Control support device

Abstract

The present disclosure pertains to a control support device, the device comprising: a fluctuation factor extraction unit that extracts, from meteorological data and flight data of an aircraft, data relating to a fluctuation factor in the flight time of the aircraft to a predetermined point; a feature quantity extraction unit that extracts, from the data relating to the fluctuation factor in the flight time, a plurality of feature quantities for predicting the flight time to the predetermined point; a database that stores the plurality of feature quantities; a flight time prediction unit that constructs a flight time prediction model by means of machine learning using the plurality of feature quantities, and predicts the flight time using the flight time prediction model; and a trajectory prediction unit that predicts the trajectory of the aircraft using the flight time prediction results. The fluctuation factor extraction unit statistically extracts a factor of fluctuation in the flight time of the aircraft by means of machine learning from past flight data and meteorological data to obtain the data relating to the fluctuation factor in the flight time, and the trajectory prediction unit selects a representative trajectory from a plurality of items of trajectory data accumulated in the past on the basis of the flight time prediction results, and outputs the selected trajectory as a predicted trajectory of the aircraft.

Description

Control support equipment

[0001] The present disclosure relates to a control assistance device, and more particularly to a control assistance device that improves the accuracy of predicting aircraft flight times.

[0002] In aircraft traffic management using four-dimensional navigation, aircraft mass is an important parameter for accurately predicting flight time. However, since air traffic control agencies typically do not have access to aircraft mass information, it is not possible to utilize this information in air traffic control. Patent Document 1 discloses a technology for deriving aircraft mass from data extracted from aircraft flight parameters from a flight model and flight data obtained by a device that accepts flight data, and predicting flight time based on this mass and flight parameters. It also discloses a technology for deriving flight distance from a flight path and deriving flight time based on the flight distance and true ground speed.

[0003] Patent Literature 2 discloses a technology that uses a neural network to input accumulated aircraft position information as learning data to build a prediction model with a learning function for predicting the future position of an aircraft, and then predicts the future position (predicted position) of the aircraft by combining the acquired aircraft position information with the prediction model. An alarm is issued if the predicted position and the actual acquired position deviate by more than a threshold. It also discloses a technology that predicts a route using the acquired position and the prediction model, and predicts the future position based on that. An alarm is issued if the predicted position and the observed value deviate by more than a threshold.

[0004] JP 2021-170253 A Patent No. 4355833 A

[0005] The technology disclosed in Patent Document 1 derives an aircraft's flight time based on information on wind direction and wind speed and the aircraft's mass. However, aircraft also change their flight route depending on weather conditions, airspace congestion, and other factors, which can change the flight time. These factors are not taken into account, and it is believed that accurate prediction of flight time is difficult.

[0006] The technology disclosed in Patent Document 2 discloses that the prediction model is changed to determine the prediction range taking into account weather, time fluctuations, day of the week fluctuations, and seasonal fluctuations. However, there is no detailed disclosure of when the model should be switched, and it is thought that it would be difficult to accurately predict flight time from the predicted position.

[0007] The present disclosure has been made to solve the above-mentioned problems, and aims to provide a control assistance device that can predict the flight time of an aircraft with high accuracy.

[0008] a database for storing the extracted plurality of feature quantities; a flight time prediction model for predicting the flight time using the plurality of feature quantities through machine learning; a flight time prediction unit for predicting the flight time using the flight time prediction model; and a trajectory prediction unit for predicting a trajectory of the aircraft using the flight time prediction result, wherein the variation factor extraction unit statistically extracts factors that may have an influence on fluctuations in the flight time of the aircraft from the past flight data and the weather data through machine learning, and uses the extracted plurality of feature quantities as the data related to the flight time variation factors;

[0009] According to the air traffic control assistance device of the present disclosure, multiple feature amounts for predicting flight time to a predetermined point are extracted from data on factors that cause fluctuations in flight time, and a flight time prediction model is constructed by machine learning using the multiple feature amounts in a trajectory prediction unit, and flight time is predicted using the flight time prediction model, thereby enabling highly accurate prediction of aircraft flight time. Furthermore, predictions can take into account the control habits of air traffic controllers, such as detouring when traffic is congested and taking shortcuts when traffic is clear.

[0010] 1 is a functional block diagram showing the configuration of an air traffic control assistance device of embodiment 1. FIG. 2 is a flowchart explaining the operation of the air traffic control assistance device of embodiment 1 during learning. FIG. 3 is a flowchart explaining the operation of the air traffic control assistance device of embodiment 1 during prediction. FIG. 4 is a diagram explaining the concept of trajectory prediction. FIG. 5 is a diagram explaining the concept of a modified trajectory prediction. FIG. 6 is a diagram showing feature quantities extracted from continuous position information of a plurality of aircraft. FIG. 7 is a diagram showing feature quantities extracted from weather data observed for each airport. FIG. 8 is a diagram showing feature quantities extracted from radar information from the Japan Meteorological Agency. FIG. 9 is a diagram showing feature quantities extracted from information assigned to an aircraft. FIG. 10 is a diagram showing data stored in a feature database. FIG. 11 is a diagram showing an example of extracting a histogram by value magnitude for each divided area. FIG. 12 is a diagram schematically showing a case where a waypoint is not passed due to trajectory variation. FIG. 13 is a diagram schematically showing a process of predicting flight time using a created prediction model. FIG. 14 is a diagram schematically showing trajectory prediction in a trajectory prediction unit. FIG. 15 is a functional block diagram showing the configuration of an air traffic control assistance device of embodiment 2. FIG. 16 is a diagram schematically showing a process of predicting trajectory and flight time in the air traffic control assistance device of embodiment 2. FIG. 17 is a flowchart explaining the operation of the air traffic control assistance device of embodiment 2 during learning. 1 is a flowchart illustrating the operation of an air traffic control assistance device during prediction according to a second embodiment. FIG. 2 is a diagram schematically illustrating trajectory prediction and flight time prediction in an air traffic control assistance device according to a second embodiment. FIG. 3 is a functional block diagram illustrating the configuration of an air traffic control assistance device according to a third embodiment. FIG. 4 is a diagram schematically illustrating the processing of trajectory classification and flight time prediction in an air traffic control assistance device according to a third embodiment. FIG. 5 is a flowchart illustrating the operation of an air traffic control assistance device during learning according to a third embodiment. FIG. 6 is a flowchart illustrating the operation of an air traffic control assistance device during prediction according to a third embodiment. FIG. 7 is a diagram illustrating a hardware configuration for realizing an air traffic control assistance device according to the first to third embodiments according to the present disclosure.

