EP2685440B1

EP2685440B1 - Using aircraft trajectory data to infer aircraft intent

Info

Publication number: EP2685440B1
Application number: EP12382273.6A
Authority: EP
Inventors: Marco La Civita
Original assignee: Boeing Co
Current assignee: Boeing Co
Priority date: 2012-07-09
Filing date: 2012-07-09
Publication date: 2015-06-24
Anticipated expiration: 2032-07-09
Also published as: EP2685440A1

Description

Field of the Invention

The present invention relates to a method of providing data relating to the aircraft intent of an aircraft using observations of the aircraft's trajectory. The inferred aircraft intent may be used for predicting the future trajectory of the aircraft, for use in conflict resolution within air traffic management, or for analysing air traffic management.

Background to the Invention

The ability to describe, and also to predict, an aircraft's trajectory is useful for many reasons. By trajectory, a four-dimensional description of the aircraft's path is meant. The description may be the evolution of the aircraft's state with time, where the state may include the position of the aircraft's centre of mass and other aspects of its motion such as velocity, attitude and weight. In order to predict an aircraft's trajectory unambiguously, one must solve a set of differential equations that model both aircraft behaviour and atmospheric conditions.
Aircraft intent is described using a formal language that describes how the aircraft is to be flown. The aircraft intent contains instructions that define aspects of motion and configuration of the aircraft. The aircraft intent contains sufficient information to allow an unambiguous determination of an aircraft's trajectory, i.e. the information the aircraft intent contains closes all degrees of freedom of the aircraft's motion. The aircraft intent may be expressed as a series of instructions spanning part of a trajectory, with each instruction's length defining a flight segment. Each degree of freedom is described as a thread, and an instruction may operate on one or more threads. Thus, combinations of instructions may operate together to close all degrees of freedom of the aircraft.
The aircraft intent that comprises a structured set of instructions is used by a trajectory computation infrastructure to calculate the resulting unambiguous trajectory. As noted above, the instructions should include both configuration details of the aircraft (e.g. landing gear deployment) and procedures to be followed during manoeuvres and normal flight (e.g. track a certain turn radius or hold a given airspeed). These instructions capture the basic commands and guidance modes at the disposal of the pilot and the aircraft's flight management system to direct the operation of the aircraft. Thus, aircraft intent may be thought of as an abstraction of the way in which an aircraft is commanded to behave by the pilot and/or flight management system into a set of instructions that unambiguously capture the information and allow an unambiguous trajectory to be calculated.
EP-A-2,040,137 , also in the name of The Boeing Company, describes aircraft intent in more detail.
Knowledge of aircraft intent is incredibly useful in air traffic management. Knowledge of an aircraft's intent in advance allows accurate predictions of aircraft trajectories to be made that may be used to aid conflict detection and resolution. Furthermore, knowledge of aircraft intent after a trajectory has been flown may be of use in analysing air traffic management, for example to assess how efficiently the air traffic management performs in areas such as fuel efficiency and noise performance. While aircraft intent data may be provided by the aircraft or the aircraft operator, aircraft intent data is not always readily available to other interested parties. In such situations, it may be useful for other interested parties to be able to obtain the aircraft intent in some other way. Previous works such as Krozel et al., 'Intend Inference for Free Flight Aircraft, AIAA Guidance, Navigation and Control Conference 2000, no. 4479, pages 1-11; Krozel et al., 'Intent Interface with Path Prediction', Journal of Guidance, Control, and Dynamics, volume 29, no. 2 , pages 225-236 (March-April 2006); and Yepes et al. 'New Algorithms for Aircraft Intent Inference and Trajectory Prediction', Journal of Guidance, Control, and Dynamics, volume 30, ) relate to methods of inferring aircraft intent.

