Energy Optimal Control of Cooling and Ventilation Systems
Abstract
In the dynamic landscape of an ever changing global economy, efficient energy utilization in large scale industrial systems is the cornerstone of building a sustainable future. Cooling and ventilation systems in industrial and commercial sectors in developed countries are responsible for approximately one-third of global energy consumption. Therefore, optimizing energy usage in these systems can not only provide significant economic benefits but also play a crucial role in mitigating the environmental impact of industrialization. A large scale cooling facility typically consists of cooling towers, chillers, and air handling units. The energy efficiency of different components of a cooling facility can be optimized at different levels, including performance, operation, equipment, and technology. An upgrade in the equipment and technology can maximize energy efficiency at the price of a considerable upfront cost. However, a significant amount of energy optimization can be achieved through retrofit solutions that aim to optimize the performance and operational cost of existing systems.
This thesis investigates the energy optimization of cooling and ventilation systems from an operational viewpoint. Traditional control approaches typically prioritize operational requirements and may overlook opportunities for energy efficient operation. Given the inherent over actuated nature of these systems alongside different power consumption characteristics of actuators and complex system dynamics, leveraging advanced control design techniques can result in a significant reduction in the operational cost of the system. A design approach based on model predictive control (MPC) is proposed to enable the energy optimal operation of different cooling system components. This involves modeling system dynamics and actuators’ power consumption and solving a constrained optimization program to minimize total power consumption while meeting the desired control specifications.
In the first part of the thesis, we investigate the energy optimal control design for an induced draft cooling tower (IDCT). Depending on the weather conditions, an IDCT can be operated in one of three modes: bypass, free cooling (natural draft), and active cooling (forced draft through a fan) to cool down the incoming hot water stream. A switched affine system representation is used to capture the cooling dynamics in various operational modes. A hybrid model predictive control formulation based on solving a mixed integer quadratic problem is proposed for the simultaneous determination of the best operational mode and optimal fan speed to enable energy optimal operation of the IDCT. A two stage optimization based approach is also investigated to handle the task of mode selection and fan speed determination in a computationally efficient manner. An extension of the hybrid MPC framework was provided to account for multiple IDCTs operating with the common goal of regulating the outlet water temperature of the shared water collection basin. Practical considerations like run time balance, robustness against mode unavailability and prevention of excessive mode switching were incorporated to make the framework suitable for practical implementation.
In the second part of the thesis, we address the problem of designing an energy optimal control of air handling units (AHUs). An AHU is an essential component of a heating, ventilation, and air conditioning (HVAC) system. It conditions and circulates the air within a building or an industrial facility. The main goal of the AHU is to maintain indoor air quality and regulate the temperature and humidity within the desired range. First, we study the energy optimal control of air handling units (AHUs) that can be operated in active cooling, economizer, and active heating modes to regulate the zone temperature within the desired range. A control oriented model based on mass and energy balances is adopted and validated against the operational data. The energy optimal control for the air handling unit (AHU) is decomposed into a two layer control architecture (supervisory layer and local control layer). The supervisory layer provides a mix of direct commands and setpoints. The local controllers are employed to maintainthese setpoints. Practical considerations like minimal sensor requirements and online estimation of the disturbances (heat load) are incorporated to make the framework suitable for practical implementation. An extensive simulation study and experimental evaluation are performed to evaluate the energy saving potential across a wide range of operating conditions. The developed solution has demonstrated notable energy savings and enhanced performance during testing on a pilot plant at CERN. The developed framework is later extended to account for the humidity consideration. The humidity consideration within the problem formulation significantly complicates the problem owing to the strong coupling between temperature and humidity. To this end, a steady state hybrid model for the cooling coil is proposed, which accounts for the coupling between temperature and humidity. A novel energy optimal control framework based on generalized disjunctive programming is presented to handle the hybrid nature of the problem. The proposed framework incorporates the power consumption model of different actuators into the objective function while also accounting for the latent heat considerations. Extensive simulations were performed under different weather conditions to evaluate the effectiveness of the proposed approach across a wide range of operating conditions.
The final part of the thesis has two main important contributions: i) A comparison between various mixed integer programming (MIP) formulations for designing MPC for switched affine systems (SAS). We highlight the effectiveness of the generalized disjunctive programming formulation over the widely used mixed logical dynamical systems in designing MPC for SAS. ii) Given the combinatorial nature of the MIP problems, solving MIP online can be very challenging. To this end, we propose employing multitask deep neural networks to approximate the solutions of mixed integer programs. While these approximations can provide fast solutions and ease of embedded implementation, the safety and stability of such solutions are not guaranteed. A computational method is provided to overapproximate the reachable sets of the closed system. Sufficient conditions are then derived for the safety and stability of the SASs. The entire process, from formulating efficient mixed integer programs to approximating solutions and deriving safety and stability guarantees, is demonstrated across several real world examples.