Propulsion power consumption modeling and beam forming method for high-altitude platform

Abstract

The invention provides a propulsion power consumption modeling and beamforming method for a high-altitude platform. The method comprises the following steps: firstly, constructing a full-system power consumption model covering propulsion, communication and load; afterwards, an accurate propulsion power consumption correction model is established for aerodynamic interference between a boat body and a propeller by utilizing interactive generation type AI intelligent agent to cooperate with CFD numerical analysis, and the deviation of a traditional model is remarkably reduced; on the basis, the beamforming problem of joint optimization of the QoS satisfaction rate and the system energy efficiency is formulated, and a Q3E algorithm is provided for priority user management. And finally, quickly solving a non-convex optimization problem through an unsupervised artificial neural network integrated with constraint penalty. According to the method, through interdisciplinary modeling and AI reasoning, the problem of power redundancy allocation of the HAP in a complex flow field is effectively solved, and the communication energy efficiency is remarkably improved while the QoS of a user is guaranteed.

Description

Technical Field

[0001] This invention relates to the field of cross-disciplinary technology of wireless communication and aircraft control, specifically to a propulsion power consumption modeling and beamforming method for high-altitude platforms. Background Technology

[0002] High Altitude Platforms (HAPs), as a crucial component of 6G integrated air-space-ground networks, are deployed in the stratosphere, offering advantages such as wide coverage and low latency. However, HAPs have limited energy resources (primarily relying on solar energy and batteries), and their propulsion systems consume significant power to withstand wind fluctuations. Traditional research often overlooks the impact of aerodynamic disturbances on propulsion power consumption, leading to inaccurate propulsion power estimates and consequently limiting the power budget for communication payloads. Therefore, accurately modeling propulsion power consumption and optimizing communication power allocation under this constraint is key to improving HAP lifespan and service quality. Summary of the Invention

[0003] (a) Technical problems to be solved

[0004] The present invention aims to overcome the shortcomings of the prior art and provide a propulsion power consumption modeling and beamforming method for high-altitude platforms, which solves the problems mentioned in the background art.

[0005] (II) Technical Solution

[0006] To achieve the above objectives, the present invention specifically adopts the following technical solution:

[0007] A propulsion power consumption modeling and beamforming method for high-altitude platforms includes the following steps:

[0008] Step S1: Construct a full-system power consumption model for the High Altitude Platform (HAP), covering communication, propulsion, payload, and maintenance power consumption; Step S2: Establish a precise propulsion power consumption modeling architecture based on generative artificial intelligence (AI), and output the corrected propeller efficiency model and propulsion power consumption. Step S3: Propulsion power consumption based on step S2 Based on the total power limit of HAP, calculate the remaining available communication power budget. The process involves: 1) Constructing a beamforming optimization problem that jointly optimizes user service quality (QoS) satisfaction rate and system energy efficiency (EE); 2) Proposing an energy-efficient beamforming algorithm Q3E to enhance QoS. Based on user channel state and power requirements, users are divided into a set that fully satisfies QoS and a set that partially satisfies QoS, transforming the original optimization problem into two cascaded sub-optimization problems; 3) Constructing an unsupervised learning architecture based on artificial neural networks (ANN) to quickly solve the sub-optimization problems and output the optimal power allocation coefficients.

[0009] Furthermore, the modeling process in step S1 includes the following:

[0010] Communication system modeling: The HAP is assumed to be equipped with a uniform planar array, which has [missing information] arranged along the x-axis and y-axis. and Each antenna element, total number of antennas Power consumption of communication systems It consists of two parts: power consumption dynamically related to transmit power and power consumption of static circuitry, mathematically expressed as: ,in For power amplifier power consumption, For amplifier inefficiency factor, This refers to the power consumption of the static circuit. Total transmission power, For RF chain power consumption, For the power consumption of the local oscillator, This refers to the power consumption of baseband processing.

[0011] Propulsion system modeling: Before introducing AI correction, propulsion power consumption is calculated based on classical momentum theory. The platform's resistance to wind and flight airspeed related: ;in For propeller efficiency, For motor efficiency, wind resistance The calculation formula is , Stratospheric air density The drag coefficient, For volume;

[0012] Total power consumption synthesis: combining the power consumption of the payload and environmental maintenance power consumption Incorporating this, we obtain the power consumption equation for the entire system: .

