A safety-based aircraft path planning method

Abstract

The present invention proposes a safety-based aircraft path planning method in response to the key requirements of civil aviation airports for operational safety. Catboost, which uses short time series as input, is used as a prediction model to predict the spatial distribution of taxiing speed in the surface area, and then calculates the variable taxiing time. The method also includes a path planning algorithm for the optimal path, which uses an improved ant colony algorithm to find the optimal path that meets the safety risk conditions and complies with the operating rules between the starting point and the end point given by the user, based on the current road network status and the safety risk threshold conditions given by the user. The present invention performs path planning with the safety threshold configured by the user as the center, adds time window constraints to the sections and nodes on the path, and directly searches for the optimal path to avoid risks.

Description

Technical Field

[0001] The present invention relates to the field of path planning, and in particular to a safety-based aircraft path planning method. Background Art

[0002] At present, the taxiing arrangements for aircraft at domestic airports are adjusted at any time by tower controllers based on the real-time traffic conditions on the apron. Controllers generally plan taxiing paths for aircraft based on their personal experience, which is inefficient and lacks overall planning. They rely too much on manual scheduling, which poses a high safety risk.

[0003] In addition to the taxiing deployment of aircraft based on tower control personnel, there are also many domestic studies that apply path planning algorithms to plan paths for aircraft on the airport surface. However, most current path planning algorithms only consider goals such as the shortest distance and the shortest driving time, or additionally consider possible path conflict safety risks on the basis of these goals, and propose risk management plans. If safety risks are not handled promptly and correctly, accidents are likely to occur. These methods do not take safety as the first priority and goal when planning paths. Summary of the invention

[0004] In response to the above problems, the present invention proposes a safety-based aircraft path planning method, which integrates machine learning and path planning algorithms, aiming to provide optimal taxiing routes for airport ground aircraft under the premise of avoiding safety risks. The main contents to be solved are: prediction of aircraft variable taxiing time and aircraft path planning under the premise of avoiding safety risks.

[0005] A safety-based aircraft path planning method, the safety-based aircraft path planning method comprising a machine learning prediction model and an optimal path path planning algorithm, the machine learning prediction model adopts catboost with short time series as input as a prediction model to predict the spatial distribution of taxiing speed in a surface area, and then calculates the variable taxiing time; the optimal path path planning algorithm adopts an improved ant colony algorithm, between a starting point and an end point given by a user, according to the current road network status and the safety risk threshold condition given by the user, to find the optimal path that meets the safety risk conditions and complies with the operation rules.

[0006] Furthermore, a safety-based aircraft path planning method is provided, wherein the specific process of the safety-based aircraft path planning method comprises the following steps:

[0007] Step 1: Receive the aircraft taxiing start and end position information, receive the airport surface road network status information with a time window, and receive the user-given safety threshold;

[0008] Step 2: The machine learning model predicts the spatial distribution of aircraft taxiing speeds in different areas of the airport surface under current environmental conditions;

[0009] Step 3: Calculate the variable sliding time of the road sections in the road network according to the spatial distribution of sliding speed;

[0010] Step 4: Use the improved ant colony algorithm to search and plan an alternative route with the shortest taxiing time that meets the safety threshold conditions and avoids safety risks, and save the variable taxiing time windows of the sections and nodes on the route;

[0011] Step 5: The controller chooses to execute an alternative route;

[0012] Step 6: Record the executed path and the time windows of the segments and nodes on the path, and update the road network status based on the planned and executed path and aircraft position at the current moment.

[0013] Furthermore, in a safety-based aircraft path planning method, the construction and prediction of the machine learning prediction model includes the following sub-steps:

[0014] S1: Prepare a historical data training set, which can select typical airports or airport data that is easy to obtain;

[0015] S2: Split the training data set into a training set and a validation set, adjust the model parameters and train until the model converges and achieves acceptable prediction accuracy on the validation set;

[0016] S3: When predicting, the spatial distribution of taxiing speed in different areas of the airport surface is predicted using the data time series collected in real time at the current moment as input;

[0017] S4: estimating the taxiing time of each section in the road network according to the taxiing speed spatial distribution predicted in step S3 and the road network, and further estimating the variable taxiing time of the aircraft;

[0018] S5: Use the designed online training model update mechanism to add the actual collected data of the deployed airports to the training dataset at regular intervals, automatically retrain the model, and update the model.

