Unmanned aerial vehicle communication system and method based on narrowband Internet of Things

Abstract

The invention discloses an unmanned aerial vehicle communication system and method based on narrowband Internet of Things, and relates to the field of unmanned aerial vehicle communication, and the system comprises an unmanned aerial vehicle communication node, a narrowband Internet of Things core network and base station, a cloud or edge intelligent control platform, and a plurality of Internet of Things terminals. The unmanned aerial vehicle obtains the three-dimensional position information of the unmanned aerial vehicle and the channel quality, service load and time delay information reported by each Internet of Things terminal, and obtains the three-dimensional position information of the unmanned aerial vehicle under the control of a preset deep reinforcement learning trajectory planning model and a supervised learning resource configuration model. Jointly determining the three-dimensional flight path of the unmanned aerial vehicle in the target area and the transmitting power, the uplink repetition frequency, the resource block number and the reporting period of each Internet of Things terminal; and the cloud or edge intelligent control platform is used for receiving operation data reported by the unmanned aerial vehicle communication node and the Internet of Things terminal, performing online or offline training on the trajectory planning model and the resource configuration model, and issuing updated model parameters to the unmanned aerial vehicle communication node.

Description

Technical Field

[0001] This invention relates to the field of unmanned aerial vehicle (UAV) communication, and more specifically to a UAV communication system and method based on narrowband Internet of Things (IoT). Background Technology

[0002] With the rapid development of IoT technology, a large number of low-power, low-data-rate sensor terminals have been deployed in scenarios such as environmental monitoring, agricultural planting, smart meter reading, and urban infrastructure monitoring. Narrowband IoT (NB-IoT), as a cellular IoT technology designed for wide coverage, massive connectivity, and low power consumption, has been widely used in urban environments. However, existing NB-IoT systems typically rely on fixed terrestrial base stations for coverage. In complex or remote scenarios such as mountainous areas, islands, deserts, disaster-stricken areas, and gaps between high-rise buildings in cities, it is difficult to provide stable network services in a timely and flexible manner, easily resulting in large-scale blind spots or weak coverage areas.

[0003] To enhance coverage, some existing technologies propose using drones as aerial base stations or relay nodes. By carrying cellular communication modules in the air, they can fill in gaps or enhance coverage for ground terminals. While these solutions improve network coverage flexibility to some extent, they typically use pre-planned fixed routes or simple rule-based strategies to control drone flight altitude and landing locations, failing to fully consider the dynamic changes in factors such as terminal distribution, service load, channel conditions, and drone energy constraints. Furthermore, the wireless resource configurations in existing NB-IoT systems, such as uplink transmit power, repetition count, resource block allocation, and reporting cycle, often employ fixed configurations or empirical rules based on simple thresholds, making it difficult to make timely and adaptive adjustments based on terminal channel quality and service requirements.

[0004] In addition, while some existing solutions have begun to attempt to plan UAV trajectories using optimization or heuristic algorithms, they often consider UAV trajectory planning and narrowband IoT resource allocation separately, lacking joint modeling and collaborative optimization of "aerial platform movement trajectory" and "NB-IoT wireless resource scheduling". The few existing intelligent methods are mostly based on parameter adjustments in offline simulation environments, making it difficult to continuously train and update the model using real-time operational data in actual deployments. As a result, there is still considerable room for improvement in the system's coverage, service latency, and energy utilization efficiency in complex scenarios. Summary of the Invention

[0005] To achieve the above and other related objectives, this invention discloses a drone communication system based on narrowband Internet of Things, comprising:

[0006] At least one drone communication node, a narrowband IoT core network and base station, a cloud or edge intelligent control platform, and multiple IoT terminals;

