Bike sharing systems offer a convenient, ecologic, and economic transport mode that has been increasingly adopted. However, the distribution of bikes is often unbalanced, which decreases user satisfaction and potential revenues. Moreover, bike sharing literature is mostly focused on the prediction of demand on large scale systems and uses simulations for the assessment of relocation operations to increase the number of utilizations. We propose prediction models based on machine learning approaches to improve the bike sharing re-balancing in a small city of Portugal. The algorithm aims to improve three metrics, namely (1) increase the number of utilizations, (2) reduce the number of stations without bikes, (3) reduce the time without available bikes in the stations. The relocation operations are validated using real data. Our findings show that (a) the estimated number of utilizations created by this system is substantially higher than the current system by 223%, (b) our model allows the correct identification of more 70%, 165%, 249% empty stations with the same or substantially higher precision than the existing approach, (c) the total time of bike unavailability reduced by the predictive model is 283% higher than the time reduced by current approach (1,394,454 vs 363,971 minutes).

The design of the communication protocols in wireless sensor networks (WSNs) often neglects several key characteristics of the sensor's hardware, while assuming that the number of transmitted bits is the dominating factor behind the system's energy consumption. A closer look at the hardware specifications of common sensors reveals, however, that other equally important culprits exist, such as the reception and processing energy. Hence, there is a need for a more complete hardware abstraction of a sensor node to reduce effectively the total energy consumption of the network by designing energy-efficient protocols that use such an abstraction, as well as mechanisms to optimize a communication protocol in terms of energy consumption. The problem is modeled for different feedback-based techniques, where sensors are connected to a base station, either directly or through relays. We show that for four example platforms, the use of relays may decrease up to 4.5 times the total energy consumption when the protocol and the hardware are carefully matched. We conclude that: 1) the energy budget for a communication protocol varies significantly on different sensor platforms; and 2) the protocols can be judiciously adapted to the underlying hardware. The results are cross-validated using real-life measurements.

A fundamental understanding of the delay behavior of network coding is key towards its successful application in real-time applications with strict message deadlines. Previous contributions focused mostly on the average decoding delay, which although useful in various scenarios of interest is not sufficient for providing worst-case delay guarantees. To overcome this challenge, we investigate the entire delay distribution of random linear network coding for any field size and arbitrary number of encoded symbols (or generation size). By introducing a Markov chain model we are able to obtain a complete solution for the erasure broadcast channel with two receivers. A comparison with Automatic Repeat reQuest (ARQ) with perfect feedback, round robin scheduling and a class of fountain codes reveals that network coding on GF(24) offers the best delay performance for two receivers. We also conclude that GF(2) induces a heavy tail in the delay distribution, which implies that network coding based on XOR operations although simple to implement bears a relevant cost in terms of worst-case delay. For the case of three receivers, which is mathematically challenging, we propose a brute-force methodology that gives the delay distribution of network coding for small generations and field size up to GF(24).