Multi-layer air network energy consumption optimization method based on sensing, communication, calculation and storage mechanism

Abstract

The invention discloses a multi-layer air network energy consumption optimization method based on a sensing, communication, calculation and storage mechanism. The method comprises the following steps: S1, establishing a multi-layer air network system model; s2, respectively establishing an unmanned aerial vehicle-wireless equipment perception model, a wireless equipment-unmanned aerial vehicle-high-altitude platform communication model and an unmanned aerial vehicle-high-altitude platform cache model; s3, establishing a system delay calculation model; s4, establishing a system energy consumption calculation model; s5, establishing an optimization model for jointly optimizing a time scheduling proportion, a task unloading proportion, a compression selection proportion, a service cache decision and a UAV trajectory; and S6, carrying out iterative solution to obtain the optimal time scheduling proportion, task unloading proportion, compression selection proportion, service cache decision and UAV trajectory so as to enable the total energy consumption of the multi-layer air network system model to be minimum. According to the method, perception, communication, calculation and caching are integrated into a unified system model, network functions are re-conceived fundamentally, the problem that a traditional ground network is limited is solved, the method is more flexible, and time delay and energy consumption are reduced while service efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the field of mobile communication technology, and in particular to a multi-layer air network energy consumption optimization method based on a sensing, communication, computing and storage mechanism. Background Art

[0002] As the digital economy and the real economy continue to deepen, traditional communication networks are at the forefront of technological change. Future networks will not only require the efficient interconnection of hundreds of billions of intelligent terminals, but also millisecond-level perception and precise control of the physical world. This poses severe challenges to the real-time, reliability, and intelligence of the network. Against this backdrop, communication networks are rapidly evolving from single information transmission pipelines to multifunctional converged infrastructures. Among them, integrated sensing and communication (ISAC) technology, by sharing hardware platforms, spectrum resources, and signal processing units, essentially breaks the boundaries between communication and perception systems, and has become one of the core enabling technologies for sixth-generation mobile communications (6G).

[0003] However, with the booming development of emerging applications such as artificial intelligence and the metaverse, wireless devices (WDs) are no longer merely passive sensing objects. Instead, they have become active participants in the continuous generation of computationally intensive tasks. This dual nature makes the introduction of efficient computational support in ISAC systems inevitable. While mobile edge computing (MEC) has proven to be an effective means of improving computational efficiency by sinking multidimensional resources to the network edge, its deep integration with ISAC still faces unique challenges. First, perception and computation tasks compete for limited time-frequency resources, requiring a delicate balance between perception accuracy and computational efficiency. Second, because task offloading and data transmission strategies rely heavily on network status information obtained through perception, perception, communication, and computation functions exhibit a complex coupling relationship. More importantly, task processing in MEC environments relies heavily on specific service support, such as deep learning models and algorithm libraries. When these services are not cached on edge nodes, they must be retrieved from remote data centers, significantly increasing end-to-end latency. Therefore, maximizing local service availability within the limited edge cache space becomes a key challenge. Furthermore, the transmission overhead caused by task offloading is a major performance bottleneck in MEC systems, yet data compression techniques, widely adopted in wireless communications, have received little attention in this context. In particular, given the heterogeneity of task data and the cost of compression and decompression operations, universal compression for all offloaded data is suboptimal. This emphasizes the need to design more flexible compression mechanisms in MEC systems. Ultimately, these intertwined factors, coupled with diverse user needs, necessitate a comprehensive framework to orchestrate sensing, communication, computation, caching, and compression functions. On the other hand, in traditional terrestrial networks, base stations are susceptible to infrastructure failures, building obstructions, or natural disasters, making it difficult to provide flexible and robust services to WDs in remote or hotspot areas. Increasing base station density also incurs significant deployment and maintenance costs. Summary of the Invention

[0004] The present invention provides a multi-layer air network energy consumption optimization method based on the sensor communication computing and storage mechanism to overcome the above technical problems.

[0005] In order to achieve the above object, the technical solution of the present invention is:

[0006] A multi-layer air network energy consumption optimization method based on a sensory communication computing and storage mechanism, the specific steps include:

[0007] S1: Build a multi-layer aerial network system model for integrating wireless devices, drones, and high-altitude platforms;

[0008] S2: Based on the multi-layer aerial network system model, a drone-wireless device perception model, a wireless device-drone-high-altitude platform communication model, and a drone-high-altitude platform cache model are respectively established;

[0009] S3: Establishing a system delay calculation model based on the wireless device-drone-high-altitude platform communication model and the drone-high-altitude platform cache model;

[0010] S4: Establishing a system energy consumption calculation model based on the system delay calculation model;

[0011] S5: Based on the multi-layer aerial network system model, the UAV-wireless device perception model, the wireless device-UAV-high-altitude platform communication model, the UAV-high-altitude platform cache model, the system delay calculation model, and the system energy consumption calculation model, an optimization model is established for jointly optimizing the time scheduling ratio, the task offloading ratio, the compression selection ratio, the service cache decision, and the UAV trajectory to minimize the total energy consumption of the multi-layer aerial network system model;

[0012] S6: Decompose the optimization model into several sub-models, and iteratively solve the sub-models using corresponding algorithms to obtain the optimal time scheduling ratio, task offloading ratio, compression selection ratio, service cache decision and UAV trajectory, so as to minimize the total energy consumption of the multi-layer aerial network system model.

[0013] Beneficial effects: The present invention establishes a multi-layer aerial network system model for integrating wireless devices, drones and high-altitude platforms. Based on the multi-layer aerial network system model, a drone-wireless device perception model, a wireless device-drone-high-altitude platform communication model and a drone-high-altitude platform cache model are respectively established; and a system delay calculation model and a system energy consumption calculation model are correspondingly established. Finally, an optimization model for jointly optimizing the time scheduling ratio, task offloading ratio, compression selection ratio, service cache decision and UAV trajectory is established to minimize the total energy consumption of the multi-layer aerial network system model; by solving the optimization model through multiple iterations, the optimal time scheduling ratio, task offloading ratio, compression selection ratio, service cache decision and UAV trajectory are obtained to minimize the total energy consumption of the multi-layer aerial network system model. The present invention integrates perception, communication, computing and caching into a unified system model through an innovative multi-layer aerial architecture, fundamentally reimagining network functions, solving the problems limited by traditional ground networks, being more flexible, improving service efficiency while reducing latency and energy consumption. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following is a brief introduction to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0015] Figure 1 This is a flow chart of a multi-layer air network energy consumption optimization method based on the sensory communication computing and storage mechanism in the present invention;

[0016] Figure 2 This is a diagram of a multi-layer aerial system model architecture that integrates wireless devices, drones, and high-altitude platforms in an embodiment of the present invention;

