete task execution process between sensing devices, drones, and execution devices in a swarm intelligence sensing system. Based on these mathematical models, the correlation factors, communication resource allocation, computational resource allocation, and drone flight trajectory between the drone and ground equipment are collaboratively optimized. First, the global search capability of a genetic algorithm is used to provide a high-quality initial exploration area. Second, a deep deterministic policy gradient is used to further train and explore this area to obtain the optimal solution that reduces system communication latency and saves energy consumption, thereby ensuring the high efficiency and low energy consumption of the "sensing-communication-computation-execution" task closed loop and minimizing overall energy consumption. Attached Figure Description

[0013] The accompanying drawings, which are included to provide a further understanding of this disclosure and form part of this application, illustrate exemplary embodiments of this disclosure and are used to explain this disclosure, but do not constitute an undue limitation of this disclosure. In the drawings: Figure 1 This is a hardware structure block diagram of a computing device for implementing a two-stage optimization method for an end-to-end swarm intelligence sensing system according to Embodiment 1 of this disclosure; Figure 2 This is a schematic diagram for implementing the end-to-end crowd sensing system according to Embodiment 1 of this disclosure; Figure 3 This is a flowchart illustrating a two-stage optimization method for an end-to-end swarm intelligence sensing system according to the first aspect of Embodiment 1 of this disclosure. Figure 4 This is a schematic diagram of a time slot model according to the first aspect of Embodiment 1 of this disclosure; Figure 5 This is a flowchart illustrating a two-stage optimization process according to the first aspect of Embodiment 1 of this disclosure; Figure 6 This is a schematic diagram of a two-stage optimization device for an end-to-end crowd sensing system according to Embodiment 2 of this disclosure; and Figure 7 This is a schematic diagram of a two-stage optimization device for an end-to-end crowd sensing system according to Embodiment 3 of this disclosure. Detailed Implementation

[0014] To enable those skilled in the art to better understand the technical solutions of this disclosure, the technical solutions of the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this disclosure, and not all embodiments. Based on the embodiments of this disclosure, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this disclosure.

[0015] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this disclosure are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this disclosure described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0016] Example 1

[0017] According to this embodiment, a method embodiment of a two-stage optimization method for an end-to-end crowd sensing system is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0018] The method embodiments provided in this example can be executed on a computer terminal, server, or similar computing device. Figure 1 A hardware block diagram of a computing device for a two-stage optimization method for end-to-end crowd sensing systems is shown. Figure 1 As shown, a computing device may include one or more processors (processors may include, but are not limited to, microprocessors such as MCUs or programmable logic devices such as FPGAs), memory for storing data, transmission devices for communication functions, and input / output interfaces. The memory, transmission devices, and input / output interfaces are connected to the processor via a bus. In addition, it may also include a display, keyboard, and cursor control device connected to the input / output interfaces. Those skilled in the art will understand that... Figure 1 The structure shown is for illustrative purposes only and does not limit the structure of the aforementioned electronic device. For example, a computing device may also include... Figure 1The more or fewer components shown, or having the same Figure 1 The different configurations shown.

[0019] Figure 2 This is a schematic diagram of the end-to-end crowd sensing system according to this embodiment.

[0020] Reference Figure 2 As shown, the drone receives sensing information transmitted by sensing devices. Then, the drone flies and processes the received sensing information. When the drone reaches a location convenient for transmitting the data to the execution device, it transmits the processed data to the execution device on the ground. This provides an example of a business scenario using a swarm intelligence sensing system; however, in practical applications, the specific business scenario of this swarm intelligence sensing system is not limited. For example, in an earthquake relief scenario, there are personnel who cannot easily access the disaster area. Therefore, these personnel can send relevant sensing information to a drone using their own sensing devices. The drone flies to the disaster area and performs the necessary calculations on the received sensing data. Then, the drone transmits the processed data to the execution device (such as a rescue vehicle) on the ground in the disaster area.

[0021] Therefore, in the swarm intelligence sensing system provided in this embodiment, the UAV can first fly over the sensing device to receive the sensing information sent by the sensing device, and then fly over the execution device to transmit the data that needs to be sent to the execution device. For this swarm intelligence sensing system, the flight trajectory of the UAV and the use of communication and computing resources can be optimized to meet the requirements of task closure and efficient task completion. Therefore, the optimization method provided in this description is a method for jointly optimizing the flight trajectory of the UAV and the communication and computing resources used by the UAV. The specific optimization method will be explained later.

[0022] It should be noted that the aforementioned one or more processors and / or other data processing circuits are generally referred to herein as "data processing circuits". These data processing circuits may be embodied, in whole or in part, in software, hardware, firmware, or any other combination thereof. Furthermore, the data processing circuits may be a single, independent processing module, or may be integrated, in whole or in part, into any other element in a computing device. As involved in the embodiments of this disclosure, the data processing circuits serve as processor control (e.g., selection of a variable resistor termination path connected to an interface).

[0023] The memory can be used to store software programs and modules of application software, such as the program instruction / data storage device corresponding to the two-stage optimization method for end-to-end crowd sensing systems in the embodiments of this disclosure. The processor executes various functional applications and data processing by running the software programs and modules stored in the memory, thereby realizing the two-stage optimization method for end-to-end crowd sensing systems described above. The memory may include high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory may further include memory remotely located relative to the processor, and these remote memories can be connected to the computing device via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0024] The transmission device is used to receive or send data via a network. Specific examples of the network described above may include a wireless network provided by the computing device's communication provider. In one example, the transmission device includes a Network Interface Controller (NIC), which can connect to other network devices via a base station to communicate with the Internet. In another example, the transmission device may be a Radio Frequency (RF) module used for wireless communication with the Internet.

[0025] The display can be, for example, a touchscreen liquid crystal display (LCD), which allows users to interact with the user interface of the computing device.

[0026] It should be noted here that, in some optional embodiments, the above... Figure 1 The computing device shown may include hardware elements (including circuitry), software elements (including computer code stored on a computer-readable medium), or a combination of both hardware and software elements. It should be noted that... Figure 1 This is only one instance of a specific particular instance, and is intended to illustrate the types of components that may exist in the aforementioned computing devices.

[0027] Under the above operating environment, according to the first aspect of this embodiment, a two-stage optimization method for end-to-end crowd sensing systems is provided. Figure 3 A flowchart illustrating the method is shown below. (Refer to...) Figure 3 As shown, the method includes: S302: Initialize the population. Each individual in the population is used to represent a solution vector. The solution vector includes the association state information between the UAV, the sensing device and the execution device, the UAV trajectory, communication parameters and computing resource parameters. S304: Based on the pre-built time slot model, communication model, computing model and energy consumption model, determine the total energy consumption corresponding to an individual, as the fitness corresponding to the individual; S306: Using a genetic algorithm, the optimal individual is searched by minimizing the fitness value of each individual. S308: Using the optimal individual as the initial state, the optimal solution is determined through a deep deterministic policy gradient algorithm. The optimal solution is used by several UAVs, several sensing devices, and several execution devices to perform tasks in the swarm intelligence sensing system.

