Unmanned aerial vehicle traffic management method based on machine learning and decentralized sharing

Abstract

The invention provides an unmanned aerial vehicle traffic management method based on machine learning and decentralized sharing, and the method comprises the steps: allowing unmanned aerial vehicles to directly communicate with each other through a decentralized network, sharing the flight path information, and determining the spacing distance between the unmanned aerial vehicles; predicting a potential traffic jam area by using a machine learning model trained based on shared flight path information; using a traffic management module to adjust the flight path of the unmanned aerial vehicle to avoid the predicted congestion area; the adjusted flight path data and the actual congestion condition are collected in real time and input into a training set of the machine learning model, and the machine learning model is updated in real time. The decentration method allows the system to expand along with the increase of the number of unmanned aerial vehicles, has no single-point fault risk, can quickly adapt to changing traffic conditions by continuously updating a machine learning model, and can optimize a flight path and reduce travel time and energy consumption by predicting and avoiding congestion.

Description

Technical Field

[0001] The present invention belongs to the technical field of drones, and in particular relates to a drone traffic management method based on machine learning and decentralized sharing. Background Art

[0002] The use of drones for delivery, surveillance, and inspection has increased significantly in recent years, raising the question of how to manage drone traffic to avoid congestion and ensure safety. Traditional air traffic management systems, designed primarily for manned aircraft, may not be well-suited for drones, especially in low-altitude airspace. Existing drone traffic management systems typically rely on centralized servers to process flight data, which is inefficient and prone to single points of failure. Furthermore, these systems focus more on real-time collision avoidance rather than predicting and proactively managing traffic congestion. Summary of the invention

[0003] The purpose of this invention is to provide a drone traffic management method based on machine learning and decentralized sharing, aiming to solve the problems of low efficiency and proneness to single point failures in existing drone traffic management solutions.

[0004] The present invention is implemented as follows: a method for drone traffic management based on machine learning and decentralized sharing, the method comprising the following steps: Step S1: Allow drones to communicate directly with each other through a decentralized network, share flight path information, and determine the distance between drones; Step S2: using a machine learning model trained based on shared flight path information to predict potential traffic congestion areas; Step S3: Use the traffic management module to adjust the UAV flight path to avoid the predicted congested area; Step S4: Collect the adjusted flight path data and actual congestion conditions in real time, input them into the training set of the machine learning model, and update the machine learning model in real time.

[0005] A further technical solution of the present invention is: step S1 includes the following steps: Step S10: Each drone is considered a node in the network, equipped with a communication module, and communicates with each other using a communication protocol. Each drone generates a unique identifier; Step S11: The drone collects flight path information in real time through built-in sensors, broadcasts its flight path information to the network at fixed time intervals, and uses encryption technology to sign the data; Step S12: Each drone monitors the network and receives broadcast data from nearby drones. The data is stored in a local distributed database. The distance between drones is measured by Calculate, where d is the Euclidean distance between two drones, Lat1, Lon1, and Alt1 are the latitude, longitude, and altitude of the first drone respectively, and Lat2, Lon2, and Alt2 are the latitude, longitude, and altitude of the second drone respectively.

[0006] A further technical solution of the present invention is that in step S10, the communication protocol adopts a point-to-point communication protocol, or a protocol based on blockchain technology, or a distributed hash table protocol.

[0007] A further technical solution of the present invention is that in step S11, the flight path information includes the current position, speed, planned flight path, and timestamp.

[0008] A further technical solution of the present invention is that step S2 includes the following steps: Step S20: The machine learning model collects all shared flight path information, converts it into a time series data set, cleans the data, removes outliers, and normalizes the data. The standard value is calculated by the formula Calculate, where x norm is the standard value of the data, x is the original value, x min and x max are the minimum and maximum values of the data set; Step S21: Use a long short-term memory network model, input the features of the current position, speed, coordinate sequence of the planned path, and timestamp, and output the congestion probability P congestion ; Step S22: Each drone regularly runs the long short-term memory network model, inputs real-time shared data, and the model outputs the congestion probability of each airspace grid. The congestion definition formula is , where D(t) is the drone density in the area at time t, N(t) is the number of drones, and V is the volume of the area; Step S23: Set the density threshold D threshold , if D(t) > D threshold or P congestion > 0.7, then mark the area as a potential congestion area.