[0017] Figure 3 Schematic diagram of a layered time division multiple access scheme for WD-UAV link sensing and offloading in an embodiment of the present invention;

[0018] Figure 4 Optimized trajectory diagram of the UAV at different HAP horizontal positions according to an embodiment of the present invention;

[0019] Figure 5 Optimized trajectory diagram of the UAV under different flight periods T in an embodiment of the present invention;

[0020] Figure 6 The trajectory planning diagram of the UAV under different mission data volume requirements in the embodiment of the present invention;

[0021] Figure 7 Graph showing the speed change of the UAV under different mission data volume requirements in an embodiment of the present invention;

[0022] Figure 8 Graphs showing convergence characteristics of optimization algorithms under different system configurations in an embodiment of the present invention;

[0023] Figure 9 This is a comparison chart of the impact of UAV buffer capacity on total energy consumption under different comparison schemes in an embodiment of the present invention;

[0024] Figure 10 This is a comparison chart of total energy consumption changes of different comparison schemes at WD scale in the embodiments of the present invention. DETAILED DESCRIPTION

[0025] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0026] Airborne platforms, including high-altitude platforms (HAPs) and unmanned aerial vehicles (UAVs), offer promising solutions for the rapid and flexible deployment of the Internet of Things. HAPs operate quasi-statically in the stratosphere, providing a service range spanning hundreds of kilometers, while UAVs maneuver at altitudes of hundreds of meters, rapidly adjusting their positions to provide on-demand localized services. The combination of these two forms a multi-layered airborne network. Through collaboration and resource sharing among platforms at different altitudes, this network overcomes the limitations of a single platform in terms of coverage, maneuverability, and resource capacity, while also enabling differentiated service deployment based on the strengths of each platform.

[0027] Based on these considerations, this embodiment proposes an innovative multi-layered aerial architecture that fundamentally reimagines network functionality by integrating perception, communication, computing, and caching into a unified system model. Multifunctional drones collaborate with HAP data centers to support resource-constrained ground-based WDs. Drones gain environmental awareness through radar perception, providing information for trajectory optimization and improving service efficiency. Computational tasks are adaptively partitioned between local WD processing and drone execution, and strategic data compression and service caching mechanisms are employed to further reduce latency and energy consumption.

[0028] This embodiment provides a multi-layer air network energy consumption optimization method based on the sensor communication computing storage mechanism, such as Figure 1 As shown, the specific steps include:

[0029] S1: Build a multi-layer aerial network system model for integrating wireless devices, drones, and high-altitude platforms;

[0030] S2: Based on the multi-layer aerial network system model, a drone-wireless device perception model, a wireless device-drone-high-altitude platform communication model, and a drone-high-altitude platform cache model are respectively established;

[0031] S3: Establishing a system delay calculation model based on the wireless device-drone-high-altitude platform communication model and the drone-high-altitude platform cache model;

[0032] S4: Establishing a system energy consumption calculation model based on the system delay calculation model;

[0033] S5: Based on the multi-layer aerial network system model, the UAV-wireless device perception model, the wireless device-UAV-high-altitude platform communication model, the UAV-high-altitude platform cache model, the system delay calculation model, and the system energy consumption calculation model, an optimization model is established for jointly optimizing the time scheduling ratio, the task offloading ratio, the compression selection ratio, the service cache decision, and the UAV trajectory to minimize the total energy consumption of the multi-layer aerial network system model;

[0034] S6: Decompose the optimization model into several sub-models, and iteratively solve the sub-models using corresponding algorithms to obtain the optimal time scheduling ratio, task offloading ratio, compression selection ratio, service cache decision and UAV trajectory, so as to minimize the total energy consumption of the multi-layer aerial network system model.

[0035] In a specific embodiment, in S1, as Figure 2 As shown in Figure 1, the specific steps for building a multi-layer aerial system model integrating wireless devices, drones, and high-altitude platforms include:

[0036] Establish a three-dimensional Cartesian coordinate system, and in the three-dimensional Cartesian coordinate system,

[0037] Establish the ground layer: It consists of K wireless devices (WDs) and uses the set express;

[0038] Specifically, the wireless device is both a sensed target and an end user with computing needs, and can periodically generate computing-intensive tasks that require corresponding service programs to execute.

[0039] Establishing a low-altitude layer: deploying a multi-functional UAV that integrates sensing, communication, computing, and caching capabilities;

[0040] Establishing a high-altitude layer: using HAP as a central data center to store all service programs;

[0041] The process of setting up a multi-layer air network system model includes:

[0042] 1) Perception and task initialization phase: The UAV performs real-time radar perception to monitor ground WDs, and each WD uses a dynamic task division method to determine the proportion of tasks offloaded to the UAV.

[0043] 2) Preprocessing and unloading stage: Each WD compresses the task data it unloads to the UAV according to the partial compression mechanism, and then transmits the compressed data to the UAV.

[0044] 3) UAV task processing stage: The UAV decompresses the received task data and checks whether the required service program exists in its local cache. If so, it executes the task immediately. If not, it retrieves the necessary service program from the HAP and downloads it before executing the task.

[0045] In a specific embodiment, the specific steps of establishing a multi-layer aerial system model integrating wireless devices, drones, and high-altitude platforms further include:

[0046] Establish a three-dimensional Cartesian coordinate system and set the HAP static hovering height to h H , the altitude of UAV dynamic flight is h U The period of the multi-layer aerial system model is T, and the horizontal coordinates of WDk and HAP are represented by w k and w H , w k =(x k ,y k ) T , w H =(x H ,y H ) T ,

[0047] Discretize the period T into N time slots of equal duration δ, δ = T / N, where The value of δ is set small enough to ensure that the UAV position remains constant within each time slot and can be updated when the time slot transition occurs.