[0013] Furthermore, the specific content of step S2 includes: constructing a generative AI agent: the AI ​​agent is constructed based on a large language model and integrates a retrieval enhancement generation module;

[0014] Collaborative CFD numerical analysis: The AI ​​agent receives the geometric parameters of the HAP and the current flight conditions, including airspeed. The angle of attack is determined, control scripts are automatically written, and computational fluid dynamics (CFD) simulation software is called to analyze the velocity field distribution at the stern of the hull.

[0015] Derivation of the modified model: The AI ​​agent utilizes its thought chain reasoning ability to perform nonlinear regression analysis on the discrete data points output by CFD, targeting... From the airspeed range, the propeller efficiency considering aerodynamic interference is derived. airspeed Exponential correction model: ;

[0016] Calculating precise propulsion power consumption: By coupling the efficiency model with the HAP drag coefficient model, the final precise propulsion power consumption formula is derived.

[0017]

[0018] in, This refers to the density of air in the stratosphere. Airspeed; The aerodynamic viscosity coefficient; For HAP volume; For motor efficiency; This is the tail fin drag correction factor; This refers to the length of the HAP hull. This is the maximum width of the HAP hull.

[0019] Furthermore, step S3 specifically includes: determining the power budget: setting the total energy supply capacity of the HAP to... The precise propulsion power consumption calculated in step S2 After deducting fixed load and maintenance power consumption, the remaining available communication power budget is obtained. :

[0020]

[0021] Establish a channel model: Assume that the ground has The user, HAP and the first Channel vector between users Modeled as a Ricean fading channel, it includes a line-of-sight component (LoS) and a scattering component (NLoS).

[0022] Define the optimization problem:

[0023] Objective function: Jointly maximize system energy efficiency (EE) and QoS satisfaction rate. EE is defined as the ratio of total transmission rate to total power consumption; QoS satisfaction rate is defined as the rate at which the actual transmission rate reaches the required rate. The proportion of users;

[0024] Constraints:

[0025] Total transmission power shall not exceed the budget: ,in For power coefficient, Assign beamforming vectors;

[0026] User QoS constraints: For users within the QoS satisfaction set.

[0027] Furthermore, the specific content of step S4 is as follows: This step first receives the remaining communication available power budget output in step S3. As the upper limit of the system's total energy constraint, it also receives the beamforming optimization problem constructed in step S3 as the objective to be solved; because Due to the relatively small power consumption constraint, directly solving the optimization problem defined in step S3 leads to constraint conflicts. Therefore, the Q3E algorithm is used to classify users. The specific implementation process includes:

[0028] Calculate the minimum power threshold: for each user Calculate the minimum QoS rate required to meet the requirements. The minimum required signal-to-interference-plus-noise ratio (SINR) is used to derive the minimum transmit power. ; ,in For noise power, For gain, its expression is:

[0029]

[0030] in, For HAP transmit antenna gain, For ground user receiving antenna gain, The number of transmitting antennas, At the speed of light, For carrier frequency, HAP flight altitude Total number of ground users;

[0031] System-wide power verification: Calculate the sum of the minimum power for all users. ;

[0032] Situation Classification and Handling:

[0033] Scenario 1: Sufficient budget If all users can be satisfied, then the optimization goal is to satisfy all users. Under the premise of maximizing system energy efficiency (EE), all users are grouped into a set. ;

[0034] Scenario 2: Insufficient budget To discard users, the algorithm executes a priority sorting strategy, ranking all users according to their minimum power requirements. Sort in ascending order from lowest to highest;

[0035] Greedy acceptance: Adds users to the service set in order of priority. And accumulate its power demand until the accumulated value exceeds Users who were not selected were grouped into a set. ;

[0036] At this point, the optimization objective becomes: prioritize maximizing the set. The cardinality, and secondly in the set Maximize energy efficiency (EE) within the system.

