Unintentional falls can cause severe injuries and even death, especially if no immediate assistance is given. A fall detection system aims to detect a fall as soon as it occurs, therefore issuing an automatic assistance request. Wearable embedded sensors are emerging as the most viable solution for continuous monitoring since they are more effective, less intrusive and less expensive than other systems. Tailoring a deep learning method to the requirements of microcontrollers entails matching very stringent constraints in terms of both memory and computational power. In addition, datasets acquired with wearable devices are relatively scarce and not necessarily devised for supervised learning. In this work, we discuss the design of a software architecture based on recurrent neural networks which can be effective for fall detection while running entirely onboard a wearable device. The well-known and publicly-available SisFall dataset was adopted and extended with fine-grained temporal annotations to cope with the supervised training of recurrent neural networks. We then show that an appropriate process of architectural minimization together with accurate hyperparameters selection leads to a workable model which compares favorably with other detection techniques. The embedding of the resulting architecture has been validated using a state-of-art hardware device.

This paper presents an innovative approach to use real-time scheduling techniques for the automation of electric loads in Cyber-Physical Power Systems. The goal is to balance the electric power usage to achieve an optimized upper bound on the power peak load, while guaranteeing specific constraints on the physical process controlled by the electric loads. Timing parameters derived from the scheduling discipline of real-time computing systems are used to model electric devices. Real-time scheduling algorithms can be exploited to achieve the upper bound by predictably and timely switching on/off the devices composing the electrical system. The paper shows the relevance of electric load balancing in power systems to motivate the use of real-time techniques to achieve predictability of electric loads scheduling. Real-Time Physical Systems (RTPS) are introduced as a novel modeling methodology of a physical system based on real-time parameters. They enable the use of traditional real-time system models and scheduling algorithms, with adequate adaptations, to manage loads activation/deactivation. The model of the physical process considered in this work is characterized by uncertainties that are compensated by a suitable feedback control policy, based on the dynamic adaptation of real-time parameter values. A number of relevant relationships between real-time and physical parameters are derived.

This paper presents a method for the real-time management of electric vehicles (EVs) charging processes. The proposed method aims to limit the peak load and to increase the number of rechargeable EVs with respect to the scenario in which no coordination action is performed, while achieving given constraints on the power distribution system. The approach is based on a tight interaction between a scheduling algorithm and a power-flow evaluation procedure. The scheduling algorithm finds the best charging periods for each EV. The power flow procedure checks the achievement of electrical constraints and evaluates the operational parameters of the grid. Simulations are carried out on a real electricity distribution system of a medium-sized Italian city. The results show that the proposed approach increases the number of rechargeable EVs up to 33%. At the same time, the peak load is reduced by 25%. The scheduling algorithm requires an average of 50 ms to evaluate each charge request on an ordinary computer, therefore allowing its use in real-time conditions.