[0011] <Embodiment 1> <Device Configuration> When an aircraft approaches an airport, an airport controller controls traffic based on information about the aircraft, and important information includes the trajectory and flight time of the aircraft to the airport. A control support device according to the present disclosure is a device that supports air traffic control by predicting the trajectory and flight time of the aircraft to a predetermined point such as an airport and providing the information to the controller.

[0012] 1 is a functional block diagram showing the configuration of an air traffic control assistance device 100 according to embodiment 1. As shown in FIG. 1, the air traffic control assistance device 100 has a flight data receiving unit 1 that receives various flight data related to aircraft flight, and a weather data receiving unit 2 that receives various weather data related to weather. The flight data and weather data are input to a variation factor extraction unit 3 that extracts data related to factors that cause flight time to vary, and the factors that cause flight time to vary are extracted.

[0013] The extracted factors that cause fluctuations in flight time are input to a feature extraction unit 4 that extracts features for constructing a flight time prediction model that predicts flight time, and the extracted features are stored in a feature database 5.

[0014] The features stored in the feature database 5 are used by the flight time prediction unit 6 to construct and predict a flight time prediction model using machine learning, and the flight time until the aircraft lands at an airport or the flight time to a predetermined waypoint, a point where air traffic control instructions are likely to occur, a point where wind direction changes, and a waypoint which is a waypoint on the standard route are predicted. Standard routes include the Standard Instrument Arrival Route (STAR) and the Standard Instrument Departure (SID), and there are multiple standard routes for each airport.

[0015] The flight time prediction result is input to the trajectory prediction unit 7 and used to predict the trajectory of the aircraft. The trajectory prediction unit 7 selects and outputs a representative trajectory from track data accumulated in the past based on the flight time prediction result and other feature quantities.

[0016] The predicted flight time and trajectory of the aircraft are output to an air traffic control system and displayed via an output unit 8 and are used for air traffic control.

[0017] <Operation> Next, the operation of the air traffic control assistance device 100 will be described using the flowcharts shown in Figures 2 and 3, while referring to Figure 1. Figure 2 is a flowchart explaining the operation during learning using artificial intelligence (AI) in the air traffic control assistance device 100, and Figure 3 is a flowchart explaining the operation during prediction, which predicts flight time using a prediction model created during learning.

[0018] First, the operation during learning will be described with reference to Figure 2. When the operation starts, the fluctuation factor extraction unit 3 extracts data related to fluctuation factors of flight time from the flight data received by the flight data receiving unit 1 and the weather data received by the weather data receiving unit 2 (step S1).

[0019] The variation factor extraction unit 3 uses AI and machine learning technology to statistically extract factors that are thought to have the greatest influence on variations in aircraft flight time from a large amount of past data. For example, factors that are thought to cause variations in required flight time and flight path include changes in aircraft flight pattern due to weather conditions such as wind direction and wind strength, and changes in aircraft flight pattern due to the number of aircraft flying on the same route, i.e., congestion. However, it is difficult to detect all individual factors and their effects and obtain a mathematical formula (model) for predicting flight path and flight time from the factors, so variation factors are extracted using a statistical approach to big data processing using AI.

[0020] Next, the feature extraction unit 4 extracts feature values ​​from the multiple variation factors extracted by the variation factor extraction unit 3 (step S2). Here, feature values ​​refer to portions of the weather data and flight data of surrounding aircraft acquired by the flight data receiving unit 1 and the weather data receiving unit 2, respectively, that are useful for predicting the required flight time and flight path. While weather data and flight data of surrounding aircraft are collected as a wide range of data, the portions that are actually useful for predicting the required flight time and flight path of the aircraft are limited. In other words, the useful portions, i.e., feature values, vary depending on the aircraft's position at the start of the prediction, and a process is performed to extract the portions to be used for prediction from the acquired data. For example, if the variation factor is weather data, the weather data from the prediction start position to the airport is useful from the acquired weather data, while weather data before the prediction start position is unnecessary. Therefore, the process of extracting weather data from the prediction start position to the airport is the process of extracting feature values, and the processing circuit that performs this programmatic extraction process is referred to as the feature extraction unit 4.

[0021] When the fluctuation factor is flight data of nearby aircraft, the extracted feature amount may include flight data of aircraft on the same standard route or other merging routes.

[0022] The features extracted by the feature extractor 4 are stored in the feature database 5 (step S3).

[0023] The feature database 5 stores a large number of feature quantities (explanatory variables) and their corresponding flight times (objective variables), and the flight time prediction unit 6 associates and learns the explanatory variables and the objective variables through regression analysis using supervised learning, and statistically processes these past data to create a prediction model that obtains the objective variable, flight time, from the explanatory variables. That is, as described in the variation factor extraction unit 3, a flight time prediction model is constructed using AI and machine learning technology from a large amount of past data (step S4).

[0024] For example, there are causal relationships such as bad weather conditions on an air route causing a detour and increasing flight time, but it is difficult to grasp all of these causal relationships and create a flight time prediction model that appropriately reflects them. Therefore, through statistical processing, machine learning is used to find relationships in which the values ​​of the explanatory variables have a strong influence on the values ​​of the objective variables, and a model is created that calculates the objective variables from the explanatory variables. This allows for the rapid construction of flight time prediction models.

[0025] Next, the trajectory prediction unit 7 selects a representative trajectory corresponding to the past flight time and approach position (step S5). Here, the approach position refers to the position when the aircraft arrives within 50 nautical miles (NM) from the airport, for example, when predicting the flight time and trajectory from 50 NM from the airport to landing. A representative trajectory is selected from multiple trajectories that the aircraft has taken in the past from the approach position to the airport, and is prepared for prediction.

