Table 1 Nomenclature.
Symbol | Description |
|---|---|
Problem formulation | |
\({\varvec{X}}=\{{{\varvec{x}}}_{1},...,{{\varvec{x}}}_{{\varvec{d}}}\}\) | Complete feature set from SCADA sensors |
\({\varvec{S}}\subseteq {\varvec{X}}\) | Selected feature subset after optimization |
\({\varvec{z}}\in \{0,1{\}}^{{\varvec{d}}}\) | Binary selection vector (\({z}_{i}=1\) if feature included) |
\({\varvec{Y}}\) | Target variable (active power or rotor speed) |
\({{\varvec{n}}}_{{\varvec{m}}{\varvec{i}}{\varvec{n}}},{{\varvec{n}}}_{{\varvec{m}}{\varvec{a}}{\varvec{x}}}\) | Feature count constraints (minimum for stability, maximum for computational feasibility) |
Objective functions | |
\({{\varvec{f}}}_{1}({\varvec{S}})\) | Validation loss (prediction accuracy) |
\({{\varvec{f}}}_{2}({\varvec{S}})\) | Normalized subset size $( |
\({\varvec{L}}({\varvec{M}},{\varvec{S}})\) | Cross-validation MSE of model trained on subset \(S\) |
Mutual information | |
\({\varvec{M}}{\varvec{I}}({{\varvec{X}}}_{{\varvec{i}}},{\varvec{Y}})\) | Mutual information between feature \({X}_{i}\) and target |
\({\varvec{H}}(\cdot )\) | Shannon entropy |
\({\varvec{\beta}}\) | Retention ratio (proportion of top-ranked features retained) |
\({\varvec{\theta}}\) | MI filtering threshold |
\({\varvec{h}}\) | Silverman’s bandwidth parameter |
AMO-BHS parameters | |
HMCR \(\left({\varvec{t}}\right)\) | Harmony memory considering rate (exploration–exploitation balance) |
PAR \(\left({\varvec{t}}\right)\) | Pitch adjustment rate (local search intensity) |
\({\boldsymbol{\alpha }}_{{\varvec{H}}{\varvec{M}}{\varvec{C}}{\varvec{R}}}\) | Power-law exponent for HMCR adaptation (Eq. 23) |
\({\boldsymbol{\alpha }}_{{\varvec{P}}{\varvec{A}}{\varvec{R}}}\) | Exponential decay coefficient for PAR adaptation (Eq. 24) |
\({\varvec{H}}{\varvec{M}}\) | Harmony memory (current solution population) |
\({\varvec{E}}\) | Elite archive (non-dominated solutions) |
\({\mathcal{P}}^{\boldsymbol{*}}\) | Final Pareto front |