[0007] The UAV communication node is equipped with a narrowband IoT radio frequency and baseband module, a position and attitude acquisition module, an airborne computing module, and a storage module. The airborne computing module is configured to acquire the UAV's three-dimensional position information and the channel quality, service load, and latency information reported by each IoT terminal. Under the control of the deep reinforcement learning trajectory planning model and the supervised learning resource configuration model pre-set in the storage module, it jointly determines the UAV's three-dimensional flight trajectory in the target area and the transmit power, uplink repetition count, resource block quantity, and reporting cycle of each IoT terminal. It also performs access control and data forwarding to the IoT terminals through the narrowband IoT air interface. The cloud or edge intelligent control platform is used to receive operational data reported by UAV communication nodes and IoT terminals, perform online or offline training on the trajectory planning model and the resource configuration model, and send the updated model parameters to the UAV communication nodes to improve the coverage, service latency performance and energy utilization efficiency of IoT terminals in remote or blind spot scenarios.

[0008] Secondly, this invention discloses a drone communication method based on narrowband Internet of Things, applied to the above-mentioned system, comprising the following steps: S1: Initialize the deep reinforcement learning trajectory planning model and the supervised learning resource configuration model in the cloud or edge intelligent control platform, obtain the initial model parameters, and send them to the UAV communication node; S2: The UAV communication node flies within the target area according to the initial trajectory, establishes connections with multiple IoT terminals through the narrowband IoT air interface, and collects operational data such as terminal channel quality, service load, latency, UAV position, and battery level. S3: Upload the running data to the cloud or edge intelligent control platform. The cloud or edge intelligent control platform trains the trajectory planning model and resource allocation model based on the running data to obtain updated model parameters. S4: The updated model parameters are sent to the UAV communication nodes. During flight, the UAV communication nodes output the next flight trajectory and transmission power based on the current state from the trajectory planning model. The resource configuration model outputs the uplink repetition count, resource block quantity and reporting cycle of each IoT terminal, and performs access control and data scheduling for the IoT terminals accordingly.

[0009] By adopting the above technical solution, and using drones as aerial communication nodes equipped with narrowband IoT modules, a deep reinforcement learning trajectory planning model and a supervised learning resource allocation model are introduced. The model parameters are continuously trained and updated using actual operational data in the cloud or at the edge. This achieves coordinated optimization of the drone's three-dimensional flight trajectory and wireless resources such as NB-IoT uplink transmission power, repetition count, resource block allocation, and reporting cycle. Thus, in complex or remote scenarios, compared with traditional solutions that rely on fixed ground base stations and rule-based configuration, it can more flexibly fill blind spots and enhance coverage, more adaptively match the channel status and service requirements of the terminal, and effectively balance coverage, service latency, and drone energy utilization, thereby improving the overall service capability and operational stability of the system under multiple scenarios and service conditions. Attached Figure Description

[0010] The above and other features, advantages, and aspects of the embodiments of this disclosure will become more apparent from the accompanying drawings and the following detailed description. The drawings are provided for a better understanding of the invention and are not intended to limit the scope of this disclosure. In the drawings, the same or similar reference numerals denote the same or similar elements, wherein: Figure 1 This is a system block diagram of the present invention. Detailed Implementation

[0011] 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.

[0012] Reference Figure 1 This invention provides a drone communication system based on narrowband Internet of Things, comprising: At least one drone communication node, a narrowband IoT core network and base station, a cloud or edge intelligent control platform, and multiple IoT terminals; The UAV communication node is equipped with a narrowband IoT radio frequency and baseband module, a position and attitude acquisition module, an airborne computing module, and a storage module. The airborne computing module is configured to acquire the UAV's three-dimensional position information and the channel quality, service load, and latency information reported by each IoT terminal. Under the control of the deep reinforcement learning trajectory planning model and the supervised learning resource configuration model pre-set in the storage module, it jointly determines the UAV's three-dimensional flight trajectory in the target area and the transmit power, uplink repetition count, resource block quantity, and reporting cycle of each IoT terminal. It also performs access control and data forwarding to the IoT terminals through the narrowband IoT air interface. The cloud or edge intelligent control platform is used to receive operational data reported by UAV communication nodes and IoT terminals, perform online or offline training on the trajectory planning model and the resource configuration model, and send the updated model parameters to the UAV communication nodes to improve the coverage, service latency performance and energy utilization efficiency of IoT terminals in remote or blind spot scenarios.

