Periodic Reporting for period 2 - ISOBAR (Artificial Intelligence Solutions to Meteo-Based DCB Imbalances for Network Operations Planning)
Período documentado: 2021-06-01 hasta 2022-11-30
European airspace capacity is near its limit, with all-causes ATFM (Air Traffic Flow Management) delays doubling from 2014 to 2018. It is critical for network management to address traffic demand and capacity balancing in an efficient way, and ensure that associated measures contribute to minimize delays at network and local levels.
Network prediction and performance is very sensitive to weather and the uncertainty in its prediction. In addition, current ATFCM operations are not evaluated from a systematic perspective. These two factors together lead to a strong dependency on the experience of human operators. ISOBAR addresses these challenges through the contribution to an Artificial Intelligence (AI)-based Network Operations Plan, by including in its scope an enhanced weather prediction tailored to ATFCM, ATM and weather data integration, demand and capacity (DC) imbalance characterization and imbalance mitigation prescription.
To achieve this vision, four objectives are set:
a) Reinforce collaborative ATFCM processes at pre-tactical and tactical levels into the LTM (local) and Network Management (network) roles integrating dynamic weather cells.
b) Characterisation of demand and capacity imbalances at pre-tactical level [-1D, -30min] depending on the input of probabilistic weather cells by using applied AI methods and ATM and weather data integration.
c) User-driven mitigation plan considering AUs priorities (and fluctuations in demand based on weather forecasts) and predicted effectiveness of ATFCM regulations, considering flow constraints and network effects.
d) Develop an operational and technical roadmap for the integration of ancillary services (providing AI-based hotspot detection and adaptative mitigation measures) into the NM platform, by defining interfaces, functional and performance requirements.
Aviation provides the only rapid worldwide transportation network, which makes it essential for global business: economic growth, creation of jobs and international trade and tourism. The project contributes to a more efficient air transport system, which consequently promotes an improved quality of life and helps to improve living standards.
Network prediction and performance is very sensitive to weather and the uncertainty in its prediction. In addition, current ATFCM operations are not evaluated from a systematic perspective. These two factors together lead to a strong dependency on the experience of human operators. ISOBAR addresses these challenges through the contribution to an Artificial Intelligence (AI)-based Network Operations Plan, by including in its scope an enhanced weather prediction tailored to ATFCM, ATM and weather data integration, demand and capacity (DC) imbalance characterization and imbalance mitigation prescription.
To achieve this vision, four objectives are set:
a) Reinforce collaborative ATFCM processes at pre-tactical and tactical levels into the LTM (local) and Network Management (network) roles integrating dynamic weather cells.
b) Characterisation of demand and capacity imbalances at pre-tactical level [-1D, -30min] depending on the input of probabilistic weather cells by using applied AI methods and ATM and weather data integration.
c) User-driven mitigation plan considering AUs priorities (and fluctuations in demand based on weather forecasts) and predicted effectiveness of ATFCM regulations, considering flow constraints and network effects.
d) Develop an operational and technical roadmap for the integration of ancillary services (providing AI-based hotspot detection and adaptative mitigation measures) into the NM platform, by defining interfaces, functional and performance requirements.
Aviation provides the only rapid worldwide transportation network, which makes it essential for global business: economic growth, creation of jobs and international trade and tourism. The project contributes to a more efficient air transport system, which consequently promotes an improved quality of life and helps to improve living standards.
WP1 Enhanced ATFCM process
ISOBAR has proposed an enhancement of the Collaborative ATFCM process both at local (FMP) and network (NM) level by describing a process based on the integration of convective weather probabilistic forecast tailored to the spatial and temporal granularities of the ATFCM process. The enhanced ATFCM process has been described in a SESAR OSED, including the Gate concept, the Netspot concept and the operational integration of new AI components. A global communication prototype has been developed over a real Network Manager dashboard.
WP2 Probabilistic Convection Input to ATFCM
The ISOBAR MetEngine provides meteorological forecasts at network and local levels based on the use of Artificial Intelligence along with Numerical Weather Prediction (NWP). Three different NWP products have been chosen:
• AROME-EPS for France region,
• the AEMET (γSREPS for the Iberian Peninsula,
• and the ECMWF (European Centre for Medium-Range Weather Forecasts) EPS for Western Europe.