[0037] Furthermore, the specific content of step S5 is as follows:

[0038] Inherited input and network input design: The high-priority user set selected in step S4... The service targets identified for this optimization were extracted. UPA array response vector of the middle user The remaining available communication power budget determined in step S3 and user QoS requirements As input features of ANN;

[0039] Dynamic network architecture construction: Building a five-layer fully connected feedforward neural network (FNN) that adapts to changes in the number of users;

[0040] Input layer: The number of neurons is adapted to the dimension of the input features;

[0041] Hidden layers: Three hidden layers are set, with the preferred number of neurons being 64, 64, and 32 respectively. The ReLU activation function is used between layers to introduce nonlinearity and prevent gradient vanishing.

[0042] Output layer: Settings 1 neuron, corresponding to Power allocation coefficient for each user The output layer uses the Sigmoid activation function, forcibly mapping the output to... The interval represents the normalized power ratio;

[0043] To achieve unsupervised learning, this invention designs a special loss function that transforms constraints into penalty terms. First, the optimization objective term is defined. In other words, system energy efficiency:

[0044]

[0045] in , This represents the total communication power consumption at the current moment.

[0046] Based on this, a loss function is constructed. :

[0047]

[0048] in, and The penalty coefficient is used to penalize QoS violations through the ReLU function, and the logarithmic barrier function forces the network output to strictly meet the total power constraint. This enables unsupervised learning without the need for labeled data.

[0049] The first objective is to maximize energy efficiency, which can be transformed into minimizing losses by taking a negative sign.

[0050] Second item: QoS penalty, user rate Below requirements If it results in a positive loss, then it is zero; otherwise, it is zero.

[0051] The third term: Logarithmic barrier term, ensuring that the total power is strictly less than... Otherwise, the losses will tend towards infinity;

[0052] Model inference: The trained ANN model is deployed on the HAP onboard processor, and the current channel state and power budget are input in real time. After forward propagation, the network directly outputs the optimal power allocation coefficients that satisfy the constraints. .

[0053] (III) Beneficial Effects

[0054] Compared with existing technologies, this invention provides a propulsion power consumption modeling and beamforming method for high-altitude platforms, which has the following beneficial effects:

[0055] Existing technologies typically ignore the aerodynamic interference (hull-induced turbulence effect) caused by the massive hull of the High Altitude Platform (HAP) on the propeller inflow, leading to significant errors in propulsion power consumption estimation and consequently misjudgment of communication power budget. This invention introduces a generative AI agent and a CFD simulation collaborative mechanism in step S2 to establish a nonlinear propeller efficiency correction model. Simulation results show that this correction model can accurately characterize the energy consumption characteristics under complex flow fields, reducing the propulsion power consumption prediction deviation by an average of 84.3% compared to traditional baseline models. This high-precision modeling provides an accurate "residual power budget" for subsequent communication systems, avoiding the risk of wasting communication resources due to excessive redundant power reserves or causing system downtime due to insufficient reserves.

[0056] For scenarios where high-altitude platforms have limited energy resources but need to serve a large number of ground users, traditional optimization algorithms often fail to converge or simply drop all users when the power budget is insufficient. This invention, through the Enhanced QoS Energy Efficiency (Q3E) algorithm proposed in step S4, creatively introduces a priority ranking mechanism based on minimum power requirements. When energy is scarce, the algorithm prioritizes access for users with low power consumption needs, maximizing the number of users served. Experimental data shows that, under the same power constraints, this invention improves the QoS satisfaction rate of users by up to 33.3%, effectively solving the service interruption problem under resource contention.

[0057] Traditional beamforming optimization typically relies on highly complex iterative algorithms (such as SLSQP) or supervised learning models requiring massive amounts of manually labeled data, making it difficult to meet the millisecond-level real-time communication requirements of HAP. This invention constructs an unsupervised artificial neural network (ANN) in step S5. By designing a composite loss function including a "soft loss" term, rigid power and QoS constraints are transformed into differentiable penalty terms, directly finding the optimal solution during backpropagation. This method avoids the expensive cost of labeled data acquisition and reduces the computation time of beamforming weights by two orders of magnitude, enabling online real-time inference.