[0026] Next, the operation during prediction will be described with reference to Figure 3. When the operation starts, the variation factor extraction unit 3 extracts data related to the variation factors of flight time from the flight data received by the flight data receiving unit 1 and the weather data received by the weather data receiving unit 2 (step S11). This process is the same as the process of step S1 shown in Figure 2, but since it is applied to flights or weather conditions different from those during learning, the extracted data will be different as a result.

[0027] Next, the feature extraction unit 4 extracts feature amounts from the plurality of fluctuation factors extracted by the fluctuation factor extraction unit 3 (step S12). This process is also the same as the process of step S1 shown in Fig. 2, but the extracted feature amounts are feature amounts based on the current situation.

[0028] The feature amounts extracted by the feature amount extraction unit 4 are stored in the feature amount database 5 (step S13).

[0029] The flight time prediction unit 6 uses the flight time prediction model created during learning to predict the flight time by inputting the feature values ​​based on the current situation (step S14).

[0030] Next, the trajectory prediction unit 7 predicts the trajectory by selecting a representative trajectory corresponding to the past flight time and entry position based on the flight time and features predicted by the flight time prediction unit 6 (step S15).

[0031] This makes it possible to predict flight times that reflect trends in changes in flight time due to weather factors such as wind direction, wind speed, cloud cover, and rainfall, as well as airspace congestion. Machine learning is also possible that takes into account changes in aircraft speed due to wind direction and wind speed, making it possible to predict flight times that reflect changes in aircraft speed.

[0032] <Concept of Trajectory Prediction> Next, the concept of trajectory prediction will be described with reference to Fig. 4. As shown in Fig. 4, the air traffic control assistance device 100 of the first embodiment predicts a predicted trajectory PJ based on the predicted flight time and feature values ​​from a predicted position PP, which is a predetermined distance from the airport AP, to the airport AP, relative to the aircraft's originally planned trajectory TJ. Although the initially planned trajectory TJ and the predicted trajectory PJ are significantly different, the predicted trajectory PJ reflects the tendency of changes in flight time due to weather and airspace congestion.

[0033] In the first embodiment, the air traffic control assistance device 100 has been described as predicting a predicted trajectory PJ from a predicted position PP at a predetermined distance from the airport AP to the airport AP. However, as a variant, it is also possible to learn separately for each of a plurality of predetermined points, construct a flight time prediction model, and predict a trajectory for each predetermined point.

[0034] The concept of this modified trajectory prediction method will be explained using FIG. 5 . As shown in FIG. 5 , for example, a flight time prediction model (first flight time prediction model) is constructed from a predicted position PP1 (first position) located 100 NM (first distance) from the airport AP to the airport AP, and the trajectory prediction unit 7 predicts a predicted trajectory PJ1 (first predicted trajectory) based on the predicted flight time (first prediction result). The predicted trajectory PJ is significantly different from the initial planned trajectory TJ. Next, a flight time prediction model (second flight time prediction model) is constructed from a predicted position PP2 (second position) located 90 NM (second distance) from the airport AP to the airport AP, and the trajectory prediction unit 7 predicts a predicted trajectory PJ2 (second predicted trajectory) based on the predicted flight time (second prediction result). The predicted trajectory PJ2 is similar to the initial planned trajectory TJ and is output as the final predicted trajectory. In this way, by recalculating the predicted trajectory each time according to the actual route taken by the aircraft, the prediction can be corrected to accommodate deviations of the actual aircraft position from the predicted trajectory.

[0035] By adding features such as distance from the airport and learning each feature for each distance, it becomes possible to correct the prediction results for each specified point, thereby improving the accuracy of the predicted trajectory.

[0036] <Specific Examples of Feature Amounts> Next, specific examples of feature amounts are shown in Figs. 6 to 9. Fig. 6 shows feature amounts extracted from continuous position information of multiple aircraft. Fig. 6 shows, in table format from top to bottom, "time [hour]," "time [minute]," "number of aircraft on the same standard route," "number of surrounding aircraft," "speed," "position," "altitude," "number of aircraft on the same runway but on different standard routes," "number of aircraft on the same runway," "distance from preceding aircraft," and "distance from airport."

[0037] "Time" is the time at which the prediction is made, and the timing of making the prediction can be determined based on the time of entry into a specified airspace and the distance from the airport.

[0038] "Position" represents the approximate location information of the aircraft. Latitude and longitude can be used directly as input features in machine learning, but they can also be treated as approximate location information as a categorical variable that handles non-quantitative data, such that the aircraft's position is assigned a value of 1 when it falls within a certain range of longitude and latitude, and a value of 2 when it falls within a different range of longitude and latitude. In this case, the value is called a position ID calculated according to the latitude and longitude, and the position ID can also be used as "position."

[0039] The "number of aircraft" can be not only the number of aircraft flying the same standard route, but also the number of aircraft joining the same approach from different standard routes. The total value of these can also be added to the feature. Here, an approach refers to a route in an area closer to the airport than the standard route, and there are multiple approaches, each of which differs depending on the landing runway.

[0040] The orientation and "speed" of an aircraft can be determined from continuous time-series data of the aircraft. If the measurement error in the position information is large, it will affect the extraction of the orientation and speed of the aircraft. Therefore, rather than determining it from data at two consecutive points, it can also be determined using data at a predetermined number of points away. It is also possible to correct the position information by averaging the data at a certain point and the values ​​of the data at points before and after it, and then use the corrected position information at the two points to calculate the orientation and speed.

[0041] Figure 7 shows the feature values ​​extracted from the weather data observed for each airport. From top to bottom, Figure 7 shows in table format the following: "Wind direction," "Wind speed," "Gust," "Wind direction fluctuation range," "Visibility," "Runway visual range [average]," "Runway visual range [maximum]," "Weather," "Cloud cover," "Cloud shape," "Cloud base height," "Dew point temperature," "Air pressure," and "Wind shear."

[0042] Since weather data is acquired at regular intervals, the most recent data is selected and used at the time of forecasting.In addition to the most recent data, feature quantities from the past (several minutes or hours ago), as well as differential values ​​and average values ​​from past feature quantities, can also be included in the feature quantities.