Summary of the Invention

The present invention provides a computer-implemented method of inferring the aircraft intent of an aircraft from an observed trajectory of the aircraft according to claim 1.
Against this background and from a first aspect, the present invention resides in a computer-implemented method of inferring the aircraft intent of an aircraft from an observed trajectory of the aircraft. The method comprises obtaining the observed trajectory of the aircraft. The observed trajectory may comprise a time-evolving sequence of positions of the aircraft. The observed trajectory may be determined from radar data, or ADS data like ADS-B data or ADS-C data.
The method also comprises determining the type of the aircraft and retrieving from memory aircraft performance data relating to that type of aircraft. For example, radar traces or other radar/ADS data may be used to identify each aircraft's call sign or tail number, and this may be correlated against records to determine the aircraft's type (e.g. Boeing 777).
Data regarding the atmospheric conditions along the observed trajectory are retrieved. For example, the observed trajectory may be used to retrieve only the atmospheric conditions pertaining to that trajectory from a database of atmospheric conditions covering the general region of airspace through which the aircraft passed.
The aircraft intent is inferred using evolutionary algorithms. The starting point is generating an initial set of candidate aircraft intents. Each candidate aircraft intent comprises instructions that define aspects of the motion and configuration of the aircraft such that the aircraft intent contains sufficient information to allow an unambiguous determination of the aircraft's trajectory so that the information the aircraft intent contains closes all degrees of freedom of the aircraft's motion. Although the initial set of candidate aircraft intents are unlikely to produce trajectories that match the observed trajectory with any accuracy, they may be evolved into candidate aircraft intents with trajectories that match well the observed trajectory.
The method further comprises providing as inputs to a computer system the initial set of candidate aircraft intents, the aircraft performance data and the atmospheric conditions data. Then, the computer system is used to calculate from the inputs a calculated trajectory defined by each candidate aircraft intent. The evolutionary algorithm requires a cost function to allow it to measure the goodness of the candidate aircraft intents it produces. This allows the best candidate aircraft intents to be retained for further evolution towards better and better candidate aircraft intents. Thus, the method comprises the computer system forming a cost function from a comparison of each calculated trajectory to the observed trajectory.
Then, the computer system uses the evolutionary algorithm to evolve the set of initial candidate aircraft intents into an evolved set of candidate aircraft intents. Iterations of the evolutionary algorithm are repeated to evolve further the candidate aircraft intents of the evolved set. The evolutionary algorithm uses the multi-objective cost function to obtain a cost function value that measures the goodness of each candidate aircraft intent.
The evolutionary algorithm terminates, and the computer system provides one or more candidate aircraft intents with the best cost function value or values respectively. Many different schemes may be used to determine when to terminate the evolutionary algorithm. Examples, which may or may not be combined, include: performing a set number of iterations, repeating until the cost function value of one or a selected number of candidate aircraft intents meet a required value, repeating until the average cost function value of a set number of candidate aircraft intents meets a required value, and waiting for the evolution to converge as measured by the change in cost function values of all or a set number of candidate aircraft intents. Where a set number of candidate aircraft intents are used, preferably a set number of the candidate aircraft intents with the best cost function values are used.
Thus, the present invention makes use of indirect measurements, which are readily available in today's airspace management systems, to derive information about the aircraft intent of the aircraft. There are no requirements for special airborne equipment or for an additional data communication infrastructure. Moreover, the aircraft are only passively involved and so there is no requirement for the aircraft to collaborate actively. This aircraft intent may be used in a predictive and/or in an analytical way.
For example, analysis of the performance of air traffic management may be performed. This may be done to measure efficiency, in terms of throughput, delays to aircraft, fuel efficiency and noise minimisation.
The aircraft intent is preferably expressed using a formal language.
Predictive uses require fast analysis of a trajectory of an aircraft still in flight to allow the future aircraft intent of that aircraft to be predicted, and hence its future trajectory to be predicted. For example, the inferred aircraft intent may conform to that required by a standard terminal arrival route (STAR) or a standard instrument departure (SID). Then, it may be assumed that the aircraft will continue to follow the identified STAR or SID.
The method may further comprise providing as a further input to the computer system initial conditions of the aircraft. Then, the calculated trajectory defined by each candidate aircraft intent may be calculated from the inputs and the further input. The method may comprise generating a common set of initial conditions from the observed trajectory. That is, the observed trajectory may be used to determine the initial position of the aircraft. Moreover, the rate of change of position of the aircraft during the initial part of the observed trajectory may be used to derive speed information. Alternatively, the method may comprise generating different sets of initial conditions from the candidate aircraft intents. For example, the candidate aircraft intents will by their nature provide a definition of the initial conditions of the aircraft including position and speed.
Preferably, the method includes a random element in how the initial set of candidate aircraft intents are generated. For example, the method may further comprise retrieving a set of bounds. Then, the initial set of candidate aircraft intents may be generated to include randomly-generated values that are constrained to remain within the bounds. For example, the randomly-generated values may correspond to values of airspeeds, rates of climb, bank angles, and high lift device settings. The bounds may provide safe or usual limits to these values, such as limiting the airspeeds to those recommended for the aircraft type or bank angles to a range that ensures passenger comfort. Furthermore, the initial set of aircraft intents may be randomly generated while being guided to provide a broad range of candidate aircraft intents. That is values with broad ranges of variation may be produced to ensure diversity in the initial set of candidate aircraft intents.
Different metrics may be used to derive the cost function. For example, the cost function may be based upon a point-by-point score derived from summing the deviation of the respective calculated trajectory from the observed trajectory at each of a number of points sampled along the observed trajectory. This rewards a low average deviation. The cost function may be based upon an overall consistency score derived from the length of the respective calculated trajectory that deviates from the observed trajectory by less than a threshold value. This may reward calculated trajectories that deviate from the observed trajectory by less than a threshold value over the greatest part of their extent.
The above two metrics may be combined. This is useful as the two different metrics reward two different behaviours. Use of the first metric will drive the candidate aircraft intents to produce trajectories with the lowest average deviation, but this may produce trajectories with very low deviations in some parts and very high deviations in other parts which it is not desirable. In contrast, the second metric punishes trajectories with extremely low and high deviations as it promotes trajectories that produce reasonable matches to the observed trajectory for as much of the observed trajectory as possible.
The candidate aircraft intents may comprise threads, each thread defining one of the degrees of freedom of the aircraft. Each thread may extend from the start of the trajectory to the end of the trajectory. Generating the initial set of candidate aircraft intents may comprise, for each candidate aircraft intent, filling each thread with one or more instructions. As each thread is then completely defined, and as the threads together define all degrees of freedom of the aircraft, this method necessarily closes all degrees of freedom of the aircraft throughout the observed trajectory and so gives rise to an unambiguous trajectory.
The degrees of freedom may comprise both degrees of freedom of motion of the aircraft and degrees of freedom of configuration of the aircraft. There may be three degrees of freedom of motion, for example two directional degrees (to define lateral and vertical motion) and a speed degree most likely defined with reference to the throttle setting. There may be three degrees of freedom of configuration, for example speed brakes, high lift devices and landing gear. Hence, this would provide six threads.
An instruction may cover more than a single thread, i.e. an instruction may define aspects relating to more than a single degree of freedom. For example, a VNAV vertical navigation instruction may define vertical motion and speed so occupying one of the threads relating to motion as well as a thread relating to throttle setting.