[0013] Preferably, the UAV communication node is equipped with a channel measurement module, which is configured to perform a decision at the time of decision-making. The spatial distance between the drone and the IoT terminal is calculated based on their three-dimensional coordinates. The spatial distance satisfies: , in, Indicates the first The decision moment The drone and the first Spatial distance between IoT terminals; , , They represent the first The decision moment A drone in the ground coordinate system Axial direction, Coordinates along the axial direction and the height direction; , They represent the first An IoT terminal in the ground coordinate system Axial direction and Coordinates along the axis.

[0014] Preferably, the channel measurement module is further configured to estimate the path loss between the UAV and the IoT terminal using a logarithmic distance path loss model. The path loss satisfies: , in, Indicates the first The decision moment The drone and the first Path loss between IoT terminals Indicates reference distance Path loss constant at that location This represents the path loss index. This represents the reference distance used for normalization.

[0015] Preferably, the onboard computing module is further configured to calculate the received signal-to-noise ratio of each IoT terminal based on path loss. The received signal-to-noise ratio satisfies: , The effective coverage terminals are determined based on the following coverage decision conditions: ,, in, Indicates the first The decision moment The signal-to-noise ratio of an IoT terminal Indicates the first The decision moment The transmit power of each drone, Indicates the first The decision moment The drone and the first Path loss between IoT terminals Indicates the noise power at the receiving end. This represents the minimum acceptable signal-to-noise ratio threshold preset by the system. When a terminal satisfies the above inequality, the terminal is determined to be a valid coverage terminal.

[0016] Preferably, the airborne computing module is configured to calculate the system coverage rate at each decision-making moment based on the number of effectively covered terminals and the UAV's energy consumption. and overall utility The system coverage satisfies: , Overall utility is satisfied: , in, Indicates the first System coverage at each decision moment Indicates the first The number of valid covered terminals that meet the coverage decision conditions at each decision moment. This indicates the total number of IoT terminals in the system; Indicates the first The overall utility value at each decision moment. Indicates the first The average uplink latency of the system at each decision moment Indicates the relationship with the first The decision moment corresponds to the time period of the first decision moment. Energy consumption of a drone This indicates the maximum total energy available to the drone. , and These represent the weighting coefficients corresponding to coverage, latency, and energy consumption, respectively.

[0017] Preferably, the cloud or edge intelligent control platform is configured to train the UAV trajectory planning strategy using a deep reinforcement learning algorithm to improve overall effectiveness. As a reward signal, optimize the policy network parameters. To maximize the following trajectory return objective function: , in, This indicates that when the strategy parameter is Expected return of trajectory planning in time This represents the parameter vector of the deep reinforcement learning policy network. This represents the upper limit of the decision steps in trajectory planning. This represents the discount factor in reinforcement learning. Indicates the first The overall utility value at each decision moment. The mathematical expectation operator represents the randomness of the environment and the randomness of the strategy.

[0018] Preferably, the cloud or edge intelligent control platform is further configured to train a supervised learning resource allocation model and construct a resource allocation output vector for each training sample. and its optimal or near-optimal label vector The model parameters are optimized using the following mean squared error loss function: , in, This represents the loss function value of the resource allocation supervised learning model. This represents the total number of samples used for training. This indicates that the model represents the i-th terminal at the k-th decision time. Resource allocation output vectors on each training sample It represents the i-th terminal at the k-th decision time. The optimal or near-optimal resource allocation label vector corresponding to each training sample. This represents the 2-norm operation for calculating the Euclidean distance between two vectors.

[0019] In a preferred embodiment, the UAV communication node is equipped with a narrowband IoT radio frequency and baseband module, a position and attitude acquisition module (e.g., a GNSS module and an IMU module), an onboard computing module, and a storage module. The onboard computing module can employ an embedded processor, an edge AI chip, or a system-on-a-chip with an AI acceleration unit to execute deep reinforcement learning trajectory planning models and supervised learning resource allocation models during flight, and to adjust the UAV's three-dimensional flight trajectory and the uplink resources of each IoT terminal in real time.