[0048] Assume that UAV can move from the initial position to the Fly to final position Given the maximum flight speed V max , set the UAV's dynamic constraints as:

[0049]

[0050] Among them, q U [n] is the discrete time position of the UAV, which is used to characterize the horizontal trajectory of the UAV, and q U [n]=(x U [n],y U [n]) T , v[n] represents the flight speed of the UAV in time slot n;

[0051] At time slot n, the three-dimensional Euclidean distances between WDk and UAV, and between UAV and HAP are respectively expressed as:

[0052]

[0053] In a specific embodiment, in S2, the UAV performs active radar sensing by emitting electromagnetic waves and analyzing their reflection from the WD, obtaining key parameters such as operating status, spatial coordinates, and environmental conditions, facilitating subsequent communication, computing, and caching services. The specific steps for establishing a UAV-wireless device perception model based on the multi-layer air system model include:

[0054] This embodiment uses a radar model to model the UAV's perception process, where the amount of perception data collected by the UAV in time slot n is assumed to be:

[0055]

[0056] in, is a constant related to data redundancy, v U Indicates the switching speed of the radar beam, N θ Indicates the quantized angle, f s [n] is the sampling rate of the radar model at time slot n, is the number of quantization bits for each collected sample;

[0057] According to electromagnetic propagation theory, assuming that the Doppler shift caused by the UAV motion is constant during the radar pulse repetition period, allowing complete Doppler compensation, the two-way channel gain between the UAV and WDk at time slot n is expressed as:

[0058]

[0059] in, and are the UAV transmitting and receiving antenna gains, respectively, λ is the radar signal wavelength, λ=c / f c , c is the speed of light, f c is the carrier frequency, σ rcs is the radar cross section;

[0060] To evaluate UAVs' ability to perceive WDs, the radar's mutual information rate (MIR) is used as a performance metric to characterize the efficiency of extracting target features from radar echoes. Compared to traditional metrics such as detection probability and positioning accuracy, MIR provides a unified measure of perception capability from an information-theoretic perspective and is more suitable for integrated synaesthesia scenarios.

[0061] At time slot n, the radar mutual information rate between UAV and WDk is set to:

[0062]

[0063] Among them, B U is the WD-UAV communication bandwidth, represents the perceived transmit power of the UAV, represents the receiver noise power of the UAV;

[0064] In order to maintain the perceptual quality, the total amount of information collected by the UAV from all WDs in each time slot is set to meet a predetermined threshold, which is expressed as:

[0065]

[0066] In this embodiment, the WD-UAV and HAP-UAV links operate in orthogonal frequency bands to avoid cross-layer interference. The WD-UAV link supports perception and communication, and the HAP-UAV link supports service program downloading. Figure 3 The WD-UAV link adopts a hierarchical time division multiple access scheme, where each time slot δ is divided into K equal sub-slots, and the duration of each sub-slot is δ / K. In each sub-slot assigned to a WD, the scheduling ratio is set to ρ k [n], the scheduling ratio is used to control the time allocated to sensing, ρ k [n]∈[0,1], the remaining time is used for task offloading.

[0067] Specifically, this embodiment uses a multifunctional drone equipped with radar, communication units, and edge servers to sense WDs and provide them with computing and caching services. WDs can flexibly divide task data and adaptively compress offloaded data to reduce transmission overhead.

[0068] In a specific embodiment, in S2, the specific steps of establishing a wireless device-drone-high-altitude platform communication model based on the multi-layer aerial system model include:

[0069] 1) Establishing the communication model for the WD-UAV link: Based on the free-space propagation model widely adopted in standards such as ITU-R P.1411, WD-UAV links typically maintain a line-of-sight (LoS) path in open space. Therefore, at time slot n, the channel power gain from WDk to UAV is set to:

[0070]

[0071] in, Indicates the WD transmitting antenna gain;

[0072] At time slot n, the achievable data rate from WDk to UAV is set to:

[0073]

[0074] Among them, pk is the transmission power of WDk;

[0075] 2) Establish the communication model of the HAP-UAV link: Unlike the WD-UAV link, the HAP-UAV channel experiences large-scale path loss and small-scale fading due to long-range stratospheric propagation. At time slot n, the channel power gain from UAV to HAP is set to:

[0076]

[0077] in, represents the transmitting antenna gain of HAP, L atm represents the atmospheric attenuation coefficient;

[0078] Small-scale fading coefficient Obey the Rice distribution, expressed as:

[0079]

[0080] Among them, K c is a coefficient representing the power ratio between the LoS and NLoS components;

[0081] At time slot n, the achievable data rate from HAP to UAV is:

[0082]

[0083] Among them, B H represents the bandwidth of HAP-UAV communication, p H Indicates the transmit power of the HAP.

[0084] In a specific embodiment, in S2, the specific steps of establishing the UAV-high-altitude platform cache model based on the multi-layer aerial system model include:

[0085] set up is a finite set of service programs, Among them, the size of program i is o i (in bits), The cache capacity of HAP is 0 H , capable of storing a complete set of service programs to meet In contrast, UAVs have limited cache capacity. U , where it is necessary to selectively cache program i according to the task data received by WD.

[0086] Define the binary decision variable χ i [n], used to represent the cache status of program i in time slot n, χ i [n]∈{0,1} where if program i is cached in UAV, then χi [n]=1; otherwise, χ i [n] = 0, set the cache capacity constraint to:

[0087]

[0088] In a specific embodiment, in S3, the specific steps of establishing a system delay calculation model based on the wireless device-drone-high-altitude platform communication model and the drone-high-altitude platform cache model include:

[0089] Assume that the service program corresponding to the task received by WDk in time slot n is program i, and the task of WDk is expressed as in Indicates the amount of input data, and Represents the CPU cycles required to calculate and compress each bit of task data;

[0090] Define the unloading ratio as Each task is based on the offloading ratio Part of it is offloaded to the UAV for processing, and the rest is executed in the local WD;

[0091] Since the offloaded tasks usually contain compressible data (e.g., text, images) and incompressible data (e.g., encrypted content), if unified compression is inefficient, in order to reduce communication overhead, this embodiment sets a partial compression mechanism, that is, before transmitting to the UAV, the WD only compresses a portion of the offloaded data. The compression process is characterized by two parameters: compression selection ratio η and compression selection ratio η. k [n], used to indicate the portion of unloaded data to be compressed, η k [n]∈[0,1]; and a fixed compression ratio β, used to represent the amount of data reduction after compression, β∈(0,1). The amount of data unloaded after compression becomes

[0092] 1) Establish a local computing delay model, that is, the local computing delay for each WD: Let f k Denotes the CPU frequency of WDk, then the local computation delay at time slot n is expressed as:

[0093]

[0094] Since the tasks generated by the WD have a predictable periodic pattern, the required service programs are pre-installed on the WD, eliminating the need to download the programs from the HAP.