[0058] Thanks to precise propulsion power deduction at the front end and intelligent filtering of the user set by the Q3E algorithm at the back end, this invention can allocate every milliwatt of power to the user with the highest channel gain and optimal performance while ensuring the quality of service for critical users. Simulation results show that, under typical stratospheric flight conditions, the method proposed in this invention improves the overall system energy efficiency (EE) by 45.4% compared to similar benchmark algorithms, significantly extends the mission loiter time of the high-altitude platform, and improves communication coverage quality. Attached Figure Description

[0059] Figure 1 This is the overall main flowchart of the method of the present invention;

[0060] Figure 2 A schematic diagram of the network structure and power consumption of a high-altitude platform;

[0061] Figure 3 Flowchart for power consumption modeling driven by generative AI agents;

[0062] Figure 4 A comparison curve before and after power consumption model correction is provided to advance the process. Detailed Implementation

[0063] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0064] like Figure 2 As shown in the figure, the high-altitude platform network architecture proposed in this embodiment covers core components such as propulsion system, communication system and environmental maintenance system, providing a basic physical model for power distribution.

[0065] like Figure 1 As shown, an embodiment of the present invention proposes a propulsion power consumption modeling and beamforming method for a high-altitude platform, comprising the following steps:

[0066] Step S1: Construct a full-system power consumption model of the High Altitude Platform (HAP), covering propulsion, communication, payload, and maintenance power consumption;

[0067] This step aims to establish the energy budget baseline for the high-altitude platform, providing a mathematical foundation for subsequent optimization. In this embodiment, the high-altitude platform (HAP) is defined as a solar / cell-powered aircraft deployed in the stratosphere (approximately 17-25 km altitude) and propelled by a quadcopter. The specific modeling process is as follows:

[0068] Communication system modeling: Assume the HAP is equipped with a Uniform Planar Array (UPA), which has arrays configured along the x-axis and y-axis. and Each antenna element, total number of antennas Power consumption of communication systems It consists of two parts: power consumption dynamically related to transmit power (generated by the power amplifier PA) and power consumption of the static circuit. The mathematical expression is: in, For power amplifier power consumption, This is the amplifier's inefficiency factor (i.e., the reciprocal of its efficiency). This represents the total transmission power. For RF chain power consumption, For the power consumption of the local oscillator, This refers to the power consumption of baseband processing.

[0069] Propulsion system modeling: Before introducing AI correction, propulsion power consumption is calculated based on classical momentum theory. The platform's resistance to wind and flight airspeed related: ;in, For propeller efficiency, Motor efficiency. Wind resistance. The calculation formula is , Stratospheric air density The drag coefficient, For volume.

[0070] Total power consumption synthesis: combining the power consumption of the payload (e.g., cameras, radar) and environmental maintenance power consumption Including (e.g., thermal control system), we obtain the power consumption equation for the entire system: .

[0071] Output: This step outputs an initial full-system power consumption mathematical model containing the definitions of various parameters. This model is used as the basic framework input to step S2.

[0072] Step S2: Establish a precise propulsion power consumption modeling architecture based on generative artificial intelligence (AI), and output the corrected propeller efficiency model and propulsion power consumption. ;

[0073] Traditional propulsion models typically assume propeller efficiency The constant value ignores the "hull-induced turbulence" generated by the massive hull of the HAP during flight, leading to significant errors in power consumption estimation. This step, in constructing the propulsion power consumption model, refers to... Figure 3 The AI-driven modeling process shown solves this physical modeling challenge.

[0074] The specific implementation process includes:

[0075] Constructing a Generative AI Agent: This agent is built upon a Large Language Model (LLM, such as GPT-4) to create its "brain," and integrates a Retrieval-Enhanced Generation (RAG) module. RAG technology is an architecture that enhances the capabilities of an LLM by retrieving relevant information from an external knowledge base. In this embodiment, RAG addresses the "illusion" problem that may arise when a general-purpose LLM deals with specialized aerodynamic problems. It inputs the blade element theory formula and the NACA airfoil database as external knowledge context to the LLM, ensuring that the AI-generated CFD control scripts conform to the laws of physics, rather than being mere textual probabilistic predictions.

