José A. Taboada

Virtual integration of sensor observation data

This paper discusses the design, implementation and evaluation of a framework that enables the virtual integration of heterogeneous observation data sources through a Sensor Observation Service (SOS) standard interface. Currently available SOS implementations follow a data warehouse design approach for data integration. Contrary to this, the present framework uses a well-known Mediator/Wrapper virtual data integration architecture, enabling the direct access to the current data supplied by the data sources. Currently, the framework is being validated as the OGC compliant technology to publish the meteorological and oceanographic observation data generated by two public agencies of the regional government of Galicia (Northwest of Spain). HighlightsVirtual observation data integration vs. prevailing data warehouse approaches.Flexible combination of Mediator/Wrapper architecture with OGC SWE interfaces.In situ and remote static and mobile sensors producing vector and raster data.Multi-thread implementation to leverage current hardware multicore architectures.Under validation in two Spanish public meteorological and oceanographic agencies.

Sensor Observation Service Semantic Mediation: Generic Wrappers for In-Situ and Remote Devices

In-situ and remote sensors produce data that fit different data modeling paradigms, namely, Entity/Relationship paradigm for the former and Multidimensional Array paradigm for the latter. Besides, different standardized data access services are used in practice. Therefore their integrated access is still a major challenge. This paper describes a solution for the development of generic semantic data access wrappers for observation datasets generated by in-situ and remote sensing devices. Those wrappers are key components of data mediation architectures designed for the semantic integrated publishing of observation data.

Heterogeneous sensor data integration for crowdsensing applications

This paper describes a conceptual solution for heterogeneous sensor data integration in crowdsensing applications and one experimental implementation for a health monitoring system in an educational environment using a low cost hardware solution. Three kinds of protocols are integrated in this solution: HL7 for medical data, Observations and Measurements model for environmental data and BACnet for buildings monitoring. This last protocol has the particularity that manages sensoring and acting. A Common Data Model is described for the integration of three kinds of data and protocols, and a validation test application is described.

SODA: A framework for spatial observation data analysis

Very large amounts of geospatial data are daily generated by many observation processes in different application domains. The amount of produced data is increasing due to the advances in the use of modern automatic sensing devices and also in the facilities available to promote crowdsourcing data collection initiatives. Spatial observation data includes both data of conventional entities and also samplings over multi-dimensional spaces. Existing observation data management solutions lack declarative specification of spatio-temporal analytics. On the other hand, current data management technologies miss observation data semantics and fail to integrate the management of entities and samplings in a single data modeling solution. The present paper presents the design of a framework that enables spatio-temporal declarative analysis over large warehouses of observation data. It integrates the management of entities and samplings within a simple data model based on the well known mathematical concept of function. Observation data semantics are incorporated into the model with appropriate metadata structures.

Comparative Study of Artificial Neural Network Models for Forecasting the Indoor Temperature in Smart Buildings

The implementation of efficient building energy management plans is key to the road-map of the European Union for reducing the effects of the climate change. Firstly, accurate models of the currently energy systems need to be developed. In particular, simulations of Heating, Ventilation and Air Conditioning (HVAC) systems are essential since they have a relevant impact in both energy consumption and building comfort. This paper presents a comparative of four different machine learning approaches, based on Artificial Neural Networks (ANNs), for modeling an HVAC system. The developed models have been tuned to forecast three consecutive hours of the indoor temperature of a public research building. Tests revealed that an on-line learning ANN, which is also fully trained weekly, is less affected by sensor noise and anomalies than the remaining approaches. Moreover, it can be also automatically adapted to deal with specific environmental conditions.