[0028] It should be noted that the executing entity of this method can be, for example, a server. For instance, the server can solve for the optimal solution, which includes the associated state information between the drone, the sensing device, and the executing device, the drone's trajectory, communication parameters, and computing resource parameters. The server then transmits this solution to the drone, the sensing device, and the executing device, enabling them to perform the corresponding tasks. Of course, the executing entity can also be, for example, a computing device such as a processor deployed within the drone, the sensing device, or the executing device; this specification does not limit the executing entity of this method.

[0029] The core principle of this method is based on an end-to-end swarm intelligence sensing system scenario, constructing a system communication, computing, and energy consumption model (i.e., time-slot model, communication model, computing model, and energy consumption model) covering the entire process of "sensing-communication-computation-execution". By jointly optimizing the UAV flight path and multi-dimensional resource allocation (bandwidth, power, etc.), the optimization problem is represented as a mixed-integer nonlinear programming problem, which is a non-convex nonlinear combinatorial problem. Therefore, a two-stage hybrid optimization strategy is adopted: first, a global search is performed to obtain the wide-area solution space, and then local refinement is performed for further optimization, thereby finding the minimum overall energy consumption of the system and ensuring the efficient completion of the mission loop.

[0030] Specifically, before the two-stage optimization process, a time-slot model, a communication model, a computational model, and an energy consumption model can be constructed. These models represent the total energy consumption of the complete task execution process among the sensing devices, drones, and execution devices in the swarm intelligence sensing system. In the subsequent two-stage optimization process, the total energy consumption and relevant constraints can be calculated using the aforementioned four models. Then, the optimization can aim to minimize the total energy consumption and satisfy the corresponding constraints, thereby determining the required parameters for the sensing devices, drones, and execution devices in the swarm intelligence sensing system.

[0031] First, the server can initialize a population, where each individual represents a solution vector (S302). The solution vector includes the association state information between the UAV, sensing devices, and execution devices, the UAV's flight trajectory information, communication parameters, and computational resource parameters.

[0032] Among them, the associated status information can represent the pairing relationship between the UAV and the sensing device in the mission, as well as the pairing relationship between the UAV and the execution device. The communication parameters represent the communication resources allocated by the system between the UAV and the sensing device for transmitting sensing information, and the communication resources allocated by the system between the UAV and the execution device for transmitting data. The computing resource parameters represent the computing resources used by the UAV for data processing.

[0033] The server can determine the total energy consumption corresponding to an individual based on the pre-built time slot model, communication model, computing model and energy consumption model, and use it as the fitness corresponding to the individual (S304).

[0034] The formulas and related constraints for the time slot model, communication model, computation model, and energy consumption model will be explained in detail later. Here, we will only give a brief introduction.

[0035] In this context, the time-slot model is used to represent different stages in a task within a swarm intelligence sensing system. (Reference) Figure 4 As shown, a round of tasks in a swarm intelligence sensing system includes an uplink data transmission phase where the sensing device sends sensing information to the drone, a computation phase where the drone processes the received data, and a downlink data transmission phase where the drone sends the processed data to the execution device. Therefore, a time-slot model can be used to represent the time periods corresponding to the uplink data transmission phase, the computation phase, and the downlink data transmission phase.

[0036] The communication model characterizes the communication resources used by sensing devices to upload sensing information to UAVs and by UAVs to transmit data to execution devices. The computational model characterizes the computational resources used by UAVs. The energy consumption model quantifies the uplink and downlink communication energy consumption, UAV computational energy consumption, and flight energy consumption during the operation of the swarm intelligence sensing system. In the energy consumption model, by combining the time-slot model, communication model, and computational model, the total energy consumption of the swarm intelligence sensing system can be determined.

[0037] The two-stage optimization method provided in this manual determines the total energy consumption in both stages using the above model, thereby determining the required parameters in the swarm intelligence sensing system with the goal of minimizing the total energy consumption.

[0038] Then, the server can use a genetic algorithm to minimize the fitness value of an individual and search for the optimal individual (S306).

[0039] The server determines the optimal solution vector based on the optimal individual using a deep deterministic policy gradient algorithm. The optimal solution vector is used by several UAVs, several sensing devices, and several execution devices to perform tasks in the swarm intelligence sensing system (S308).

[0040] The optimal individual mentioned above refers to the individual with the lowest fitness value determined by the genetic algorithm. In other words, the genetic algorithm identifies the individual that minimizes energy consumption as the optimal individual. This optimal individual serves as the initial locally optimal solution found in the search. Then, this optimal individual is used as the initial state in the deep deterministic policy gradient algorithm (DPRA), which continues the search to determine the optimal solution vector. In other words, the DPRA performs a local search within the local solution space corresponding to the optimal individual to find a further optimal solution vector, namely the aforementioned optimal solution vector, which minimizes the total energy consumption of the swarm intelligence sensing system.

[0041] As described in the background section, drone-assisted swarm sensing technology can improve communication conditions and increase transmission efficiency. In existing technologies, drones are typically deployed in fixed locations, and optimizations of drone-assisted swarm sensing technology usually only consider optimizations at a single dimension, such as optimizing the fixed deployment location of the drones or optimizing the use of communication resources in the entire swarm sensing system.

[0042] In view of this, this method establishes time-slot models, computational models, communication models, and energy consumption models representing the complete task execution process between sensing devices, UAVs, and execution devices in a swarm intelligence sensing system. Based on these mathematical models, the correlation factors between UAVs and ground equipment, communication resource allocation, computational resource allocation, and UAV flight trajectory are collaboratively optimized. First, the global search capability of a genetic algorithm is used to provide a high-quality initial exploration region. Second, a deep deterministic policy gradient is used to further train and explore this region to obtain the optimal solution that reduces system communication latency and saves energy consumption, thereby ensuring the high efficiency and low energy consumption of the "sensing-communication-computation-execution" task closed loop and minimizing overall energy consumption.

[0043] The time slot model, communication model, computation model, and energy consumption model will be explained one by one below, and the relevant constraints (pre-defined constraints mentioned later) will also be explained.

[0044] Optionally, one round of tasks for the swarm intelligence sensing system in the time-slot model includes: uplink data transmission period. Calculation period and downlink data transmission period The service time T of one round of tasks in the crowd-sensing system is divided into N time segments. , , The duration of a time segment, the uplink data transmission period. Calculation period Downlink data transmission period ,in, This indicates the start time of the uplink transmission from sensing device k. This represents the time it takes for the drone corresponding to sensor device k to complete data processing. This indicates the uplink transmission end time of sensor device k.