[0076] Collaborative CFD numerical analysis: The AI ​​agent receives the geometric parameters of the HAP and the current flight conditions (airspeed). The system automatically generates control scripts and calls computational fluid dynamics (CFD) simulation software (angle of attack). The CFD simulation focuses on analyzing the velocity field distribution at the stern of the hull and finds that the wake of the hull causes distortion of the inflow velocity at the propeller, thereby reducing propulsion efficiency.

[0077] Derivation of the modified model: The AI ​​agent utilizes its thought chain reasoning ability to perform nonlinear regression analysis on the discrete data points output by CFD. In this embodiment, for... Based on the airspeed range, the propeller efficiency considering aerodynamic interference was derived. airspeed Exponential correction model: This formula precisely describes the nonlinear suppression effect of aerodynamic disturbances on efficiency as airspeed increases.

[0078] Calculate precise propulsion power consumption: Combine the above efficiency model with the HAP drag coefficient model (incorporating tail fin correction factors). By coupling the Reynolds number-related terms, the final precise propulsion power consumption formula is derived:

[0079]

[0080] in, This refers to the density of air in the stratosphere. Airspeed; The aerodynamic viscosity coefficient; For HAP volume; For motor efficiency; This is the tail fin drag correction factor; This refers to the length of the HAP hull. This is the maximum width of the HAP hull.

[0081] This formula precisely describes the nonlinear suppression effect of aerodynamic disturbances on efficiency as airspeed increases.

[0082] Output: This step outputs the high-precision propulsion power consumption after correction based on the S1 physical framework. This value corrects for the error in the initial model of step S1 caused by neglecting aerodynamic disturbances (the deviation is reduced by an average of 84.3%), and is input into step S3 as an immutable "rigid power consumption deduction item" to calculate the remaining available communication power budget. .

[0083] Step S3: Construct a beamforming optimization model for the high-altitude platform communication system based on user quality of service (QoS) requirements and system energy efficiency (EE), and determine the remaining available communication power budget;

[0084] This step uses the propulsion power consumption output from step S2 as a known constraint to determine how much power the communication system still has available and to establish an optimization objective. The specific implementation process includes:

[0085] Determine the power budget: Assume the total energy supply capacity of the HAP (determined by the solar array and energy storage batteries) is: The precise propulsion power consumption calculated in step S2 After deducting fixed load and maintenance power consumption, the remaining available communication power budget is obtained. :

[0086]

[0087] The budget is dynamic and changes with the HAP flight speed (wind resistance requirements).

[0088] Establish a channel model: Assume there is a channel on the ground. The user, HAP and the first Channel vector between users It is modeled as a Rician fading channel, which includes a line-of-sight (LoS) component and a scattering (NLoS) component.

[0089] Define the optimization problem:

[0090] Objective function: Jointly maximize system energy efficiency (EE) and QoS satisfaction rate. EE is defined as the ratio of total transmission rate to total power consumption; QoS satisfaction rate is defined as the rate at which the actual transmission rate reaches the required rate. The proportion of users.

[0091] Constraints:

[0092] Total transmission power shall not exceed the budget: (in For power coefficient, (Beamforming vector).

[0093] User QoS constraints: (For users within the QoS satisfaction set).

[0094] Output: This step outputs the explicit remaining communication power budget. The mathematical optimization problem P to be solved is then passed as input to the next level algorithm.

[0095] Step S4: Propose the energy-efficient beamforming algorithm Q3E to enhance QoS, transforming the original optimization problem into a sub-optimization problem under two conditions: complete and partial satisfaction of user QoS.

[0096] Input reception: This step first receives the communication available power budget output from step S3. As the upper limit of the system's total energy constraint, it also receives the beamforming optimization problem constructed in step S3 as the objective to be solved. Because Due to the pressure of power consumption, the numerical value may be small, and directly solving the optimization problem defined by S3 may lead to constraint conflicts (i.e., no solution). Therefore, this step uses the Q3E algorithm to classify users.

[0097] The specific implementation process includes:

[0098] Due to communication power budget Due to power consumption constraints, directly solving the above optimization problem may result in no solution or difficulty in convergence. This step uses the Q3E algorithm to classify users. The specific implementation process includes:

[0099] Calculate the minimum power threshold: for each user Calculate the minimum QoS rate required to meet the requirements. The minimum required signal-to-interference-plus-noise ratio (SINR) is then used to derive the minimum transmit power. : in, For noise power, For gain, its expression is:

[0100]

[0101] in, For HAP transmit antenna gain, For ground user receiving antenna gain, The number of transmitting antennas, At the speed of light, For carrier frequency, HAP flight altitude This represents the total number of ground users.

