The distribution of high-quality drinking water is a major challenge for water distribution network operators. In this context, we propose an algorithm for contamination warning systems that partitions the network into areas of homogeneous water quality, thereby facilitating the identification of contaminated areas. To this end, we introduce a dedicated model-based approach that simultaneously clusters and aligns the conductivity signals measured at the network probes. We use the shifts inferred by the model to summarize the water propagation in a cluster. To produce physically meaningful clusters, we integrate expert knowledge about the physics of the signals into the curve clustering problem using a Hidden Markov Random Field (HMRF). We present a methodology for model selection and apply our approach to real conductivity curves from a large water distribution network. • Incorporate prior knowledge in the simultaneous clustering and alignment of curves. • Apply to water conductivity curves to find areas of homogeneous water quality. • Summarize the water propagation in a cluster using the inferred shifts.

The implementation of real-time gait mode detection is paramount in effectively providing tailored support for individuals utilizing actuated ankle-foot orthoses (AAFOs), thereby enhancing their walking capabilities and overall mobility. However, existing systems often rely on multiple sensors and struggle with accurate and prompt detection of gait transitions, especially in varied and challenging environments. This study aims to develop a novel real-time gait mode detection system that accurately identifies five essential daily living gait modes, namely level walking, ramp ascent/descent, and stair ascent/descent using only two foot-mounted inertial measurement units (IMUs). By using a long short-term memory (LSTM)-based algorithm trained on data collected from ten healthy subjects, the system extracts six kinematic features to predict gait modes with high accuracy. The proposed method integrates this detection system with a task-oriented control strategy to adapt the control of the AAFO based on the identified gait modes. The real-time experiments involving three healthy participants demonstrated robust gait mode detection, achieving an average estimation accuracy of$98 \pm 1$% across the five gait modes, even with the application of assistive torque. In cases mimicking abnormal gait, the system maintained an accuracy of$93 \pm 3$%. Additionally, each transition delay between gait modes was analyzed, showing that gait mode detection can occur between the transitions of the leading and trailing foot. The results of the control strategy showed a reduction in muscle activation of the dorsiflexor and plantarflexor muscles as measured by EMG, as well as improved tracking performance during the swing phase. Gait mode detection robustness was further evaluated by including walking with obstacles and changes in environmental dimensions.

Forecasting counts data in transportation areas can enrich passenger information for public transport passengers, who may thus better plan their trips. Moreover, forecasting with uncertainty is particularly important in the transportation domain, where the risk of poorly managed high demand is to be avoided. In this paper, we propose a new probabilistic prediction model well-suited for multivariate, overdispersed, and possibly correlated count data. This model combines the strength of the deep learning framework with the modeling of counts data allowed by “sums and shares” distributions. Indeed, deep learning models can handle uncertainty by relying on an abstraction of contextual data and by assuming output distributions. Our model learns a latent representation of the input data with the help of a recurrent neural network and then translates it into multivariate count predictions with a “sums and shares” distribution, well suited to tackle multivariate overdispersed and correlated count data. An extensive benchmark of the proposed model is carried out. We compare this model with seven others from the state-of-the-art probabilistic forecasting models using five open-source data (bikes, taxis, railways, traffic, wikipedia) and a specific use case on pedestrian counts within a multimodal transport hub in the Paris Region. Our model outperforms other models in situations where the data present temporal regularities. The results also highlight the potential of our model in the specific use case. Moreover, this forecasting represents an interesting way to predict short-term pedestrian counts in response to different events, such as concerts or transport disruptions.