[0045] For details, please refer to Figure 4 As shown, a task round can be divided into an uplink data transmission phase, a computation phase, and a downlink data transmission phase. In the time-slot model, a task round is divided into N time segments (i.e., N time slots), and the start time of the uplink data transmission phase is modeled as... The end time of the uplink data transmission phase is modeled as Model the completion time of the computation phase as Therefore, the uplink data transmission phase is represented as... The calculation stage is represented as The downlink data transmission phase is represented as Using a time-slot model, it can be shown that the drone only... Collecting sensory information, the drone only Calculation data, and drones only Transmit data to the execution device.

[0046] After establishing the time slot model, communication, computation, and energy consumption models can be established.

[0047] The communication model supports the data flow in the "perception-communication-computation-execution" task closed loop, covering two stages: uploading and downloading of perception information. In the uploading stage, the perception device transmits the collected perception information to the UAV node, completing the connection between perception and communication. In the downloading stage, the UAV processes the received data and transmits the processing results to the execution device user, thereby driving the final execution of the task and completing the entire closed loop of the perception task.

[0048] The devices in the crowd sensing system include drones Individual sensing devices, There are three execution devices, m∈{1,...,M},k∈{1,...,K},i∈{1,...,I}.

[0049] Optionally, the communication model includes: an uplink communication link model and a downlink communication link model; The expression for the uplink communication link model is: ; The expression for the downlink communication link model is: ; n represents any time segment. This represents the uplink transmission rate at which sensing device k uploads sensing information to drone m during the corresponding time segment. This represents the bandwidth allocated by the system to sensing device k for uploading sensing information. This represents the uplink transmit power of sensing device k. Represents user-related variables in the uplink. This represents the channel gain between the sensing device k and the drone m. Indicates the noise power spectral density; This represents the downlink transmission rate at which the drone m sends data to the execution device i during the corresponding time segment. This indicates the bandwidth allocated by the system to the drone m for sending data to the execution device i. This indicates the downlink transmit power of the UAV m. This represents the channel gain between the execution device i and the drone m. Represents user-related variables in the downlink; The expression for the computational model is:

[0050] in, Let m represent the total computing resources allocated to the drone and k sensing devices. Represents the total computing resources of drone m The computing resources allocated to the sensing device k.

[0051] Among them, the user-related variables of the uplink This can indicate which sensing device transmits sensing information to which drone, and is a user-related variable in the downlink. This indicates which drone sends its processed data to which execution device. Therefore, , .

[0052] In the uplink (uplink communication link), each sensing device can only be associated with one drone. In the downlink (downlink communication link), a drone can serve multiple execution devices, but can only serve one execution device at a time.

[0053] Therefore, the predefined constraints include the following user-related variables. User-related variables Constraints: ; ; .

[0054] in, This indicates that sensing device k can only establish a connection and transmit data with one drone at a time; This indicates that device i can only establish a connection with one drone at a time. This means that each drone m can serve multiple execution devices, with a maximum of I execution devices.

[0055] The predetermined constraints also include the following constraints for the communication model: ; ; ; ; .

[0056] This indicates that the total bandwidth of each drone in the uplink is B, and the total bandwidth allocation of the links connected to that drone is less than the total bandwidth of the drone. This indicates that the data upload time must be within the upload time slot phase (uplink data transmission phase); This indicates that the data download time must be within the download time slot phase (downlink data transmission phase); and These represent the constraint ranges for the uplink transmission power of the sensing device and the downlink transmission power of the UAV, respectively.

[0057] In this swarm intelligence sensing system, the total computing resources of drone m are (CPU cycles / s), the computing power allocated to sensing device k is (CPU cycles / s), the amount of data that sensing device k needs to transmit is The number of CPU cycles required for a drone to complete a 1-bit task is (CPU cycles / bit), the amount of data output after completing the perception task (bits), of which This indicates the compression ratio after the calculation and processing are complete.

[0058] The preset constraints also include the following constraints for the computational model: ; .

[0059] in, This indicates that the total computing resources allocated to the sensing devices by the drone m do not exceed the total computing resources of the drone m.

[0060] This indicates that the data uploaded by the drone m to the sensing device k must be within the computation time slot phase (computation phase). .

[0061] In addition, the predetermined constraints also include the following constraints for drone flight.

[0062] This means that the distance between any two drones is greater than the collision distance. This serves as a constraint to prevent collisions between drones. A preset distance threshold can be used. This indicates the position of drone m.

[0063] Alternatively, the expression for the energy consumption model is:

[0064] in, Indicates total energy consumption. This represents the uplink transmission energy consumption of sensing device k when transmitting sensing information to UAV m during the uplink data transmission period. This represents the downlink transmission energy consumption of the drone m when transmitting data to the execution device i during the downlink data transmission period. This represents the computational energy consumption of drone m during the computation period, where The coefficient representing the energy calculation. This represents the flight energy consumption of the drone m. This represents the flight speed of the drone, m.

[0065] Optionally, the formula for calculating the flight energy consumption of UAV m in one mission round is as follows:

[0066] in, The tip velocity of the UAV rotor blades. The average rotor speed of the drone while it is hovering. , , These are the energy consumption coefficients for drone flight. This refers to the flight speed of the drone.

[0067] In the above energy consumption model, the uplink transmit power of sensing device k is: Data transmission time is And in Within the range, therefore the uplink transmission power consumption can be obtained as follows: .

[0068] Similarly, the downlink transmit power from the UAV to the execution device i is Data transmission time is And in Within the range, the downlink transmission power consumption can be obtained as follows: .

[0069] Finally, the computational model shows that the processing time for the perceived data is... And in Within the range, therefore the amount of data processed on the drone The energy consumed is ,in This represents the coefficient used to calculate energy.

[0070] The solution vector includes the following optimization variables: , , , , .

[0071] Therefore, the problem can ultimately be summarized into the following mathematical problem: .in, Corresponding to the associated state information in the solution vector, and Corresponding to the communication parameters in the solution vector, Corresponding to the calculation of resource parameters in the solution vector, This corresponds to the UAV's flight trajectory information in the solution vector.

[0072] With the time slot model, computational model, communication model, and energy consumption model established, and the corresponding predetermined constraints defined, the above mathematical models and constraints can be used as a reference. Figure 5 As shown, the optimal parameters required by the system are determined through a two-stage optimization method (i.e., firstly, a global search is performed using a genetic algorithm to locate one or more preliminary optimal solutions, and then, using these preliminary optimal solutions as the initial state, further optimization is performed using a deep deterministic policy gradient algorithm to obtain a more accurate solution vector).