Evolving the candidate aircraft intents may comprise evolving the instructions. For example, the types of instructions may vary, or the values an instruction defines may vary.
Generating the initial set of candidate aircraft intents may comprise, for each candidate aircraft intent, filling each thread with an instruction such that each thread contains only a single instruction spanning the entire trajectory. The calculated trajectories may be divided into flight segments defined by the instructions. That is, the start and end of the flight segments may be defined by the starts and ends of the instructions. For example, the trajectory may be divided according to the instructions: whenever an instruction in any of the threads ends, a break between flight segments may be provided. Not all instructions need necessarily start and end together. However, the division into flight segments allows each flight segment to be defined by the instructions it contains. The instructions of some threads may not change between consecutive flight segments, although at least one instruction must change between flight segments.
Then, the method may comprise using the evolutionary algorithm to evolve a set of evolved candidate aircraft intents in a stepwise manner. Each step comprises optimising the candidate aircraft intents for one flight segment, and proceeding one flight segment at a time. Optionally, the first flight segment is evolved first, that is the flight segment starting with the start of the observed trajectory. Then the next flight segment may be evolved, then the next, and so on.
To this end, the method may comprise evolving the initial set of candidate aircraft intents iteratively to form the evolved set of candidate aircraft intents while allowing the length of the instructions to vary while keeping the start of each instruction tied to the start of the observed trajectory. That is, the initial set of candidate aircraft instructions that have a single instruction per thread are evolved such that they are constrained to have only a single instruction per thread, but with the length of that instruction being allowed to vary: as the start of the instruction must match the start of the observed trajectory, a variable length flight segment results determined by how long the instruction remains active. Not all instructions of a candidate aircraft intent need have the same length, in which case the length of the first flight segment will be determined by the shortest instruction.
As the candidate aircraft intents no longer span the entire observed trajectory, the evolutionary algorithm uses the multi-objective cost function to obtain a cost function value that measures the goodness of each candidate aircraft intent based upon a comparison of the calculated trajectory with the corresponding portion of the observed trajectory that the flight segment spans.
Once the evolutionary algorithm has evolved the evolved set of candidate aircraft intents for the first flight segment, the computer system retains the candidate aircraft intents with the best cost function values. The computer system then uses the evolutionary algorithm to perform outer loops of iterations and inner loops of iterations, as follows.
The outer loop of iterations comprises generating a further initial set of candidate aircraft intents. This is done by generating multiple copies of the retained aircraft intents. In general, each aircraft intent will contain a flight segment ending before the end of the observed trajectory. Any retained candidate aircraft intents having complete trajectories (i.e. with each thread filled by an instruction), no further changes are necessary. For the other retained candidate aircraft intents, the threads are supplemented with instructions to extend from the end of the last flight segment to the end of the trajectory such that each thread is again filled by instructions spanning the entire trajectory. Each instruction may be generated randomly, as described before.
Next, each iteration of the outer loop performs repeated iterations of the inner loop. Each iteration of the inner loop may comprise evolving the further initial set of candidate aircraft intents to form further evolved sets of candidate aircraft intents. The evolution allows the length of the instructions occupying the final flight segment to vary while keeping the start of each instruction tied to the end of the previous instruction. That is, again and again the instructions are allowed to move away from the end of the observed trajectory to define a new flight segment.
In this way, the actual flight segments in the observed trajectory may be replicated, for example changes in flight segments corresponding to the pilot switching autopilot guidance modes may be replicated. As before, the evolutionary algorithm may use the multi-objective cost function to obtain a cost function value that measures the goodness of each candidate aircraft intent based upon a comparison of the calculated trajectory calculated from the start of the observed trajectory to the end of the final flight segment with the corresponding portion of the observed trajectory. Alternatively, the comparison may be made of the calculated trajectory for the latest flight segment with the corresponding portion of the observed trajectory.
The end of each iteration of the outer loop may see the candidate aircraft intents with the best cost function values retained. Iterations of the outer loop are repeated until an evolved set of candidate solutions is produced that includes candidate aircraft intents with threads that are filled with instructions to span the entire observed trajectory. This may be implemented in different ways. For example, the outer loop may terminate when all retained candidate aircraft intents have full threads, or when one candidate aircraft intent has full threads or when some predetermined number of candidate aircraft intents have full threads.
When the outer loop terminates, all or some of the retained candidate aircraft intents may be provided.
The use of flight segments allows another metric to be used when forming the cost function. A metric may be included that rewards candidate aircraft intents with fewer flight segments and penalises candidate aircraft intents with more flight segments. This helps constrain the evolutionary algorithm that may otherwise excessively segment a trajectory to achieve better and better matches between the calculated trajectory and the observed trajectory.
How evolutionary algorithms operate to evolve solutions (solutions equating to candidate aircraft intents in the present context) is well understood. For example, strategies like retention of the fittest solutions, cross-breeding (particularly between fittest solutions) and maintenance of diversity through iterations by injecting mutations into solutions and even injecting new, random solutions are well known and may all be used with the present method. Further details are provided below.
Any of the above methods may comprise providing multiple candidate aircraft intents with the best cost function values to a user for the user to select a preferred candidate aircraft intent. This allows the user to select the most appropriate candidate aircraft intent. Many candidate aircraft intents may be suitable for selection. This may be due to the accuracy of the evolutionary algorithm providing alternatives that all fit the trajectories reasonably well, but in different ways. This may partly or wholly arise from competing objectives in the cost function. For example, the contention between the point-to-point score and the overall score may drive the evolutionary algorithm to generate Pareto optimal solutions, and solutions on the optimal Pareto front may be considered to be equally good. Also, alternative candidate aircraft intents may be equally valid, remembering that an aircraft intent can only produce a single trajectory without ambiguity but that the reverse is not true and the same trajectory may result from different aircraft intents.
In the method described above comprising outer and inner loops, the end of each iteration of the outer loop may see the best candidate aircraft intents presented to a user so that the user may select which candidate aircraft intents should be retained for the next iteration of the outer loop.
The method may further comprise ranking the provided candidate aircraft intents and the step of providing one or more candidate aircraft intents with the best cost function value or values respectively may comprise either (a) providing a ranked list of candidate aircraft intents or (b) providing the highest ranked candidate aircraft intent. Ranking the provided candidate aircraft intents may comprise at least one of: ranking according to the cost function values, ranking according to number of flight segments, and ranking according to the frequency with which that candidate aircraft intent appears in the evolved set. Ranking may reward the candidate aircraft intents with the fewest flight segments, for the same reasons as described above. Ranking according to the frequency with which that candidate aircraft intent appears in the evolved set rewards candidate aircraft intents that have been evolved independently multiple times by the evolutionary algorithm. The ranking may be used to determine which candidate aircraft intents are to be retained at the end of each iteration of the outer loop, when that method is being employed.
The present invention also resides in a computer system programmed to implement any of the methods described above. The present invention also resides in a computer program comprising computer program instructions that, when executed, cause a computer system to implement any of the methods described above. The present invention also resides computer readable medium having stored therein such a computer program. For example, the computer readable medium may comprise memory. Accordingly, the present invention also resides in a computer system comprising such a memory.