[0043] If the flight distance of the aircraft to be predicted is long, weather data observed at multiple locations and multiple airports can also be used.

[0044] Not all features are essential, and by selecting features based on what can be obtained, their importance, etc., and reducing the number of features used, the processing speed of the air traffic control assistance device 100 can be improved.

[0045] Figure 8 shows feature quantities extracted from radar information from the Japan Meteorological Agency. Figure 8 shows data such as rain intensity and cloud height that can be acquired for each fixed location. Data on rain intensity and cloud height is acquired at fixed intervals according to latitude and longitude.

[0046] Fig. 9 shows feature quantities extracted from information assigned to aircraft. In Fig. 9, from top to bottom, "standard route name," "approach name," "weight category," and "weight category (preceding aircraft)" are shown in table format.

[0047] 6 to 9 described above, if information on the controller in charge of control is available, the controller information can be used as a feature. Since the control method may differ depending on the controller, using the controller information as a feature allows predictions that take into account the controller's control method.

[0048] Furthermore, a flight time prediction model can be created by machine learning features without obtaining controller information, making it possible to create a prediction model that learns the controller's control know-how and habits, such as taking detours when it is congested and shortcuts when it is clear.

[0049] Furthermore, machine learning can be accelerated by selecting and using only features that are highly important and correlated with the prediction results. For example, weather data is obtained by dividing the airspace into a mesh of multiple divided areas at predetermined intervals, but the weather data for divided areas where the aircraft being predicted will not fly has a low correlation with the required time and flight path, and has no impact and is therefore of low importance. On the other hand, weather data near the flight path is thought to have a significant impact. Therefore, only features that are highly important and correlated with such prediction results are selected and used.

[0050] Furthermore, by inputting the names of the standard routes and approaches as categorical variables, predictions can be made for various routes. For example, information about which of multiple standard routes an aircraft will take is not numerical data. In this case, standard route A is given a value of 1, standard route B a value of 2, and so on, and this is converted into a discrete numerical value that can be handled as a feature in machine learning. By handling the standard route information as a feature, it becomes possible to predict flight time by route.

[0051] In addition, machine learning can be performed for each standard pathway and approach name to create a predictive model for each. This means that a predictive model using machine learning is created by dividing the pathway into cases, and when making predictions, the predictive model to be used is selected by dividing the cases.

[0052] The features described above are extracted by the feature extraction unit 4 ( FIG. 1 ) and stored in the feature database 5. FIG. 10 shows the features (explanatory variables) and flight times (objective variables) stored in the feature database 5 in a table format.

[0053] In Figure 10, "hour," "minute," "number of aircraft on the same route," "weight category," "position," "wind speed," "wind direction," and "distance from airport" are used as examples of explanatory variables, and "flight time" calculated as the objective variable is shown, with the feature quantities below the number of aircraft being stored for each time when prediction is made.

[0054] 8, data on rain intensity and cloud height are acquired at regular intervals according to latitude and longitude, and the amount of information is large. Therefore, when extracting cloud-related features in the feature extractor 4 (FIG. 1), the airspace is divided into predetermined intervals, and the average, standard deviation, maximum value, minimum value, number of data (amount of cloud), etc. are extracted for each divided area.

[0055] An example of extracting a histogram by the magnitude of values ​​for each divided area is shown in Fig. 11. In the left diagram of Fig. 11, the horizontal axis represents the intensity of rain and the height of clouds, both on a five-level scale, and the vertical axis represents the frequency, i.e., the number of points with that rainfall and the number of points with that cloud height.

[0056] 11 shows an example of dividing an airspace into multiple areas, with standard routes SR1 and SR2 on opposite sides of the airport AP. The histogram shown on the left shows the feature amounts corresponding to one of the multiple divided areas shown on the right.

[0057] 11, it can be seen that the feature values ​​at which positions are useful for prediction differ depending on the standard route of the aircraft. When building a prediction model for each standard route, it is possible to select feature values ​​of divided areas around the corresponding standard route or feature values ​​of divided areas that are highly important based on the importance of feature values ​​during learning, etc.

[0058] <When the aircraft does not pass directly above a waypoint> The prediction target can be not only the flight time to landing but also the flight time to a waypoint, which is a passing point on the standard route. Processing when the aircraft does not pass over a waypoint will be described using Figure 12.

[0059] 12 shows a schematic diagram of a case where a waypoint is not passed due to trajectory variations. When the aircraft's trajectory AT deviates from the standard route SR, a perpendicular line is drawn from the waypoint WP to the trajectory AT, and the flight time to the intersection point CP where the perpendicular line and the trajectory AT intersect at a 90-degree angle is predicted. In this case, the difference in elevation between the trajectory AT and the standard route SR is not taken into account, and processing is performed assuming that the trajectory AT and the standard route SR are on the same plane. This makes it possible to predict the flight time to a waypoint even when the aircraft does not pass through it.

[0060] As explained above, the flight time prediction unit 6 (FIG. 1) learns by linking explanatory variables (feature values) and objective variables (flight time) through regression analysis using supervised learning, and statistically processes these past data to create a prediction model that obtains the objective variable from the explanatory variables. FIG. 13 is a diagram schematically illustrating the process of predicting flight time using the created prediction model.

[0061] As shown in FIG. 13, during prediction, the time-of-flight prediction unit 6 inputs the feature value FV into a regression model (prediction model) RM to output (predict) the time-of-flight FT.

[0062] <Prediction by trajectory prediction unit> As explained above, the trajectory prediction unit 7 (Fig. 1) predicts a trajectory by selecting a representative trajectory corresponding to the past flight time and entry position based on the flight time and feature amount predicted by the flight time prediction unit 6. Fig. 14 is a diagram schematically showing the prediction of a trajectory by the trajectory prediction unit 7.

[0063] As shown in Figure 14, in trajectory prediction, possible trajectories vary depending on the entry position of the aircraft being predicted and the flight time predicted by the flight time prediction unit 6. That is, entry position A is linked to multiple past trajectory data corresponding to flight time prediction results A, B, C, ..., and entry position B is linked to multiple past trajectory data corresponding to flight time prediction results A, B, C, .... The multiple trajectory data linked to the combination of entry position and flight time prediction results is stored in the trajectory database TD. Furthermore, from the multiple linked trajectory data, the trajectory with the median flight distance is selected as a representative trajectory and stored in the trajectory database RT. These databases can be provided in a storage device (not shown) within the air traffic control assistance device 100, or in a storage device external to the air traffic control assistance device 100.