[0073] Optionally, the population is initialized under predetermined constraints, wherein each individual in the population satisfies the predetermined constraints, which include: ; ; ; ; ; ; ; ; ; ; ; ; ; in, This means that sensing device k can only establish a connection with one drone at a time for data transmission; This indicates that device i can only establish a connection with one drone at a time; This indicates that each drone can serve multiple execution devices, where I is the number of execution devices; This indicates that the total bandwidth allocation for each drone in the uplink is less than the total bandwidth of the drone, and B represents the total bandwidth of each drone in the uplink. This indicates that the data upload time occurred during the upstream data transmission period; This indicates that the data download time is within the downlink data transmission period; and These represent the constraint ranges for the uplink transmit power of the sensing device and the downlink transmit power of the UAV, respectively; where, This indicates that the total computing resources allocated to the sensing devices by drone m are less than the total computing resources of drone m. This indicates that the time of data uploaded by the drone m to the sensing device k falls within the calculation period; This means that the distance between any two drones is greater than the collision distance. ; This indicates the position of drone m.

[0074] Optionally, a genetic algorithm is used to search for the optimal individual with the objective of minimizing the fitness value of each individual. This includes: in each iteration, determining the fitness value of each individual in the initial population of the current iteration; determining the roulette wheel selection probability for each individual based on the inverse fitness method; selecting individuals from the initial population according to the roulette wheel selection probability, and performing crossover and mutation on the selected individuals to obtain a subpopulation in the current iteration; merging the initial population and the subpopulation to obtain a merged population, and selecting the top-P dominant individuals from the merged population, sorted by fitness value from smallest to largest, to enter the next iteration, where P is a preset number; and, upon completion of the iterations, selecting the individual with the smallest fitness value in the entire iteration as the optimal individual.

[0075] In other words, in the genetic algorithm, the goal is to minimize the total energy consumption mentioned above. The goal is to obtain the optimal individual, where all individuals, whether in the initial population or newly obtained in each iteration, must satisfy the aforementioned predetermined constraints.

[0076] In each iteration, the individuals that entered the current iteration after the previous iteration are used as the initial population for the current iteration, and the total energy consumption corresponding to each individual in the initial population is determined. The population is then updated to identify individuals that need to proceed to the next iteration.

[0077] Specifically, the optimization process at this stage is as follows: The problem is solved by performing a global exploration in the initial stage. Through selection, crossover, and mutation operations using a genetic algorithm, a broad search is conducted throughout the solution space to locate the neighborhood of one or more high-quality solutions. Each solution in the solution space contains various parameters that need to be optimized for the total system energy consumption in the aforementioned optimization problem, including parameters related to the association state between the UAV and the equipment, data communication, and energy consumption calculation. Specifically: Step 1: Initialize the population within the variable range. The information of each individual in the population includes the association state of the sensing device-UAV-execution device, as well as the amount of resources required for various communications and calculations. Then, the fitness value of each individual (i.e. the total energy consumption of the system with penalty constraints) can be calculated using the objective function.

[0078] Step 2: Enter the individual selection phase. In each iteration, individuals from the parent generation (i.e., the initial population in this iteration) need to be selected with a certain probability. To increase the probability of obtaining a optimal solution in the selection phase, a roulette wheel selection probability is constructed using the inverse of fitness. The probability formula is as follows.

[0079] ,

[0080] in, Let be the roulette wheel selection probability for the i-th individual. Let P be the fitness value corresponding to the i-th individual, and let P represent the total number of individuals in the population. This is a preset value.

[0081] A number of individuals can be selected from the initial population based on the probability of selection in roulette.

[0082] Step 3: After obtaining the selected individuals, crossover and mutation can be performed on them to obtain the subpopulation for this iteration. Offspring are generated by simulating binary crossover, and polynomial mutation is used to perturb some variables to increase diversity. After evaluating the fitness of the parents and offspring, the populations are merged, and the top P dominant individuals with the smallest fitness values ​​are selected into the next generation using an elite retention strategy. The fitness values ​​of these P dominant individuals are compared, and the individual with the smallest fitness value is the optimal individual in this generation. This process is continued iteratively until the optimal individual (i.e., the near-optimal individual solution) of the genetic algorithm stage is obtained.

[0083] Then, the optimal individual is used as the initial state in the deep deterministic policy gradient algorithm, so that the search starts from the optimal individual. In the deep deterministic policy gradient algorithm, by making minor updates to the state at each time step, and determining the total energy consumption and whether the constraints are violated after each update, a better solution than the optimal individual can be found.

[0084] Optionally, the operation of determining the optimal solution vector using the optimal individual as the initial state through the deep deterministic policy gradient algorithm includes: using the optimal individual as the initial state, and initializing the action network, evaluation network, and target network corresponding to the action network and evaluation network, respectively, wherein any state in the deep deterministic policy gradient algorithm represents a solution vector; at each time step, determining the corresponding action information by inputting the state information corresponding to the corresponding time step into the action network, the action information representing the magnitude of fine-tuning the solution vector corresponding to the state at the corresponding time step; adjusting the state information at the corresponding time step according to the action information to obtain the next... The process involves determining the state information corresponding to each time step and the total energy consumption for determining the state information corresponding to the next time step. Based on the total energy consumption and predetermined constraints, the reward value for the action information is determined, where a higher total energy consumption results in a lower reward value. If the state information determined through the action information violates the predetermined constraints, the reward value will be penalized by a preset penalty term. Based on the reward value, with the goal of minimizing the total energy consumption, the evaluation network is updated using the minimum mean square error loss function, the action network is updated using a deterministic policy gradient ascent, and the target network is updated using a soft update method. Through continuous iteration over multiple time steps, the optimal solution vector is determined based on the converged action network and evaluation network.

[0085] Specifically, the solution obtained from global exploration (the optimal individual) is used as the initial state of the deep deterministic policy gradient algorithm. Starting from this initial state, the deep deterministic policy gradient algorithm is used for further optimization. By alternately executing deterministic policy gradient ascent and Q-value regression in the state and action spaces, and using soft updates to ensure the stability of the target network, the optimal policy is obtained, and the overall energy consumption of the system is minimized.

[0086] In the deep deterministic policy gradient algorithm, the reward function is formulated as follows: r = -e(1 + λ·η) Where e represents the total energy consumption of the system determined based on the above energy consumption model, λ is the weight parameter, and η is the penalty term, which is used to reduce the reward value of the action information (policy) output by the action network when the above predetermined constraints are violated.

[0087] The following describes the process of a single state update in the deep deterministic policy gradient algorithm.