[0009] A further technical solution of the present invention is that step S3 includes the following steps: Step S30: The traffic management module obtains the current drone position, target position, and predicted congestion area, divides the airspace into a three-dimensional grid map, and each grid is marked as passable or congested; Step S31: Calculate the optimal path in the three-dimensional grid map, and the cost function , where g(n) is the path cost from the starting point to the current node n, h(n) is the heuristic estimate from n to the target, avoiding all grids marked as congested, and finally outputting an alternative path composed of a series of new coordinate points; Step S32: Send the new path to the drone and update its flight plan; Step S33: Broadcast the new path through a decentralized network to ensure that neighboring drones avoid conflicts, and use a time scheduling algorithm to stagger the arrival times.

[0010] A further technical solution of the present invention is: in step S32, if the congested area cannot be completely avoided, reduce the speed to reduce the density, and the new speed , where is an adjustment coefficient, is the current speed of the drone, and D(t) is the drone density in the area at time t.

[0011] A further technical solution of the present invention is: in step S32, if the congested area cannot be completely avoided, assign the drones to different altitude layers.

[0012] A further technical solution of the present invention is: step S4 includes the following steps: Step S40: Collect the adjusted flight path data and the actual congestion situation in real time, verify through density calculation, and add the new data to the training set of the machine learning model; Step S41: Use incremental learning technology to update the parameters of the long short-term memory network model, and update the model every certain time or when the data volume reaches the threshold; Step S42: Calculate the prediction accuracy rate, using the cross-entropy loss formula , where y i is the actual congestion label, is the predicted probability, and N is the number of samples.

[0013] The beneficial effects of the present invention are: the decentralized method allows the system to scale as the number of drones increases, without the risk of a single point of failure. The system can quickly adapt to changing traffic conditions by continuously updating the machine learning model. By predicting and avoiding congestion, the system can optimize the flight path, reduce travel time and energy consumption. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 is the main flowchart of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0015] In order to make the objectives, technical solutions and advantages of the present invention clearer and more understandable, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments.

[0016] The present invention provides a method for unmanned aerial vehicle (UAV) traffic management based on machine learning and decentralized sharing, and the method includes the following steps: Step S1: Allow UAVs to directly communicate with each other through a decentralized network, share flight path information, and determine the spacing distance between UAVs; Step S2: Use a machine learning model trained based on the shared flight path information to predict potential traffic congestion areas; Step S3: Use a traffic management module to adjust the UAV flight paths to avoid the predicted congestion areas; Step S4: Real-time collect the adjusted flight path data and actual congestion conditions, input them into the training set of the machine learning model, and update the machine learning model in real time.

[0017] A system applying the method for UAV traffic management based on machine learning and decentralized sharing includes a decentralized network that allows UAVs to share flight path information (such as position, speed, direction) in a peer-to-peer manner without relying on a central server; a machine learning model that predicts potential traffic congestion areas based on the shared flight path information, and the model is trained by identifying traffic patterns and trends; and a traffic management module that uses the prediction results of the machine learning model to adjust the UAV flight paths, avoid congestion areas, and optimize traffic flow.

[0018] The decentralized network allows UAVs to directly communicate with each other and share flight path information. It can be implemented through blockchain technology or other peer-to-peer network protocols. Each UAV acts as a network node and broadcasts data to nearby UAVs or the entire network. The shared data includes, but is not limited to: current position (latitude, longitude, altitude), speed (speed and direction), planned flight path, and timestamp, and this data provides insights into the current and future UAV traffic states for the machine learning model.

[0019] The machine learning model is designed to predict areas where traffic congestion may occur based on the shared flight path information. Algorithms such as deep neural networks, recurrent neural networks (RNNs), or long short-term memory networks (LSTMs) can be used, and these algorithms are suitable for processing time series data and capturing time dependencies. The model is trained using historical data containing traffic congestion instances and corresponding flight path information. During operation, the model processes real-time data from the decentralized network and predicts potential congestion in the near future.