[0095] The compressed delay of WDk at time slot n is expressed as:

[0096]

[0097] Therefore, the transmission delay of unloading compressed data from WDk to UAV at time slot n is:

[0098]

[0099] in, Time scheduling constraints need to be met:

[0100]

[0101] 2) Establishing UAV edge computing delay model: After receiving the unloaded data from WDk, UAV first decompresses the task data. Based on the previous compression operation, the amount of data that needs to be decompressed is expressed as Assume that the CPU frequency assigned to WDk by UAV is f k,U , then the corresponding decompression delay is expressed as:

[0102]

[0103] Among them, J U [n] represents the CPU cycles required to decompress each bit of data;

[0104] Subsequently, the UAV checks whether the required service program i exists in its cache. If so, the UAV directly executes the task. At this time, the computational delay of the task data unloaded by the UAV executing WDk in time slot n is:

[0105]

[0106] If the required service program does not exist on the UAV, the UAV needs to download the corresponding service program from the HAP before executing the task. In time slot n, the transmission delay from the HAP to the UAV is:

[0107]

[0108] Due to common computing tasks, the output data volume is usually several orders of magnitude smaller than the input data. At the same time, due to the HAP's higher transmission power and better antenna configuration, the HAP's downlink capacity significantly exceeds the uplink capacity. Therefore, if there is no transmission delay from HAP to WD, the task is completed. The total delay is expressed as:

[0109]

[0110] In a specific embodiment, in S4, the specific steps of establishing a system energy consumption calculation model based on the system delay calculation model include:

[0111] 1) Establish WD energy consumption model: define κ1 as the effective capacitance coefficient of WD, then calculate the local Task compression and transmission energy consumption Respectively expressed as:

[0112]

[0113]

[0114] Therefore, the total energy consumption of WDk in time slot n is expressed as:

[0115]

[0116] 2) Establish UAV energy consumption model: At time slot n, the energy consumption of UAV’s perception of WDk is expressed as:

[0117]

[0118] At time slot n, the energy consumption of unloading task data from WDk is expressed as:

[0119]

[0120] Define κ2 as the effective capacitance coefficient of UAV, then in time slot n, the decompression energy consumption of UAV for WDk offloading task is expressed as:

[0121]

[0122] Based on the blade element momentum theory, the UAV propulsion power is expressed as:

[0123]

[0124] Among them, P1 and P2 represent the blade profile power and induced power in the hovering state respectively, U top is the rotor blade tip speed, v0 represents the average induced speed during hovering, d0 is the fuselage drag ratio, ρ0, s, and A represent the air density, rotor solidity, and rotor disk area, respectively;

[0125] The flight energy consumption of UAV in time slot n is expressed as:

[0126]

[0127] Therefore, the total energy consumption of the UAV in time slot n is expressed as:

[0128]

[0129] 3) Establishing a HAP energy consumption model: With stable solar power supply, the HAP is self-sufficient in terms of operating power. Therefore, this embodiment only focuses on its transmission energy consumption. At time slot n, the transmission energy consumption of the HAP transmitting service program i is:

[0130]

[0131] In a specific embodiment, in S5, the optimization model is expressed as:

[0132]

[0133] in, is the time scheduling ratio, Ω is the task offloading ratio, Λ is the compression selection ratio, X is the service cache decision, Q is the UAV trajectory, ω is a weight factor; C1 is a constraint for enforcing the completion time of each task; C2 is a constraint for ensuring that the cached service program does not exceed the storage capacity of the UAV; C3 is a constraint for guaranteeing the minimum amount of required sensory data; C4 is a constraint for limiting the offloading delay to the allocated duration of each sub-time slot; C5 represents the binary cache decision constraint; C6-C8 are constraints for limiting the time scheduling, task offloading, and compression selection ratio to within [0,1]; C9 is a constraint for specifying the initial and final positions of the UAV; C10 is a constraint for limiting the displacement of the UAV between adjacent time slots; and C11 is a constraint for limiting the maximum speed of the UAV.

[0134] Specifically, the optimization model is formulated as a mixed integer nonlinear programming (MINLP), subject to practical constraints such as task completion delay, UAV storage capacity, and perception quality. This embodiment solves the MINLP problem by designing a low-complexity iterative optimization algorithm.

[0135] In a specific embodiment, the joint optimization model is transformed into It is decomposed into several sub-models, including time scheduling ratio optimization model, task offloading ratio optimization model, compression selection ratio optimization model, service cache decision optimization model and UAV trajectory optimization model.

[0136] In this embodiment, each sub-model is optimized sequentially until convergence, and each sub-model optimizes a specific subset of variables, while the remaining variables are fixed. The solution method is selectively applied according to the mathematical properties of each sub-model. The interior point method is used to solve the time scheduling ratio optimization model for time scheduling ratio optimization, the CVX toolbox is used to solve the task offloading ratio optimization model for task offloading ratio optimization, the interior point method is used to solve the compression selection ratio optimization model for compression selection ratio optimization, the Lagrangian dual algorithm is used to solve the service cache decision optimization model for optimizing the service cache decision, and the SCA algorithm is used to solve the UAV trajectory optimization model for optimizing the UAV trajectory. The proposed algorithm ensures convergence within a finite number of iterations and obtains near-optimal performance. The specific steps include:

[0137] For a given task offloading ratio Ω, compression selection ratio Λ, service cache decision X, and UAV trajectory Q, the time scheduling ratio optimization model is defined as:

[0138]

[0139] Specifically, This is a standard linear programming problem. This embodiment uses a mature linear programming method (such as the simplex method or the interior point method) to solve it efficiently.

[0140] For a given compression selection ratio Λ, service cache decision X, UAV trajectory Q, and optimized time scheduling ratio The task offloading ratio optimization model is defined as:

[0141]

[0142] Specifically, This is a standard convex optimization problem. This embodiment uses a mature convex optimization toolbox such as CVX to solve it efficiently.

[0143] For a given service cache decision X, UAV trajectory Q, and optimized time scheduling ratio and the task offloading ratio Ω, the compression selection ratio optimization model is defined as:

[0144]

[0145] Specifically, This is a standard linear programming problem. This embodiment uses a mature convex optimization toolbox such as CVX to solve it efficiently.

[0146] For a given UAV trajectory Q, and the optimized time scheduling ratio The task offloading ratio Ω and the compression selection ratio Λ define the service cache decision optimization model as:

[0147]

[0148] Specifically, It is a binary integer linear programming (BILP) problem. Due to the discrete nature of the decision variables, It becomes an NP-hard problem. Although the branch-and-bound method can guarantee global optimality, its computational complexity increases exponentially with the size of the problem. In order to strike a balance between solution quality and computational efficiency, The problem is processed as follows. First, since the max operation of C1 leads to a non-differentiable Lagrangian function, this embodiment introduces an auxiliary variable η k [n]Re-state C1 as:

[0149]

[0150] in,

[0151] Secondly, χ i The binary constraint of [n] is relaxed to therefore, Redefined as:

[0152]

[0153] After the above steps, It becomes a continuous optimization problem with linear constraints. The Lagrange dual method can be used to solve the problem. The Lagrange multipliers υ={υ k,n ≥0}, λ={λ k,n ≥0}, γ={γ k,n ≥0} and μ={μ n ≥0}, The Lagrangian function is defined as:

[0154]

[0155] in,

[0156]

[0157] Define the dual function g(v,λ,γ,μ) as the Lagrangian function Relative to the original variable in its feasible region The lower bound of is expressed as:

[0158]

[0159] The dual problem is to seek the maximization of the dual function with respect to the Lagrange multiplier, which can be expressed as:

[0160]

[0161] Specifically, the Lagrange dual algorithm is used to The solution consists of the following iterative steps:

[0162] 1) Primal Variable Update: The Lagrangian function of variables and The part of has linear characteristics, so there is no need to find the zero point by explicitly solving the partial derivatives. Combined with the constraint range of boundary conditions, the optimal solution can be obtained by direct comparison To determine. Because the capacity constraint is linear, the optimal solution can be quickly found using a sorting method combined with a greedy allocation strategy. This approach exploits the structural characteristics of the problem, significantly simplifying the computational complexity while ensuring a globally optimal solution.