[0088] At time step t, the corresponding state information s t The input is fed into the action network to obtain the action information a at time step t. t Then based on the action information a t Adjust s t , obtain the state information s of the next time step t+1. t+1 The form of the state information is similar to that of the individuals in the aforementioned genetic algorithm; both are in the form of solution vectors. That is, the state information at a time step includes the association state information between the UAV, sensing devices, and execution devices, the UAV's flight trajectory information, communication parameters, and computational resource parameters. Therefore, by using the state information s... t+1 Substituting this into the energy consumption model above, we can determine the relationship between the state information s and the given information. t+1 The corresponding total energy consumption e. Then, based on the aforementioned predetermined constraints, the state information s can be determined. t+1 Has any of the predetermined constraints been violated? Based on this status information s t+1 Whether any of the predetermined constraints is violated determines the corresponding penalty term η. Therefore, based on this state information s t+1 And based on the formula for the reward function mentioned above, the relationship with action information a can be determined. t The corresponding reward value r t .

[0089] Through reward value r t The evaluation value Q can be determined, and the action network, evaluation network, and their corresponding target networks (i.e., the target action network and the target evaluation network) can be updated accordingly. After all networks have been updated and convergence is achieved, the action information that minimizes the energy consumption corresponding to the state information can be determined. The state information with the lowest energy consumption can then be taken as the optimal solution. This allows us to obtain the parameters required for the swarm intelligence perception system to achieve minimum energy consumption and task closure.

[0090] Specifically, the process begins by randomly initializing the Actor and Critic networks, followed by initializing the target network and the empirical replay pool. Further, the near-optimal individual solutions are used as the initial states in the deep deterministic policy gradient algorithm, and noisy exploration is employed to obtain and store the transformation tuples. Then, d mini-batches are sampled from the empirical replay pool to calculate the TD target. The Critic network is updated using the minimum mean squared error loss function, the Actor network is updated using deterministic policy gradient ascent, and the target network is updated using a soft update method. At each time step, a state s is obtained, containing the associated state factors of the sensing device, the UAV, and the execution device, the UAV's flight trajectory, and the resources required for various communications and computations. Using this state s, the total system energy consumption can be calculated. Through continuous iterative comparison, the minimum overall system energy consumption value, i.e., the optimal solution to the aforementioned problem, is finally obtained.

[0091] In addition, refer to Figure 1 As shown, according to a second aspect of this embodiment, a storage medium is provided. The storage medium includes a stored program, wherein, when the program is executed, a processor performs any of the methods described above.

[0092] Therefore, according to this embodiment, compared with the prior art, this method has the following beneficial effects: 1. This method addresses the problem in the prior art that it does not fully consider the insufficient computing power of ground sensing equipment and the susceptibility of end-to-end links to obstruction. It uses UAVs with high computing power in conjunction with end-to-end computing processing to achieve aerial relay and computing compensation in the "sensing-computing" link, effectively improving data processing speed and enhancing the connection reliability of the "sensing-communication" link, ensuring the efficient execution of the mission loop. 2. This method addresses the shortcoming of the air-to-ground transmission link being easily obstructed due to the fixed-location deployment of UAVs in the end-to-end swarm sensing system. It considers the path planning of UAVs to ensure the reliability of the air-to-ground link. 3. This method addresses the problem of limited network resources in the end-to-end swarm sensing system by jointly optimizing communication and computing resources to achieve effective resource allocation. While ensuring the performance of the entire mission loop of "sensing-communication-computing-execution", it significantly reduces the overall operating energy consumption of the system.

[0093] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that the present invention is not limited to the described order of actions, because according to the present invention, some steps can be performed in other orders or simultaneously. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to the present invention.

[0094] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods according to the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods described in the various embodiments of the present invention.

[0095] Example 2

[0096] Figure 6 A two-stage optimization apparatus for an end-to-end swarm intelligence sensing system according to this embodiment is shown, which corresponds to the method described according to the first aspect of Embodiment 1. (See reference...) Figure 6 As shown, the two-stage optimization device for an end-to-end swarm intelligence sensing system includes: an initialization module 610, used to initialize the population, where each individual in the population represents a solution vector, which includes the association state information between the UAV, sensing devices, and execution devices, the UAV's flight trajectory information, communication parameters, and computing resource parameters; and an energy consumption determination module 620, used to determine the total energy consumption corresponding to an individual based on a pre-constructed time slot model, communication model, computing model, and energy consumption model, as the fitness corresponding to that individual. The time slot model represents different stages in a round of tasks in the swarm intelligence sensing system, and the communication model characterizes the communication between the sensing devices and the UAV. The system includes a computational model to characterize the computational resources used by the drones in uploading sensing information and transmitting data from the drones to the execution devices, and an energy consumption model to quantify the uplink and downlink communication energy consumption, drone computing energy consumption, and flight energy consumption during the operation of the swarm intelligence sensing system. A first optimization module 630 uses a genetic algorithm to search for the optimal individual with the objective of minimizing its fitness value. A second optimization module 640 uses the optimal individual as the initial state and a deep deterministic policy gradient algorithm to determine the optimal solution vector. This optimal solution vector is used by several drones, several sensing devices, and several execution devices to perform tasks in the swarm intelligence sensing system.

[0097] Optionally, one round of tasks for the swarm intelligence sensing system in the time-slot model includes: uplink data transmission period. Calculation period and downlink data transmission period The service time T of one round of tasks in the crowd-sensing system is divided into N time segments. , , The duration of a time segment, the uplink data transmission period. Calculation period Downlink data transmission period ,in, This indicates the start time of the uplink transmission from sensing device k. This represents the time it takes for the drone corresponding to sensor device k to complete data processing. This indicates the uplink transmission end time of sensor device k.

[0098] Optionally, the communication model includes: an uplink communication link model and a downlink communication link model; The expression for the uplink communication link model is: ; The expression for the downlink communication link model is: ; Where n represents any time segment, This represents the uplink transmission rate at which sensing device k uploads sensing information to drone m during the corresponding time segment. This represents the bandwidth allocated by the system to sensing device k for uploading sensing information. This represents the uplink transmit power of sensing device k. Represents user-related variables in the uplink. This represents the channel gain between the sensing device k and the drone m. Indicates the noise power spectral density; This represents the downlink transmission rate at which the drone m sends data to the execution device i during the corresponding time segment. This indicates the bandwidth allocated by the system to the drone m for sending data to the execution device i. This indicates the downlink transmit power of the UAV m. This represents the channel gain between the execution device i and the drone m. Represents user-related variables in the downlink; The expression for the computational model is:

[0099] in, Let m represent the total computing resources allocated to the drone and k sensing devices. Represents the total computing resources of drone m The computing resources allocated to the sensing device k.

