Smart Transport and Smart Buildings for Sustainable City Francesco Marcelloni Dipartimento di Ingegneria dell Informazione University of Pisa, Italy E-mail: francesco.marcelloni@unipi.it AI*IA 2014 - Smart Semantic Cities, Pisa, December 12, 2014
Outline Smart Transport The Smarty Project Urban Sensing Social Sensing Analysis of GPS traces Smart Buidings The E-tutor Project The SINCRO Project Smart Building Francesco Marcelloni
The Smarty Project SMARTY - SMArt Transport for sustainable city, funded by the Tuscany Region in the framework of Bando Unico R&S - 2012
The Smarty Project SMARTY - SMArt Transport for sustainable city, funded by the Tuscany Region in the framework of Bando Unico R&S - 2012
Our role in the Smarty project Urban Sensing Efficient data collection of air pollution data from low-cost sensors deployed in several strategical points (in cooperation with G. Anastasi and P. Bruschi) Low effort support to urban parking Social Sensing Real-Time Detection of Traffic from Twitter Stream Analysis Critical event detection from Facebook events analysis GPS trace analysis Real-time detection of traffic from GPS traces Francesco Marcelloni
Urban sensing Air quality typically monitored through large and expensive sensing stations Located in (few) strategic locations Accurate monitoring, but limited to specific areas Francesco Marcelloni
Urban Sensing Sensing stations are managed by public authorities pollution data are often not (promptly) available to citizens or they can be difficult to understand People are really interested in knowing air quality in places where they live street where their home is located school of their kids working place public gardens Francesco Marcelloni
Our Solution Based on low-cost sensor nodes equipped with appropriate gas sensors privately installed by citizens (Balcony, Garden, ) Sensor nodes are powered by batteries flexible deployment and easy relocation Users can share their measurements through social networks (cooperating sensing). Real-time and fine-grained monitoring Many low-cost sensors G. Anastasi, P. Bruschi, F. Marcelloni, U-Sense, A Cooperative Sensing System for Monitoring Air Quality in Urban Areas, ERCIM News, ISSN 0926-4981, July 2014, No. 98, pp.34-35. Francesco Marcelloni
usense Architecture Data collected by private sensors Consumed locally Made available to the community Services to city users Opportunistic communication Francesco Marcelloni
Services to City Users Through a web interface, a user can Check sensor locations View pollution map
Services to City Users Check gas concentrations in real-time
Services to City Users or plot their trend over time
Services to City Users Search for the less polluted routes
Urban parking Parking represents a crucial problem in urban life Wastes time and fuel Contributes to pollution and traffic jam A survey consisting of 1,400 questionnaires and proposed to citizens of Nicosia in Cyprus and Chania in Greece, shows that about 37% of the drivers spend more than 10 minutes searching for an available parking spot. Moreover, in 34% of the cases the drivers behave negatively in case no available parking spot is found nearby: Some park illegally causing traffic problems, whereas others leave the area cancelling their activities. Francesco Marcelloni
Urban parking We developed a system for managing occupation, booking, payment,.
Crucial aspect: spot identification Good solutions often fail to find widespread acceptance because of the cost of in-place deployment and maintenance
Crucial aspect: spot identification In the next future GPS coordinates. Now, for instance, the use of QR-Code SMARTY Platform Cost Euro: 2.10 Thank you Two hours later. Slot 12987 Free Slot 12987 Time 13:15 User: 34678 Slot 12987 Time 15:25 User: 34678 A. Bechini, F. Marcelloni, A. Segatori, Low-Effort Support to Efficient Urban Parking in a Smarty City Perspective, in Advances onto the Internet of Things, Advances in Intelligent Systems and Computing Series, Springer, Vol. 260, 2014, pp 233-252.