Understanding driver-vehicle interactions remains a challenge, particularly in the case of cornering. This is particularly the case for powered two-wheeler vehicle (PTWs) users, perhaps because PTW drivers play a greater role in controlling the stability of their vehicles than do four-wheeled vehicle drivers. This difficulty stems from the variety of practices of this population of road users when entering, controlling their path, and exiting turns. Thus, observing the evolution of rider behavior during a cornering maneuver is an essential step in identifying road environment features that are risk factors for this category of road users. The real data set used for the experiments reported here was collected in the framework of the VIROLO++ collaborative project to improve the knowledge of the real practices of PTWs drivers, particularly during cornering. The in-depth analysis of these data in order to better understand motorcyclists’ behavior can therefore be considered a challenge. For this purpose, a two-step methodology was applied: (1) a data segmentation and feature extraction step in which the multidimensional time series of roll angle and roll velocity data were segmented using a multiple regression hidden logistic process (MRHLP), and (2) a clustering step in which the two detected segments characterizing the curve entry were assigned to different clusters regarding the curve initiation and the control actions set up during the different phases of the turning maneuver based on the hierarchical clustering algorithm. The results obtained show the effectiveness of the proposed methodology.

Clustering is used in many applicative fields to summarize information into a small number of groups.Motivated by behavioral extraction issues from urban data, the interest of this paper is to propose a classification method that allows modeling the evolution of cluster profiles over time while considering common regressive effects.The parameters of the proposed model are estimated using variational approximation because maximum likelihood estimation is not suitable in this case.The ability of the model to estimate parameters is evaluated using various simulated data and compared with two other models.

The last few decades have seen a faster development of digital systems for observing the mobility of people and goods.Various sensing systems -such as radio communication, Wi-Fi, Bluetooth, validation of smart cards, mobile phone, and road traffic monitoring systems -have enabled researchers and practitioners to acquire large amounts of data, which generally refer to individual and collective trajectories.The mobility data can be further enriched with side information, such as text corpora from social media, survey data, and weather information.These massive data, temporally and spatially structured, can benefit from advanced machine learning and data mining methods, providing decision aid tools, and contributing to the development of safer, cleaner, and more efficient transportation systems.They can also help to implement new mobility services for the user.This article provides an overview of methodological advances in temporal and spatial mobility data processing. Structural time series decomposition applied to mobility dataConsidering mobility data as time series, structural models [1] have been used to extract from them multiple latent components, each representing an aspect of 453

Personal trajectory data are becoming more and more accessible and have a high value in transport planning and mobility characterisation, at the cost of a risk for user's privacy.Addressing this risk is usually computationally expensive and can lead to losing most of the data utility.We explore a new, light-weight approach to Origin/Destinationmatrix anonymization that is easily scalable.We apply it to trip records from New York City Taxi and Limousine Commission (TLC) to illustrate how it can combine foolproof anonymity with a good spatial precision for a reasonable computational cost.

The emergence of smart meters has fostered the collection of massive data that support a better understanding of consumer behaviors and better management of water resources and networks. The main focus of this article is to analyze consumption behavior over time; thus, we first identify the main weekly consumption patterns. This approach allows each meter to be represented by a categorical series, where each category corresponds to a weekly consumption behavior. By considering the resulting consumption behavior sequences, we propose a new methodology based on a mixture of nonhomogeneous Markov models to cluster these categorical time series. Using this method, the meters are described by the Markovian dynamics of their cluster. The latent variable that controls cluster membership is estimated alongside the parameters of the Markov model using a novel classification expectation maximization algorithm. A specific entropy measure is formulated to evaluate the quality of the estimated partition by considering the joint Markovian dynamics. The proposed clustering model can also be used to predict future consumption behaviors within each cluster. Numerical experiments using real water consumption data provided by a water utility in France and gathered over 19 months are conducted to evaluate the performance of the proposed approach in terms of both clustering and prediction. The results demonstrate the effectiveness of the proposed method.