[0100] Alternatively, the expression for the energy consumption model is: ,in, Indicates total energy consumption. This represents the uplink transmission energy consumption of sensing device k when transmitting sensing information to UAV m during the uplink data transmission period. This represents the downlink transmission energy consumption of the drone m when transmitting data to the execution device i during the downlink data transmission period. This represents the computational energy consumption of drone m during the computation period, where The coefficient representing the energy calculation. This represents the flight energy consumption of the drone m; Furthermore, the formula for calculating the flight energy consumption of the drone m is as follows:

[0101] in, The tip velocity of the UAV rotor blades. The average rotor speed of the drone while it is hovering. For the drone's flight speed, , , These are the energy consumption coefficients for drone flight.

[0102] Optionally, the first optimization module 630 is used to determine the fitness value of each individual in the initial population of the current iteration in each iteration; determine the roulette wheel selection probability of each individual based on the inverse fitness method; select individuals from the initial population according to the roulette wheel selection probability, and perform crossover and mutation on the selected individuals to obtain the subpopulation in the current iteration; merge the initial population and the subpopulation to obtain the merged population, and select the top-P dominant individuals in the merged population arranged in ascending order of fitness value to enter the next iteration, where P is a preset number; and, if the iteration is completed, select the individual with the smallest fitness value in the entire iteration as the optimal individual.

[0103] Optionally, the second optimization module 640 is used to initialize the optimal individual as the initial state, and to initialize the action network, evaluation network, and target network corresponding to the action network and evaluation network, respectively. In the deep deterministic policy gradient algorithm, any state represents a solution vector. At each time step, by inputting the state information corresponding to the corresponding time step into the action network, the corresponding action information is determined. The action information represents the magnitude of fine-tuning the solution vector corresponding to the state at the corresponding time step. Based on the action information, the state information at the corresponding time step is adjusted to obtain the state information corresponding to the next time step, and the total energy consumption of the state information corresponding to the next time step is determined. Based on the total energy consumption and predetermined constraints, the reward value of the action information is determined, where a higher total energy consumption results in a lower reward value. If the state information determined by the action information violates the predetermined constraints, the reward value will be penalized by a preset penalty term. Based on the reward value, with the goal of minimizing the total energy consumption, the evaluation network is updated using the minimum mean squared error loss function, the action network is updated using deterministic policy gradient ascent, and the target network is updated using a soft update method. Through continuous iteration over multiple time steps, the optimal solution vector is determined based on the converged action network and evaluation network.

[0104] Therefore, according to the technical solution of this embodiment, this embodiment has the following technical effects: 1. This embodiment addresses the problem in the prior art that the computing power of ground sensing equipment is insufficient and the end-to-end link is easily affected by obstruction. It adopts a UAV with high computing power combined with end-to-end computing processing to realize aerial relay and computing compensation in the "sensing-computing" link, effectively improving data processing speed and enhancing the connection reliability of the "sensing-communication" link, ensuring the efficient execution of the task loop. 2. This embodiment addresses the disadvantage of the air-to-ground transmission link being easily obstructed due to the fixed-position deployment of UAVs in the end-to-end swarm sensing system. It considers the path planning of UAVs to ensure the reliability of the air-to-ground link. 3. This embodiment addresses the problem of limited network resources in the end-to-end swarm sensing system by jointly optimizing communication and computing resources to achieve effective resource allocation. While ensuring the performance of the entire task loop of "sensing-communication-computing-execution", it significantly reduces the overall operating energy consumption of the system.

[0105] Example 3

[0106] Figure 7 A two-stage optimization apparatus for an end-to-end swarm intelligence sensing system according to this embodiment is shown, which corresponds to the method described according to the first aspect of Embodiment 1. (See reference...) Figure 7As shown, the two-stage optimization device for an end-to-end swarm intelligence sensing system includes: a processor 710; and a memory 720 connected to the processor 710, used to provide the processor 710 with instructions to process the following steps: initializing a population, where each individual in the population represents a solution vector, the solution vector including the association state information between the UAV, sensing devices, and execution devices, the UAV's flight trajectory information, communication parameters, and computing resource parameters; and determining the total energy consumption corresponding to each individual based on a pre-constructed time-slot model, communication model, computing model, and energy consumption model, as the fitness corresponding to that individual, wherein the time-slot model is used to represent the swarm intelligence sensing system. In different stages of the first round of the task, the communication model is used to characterize the communication resources used by the sensing devices to upload sensing information to the UAVs and by the UAVs to transmit data to the execution devices. The computation model is used to characterize the computational resources used by the UAVs. The energy consumption model is used to quantify the uplink and downlink communication energy consumption, UAV computational energy consumption, and flight energy consumption during the operation of the swarm intelligence sensing system. Using a genetic algorithm, the optimal individual is searched with the goal of minimizing the fitness value of the individual. The optimal individual is used as the initial state, and the optimal solution vector is determined by the deep deterministic policy gradient algorithm. The optimal solution vector is used by several UAVs, several sensing devices, and several execution devices to perform tasks in the swarm intelligence sensing system.

[0107] Optionally, one round of tasks for the swarm intelligence sensing system in the time-slot model includes: uplink data transmission period. Calculation period and downlink data transmission period The service time T of one round of tasks in the crowd-sensing system is divided into N time segments. , , The duration of a time segment, the uplink data transmission period. Calculation period Downlink data transmission period ,in, This indicates the start time of the uplink transmission from sensing device k. This represents the time it takes for the drone corresponding to sensor device k to complete data processing. This indicates the uplink transmission end time of sensor device k.

[0108] Optionally, the communication model includes: an uplink communication link model and a downlink communication link model; The expression for the uplink communication link model is: ; The expression for the downlink communication link model is: ; Where n represents any time segment, This represents the uplink transmission rate at which sensing device k uploads sensing information to drone m during the corresponding time segment. This represents the bandwidth allocated by the system to sensing device k for uploading sensing information. This represents the uplink transmit power of sensing device k. Represents user-related variables in the uplink. This represents the channel gain between the sensing device k and the drone m. Indicates the noise power spectral density; This represents the downlink transmission rate at which the drone m sends data to the execution device i during the corresponding time segment. This indicates the bandwidth allocated by the system to the drone m for sending data to the execution device i. This indicates the downlink transmit power of the UAV m. This represents the channel gain between the execution device i and the drone m. Represents user-related variables in the downlink; The expression for the computational model is:

[0109] in, Let m represent the total computing resources allocated to the drone and k sensing devices. Represents the total computing resources of drone m The computing resources allocated to the sensing device k.