[0020] The traffic management module adjusts the flight paths of drones using the prediction results of machine learning models. This can be achieved in the following ways: recommending alternative routes to drones heading towards predicted congested areas; dynamically adjusting the speed or altitude of drones to reduce traffic density in certain areas; coordinating the flight times of drones to stagger arrival times in busy areas. The module can use path planning algorithms (such as the A* or Dijkstra algorithms) to find optimized paths to avoid congestion and adopt optimization techniques to minimize the overall travel time or energy consumption.

[0021] The specific operation process of step S1 is as follows: First, initialize the decentralized network. Each drone is regarded as a node in the network and is equipped with a communication module (such as a wireless communication protocol, e.g., WiFi, LoRa, or 5G). The network adopts a peer-to-peer (P2P) communication protocol, which may be based on blockchain technology (such as Ethereum or Hyperledger) or a distributed hash table (DHT) protocol to ensure data integrity and decentralization characteristics. Each drone generates a unique identifier (ID) for identification and communication in the network; Secondly, perform data collection and formatting. Drones collect flight path information in real-time through built-in sensors (such as GPS, accelerometers, altimeters), including: current position: latitude (Lat), longitude (Lon), altitude (Alt); speed: magnitude (v) and direction (θ), planned flight path: a series of predetermined coordinate points (Lat i , Lon i， Alt i ), and timestamp (t). The data is formatted into a standard structure, such as JSON format; Then, perform data broadcasting. Drones broadcast their flight path information to the network at fixed time intervals (e.g., every second or every 5 seconds). Encryption technology (such as public key encryption) is used to sign the data to ensure that the data source is trustworthy and has not been tampered with. The data broadcasting range can be restricted to a certain geographical area (e.g., drones within a radius of 500 meters) to reduce communication overhead; Finally, perform data reception and storage. Each drone listens to the network and receives the broadcast data from nearby drones. The data is stored in a local distributed database (such as IPFS or local cache) for subsequent analysis; Calculation of the distance between drones (for determining the broadcast range): where d is the Euclidean distance between two drones, in kilometers or meters, Lat1, Lon1, and Alt1 are the latitude, longitude, and altitude of the first drone respectively, and Lat2, Lon2, and Alt2 are the latitude, longitude, and altitude of the second drone respectively.

[0022] The specific operation process of step S2 is as follows: First, perform data preprocessing. Collect all shared flight path information and convert it into a time series data set. Clean the data to remove outliers (such as records with speeds outside the reasonable range). Normalize the data. For example, normalize the position coordinates to the interval [0, 1]: , where x norm is the standardized value of the data, x is the original value, x min and x max are the minimum and maximum values of the data set; Secondly, select and train the model. Use a Long Short-Term Memory Network (LSTM) because it is good at processing time series data. The input features include: the current position (latitude, longitude, and altitude), speed (v, θ), the coordinate sequence of the planned path, and the timestamp (t). The output target is the congestion probability P within a future time window (such as the next 10 minutes). congestion . The training data is sourced from historical flight records, and congestion events are labeled (for example, when the density of drones in a certain area per unit time exceeds a threshold, such as 10 drones per cubic kilometer). Then, perform real-time prediction. Each drone (or computing node in the network) regularly runs the LSTM model and inputs real-time shared data. The model outputs the congestion probability for each airspace grid (for example, divided into 1km×1km×50m cubes).

[0023] Congestion definition formula: where D(t) is the drone density in a certain area at time t, N(t) is the number of drones, and V is the volume of the area; Finally, identify the congested areas. Set a density threshold D threshold (such as 15 drones per cubic kilometer). If D(t) > D threshold or P congestion > 0.7, mark this area as a potentially congested area.

[0024] The specific operation process of step S3 is as follows: First, prepare the path planning input. Obtain the current drone position, target position, and predicted congested areas. Divide the airspace into a three-dimensional grid map, and each grid is marked as "passable" or "congested".