In the domain of water, companies are deploying innovative infrastructures to instrument buildings with smart meters that collect in particular consumption readings with a fine granularity. Advanced analytics are investigated on these massive and complex data to address different objectives e.g., water consumption forecasting, water leak detection and consumption behavior analysis. This paper is an extension of a previous study aiming to model the dynamics of water consumption time series issued from smart meters in order to predict the future consumption behaviors. We propose a mixture of non-homogeneous hidden Markov models (MixNHMM) to cluster the consumption behavior series in appropriate groups, which share the same transition dynamics, and to forecast the future behaviors in each group of consumers separately. The proposed methodology is applied on a real data gathered during one year and provided by a water utility in France. The experimentation results demonstrate the effectiveness of the proposed approach for the forecasting task. In addition, the clustering capability of the proposed method is confirmed using numeric metrics and visualization.

Posture analysis in quiet standing is an essential element in evaluating human balance control. Many factors enhance the human control system's ability to maintain stability, such as the visual system and base of support (feet) placement. In contrast, many neural pathologies, such as Parkinson's disease (PD) and cerebellar disorder, disturb human stability. This paper addresses the problem of the automatic segmentation of stabilometric signals recorded under four different conditions related to vision and foot position. This is achieved for both control subjects and PD subjects. A hidden Markov model (HMM)regression-based approach is used to carry out the segmentation between the different conditions using simple and multiple regression processes. Twenty-eight control subjects and thirty-two PD subjects participated in this study. They were asked to stand upright while recording stabilometric signals in mediolateral and anteroposterior directions under two permutations: feet apart and together with eyes open or closed. The results show high values for the correct segmentation rates, up to 98%, for the separation between the different conditions. The present findings could help clinicians better understand the motor strategies used by the patients during their orthostatic postures and may guide the rehabilitation process. The proposed method compares favorably with standard segmentation approaches.

Driving errors are considered to be the greatest contributory cause in all road accidents and an important contributory cause of most fatal accidents. This is particularly the case for the users of powered two-wheeled vehicles (PTWs), perhaps because PTW riders play a greater role in the control of their vehicles' stability than four-wheeled vehicle drivers. Thus, observing and analyzing the evolution of riders' behavior in a real-life context is an important step in the identification of the road environment characteristics that constitute a risk factor for PTW riders. A relevant research issue in naturalistic studies is related to the detection and identification of critical riding events from among the vast amount of data recorded during the experiment. In this paper, two approaches were used to automatically detect such critical riding events. First, we formalized this problem in terms of detecting changes in the mean and variance of the signals generated by the acceleration and angular velocity sensors. For this purpose, two steps were performed: (1) a data segmentation and feature extraction step in which the multidimensional time series of accelerometer and angular velocity data were segmented and modeled using a Gaussian mixture model with quadratic logistic proportions; and (2) a classification step in which each detected segment was assigned to the appropriate riding sequence, whether “naturalistic” or “critical” (i.e., fall or near fall), using the k-nearest neighbor algorithm. The second approach was based on online fall detection. This methodology used control charts (a multivariate cumulative sum), an approach that has been traditionally employed for sequential detection. These two algorithms were applied to a database composed of data from a real-life driving experiment. The obtained results show the effectiveness of both methodologies.

The analysis of time series data issued from smart meters has been studied relatively extensively in the electricity domain. Meanwhile, analysis of medium resolution water consumption data collected via smart meters has become possible recently, and the research has tried to develop statistical and machine learning tools in order to respond to different requirements of the domain, e.g., better understanding of water consumption behaviors and prediction of consumption. In the present paper, we propose a new predictive approach based on Non-homogeneous Markov Models in order to learn the dynamics of water consumption behavior and be able to predict future consumption behaviors with a daily time-step based on different relevant exogenous covariates. The data used for this purpose are categorical time series, where each series corresponds to a smart meter and each category corresponds to a specific daily consumption behavior. The experiments are performed on a real data set provided by a water utility in France. Prediction results obtained with the proposed model are compared to those provided by two models, namely, the state independent model and homogeneous Markov model. This predictive classification can be helpful for water utilities in order to better manage the water resources and respond to consumer requirements.