Brief Description of the Drawings

In order that the present invention may be more readily understood, preferred embodiments will now be described, by way of example only, with reference to the following drawings in which:

Figure 1 is a system for computing an aircraft's trajectory using flight intent and aircraft intent;
Figure 2 shows the system of Figure 1 in greater detail;
Figures 3a to 3c are three schematic representations of aircraft intent;
Figure 4 shows the horizontal and vertical profiles of a trajectory, and how the trajectory may be divided into flight segments;
Figure 5 is a schematic representation of a method of providing aircraft intent data of an aircraft according to a first embodiment of the present invention;
Figure 6 is a schematic representation to show part of the method of Figure 5 in greater detail; and
Figure 7 shows an alternative arrangement to Figure 5.

Detailed Description of the Invention

To appreciate the present invention, it is necessary to explain several details related to trajectory prediction. In essence, the process of predicting the trajectory of an air vehicle involves the following pieces of information.
Aircraft intent (AI) is a formal description of how the air vehicle is to be operated during the time interval for which the predicted trajectory is valid. This information, which can be expressed by means of an aircraft intent description language (AIDL) captures all the details related to the specific aircraft motion that gives rise to the trajectory of interest.
The aircraft performance model (APM) provides all the aircraft-specific details that influence aircraft response. For example, it may include aerodynamic and propulsive characteristics, preferred and default flap schedules and other manoeuvres, operational limitations, etc.
The Earth model (EM) provides all the details necessary to represent mathematically the influences of the environment on the aircraft motion. For example, it may reflect gravity, geopotential, atmospheric conditions, etc. The Earth model provides 4D predictions (i.e. for any given position and time) of wind velocity, pressure and temperature.
The prediction process also requires a formulation of the initial conditions (IC), namely the parameters that characterise the aircraft state (position, velocity, attitude) at the beginning of the time interval of interest.
The trajectory computation process essentially consists of solving a mathematical problem that arises from a certain formulation of the air vehicle's motion physics made out in terms of the aforementioned elements. The result is the aircraft trajectory (AT), i.e. a discrete sequence of aircraft states (position and velocity, and optionally attitude and other additional characteristics of the motion).
In order to predict aircraft trajectory unambiguously, one must solve a set of differential equations that model both aircraft behaviour and atmospheric conditions. The computation process requires inputs corresponding to the aircraft intent, often derived from flight intent.
Aircraft intent must be distinguished from flight intent. Flight intent may be thought of as a generalisation of the concept of a flight plan, and so will reflect operational constraints and objectives such as intended or required route and operator preferences. Generally, flight intent will not unambiguously define an aircraft's trajectory, as the information it contains need not close all degrees of freedom of the aircraft's motion. Flight intent also comprises instructions that span flight segments. However, instances of flight intent may not contain instructions that cover all threads and so some degrees of freedom of the aircraft may be left open. Therefore, there may be many aircraft trajectories that would satisfy a given flight intent. Flight intent may be regarded as a basic blueprint for a flight or a set of requirements that the ensuing aircraft trajectory must comply with, but that lacks the specific details that are present in the aircraft intent and that are required to determine unambiguously how the aircraft will behave as to meet these requirements (i.e. as to exhibit a particular trajectory).
For example, the instructions to be followed during a standard terminal arrival route (STAR) or a standard instrument departure (SID) route that typically appear in the airport information publication (AIP) departure/arrival charts would correspond to an example of flight intent. In addition, airline preferences such as a prescribed cost index (e.g. describing a preference towards prioritising punctuality of flights against economic cost of flights) to be achieved may also form an example of flight intent. To derive aircraft intent instances from flight intent like a SID procedure, the airline's operational preferences and the actual pilot's decision making process can be combined.
Figure 1 shows a basic structure 100 to derive aircraft intent, and how aircraft intent may be used to determine and compute the aircraft's trajectory univocally. In essence, flight intent 101 is provided as an input to an intent generation infrastructure 103. The intent generation infrastructure 103 determines aircraft intent 114 using the instructions provided by the flight intent 101 and other inputs to ensure a set of aircraft intent instructions is provided that will allow an unambiguous determination of the trajectory. The aircraft intent 114 output by the intent generation infrastructure 103 may then be used as an input to a trajectory computation infrastructure 110. The trajectory computation infrastructure 110 calculates the resulting trajectory unambiguously determined beforehand by the aircraft intent 114, for which additional inputs are required to solve the equations of motion of the aircraft.
Figure 2 shows the structure 100 of Figure 1 in further detail. As can be seen, the intent generation infrastructure 103 receives a description of the flight intent 101 as an input along with a description of the initial state 102 of the aircraft (the initial state of the aircraft may be defined as part of the flight intent, in which case these two inputs are effectively one and the same). The intent generation infrastructure 103 comprises an intent generation engine 104 and a pair of models, one encompassing the user preferences 105 and one representing the operational context 106 of the aircraft.
The user preferences model 105 embodies the preferred operational strategies governing the aircraft, e.g. the preferences of an airline with respect to loads (both payload and fuel); how to react to meteorological conditions such as temperature, wind speeds, altitude, jet stream, thunderstorms and turbulence as this will affect the horizontal and vertical path of the aircraft as well as its speed profile; cost structure such as minimising time of flight or cost of flight, maintenance costs, environmental impact; communication capabilities; passenger comfort requirements and security considerations.
The operational context model 106 embodies constraints on use of airspace, e.g. those imposed by the air traffic control to all flights operating within the given airspace such as speed, altitude restrictions, etc. The intent generation engine 104 uses the flight intent 101, initial state 102, user preferences model 105 and operational context model 106 to provide the aircraft intent 114 as its output. The process performed by the intent generation infrastructure 103 can be seen as determining a particular way of commanding the aircraft such that the resulting trajectory 122 meets the requirements set by the specific flight intent 101 and those generally imposed by the user 105 and the operational context 106.
Figure 2 also shows that the trajectory computation infrastructure 110 comprises a trajectory engine 112. The trajectory engine 112 requires as inputs both the aircraft intent description 114 explained above and also the initial state 116 of the aircraft. The initial state 116 of the aircraft may be defined as part of the aircraft intent 114 in which case these two inputs are effectively one and the same. The initial state 116 may consist of the same information as the initial state input 102 provided to the intent generation infrastructure 103 or it may be augmented to include further aspects of the initial aircraft state as needed to determine an unambiguous trajectory. For the trajectory engine 112 to provide a description of the computed trajectory 122 for the aircraft, the trajectory engine 112 uses two models: an aircraft performance model 118 and an Earth model 120.