[0064] For example, if the predicted flight time at entry position B is C, the trajectory prediction unit 7 predicts the trajectory by selecting from the trajectory database RT a representative trajectory from among multiple past trajectory data for a flight time corresponding to prediction result C. This makes it possible to predict a realistic trajectory supported by multiple past trajectory data.

[0065] <Effects> The effects of predicting flight time are explained below. It can contribute to the time management and trajectory management required for Trajectory Based Operation (TBO), which plans and operates an aircraft's four-dimensional trajectory, i.e., three-dimensional position and time, based on that trajectory. TBO requires four-dimensional data, and operations are managed by determining the time of passing each point, so being able to predict flight time is useful.

[0066] It can also contribute to a method known as TBS (Time Based Separation), which manages the separation between aircraft by time. TBS is a method of managing flight operations based on the time interval between aircraft flying on the same trajectory. Conventionally, flight operations were managed by distance, but in the case of strong headwinds, aircraft speeds would decrease, and if the separation requirements were about to be violated, irregular operations such as detouring or aborting landings could occur. By operating based on time intervals, it becomes possible to shorten the separation even in the case of a decrease in speed. Being able to predict flight times is useful for flight operations based on time intervals.

[0067] Furthermore, although the above describes an example of predicting the flight time to an airport, by predicting the flight time to a predetermined location where air traffic control instructions are likely to be given, such as a waypoint on a standard route or a point where the aircraft's direction changes, rather than an airport, it is possible to inform the controller in advance of the timing of issuing instructions.

[0068] In addition, calculating landing times from predicted flight times can help ensure efficient use of runways.

[0069] In addition, the trajectory can be predicted from the flight time prediction results, and the predicted trajectory results can be presented to the controller, thereby assisting the controller in issuing control instructions.

[0070] <Embodiment 2> <Device Configuration> Figure 15 is a functional block diagram showing the configuration of an air traffic control assistance device 200 according to embodiment 2. As shown in Figure 15, in air traffic control assistance device 200, the functions of flight data receiving unit 1, weather data receiving unit 2, fluctuation factor extraction unit 3, feature quantity extraction unit 4, and feature quantity database 5 are the same as those of air traffic control assistance device 100 shown in Figure 1, but the arrangement of flight time prediction unit 6 and trajectory prediction unit 7 is opposite to that of air traffic control assistance device 100. That is, air traffic control assistance device 200 differs from air traffic control assistance device 100 in that the trajectory of the aircraft is first predicted by trajectory prediction unit 7, and the flight time is predicted by flight time prediction unit 6 using the trajectory prediction result as a feature quantity.

[0071] The flight data and weather data are input into the variation factor extraction unit 3, and using AI-based machine learning technology, the variation factors that are thought to have the greatest impact on variations in aircraft flight time are statistically extracted.

[0072] The extracted flight time variation factors are input to a feature extraction unit 4 that extracts features for constructing a flight time prediction model and a trajectory prediction model, and the extracted features are stored in a feature database 5. In addition to the feature database 5 of the air traffic control assistance device 100, the feature database 5 of the air traffic control assistance device 200 also includes the trajectory as a target variable.

[0073] The features stored in the feature database 5 are used by the trajectory prediction unit 7 to construct and predict a trajectory prediction model using machine learning, and the trajectory of the aircraft until it lands at the airport is predicted.

[0074] The predicted aircraft trajectory is input to the flight time prediction unit 6, and a flight time prediction model is constructed using machine learning with the features stored in the feature database 5 and the predicted trajectory as features, and the time required to fly the input trajectory (flight time) is predicted.

[0075] The predicted aircraft trajectory and flight time are output to an air traffic control system and a display via an output unit 8 and are used for air traffic control.

[0076] 16 is a diagram schematically illustrating the process of predicting the trajectory and flight time described above. As shown in FIG. 16, during prediction, the trajectory prediction unit 7 inputs the feature values ​​FV stored in the feature value database 5 into a constructed regression model (prediction model) RM1, thereby outputting (predicting) a trajectory TJ. The trajectory TJ, together with the feature values ​​FV stored in the feature value database 5, is input into a regression model RM2 constructed by the flight time prediction unit 6, which outputs the flight time FT.

[0077] Here, the trajectory prediction results from the trajectory prediction unit 7 include information such as the route ID, flight distance, distance from the standard route, and number of waypoints passed, and the flight time prediction unit 6 uses this information as features to construct a flight time prediction model using machine learning and predict the flight time using it.

[0078] The route ID is a value for treating the predicted trajectory as a categorical variable, and is used as a feature when the flight time prediction unit 6 predicts the time required to fly the predicted trajectory (flight time).

[0079] <Operation> Next, the operation of the air traffic control assistance device 200 will be described using the flowcharts shown in Fig. 17 and Fig. 18, while referring to Fig. 15. Fig. 17 is a flowchart explaining the operation during learning using AI in the air traffic control assistance device 200, and Fig. 18 is a flowchart explaining the operation during prediction, in which flight time is predicted using a prediction model created during learning.

[0080] First, the operation during learning will be described with reference to Fig. 17. Steps S1 to S3 are the same as steps S1 to S3 shown in Fig. 2.

[0081] The feature database 5 stores a large number of feature quantities (explanatory variables) and trajectories (objective variables). The trajectory prediction unit 7 learns by linking the explanatory variables and the objective variables through regression analysis using supervised learning, and statistically processes these past data to construct a trajectory prediction model that obtains the objective variable, i.e., the trajectory, from the explanatory variables (step S6).

[0082] This involves using statistical processing to discover relationships where the values ​​of the explanatory variables have a strong influence on the values ​​of the objective variables, and then using machine learning to create a model that calculates the objective variables from the explanatory variables, thereby enabling the rapid construction of a trajectory prediction model.

