Energy Digital Twin of a Steel Plant for Load Forecasting and Energy-Aware Production Scheduling

Electricity is one of the largest controllable costs of a steel plant, and time-of-use tariffs and demand charges reward moving flexible jobs to cheaper hours. This project builds an energy digital twin from a year of real 15-minute meter data of a Korean steel plant (UCI Steel Industry Energy Consumption, 2018): an honest day-ahead load forecast (seasonal naive, calendar regression and SARIMA-type baselines against HistGradientBoosting, LightGBM with quantiles and an LSTM), a time-indexed MILP that places simulated flexible production jobs on top of the forecast base load to minimise time-of-use energy cost plus demand charge under a peak cap (HiGHS, cross-checked with CBC and CP-SAT), and a rolling-horizon twin that updates the forecast after every meter reading and re-optimises the jobs that have not started. A web application forecasts, schedules, compares what-if tariffs, replays days and dispatches approved schedules; Node-RED replays meter data live. A complete MATLAB version is included.

Category: Industrial Engineering

Industrial Engineering Project IE08 from ProTechSC Academy, with graphical abstract, video demonstration and project guidance.