The project investigates state estimation techniques for a nonlinear quadrotor model, serving as a benchmark system for evaluating control and estimation algorithms. The project involved:

• Deriving nonlinear and linearised state-space models of quadrotor dynamics

• Implemented and compared three estimators

  • Kalman Filter

  • Moving Horizon Estimator

  • Unscented Kalman Filter

• Assessed the accuracy and computational efficiency in estimating the quadrotor states in hover condition.

This project explores advanced deep learning methods for financial time series forecasting, focusing on the challenges of market complexity and "concept drift." 

• Implemented and compared LSTMs, GRUs, Transformers, and a Hybrid (LSTM-GRU) model.

• Used continual learning strategies (like EWC) to update models with new market data without "catastrophic forgetting" of past information.

• Continual learning approach consistently outperformed standard retraining methods when tested on market data

This project analyzes power grid stability by modeling grid components (generators and loads) as coupled nonlinear oscillators (Kuramoto Oscillators).

• We applied a second-order Kuramoto model to a simplified three-node power grid (two generators and one machine) to simulate its dynamics.

• The study tested the grid's robustness by introducing sudden spikes in power demand (perturbations) at the machine node and demonstrated critical threshold.

• Below this threshold, the system is resilient and recovers synchronization after a disturbance and above this critical threshold, the perturbation causes the system to become unstable and incoherent, leading to a loss of synchronization and modeling a grid failure.