A Central Chilled Water Plant Model for Designing Learning-Based Controllers

TL;DR

This paper presents a comprehensive and validated model of a central chilled water plant that enhances existing models with capacity-respecting constraints, enabling more reliable training of learning-based control systems.

Contribution

It introduces a constrained optimization framework for modeling CCWPs, improving model validity across wider operating ranges for learning-based controller development.

Findings

Model respects heat exchanger capacities under all inputs.

Enhanced model validity beyond normal operating ranges.

Publicly available Matlab implementation.

Abstract

We describe a framework of modeling a central chilled water plant (CCWP) that consists of an aggregate cooling coil, a number of heterogeneous chillers and cooling towers, and a chilled water-based thermal energy storage system. We improve upon existing component models from the open literature using a constrained optimization-based framework to ensure that the models respect capacities of all the heat exchangers (cooling coils, chillers, and cooling towers) irrespective of the inputs provided. As a result, the proposed model has a wider range of validity compared to existing models; the latter can produce highly erroneous outputs when inputs are not within normal operating range. This feature is essential for training learning-based controllers that can choose inputs beyond normal operating conditions and is lacking in currently available models. The overall plant model is implemented in Matlab and is made publicly available. Simulation of a CCWP with closed loop control is provided as an illustration.