[0163] Given the Lagrange multiplier υ k,n ,λ k,n , γ k,n and μ n , the Lagrangian function can be decomposed into two parts, Related parts and The relevant part, The relevant sub-questions are:

[0164]

[0165] By directly analyzing the objective function Coefficient -υ k,n -λ k,n +γ k,n , the optimal solution is:

[0166]

[0167] and The relevant sub-questions are:

[0168]

[0169] The objective function can be organized as Combined with capacity constraints This is a linear programming problem with linear constraints, so it can be solved efficiently by the following greedy algorithms, including:

[0170] First, follow Sort the service program i in non-increasing order. Then, assign Until capacity limit O U Finally, if the remaining capacity is insufficient to fully allocate program i, a partial value is allocated according to the remaining capacity, i.e.

[0171] 2) Lagrange Multiplier Update: Since the dual function is usually non-differentiable at the maximum point, traditional gradient methods cannot be directly applied. Subgradient methods can efficiently handle non-smooth concave function optimization problems by utilizing the subgradient information of the dual function. Therefore, at the rth iteration, the update of the multiplier rule is:

[0172]

[0173] Where [x] + =max{0,x} represents a non-negative orthogonal projection operation, τ (r) is the step size of the rth iteration,

[0174] 3) Algorithm Stopping Criterion: Due to the original problem Convexity and Slater conditions are satisfied, and strong duality ensures that the duality gap gradually converges to zero. Therefore, we can define:

[0175]

[0176] in, is the duality gap at iteration r, g(υ (r) ,λ (r) ,γ (r) ,μ (r) ) is the dual function value of the rth iteration, is the original problem objective function value of the rth iteration, ∈1 is the preset tolerance threshold, ∈1>0. In addition, the multiplier change is defined as To measure the convergence of the Lagrange multiplier, it is expressed as:

[0177]

[0178] Among them, ‖·‖ represents the 2-norm of the vector, ∈2 is the set threshold, ∈2>0;

[0179] Set the maximum number of iterations r max , when r≥r max When , the algorithm is forced to terminate and the current optimal solution is output.

[0180] 4) Binary Variables Recover: To recover binary variables, firstly based on the relaxation solution And use the thresholding rule for initial mapping, namely:

[0181]

[0182] If the initial recovery solution exceeds the capacity limit, that is, The following steps are used to correct the problem: First, according to the unit benefit Variables are sorted from lowest to highest. Then, starting with the variable with the lowest unit benefit, they are reset to zero one by one until the capacity constraint is satisfied. This method combines initial recovery with subsequent corrections to ensure the feasibility of the solution and the optimization of the objective function.

[0183] For a given optimized time scheduling ratio With the task offloading ratio Ω, compression selection ratio Λ and service cache decision X, the UAV trajectory optimization model is defined as:

[0184]

[0185] Specifically, This is a non-convex optimization problem. This embodiment uses the successive convex approximation (SCA) technique to solve the problem. The basic idea is to perform a local convex approximation on the non-convex objective function and constraints in each iteration. The problem is transformed into a series of convex subproblems. First, deal with the non-convexity in C1, C3, and C4. Let Rewrite (10) as:

[0186]

[0187] At the local point of the tth iteration right Perform a first-order Taylor expansion and obtain the lower bound expression as follows:

[0188]

[0189] in,

[0190]

[0191] By using replace Convert C4 into a convex constraint. Similarly, let Rewrite equation (7) as:

[0192]

[0193] At the local point of the tth iteration right Perform a first-order Taylor expansion and obtain the lower bound expression as follows:

[0194]

[0195] in,

[0196]

[0197]

[0198] On this basis, formula (13) is rewritten as:

[0199]

[0200] in,

[0201] At the local point of the tth iteration R H,U [n] performs a first-order Taylor expansion and obtains the lower bound expression as:

[0202]

[0203] in,

[0204]

[0205] By using replace replace Instead of R H,U [n], C1, C3, and C4 are all converted into convex constraints.

[0206] Based on the non-convexity problem in the objective function, it is known that the non-convexity is currently only caused by e fiy The last term of [m] is caused by the equation (30). In order to solve this problem, two auxiliary variables are introduced: and To relax this term, we get:

[0207]

[0208] The expression in formula (65) can be further equivalently converted to:

[0209]

[0210] Although constraint (66) is non-convex, its right-hand side is about and Similarly, through the SCA technology, Equation (66) is approximately converted into the following convex form:

[0211]

[0212] in, and is a given arbitrary local point. Therefore, Equation (31) is finally transformed into the following convex expression:

[0213]

[0214] So far, it has been completely solved Contains all non-convexities of the objective function and constraints. Let Denotes the set of auxiliary variables introduced. By substituting the derived lower bound function and the approximate convex expression into This can be reformulated as:

[0215]

[0216] in,

[0217] It can be seen that This is a convex optimization problem, and this embodiment uses the CVX tool to solve it.

[0218] For comprehensive performance evaluation, the proposed method in this example is compared with four baselines, including:

[0219] Fixed Trajectory (FT): The UAV flies along a preset straight path from an initial position to a final position at a constant speed, and other variables are optimized using the BCD algorithm.

[0220] No Compression (NC): No data compression is performed on the compute task when it is offloaded, and other variables are optimized using the BCD algorithm.

[0221] No Cache (NCH): The UAV does not have local cache capabilities. All service programs need to be obtained from the HAP. Other variables are optimized using the BCD algorithm.

[0222] Local Execution (LE): The UAV does not have edge computing capabilities. All computing tasks are executed locally on the WD, and other variables are optimized using the BCD algorithm.

[0223] Fixed TDMA (FTDMA): The UAV adopts a fixed time scheduling strategy, allocating equal sensing and offloading durations to each WD in each time slot, and other variables are optimized using the BCD algorithm.