The aircraft performance model 118 provides the values of the aircraft performance aspects required by the trajectory engine 112 to integrate the equations of motion. These values depend on the aircraft type for which the trajectory is being computed, the aircraft's current motion state (position, velocity, weight, etc) and the current local atmospheric conditions. In addition, the performance values may depend on the intended operation of the aircraft, i.e. on the aircraft intent 114. For example, a trajectory engine 112 may use the aircraft performance model 118 to provide a value of the instantaneous rate of descent corresponding to a certain aircraft weight, atmospheric conditions (pressure altitude and temperature) and intended speed schedule (e.g. constant calibrated airspeed). The trajectory engine 112 will also request from the aircraft performance model 118 the values of the applicable limitations so as to ensure that the aircraft motion remains within the flight envelope. The aircraft performance model 118 is also responsible for providing the trajectory engine 112 with other performance-related aspects that are intrinsic to the aircraft, such as flap and landing gear deployment times.
The Earth model 120 provides information relating to environmental conditions, such as the state of the atmosphere, weather conditions, gravity and magnetic variation.
The trajectory engine 112 uses the inputs 114 and 116, the aircraft performance model 118 and the Earth model 120 to solve a set of equations of motion. Many different sets of equations of motion are available that vary in complexity, and that may reduce the aircraft's motion to fewer degrees of freedom by means of a certain set of simplifying assumptions. Any of these sets of equations of motion may be used with the present invention.
The trajectory computation infrastructure 110 may be air-based or land-based. For example, the trajectory computation infrastructure 110 may be associated with an aircraft's flight management system that controls the aircraft on the basis of a predicted trajectory that captures the airline operating preferences and business objectives. The primary role for land-based trajectory computation infrastructures 110 is for air traffic management.
The basic relationship connecting aircraft intent 114, the aircraft performance model 118, the Earth model 120, the initial conditions 116 and the aircraft trajectory 122 can be written, using set notation, as {AI, APM, EM, IC} ⇒ AT. In other words, given an instance of aircraft intent 114, for a specific aircraft (as reflected by the aircraft performance model 118) performing in a certain environment (reflected by the Earth model 120), starting from the given initial conditions (116), the resulting aircraft trajectory (122) is univocally determined.
Formally speaking, the opposite is not true. That is to say, given an aircraft trajectory 122 (and hence initial conditions because IC ⊂ AT) that belongs to a specific aircraft model (from the aircraft performance model 118) performing in a certain environment (from the Earth model 120), there are in principle many aircraft intents 114 that could give rise to that aircraft trajectory 122. This is because the relationship between {APM, AT, EM} and AI is not bijective. So, when determining aircraft intent from a trajectory, it is entirely possible that more than a single solution might exist (i.e. the calculation may well find multiple aircraft intents, each of which could give rise to the trajectory seen).
Figures 3a to 3c show how aircraft intent may be visualised. In this example, an aircraft is considered to be defined by six degrees of freedom, three of which relate to degrees of freedom of aircraft motion and three of which relate to degrees of freedom of aircraft configuration. For example, one degree may relate to horizontal motion and one degree might relate to landing gear configuration. The six degrees are shown as six threads (thread 1 to thread 6) that extend throughout the time of the trajectory being considered. The time elapsed through the trajectory is shown from left to right in Figures 3a to 3c, with the start and end times indicated.
Each thread is filled by one or more instructions that define that degree of freedom. In principle, a single instruction may be all that is required to define one thread through the trajectory being considered as illustrated in Figure 3a. In practice, normally more than a single instruction is required per thread. Figure 3b shows an example of two instructions per thread, and Figure 3c shows a general example comprising n instructions per thread. Where there are multiple instructions per thread, this divides the trajectory into flight segments, with the end of an instruction creating an end to a flight segment.
In Figures 3a to 3c, instructions of equal length are shown such that, at the start of each flight segment, all threads also start with a new instruction. This need not be the case. Some instructions may span multiple flight segments. For example, an instruction to maintain a constant Mach speed may span three flight segments where other instructions first define maintaining a heading, then define turning to a new heading and then define maintaining the new heading. In addition, a single instruction may span more than a single thread. An example is shown in Figure 3c where the second flight segment (segment b) contains an instruction (instruction 1b) that defines both the first and second threads. For example, the instruction may be a vertical navigation guidance mode (VNAV) that may define three threads relating to vertical motion, throttle configuration and high lift devices configuration.
Figure 4 shows an example of an aircraft 400 flying an approach into an airport and shows the resulting trajectory 410. The horizontal profile is shown in the upper part of the figure and the vertical profile is shown in the lower part of the figure. The figure also shows how the trajectory may be decomposed into flight segments according to the instructions of the aircraft intent.
The horizontal profile 410a comprises three sections of constant heading divided by two left turns shown at 411 and 412. The vertical profile 410b comprises an initial constant rate of descent 413, followed by a phase of maintaining altitude 414 and then a phase of maintaining a constant rate of descent 415. The phase of flight where altitude is held 414 begins shortly before turn 412 is started and ends shortly after turn 412 is completed. These changes in motion lead to the following segments in the aircraft intent. Segment a starts with the start of the trajectory and requires a heading and a constant rate of descent to be maintained. Segment b is triggered by a change to turn to a new heading instruction, the start of the turn marking the start of segment b and the end of the turn marking the end of segment b. Segment c sees heading and rate of descent maintained. Segment d is triggered by an instruction to maintain altitude. Segment e corresponds to the instruction to change heading, while the maintain altitude instruction remains in force. Segment f sees an instruction to maintain the new heading while still maintaining altitude, which ends when the new maintain a constant rate of descent instruction triggers segment g.
The above example is simplified, and further instructions are likely and so further flight segments may arise. For example, configuration changes may be required. By way of illustration, segment h is shown in Figure 4 that corresponds to an extend landing gear instruction at 416. In addition, speed control will be required during an approach to an airport, and this will give rise to further instructions not illustrated in Figure 4.
Figure 5 shows an embodiment of a method of inferring aircraft intent using observations of a trajectory 410 flown by an aircraft 400. The method may be implemented using a computer system to perform the calculations required. Any type of commonly-available computer system may be used to implement the method, provided it has sufficient performance to perform the calculations at the required speed. The computer system may be conventional in providing input devices like keyboards, computer mice, touch screens and microphones, and providing output devices like monitors, displays and speakers. Data storage means may be provided like volatile and non-volatile memory. The computer system may be centralised, or may be distributed with connections provided by a network. For example, the computer system may be located at an airport and may ingest trajectory surveillance data related to the airspace around the airport, such as a defined terminal manoeuvring area. The computer system may be required to determine the aircraft intent of multiple aircraft, and so the performance of the computer system may be specified with this in mind.
Turning now to Figure 5, the method of inferring aircraft intent begins at step 510 where the trajectory 410 of an aircraft of interest 400 is collected from the appropriate surveillance source. This may be done using any convenient means. For example, radar data may be used to provide the aircraft trajectory 410, and ADS data such as ADS-A, ADS-B or ADS-C data may be used. The trajectory 410 may be saved into memory.
At 520, the aircraft type is determined. For example, the radar trace may be used to identify the aircraft's call sign or tail number, and this may be correlated against records to determine the aircraft's type (e.g. Boeing 777). Data pertaining to the aircraft type is retrieved from the aircraft performance model 118 and associated with the aircraft's trajectory 410 saved in memory.
