Merges Multi-Objective Reinforcement Learning with Lightweight Surrogate Models to Dynamically Throttle Machine Power States Based on Real-Time Edge Analytics and Dynamic Energy Grid Tariffs

• Ada John

  ladoke Akintola university of technology

  Author

Manufacturing accounts for a substantial share of global electricity consumption, and dynamic tariff mechanisms create economic incentives for intelligent load regulation. However, existing machine tool power management methods struggle to respond in real time at the edge to the multi-objective trade-off between tariff fluctuations and production constraints. This study proposes a framework that merges multi-objective reinforcement learning with lightweight surrogate models to dynamically throttle machine power states within an edge analytics environment. A design science research methodology is adopted, and the framework is validated in a simulated edge computing environment. Results show that the proposed framework achieves 89.4% classification accuracy in predicting optimal power states, reduces energy costs by 23.7% compared with a static scheduling baseline, and keeps production constraint violation rates below 2.1%. The study concludes that a surrogate-assisted multi-objective reinforcement learning framework can deliver effective power state throttling decisions within millisecond-level time budgets, offering a deployable technical pathway for manufacturers to participate in dynamic tariff response.

• pdf

Copyright (c) 2026 Ada John (Author)

This work is licensed under a Creative Commons Attribution 4.0 International License.