[0019] Furthermore, in a safety-based aircraft path planning method, the improved ant colony algorithm described in step 4 further includes the following sub-steps:

[0020] Step A: Set algorithm parameters;

[0021] Step B: Initialize the pheromone matrix τ ij (0);

[0022] Step C: Set the number of algorithm iterations to loop for each ant in the ant colony, place the current ant at the starting point, update the pheromone, calculate the state transition probability and select the next node according to the roulette strategy, introduce heuristic information when selecting the next edge and update the state transition probability formula, and store the current route and evaluation in the result set;

[0023] Step D: Release and volatilize pheromones according to the quality of all solutions in the result set, and update the pheromone matrix τ ij (t).

[0024] Furthermore, a safety-based aircraft path planning method, the calculation formula for updating pheromones in the improved ant colony algorithm is:

[0025] τ ij (n+1)=(1-ρ)*τ ij (n)+△τ ij

[0026] Among them, τ ij is the amount of pheromone on edge (i, j), ρ is the pheromone volatility coefficient, and n represents the number of road sections in the airport road network.

[0027] Furthermore, in a safety-based aircraft path planning method, the calculation formula for calculating the state transition probability is:

[0028]

[0029] Among them, p ij represents the probability of selecting the next node j from node i, the set next represents the set of nodes that can be reached from point i in the next step, and x represents the node visited by the ant.

[0030] Furthermore, in a safety-based aircraft path planning method, the heuristic information evaluates the selectable edges according to the characteristics of the problem to be solved, and provides guidance to the ants when making decisions with certain weights; the calculation formula of the heuristic information is:

[0031]

[0032] Among them, η ijt represents the inspiration information from point i to point j at time t, t ij represents the sliding time from point i to point j, represents the shortest distance from point j to the destination, C ijt It represents the conflict penalty function value from point i to point j at time t, and a, b, and c are the weights of each heuristic information.

[0033] Furthermore, a safety-based aircraft path planning method updates the state transition probability formula as follows:

[0034]

[0035] Among them, α is the pheromone factor and β is the heuristic information factor.

[0036] Beneficial effects of the present invention: The present invention proposes an optimal path planning method with safety as the goal, in response to the key requirements of civil aviation airports for operational safety. The designed machine learning algorithm can more accurately predict the spatial distribution of taxiing speeds in the airport surface area, and thus more accurately predict the variable taxiing time of the aircraft. The designed risk-avoiding optimal path planning algorithm with time window constraints can provide the aircraft with the optimal path under current conditions, ensuring safe and coherent taxiing of the aircraft. Safety-centered path planning, with safety thresholds configured by the user. Through the time window constraints of the sections and nodes on the path, the optimal path that avoids risks is directly searched. The application of machine learning models to real-time prediction of taxiing speeds in the airport surface area improves the accuracy of the estimation of the variable taxiing time of the aircraft. The taxiing speed prediction model is highly flexible and can be learned online and updated in a timely manner according to the different operating conditions of each airport. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 It is an overall flow chart of the safety-based aircraft path planning method.

[0038] Figure 2 It is a flowchart of constructing a prediction model in a safety-based aircraft path planning method.

[0039] Figure 3 This is a schematic diagram of the spatial distribution of taxiing speed on the scene predicted by the prediction model in the safe aircraft path planning method.

[0040] Figure 4 This is a flowchart of the improved ant colony algorithm in the path planning algorithm of the safety-based aircraft path planning method. DETAILED DESCRIPTION

[0041] In order to have a clearer understanding of the technical features, purposes and effects of the present invention, specific embodiments of the present invention are now described with reference to the accompanying drawings.

[0042] The present invention proposes a safety-based aircraft path planning method, including a machine learning prediction model and an optimal path path planning algorithm. The machine learning prediction model uses catboost with short time series as input as a prediction model to predict the spatial distribution of taxiing speed in a surface area, and then calculates the variable taxiing time; the optimal path path planning algorithm uses an improved ant colony algorithm to find the optimal path that meets the safety risk conditions and complies with the operating rules between a starting point and an end point given by a user, according to the current road network status and the safety risk threshold conditions given by the user.

