The iLU project is the first initiative combining heterogeneous sources of mobility data with a broad diversity of situational context factors (such as public events, weather, accidents, public transport plans, construction works, parking density, and city ornament) for the comprehensive characterization and prediction of mobility patterns; and their proper translation into strategic and real-time public initiatives, using a selected study area.

To this end, iLU aims to learn advanced descriptive, predictive and prescriptive models from urban data produced by different sensor modalities spread across the Lisbon city, including fixed sensors (such as loop counters, surveillance cameras or machinery at public stations) and mobile sensors (carried by drivers, pedestrians and within public transports), as well as context data from complementary public repositories. 

Despite the recent efforts established by the Lisbon City Council (CML) and advances on urban computing research, the purpose of the iLU project is challenged by six major problems:
1) although urban data in Lisbon city is abundant and maintained by CML in Plataforma de Gestão Inteligente de Lisboa (PGIL), it is not yet consolidated. Data integration fails at three levels: 
(i) different sensor technologies (such as geolocalized speed and loop counters for traffic monitoring); 
(ii) heterogeneous modalities of mobility (such as private, bus, subway, train and bike modalities of transportation); 
(iii) urban data augmentation with their situational context (such as public events, interdictions, or surrounding buildings of interest like schools);
2) descriptive analytics from urban data are mostly focused on a single data source and fail to find non-trivial and actionable patterns of mobility;
3) traffic forecasters are unable to make accurate predictions for time horizons above 15 minutes due to uncertainty and efficiency factors;
4) predictive models of city traffic disregard situational context data;
5) state-of-the-art networks of agents for traffic flow analysis disregard predictive and context information;
6) the utility of data-centric simulation and control studies for mobility optimization remains unexplored.

These challenges prevent a comprehensive understanding of non-trivial and actionable traffic dynamics in Lisbon. 
In this context, the iLU project aims to address these challenges by accomplishing the following six major objectives:

1) develop an integrative data repository that consolidates available urban data in the PGIL platform with potential impact on city traffic analysis. To this end, well-established principles on data integration and cleaning will be considered to: (a) guarantee the consolidation of shared attributes between different sources, (b) support automatic updates in the presence of more recent data, (c) remove errors, and (d) facilitate the subsequent development of data mining algorithms;

2) propose algorithms to learn comprehensive descriptive models of mobility dynamics and, more importantly, mine actionable traffic patterns with a focus on: (a) frequent and periodic patterns; (b) emerging patterns; (c) deviant patterns; and (d) multi-source correlations. To this end, state-of-the-art principles on urban data fusion and analytics, relational data mining, and spatiotemporal pattern mining will be considered;

3) propose superior predictive algorithms of traffic circulation from multiple sensor data sources using pattern-centric views (such as associative classifiers) and deep learning views (such as deep recurrent neural networks);

4) propose algorithms to describe and anticipate traffic dynamics in the presence of situational context data. Here, iLU focus will be placed on: (a) discovering comprehensive correlations between traffic states and situational context factors using subspace clustering; and (b) enhancing predictive algorithms in the presence of future situational context such as upcoming events or weather forecasts;

5) propose networks of reinforcement learning agents that are able to optimize traffic flow and adapt to context information (given by public events, weather forecasts, construction works). In this context, iLU will combine state-of-the-art architectures for traffic simulation with recent advances on deep reinforcement learning;

6) integrate previous descriptive, predictive and prescriptive contributions within a functional prototype and comprehensively assess its ability to support strategic decisions (such as city mobility planning) and real-time decisions (such as positive traffic conditioning through intelligent traffic lights and road message panels) using pilot initiatives.

The listed objectives build upon each other: urban data is consolidated (1) for integrative descriptive analytics (2) and the discovery of context-sensitive mobility patterns (3) that support the learning of robust predictive models (4) used within control studies (5) to support real-world decisions (6).

In line with the six listed contributions, our research plan is organized in six vertical tasks (Tasks 3-8) and two transversal tasks (Tasks 1 and 2).
Task 1 guarantees the proper operation of administrative, technical, and financial aspects of the iLU project.
Task 2 establishes all relevant specifications and evaluation metrics to guarantee that contributions are developed in accordance with a well-defined set of requirements and design principles, revised and agreed by all participants.
Tasks 3 and 4 provide the infrastructural and computational means to update, extract, navigate and explore the wide-diversity of available urban and situational data.
Tasks 5 and 6 propose advanced descriptive and predictive models of traffic dynamics able to leverage on heterogeneous data sources and on situational context data.
Task 7 uses previous statistical models in the context of control agents to answer traffic-related decision problems.
Task 8 integrates previous contributions within a recommendation system and assesses its utility in the context of pilot demonstrations. 

The aforementioned tasks will culminate in five major artifacts (milestones): 
detailed specifications of available data inputs, target problems ranked by priority, and desirable outputs;
M2) integrative data repository consolidating available mobility data and situational context data;
M3) context-sensitive descriptive and predictive algorithms to explore and forecast mobility dynamics and bottlenecks;
M4) simulation and control algorithms to support traffic-related decisions;
M5) recommendation system integrating all previous computational artifacts to support strategic and real-time decisions.

The aforementioned contributions will provide actionable descriptive, predictive and prescriptive models to support: 
1) strategic decisions, including sustainable city mobility planning, placement of sensors and message panels, and public transport routes;
2) positive traffic conditioning, including dynamic routing constraints, real-time messaging using alerts and road panels, or intelligent traffic light control;
3) context-sensitive waiting time recommendations for public and private transports;
4) effective coordination between authorities, municipalities and public transport operators. 

The iLU project will primarily focus on positive traffic conditioning and on additional one-to-three use cases/applications to validate the scientific contributions. Nevertheless, the target DSS will be designed and implemented using generalization, adaptability and extensibility principles so that iLU contributions can be easily refined to answer alternative applications. 

The computational contributions will be integrated within an auditable decision support system (DSS) to be promptly used by CML and its mobility partners. Given the high volume of available urban data and the need to learn comprehensive and actionable statistical models, the computational contributions of iLU will consider distribution principles and the DSS will be attemptively deployed over the National Infrastructure for Distributed Computing (, of which LNEC is a founding member.

To accomplish the established objectives, iLU relies on a multidisciplinary team joining experts from INESC-ID and LNEC, and a team of mobility policy entrepeuners at CML:
(a) INESC-ID gathers researchers from IDSS and GAIPS groups. The Intelligent Decision Support Systems (IDSS) group has extensive expertise on data mining, including numerous publications on heterogeneous and temporal data analysis (Henriques, 2015a; Henriques, 2015b); while the Intelligent Agents and Synthetic Characters (GAIPS) group has renowned researchers on reinforcement learning and multi-agent control systems (Melo, 2016);
(b) LNEC gathers highly qualified researchers from the Mobility Department with extensive expertise on sustainable city mobility, road safety policies, transport planning, intermodal mobility, and urban sensor systems (Arsenio, 2016);
(c) CML gathers mobility strategists and the coordinators of the Urban Data Lab and the Open Data Portal initiative ( developed by the Centro Operacional Intregado (COI), a division responsible to create an intelligent decision support platform from city data (

The project structure and experience of INESC-ID, LNEC and CML will ensure that iLU establishes the necessary foundations to support both strategic and real-time decisions with impact on city mobility. In particular, the prototype development of a recommendation system combining the consolidated data infrastructure and the learning algorithms, together with a pilot demonstration to be undertaken with relevant stakeholders, will guarantee the prompt applicability of iLU contributions in real-world administrative and operational decisions on mobility.