[0110] Alternatively, the expression for the energy consumption model is:

[0111] in, Indicates total energy consumption. This represents the uplink transmission energy consumption of sensing device k when transmitting sensing information to UAV m during the uplink data transmission period. This represents the downlink transmission energy consumption of the drone m when transmitting data to the execution device i during the downlink data transmission period. This represents the computational energy consumption of drone m during the computation period, where The coefficient representing the energy calculation. This represents the flight energy consumption of the drone m; Furthermore, the formula for calculating the flight energy consumption of the drone m is as follows:

[0112] in, The tip velocity of the UAV rotor blades. The average rotor speed of the drone while it is hovering. For the drone's flight speed, , , These are the energy consumption coefficients for drone flight.

[0113] Optionally, a genetic algorithm is used to search for the optimal individual with the objective of minimizing the fitness value of each individual. This includes: in each iteration, determining the fitness value of each individual in the initial population of the current iteration; determining the roulette wheel selection probability for each individual based on the inverse fitness method; selecting individuals from the initial population according to the roulette wheel selection probability, and performing crossover and mutation on the selected individuals to obtain a subpopulation in the current iteration; merging the initial population and the subpopulation to obtain a merged population, and selecting the top-P dominant individuals from the merged population, sorted by fitness value from smallest to largest, to enter the next iteration, where P is a preset number; and, upon completion of the iterations, selecting the individual with the smallest fitness value in the entire iteration as the optimal individual.

[0114] Optionally, the operation of determining the optimal solution vector by using the optimal individual as the initial state and employing a deep deterministic policy gradient algorithm includes: The optimal individual is used as the initial state, along with the initial action network, evaluation network, and target network corresponding to the action and evaluation networks, respectively. In the deep deterministic policy gradient algorithm, any state represents a solution vector. At each time step, the corresponding action information is determined by inputting the state information corresponding to that time step into the action network. The action information represents the magnitude of fine-tuning the solution vector corresponding to the state at that time step. Based on the action information, the state information at the corresponding time step is adjusted to obtain the state information for the next time step, and the total energy consumption for the state information at the next time step is determined. Based on the total energy consumption and predetermined constraints, the reward value for the action information is determined, where higher total energy consumption results in a lower reward value. If the state information determined by the action information violates the predetermined constraints, the reward value will be penalized by a preset penalty term. Based on the reward value, with the goal of minimizing the total energy consumption, the evaluation network is updated using the minimum mean squared error loss function, the action network is updated using deterministic policy gradient ascent, and the target network is updated using a soft update method. Through continuous iteration over multiple time steps, the optimal solution vector is determined based on the converged action network and evaluation network.

[0115] Therefore, according to the technical solution of this embodiment, this embodiment has the following technical effects: 1. This embodiment addresses the problem in the prior art that the computing power of ground sensing equipment is insufficient and the end-to-end link is easily affected by obstruction. It adopts a UAV with high computing power combined with end-to-end computing processing to realize aerial relay and computing compensation in the "sensing-computing" link, effectively improving data processing speed and enhancing the connection reliability of the "sensing-communication" link, ensuring the efficient execution of the task loop. 2. This embodiment addresses the disadvantage of the air-to-ground transmission link being easily obstructed due to the fixed-position deployment of UAVs in the end-to-end swarm sensing system. It considers the path planning of UAVs to ensure the reliability of the air-to-ground link. 3. This embodiment addresses the problem of limited network resources in the end-to-end swarm sensing system by jointly optimizing communication and computing resources to achieve effective resource allocation. While ensuring the performance of the entire task loop of "sensing-communication-computing-execution", it significantly reduces the overall operating energy consumption of the system.

[0116] The sequence numbers of the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0117] In the above embodiments of the present invention, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0118] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual coupling, direct coupling, or communication connection may be through some interfaces; the indirect coupling or communication connection between units or modules may be electrical or other forms.

[0119] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0120] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0121] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0122] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A two-stage optimization method for end-to-end crowd sensing systems, characterized in that, The swarm intelligence sensing system includes several drones, several sensing devices, and several execution devices; the method includes: Initialize the population, where each individual represents a solution vector. The solution vector includes the association state information between the UAV, the sensing device and the execution device, the flight trajectory information of the UAV, communication parameters and computing resource parameters. Based on the pre-constructed time-slot model, communication model, computing model, and energy consumption model, the total energy consumption corresponding to an individual is determined as the fitness corresponding to the individual. The time-slot model is used to represent different stages in a round of tasks in the swarm intelligence perception system. The communication model is used to characterize the communication resources used by the sensing device to upload sensing information to the UAV and by the UAV to transmit data to the execution device. The computing model is used to characterize the computing resources used by the UAV. The energy consumption model is used to quantify the uplink and downlink communication energy consumption, UAV computing energy consumption, and flight energy consumption during the operation of the swarm intelligence perception system. Using a genetic algorithm, the optimal individual is searched by minimizing the fitness value of each individual. Using the optimal individual as the initial state, the optimal solution vector is determined through a deep deterministic policy gradient algorithm. The optimal solution vector is used by the plurality of UAVs, the plurality of sensing devices, and the plurality of execution devices to perform tasks in the swarm intelligence sensing system.

2. The method according to claim 1, characterized in that, One round of tasks for the swarm intelligence sensing system in the time slot model includes: uplink data transmission period. Calculation period and downlink data transmission period The service time T of one round of tasks in the crowd-sensing system is divided into N time segments. , , The duration of a time segment, the uplink data transmission period. Calculation period Downlink data transmission period ,in, This indicates the start time of the uplink transmission from sensing device k. This represents the time it takes for the drone corresponding to sensor device k to complete data processing. This indicates the uplink transmission end time of sensor device k.

3. The method according to claim 2, characterized in that, The communication model includes: an uplink communication link model and a downlink communication link model; The expression for the uplink communication link model is: ； The expression for the downlink communication link model is: ； Where n represents any time segment, This represents the uplink transmission rate at which sensing device k uploads sensing information to drone m during the corresponding time segment. This represents the bandwidth allocated by the system to sensing device k for uploading sensing information. This represents the uplink transmit power of sensing device k. Represents user-related variables in the uplink. This represents the channel gain between the sensing device k and the drone m. Indicates the noise power spectral density; This represents the downlink transmission rate at which the drone m sends data to the execution device i during the corresponding time segment. This indicates the bandwidth allocated by the system to the drone m for sending data to the execution device i. This indicates the downlink transmit power of the UAV m. This represents the channel gain between the execution device i and the drone m. Represents user-related variables in the downlink; The expression for the computational model is: ； in, Let m represent the total computing resources allocated to the drone and k sensing devices. Represents the total computing resources of drone m The computing resources allocated to the sensing device k.