[0224] like Figure 4 As shown in the figure, in this simulation scenario, the UAV's initial and final positions are fixed at coordinates (0, 0) and (400m, 400m), respectively. Three different HAP deployment scenarios are considered: a central location (200m, 200m), an upward location (200m, 300m), and a downward location (200m, 100m). When the HAP is located at the center of the area, the UAV's trajectory exhibits a relatively symmetrical "S"-shaped curve. This trajectory design ensures uniform coverage of ground WDs while maintaining stable communication link quality with the HAP. However, as the HAP position shifts, the UAV's optimal trajectory shifts accordingly, tending to converge toward the HAP. This adaptive behavior reflects the trade-off between HAP-UAV communication quality during the optimization process of the multi-layer aerial network system. Specifically, when the HAP position shifts upward, the optimized trajectory bends upward to reduce the average distance to the HAP. Similarly, when the HAP position shifts downward, the trajectory shifts downward. Notably, although the trajectory shifts with the HAP's position, the UAV never completely strays from its service area. This demonstrates that the method proposed in this embodiment maintains service quality for WDs while pursuing HAP-UAV communication efficiency. Curved portions of the trajectory typically correspond to brief UAV stays in areas with high WD density, which helps provide better computing services and perception coverage. Through this flexible trajectory planning, the multi-layer aerial network system achieves a multi-dimensional balance between computing efficiency, communication quality, perception performance, and energy efficiency.

[0225] like Figure 5As shown, the HAP is fixed at the center (400m, 200m), and the three WDs (WD1, WD2, and WD3) are distributed in a triangular pattern. This example examines five different flight cycle settings, requiring the UAV to fly from (0, 200m) to (800m, 200m). Trajectory evolution trends show that when the flight cycle is short (T = 45s), the UAV's trajectory has the lowest peak and the flattest arc, indicating that under strict time constraints, the UAV is forced to adopt a relatively conservative coverage strategy. As the cycle lengthens from T = 45s to T = 60s, the UAV's "double-peaked" trajectory exhibits a clear, gradual change: the peak gradually increases and the coverage area continues to expand. This is because the longer flight cycle provides the UAV with more maneuvering time, enabling it to better approach each WD and thus provide better service. However, when the cycle is further increased to T = 65s, the trajectory characteristics almost overlap with those of the T = 60s case, indicating that the system performance improvement has reached saturation. The emergence of this marginal effect has profound physical significance: it suggests that, given a given system configuration, there exists an optimal point for allocating time resources. Beyond this point, the performance gains from further increasing flight cycles become negligible. Notably, all trajectories decrease in altitude as they approach WD2, forming a "valley" in the middle of the trajectory. This behavior reflects a key system trade-off: while reaching the valley increases the distance between the UAV and the HAP, reducing the service program download rate, this trade-off results in a shorter communication distance with WD2, significantly improving offloading and sensing efficiency.

[0226] like Figure 6 and Figure 7Figure 2 shows the horizontal trajectories of UAVs under four representative load distribution scenarios and the corresponding UAV speed changes. In single-user demand scenarios such as (0, 0, 8) Mbits and (8, 0, 0) Mbits, the UAV exhibits a basic "approach-stay-leave" behavior pattern. Specifically, when approaching a WD with computational needs, the UAV significantly reduces its flight speed and provides service at an appropriate distance; when passing through areas without computational needs, it maintains a high speed and passes quickly. In dual-user demand scenarios such as (8, 8, 0) Mbits and (0, 8, 8) Mbits, the UAV adopts a "serial service" mode, sequentially visiting WDs with computational needs. These trajectory characteristics stem from the nature of the optimization problem: a multi-layer aerial network system must meet the constraints of perceived quality and task completion time while minimizing total energy consumption. Specifically, the UAV approaches the WD with computational needs to provide the necessary communication and perception quality, but maintains an appropriate distance from the WD to balance energy consumption. Furthermore, each UAV position adjustment must consider the quality of communication with the regional HAP to ensure efficient access to required service programs in the event of a cache miss. Simultaneously, overall trajectory planning must ensure that all perception and computational service tasks are completed within a limited period. This is reflected in the fact that all UAVs following these trajectories ultimately reach their destination on time. This differentiated trajectory planning strategy not only demonstrates the multi-layered aerial network system's adaptive responsiveness to spatial task distribution but also demonstrates the efficiency of the proposed method in spatiotemporal resource scheduling.

[0227] like Figure 8As shown in Figure 2, three scenarios were evaluated: 5 WDs with 3MB tasks, 5 WDs with 6MB tasks, and 10 WDs with 6MB tasks. The results show that these scenarios exhibit significantly different convergence patterns and energy consumption characteristics. For the baseline scenario of 5 WDs with 3MB tasks, the method proposed in this embodiment exhibits rapid convergence, with total energy consumption dropping from 100.45J to approximately 40.50J, and reaching a steady state in only 6-7 iterations. When the task size increases to 6MB while the number of WDs remains unchanged, the convergence speed slows down moderately, requiring 8-9 iterations to stabilize, and the final energy consumption stabilizes at 71.20J, a 75% increase compared to the baseline scenario. This proportional increase is consistent with the energy consumption overhead brought about by the larger data transmission and processing requirements. In the most challenging scenario (10 WDs, 6MB tasks), convergence requires 11-12 iterations, and the final energy consumption stabilizes at 121.20J. Compared to the five-WD scenario with the same task size, energy consumption increased by 70%, reflecting the additional overhead of coordinating more WDs, including increased perception operations, communication interactions, and trajectory optimization complexity. Notably, across all scenarios, the algorithm maintained a consistent convergence pattern: rapid initial improvement followed by progressive optimization, and minimal fluctuations in the final stage, fully demonstrating the robustness of the proposed method.

[0228] like Figure 9As shown, the method proposed in this embodiment achieves optimal energy efficiency under various cache capacity configurations and exhibits a significant downward trend with increasing cache capacity, with a reduction of up to 28%. This is due to the method proposed in this embodiment's ability to fully utilize the increased cache space and reduce system overhead by dynamically adjusting cache decisions, computation offloading, compression selection, and trajectory planning. Specifically, the larger cache capacity enables the UAV to store more service programs locally, reducing frequent data exchanges with the HAP and significantly reducing communication energy consumption on the backhaul link. Furthermore, the improved local cache hit rate shortens task processing latency, enabling the UAV to adopt more flexible flight trajectories to optimize communication with the WD and sensor quality. The FTDMA scheme exhibits a similar, but less pronounced, downward trend due to its fixed time allocation strategy, which results in suboptimal system resource utilization. The NC scheme, lacking a data compression mechanism, incurs significant transmission overhead under heavy communication loads, maintaining high total energy consumption. The NCH scheme, on the other hand, is completely insensitive to changes in cache capacity because all service programs must be retrieved from the HAP in real time. While the FT solution also shows a decrease in energy consumption as cache capacity increases, its fixed trajectory limits the system's adaptability and prevents it from fully leveraging the benefits of cache expansion. Finally, the LE solution maintains the highest energy consumption and does not change with cache capacity, reflecting the inherent limitations of pure local execution strategies for computationally intensive tasks.