[0083] In addition, features different from those used to construct the flight time prediction model are used to construct the trajectory prediction model, such as route ID, flight distance, distance from the standard route, and number of waypoints passed.

[0084] Next, the flight time prediction unit 6 uses the features (explanatory variables) and flight time (objective variable) stored in the feature database 5 to learn by linking the explanatory variables and the objective variable through regression analysis using supervised learning, and statistically processes these past data to construct a flight time prediction model that obtains the objective variable, i.e., the flight time, from the explanatory variables (step S7).

[0085] This involves using statistical processing to discover relationships where the values ​​of the explanatory variables have a strong influence on the values ​​of the objective variable through machine learning, and then creating a model that calculates the objective variable from the explanatory variables, thereby enabling the rapid construction of a flight time prediction model.

[0086] Next, the operation at the time of prediction will be described with reference to Fig. 18. Steps S11 to S13 are the same as steps S11 to S13 shown in Fig. 3.

[0087] The trajectory prediction unit 7 uses the trajectory prediction model created during learning to predict the trajectory by inputting the feature amount based on the current situation (step S16).

[0088] Next, the flight time prediction unit 6 uses the flight time prediction model created during learning to predict the flight time by inputting as features based on the current situation and information contained in the trajectory prediction result by the trajectory prediction unit 7, such as the route ID, flight distance, distance from the standard route, and number of waypoints passed (step S17).

[0089] Fig. 19 is a diagram showing a schematic diagram of trajectory prediction and flight time prediction in the air traffic control assistance device 200. As shown in Fig. 19, the flight time varies depending on the trajectory predicted by the trajectory prediction unit 7 of the aircraft that is the prediction target. That is, for predicted trajectory A, the predicted flight time results vary as A, B, C, etc., and for predicted trajectory B, the predicted flight time results vary as A, B, C, etc. For example, for predicted trajectory B, the predicted flight time result is C.

[0090] <Effects> In the air traffic control assistance device 200 according to the second embodiment, the trajectory predicted by the trajectory prediction unit 7 is input to the flight time prediction unit 6 and used as one of the feature quantities for predicting the flight time, thereby improving the accuracy of predicting the flight time.

[0091] <Embodiment 3> <Device Configuration> Figure 20 is a functional block diagram showing the configuration of an air traffic control assistance device 300 according to embodiment 3. As shown in Figure 20, in air traffic control assistance device 300, the functions of flight data receiving unit 1, weather data receiving unit 2, fluctuation factor extraction unit 3, feature extraction unit 4, and feature database 5 are the same as those of air traffic control assistance device 100 shown in Figure 1, but the features stored in feature database 5 are input to a first trajectory prediction unit 71.

[0092] The first trajectory prediction unit 71 classifies trajectories into a plurality of trajectory classes based on distance, flight time, number of waypoints passed, etc., and constructs a prediction model (classification model) for the trajectory classes.

[0093] As with the flight time prediction model, various feature quantities can be used as feature quantities for the trajectory classification model, but feature quantities different from those for the flight time prediction model can also be used.

[0094] For example, when predicting a route using a classification model, features related to airspace congestion, such as the number of aircraft on the same standard route, are useful, and when predicting flight time using a regression model for each route, features that affect flight time, such as aircraft speed, wind speed, and wind direction, may be useful.Therefore, by changing the features used in the classification model and the regression model, more accurate predictions can be achieved.

[0095] When making a prediction, the constructed classification model is used to predict the class of the trajectory of the aircraft to be predicted, and the trajectory of the predicted class (classified trajectory) is input to the flight time prediction unit 6.

[0096] The first trajectory prediction unit 71 only roughly classifies trajectories. For example, Class 1 is a shortcut trajectory, Class 2 is a trajectory with an average flight time, and Class 3 is a detour trajectory. Note that the classification of trajectories is not limited to the three types described above, and trajectories can also be classified by flight distance and flight time. Flight distance can be defined as the flight distance on a two-dimensional plane ignoring altitude, or it can be defined as the flight distance in three dimensions including altitude.

[0097] Furthermore, trajectories can be classified by the distance from the standard route, i.e., how far it is from the standard route. For example, if it is close to the standard route, it can be a detour trajectory, and if it is far from the standard route, it can be a shortcut trajectory. It can also be classified by the number of shortcut waypoints.

[0098] Also, since taking a shortcut shortens the flight time and taking a detour lengthens the flight time, the flight time can be classified into stages.

[0099] The flight time prediction unit 6 constructs a flight time prediction model corresponding to each of these classes. That is, using the features stored in the feature database 5 and the predicted trajectory for each class as features, a flight time prediction model is constructed for each class predicted by machine learning, and during prediction, the constructed flight time prediction model is used to predict the time required to fly the input trajectory (flight time). This allows the flight time prediction model to be constructed quickly.

[0100] The flight time prediction result is input to the second trajectory prediction unit 72 and used to predict the trajectory of the aircraft. Similar to the trajectory prediction unit 7 in the first embodiment, the second trajectory prediction unit 72 selects and outputs a representative trajectory from track data accumulated in the past based on the flight time prediction result and other feature quantities.

[0101] The predicted flight time and trajectory of the aircraft are output to an air traffic control system and displayed via an output unit 8 and are used for air traffic control.

[0102] 21 is a diagram schematically illustrating the process of predicting the trajectory classification and flight time described above. As shown in FIG. 21, during prediction, the first trajectory prediction unit 71 inputs the features FV stored in the feature database 5 into the constructed classification model, thereby outputting (predicting) the class of the trajectory of the aircraft to be predicted.

[0103] The flight time prediction unit 6 constructs a flight time prediction model for each trajectory of the input class. That is, regression model A, regression model B, and regression model C are provided corresponding to trajectories A, B, and C, respectively, and outputs the flight time FT for each regression model to which a trajectory is input.

[0104] <Operation> Next, the operation of the air traffic control assistance device 300 will be described using the flowcharts shown in Figures 22 and 23 while referring to Figure 20. First, the operation during learning will be described using Figure 22. Steps S1 to S3 are the same as steps S1 to S3 shown in Figure 2.