4. The method according to claim 3, characterized in that, The expression for the energy consumption model is: ； in, This represents the total energy consumption. This represents the uplink transmission energy consumption of sensing device k in transmitting sensing information to UAV m during the uplink data transmission period. This represents the downlink transmission power consumption of the drone m transmitting data to the execution device i during the downlink data transmission period. This represents the computational energy consumption of UAV m during the computation period, where The coefficient representing the energy calculation. This represents the flight energy consumption of the drone m; Furthermore, the formula for calculating the flight energy consumption of the drone m is as follows: ； in, The tip velocity of the UAV rotor blades. The average rotor speed of the drone while it is hovering. For the drone's flight speed, , , These are the energy consumption coefficients for drone flight.

5. The method according to claim 3, characterized in that, The operations for initializing the population include: Under predetermined constraints, the population is initialized, wherein each individual in the population satisfies the predetermined constraints, which include: ； ； ； ； ； ； ； ； ； ； ； ； ； in, This indicates that sensing device k can only establish a connection with one drone at a time for data transmission. This indicates that device i can only establish a connection with one drone at a time; This indicates that each drone can serve multiple execution devices, where I is the number of execution devices; This indicates that the total bandwidth allocation for each drone in the uplink is less than the total bandwidth of the drone, and B represents the total bandwidth of each drone in the uplink. This indicates that the data upload time falls within the uplink data transmission period. This represents the amount of data that sensing device k needs to transmit; This indicates that the data download time falls within the downlink data transmission period. This represents the amount of data output by the drone m after data processing. and These represent the constraint ranges for the uplink transmit power of the sensing device and the downlink transmit power of the UAV, respectively; where, This indicates that the total computing resources allocated to the sensing devices by drone m do not exceed the total computing resources of drone m. This indicates that the time period during which the drone m calculates and the sensing device k uploads the data falls within the calculation time period. , Indicates the task processing speed of the drone; This means that the distance between any two drones is greater than the collision distance. ; This indicates the position of drone m.

6. The method according to claim 1, characterized in that, The operation of using a genetic algorithm to search for the optimal individual with the objective of minimizing the fitness value of an individual includes: In each iteration, the fitness value of each individual in the initial population of the current iteration is determined; Based on the inverse fitness method, the probability of choosing the roulette wheel for each individual is determined. Based on the selection probability of the roulette wheel, individuals in the initial population are selected, and the selected individuals are cross-crossed and mutated to obtain a subpopulation in this round of iteration. The initial population and the subpopulation are merged to obtain a merged population, and the top-P dominant individuals in the merged population, arranged in ascending order of fitness value, are selected to enter the next round of iteration, where P is a preset number. If the iteration is complete, the individual with the smallest fitness value in the entire round of iterations is taken as the optimal individual.

7. The method according to claim 1, characterized in that, The operation of determining the optimal solution vector using the optimal individual as the initial state and through the deep deterministic policy gradient algorithm includes: The optimal individual is taken as the initial state, and the action network, evaluation network, and target network corresponding to the action network and evaluation network are initialized respectively, wherein any state in the deep deterministic policy gradient algorithm represents a solution vector; At each time step, the corresponding action information is determined by inputting the state information corresponding to the corresponding time step into the action network. The action information represents the magnitude of fine-tuning of the solution vector corresponding to the state of the corresponding time step. Based on the action information, the state information of the corresponding time step is adjusted to obtain the state information corresponding to the next time step, and the total energy consumption of the state information corresponding to the next time step is determined. The reward value of the action information is determined based on the total energy consumption and the predetermined constraints. The higher the total energy consumption, the lower the reward value. If the state information determined by the action information violates the predetermined constraints, the reward value will be penalized by a preset penalty item. Based on the reward value, with the goal of minimizing total energy consumption, the evaluation network is updated using the minimum mean square error loss function, the action network is updated using the deterministic policy gradient ascent, and the target network is updated using a soft update method. Through continuous iteration over multiple time steps, the optimal solution vector is determined based on the converged action network and evaluation network.

8. A storage medium, characterized in that, The storage medium includes a stored program, wherein, when the program is executed, the method described in any one of claims 1 to 7 is performed by a processor.

9. A two-stage optimization device for an end-to-end swarm intelligence sensing system, characterized in that, The swarm intelligence sensing system includes several drones, several sensing devices, and several execution devices, including: An initialization module is used to initialize the population. Each individual in the population represents a solution vector, which includes the association status information between the UAV, the sensing device and the execution device, the flight trajectory information of the UAV, communication parameters and computing resource parameters. The energy consumption determination module is used to determine the total energy consumption corresponding to an individual based on a pre-constructed time slot model, communication model, computing model, and energy consumption model, which serves as the fitness of the individual. The time slot model represents different stages in a round of tasks in the swarm intelligence perception system. The communication model characterizes the communication resources used by the perception device to upload perception information to the drone and by the drone to transmit data to the execution device. The computing model characterizes the computing resources used by the drone. The energy consumption model quantifies the uplink and downlink communication energy consumption, drone computing energy consumption, and flight energy consumption during the operation of the swarm intelligence perception system. The first optimization module is used to search for the optimal individual by using a genetic algorithm with the goal of minimizing the fitness value of the individual. The second optimization module is used to take the optimal individual as the initial state and determine the optimal solution vector through a deep deterministic policy gradient algorithm. The optimal solution vector is used by the plurality of UAVs, the plurality of sensing devices and the plurality of execution devices to perform tasks in the swarm intelligence sensing system.

10. A two-stage optimization device for an end-to-end swarm intelligence sensing system, characterized in that, The swarm intelligence sensing system includes several drones, several sensing devices, and several execution devices, including: Processor; and A memory, connected to the processor, for providing the processor with instructions to perform the following processing steps: Initialize the population, where each individual represents a solution vector. The solution vector includes the association state information between the UAV, the sensing device and the execution device, the flight trajectory information of the UAV, communication parameters and computing resource parameters. Based on the pre-constructed time-slot model, communication model, computing model, and energy consumption model, the total energy consumption corresponding to an individual is determined as the fitness corresponding to the individual. The time-slot model is used to represent different stages in a round of tasks in the swarm intelligence perception system. The communication model is used to characterize the communication resources used by the sensing device to upload sensing information to the UAV and by the UAV to transmit data to the execution device. The computing model is used to characterize the computing resources used by the UAV. The energy consumption model is used to quantify the uplink and downlink communication energy consumption, UAV computing energy consumption, and flight energy consumption during the operation of the swarm intelligence perception system. Using a genetic algorithm, the optimal individual is searched by minimizing the fitness value of each individual. Using the optimal individual as the initial state, the optimal solution vector is determined through a deep deterministic policy gradient algorithm. The optimal solution vector is used by the plurality of UAVs, the plurality of sensing devices, and the plurality of execution devices to perform tasks in the swarm intelligence sensing system.

Images