[0043] During the initial path planning (for one aircraft), all sections of the airport surface road network are unoccupied, all sections and nodes are available, and the path planning is the same as the ordinary path planning. In this embodiment, during the non-initial path planning (for one aircraft), some sections and nodes in the airport road network have been occupied or will be occupied in a certain period of time in the future. The time window of the sections and nodes being occupied constitutes the constraint of this path planning. Under the current road network status, the specific process of the aircraft path planning method is as follows: Figure 1 As shown:

[0044] Step 1: Receive the aircraft taxiing start and end position information, receive the airport surface road network status information with a time window, and receive the user-given safety threshold;

[0045] Step 2: The machine learning model predicts the spatial distribution of aircraft taxiing speeds in different areas of the airport surface under current environmental conditions;

[0046] Step 3: Calculate the variable sliding time of the road sections in the road network according to the spatial distribution of sliding speed;

[0047] Step 4: The ant colony algorithm searches and plans an alternative route with the shortest taxiing time that meets the safety threshold conditions and avoids safety risks, and saves the variable taxiing time windows of the sections and nodes on the route;

[0048] Step 5: The controller chooses to execute an alternative route;

[0049] Step 6: Record the executed path and the time windows of the segments and nodes on the path. Update the network status according to the currently planned and executed paths and aircraft positions;

[0050] Repeat steps 1 to 6 the next time you plan.

[0051] In this embodiment, the digital road network is the basis of the entire algorithm. Runways, service lanes, and taxiways are represented as arcs in the airport road network. The road arc data attributes include parameter information such as name, starting node number, ending node number, road length, road direction, whether it is passable, and road type. A specific aircraft taxiway arc contains at least two nodes, and the nodes can uniquely identify the starting, ending, turning, and other parameter information of the taxiway. The road network status refers to whether the road sections and nodes in the road network are currently available, whether they will be available in the future, whether they are currently occupied, whether they will be occupied in the future, and the time window for occupation and release. In this embodiment, the safety risk threshold is defined as the time interval between two consecutive aircraft passing through the same road section or node. The larger the time interval, the smaller the safety risk; the smaller the time interval, the greater the safety risk. For example, the time for aircraft A to pass through section 1 at time t1 is 5 minutes (calculated by the length of the section and the taxiing speed), and the safety threshold is set to 1 minute. Then, aircraft B starts to pass through the same section at time t>t1+1, and ends passing through the section at t>t1+1+5, which is safe. If the safety threshold is set to 2 minutes, then aircraft B starts to pass through the same section at time t>t1+2, and ends passing through the section at t>t1+2+5, which is safe. The safety threshold is determined by the airport operation and dispatching department according to the actual operation of the airport. In addition to being limited by the regulations of the operation department and the aircraft's operating manual, the taxiing speed of the aircraft is also affected by environmental factors. For example, the airport's busyness, weather, visibility, airport road conditions, etc. These factors vary greatly even within a day. Therefore, accurately predicting the taxiing speed of aircraft taxiing in different areas of the airport under different environments and time periods is of great significance for rationally planning aircraft paths and accurately estimating the variable taxiing time of aircraft. In this embodiment, a spatial distribution prediction model of airport surface taxiing speed based on the boosting algorithm is designed, and on this basis, the variable taxiing time of the aircraft from the starting point to the destination is calculated. If there are links such as fixed-point deicing, the variable taxiing time needs to be added to the deicing time accordingly. The combined experience of business personnel, weather conditions, airport busyness, and whether there are field maintenance and emergencies have a significant impact on the aircraft taxiing speed. These factors can be described by the characteristic quantities in Table 1.

