AI for Power Electronics — Algorithm Selector

Open the web app →
_{If the hosted link is unavailable, see Run locally.}

---

Authorship & status

• Web app author: Xinze Li
• Review article: Xinze Li, Fanfan Lin, Juan J. Rodríguez-Andina, Sergio Vazquez, Homer Alan Mantooth, Leopoldo García Franquelo, “Fundamentals of Artificial Intelligence for Power Electronics,” IEEE Transactions on Industrial Electronics, 2026.

This app is an additional material accompanying the IEEE TIE review article. It provides step-by-step guidance for AI algorithm selection, design, and tuning for Power Electronics scenarios, aligned with the review article’s What–Which–How framework. GitHub Jupyter Notebook Materials: Fundamentals_of_AI_for_PE.

---

About

This interactive wizard recommends AI/ML algorithms and provides tailored guidance using the review article’s What–Which–How framework:

• What (Sec. I–II): Identify your PE lifecycle phase (design / control / maintenance), specific task, and data modality (tabular, signal, field, graph, hybrid).
• Which (Sec. III–VI): Match the recommended AI model family (classic ML, NN, PIML, MHA, RL, agentic AI) to your task and data structure.
• How (Sec. III–VI): Apply PE-specific tuning practices — feature scaling, EDA, physics-informed constraints, and statistical comparison across runs.
• Case studies (Sec. VII): Reproduce the closest D1–D7 notebook from Fundamentals_of_AI_for_PE before adapting to your hardware or dataset.

Section numbers in reports follow the revised manuscript (Roman I–VIII), aligned with the repo README.

ChatGPT companion: For deeper Q&A and report-style help, a button below the wizard and report opens the Fundamentals of AI for PE Custom GPT in a new tab. Use the “Copy prompt” button to send your wizard selections and report directly to the assistant.

---

Run locally

The app uses ES modules; opening index.html as a file will not work — use a local HTTP server.

Windows: double-click start-server.bat (or run start-server.ps1), then open http://127.0.0.1:8765/.

Command line:

cd AI_for_PE_Algorithm_Selector
python -m http.server 8765

Then open http://127.0.0.1:8765/ in your browser.

If the port is in use, choose another (e.g. 8080) or use:

npx --yes serve -l 8765

Troubleshooting: confirm the terminal shows the server listening; try 127.0.0.1 instead of localhost; allow Python through the firewall if needed; check python --version.

---

Repository layout

Path	Role
index.html	Page shell
css/styles.css	Styling
js/data.js	Glossary, recommendations, GitHub URLs, review-article section mapping
js/app.js	Wizard and report UI
start-server.bat / start-server.ps1	Local server (Windows)

---

Notes

• Article ↔ repo mapping: MHA Sec. V; classic ML / ensembles Sec. III-B–III-E; NNs Sec. II + Sec. III-F–III-G; PIML Sec. IV; RL Sec. III-D; simulation automation Sec. III-A; agentic AI Sec. VI; case studies Sec. VII-A–VII-G — see the README alignment table.
• RL: the course ships 7_Reinforcement_Learning/RL_buck_control.ipynb (pedagogical DQN) and 7_Reinforcement_Learning/DDPG_buck_control.ipynb (DDPG, same averaged-buck task). Broader RL at scale still often uses external libraries; the selector links Stable-Baselines3 where relevant.
• PINN: 5_PIML/PINN/pinn_ode.ipynb (Newton cooling ODE) and 5_PIML/PINN/pinn_pde.ipynb (Burgers PDE) follow the same stability-oriented pattern: fixed collocation / IC grids, soft (MSE) constraints, weighted composite losses, Adam with gradient clipping + ReduceLROnPlateau, and optional L-BFGS polish.
• GP + BO (tabular surrogate): 2_Classic_ML/gaussian_process_bayesian_optimization.ipynb demonstrates Gaussian process regression with predictive uncertainty on fetch_california_housing, and an expected-improvement Bayesian optimization loop for hyperparameters—useful pattern when PE simulations/experiments are expensive.
• 3D field / thermal maps: 4_Neural_Network/Field_Data/field_temperature_residual_fnn.ipynb trains a residual FNN on downsampled Tfield_* CSVs (x,y,z,T; loss and Tamb from filenames), with per-file train/val/test splits and 3-D residual plots—complements tabular FNN case studies in the same chapter.
• Some paths (e.g. full GNN training) point to external libraries or papers when the course has no matching notebook.
• Notebook links in the app point to https://github.com/XinzeLee/Fundamentals_of_AI_for_PE/... — keep that repo public or change REPO_ROOT in js/data.js.