[0229] like Figure 10As shown, the method proposed in this embodiment demonstrates optimal performance, minimizing energy consumption and achieving the flattest growth curve. While FTDMA is more conservative in resource scheduling, it ensures basic service fairness, resulting in relatively stable, near-linear growth. When the number of WDs is less than eight, the energy consumption differences among the various schemes are relatively modest, and the growth trends are relatively consistent. This is because under light load conditions, system resources are relatively abundant, and the differences between the various optimization strategies are not fully apparent. However, when the number of WDs exceeds eight, the performance of the different schemes begins to diverge significantly. Particularly noteworthy is the crossover between the NC and FT schemes: when the number of WDs approaches nine, the NC scheme's energy consumption exceeds that of the FT scheme. This shift reflects the critical role of data compression in high-load scenarios: while the FT scheme is limited by fixed trajectories and cannot achieve optimal spatial coverage, its compression mechanism effectively reduces the transmission load. In contrast, the NC scheme requires transmitting larger volumes of raw data as the WD density increases, resulting in a rapid accumulation of communication burden, ultimately overcoming the performance loss associated with fixed trajectories. The NCH solution performs even worse, reflecting the importance of caching mechanisms in multi-user scenarios: frequent program downloads not only directly increase communication overhead but also indirectly impact task processing efficiency. Finally, the LE solution exhibits the most significant performance degradation as the user scale increases. This is due to the cumulative effect of processing all computationally intensive tasks locally on the resource-constrained WD. At a WD of 14, its energy consumption is nearly 2.8 times that of the proposed solution.

[0230] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A multi-layer air network energy consumption optimization method based on the sensory communication computing and storage mechanism, characterized in that: The specific steps include: S1: Build a multi-layer aerial network system model for integrating wireless devices, drones, and high-altitude platforms; S2: Based on the multi-layer aerial network system model, a drone-wireless device perception model, a wireless device-drone-high-altitude platform communication model, and a drone-high-altitude platform cache model are respectively established; S3: Establishing a system delay calculation model based on the wireless device-drone-high-altitude platform communication model and the drone-high-altitude platform cache model; S4: Establishing a system energy consumption calculation model based on the system delay calculation model; S5: Based on the multi-layer aerial network system model, the UAV-wireless device perception model, the wireless device-UAV-high-altitude platform communication model, the UAV-high-altitude platform cache model, the system delay calculation model, and the system energy consumption calculation model, an optimization model is established for jointly optimizing the time scheduling ratio, the task offloading ratio, the compression selection ratio, the service cache decision, and the UAV trajectory to minimize the total energy consumption of the multi-layer aerial network system model; S6: Decompose the optimization model into several sub-models, and iteratively solve the sub-models using corresponding algorithms to obtain the optimal time scheduling ratio, task offloading ratio, compression selection ratio, service cache decision and UAV trajectory, so as to minimize the total energy consumption of the multi-layer aerial network system model.

2. The multi-layer air network energy consumption optimization method based on the sensory communication computing and storage mechanism according to claim 1 is characterized in that: In S1, the specific steps for building a multi-layer aerial system model integrating wireless devices, drones, and high-altitude platforms include: Establish a three-dimensional Cartesian coordinate system, and in the three-dimensional Cartesian coordinate system, Establish the ground layer: composed of K WDs, using the set express; Establishing a low-altitude layer: deploying a UAV with integrated sensing, communication, computing, and caching capabilities; Establishing a high-altitude layer: using HAP as a central data center to store all service programs; The process of setting up a multi-layer air network system model includes: 1) Perception and task initialization phase: The UAV performs real-time radar perception to monitor wireless devices, and each WD determines the proportion of tasks to be offloaded to the UAV; 2) Preprocessing and offloading phase: Each WD compresses the task data it offloads to the UAV according to the partial compression mechanism and transmits it to the UAV; 3) UAV task processing stage: The UAV decompresses the received task data and checks whether the required service program exists in its local cache. If so, it executes the task immediately. If not, it obtains the corresponding service program from the HAP and then executes the task.

3. The multi-layer air network energy consumption optimization method based on the sensory communication computing and storage mechanism according to claim 2 is characterized in that: In S1, the specific steps for building a multi-layer aerial system model integrating wireless devices, drones, and high-altitude platforms include: Set the HAP static hover height to h H , the altitude of UAV dynamic flight is h U The period of the multi-layer aerial system model is T, and the horizontal coordinates of WDk and HAP are represented by w k and w h , w k =(x k ,y k ) T , w H =(x H ,y H ) T , Discretize the period T into N time slots of equal duration δ, δ = T / N, where Assume that UAV can move from the initial position to the Fly to final position Given the maximum flight speed V max , set the UAV's dynamic constraints as: Among them, q U [n] is the discrete time position of the UAV, which is used to characterize the horizontal trajectory of the UAV, and q U [n]=(x U [n],y U [n]) T , v[n] represents the flight speed of the UAV in time slot n; At time slot n, the three-dimensional Euclidean distances between WDk and UAV, and between UAV and HAP are respectively expressed as:

4. The multi-layer air network energy consumption optimization method based on the sensory communication computing and storage mechanism according to claim 3 is characterized in that: In S2, the specific steps of establishing the drone-wireless device perception model based on the multi-layer air system model include: The radar model is used to model the UAV’s perception process, where the amount of perception data collected by the UAV in time slot n is expressed as: in, is a constant related to data redundancy, v U Indicates the switching speed of the radar beam, N θ Indicates the quantized angle, f s [n] is the sampling rate of the radar model at time slot n, θ U is the number of quantization bits for each collected sample; According to electromagnetic propagation theory, assuming that the Doppler shift caused by the UAV motion is constant during the radar pulse repetition period, allowing complete Doppler compensation, the two-way channel gain between the UAV and WDk at time slot n is expressed as: in, and are the UAV transmitting and receiving antenna gains, respectively, λ is the radar signal wavelength, λ=c / f c , c is the speed of light, f c is the carrier frequency, σ rcs is the radar cross section; At time slot n, the radar mutual information rate between UAV and WDk is set to: Among them, B U is the WD-UAV communication bandwidth, represents the perceived transmit power of the UAV, represents the receiver noise power of the UAV; In order to maintain the perceptual quality, the total amount of information collected by the UAV from all WDs in each time slot is set to meet a predetermined threshold, which is expressed as:

5. The multi-layer air network energy consumption optimization method based on the sensory communication computing and storage mechanism according to claim 4 is characterized in that: In S2, the specific steps of establishing a wireless device-drone-high-altitude platform communication model based on the multi-layer aerial system model include: 1) Establish the communication model of the WD-UAV link: At time slot n, set the channel power gain from WDk to UAV to be: in, Indicates the WD transmitting antenna gain; At time slot n, the achievable data rate from WDk to UAV is set to: Among them, p k is the transmission power of WDk; 2) Establish the communication model of the HAP-UAV link: At time slot n, the channel power gain from UAV to HAP is set to: in, represents the transmitting antenna gain of HAP, L atm represents the atmospheric attenuation coefficient; Small-scale fading coefficient Obey the Rice distribution, expressed as: Among them, K c is a coefficient representing the power ratio between the LoS and NLoS components; At time slot n, the achievable data rate from HAP to UAV is: Among them, B H represents the bandwidth of HAP-UAV communication, p H Indicates the transmit power of the HAP.

