predicting car park occupancy rates in smart cities

PREDICTING CAR PARK OCCUPANCY RATES

IN SMART CITIES

Daniel H. Stolfi1

[email protected]

Enrique Alba1

[email protected]

Xin Yao2

[email protected]

1Departamento de Lenguajes y Ciencias de la Computación,University of Malaga, Spain

2CERCIA, School of Computer Science,University of Birmingham, Birmingham, U.K.

International Conference on Smart CitiesSmart-CT 2017

Málaga, SpainJune 14-16 2017

CONTENTS

1 INTRODUCTION

2 OUR PROPOSAL

3 EXPERIMENTS

4 CONCLUSIONS & FUTURE WORK

Daniel H. Stolfi, Enrique Alba & Xin Yao Predicting Car Park Occupancy Rates in SC 1 / 21

CONTENTS

1 INTRODUCTION

2 OUR PROPOSAL

3 EXPERIMENTS

CONTENTS

1 INTRODUCTION

2 OUR PROPOSAL

3 EXPERIMENTS

CONTENTS

1 INTRODUCTION

2 OUR PROPOSAL

3 EXPERIMENTS

IntroductionOur ProposalExperiments

Conclusions & Future Work

Road TrafficParking in a Big City

INTRODUCTION

Nowadays in our cities. . .

There is a larger number of vehicles in the streetsFinding an available parking space is hardTime and Fuel are wasted in finding a free spaceTons of greenhouse gases are emitted to theatmosphereThe citizens’ quality of life is decreasing

INTRODUCTION

SENSORS

Sensors reporting car park occupancy

Predicting Car Park Occupancy RatesCase Study: Birmingham U.K.Predictors Analyzed

SYSTEM ARCHITECTURE

BIRMINGHAM, U.K.

32 Car Parks32 Car ParksDaniel H. Stolfi, Enrique Alba & Xin Yao Predicting Car Park Occupancy Rates in SC 5 / 21

DATA SOURCE

Data set:Oct 4th to Dec 19th

From 9am to 5pm18 measures per day

32 car parks36,285 occupancy measures

PREDICTORS

POLYNOMIALS, FOURIER, TIME SERIES & K-MEANS

Polynomial Fitting

Fourier Series

Polynomial Fitting

Fourier Series

Time Series

Polynomial Fitting

Fourier Series

Time Series

K-Means

KM-POLYNOMIALS AND SHIFT & PHASE

F (x) = a0 + a1 · x + a2 · x2 + . . .+ an · xn

F (x) =(a0 + δ) + a1 · (x + φ) + a2 · (x + φ)2 + . . .+ an · (x + φ)n

δ = Shift, φ = Phase

F (x) = a0 + a1 · x + a2 · x2 + . . .+ an · xn

TrainingPredictingPrototype

TRAINING

Training

K-FOLD CROSS VALIDATION

MEAN SQUARED ERROR (MSE)

MSE = 1n∑

i(yi − fi)2

K-FOLD TRAINING RESULTS (I)

Polynomial Fitting

Polynomial Fitting Fourier Series

K-FOLD TRAINING RESULTS (II)

K-Means

K-Means KM-Polynomials

SHIFT & PHASE AND TIME SERIES

Shift & Phase

Shift & Phase Time Series

PREDICTION RESULTS: UNSEEN WEEK

Which car park will be best for metomorrow?

And the day after tomorrow?

PREDICTION EXAMPLES

Working Days

PREDICTION EXAMPLES

Working Days Weekends

PARKING IN BIRMINGHAM

PREDICTION STATS

ConclusionsFuture Work

CONCLUSIONS

Six predictor for forecasting car park occupancy ratesTrained by using real parking dataTested with one week of unseen parking data

Time Series shows the best results during working daysShift & Phase has good results during weekends

We have presented a web based prototype

CONCLUSIONS

FUTURE WORK

Repeat this study using a larger training data set andother citiesInclude new predictors in the comparison

Develop an application for mobile phones

FUTURE WORK

QUESTIONS

Predicting Car Park Occupancy Rates in Smart Cities

Prototype: http://mallba3.lcc.uma.es/parking/

Questions?Daniel H. [email protected]

Enrique [email protected]

Xin Yao http://[email protected] http://danielstolfi.com

Acknowledgements: This research has been partially funded by Spanish MINECO project TIN2014-57341-R (moveON). Daniel H. Stolfi issupported by a FPU grant (FPU13/00954) from the Spanish Ministry of Education, Culture and Sports. University of Malaga. InternationalCampus of Excellence Andalucia TECH.

PARAMETERIZATION

Training days: Oct 4th to Dec 12th

Testing Week: Dec 13th to Dec 19th

Predictor Parameter TrainingPolynomials: 2o Degree Fold: 1

Fourier Series: 3 Components Fold: 1K-Means: 3 Clusters Fold: 1

KM-Polynomials: 2o Degree Fold: 1Shift & Phase : - Fold: 1

Time Series : - Weeks: 8

THREE CLUSTERS

Weekdays in eachcluster

THREE CLUSTERS

Occupancy values ineach cluster

THREE CLUSTERS

Occupancy values ineach cluster

KM-Polynomials andShift & Phase fitting