[0052] Table 1. Characteristic quantities of external influencing factors

[0053]

[0054] In this embodiment, catboost in the boosting algorithm is selected as the prediction model. Mainly considering the interpretability of the model and the simultaneous existence of continuous variables and categorical variables in the feature vector. Whether there are emergencies in the region belongs to typical categorical features. Although features such as visibility and precipitation are continuous variables, they are only graded in some airports. For example, visibility can be divided into excellent visibility, good visibility, poor visibility, etc. In order to cope with the situation of different airports and maintain the flexibility of the algorithm, the catboost algorithm that can better handle feature data containing both continuous variables and categorical variables is selected. After the model is trained, the importance of the model prediction standard and related influencing factors can be intuitively explained by sorting the importance of the features. The taxiing speed at the current moment is not only affected by the environmental factors at the current moment, but also by historical factors. For example, if the rainfall reaches the level of heavy rain in the past hour, water will accumulate on the road surface, which will definitely affect the taxiing speed of the aircraft under the safe taxiing conditions at the current moment. But this influence will not exist for a long time. Therefore, in the embodiment, in order to fully mine historical data information, a short time series (the past hour, 5 minutes interval) is used as the input of the machine learning algorithm.

[0055] In order to predict the spatial distribution of taxiing speed, the entire airport area is partitioned. The airport is divided into domestic terminal area, international terminal area, remote parking area, runway area, taxiing area, etc. according to the operating specifications. The regional division needs to be combined with the actual operation process of the airport, and the division granularity is determined by the airport business. The finer the granularity, the more detailed the spatial distribution of taxiing speed. In practical applications, for each divided area, a high-dimensional feature vector sequence and a speed vector sequence are created.

[0056] x=([temperature, visibility, ..., regional emergencies] t1 ,...,

[0057] [Temperature, visibility, ..., regional emergencies] tn )

[0058] y=(v t1 ,...,v tn )

[0059] where t n is the time series number. Then we get the high-dimensional feature vector matrix and velocity matrix of the entire airport scene

[0060] X=[x1,...,x k ]

[0061] Y=[v1,...,v k ]

[0062] Where k is the airport area number. The (X, Y) sequence set constitutes the training data of the machine learning algorithm. By learning the data, a prediction model is constructed to predict the safe taxiing speed of each area in real time. The algorithm process is as follows: Figure 2 As shown:

[0063] S1: Prepare a historical data training set, which can select typical airports or airport data that is easy to obtain;

[0064] S2: Split the training data set into a training set and a validation set, adjust the model parameters and train until the model converges and achieves acceptable prediction accuracy on the validation set;

[0065] S3: When predicting, the spatial distribution of taxiing speed in different areas of the airport surface is predicted using the data time series collected in real time at the current moment as input;

[0066] S4: estimating the taxiing time of each section in the road network according to the taxiing speed spatial distribution predicted in step S3 and the road network, and further estimating the variable taxiing time of the aircraft;

[0067] S5: Design an online training model and update the model mechanism. Add the actual collected data of the deployed airport to the training data set at regular intervals, automatically retrain the model, and update the model. Figure 3 shown.

[0068] The present invention proposes to use an improved ant colony algorithm to plan aircraft paths with the goal of minimizing travel time and avoiding conflicts. In this embodiment, during the calculation process, each ant will release pheromones on the path it passes through, which is usually inversely proportional to the total length of the path passed by the ant. Therefore, the pheromone increment on any edge can be expressed as the sum of the pheromones released by all ants on the edge in the current cycle. At the same time, pheromones will continue to evaporate over time, so the original pheromones need to be evaporated after each cycle. The update of pheromones can be expressed as follows:

[0069] τ ij (n+1)=(1-ρ)*τ ij (n)+△τ ij

[0070] Based on the pheromones left by the historical ant colony, when a new ant is looking for its own path, it will randomly select with a certain probability according to the concentration of pheromones on each edge. The higher the pheromone concentration, the greater the probability of selecting this edge. The specific selection method usually adopts the roulette selection method. The state transition probability calculation formula for selecting any edge can be defined as:

[0071]

[0072] where p ij represents the probability of selecting the next node j from node i, and the set next represents the set of nodes that can be reached from point i in the next step. This method can ensure that the entire selection process is random, and the edges with high pheromone concentration can be selected with a higher probability, thereby ensuring the global search and convergence of the algorithm.

[0073] With the continuous updating of pheromones, the entire environment can eventually reach a stable state. At this time, the path with the highest pheromone concentration is the optimal path. However, this random search is inefficient and the algorithm converges slowly. To improve the search efficiency, the concept of heuristic information can be introduced when selecting the next edge. This information evaluates the selectable edges according to the characteristics of the problem being solved, and guides the ants in making decisions with a certain weight, thereby improving the algorithm search efficiency. For this problem, the heuristic information can include the travel time between two points, potential conflicts, and the distance from the next point to the end point.