[0020] Narrowband IoT core network and base stations are used to provide basic cellular network access and control plane functions. IoT terminals access the core network through drone nodes or ground base stations to realize the reporting and transmission of business data.

[0021] The cloud or edge intelligent control platform is connected to the core network to centrally store the operational data reported by drones and IoT terminals. Based on this data, the platform can train trajectory planning models and resource allocation models offline or online. The updated model parameters are then distributed to the drone communication nodes through the core network, thereby continuously improving coverage, service latency performance, and energy utilization efficiency.

[0022] In typical application scenarios, the system is deployed in mountainous areas, islands, disaster-stricken areas, or urban signal blind spots, with the number of IoT terminals reaching thousands or tens of thousands. Drones provide narrowband IoT access services to fill gaps or enhance coverage for a large number of ground terminals by moving and hovering in the air.

[0023] Preferably, in order to achieve accurate estimation of terminal channel quality, in this embodiment, the airborne computing module first calculates the spatial distance between the UAV and the IoT terminal based on their positions, and then uses the logarithmic distance path loss model to estimate the path loss.

[0024] At the discrete decision time Let the position of the UAV in the ground coordinate system be... , No. The plane coordinates of the IoT terminal on the ground are: Spatial distance between drone and terminal It can be calculated using the following formula: , in, Indicates the first The decision moment and the drone Spatial distance between IoT terminals; , , They represent the first At each decision moment, the drone is in the ground coordinate system. Axial direction, Coordinates along the axial direction and the height direction; , They represent the first An IoT terminal in the ground coordinate system Axial direction and Coordinates along the axis.

[0025] Based on the aforementioned spatial distance, the channel measurement module uses a logarithmic distance path loss model to estimate the path loss at the decision time. At that time, the drone and the first Path loss between terminals It can be represented as: , in, Indicates the first The decision moment and the drone Path loss between IoT terminals (in dB). Indicates reference distance Path loss constant at the location (in dB); This represents the path loss index, used to characterize the fading characteristics of the propagation environment. Indicates the reference distance used for normalization (e.g., 1 meter); The spatial distance is defined by the previous equation.

[0026] After completing the path loss estimation, the onboard computing module calculates the received signal-to-noise ratio of each terminal based on the current transmit power of the UAV to determine whether the terminal is within the effective coverage range.

[0027] At the moment of decision At that time, the first The received signal-to-noise ratio of each terminal It can be expressed by the following formula: , in, Indicates the first The decision moment Signal-to-noise ratio (in dB) of an IoT terminal. Indicates the first Transmit power of the UAV at each decision moment (in dBm). Indicates the first The decision moment and the drone Path loss between IoT terminals (in dB). This indicates the noise power at the receiving end (unit: dBm).

[0028] The system pre-sets the minimum signal-to-noise ratio threshold for NB-IoT coverage determination. A terminal is considered to be in a state of effective coverage at the current moment when it satisfies the following formula: , At the moment of decision Let the total number of IoT terminals in the system be . The number of terminals that meet the coverage requirements is Then the system coverage It can be calculated using the following formula: , in, Indicates the first System coverage at each decision-making moment; Indicates the first Satisfy at each decision moment The number of terminals with effective coverage; This indicates the total number of IoT terminals in the system.

[0029] To balance coverage performance and drone battery life, this embodiment models the energy consumption of the drone in each decision cycle and constructs a comprehensive utility function as a reward signal in reinforcement learning.

[0030] Assuming during the decision-making cycle Within, the time length is The average power consumption of a drone while hovering or flying is (Unit: W), the linear power of wireless transmission is (Unit: W) represents the energy consumption of the drone during this period. It can be represented as: , in, Indicates the relationship with the first The energy consumption of the drone (in J) during the time period corresponding to each decision moment. This represents the average flight or hovering power consumption of the drone during that time period (in W). Indicates the first Average linear transmit power of the UAV at each decision moment (in W). Indicates the first The length of a decision cycle (in seconds).

