Calibration automatique d'un réseau de capteurs sans fil à effet Hall mesurant la consommation énergétique résidentielle

Abstract

A wireless sensor network monitoring the residential energy consumption of each breaker in the main distribution panel of a house is a useful tool, but such a system can prove to be expensive. In order to reduce the cost of this product, a network of Hall effect sensors is proposed instead of the traditional current transformers. A system using many Hall effect sensors is more affordable and easier to install due to its miniaturization potential. On the other hand, the accuracy of Hall effect sensors is not the same as current transformers because the sensitivity is dependent on the distance between the sensor and the conductor that is measured, which causes unknown variations on the gain of each sensor. This thesis aims at enhancing the accuracy of the measures taken by the Hall effect sensors by proposing two numerical algorithms in order to calibrate the sensors. The first algorithm is the least mean square (LMS) algorithm that is simple and able to calibrate the sensors when no crosstalk is present while the second is fast independent component analysis (FastICA) which is a more complex algorithm, but able to attenuate the crosstalk problem when it is present. Following the presentation of the system and of the theory linked to the algorithms, the simulation and experimental results are presented for both algorithms. For the simulation results with LMS, the mean of the error for the measured current for each sensor for 3 to 30 sensors without crosstalk and with noise is 0.53 ARMS. For the combination of FastICA and LMS, the mean of the error for the measured current with crosstalk and noise is 0.55 ARMS for 3 sensors going up to 0.83 ARMS for 30 sensors. Two types of installations are tested for the experimental results. The LMS algorithm for the tests without crosstalk is able to reduce the mean of the error by sensor for 3 sensors to 0.47 ARMS for the first installation (coffee maker, heat gun and air purifier) and to 0.49 ARMS for the second installation (coffee maker, heat gun and electric baseboards controlled by an electronic thermostat). Due to the Noisy measures of one of the appliance used for the first installation, the combination of FastICA and LMS for the tests with crosstalk has only reduced the mean of the error by sensor for 3 sensors to 3.16 ARMS. Nevertheless, the error is reduced to 0.85 ARMS for the second installation.

Date	11 Nov 2014
Original language	French
Awarding Institution	École de technologie supérieure
Supervisor	Ghyslain Gagnon (Supervisor) & François Gagnon (Co-supervisor)