[0074] Considering the variable taxi time, the travel time of a road section is also likely to be proportional to the distance. Therefore, the shortest distance between any two points can be calculated in advance based on the airport road network data, so that the shortest distance from the next feasible point to the destination can be used as one of the heuristic information. The calculation of the heuristic information of any edge can be described as follows:

[0075]

[0076] η ijt represents the inspiration information from point i to point j at time t, t ij represents the sliding time from point i to point j, represents the shortest distance from point j to the destination, C ijt represents the conflict penalty function value from point i to point j at time t, and a, b, c are the weights of each heuristic information. Based on the above heuristic information, the state transition probability formula can be updated as follows:

[0077]

[0078] Where α is the pheromone factor, and β is the heuristic information factor / visibility. In summary, the ant colony algorithm process steps designed in this embodiment are as follows: Figure 4 Shown are:

[0079] Step A: Set algorithm parameters;

[0080] Step B: Initialize the pheromone matrix τ ij (0);

[0081] Step C: Set the number of algorithm iterations to loop each ant in the ant colony, place the current ant at the starting point, update the taboo table and accessible table, calculate the state transition probability and select the next node according to the roulette strategy, and store the current route and evaluation in the result set;

[0082] Step D: Release and volatilize pheromones according to the quality of all solutions in the result set, and update the pheromone matrix τ ij (t).

[0083] In this embodiment, the path planned for each aircraft has a time window. After the controller chooses to execute a planned path, the path and the time window of each section and node on the path will be recorded at the same time and updated to the road network. The resources of the sections and nodes that the aircraft has passed through are automatically released, the time windows are deleted, and the resources can be used by other aircraft. If an emergency occurs, the aircraft, section, and node where the emergency occurs are frozen, and the resources are occupied until the emergency is handled and the resources are released again. When an aircraft is traveling along a planned route, it may need to change the route or destination for some reason. At this time, the time window of the original planned path of the aircraft is immediately deleted, and the resources are immediately released. The path of the aircraft is re-planned. When the controller chooses to execute the path, the newly planned path and the time window it carries are automatically added to the road network, and the road network status is updated immediately.

[0084] In addition, the risks avoided by the present invention include several categories: (1) intersection conflict risk. An intersection refers to a position where aircraft from different directions of travel intersect each other at a large angle (or close to 90 degrees). On a road, this intersection is generally at the intersection between roads, such as a crossroads. When the time interval between two aircraft passing through this intersection is less than a certain time, a collision accident between vehicles will occur. (2) Head-on collision risk. A head-on collision refers to a dangerous situation where two aircraft are traveling in opposite directions on the same road, which may eventually lead to a head-on collision. On an airport taxiway, each taxiway is a lane, and two aircraft cannot be allowed to run in parallel. Therefore, when two aircraft occupy the same taxiway and are traveling in each other's direction, it is impossible for the two aircraft to avoid conflict at this time. (3) Rear-end collision risk. A rear-end collision refers to a situation where two aircraft are traveling in the same direction on the same road, and the speed of the rear aircraft is greater than that of the front aircraft, which may cause the nose of the rear aircraft to collide with the tail of the front aircraft. The taxiing speeds of different aircraft at an airport may be different. Although two aircraft occupying the same road and traveling in the same direction will not cause head-on collisions or cross-collisions, and the risk is relatively low, collisions are still possible. When the pilot makes mistakes in controlling the distance and relative speed between the two aircraft, danger may occur. (4) Runway runway risk. Runway runway risk refers to the risk of an aircraft sliding off the road when it is traveling at a speed that exceeds the safe taxiing speed of the current road section.

[0085] The present invention proposes a safety-based aircraft optimal path planning model and method in response to the key requirements of civil aviation airports for operational safety. The designed machine learning model can more accurately predict the spatial distribution of taxiing speed in the airport surface area, and then more accurately predict the variable taxiing time of the aircraft. The designed risk-avoiding optimal path planning algorithm with time window constraints can provide the aircraft with the optimal path under current conditions to ensure safe and coherent taxiing of the aircraft. The present invention performs path planning with safety as the center, and the safety threshold is configured by the user. Through the time window constraints of the sections and nodes on the path, the optimal path that avoids risks is directly searched. The machine learning model is applied to the real-time prediction of the taxiing speed in the airport surface area, which improves the accuracy of the estimation of the variable taxiing time of the aircraft. The taxiing speed prediction model is highly flexible and can be learned online and updated in time according to the different operating conditions of each airport.