[0031] Based on this, define the comprehensive utility function. Simultaneously considering three factors: coverage, average latency, and energy consumption: , in, Indicates the first The overall utility value at each decision moment; Indicates the first System coverage at each decision-making moment; Indicates the first The average uplink latency of the system at each decision moment; Indicates the first The energy consumption of the drone during the time period corresponding to each decision moment; This indicates the maximum total energy available to the drone (e.g., calculated based on battery capacity). , and These represent the weighting coefficients for coverage, latency, and energy consumption, respectively, used to adjust the importance or penalty intensity of each indicator in the overall utility.

[0032] In this embodiment, the optimization of the UAV's 3D trajectory is achieved using a deep reinforcement learning method. This method optimizes the system's decision-making process at each step. The state is modeled as a state vector The control operations that the drone can perform at that moment are modeled as action vectors. .

[0033] In one optional implementation, the state vector may include UAV location information, coverage, average signal-to-noise ratio, payload information, and remaining energy ratio, specifically represented as follows: , in, Indicates the first The state vector at each decision moment; , , They represent the first The three-dimensional coordinates of the drone at each decision moment; Indicates the first System coverage at each decision-making moment; Indicates the first The average signal-to-noise ratio received by all connected terminals in the system at each decision moment; Indicates the first The average workload metric of the system at each decision moment (e.g., the average number of data packets arriving per unit time). Indicates the first The percentage of the drone's remaining battery power at each decision-making moment relative to its total battery power.

[0034] The motion vector can include the displacement increment of the UAV in three-dimensional space and the adjustment of the transmission power, for example: , in, Indicates the first Action vector at each decision moment; , , They represent the first At the moment of decision, the drone axis, Displacement increments in the axial and height directions; Indicates the first The adjustment amount of the UAV's transmit power at each decision moment (e.g., in dB).

[0035] The goal of deep reinforcement learning is to adjust the policy network parameters. To maximize long-term discounted returns, the expected return of trajectory planning can be expressed as: , in, This indicates that when the strategy parameter is Expected return for trajectory planning in time; This represents the parameter vector of the deep reinforcement learning policy network; This represents the upper limit of the decision steps in trajectory planning; This represents the discount factor in reinforcement learning, and its value typically ranges from 1 to 2. ; Indicates the first The overall utility value at each decision moment; The mathematical expectation operator represents the randomness of the environment and the randomness of the strategy.

[0036] In practical implementation, reinforcement learning algorithms such as Deep Q-Network (DQN) and Proximal Policy Optimization (PPO) can be used. Through the iterative process of "trajectory acquisition - parameter update", the policy network is trained on the cloud or edge server, and the updated parameters are periodically downloaded to the UAV communication node for online inference and trajectory decision-making.

[0037] To further improve link reliability and spectrum utilization efficiency, this embodiment uses a supervised learning model to predict and optimize NB-IoT uplink resource configuration.

[0038] For each terminal At the moment of decision Construct feature vectors Information including distance, instantaneous signal-to-noise ratio, service arrival rate, and queue length: , in, Indicates the first The decision moment Feature vectors of each terminal; Indicates the first The decision moment and the drone Spatial distance between terminals; Indicates the first The decision moment The received signal-to-noise ratio of each terminal; Indicates the first The decision moment Average service delivery rate per terminal (e.g., packets per second); Indicates the first The decision moment The length of the buffer queue for each terminal.

[0039] The output of the supervised learning model is a resource allocation vector. This includes the number of uplink repetitions, the number of allocated resource blocks, and the reporting cycle, etc. , in, Indicates the first The decision moment Resource allocation vectors for each terminal; Indicates the first The decision moment The number of uplink repetitions for each terminal; Indicates the first The decision time is assigned to the first The number of resource blocks per terminal; Indicates the first The decision moment The reporting cycle or scheduling cycle of each terminal.