[0025] Secondly, perform the path optimization algorithm. Use the A* algorithm to calculate the optimal path, and the cost function: Among them, g(n) is the path cost from the starting point to the current node n (such as distance or energy consumption), and h(n) is the heuristic estimate from n to the target. Commonly used heuristic estimate functions include Euclidean distance and Manhattan distance. Euclidean distance is applicable to path planning on a two-dimensional plane, while Manhattan distance is applicable to grid graphs. Selecting an appropriate heuristic function can improve search efficiency; the constraint is to avoid all grids marked as "congested", and finally output an alternative path composed of a series of new coordinate points; Then, a path adjustment strategy is carried out. Send the new path to the drone and update its flight plan. Speed adjustment: If it is impossible to completely avoid the congested area, reduce the speed to reduce density: , where, is an adjustment coefficient (such as 0.1), is the current speed of the drone, and D(t) is the density of drones in the area at time t. Altitude adjustment: Assign the drones to different altitude layers (such as one layer every 50 meters) to reduce conflicts in the same plane.

[0026] It is best to carry out path coordination. Broadcast the new path through a decentralized network to ensure that neighboring drones avoid conflicts. Use a time scheduling algorithm (such as the greedy algorithm) to stagger the arrival times.

[0027] The specific operation process of step S4 is as follows: First, collect the adjusted flight path data and the actual congestion situation in real time (verified by density calculation), and add the new data to the training set.

[0028] Then, use incremental learning techniques (such as online gradient descent) to update the LSTM model parameters. Update the model at regular intervals (such as every hour) or when the data volume reaches the threshold.

[0029] Calculate the prediction accuracy. For example, use cross-entropy loss, and the formula is as follows: , where, y i is the actual congestion label, is the predicted probability, and N is the number of samples.

[0030] The following uses two specific embodiments to illustrate the present solution.

[0031] Embodiment 1: Urban low-altitude delivery: Scenario: Ten delivery drones are flying over the city.

[0032] The drone broadcasts its position and path every second, with a network coverage radius of 1 km. The LSTM model predicts that the density in a certain area (Lat: 39.9050, Lon: 116.4080, Alt: 50m) will exceed 15 drones / km³ after 5 minutes. The traffic management module calculates a detour path (such as ascending to 100m) for the drones about to enter this area and reduces the speed of two drones to 4 m / s. In the urban low-altitude delivery scenario, multiple delivery drones share their flight paths and real-time position data through a decentralized network. The system analyzes the congestion status of the current airspace based on a machine learning model, predicts possible future traffic bottlenecks, and provides the best path planning for the drones about to fly. For example, when congestion is about to occur in a certain airspace, the system will adjust the flight paths of the drones in this airspace in advance to avoid collisions and traffic jams among the drones.

[0033] Embodiment 2: Disaster relief: Scenario: Twenty drones are performing tasks in the disaster area. The drones share data through a blockchain network to ensure that mission-critical data is not lost. The model identifies that the airspace near a certain rescue point will be congested within 10 minutes. Adjust the paths of five drones, allocate different altitude levels (50m, 100m, 150m), and delay the takeoff time of two drones. In the disaster relief scenario, multiple drones perform tasks in the disaster area simultaneously. The drones share their flight data through a decentralized network and provide real-time feedback on the mission progress. The system uses a machine learning model to analyze the traffic flow in the airspace and adjusts the flight paths of the drones according to actual needs to ensure the efficient progress of the rescue mission and avoid unnecessary traffic jams in the airspace.

[0034] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, and improvements made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A method for unmanned aerial vehicle traffic management based on machine learning and decentralized sharing, characterized in that, The method includes the following steps: Step S1: Allow direct communication between drones through a decentralized network, share flight path information, and determine the spacing distance between drones; Step S2: Use a machine learning model trained based on the shared flight path information to predict potential traffic congestion areas; Step S3: Use a traffic management module to adjust the drone flight path to avoid the predicted congestion areas; Step S4: Real-time collect the adjusted flight path data and actual congestion situation, input them into the training set of the machine learning model, and update the machine learning model in real-time.

