2022—2023 — Knowledge and Data
Some of the current water distribution infrastructure in the Netherlands is covered with a decent number of sensors to measure the real-time state of the network. Most water network companies use physics-based water network simulations to predict the water network conditions. In order to be accurate, these simulations require careful parameter tuning to ensure conditions that correspond to the real world. The student’s task in this master’s project would be to use supervised machine learning methods to construct a model that predicts the parameters of the physical simulation system, based on the specific real-time sensor values.
Some of the current water distribution infrastructure in the Netherlands is covered with a decent number of sensors to measure the real-time state of the network. Historical dataset of water network conditions can provide us with many insights into how water distribution networks are being used, their supply/demand patterns, seasonal variations, etc. The goal of this project is to use unsupervised methods to create an algorithm for profiling and pattern extraction from historical sensor data of the network. The dataset contains real data, collected from the Dutch water distribution network over several recent years.
Growing network of charging stations for electric vehicles in the Netherlands produces considerable datasets that allow us to analyse charging patterns of the population. The student’s task in this master’s project would be to analyse the dataset of charging sessions in the Netherlands, reconstruct charging demand patterns, and use either supervised methods for changing demand predictions or unsupervised methods for profiling of different usage types. The real charging data is provided by Shell’s EV charging network, and the project will be done in collaboration with Shell’s data scientists.
Systems such as smart homes usually depend heavily on the existence of sufficient logical rules that describe the expected system’s behavior. These rules are largely designed manually (e.g. IFTTT rules), but this requires considerable effort and prevents further mass acceptance of such systems. In this project, we would like to investigate and apply rule learning algorithms, such as frequent itemsets, to extract behavior patterns in a form of logical rules from an activity dataset with raw sensor readings.