A set of web-based ISOBAR MetEngine dashboards has been developed.
WP3 Demand and Capacity Prediction
The work performed has been split in two threads:
• AUs Alternative Trajectories Demand Characterisation. AU preference has been adopted through two following factors: Aircraft Type and Aircraft Operator. The predictions have been obtained for triplets (city-par, aircraft type and aircraft operator)
• Capacity Decay Prediction. Three approaches were considered: Using gates, where a set of gates were defined over the European airspace, and capacity was defined as the volume of traffic crossing a gate. Using a traffic flow approach, where the flight trajectories were used to reconstruct an image of the traffic patterns over a region. Using sectors.
WP4 Automated DCB Solver Suite
The work on DCB solvers has been focused on delivering artificial-intelligence-based solvers based on two different artificial-intelligence paradigms:
• Multi-Agent Reinforcement Learning (MARL);
• Simulated-Annealing-based Hyper-Heuristic.
In addition, 4 non-AI solvers have been developed: Greedy regulations, Cherry-picking, Optimised regulations and Hybrid.
WP5 Data integration and architecture
This work has corresponded to data Extract, Transform and Load (ETL) in the project. Data ETL has served the integration of the new proposed services into a local or a NM platform by processing inputs and providing early architecture analysis. The results have crystallised in a set of diagrams depicting the operational dataflows between the ISOBAR solution components that would form the target operational architecture.
WP5 & WP6 ISOBAR Solution Prototype and Evaluation
The validation of ISOBAR concept and technical developments has been addressed through validation tasks, organised around four activities and two exercises:
• ACT01: to assess the performance of a Machine Learning model providing probabilistic convective weather information.
• ACT02: model providing weather-related capacity reduction and imbalance prediction.
• ACT03: model providing mitigation plans.
• ACT04: model providing AU Preferred Trajectory alternatives.
• EXE01: operational evaluation of ISOBAR Collaborative Framework.
• EXE02: Fast-Time Validation Exercise to assess the performance benefits of the global solution.
The operational evaluation has proven realistic behaviour and logical solutions.
The performance evaluation has addressed four KPAs: capacity, predictability, environment and safety.
ISOBAR has proposed an enhancement of the Collaborative ATFCM process both at local (FMP) and network (NM) level by describing a process based on the integration of convective weather probabilistic forecast tailored to the spatial and temporal granularities of the ATFCM process. The enhanced ATFCM process has been described in a SESAR OSED, including the Gate concept, the Netspot concept and the operational integration of new AI components. A global communication prototype has been developed over a real Network Manager dashboard.
WP2 Probabilistic Convection Input to ATFCM
The ISOBAR MetEngine provides meteorological forecasts at network and local levels based on the use of Artificial Intelligence along with Numerical Weather Prediction (NWP). Three different NWP products have been chosen:
• AROME-EPS for France region,
• the AEMET (γSREPS for the Iberian Peninsula,
• and the ECMWF (European Centre for Medium-Range Weather Forecasts) EPS for Western Europe.
A set of web-based ISOBAR MetEngine dashboards has been developed.
WP3 Demand and Capacity Prediction
The work performed has been split in two threads:
• AUs Alternative Trajectories Demand Characterisation. AU preference has been adopted through two following factors: Aircraft Type and Aircraft Operator. The predictions have been obtained for triplets (city-par, aircraft type and aircraft operator)
• Capacity Decay Prediction. Three approaches were considered: Using gates, where a set of gates were defined over the European airspace, and capacity was defined as the volume of traffic crossing a gate. Using a traffic flow approach, where the flight trajectories were used to reconstruct an image of the traffic patterns over a region. Using sectors.
WP4 Automated DCB Solver Suite
The work on DCB solvers has been focused on delivering artificial-intelligence-based solvers based on two different artificial-intelligence paradigms:
• Multi-Agent Reinforcement Learning (MARL);
• Simulated-Annealing-based Hyper-Heuristic.
