Electricity Load Forecasting via State-Space Models
An electricity load forecasting system where each base model adapts continuously via a Kalman filter, with automatic variance tuning (VIKING) and robust online aggregation of 28 experts.
Stochastic modeling, time series forecasting and risk-constrained optimization, applied to energy and financial markets.
An electricity load forecasting system where each base model adapts continuously via a Kalman filter, with automatic variance tuning (VIKING) and robust online aggregation of 28 experts.
A segment-based forecasting architecture for a heterogeneous gas portfolio: temperature-sensitive correction for the profiled share, regime classification (k-means + random forest) for the industrial share with discontinuous behavior.
Pricing a wind power purchase agreement formulated as a stochastic linear program under a CVaR constraint, with the capture rate modeled via a Gaussian copula to capture the cannibalization effect between production and price.
How to make a Kalman filter self-adaptive by treating its variances as latent states, estimated via a mean-field variational approximation rather than tuned by hand.
Numerical implementation of the one-factor Hull-White model for zero-coupon bond pricing. PDE-based approach (Feynman-Kac) with implicit finite-difference discretization.
Study and implementation of bootstrapping and interpolation methods for yield curve construction. Critical analysis of cubic spline, monotone convex, and other approaches.
Study and numerical implementation of the augmented anti-symmetric formulation of Bresch et al. for the Shallow-Water equations with surface tension, via an implicit-explicit finite-volume scheme.