[0102] System-wide power verification: Calculate the sum of the minimum power for all users. .

[0103] Situation Classification and Handling:

[0104] Scenario 1 (Ample budget, ): All users can be satisfied. At this point, the optimization objective is to satisfy all... Under the premise of maximizing system energy efficiency (EE), all users are grouped into a set. .

[0105] Scenario 2 (Insufficient budget) (This requires "user discarding"). The algorithm executes a priority sorting strategy: all users are sorted according to their minimum power requirement. Sort in ascending order from lowest to highest.

[0106] Greedy acceptance: Adds users to the service set in order of priority. And accumulate its power demand until the accumulated value exceeds Users who were not selected are grouped into a set. (We will do our best to provide service, but QoS is not guaranteed).

[0107] At this point, the optimization objective becomes: prioritize maximizing the set. The cardinality (i.e., satisfying as many people as possible), and secondly in the set Maximize energy efficiency (EE) within the system.

[0108] Output: This step outputs the categorized user set (the set that satisfies the criteria). and non-satisfied sets This transforms a potentially unsolvable non-convex problem into a feasible optimization problem for a specific subset of users.

[0109] Step S5: Construct an unsupervised learning architecture based on artificial neural networks (ANN) to quickly solve the sub-optimization problem and output the optimal power allocation coefficient.

[0110] Traditional convex optimization algorithms (such as SLSQP) have high computational complexity, making them unsuitable for the real-time communication requirements of HAP. This step designs an unsupervised ANN network to output results in seconds. The specific implementation process includes: 1. Inherited input and network input design: The high-priority user set selected in step S4... (If the budget allows, the entire series will be included) The target service for this optimization was identified. The set was extracted. UPA array response vector of the middle user The remaining available communication power budget determined in step S3 and user QoS requirements As the input feature vector of ANN.

[0111] 2. Dynamic Network Architecture Construction: Construct a five-layer fully connected feedforward neural network (FNN) that can adapt to changes in the number of users.

[0112] Input layer: The number of neurons is adapted to the dimension of the input features.

[0113] Hidden layers: Three hidden layers are set, with the preferred number of neurons being 64, 64, and 32 respectively. The ReLU activation function is used between layers to introduce nonlinearity and prevent gradient vanishing.

[0114] Output layer: Settings 1 neuron, corresponding to Power allocation coefficient for each user The key lies in using the Sigmoid activation function in the output layer, which forces the output to be mapped to... The interval represents the normalized power ratio.

[0115] Constructing a composite soft loss function: The so-called "soft loss" refers to transforming the originally rigid mathematical constraints (such as power not exceeding) into a soft loss function. This is transformed into a penalty term in the loss function.

[0116] To achieve unsupervised learning, this embodiment designs a special loss function that transforms constraints into penalty terms. First, the optimization objective term is defined. In other words, system energy efficiency:

[0117]

[0118] in , This represents the total communication power consumption at the current moment.

[0119] Based on this, a loss function is constructed. :

[0120]

[0121] in, and The penalty coefficient is used to penalize QoS violations through the ReLU function, and the logarithmic barrier function forces the network output to strictly meet the total power constraint. This enables unsupervised learning without the need for labeled data.

[0122] The first item: Maximize energy efficiency (taking the negative sign to minimize loss).

[0123] The second item: QoS penalty. If the user rate... Below requirements If it results in a positive loss, then it is zero; otherwise, it is zero.

[0124] The third term: the logarithmic barrier term. This ensures the total power is strictly less than... Otherwise, the losses will tend to be infinite.

[0125] Model inference: The trained ANN model is deployed on the HAP onboard processor. With the current channel state and power budget input in real time, the network can directly output the optimal power allocation coefficients that satisfy the constraints after forward propagation. .