[0040] Based on extensive simulation or historical optimization results, cloud-based or edge-based intelligent control platforms construct "optimal or near-optimal" resource configuration label vectors for each training sample. Then, the parameters of the supervised learning model are trained using the mean squared error loss function: , in, This represents the loss function value of the resource allocation supervised learning model; This indicates the total number of samples used for training; The model represents the first time. Resource allocation output vectors on each training sample; Indicates the first The optimal or near-optimal resource allocation label vector corresponding to each training sample; This represents the 2-norm operation for calculating the Euclidean distance between two vectors.

[0041] Secondly, this invention discloses a drone communication method based on narrowband Internet of Things, applied to the above-mentioned system, comprising the following steps: S1: Initialize the deep reinforcement learning trajectory planning model and the supervised learning resource configuration model in the cloud or edge intelligent control platform, obtain the initial model parameters, and send them to the UAV communication node; S2: The UAV communication node flies within the target area according to the initial trajectory, establishes connections with multiple IoT terminals through the narrowband IoT air interface, and collects operational data such as terminal channel quality, service load, latency, UAV position, and battery level. S3: Upload the running data to the cloud or edge intelligent control platform. The cloud or edge intelligent control platform trains the trajectory planning model and resource allocation model based on the running data to obtain updated model parameters. S4: The updated model parameters are sent to the UAV communication nodes. During flight, the UAV communication nodes output the next flight trajectory and transmission power based on the current state from the trajectory planning model. The resource configuration model outputs the uplink repetition count, resource block quantity and reporting cycle of each IoT terminal, and performs access control and data scheduling for the IoT terminals accordingly.

[0042] In a preferred embodiment, the method includes the following steps: Deep reinforcement learning trajectory planning models and supervised learning resource allocation models are pre-trained on cloud or edge intelligent control platforms to obtain initial parameters, which are then downloaded to the UAV communication nodes; ground base stations and core networks are deployed, and IoT terminals complete network registration and authentication. The drone flies to the target service area according to the preset initial route, hovers at an appropriate altitude or moves slowly, establishes a connection with the ground IoT terminal through the NB-IoT air interface, and begins to collect operational data; At each decision-making moment, the onboard computing module calculates the distance between each terminal based on the channel quality, traffic volume, and latency information reported by the terminals, combined with the UAV's own position and battery information. Path loss and signal-to-noise ratio This leads to the system coverage rate. and overall utility Subsequently, the trajectory planning model is based on the state vector. Output action vector Controlling the three-dimensional displacement and transmission power of the UAV; the resource allocation model is based on the feature vectors of each terminal. Output resource configuration vector Based on this, configure the uplink repetition count, resource block count, and reporting cycle; The drone packages the state, actions, comprehensive utility, terminal-level features and resource allocation results of each decision cycle into training samples, and transmits them back to the cloud or edge intelligent control platform through the core network for subsequent model training. Cloud-based or edge-based intelligent control platforms iteratively train deep reinforcement learning trajectory planning models based on a large amount of collected sample data to maximize trajectory rewards. Simultaneously, the supervised learning resource allocation model is trained to minimize the loss function. Once the training reaches a preset number of epochs or the performance improvement reaches a threshold, new model parameters are generated. The system sends the updated model parameters to the UAV communication nodes, replacing the old version parameters, and realizes the online evolution of trajectory planning and resource allocation strategies, thereby continuously improving coverage, service latency performance and UAV energy utilization efficiency during long-term operation.

[0043] In this invention, the narrowband IoT radio frequency includes at least multiple power modules. These functional modules include a modulation module, a power amplification module unit, and a transceiver antenna module. A matching control network is provided between adjacent modules. , To match the parameter vectors of multiple controllable elements in the current k control network, The current measurement status, , The impedance of the current k is The current operating frequency of k is... Given the current bandwidth of k, The Q-value profile of the current k; Current adjustment instructions.