At 530, the earth model is used to provide the atmospheric conditions affecting the aircraft 400 while flying its trajectory 410. For example, the four dimensional trajectory 410 observed for the aircraft 400 may be used to identify positions of the aircraft 400 along the trajectory 410, and data retrieved that describes the atmospheric conditions prevailing at each position at the time the aircraft 400 was at that position.
At 540, the computer system generates a set of initial candidate aircraft intents that might describe the trajectory 410 flown by the aircraft 400. This initial population of aircraft intents is generated randomly, as is described in more detail below.
At 550, a trajectory is calculated for each initial candidate aircraft intent generated. That is, the initial candidate aircraft intent, the associated initial conditions implied by the candidate aircraft intent, the aircraft performance data and the atmospheric conditions data are used by the trajectory computation engine 112 to calculate a corresponding calculated trajectory.
Next, at 560, each of the calculated trajectories is compared to the observed trajectory 410. This process is used to identify the best of the initial candidate aircraft intents and to evolve the candidate aircraft intents to one that is most likely to match the actual aircraft intent of the aircraft 400 that gave rise to the observed trajectory 410. That is, successfully matching a calculated trajectory to the observed trajectory 410 is used as an indication that the associated candidate aircraft intent accurately describes the actual aircraft intent.
This comparison of trajectories is used to generate a multi-objective cost function that characterises the difference between each calculated trajectory and the observed trajectory 410. This multi-objective cost function uses two principal, competing measures to characterise how well the calculated trajectory matches the observed trajectory 410. The first measure is the root mean square deviation between the trajectories, and how this compares to a threshold value. The second measure counts the number of time intervals over which the root mean square deviation is less than the threshold. The time intervals may correspond to flight segments, although this need not be the case.
At 570, the candidate aircraft intents are evolved by the computer system. The calculation uses evolutionary algorithms to evolve the candidate aircraft intents to reflect the actual aircraft intent better. The evolutionary algorithms use the multi-objective cost function to drive the evolution. That is, the evolutionary algorithm evolves new candidate aircraft intents, calculates the corresponding trajectories, and calculates a new value for the cost function with the aim of minimising the cost function value. The evolutionary algorithm seeks on the one hand to minimise the root mean square deviation between the trajectories and to ensure the root mean square difference is within the threshold value while, on the other hand, the evolutionary algorithm seeks to maximise the number of time intervals over which the root mean square deviation falls within the threshold. These two competing demands lead to candidate aircraft intents that form a set of Pareto solutions occupying a Pareto front. These candidate aircraft intents are equally valid solutions, and may arise either because they represent equally good estimates or because they both represent valid aircraft intents (remembering that multiple aircraft intents may give rise to any particular trajectory). Further explanation of this step follows below.
With a set of solutions determined, the set of candidate aircraft intents having the lowest cost function values are selected at step 580 and, at step 590, these candidate solutions are provided to a user for the user to select one of the candidate aircraft intents as being the inferred aircraft intent. The candidate aircraft intents may be presented in different ways. The candidate aircraft intents may be ranked. For example, candidate solutions having a range of the lowest cost function values may be presented, with the lowest values being ranked highest. Other criteria may be used for ranking, such as the least number of flight segments or the number of times that candidate aircraft intent appears in the results (the nature of evolutionary algorithms means that the same solution may be evolved independently more than once). Of course, any of these criteria may be used to select a candidate aircraft intent automatically rather than allowing a user to select an inferred aircraft intent.
Figure 6 shows the method of inferring aircraft intent of Figure 5 in more detail. The method of Figure 6 starts with steps 510, 520 and 530 as already described above with respect to Figure 5. Thus, the observed aircraft trajectory 410 is retrieved, along with aircraft performance data relating to the aircraft 400 and the relevant atmospheric conditions.
At step 540, candidate aircraft intents are generated. To do this, at step 542, bounds for the aircraft intent instructions are retrieved. For example, the aircraft performance data may be used to define upper and lower airspeeds for the aircraft, and to determine maximum speeds when landing gear and high lift devices are deployed. Next, at step 544, the computer system generates a pre-determined number of candidate aircraft intents. The generation is random, within bounds. Specifically, the aircraft intents may be generated with any set of instructions although parameters must be set within the bounds retrieved at step 542. Consequently, the trajectory may be divided into any number of flight segments of any length, and the instructions for each segment are generated randomly. Strategies may be used to ensure that the initial population of candidate aircraft intents is suitably diverse, as is well known in the art. Although most initial guesses will bear little resemblance to the actual aircraft intent, the evolutionary process will quickly evolve the candidate aircraft intents closely resembling the actual aircraft intent and will then refine those candidate solutions to ensure a close match.
At step 550, each candidate aircraft intent is used to calculate a corresponding trajectory as already described with respect to step 550 of Figure 5. Step 560 of Figure 5 is followed next, that is step 560 of Figure 6 sees a multi-objective cost function formed from a comparison of the calculated trajectories with the observed trajectory 410.
At step 570, the candidate aircraft intents are evolved by the computer system using evolutionary algorithms. That is, repeated iterations are used to evolve the candidate aircraft intents to provide calculated trajectories that match better the observed trajectory. This may be done in any well-known way. For example, at the end of each iteration, several well-known selection criteria are applied. Some of the better candidate aircraft intents are retained for the next iteration. In this embodiment, the top 20% are retained as ranked by the lowest cost function value. Then, a random selection of 60% are cross-bred (i.e. portions of their aircraft intents are swapped) and the resulting aircraft intents used in the next iteration. A further 10% of the aircraft intents used in the next iteration are derived by randomly selecting 10% of the candidate aircraft intents from the previous iteration and randomly introducing mutations, i.e. removing a portion of the aircraft intent and replacing it with a randomly generated portion. The remaining 10% of aircraft intents used in the next iteration are randomly generated. Thus, genetic diversity is introduced into the evolutionary process, and the selection process ensures that the better candidate aircraft intents are retained.
Step 570 may end after a pre-determined number of iterations of the evolutionary algorithm have been performed. However, in this embodiment, it is preferred to end the evolution process according to a test performed at the end of each iteration of the evolutionary algorithm. The test may use the cost function values of the candidate solutions, or a range of the best candidate solutions. How the cost function values change from iteration to iteration may be used to indicate convergence of the evolutionary process, for example when the cost function values improve by less than a convergence threshold, step 570 exits. Alternatively, the cost function values themselves may be used to cause step 570 to exit, for example when the best candidate solution has a cost function value below a limit or an average cost function value is below a limit or a certain number of candidate solutions have cost function values below a certain limit. Combinations of these tests may also be used to determine when step 570 exits.
When step 570 exits, the best candidate aircraft intents are selected at step 580 and presented to the user at step 590, in accordance with any of the methods and variations described with respect to steps 580 and 590 in Figure 5.
