Benchmark Data and MATLAB Code for Reliability-Aware Learned Residual Model Predictive Control of Permanent-Magnet Synchronous Motor Drives under Distribution Shift
Year: 2026
Type: dataset
Abstract: This repository provides the reproducibility package and benchmark data supporting the study “Reliability-Aware Learned Residual Model Predictive Control for Permanent-Magnet Synchronous Motor Drives under Distribution Shift.” The study investigates selective trust between a strong causal one-step residual-persistence predictor and a learned five-member episode-bootstrap neural residual ensemble within a physics-based continuous-control-set model predictive control framework for permanent-magnet synchronous motor drives.The archive contains MATLAB R2020b source scripts, benchmark definitions, nominal motor and controller parameters, controlled distribution-shift definitions, fixed random seeds, tabulated benchmark results, and documentation required to reproduce the principal analyses reported in the associated article. The evaluation covers Final-ID, Near-OOD, Far-OOD, and deterministic Corner-OOD conditions. The final untouched benchmark comprises 64 scenarios evaluated with four controller configurations, corresponding to 256 closed-loop simulation runs.The final benchmark uses parameterSeed = 910247 and operatingSeed = 731059. The reliability-aware architecture is frozen before this independently seeded benchmark. Additional post-freeze analyses include selector feature-group ablation, q-axis current-reference-limit sensitivity, measurement-noise stress testing, causal residual filtering, controlled voltage-actuation mismatch, combined stress testing, computational timing analysis, and optimized neural inference.The parameter variations used in the benchmark represent controlled distribution shifts for robustness evaluation and should not be interpreted as manufacturing tolerances or fault statistics. The 7.7-A current quantity used in the study is a q-axis current-reference clamp rather than a hard MPC current-state constraint.Large intermediate MAT files and generated neural-network training objects are not included in this compact archive. The supplied MATLAB scripts define the generation of the required intermediate data and learned models.Software environment: MATLAB R2020b. MATLAB source code is distributed under the BSD 3-Clause License, while benchmark data, configuration files, tabulated results, and documentation are provided under CC BY 4.0.
Author: KARABACAK, YUSUF
Institution: National Defence University
Language: English
Subfield: Control and Systems Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7220469474