[0044] In this invention, according to Selecting controllable components and forming a connection network, the process of matching adjacent modules in the connection network includes: obtaining the current state of the connection network k. The adjustment instructions are obtained according to the following formula. : ; According to the adjustment instructions Get the state of the next k+1 and get rewards ; Update according to the following formula: , In the formula, coefficient To The gradient; It is an approximate regression; For discounts.

[0045] In the first embodiment of the present invention, When the value is less than or equal to the threshold, It is positive if it is positive, otherwise it is zero.

[0046] The present invention can achieve the following beneficial effects through the above technical solution: when any module of the modulation module, power amplification module unit and transceiver antenna module is replaced, the new module can achieve adaptive matching with the old module that has not been replaced.

[0047] It will be understood by those skilled in the art that, unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. It should also be understood that terms such as those defined in general dictionaries should be understood to have the meaning consistent with their meaning in the context of the prior art, and should not be interpreted in an idealized or overly formal sense unless specifically defined.

[0048] For the sake of simplicity, the method embodiments are described as a series of actions. However, those skilled in the art should understand that the embodiments of the present invention are not limited to the described order of actions, because according to the embodiments of the present invention, some steps can be performed in other orders or simultaneously. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions involved are not necessarily essential to the embodiments of the present invention.

[0049] As can be seen from the above description of the embodiments, those skilled in the art can clearly understand that this application can be implemented by means of software plus necessary general-purpose hardware platforms. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in various embodiments or some parts of the embodiments of this application.

[0050] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions 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 narrowband internet of things (NB-IoT) based unmanned aerial vehicle (UAV) communication system, characterized in that, The application relates to a UAV communication node, a NB-IoT core network and base station, a cloud or edge intelligent control platform and a plurality of IoT terminals. The UAV communication node is provided with a NB-IoT radio frequency and baseband module, a position and attitude acquisition module, an onboard computing module and a storage module, the onboard computing module is configured to acquire three-dimensional position information of the UAV and channel quality, service load and time delay information reported by the IoT terminals, under the control of a pre-installed deep reinforcement learning trajectory planning model and a supervised learning resource configuration model in the storage module, three-dimensional flight trajectories of the UAV in a target area and transmission power, uplink repetition times, resource block quantities and reporting periods of the IoT terminals are determined, and the UAV communication node controls access and data forwarding of the IoT terminals through a NB-IoT air interface. The cloud or edge intelligent control platform is used for receiving operation data reported by the UAV communication node and the IoT terminals, training the trajectory planning model and the resource configuration model on-line or off-line, and delivering updated model parameters to the UAV communication node, so that the coverage rate, service time delay performance and energy utilization efficiency of the IoT terminals in a remote or blind area are improved. And the effective coverage terminal is determined based on the following coverage decision condition:

2. The system of claim 1, wherein, The unmanned aerial vehicle communication node is provided with a channel measurement module, which is configured to measure the channel at the decision moment According to the unmanned aerial vehicle three-dimensional coordinates and the Internet of Things terminal plane coordinates, the spatial distance between the two is calculated The spatial distance satisfies: ， wherein, denotes the spatial distance between the th unmanned aerial vehicle and the th Internet of Things terminal at the th decision moment; , , denote the coordinates of the th unmanned aerial vehicle in the ground coordinate system direction, direction and height direction at the th decision moment, respectively; , denote the coordinates of the th Internet of Things terminal in the ground coordinate system direction and direction, respectively.

3. The system of claim 1, wherein, The channel measurement module is further configured to estimate a path loss between the UAV and the IoT terminal using a log-distance path loss model that satisfies: ， wherein, denotes the path loss between the th unmanned aerial vehicle and the th Internet of Things terminal at the th decision instant, denotes the path loss constant at a reference distance denotes the path loss exponent, denotes the path loss exponent, denotes the reference distance used for normalization, is the spatial distance between the th unmanned aerial vehicle and the th Internet of Things terminal at the th decision instant.