Figure 7 shows an alternative embodiment to that shown in Figure 6. As many parts are the same, the following description focuses on the differences. The principal difference is that the method of Figure 7 sees the evolutionary algorithm stage operate one flight segment at a time. That is, the candidate aircraft intents are optimised for a particular flight segment, before the evolutionary algorithm progresses to optimising the aircraft intents for the next flight segment, and so on.
The observed aircraft trajectory 410, the aircraft performance data and the atmospheric conditions are retrieved in steps 510, 520 and 530 as has already been described. As before, step 520 begins with bounds being retrieved at 542.
Step 544 sees a difference to the previously-described methods. A further restriction is placed on the random generation of candidate aircraft intents: all instructions generated are required to span the entire observed trajectory 410 such that there is no segmentation of the observed trajectory 410. Instructions may be allowed to apply to more than a single thread, but still within the requirement that the instruction will apply to all the applicable threads for the entire length of the observed trajectory. Thus, the method initially tries to derive an aircraft intent with a single flight segment occupying the entire observed trajectory 410.
With the candidate aircraft intents generated at step 540, the method progresses through steps 550 and 560 where trajectories are calculated and the cost function is formed as has been described previously.
The candidate aircraft intents are evolved at step 570. The evolutionary algorithm is constrained to keep a single instruction per thread, but is free to allow the length of that instruction to vary subject to beginning at the start time of the observed trajectory 410. Consequently, step 570 evolves candidate aircraft intents that define the initial part of the observed trajectory 410 only, while leaving the subsequent part of the observed trajectory undefined. Hence, the evolutionary algorithm produces a set of candidate aircraft intents at step 570 that produce part-trajectories that match the initial part of the observed trajectory 410 well. As indicated at step 571, the results of each iteration of the evolutionary algorithm are tested to see whether a further iteration should be performed. This test may be performed as described above, albeit only in respect of the initial part of the observed trajectory 410 defined by the part-aircraft intent.
When the test at step 571 is passed, and evolution step 570 is exited, the method continues to step 535 where the candidate aircraft intents with the lowest cost function values are selected to be retained for the next part of the method. In the next part of the method, the part-aircraft intents of the selected candidate aircraft intents are developed to match the next part of the observed trajectory 410.
This is achieved, starting at step 545 where further instructions are added to the candidate aircraft intents. Similarly to step 542, bounds are retrieved and used in constraining the random values taken by instruction parameters. Each instruction added is constrained to occupy the remainder of each thread, that is each instruction starts at the end of the instructions already present in the candidate aircraft intent and must end at the end of the observed trajectory 410. As before, an instruction may apply to more than a single thread. In this embodiment, the different candidate aircraft intents may not necessarily end at the same point in the observed trajectory 410. However, in other contemplated embodiments, the evolutionary algorithm is constrained to keep the end points of the instructions currently being evolved to be the same. In any event, step 545 sees a new set of candidate aircraft intents being generated where each candidate aircraft intent defines the entire observed trajectory 410 with two flight segments.
It should be noted that one or more candidate aircraft intents may arise from step 570 that occupy the entire observed trajectory: thus, a check may be made at the start of step 545 for such candidate aircraft intents and these intents may be either removed and placed into a set of final results or they may be input back into the evolutionary algorithm without the addition of further instructions.
Steps 550 and 560 are then repeated, that is a complete trajectory is calculated for each candidate aircraft intent and a cost function is formed that reflects a comparison of the complete trajectories (i.e. the trajectories resulting from both flight segments).
Step 575 sees the candidate aircraft intents evolved according to the evolutionary algorithm. This step is similar to step 570, except that now the instructions for each first flight segment are fixed and only the instructions relating to the second flight segment are allowed to vary. The length of the second flight segment is allowed to vary, subject to the requirement it must start at the end of the first flight segment. Hence, the evolutionary algorithm evolves the candidate aircraft intents to form candidate aircraft intents that describe well the first and second parts of the observed trajectory 410. The second part of each candidate aircraft intent may or may not extend till the end of the observed trajectory. The evolutionary algorithm continues through iterations as determined by test step 576 that corresponds to test step 571.
Steps 535, 545, 550, 560, 575 and 576 essentially repeat as a loop, such that with each pass through the loop, candidate aircraft intents are evolved that match better the next part of the observed trajectory 410 under consideration. That is, the evolutionary algorithm is repeatedly invoked to determine a suitable length for the next flight segment and to provide candidate aircraft intents that produce calculated trajectories that match well the observed trajectory 410 up to the end of the flight segment currently being considered. In this way, candidate aircraft intents are constructed one flight segment at a time, in chronological order.
Accordingly, the method comprises step 577: when the current evolutionary step 570 exits after the determination at step 576, a further determination is made at step 577. This second determination looks at each candidate aircraft intent to see whether the final flight segment has been defined, i.e. whether or not the latest flight segment added extends to the end of the observed trajectory. This step may be combined with the selection process of step 535 such that the length of only the best candidate aircraft intents is considered. Any candidate flight intents defining the full extent of the observed trajectory 410 may be either removed and placed into a set of final results or they may be input back into the evolutionary algorithm without the addition of further instructions.
Eventually, step 577 will determine that all candidate aircraft intents, or that the best candidate aircraft intents to be selected, define the full extent of the observed trajectory 410. When this point is reached, the method proceeds to steps 580 and 590 where, as described before, the best candidate aircraft intents are selected and presented to the user.
As will be appreciated, the above methods see the aircraft intents segmented by the evolutionary algorithm. This process can in theory lead to much segmentation in order to arrive at the lowest cost function possible (i.e. to ensure better correspondence between the calculated trajectories and the observed trajectory 410). In practice, the aircraft 400 is likely to have been flown with relatively few flight segments, so the selection process of which candidate aircraft intents to retain for the next iteration of the evolutionary algorithm may reward those candidate solutions with the fewest flight segments or, in the case of the method of Figure 7, those with the longest flight segment currently under consideration.
Those skilled in the art will appreciate that variations may be made to the above embodiments without departing from the scope of the invention that is defined by the appended claims.
Moreover, additional applications of the present invention are contemplated. For example, to communicate a trajectory, such as a predicted trajectory or a desired trajectory, it may be necessary to apply compression algorithms so that a bandwidth is not exceeded. Lossless compression algorithms are available for this purpose. However, to achieve high compression ratios, it is often necessary to apply lossy compression algorithms. Such lossy algorithms are undesirable since the resolution of the communicated trajectory is diminished.
However, as explained above, an instance of aircraft intent can unambiguously define a corresponding trajectory.
The methods of inferring aircraft intent from trajectory described above can therefore be used as a method of compression of trajectory. Any instance of aircraft intent that unambiguously defines the trajectory can be used.
The compressed aircraft intent can be transmitted from a transmitting system to a receiving system instead of the trajectory. The aircraft intent can be decompressed to express the trajectory at the receiving system.
In other words, a preferred embodiment of a method of transmitting aircraft trajectory data from a transmitting system to a receiving system comprises:

inferring aircraft intent data from the aircraft trajectory data using the transmitting system, for example in accordance with any of the methods described above;
transmitting the inferred aircraft intent data; and
deriving trajectory data from the transmitted aircraft intent data.

Preferably, the aircraft intent data is inferred from the aircraft trajectory data using an Earth model and an aircraft performance model stored by the transmitting system.
Preferably, the trajectory data is derived at the receiving system from the transmitted aircraft intent data using the same Earth model and the aircraft performance model used to infer the aircraft intent data.
In some cases, these models may already be available at the receiving system. In other cases, the method may further comprise transmitting the Earth model and the aircraft performance model from the transmitting system to the receiving system.
Optionally, the aircraft intent data can itself be compressed by the transmitting system (preferably, using lossless compression), and decompressed after transmission by the receiving system.
This does not unduly increase the amount of data to be transmitted, since one Earth model and one aircraft performance model can be used to compress and decompress an unlimited number of trajectories 122.
It has been shown experimentally that compression ratios (the ratio of size of trajectory data to size of aircraft intent data) of 55:1 can be achieved using the above-described method.

Claims

A computer-implemented method of inferring the aircraft intent (114) of an aircraft (400) from an observed trajectory (410) of the aircraft, the method comprising:
obtaining the observed trajectory of the aircraft;

determining the type of the aircraft and retrieving from memory aircraft performance data relating to that type of aircraft;

retrieving data regarding the atmospheric conditions along the observed trajectory;

generating an initial set of candidate aircraft intents wherein each aircraft intent comprises a set of threads, wherein each thread comprises one or more instructions that defines a degree of freedom of the aircraft, the set of threads together defining all degrees of freedom of the aircraft from the start of the trajectory to the end of the trajectory, thereby closing all degrees of freedom of the aircraft throughout the observed trajectory;

providing as inputs to a computer system the initial set of candidate aircraft intents, the aircraft performance data and the atmospheric conditions data;

calculating from the inputs a calculated trajectory defined by each candidate aircraft intent by solving a set of equations of motion of the aircraft;

obtaining for each calculated trajectory a cost function value from a multi-objective cost function that characterises the difference between a calculated trajectory and the observed trajectory, and retaining the initial candidate aircraft intents with the best cost function values;

using an evolutionary algorithm to evolve the set of best initial candidate aircraft intents into an evolved set of candidate aircraft intents by evolving the instructions of the candidate aircraft intents and repeating iterations of the evolutionary algorithm to evolve further the candidate aircraft intents of the evolved set, wherein the evolutionary algorithm uses the multi-objective cost function to obtain a cost function value for the calculated trajectory obtained for each the candidate aircraft intents of the evolved set; and

providing one or more candidate aircraft intents with the best cost function value or values respectively.
The method of claim 1, wherein determining the trajectory of the aircraft comprises using radar data, ADS-B data or ADS-C data.
The method of claim 1 or 2, comprising providing as a further input to the computer system initial conditions (116) of the aircraft, and wherein the calculated trajectory defined by each candidate aircraft intent is calculated from the inputs and the further input.
The method of claim 3, comprising generating a common set of initial conditions from the observed trajectory or generating different sets of initial conditions from the candidate aircraft intents.
The method of any preceding claim, further comprising retrieving a set of bounds, and randomly generating the
initial set of candidate aircraft intents to include randomly generated values within the bounds.
The method of any preceding claim, further comprising randomly generating the initial set of aircraft intents while being guided to provide a broad range of candidate aircraft intents.
The method of any preceding claim, wherein the cost function is based upon the combination of (a) a point-by-point score derived from summing the deviation of the respective calculated trajectory from the observed trajectory at each of a number of points sampled along the observed trajectory, and (b) an overall consistency score derived from the length of the respective calculated trajectory that deviates from the observed trajectory by less than a threshold value.
The method of any preceding claim, wherein generating the initial set of candidate aircraft intents comprises, for each candidate aircraft intent, filling each thread with an instruction such that each thread contains only a single instruction spanning the entire trajectory.
The method of claim 8, wherein the calculated trajectories are divided into flight segments, the start and end of the flight segments being defined by the starts and ends of the instructions, and wherein the method comprises using the evolutionary algorithm to evolve a set of evolved candidate aircraft intents in a stepwise manner, each step comprising optimising one flight segment at a time starting with the first flight segment and proceeding chronologically through the observed trajectory.
The method of claim 9, comprising:
evolving the initial set of candidate aircraft intents iteratively to form the evolved set of candidate aircraft intents while allowing the length of the instructions to vary while keeping the start of each instruction tied to the start of the observed trajectory, and wherein the evolutionary algorithm uses the multi-objective cost function to obtain a cost function value that measures the goodness of each candidate aircraft intent based upon a comparison of the calculated trajectory calculated for the flight segment with the corresponding portion of the observed trajectory;

retaining the candidate aircraft intents with the best cost function values;

and then performing outer loops of iterations and inner loops of iterations, wherein:
the outer loop of iterations comprises
generating a further initial set of candidate aircraft intents by generating multiple copies of the retained aircraft intents and adding an instruction to each thread of the copies of the retained candidate aircraft intents to extend from the end of the last flight segment to the end of the trajectory such that each thread is again filled by instructions spanning the entire trajectory,
repeated iterations of the inner loop comprising evolving the further initial set of candidate aircraft intents to form further evolved sets of candidate aircraft intents while allowing the length of the instructions occupying the final flight segment to vary subject to keeping the start of each instruction tied to the end of the previous instruction, and while the evolutionary algorithm uses the multi-objective cost function to obtain a cost function value that measures the goodness of each candidate aircraft intent based upon a comparison of the calculated trajectory calculated from the start of the observed trajectory to the end of the final flight segment with the corresponding portion of the observed trajectory, and
retaining the candidate aircraft intents with the best cost function values;

wherein the outer loop of iterations are repeated until an evolved set of candidate solutions is produced that includes candidate aircraft intents with threads that are filled with instructions to span the entire observed trajectory.
The method of any preceding claim, providing multiple candidate aircraft intents with the best cost function values to a user for the user to select a preferred candidate aircraft intent.
The method of any preceding claim, further comprising ranking the provided candidate aircraft intents and the step of providing one or more candidate aircraft intents with the best cost function value or values respectively comprises either (a) providing a ranked list of candidate aircraft intents or (b) providing the highest ranked candidate aircraft intent.
The method of claim 12, wherein ranking the provided candidate aircraft intents comprises at least one of:
ranking according to the cost function values, ranking according to number of flight segments, and ranking
according to the frequency with which that candidate aircraft intent appears in the evolved set.
A computer system programmed to implement the method of any preceding claim.
A computer program comprising computer program instructions that, when executed, cause a computer system to implement the method of any of claims 1 to 13.
A computer readable medium having stored therein a computer program according to claim 15.