[0126] Output: Step S5 finally outputs the beamforming vectors for each user. Based on this, the HAP communication system adjusts the transmitter to achieve optimal power allocation that guarantees QoS after deducting propulsion power consumption interference.

[0127] from Figure 4 As shown in the propulsion power consumption deviation comparison curve, after adopting the corrected model proposed in this invention, the growth trend of its power consumption prediction deviation with airspeed changes is significantly slowed down, verifying the accuracy of this modeling method.

[0128] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A propulsion power consumption modeling and beamforming method for a high-altitude platform, characterized in that, Includes the following steps: Step S1: Construct a full-system power consumption model for the High Altitude Platform (HAP), covering communication, propulsion, payload, and maintenance power consumption; Step S2: Establish a precise propulsion power consumption modeling architecture based on generative artificial intelligence (AI), and output the corrected propeller efficiency model and propulsion power consumption. Step S3: Propulsion power consumption based on step S2 Based on the total power limit of HAP, calculate the remaining available communication power budget. The process involves: 1) Constructing a beamforming optimization problem that jointly optimizes user service quality (QoS) satisfaction rate and system energy efficiency (EE); 2) Proposing an energy-efficient beamforming algorithm Q3E to enhance QoS. Based on user channel state and power requirements, users are divided into a set that fully satisfies QoS and a set that partially satisfies QoS, transforming the original optimization problem into two cascaded sub-optimization problems; 3) Constructing an unsupervised learning architecture based on artificial neural networks (ANN) to quickly solve the sub-optimization problems and output the optimal power allocation coefficients.

2. The propulsion power consumption modeling and beamforming method for a high-altitude platform according to claim 1, characterized in that, The modeling process in step S1 includes the following: Communication system modeling: The HAP is assumed to be equipped with a uniform planar array, which has [missing information] arranged along the x-axis and y-axis. and Each antenna element, total number of antennas Power consumption of communication systems It consists of two parts: power consumption dynamically related to transmit power and power consumption of static circuitry, mathematically expressed as: ,in For power amplifier power consumption, For amplifier inefficiency factor, This refers to the power consumption of the static circuit. Total transmission power, For RF chain power consumption, For the power consumption of the local oscillator, This refers to the power consumption of baseband processing. Propulsion system modeling: Before introducing AI correction, propulsion power consumption is calculated based on classical momentum theory. The platform's resistance to wind and flight airspeed related: ;in For propeller efficiency, For motor efficiency, wind resistance The calculation formula is , Stratospheric air density The drag coefficient, For volume; Total power consumption synthesis: combining the power consumption of the payload and environmental maintenance power consumption Incorporating this, we obtain the power consumption equation for the entire system: .

3. The propulsion power consumption modeling and beamforming method for a high-altitude platform according to claim 2, characterized in that, The specific content of step S2 includes: constructing a generative AI agent: the AI ​​agent is constructed based on a large language model and integrates a retrieval enhancement generation module; Collaborative CFD numerical analysis: The AI ​​agent receives the geometric parameters of the HAP and the current flight conditions, including airspeed. The angle of attack is determined, control scripts are automatically written, and computational fluid dynamics (CFD) simulation software is called to analyze the velocity field distribution at the stern of the hull. Derivation of the modified model: The AI ​​agent utilizes its thought chain reasoning ability to perform nonlinear regression analysis on the discrete data points output by CFD, targeting... From the airspeed range, the propeller efficiency considering aerodynamic interference is derived. airspeed Exponential correction model: ; Calculating precise propulsion power consumption: By coupling the efficiency model with the HAP drag coefficient model, the final formula for precise propulsion power consumption is derived. ; in, This refers to the density of air in the stratosphere. Airspeed; The aerodynamic viscosity coefficient; For HAP volume; For motor efficiency; This is the tail fin drag correction factor; This refers to the length of the HAP hull. This is the maximum width of the HAP hull.