6. The multi-layer air network energy consumption optimization method based on the sensory communication computing and storage mechanism according to claim 5 is characterized in that: In S2, the specific steps of establishing the UAV-high-altitude platform cache model based on the multi-layer aerial system model include: set up is a finite set of service programs, Among them, the size of program i is o i , The cache capacity of HAP is 0 H ,satisfy The UAV's cache capacity is O U ; Define the binary decision variable χ i [n], used to represent the cache status of program i in time slot n, χ i [n]∈{0,1} where if program i is cached in UAV, then χ i [n]=1; otherwise, χ i [n] = 0, set the cache capacity constraint to:

7. The multi-layer air network energy consumption optimization method based on the sensory communication computing and storage mechanism according to claim 6 is characterized in that: In S3, the specific steps of establishing a system delay calculation model based on the wireless device-drone-high-altitude platform communication model and the drone-high-altitude platform cache model include: Assume that the service program corresponding to the task received by WDk in time slot n is program i, and the task of WDk is expressed as in Indicates the amount of input data, and Represents the CPU cycles required to calculate and compress each bit of task data; Define the unloading ratio as Each task is based on the offloading ratio Part of it is offloaded to the UAV for processing, and the rest is executed in the local WD; Setting the partial compression mechanism includes: before transmitting to the UAV, the WD compresses only a portion of the unloaded data. The compression process is characterized by two parameters: compression selection ratio η k [n], used to indicate the portion of unloaded data to be compressed, η k [n]∈[0,1]; fixed compression ratio β, used to represent the amount of data reduction after compression, β∈(0,1), for the original size The amount of data unloaded after compression becomes 1) Establish a local computing delay model: Let f k Denotes the CPU frequency of WDk, then the local computation delay at time slot n is expressed as: The compressed delay of WDk at time slot n is expressed as: Therefore, the transmission delay of unloading compressed data from WDk to UAV at time slot n is: in, Time scheduling constraints need to be met: 2) Establishing UAV edge computing delay model: After receiving the unloaded data from WDk, UAV decompresses the task data. The amount of data that needs to be decompressed is expressed as Assume that the CPU frequency assigned to WDk by UAV is f k,U , then the corresponding decompression delay is expressed as: Among them, J U [n] represents the CPU cycles required to decompress each bit of data; The UAV checks whether the required service program i exists in its cache. If so, the UAV directly executes the task. The computational delay of the task data unloaded by the UAV executing WDk in time slot n is: If the required service program does not exist on the UAV, the UAV needs to download the corresponding service program from the HAP before executing the task. In time slot n, the transmission delay from the HAP to the UAV is: Will complete the task The total delay is expressed as:

8. The multi-layer air network energy consumption optimization method based on the sensory communication computing and storage mechanism according to claim 7 is characterized in that: In S4, the specific steps of establishing a system energy consumption calculation model based on the system delay calculation model include: 1) Establish WD energy consumption model: define κ1 as the effective capacitance coefficient of WD, then calculate the local Task compression and transmission energy consumption Respectively expressed as: The total energy consumption of WDk in time slot n is expressed as: 2) Establish UAV energy consumption model: At time slot n, the energy consumption of UAV’s perception of WDk is expressed as: At time slot n, the energy consumption of unloading task data from WDk is expressed as: Define κ2 as the effective capacitance coefficient of UAV, then in time slot n, the decompression energy consumption of UAV for WDk offloading task is expressed as: Based on the blade element momentum theory, the UAV propulsion power is expressed as: Among them, P1 and P2 represent the blade profile power and induced power in the hovering state respectively, U tip is the rotor blade tip speed, v0 represents the average induced speed during hovering, d0 is the fuselage drag ratio, ρ0, s, and A represent the air density, rotor solidity, and rotor disk area, respectively; The flight energy consumption of UAV in time slot n is expressed as: The total energy consumption of UAV in time slot n is expressed as: 3) Establishing HAP energy consumption model: At time slot n, the transmission energy consumption of HAP transmission service program i is expressed as:

9. The multi-layer air network energy consumption optimization method based on the sensory communication computing and storage mechanism according to claim 8 is characterized in that: In S5, the optimization model is expressed as: in, is the time scheduling ratio, Ω is the task offloading ratio, Λ is the compression selection ratio, X is the service cache decision, Q is the UAV trajectory, ω is a weight factor; C1 is a constraint for enforcing the completion time of each task; C2 is a constraint for ensuring that the cached service program does not exceed the storage capacity of the UAV; C3 is a constraint for guaranteeing the minimum amount of required sensory data; C4 is a constraint for limiting the offloading delay to the allocated duration of each sub-time slot; C5 represents the binary cache decision constraint; C6-C8 are constraints for limiting the time scheduling, task offloading, and compression selection ratio to within [0,1]; C9 is a constraint for specifying the initial and final positions of the UAV; C10 is a constraint for limiting the displacement of the UAV between adjacent time slots; and C11 is a constraint for limiting the maximum speed of the UAV.

10. The multi-layer air network energy consumption optimization method based on the sensory communication computing and storage mechanism according to claim 9 is characterized in that: The optimization model is transformed into It is decomposed into several sub-models, including time scheduling ratio optimization model, task offloading ratio optimization model, compression selection ratio optimization model, service cache decision optimization model and UAV trajectory optimization model; The interior point method is used to solve the time scheduling ratio optimization model for time scheduling ratio optimization, the CVX toolbox is used to solve the task offloading ratio optimization model for task offloading ratio optimization, the interior point method is used to solve the compression selection ratio optimization model for compression selection ratio optimization, the Lagrangian dual algorithm is used to solve the service cache decision optimization model for service cache decision optimization, and the SCA algorithm is used to solve the UAV trajectory optimization model for UAV trajectory optimization.

Images