Smart card data gathered by automated fare collection (AFC) systems are valuable resources for studying urban mobility. In this paper, we propose two approaches to cluster smart card data, which can be used to extract mobility patterns in a public transportation system. Two complementary standpoints are considered: a station-oriented operational point of view and a passenger-focused one. The first approach clusters stations based on when their activity occurs, i.e., how trips made at the stations are distributed over time. The second approach makes it possible to identify groups of passengers that have similar boarding times aggregated into weekly profiles. By applying our approaches to a real data set issued from the metropolitan area of Rennes, France, we illustrate how they can help reveal valuable insights about urban mobility, such as the presence of different station key roles, including residential stations used mostly in the mornings and work stations used only in the evening and almost exclusively during weekdays, as well as different passenger behaviors ranging from the sporadic and diffuse usage to typical commute practices. By cross comparing passenger clusters with fare types, we also highlight how certain usages are more specific to particular types of passengers.

In recent years, many research studies are conducted into the use of smart meters data for developping decision-making tools including both analytical, forecasting and display purposes. Forecasting energy generation or forecasting energy consumption demand are indeed central problems for urban stakeholders (electricity companies and urban planners). These issues are helpful to allow them ensuring an efficient planning and optimization of energy resources. This paper investigates the problem for forecasting the hourly solar irradiance within a Machine Learning (ML) framework using Similarity method (SIM), Support Vector Machine (SVM) and Neural Network (NN). These approaches rely on a methodology which takes into account the previous hours of the predicting day and also the days having the same number of sunshine hours in the history. The study is conducted on a real data set collected on the Paris suburb of Alfortville. A comparison with two time series approaches namely Naive method and Autoregressive Moving Average Model (ARMA) is performed. This study is the first step towards the development of the hourly solar irradiance forecasting hybrid models.

Smartcard data provide a great number of information that are increasingly used nowadays. In the field of transport, they offer the opportunity to study passenger behavior, leading to a better knowledge of public transit demand and thereby granting the transport operators the ability to adapt their transport offer and services accordingly, both in space and in time. In particular, an accurate characterization of mobility patterns using data mining approches has a very strong interest for transport planning purposes. This paper aims to propose a two-level generative mixture model that partitions passengers according to their temporal profiles. Using the timestamps of the passengers' transactions in the public transport network, the first level models the passengers partitioning into a reduced set of clusters, whereas the second level captures how the trips made by each cluster of passengers are distributed over time. The proposed approach is applied on real ticketing data collected from the urban transport network of Rennes Métropole (France). The obtained results show that different passenger profiles can be discovered, thus highlighting several patterns of transport demand. The crossing of the clustering results with smartcard fare types as well as city characteristics such as academic centralities is also conducted in order to identify the close link between urban mobility and the socio-economic characteristics of the city.

This paper presents clustering approaches applied on daily energy consumption curves of a building. Our aim is to identify a reduced set of consumption patterns for a tertiary building during one year. These patterns depend on the temperature throughout the year as well as the type of the day (working day, work-free day and school holidays). Two clustering approaches are used independently, namely the functional K- means algorithm, that takes into account the functional aspect of data and the Expectation-Maximization algorithm based on Gaussian Mixture Model (EM-GMM). The clustering results of the two algorithms are analyzed and compared. This study represents the first step towards the development of prediction models for energy consumption.

In this paper, a machine-learning framework is used for riding pattern recognition. The problem is formulated as a classification task to identify the class of riding patterns using data collected from 3-D accelerometer/gyroscope sensors mounted on motorcycles. These measurements constitute an experimental database used to analyze powered two-wheeler rider behavior. Several well-known machine-learning techniques are investigated, including the Gaussian mixture models, the k-nearest neighbor model, the support vector machines, the random forests, and the hidden Markov models (HMMs), for both discrete and continuous cases. Additionally, an approach for sensor selection is proposed to identify the significant measurements for improved riding pattern recognition. The experimental study, performed on a real data set, shows the effectiveness of the proposed methodology and the effectiveness of the HMM approach in riding pattern recognition. These results encourage the development of these methodologies in the context of naturalistic riding studies.