4. The system of claim 3, wherein, The on-board computing module is further configured to calculate a received signal-to-noise ratio for each Internet of Things terminal according to the path loss , which satisfies: ，， The comprehensive utility satisfies: ， wherein, denotes the received signal-to-noise ratio of the i-th Internet of Things terminal at the j-th decision moment, denotes the transmission power of the i-th unmanned aerial vehicle at the j-th decision moment, denotes the path loss between the i-th unmanned aerial vehicle and the j-th Internet of Things terminal at the j-th decision moment, denotes the received end noise power, denotes the minimum acceptable signal-to-noise ratio threshold preset by the system, when a certain terminal satisfies the above inequality, the terminal is determined as an effective coverage terminal.​​​​​​​ 5. The system of claim 4, wherein, The on-board computing module is configured to calculate system coverage at each decision-making moment according to the number of effective coverage terminals and the energy consumption of the UAV and the overall utility wherein the system coverage rate satisfies: ， The application comprises the following steps: ， wherein, denotes the system coverage at the decision moment, denotes the number of valid coverage terminals satisfying the coverage decision condition in the decision moment, denotes the total number of Internet of Things terminals in the system; denotes the comprehensive utility value at the decision moment, denotes the average uplink delay of the system at the decision moment, denotes the energy consumption of the drone in the time period corresponding to the decision moment, denotes the total energy upper limit available to the drone, , , , and respectively denote the weight coefficients corresponding to the coverage, the delay, and the energy consumption.

6. The system of claim 5, wherein, The cloud or edge intelligent control platform is configured to train a drone trajectory planning policy using a deep reinforcement learning algorithm to maximize a utility As a reward signal, the policy network parameters are optimized to maximize a trajectory return objective function: ， wherein, represents the trajectory planning expected return when the policy parameters are represents the parameter vector of the deep reinforcement learning policy network, represents the upper limit of the decision steps of the trajectory planning, represents the discount factor in reinforcement learning, represents the comprehensive utility value at the th decision moment, represents the mathematical expectation operator of the environmental randomness and the policy randomness.​ 7. The system of claim 1, wherein, The cloud or edge intelligent control platform is further configured to train a supervised learning resource configuration model, construct a resource configuration output vector for each training sample and its optimal or near-optimal label vector and optimize the model parameters by the following mean square error loss function: ， in, This represents the loss function value of the resource allocation supervised learning model. This represents the total number of samples used for training. This indicates that the model represents the i-th terminal at the k-th decision time. Resource allocation output vectors on each training sample It represents the i-th terminal at the k-th decision time. The optimal or near-optimal resource allocation label vector corresponding to each training sample. This represents the 2-norm operation for calculating the Euclidean distance between two vectors. 8.A method for UAV communication based on narrowband internet of things, applied to the system of any one of claims 1 to 7, characterized in that, S1: initializing the deep reinforcement learning trajectory planning model and the supervised learning resource configuration model in the cloud or edge intelligent control platform, obtaining initial model parameters and delivering the initial model parameters to the UAV communication node; S2: the UAV communication node flies in the target area according to the initial trajectory, establishes connection with the plurality of IoT terminals through the NB-IoT air interface, and collects terminal channel quality, service load, time delay and UAV position, power and other operation data; S3: uploading the operation data to the cloud or edge intelligent control platform, training the trajectory planning model and the resource configuration model based on the operation data, and obtaining updated model parameters; S4: delivering the updated model parameters to the UAV communication node, outputting the next flight trajectory and transmission power from the trajectory planning model and outputting the uplink repetition times, resource block quantities and reporting periods of the IoT terminals from the resource configuration model according to the current state of the UAV communication node in the flight process, and controlling access and data scheduling of the IoT terminals. ​ 9. The method of claim 8, wherein, In step S2, the UAV communication node calculates the spatial distance between the UAV and each IoT terminal based on the collected UAV location and IoT terminal location information. and path loss In step S4, the received signal-to-noise ratio of each IoT terminal is calculated. And the coverage decision results, to determine the set of valid coverage terminals.

Images