Through this project, I conducted a systematic exploration of linear model structures including FIR, ARX, OE, ARMAX, Box–Jenkins, instrumental-variable methods, and stochastic state-space models. Model orders were increased conservatively while prioritizing compactness and numerical robustness.

Residual–input cross-correlation tests were used as the primary validation metric. While several models significantly reduced correlation, a small persistent structure remained, highlighting practical limitations caused by disturbance effects and unmodeled dynamics. The final selected model represents the best achievable balance between accuracy, simplicity, and validation performance.