The emergence of new technologies allows better monitoring of traffic conditions and understanding of urban network dynamics. Bluetooth technology is becoming widespread, as it represents a cost-effective means for capturing road traffic in both arterials and motorways. Although the extraction of travel time from Bluetooth data is fairly straightforward, data reliability and processing is still challenging with the issues of penetration rate, mode discrimination, and detection quality. This paper presents a methodological contribution to the use of Bluetooth data for the spatiotemporal analysis of a large urban network (Brisbane, Australia). It introduces the concept of the Bluetooth origin-destination (B-OD) matrix, which is built from a network of 79 Bluetooth detectors located within the Brisbane urban area. The B-OD matrix describes the dynamics of a subpopulation of vehicles, between pairs of detectors. The results show that the characteristics of urban networks can be effectively represented through B-OD matrices. A comparison with loop detector data enables an assessment of the results' significance. Then, the spatiotemporal structure of the network is analyzed with two different clustering analyses, namely, latent Dirichlet allocation (LDA) and $K$-means. While LDA is used to detect a temporal pattern, the $K$-means algorithm highlights Bluetooth fundamental diagram (BFD) classes. The results show that Bluetooth data has the potential to be a reliable data source for traffic monitoring. By highlighting hidden structures of a large area, the algorithm outputs allow us to provide the road operators with a fine spatiotemporal analysis of their network, in terms of traffic conditions.

Today, more and more bicycle sharing systems (BSSs) are being introduced in big cities. These transportation systems generate sizable transportation data, the mining of which can reveal the underlying urban phenomenon linked to city dynamics. This article presents a statistical model to automatically analyze the trip data of a bike sharing system. The proposed solution partitions (i.e., clusters) the stations according to their usage profiles. To do so, count series describing the stations’s usage through departure/arrival counts per hour throughout the day are built and analyzed. The model for processing these count series is based on Poisson mixtures and introduces a station scaling factor that handles the differences between the stations’s global usage. Differences between weekday and weekend usage are also taken into account. This model identifies the latent factors that shape the geography of trips, and the results may thus offer insights into the relationships between station neighborhood type (its amenities, its demographics, etc.) and the generated mobility patterns. In other words, the proposed method brings to light the different functions in different areas that induce specific patterns in BSS data. These potentials are demonstrated through an in-depth analysis of the results obtained on the Paris Vélib’ large-scale bike sharing system.

Using supervised machine learning approaches to recognize human activities from on-body wearable accelerometers generally requires a large amount of labeled data. When ground truth information is not available, too expensive, time consuming or difficult to collect, one has to rely on unsupervised approaches. This paper presents a new unsupervised approach for human activity recognition from raw acceleration data measured using inertial wearable sensors. The proposed method is based upon joint segmentation of multidimensional time series using a Hidden Markov Model (HMM) in a multiple regression context. The model is learned in an unsupervised framework using the Expectation-Maximization (EM) algorithm where no activity labels are needed. The proposed method takes into account the sequential appearance of the data. It is therefore adapted for the temporal acceleration data to accurately detect the activities. It allows both segmentation and classification of the human activities. Experimental results are provided to demonstrate the efficiency of the proposed approach with respect to standard supervised and unsupervised classification approaches.

Unsupervised learning approaches are used in various applications such as speech recognition, image compression, information retrieval and activity recognition. This paper introduces a novel unsupervised approach for clustering multi-dimensional time series that present the 3-d acceleration data measured with body-worn accelerometers. More specifically, the proposed approach uses a statistical model based on Multiple Hidden Markov Model Regression (MHMMR) to automatically analyze the human activity. This method takes into account the sequential appearance and temporal evolution of the data to easily detect static and dynamic activities. Comparisons with existing unsupervised approaches, including the standard Gaussian Mixture Model, the k-means algorithm, the DBSCAN algorithm and the standard HMM, demonstrate the effectiveness of the proposed approach.