Social Sensing Tweet analysis aimed at detecting road traffic congestions and accidents discriminating traffic event due to an external cause (football match, procession, demonstration, flash-mob, etc.) notifying (in real-time) the users of the traffic event Facebook event analysis aimed at monitoring the number of partecipants along the time notifying the users when the event is likely to be critical
Traffic detection from Tweet analysis...i'm stuck in a 7 km...i'm...i'm stuck stuck in in a a 77 km km queue... queue... queue... Tokenization Stop-word filtering...i'm stuck in a 7 km...i'm...i'm stuck stuck in in a a 7 7 km km queue... queue... queue... TRAFFIC TRAFFIC TRAFFIC Fetch of SUMs and Pre-processing Stemming Stem filtering Classification of SUMs Feature representation Elaboration of SUMs Text of a sample tweet Sono bloccato in una coda di 7 km... il traffico è incredibile stasera! Voglio tornare a CASA!!! English translation: I'm stuck in a 7 km queue... traffic is unbelievable this night! Wanna get HOME!!! Text mining elaboration on a sample tweet tokens <sono>, <bloccato> <in>, <una>, <coda>, <di>, <7>, <km>, <il>, <traffico>, <è>, <incredibile>, <stasera>, <voglio>, <tornare>, <a>, <casa> Tokenization <sono>, <bloccato> <in>, <una>, <coda>, <di>, <7>, <km>, <il>, <traffico>, <è>, <incredibile>, <stasera>, <voglio>, <tornare>, <a>, <casa> Stop-word filtering Feature representation [arriv, blocc, caos, cod, km,...,..., stasera, traffic, vers, vial] F [0, w blocc, 0, w cod, w km,..., w stasera, w traffic, 0, 0] F Francesco Marcelloni Stem filtering <blocc>, <cod>, <7>, <km>, <traffic>, <incredibil>, <stasera>, <vogl>, <torn>, <cas> F relevant stems selected in the learning phase [arriv, blocc, caos, cod, km,..., stasera, traffic, vers, vial] F stems <bloccato>, <coda>, <7>, <km>, Stemming <traffico>, <incredibile>, <stasera>, <voglio>, <tornare>, <casa> <blocc>, <cod>, <7>, <km>, <traffic>, <incredibil>, <stasera>, <vogl>, <torn>, <cas>
Traffic detection from Tweet analysis Binary classification problem traffic vs. non-traffic tweets balanced 2-class dataset of 1330 tweets best accuracy: 95.75% using an SVM classifier Prec TP TP FP Rec TP TP FN F 2 2 Prec -score 1 Prec Rec Rec E. D Andrea, P. Ducange, B. Lazzerini, F. Marcelloni, Real-Time Detection of Traffic from Twitter Stream Analysis, Transactions on Intelligent Transportation Systems
Traffic detection from Tweet analysis Multi-class classification problem traffic due to external event vs. traffic congestion or crash vs. non-traffic tweets balanced 3-class dataset of 999 tweets best accuracy: 88.89% using an SVM classifier Francesco Marcelloni
Traffic detection from Tweet analysis Real-time monitoring campaign monitoring of several areas of the Italian road network detection of traffic events almost in real-time, often before online traffic news web sites (early detection) 4 traffic events detected on September, 26th, 2014 2 late detection events 2 early detection events During September and early Octobr 2014, 70 events detected by our system. The events are related both to highways and to urban roads. Francesco Marcelloni
Facebook event analysis Real-time monitoring of events using Facebook Critical event: at least K e persons probably will attend the event K e is determined based on the event features and context IF 2/3 * Num. Sure + 1/3 Probable > K e THEN the event is critical Analysis on the trend of the possible attendees {"type":"eventofacebookcritico","eventofb":{ "idfb":"365145446986497", "nome":"open day #master #alta #formazione", "descrizione":"una giornata di incontri ed orientamento per futuri studenti dei nostri master e corsi di alta formazione:\n\n- presentazione delle attività didattiche\n- workshop con coordinatore e docenti \n- incontro con ex alunni\n\nl\u0027\u0027\u0027\u0027open day è aperto a tutti.\n\ninizio corsi novembre 2014\niscrizioni ai corsi entro il 31 ottobre 2014 \n\npossibilità di colloqui individuali. sede: roma. \nper partecipare all\u0027\u0027\u0027\u0027open day è necessario registrarsi online http://goo.gl/rgg8wy", "owner":"75874409682", "location":"accademia di costume e di moda", "starttime":"2014-10-25t11:00:00+0200", "endtime":"data_stimata : 2014-10-25T13:00:00", "pointwkt":"point((12.468175254952 41.901237903208)", "partecipazione":{"attending":"22","maybe":"2","declined":"6"}}, "tipoeventoclassificato":"arte","angleindex":{"timeinms":67442172,"estimatedattending":22}},
GPS trace analysis (work in progress) Traffic and incident detection from GPS traces analysis Detect road traffic congestions and accidents Notify the users of a traffic alert containing Affected area Critical levels o slowed traffic o very slowed traffic o blocked traffic o incident Detected velocity of vehicles Francesco Marcelloni
GPS trace analysis Our approach Collection or simulation of GPS traces Construction of an optimised* digital map (based on Open Street Map) *further edge segmentation Matching of GPS traces on the digital map and completion of routes (direction of moving and routing) Expert system for traffic and incident detection (traffic alert) Spot traffic classification on the basis of GPS traces + Traffic alert notification on the basis of spatial and temporal analysis of classified spots Francesco Marcelloni
GPS trace analysis Spot classification based on the velocity of vehicles in the spot velocity of vehicles in the spot traffic states in the spot blocked very slowed slowed flowing absent Traffic alert notification based on spatial and temporal analysis of classified spots T=1 T=2 alert for incident with queue S very slowed very slowed blocked absent very slowed very slowed very slowed blocked absent T=3 very slowed very slowed very slowed very slowed blocked absent
GPS trace analysis Experimental results Used SUMO (Simulation of Urban Mobility) Simulations with 8157 cars (both working and non-working days) Simulated accidents were detected correctly It is also possible to detect the propagation of the traffic in the roads close to the accident False positives: just 2 in correspondence to traffic lamps Francesco Marcelloni
Smart Buildings
Non-Intrusive Load Monitoring Energy Consumption in Households A significant part of the electrical energy consumption in residential and business buildings is due to an improper use of the electrical appliances There is a growing interest in developing systems for profiling the use of electrical appliances and suggesting adequate policies for energy saving Information and Communication Technology (ICT) can play a crucial role Un Sistema INtelligente per un Consumo RespOnsabile dell'energia elettrica (SINCRO), funded by Tuscany region E-tutor: A low-cost system to monitor the use of electrical energy in buildings, funded by Fondazione Cassa di Risparmio di Lucca, Lucca, Italy Francesco Marcelloni
Intrusive Load Monitoring G. Anastasi, F. Corucci, F. Marcelloni, An Intelligent System for Electrical Energy Management in Buildings, Proc. International Conference on Intelligent Systems Design and Applications (ISDA 2011), Córdoba, Spain, November 22-24, 2011.