[0105] The first trajectory prediction unit 71 classifies past trajectories into multiple classes using machine learning based on features stored in the feature database 5, such as distance, flight time, and number of waypoints passed, and constructs a classification model for the trajectory classes (step S8).

[0106] Next, the flight time prediction unit 6 constructs a flight time prediction model for each classification model classified by the first trajectory prediction unit 71 (step S9).

[0107] Next, the second trajectory prediction unit 72 selects a representative trajectory corresponding to the past flight time and the entry position (step S5). From multiple trajectories that the aircraft has taken in the past from the entry position to the airport, the representative trajectory is selected and prepared for prediction.

[0108] Next, the operation at the time of prediction will be described with reference to Fig. 23. Steps S11 to S13 are the same as steps S11 to S13 shown in Fig. 3.

[0109] The first trajectory prediction unit 71 predicts the class of the trajectory of the aircraft to be predicted by inputting the features stored in the feature database 5 into the trajectory class classification model constructed during learning (step S18).

[0110] The flight time prediction unit 6 selects a flight time prediction model constructed during learning for each class of trajectory predicted by the first trajectory prediction unit 71, and predicts the flight time by inputting feature quantities based on the current situation (step S19). The flight time prediction unit 6 outputs a predicted value of the flight time (step S19).

[0111] Next, the second trajectory prediction unit 72 predicts a detailed trajectory by selecting a representative trajectory corresponding to the past flight time and entry position based on the flight time and feature values ​​predicted by the flight time prediction unit 6 (step S20). This makes it possible to predict a realistic trajectory supported by multiple past flight track data.

[0112] <Effects> In the air traffic control assistance device 300 of the third embodiment described above, the first trajectory prediction unit 71 roughly classifies trajectories. This is to avoid problems caused by differences in the proportion of data items depending on the aircraft trajectory. For example, a trajectory resulting in an average flight time will have a large amount of data items, while a detour-like trajectory will have a small amount of data items. For this reason, there is a possibility that a model will be constructed that is strong at predicting cases where there is a large amount of data items, but weak at predicting cases where there is a small amount of data items. However, the air traffic control assistance device 300 roughly classifies and predicts trajectories, and then predicts detailed trajectories, thereby avoiding such a situation.

[0113] <Modification> In the air traffic control assistance device 300 of the third embodiment described above, the first trajectory prediction unit 71 classifies past trajectories into multiple classes through machine learning using AI, but classification can also be performed using a method that does not use AI.

[0114] For example, if the degree of congestion on aircraft varies depending on the time of day, it is possible that it may be easier to make detours or shortcuts depending on the time of day, so it is possible to simply classify times as morning and evening times when congestion is high, and daytime times when congestion is low.

[0115] In addition, in the control assistance device 300 of embodiment 3, the flight time was predicted by selecting one of the flight time prediction models (regression models) constructed during learning using a classification model, but correction can also be made using the classification probability and each regression model.

[0116] In other words, if the classification model cannot accurately select a regression model, taking Figure 21 as an example, the classification model can be made to assign a probability (weight) to each trajectory class and output it, such that the probability that the classification model will select regression model A is 60% and the probability that the classification model will select regression model B is 40%, and the flight time predicted by regression model A and the flight time predicted by regression model B can be averaged using this weighting to obtain the flight time prediction result.

[0117] For the correction using the classification probability and each regression model, the following formula (1) can be used.

[0118]

[0119] In formula (1), n ​​is the number of orbitals, p ｉ is the probability of orbit i, t ｉ represents the predicted result of the regression model for trajectory i.

[0120] This makes it possible to prevent a decrease in prediction accuracy due to the selection of an incorrect regression model caused by misclassification.

[0121] <Hardware Configuration> Each of the components of the air traffic control assistance devices 100 to 300 according to the first to third embodiments described above can be configured using a computer and implemented by the computer executing a program. That is, the air traffic control assistance devices 100 to 300 are implemented, for example, by a processing circuit 1000 shown in FIG. 24. A processor such as a CPU (Central Processing Unit) or a DSP (Digital Signal Processor) is applied to the processing circuit 1000, and the functions of each unit are implemented by executing a program stored in a storage device.

[0122] Dedicated hardware may be applied to the processing circuit 1000. When the processing circuit 1000 is dedicated hardware, the processing circuit 1000 may be, for example, a single circuit, a composite circuit, a programmed processor, a parallel programmed processor, an ASIC (Application Specific Integrated Circuit), an FPGA (Field-Programmable Gate Array), or a combination thereof.

[0123] The functions of the components of the air traffic control assistance devices 100 to 300 can be realized by individual processing circuits, or the functions can be realized collectively by a single processing circuit.

[0124] 25 shows a hardware configuration in the case where the processing circuit 1000 is configured using a processor. In this case, the functions of each unit of the air traffic control assistance devices 100 to 300 are realized by a combination of software, etc. (software, firmware, or software and firmware). The software, etc. is written as a program and stored in memory 1002. The processor 1001 functioning as the processing circuit 1000 realizes the functions of each unit by reading and executing the program stored in memory 1002 (storage device). In other words, it can be said that this program causes a computer to execute the procedures and methods of operation of the components of the air traffic control assistance devices 100 to 300.

[0125] Here, the memory 1002 may be, for example, a non-volatile or volatile semiconductor memory such as RAM, ROM, flash memory, EPROM (Erasable Programmable Read Only Memory), EEPROM (Electrically Erasable Programmable Read Only Memory), HDD (Hard Disk Drive), magnetic disk, flexible disk, optical disk, compact disk, mini disk, DVD (Digital Versatile Disc) and its drive device, or any storage medium to be used in the future.

[0126] The above describes a configuration in which the functions of each component of the air traffic control assistance devices 100 to 300 are realized either by hardware or software. However, this is not limited to this, and some of the components of the air traffic control assistance devices 100 to 300 can be realized by dedicated hardware, and other components can be realized by software, etc. For example, some of the components can be realized by the processing circuit 1000 as dedicated hardware, and other components can be realized by the processing circuit 1000 as the processor 1001 reading and executing a program stored in the memory 1002.

[0127] As described above, the air traffic control assistance devices 100 to 300 can realize the above-mentioned functions by hardware, software, or a combination of these.

