OpenCEM Research

OpenCEM Data and Simulator

OpenCEM Algorithms

InstructMPC

InstructMPC is a novel framework that enhances traditional Model Predictive Control (MPC) by integrating real-time contextual information through its Language-to-Distribution (L2D) module, which translates contextual information—such as events, weather, and user-generated data—into predictive disturbance trajectories. These trajectories are then incorporated into the MPC optimization process.

Within OpenCEM, the abundance of real-world time series and diverse contextual data (e.g., local news, weather reports, event schedules) serve as valuable inputs for InstructMPC. By leveraging these inputs, InstructMPC can adapt its control strategy more effectively than traditional MPC methods, making it particularly powerful for energy management scenarios. The framework’s dynamic human-LLM interaction allows for real-time adjustments and refined decision-making, resulting in more responsive and context-aware control policies.

As part of the OpenCEM initiative, InstructMPC will help demonstrate how in-context human instructions and local data can be used to improve energy management decisions. Stay tuned for future updates and data releases that will further support and showcase InstructMPC in real-world settings.

R. Wu, J. Ai, and T. Li, Instructmpc: A human-llm-in-the-loop framework for context-aware control, 2025.