In addition, 4 non-AI solvers have been developed: Greedy regulations, Cherry-picking, Optimised regulations and Hybrid.
WP5 Data integration and architecture
This work has corresponded to data Extract, Transform and Load (ETL) in the project. Data ETL has served the integration of the new proposed services into a local or a NM platform by processing inputs and providing early architecture analysis. The results have crystallised in a set of diagrams depicting the operational dataflows between the ISOBAR solution components that would form the target operational architecture.
WP5 & WP6 ISOBAR Solution Prototype and Evaluation
The validation of ISOBAR concept and technical developments has been addressed through validation tasks, organised around four activities and two exercises:
• ACT01: to assess the performance of a Machine Learning model providing probabilistic convective weather information.
• ACT02: model providing weather-related capacity reduction and imbalance prediction.
• ACT03: model providing mitigation plans.
• ACT04: model providing AU Preferred Trajectory alternatives.
• EXE01: operational evaluation of ISOBAR Collaborative Framework.
• EXE02: Fast-Time Validation Exercise to assess the performance benefits of the global solution.
The operational evaluation has proven realistic behaviour and logical solutions.
The performance evaluation has addressed four KPAs: capacity, predictability, environment and safety.
The innovation in ISOBAR concept applied to network management is focused in:
• Improved Collaborative ATFCM (ConOps and AUs involvement): New flight planning services with enriched DCB and MET information will give to civil Airspace Users the flexibility to optimize their operational flight plans and to accommodate their business needs without compromising optimum ATM system outcome and the performances of all stakeholders.
• Non-Nominal Weather Situations and Probabilistic Storm Prediction: In the ISOBAR project, forecasts of probability of convection will be improved for tactical lead times by increasing the update frequency and the spatial resolution.
• AI Demand Prediction: As a progress beyond analytical and deterministic methods, Machine Learning libraries will be developed to predict probabilistic demand variability associated to probabilistic forecasts of weather cells. Moreover, demand prediction will take into account AU’s needs and their possible reactions to weather forecasts.
• Meteo and ATM Data integration: The vision of ISOBAR is to transform forecasts in tailored and consistent meteorological information with the potential to be deployed as a NM service.
• Probabilistic DCB Imbalance: Automation will be introduced to reduce uncertainty in the process and to provide operators with better situational awareness thanks to the development of the technological components of ISOBAR.
• Learning-based DCB Solution: leverage the latest advances in machine learning to present a learning-based DCB solver where the uncertainty profiles of both demand and capacity predictions and the collaborative planning are taken into account. Expected benefits are in terms of AUs’ preferences compliances, more efficient flow patterns and reduction of ATC workload.
• Improved Collaborative ATFCM (ConOps and AUs involvement): New flight planning services with enriched DCB and MET information will give to civil Airspace Users the flexibility to optimize their operational flight plans and to accommodate their business needs without compromising optimum ATM system outcome and the performances of all stakeholders.
• Non-Nominal Weather Situations and Probabilistic Storm Prediction: In the ISOBAR project, forecasts of probability of convection will be improved for tactical lead times by increasing the update frequency and the spatial resolution.
• AI Demand Prediction: As a progress beyond analytical and deterministic methods, Machine Learning libraries will be developed to predict probabilistic demand variability associated to probabilistic forecasts of weather cells. Moreover, demand prediction will take into account AU’s needs and their possible reactions to weather forecasts.
• Meteo and ATM Data integration: The vision of ISOBAR is to transform forecasts in tailored and consistent meteorological information with the potential to be deployed as a NM service.
• Probabilistic DCB Imbalance: Automation will be introduced to reduce uncertainty in the process and to provide operators with better situational awareness thanks to the development of the technological components of ISOBAR.
• Learning-based DCB Solution: leverage the latest advances in machine learning to present a learning-based DCB solver where the uncertainty profiles of both demand and capacity predictions and the collaborative planning are taken into account. Expected benefits are in terms of AUs’ preferences compliances, more efficient flow patterns and reduction of ATC workload.