[0128] Although the present disclosure has been described in detail, the above description is illustrative in all respects and does not limit the present disclosure to the above. It is understood that countless variations not illustrated can be envisioned without departing from the scope of the present disclosure.

[0129] It should be noted that, within the scope of the present disclosure, the embodiments can be freely combined, modified, or omitted as appropriate.

Claims

1. A control assistance device comprising: a variation factor extraction unit that extracts data on variation factors of the flight time of the aircraft to a predetermined point from input flight data related to the flight of the aircraft and meteorological data related to weather; a feature extraction unit that extracts a plurality of feature amounts for predicting the flight time to the predetermined point from the data on variation factors of the flight time; a database that stores the extracted plurality of feature amounts; a flight time prediction unit that constructs a flight time prediction model that predicts the flight time by machine learning using the plurality of feature amounts and predicts the flight time using the flight time prediction model; and a trajectory prediction unit that predicts a trajectory of the aircraft using a result of the flight time prediction, wherein the variation factor extraction unit statistically extracts factors that may have an influence on variation in the flight time of the aircraft from the past flight data and meteorological data by machine learning, to obtain the data on the variation factors of the flight time, and the trajectory prediction unit selects a representative trajectory from a plurality of track data accumulated in the past based on the result of the flight time prediction, and outputs the representative trajectory as the predicted trajectory of the aircraft.

2. The control assistance device of claim 1, wherein the database stores the multiple features which are explanatory variables and the flight time which is an objective variable corresponding to the multiple features, and the flight time prediction unit learns by linking the explanatory variables and the objective variable through regression analysis, and constructs the flight time prediction model which obtains the flight time which is the objective variable from the explanatory variables through statistical processing.

3. The air traffic control assistance device of claim 2, wherein the trajectory prediction unit selects a median trajectory as the representative trajectory from among the multiple track data linked to a combination of the prediction results of the aircraft's entry position defined by the distance between the aircraft and the predetermined point and the flight time.

4. The control assistance device of claim 2, wherein the flight time prediction unit constructs a first flight time prediction model from a first position that is a first distance away from the predetermined point to the predetermined point, and constructs a second flight time prediction model from a second position that is a second distance away from the predetermined point that is shorter than the first distance to the predetermined point, and the trajectory prediction unit predicts a first predicted trajectory of the aircraft based on a first prediction result predicted based on the first flight time prediction model, predicts a second predicted trajectory of the aircraft based on a second prediction result predicted based on the second flight time prediction model, and outputs the second predicted trajectory as the predicted trajectory.

5. A control assistance device comprising: a variation factor extraction unit that extracts data on variation factors of the flight time of the aircraft to a predetermined point from input flight data related to the flight of the aircraft and meteorological data related to weather; a feature extraction unit that extracts a plurality of feature amounts for predicting the trajectory of the aircraft to the predetermined point from the data on variation factors of the flight time; a database that stores the extracted plurality of feature amounts; a trajectory prediction unit that uses the plurality of feature amounts to construct a trajectory prediction model that predicts the trajectory by machine learning and predicts the trajectory using the trajectory prediction model to obtain a predicted trajectory; and a flight time prediction unit that uses the predicted trajectory and the plurality of feature amounts to construct a flight time prediction model that predicts the flight time by machine learning and predicts the flight time using the flight time prediction model, wherein the variation factor extraction unit statistically extracts factors that may have an influence on variation in the flight time of the aircraft from the past flight data and meteorological data by machine learning to obtain the data on the variation factors of the flight time.

6. The control assistance device described in claim 5, wherein the database stores the multiple feature amounts which are explanatory variables, and the flight time and the trajectory which are objective variables corresponding to the multiple feature amounts, the trajectory prediction unit learns by linking the explanatory variables and the objective variable by regression analysis, and constructs the trajectory prediction model which obtains the trajectory, which is the objective variable, from the explanatory variables by statistical processing, and the flight time prediction unit learns by linking the explanatory variables and the objective variable by regression analysis, and constructs the flight time prediction model which obtains the flight time, which is the objective variable, from the explanatory variables by statistical processing.

7. A variable factor extraction unit that extracts data on variable factors of flight time of the aircraft to a predetermined point from input flight data on the flight of the aircraft and meteorological data on weather; a feature extraction unit that extracts a plurality of feature amounts for predicting the flight time to the predetermined point from the data on variable factors of the flight time; a database that stores the extracted plurality of feature amounts; a first trajectory prediction unit that constructs a classification model that classifies a trajectory of the aircraft to the predetermined point by machine learning based on the plurality of feature amounts, and outputs the trajectory classified using the classification model as a classified trajectory; a flight time prediction unit that constructs a flight time prediction model that predicts the flight time by machine learning using a predicted trajectory of the classified trajectory and the plurality of feature amounts, and predicts the flight time using the flight time prediction model; and a second trajectory prediction unit that predicts the trajectory of the aircraft using the flight time prediction result, wherein the variable factor extraction unit: a second trajectory prediction unit that selects a representative trajectory from a plurality of track data accumulated in the past based on the prediction result of the flight time and outputs the representative trajectory as a predicted trajectory of the aircraft, the second trajectory prediction unit selecting a representative trajectory from a plurality of track data accumulated in the past based on the prediction result of the flight time and outputting the representative trajectory as a predicted trajectory of the aircraft.

8. The control assistance device described in claim 7, wherein the database stores the multiple features which are explanatory variables and the flight time which is an objective variable corresponding to the multiple features, and the flight time prediction unit learns by linking the explanatory variables and the objective variable by regression analysis, and constructs the flight time prediction model which obtains the flight time which is the objective variable from the explanatory variables by statistical processing.

9. The air traffic control assistance device of claim 8, wherein the second trajectory prediction unit selects a median trajectory as the representative trajectory from among the multiple track data linked to a combination of the prediction results of the aircraft, the predetermined location, and the flight time.

10. The air traffic control assistance device described in claim 7, wherein the first trajectory prediction unit constructs the classification model using features related to airspace congestion as the plurality of features, and the flight time prediction unit constructs the flight time prediction model using features that affect flight time as the plurality of features.

Images