[0086] The above shows and describes the basic principles and main features of the present invention and the advantages of the present invention. It should be understood by those skilled in the art that the present invention is not limited to the above embodiments. The above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention to be protected. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. A safety-based aircraft path planning method, characterized in that: The safety-based aircraft path planning method includes a machine learning prediction model and an optimal path planning algorithm. The machine learning prediction model uses catboost with short time series as input as a prediction model to predict the spatial distribution of taxiing speed in the surface area, and then calculates the variable taxiing time; the optimal path planning algorithm uses an improved ant colony algorithm to find the optimal path that meets the safety risk conditions and complies with the operating rules between the starting point and the end point given by the user according to the current road network status and the safety risk threshold conditions given by the user; the specific process of the safety-based aircraft path planning method includes the following steps: Step 1: Receive the aircraft taxiing start and end position information, receive the airport surface road network status information with a time window, and receive the user-given safety threshold; Step 2: The machine learning model predicts the spatial distribution of aircraft taxiing speeds in different areas of the airport surface under current environmental conditions; Step 3: Calculate the variable sliding time of the road sections in the road network according to the spatial distribution of sliding speed; Step 4: Using the improved ant colony algorithm to search and plan an alternative route with the shortest taxiing time that meets the safety threshold conditions and avoids safety risks, and saves the variable taxiing time windows of the sections and nodes on the route; the improved ant colony algorithm also includes the following sub-steps: Step A: Set algorithm parameters; Step B: Initialize the pheromone matrix τ ij (0); Step C: Set the number of algorithm iterations to loop for each ant in the ant colony, place the current ant at the starting point, update the pheromone, calculate the state transition probability and select the next node according to the roulette strategy, introduce heuristic information when selecting the next edge and update the state transition probability formula, and store the current route and evaluation in the result set; Step D: Release and volatilize pheromones according to the quality of all solutions in the result set, and update the pheromone matrix τ ij (t); The calculation formula of the updated pheromone is: t ij (n+1)=(1-ρ)*τ ij (n)+△τ ij Among them, τ ij is the amount of pheromone on edge (i, J), ρ is the pheromone volatility coefficient, and n represents the number of road sections in the airport road network; The calculation formula for calculating the state transition probability is: Among them, p ij represents the probability of selecting the next node j from node i, the set next represents the set of nodes that can be reached from point i in the next step, and x represents the node visited by the ant; The heuristic information evaluates the selectable edges according to the characteristics of the problem being solved, and guides the ants in making decisions with certain weights; the calculation formula of the heuristic information is: Among them, η ijt represents the inspiration information from point i to point j at time t, t ij represents the sliding time from point i to point j, represents the shortest distance from point j to the destination, C ijt represents the conflict penalty function value from point i to point j at time t, and a, b, and c are the weights of each heuristic information; The update state transition probability formula is: Among them, α is the pheromone factor, and β is the heuristic information factor; Step 5: The controller chooses to execute an alternative route; Step 6: Record the executed path and the time windows of the segments and nodes on the path, and update the road network status based on the planned and executed path and aircraft position at the current moment.

2. A safety-based aircraft path planning method according to claim 1, characterized in that: The construction and prediction of the machine learning prediction model includes the following sub-steps: S1: Prepare a historical data training set, which can select typical airports or airport data that is easy to obtain; S2: Split the training data set into a training set and a validation set, adjust the model parameters and train until the model converges and achieves acceptable prediction accuracy on the validation set; S3: When predicting, the spatial distribution of taxiing speed in different areas of the airport surface is predicted using the data time series collected in real time at the current moment as input; S4: estimating the taxiing time of each section in the road network according to the taxiing speed spatial distribution predicted in step S3 and the road network, and further estimating the variable taxiing time of the aircraft; S5: Use the designed online training model update mechanism to add the actual collected data of the deployed airports to the training dataset at regular intervals, automatically retrain the model, and update the model.

Images