Intrusive Load Monitoring Pro ILM systems are characterized by a good accuracy in measuring appliance-specific power consumptions High costs Cons: Multiple sensor configuration and installation efforts Low Scalability particularly affects the implementation of ILM systems when the monitoring scenario involves a large number of appliances Francesco Marcelloni
Our objective To design a low cost and intelligent energy monitoring system To introduce a novel technique to extract the individual power consumption of a set of appliances from aggregate measures collected by a smart meter To exploit computational intelligence advances for both appliance modeling and to extract individual appliance power consumption from aggregate measures E-tutor: A low-cost system to monitor the use of electrical energy in buildings, funded by Fondazione Cassa di Risparmio di Lucca, Lucca, Italy. Francesco Marcelloni
Our Non-Intrusive Load Monitoring The working states of each appliance are known but we do not associate them with a specific power consumption value The events that trigger a state transition are described in terms of linguistic labels, such as low, medium and high We define fuzzy linguistic variables are defined on the variations of the real (P) and reactive (Q) powers The use of fuzzy sets allows us to deal with the tolerance of the smart meters and the noise which affects the measures The linguistic terms permit to coarsely describe the events, thus enabling the modelling of appliance type rather than single appliances P. Ducange, F. Marcelloni, M. Antonelli, A Novel Approach based on Finite State Machines with Fuzzy Transitions for Non-Intrusive Home Appliance Monitoring, IEEE Transactions on Industrial Informatics, Vol. 10, N. 2, 2014, pp. 1185-1197.
Appliance modelling Linguistic variables defined for DP and DQ We model a generic fuzzy transition T(S i ->S j ) as a rule: Whenever a new couple of variations (Dp t,dq t ) is measured the firing strength of each involved rule is calculated OFF_ON rule If the firing strength is higher than a threshold then the transition associated with the corresponding rule is labeled as candidate transition.
Disaggregation Algorithm We handle several candidate transitions and make hypotheses on the actual configuration and therefore on the power consumption of each appliance The main data structures handled by the load disaggregation algorithm are: the configuration of active appliances at instant t, that is, a collection of triplets (S i,a (t), P i,a (t), Q i,a (t)) the collection CC(t) of active configurations at instant t the list LC of collections CC(t), CC(t-1),..., CC(1) appliances
Disaggregation Algorithm When the algorithm analyzes a new couple (Dp t,dq t ) at instant t, a list of candidate transitions is created for each configuration in collection CC(t-1). The list is generated by considering: for the appliances in the DB, the OFF_ON rules whose firing strengths are higher than a prefixed threshold for each active appliance in the configuration, all the possible transitions from the current state to another state allowed by the corresponding FSMFT
Disaggregation Algorithm At t=2, the new variation fires the OFF-ON rules of both a hair dryer and a lamp At t=3, the new variation is compatible only with the change of the working state of the hair dryer
Disaggregation Algorithm Installation of the proposed NILM system We collected the aggregated measures of P and Q by using the Plogg Ext CT 100 smart meter equipped with ZigBee wireless communication and an external split core 100A current transducer
Disaggregation Algorithm
Disaggregation Algorithm Hair Dryer Toaster Oven Food Cutter
Disaggregation Algorithm
Disaggregation Algorithm
Conclusions Smart City An aggregation of competences, technologies, ideas, innovations, algorithms, Why are not our cities very smart? Old infrastructures Old buildings Lack of investiments. IMPACT!!!! Thank you very much Questions? Francesco Marcelloni