2. The method for unmanned aerial vehicle traffic management based on machine learning and decentralized sharing according to claim 1, wherein, Step S1 includes the following steps: Step S10: Each drone is regarded as a node in the network, equipped with a communication module, communicates with each other using a communication protocol, and each drone generates a unique identifier; Step S11: The drone real-time collects flight path information through built-in sensors, broadcasts its flight path information to the network at a fixed time interval, and signs the data using encryption technology; Step S12: Each drone monitors the network and receives the broadcast data of nearby drones. The data is stored in the local distributed database. The distance between drones is calculated by where d is the Euclidean distance between two drones, Lat1, Lon1, and Alt1 are the latitude, longitude, and altitude of the first drone, and Lat2, Lon2, and Alt2 are the latitude, longitude, and altitude of the second drone.

3. The method for unmanned aerial vehicle traffic management based on machine learning and decentralized sharing according to claim 2, wherein In Step S10, the communication protocol uses a peer-to-peer communication protocol or a blockchain technology-based protocol or a distributed hash table protocol.

4. The method for unmanned aerial vehicle traffic management based on machine learning and decentralized sharing according to claim 2, wherein In Step S11, the flight path information includes the current position, speed, planned flight path, and timestamp.

5. The method for unmanned aerial vehicle traffic management based on machine learning and decentralized sharing according to claim 1, characterized in that, Step S2 includes the following steps: Step S20: The machine learning model collects all the shared flight path information, converts it into a time series dataset, cleans the data, removes outliers, and normalizes the data. The standard value is calculated by the formula where, x norm is the standard value of the data, x is the original value, x min and x max are the minimum and maximum values of the dataset; Step S21: Use a long short-term memory network model to input the current position, speed, coordinate sequence of the planned path, and timestamp of the feature, and output the congestion probability P within the future time window congestion ; Step S22: Each drone regularly runs a long short-term memory network model and inputs real-time shared data. The model outputs the congestion probability of each airspace grid. The congestion definition formula is , where D(t) is the drone density in the area at time t, N(t) is the number of drones, and V is the volume of the area; Step S23: Set the density threshold D threshold , if D(t) > D threshold or P congestion > 0.7, then mark this area as a potential congestion area.

6. The method for unmanned aerial vehicle traffic management based on machine learning and decentralized sharing according to claim 1, characterized in that, Step S3 includes the following steps: Step S30: The traffic management module obtains the current drone position, target position, and predicted congestion area, divides the airspace into a three-dimensional grid map, and each grid is marked as passable or congested; Step S31: Calculate the optimal path in the three-dimensional grid map, with the cost function , where g(n) is the path cost from the starting point to the current node n, h(n) is the heuristic estimate from n to the target, avoiding all grids marked as congested, and finally outputting an alternative path consisting of a series of new coordinate points; Step S32: Send the new path to the drone and update its flight plan; Step S33: Broadcast the new path through the decentralized network to ensure that neighboring drones avoid conflicts, and use a time scheduling algorithm to stagger the arrival times.

7. The method for unmanned aerial vehicle traffic management based on machine learning and decentralized sharing according to claim 6, wherein In step S32, if it is impossible to completely avoid the congested area, the speed is reduced to decrease the density, and the new speed , where is the adjustment coefficient, is the current speed of the UAV, and D(t) is the UAV density in the area at time t.

8. The method for unmanned aerial vehicle traffic management based on machine learning and decentralized sharing according to claim 6, wherein In Step S32, if the congestion area cannot be completely avoided, the drones are assigned to different altitude layers.

9. The method for unmanned aerial vehicle traffic management based on machine learning and decentralized sharing according to claim 1, wherein The said Step S4 includes the following steps: Step S40: Real-time collect the adjusted flight path data and actual congestion situation, verify through density calculation, and add the new data to the training set of the machine learning model; Step S41: Use incremental learning technology to update the long short-term memory network model parameters, and update the model at regular intervals or when the data volume reaches a threshold. Step S42: Calculate the prediction accuracy rate using the cross-entropy loss formula , where y i is the actual congestion label, is the predicted probability, and N is the number of samples.

Images