Shum, Christopher2026-05-282026-05-282026-05-282026-05-14https://hdl.handle.net/10012/23428Electrification of air transportation is an emerging technology with the potential to sig- nificantly reduce global greenhouse gas emissions. The Pipistrel Velis Electro, an all-electric fixed-wing light aircraft, is the first battery-electric aircraft to receive type certification in Canada. This technology represents an opportunity to eliminate aircraft exhaust emis- sions and reduce the cost of operation. However, similar to other electric vehicles, electric aircraft are subject to physically-limited energy density compared to liquid fuels, which makes energy-efficient operation critical to maximize their benefits. Toward this goal, the development of computational models based on system-specific performance can enable more precise operational planning. In this thesis, a series of models are developed that represent the aircraft’s subsystems that contribute to energy flow and consumption. First, a set of physics-based models was developed using established governing equations. Analogous black-box data-driven models consisting of feedforward neural networks are then trained to approximate physical relationships for aspects of the system. A neural network is also trained to approximate the residual error of the physics-based system model compared to the measured observations, then combined with the output of the physics-based model in a hybrid architecture to improve accuracy and compensate for system-specific and environment-specific deviation. The models are tuned and validated using real-world data collected from flight testing. The best-performing model is then used to estimate the energy consumption of the aircraft as a function of the change in state and exogenous variables. By mapping the state trajectory to a directed graph, graph-based optimization methods can be utilized. The projected energy consumption is mapped to the graph as edge costs, and a shortest- path algorithm is applied to find the minimum energy path with respect to the decision variables.enelectric vehiclesmachine learningoptimizationMaster Thesis