4. The propulsion power consumption modeling and beamforming method for a high-altitude platform according to claim 3, characterized in that, The specific content of step S3 includes: determining the power budget: setting the total energy supply capacity of the HAP to be... The precise propulsion power consumption calculated in step S2 After deducting fixed load and maintenance power consumption, the remaining available communication power budget is obtained. : ; Establish a channel model: Assume that the ground has The user, HAP and the first Channel vector between users Modeled as a Ricean fading channel, it includes a line-of-sight component (LoS) and a scattering component (NLoS). Define the optimization problem: Objective function: Jointly maximize system energy efficiency (EE) and QoS satisfaction rate. EE is defined as the ratio of total transmission rate to total power consumption; QoS satisfaction rate is defined as the rate at which the actual transmission rate reaches the required rate. The proportion of users; Constraints: Total transmission power shall not exceed the budget: ,in For power coefficient, Assign beamforming vectors; User QoS constraints: For users within the QoS satisfaction set.

5. The propulsion power consumption modeling and beamforming method for a high-altitude platform according to claim 4, characterized in that, The specific content of step S4 is as follows: This step first receives the remaining communication available power budget output in step S3. As the upper limit of the system's total energy constraint, it also receives the beamforming optimization problem constructed in step S3 as the objective to be solved; because Due to the relatively small power consumption constraint, directly solving the optimization problem defined in step S3 leads to constraint conflicts. Therefore, the Q3E algorithm is used to classify users. The specific implementation process includes: Calculate the minimum power threshold: for each user Calculate the minimum QoS rate required to meet the requirements. The minimum required signal-to-interference-plus-noise ratio (SINR) is used to derive the minimum transmit power. ; ,in For noise power, For gain, its expression is: ; in, For HAP transmit antenna gain, For ground user receiving antenna gain, The number of transmitting antennas, At the speed of light, For carrier frequency, HAP flight altitude Total number of ground users; System-wide power verification: Calculate the sum of the minimum power for all users. ; Situation Classification and Handling: Scenario 1: Sufficient budget If all users can be satisfied, then the optimization goal is to satisfy all users. Under the premise of maximizing system energy efficiency (EE), all users are grouped into a set. ; Scenario 2: Insufficient budget To discard users, the algorithm executes a priority sorting strategy, ranking all users according to their minimum power requirements. Sort in ascending order from lowest to highest; Greedy acceptance: Adds users to the service set in order of priority. And accumulate its power demand until the accumulated value exceeds Users who were not selected were grouped into a set. ; At this point, the optimization objective becomes: prioritize maximizing the set. The cardinality, and secondly in the set Maximize energy efficiency (EE) within the system.

6. The propulsion power consumption modeling and beamforming method for a high-altitude platform according to claim 5, characterized in that, The specific content of step S5 is as follows: Inherited input and network input design: The high-priority user set selected in step S4... The service targets identified for this optimization were extracted. UPA array response vector of the middle user The remaining available communication power budget determined in step S3 and user QoS requirements As input features of ANN; Dynamic network architecture construction: Building a five-layer fully connected feedforward neural network (FNN) that adapts to changes in the number of users; Input layer: The number of neurons is adapted to the dimension of the input features; Hidden layers: Three hidden layers are set, with the preferred number of neurons being 64, 64, and 32 respectively. The ReLU activation function is used between layers to introduce nonlinearity and prevent gradient vanishing. Output layer: Settings 1 neuron, corresponding to Power allocation coefficient for each user The output layer uses the Sigmoid activation function, forcibly mapping the output to... The interval represents the normalized power ratio; To achieve unsupervised learning, a special loss function is designed to transform constraints into penalty terms. First, the optimization objective term is defined. In other words, system energy efficiency: ; in , This represents the total communication power consumption at the current moment. Based on this, a loss function is constructed. : ; in, and The penalty coefficient is used to penalize QoS violations through the ReLU function, and the logarithmic barrier function forces the network output to strictly meet the total power constraint. This enables unsupervised learning without the need for labeled data. The first option: maximize energy efficiency, which can be converted to minimizing losses by taking a negative sign; Second item: QoS penalty, user rate Below requirements If it results in a positive loss, then it is zero; otherwise, it is zero. The third term: Logarithmic barrier term, ensuring that the total power is strictly less than... Otherwise, the losses will tend towards infinity; Model inference: The trained ANN model is deployed on the HAP onboard processor, and the current channel state and power budget are input in real time. After forward propagation, the network directly outputs the optimal power allocation coefficients that satisfy the constraints. .

Images