Deployment is a fundamental issue in wireless sensor networks. Usually the sensor locations are precomputed based on a ”perfect” sensor coverage model, whereas sensors may not always provide reliable information, either due to operational tolerance levels or environmental factors. Therefore, it is imperative to have practical considerations at the design stage to anticipate this sensing behavior. In this paper, we address four different forms of static wireless sensor networks deployment while considering an evidence-based sensor coverage model. The four problems are formalized as combinatorial optimization problems, which are NP-complete. We propose, E2BDA (Efficient Evidence-Based sensor Deployment Algorithm), a polynomial-time uncertainty-aware deployment algorithm based on a dynamic programming approach. E2BDA is able to determine the minimum number of sensors and their locations to achieve both coverage and connectivity. We compare our proposal to the state-of-the-art deployment strategies, the obtained results show that E2BDA obtains the best performances.

In this paper, we address the issue of handling uncertainty and information fusion for an efficient WSN deployment. We present a flexible framework for collaborative target detection within the transferable belief model. Using the developed framework, we propose an uncertainty-aware deployment algorithm that is able to determine the minimum number of sensors and their locations such that full area coverage is achieved. The issues of connectivity, obstacles, preferential coverage, challenging environments and sensor reliability are also discussed. Experimental results are provided to demonstrate the ability of our approach to achieve an efficient sensor deployment by exploiting a collaborative target detection scheme.

This paper describes a pattern recognition approach aiming to estimate fuel cell duration time from electrochemical impedance spectroscopy measurements. It consists in first extracting features from both real and imaginary parts of the impedance spectrum. A parametric model is considered in the case of the real part, whereas regression model with latent variables is used in the latter case. Then, a linear regression model using different subsets of extracted features is used for the estimation of fuel cell time duration. The performances of the proposed approach are evaluated on experimental data set to show its feasibility. This could lead to interesting perspectives for predictive maintenance policy of fuel cell.

Independent Factor Analysis (IFA) is a well known method used to recover independent components from their linear observed mixtures without any knowledge on the mixing process. Such recovery is possible thanks to the hypothesis that the components are mutually independent and non-Gaussians. The IFA model assumes furthermore that each component is distributed according to a mixture of Gaussians. This article investigates the possibility of incorporating prior knowledge on the mixing process and partial knowledge on the cluster belonging of some samples to estimate the IFA model. In this way, other learning contexts can be handled such as semi-supervised or partially supervised learning. Such information is valuable to enhance estimation accuracy and remove indeterminacy commonly encountered in unsupervised IFA such as the permutation of the sources. The proposed method is illustrated by a railway device diagnosis application and results are provided to show its effectiveness for this type of problem.

This paper addresses the problem of fault detection in a complex system made up of several spatially dependent subsystems. The diagnosis method consists of both detecting and localizing a defect on the system by combining the outputs scores of subclassifiers within the framework of belief function theory. This paper is focused on the coding and the combination of classifier outputs that can reflect the spatial relationship between the subsystems. In the particular case of upstream/downstream dependency, two strategies of output coding are detailed. The proposed methodology is illustrated on a railway device diagnosis application. It will be shown that the choice of an appropriate coding scheme improves the classification results

This paper introduces a diagnosis scheme of a railway infrastructure component based on a combined use of empirical mode decomposition (EMD) and Hilbert transform. This component is dedicated to track/vehicle transmission referred as track circuit. The aim is to detect its working state from one measurement signal which can be viewed as a superposition of several oscillations and periodic patterns called intrinsic mode functions (IMFs). For this application, it will be shown that physical meaning can be assigned to each mode that EMD tries to extract. Furthermore, when the Hubert transform of the IMFs is performed, we show that the changing of instantaneous frequency can be linked to the existence of defect. The performances are illustrated on both simulated and experimental signals