Profile

Project Highlights

Designed and implemented a Kalman filter-based disturbance observer, applying Bayesian estimation techniques for real-time state and disturbance tracking. Applied Kalman filtering techniques for dynamic state estimation in noisy environments, analogous to time-series forecasting and anomaly detection in AI.
Developed model-based predictive controllers (MPC) to optimize dynamic decision-making under uncertainty, conceptually aligned with reinforcement learning. Developed and implemented Model Predictive Control (MPC) algorithms for real-time optimization of microgrid operations using forecast and system state data.
Applied state-space modeling and system identification to build simplified yet physically realistic models for control and forecasting.
Validated controller performance through data-driven simulations and sensitivity analysis, akin to ML model evaluation and stress-testing.
Integrated stochastic estimation and optimization within constrained control frameworks
Integrated heterogeneous data sources through the MOSAIK co-simulation platform, enabling real-time coordination across hierarchical control layers.
Gained practical experience in AI-driven control systems, mathematical modeling, and real-time systems integration, with strong relevance to AI applications in energy, robotics, and industrial automation.

Applied in the context of smart energy systems to estimate prosumer response dynamics

Developed a data-driven modeling framework combining SINDyc (Sparse Identification of Nonlinear Dynamics with Control) and Bayesian inference (MCMC).
Applied machine learning techniques to identify sparse nonlinear dynamic models from observed data under uncertainty.
Implemented probabilistic modeling using Stan and the No-U-Turn Sampler (NUTS) to efficiently perform posterior inference.
Quantified uncertainty in model parameters, improving robustness and enabling application in stochastic model predictive control.
Gained hands-on experience in ML for dynamical systems, Bayesian statistics, and real-world system modeling constraints.

Developed a modular, object-oriented energy system modeling framework with support for hierarchical composition and optimization.
Applied AI/ML-adjacent techniques including deterministic optimization modeling (Pyomo) and time-series data integration (Pandas).
Facilitated scalable energy network representations with extensibility toward AI-driven decision-making and ML-based forecasting.
Utilized tools such as Python, Pyomo, Pandas, and GraphViz for modeling and visualization.

Developed Python-based data pipelines for high-frequency sensor data preprocessing, including cleaning, resampling, and feature engineering.
Applied statistical modeling, regression analysis, and clustering techniques to evaluate performance and detect system patterns.
Utilized kernel density estimation (KDE), quantile-based correlations, and custom performance metrics to uncover optimization opportunities in a real-world CO2 refrigeration system.
Modeled and compared control strategies and system configurations using predictive analysis and scenario-based simulation methods.

Designed and implemented a rule-based control algorithm for dynamic CO2-driven ventilation, integrating sensor data with Excel-linked control logic in TRNSYS.
Simulated and analyzed time-series energy data under varying occupancy schedules and climatic conditions, using data-driven modeling techniques.
Performed data collection, preprocessing, and regression analysis to estimate thermal system parameters (kA-values, power consumption), applying statistical curve fitting.
Applied core concepts of automation, control systems, and building simulation, building early expertise in cyber-physical systems and data-centric optimization.