Airport capacity prediction system

Patent metadata

Field	Value
Publication number	US-2021358313-A1
Application number	US-202117316509-A
Country	US
Kind code	A1
Filing date	May 10, 2021
Priority date	May 13, 2020
Publication date	Nov 18, 2021
Grant date	—

Abstract

Examples provide a method and apparatus for predicting airport utilization. A plurality of data sources provides raw flight data and weather data which is stored in a key value data store. The incoming raw data is cleaned, correlated and indexed. The processed data is stored in an interim data store. A plurality of competing predictive models use the processed data and historical data to generate predicted arrivals and departures for a selected runway during a selected time-period. The predicted data is compared to actual arrival and departures for the selected runway. The ML model generating predicted results most closely matching the actual arrival and departure data is selected. The selected ML model generates predicted future arrivals and future departures data for the runway based on real-time flight data and weather data for output via a user interface device.

First claim

What is claimed is:

1 . A system for predicting airport utilization, the system comprising:
a computing device comprising a computer-readable medium storing instructions that are operative upon execution by a processor to:
generate a plurality of competing sets of prediction results, by a plurality of competing machine learning (ML) models analyzing flight data and weather data associated with at least one of a set of aircraft and a selected airport, wherein a set of prediction results from the plurality of competing sets of prediction results comprises a set of predicted arrivals and a set of predicted departures for at least one runway during a selected time-period, each set of prediction results generated by each candidate ML model from the plurality of competing ML models;
compare the competing sets of prediction results with actual arrival and departure data for the selected airport during the selected time-period;
select a ML model from the plurality of competing ML models generating highest accuracy predictions.

2 . The system of claim 1 , wherein the instructions are further operative to:
extract a set of salient historical data from historical flight-related data using PCA to eliminate bias and variance, wherein the set of salient historical data comprises data associated with at least one of airport features, environmental features, spatial features and temporal features associated with the runway; and store the set of salient historical data in an interim data store for utilization by the plurality of competing ML models.

3 . The system of claim 1 , wherein the instructions are further operative to:
process raw data by the plurality of competing predictive models, wherein the processing of the raw data further comprises:
cleaning the raw data to remove extraneous data;
correlating the raw data;
indexing the raw data; and
storing the processed data in a key-value data store for query processing.

4 . The system of claim 1 , wherein the instructions are further operative to:
train the plurality of competing ML models with the flight data, the weather data and real-time feedback associated with aircraft arrivals and departures, wherein one or more hyperparameters associated with at least one candidate ML model is tuned specifically for airport capacity prediction during training.

5 . The system of claim 1 , wherein the instructions are further operative to:
deploy a set of microservices on a cloud server, wherein the set of microservices build, train and test the plurality of competing ML models to generate the plurality of competing sets of prediction results using real-time feedback and training data.

6 . The system of claim 1 , further comprising:
a plurality of data sources providing the flight data and the weather data, wherein the instructions are further operative to:
analyze data associated with a selected runway by the selected ML model; and
provide per-runway utilization data for the runway, wherein the per-runway utilization data comprises congestion information for the selected runway, including predicted aircraft arrival and departure counts for the selected runway at a selected airport within a selected future time-period.

7 . The system of claim 1 , wherein the instructions are further operative to:
provide real-time flight data and real-time weather data received from a plurality of data sources to the selected ML model; obtain predicted airport capacity for a runway generated by the selected ML model based on real-time flight data and weather data received from the plurality of data sources; and output the predicted airport capacity generated by the selected ML model to at least one user interface device, wherein the predicted airport capacity comprising predicted future aircraft arrival counts and predicted future aircraft departure counts for the runway during a set of pre-configured time-periods, wherein the predicted future aircraft arrival counts.

8 . The system of claim 7 , wherein the instructions are further operative to:
adjusting at least one candidate ML model in the plurality of competing ML models based on the real-time flight data and the real-time weather data.

9 . A method for predicting airport utilization, the method comprising:
utilizing a plurality of competing machine learning (ML) models configured to analyze flight data and weather data with actual arrival and departure data associated with a runway to generate competing sets of prediction results, a set of prediction results from the competing sets of prediction results comprising a set of predicted arrivals and a set of predicted departures for a runway in the runway during a selected time-period generated by a candidate ML model in the plurality of competing ML models; comparing the competing sets of prediction results generated by the plurality of competing ML models with actual arrival and departure data for the runway during the selected time-period; selecting a candidate ML model from the plurality of competing ML models generating highest accuracy predictions; and outputting predicted future arrivals and future departures data for the runway generated by the selected ML model based on real-time flight data and weather data received from a plurality of data sources.

10 . The method of claim 9 , further comprising:
extracting a set of salient historical data from historical flight-related data using PCA to eliminate bias and variance, wherein the set of salient historical data comprises data associated with at least one of airport features, environmental features, spatial features and temporal features associated with the runway.

11 . The method of claim 10 , further comprising:
storing the set of salient historical data in an interim data store for utilization by the plurality of competing ML models.

12 . The method of claim 9 , further comprising:
displaying predicted future aircraft arrival counts and predicted future aircraft departure counts for the runway during a set of pre-configured time-periods to a user interface device, wherein the predicted future aircraft arrival counts and the predicted future aircraft departure counts are generated by the selected ML model.

13 . The method of claim 9 , further comprising:
processing raw data by the plurality of competing ML models, wherein the processing of the raw data further comprises:
cleaning the raw data to remove extraneous data;
correlating the raw data;
indexing the raw data; and
storing the processed data in a key-value data store for query processing.

14 . The method of claim 9 , further comprising:
deploying a set of microservices on a cloud server, wherein the set of microservices build, train and test the plurality of competing ML models to generate the competing sets of prediction results using real-time feedback and training data.

15 . The method of claim 9 , wherein the plurality of ML models comprises at least one of a linear ML model, a non-linear ML model, an ensemble ML model, lasso ML model, elastic net regression, classification and regression trees, support vector regression, k-nearest neighbors, adaptive boosting, gradient boosting, random forest regression, extra trees regression, and recurrent neural networks.

16 . A computer program product, comprising a computer usable medium having a computer readable program code embodied therein, the computer readable program code adapted to be executed to implement a method for predicting airport utilization runway, the method comprising:
training a plurality of competing machine learning (ML) mo

Assignees

• Boeing Co

Inventors

• Ayhan Samet M
• Romani De Oliveira Italo
• Leite Alexandre Carvalho
• Costas Alvarez Pablo
• Wilson Ian Andrew Byham
• Biglin Michael
• Pinto Neto Euclides Carlos

Classifications

• G06N3/044

  Recurrent networks, e.g. Hopfield networks · CPC title

• G06N5/01

  Dynamic search techniques; Heuristics; Dynamic trees; Branch-and-bound · CPC title

• G06N3/045

  Combinations of networks · CPC title

• G06N3/0985

  Hyperparameter optimisation; Meta-learning; Learning-to-learn · CPC title

• G06N3/0442

  characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU] · CPC title