Machine Learning-based Energy Optimisation in Smart City Internet of Things

Dataset for the paper Machine Learning-based Energy Optimisation in Smart City Internet of Things accepted for publication at The First International Workshop on the Integration between Distributed Machine Learning and the Internet of Things, ACM MobiHoc 2023.

The dataset is collected from a real-world deployment of environmental sensors in the city of Bern, Switzerland. Our proposed approach can be applied to determine the tradeoff between the accuracy of temperature measurements and reducing the energy consumption for a single sensor; hence, without loss of generality, the evaluation is conducted on a dataset from a single sensor. Overall, we acquired 3697 measurements, each long 138 seconds. To correct the measurements, we set the maximum ventilation duration of 138 seconds, during which the multivariate time series of humidity and temperature sensor values are recorded together with their corresponding timestamps. The sensor values are recorded at a fixed frequency.

From this raw data, we created the training and test sets through data augmentation to simulate time series of different lengths. Namely, for each measurement, we generated 136 samples with the increasing length of measurement time-series, padding the residual time-series length with zeros until reaching a time-series length of 137.

We released the source code and trained models on the following GitHub repository https://www.github.com/ricsamikwa/ml-iot-smartcitytemp