Sponsor: Faculty Initiative Fund (FIF), LUMS
The objective of this study is to investigate the potential for demand response in unstable electricity grids, such as the one in Pakistan. This can result in increased utility for all stakeholders by reducing losses as well as infrastructure upgrade costs. The focus of this work will be residential buildings equipped with electric batteries and possibly solar PV panels. The significance of the work can be understood from the problem being tackled. The energy crisis in Pakistan, stemming from lack of enough installed capacity, is now well over a decade old and shows no sign of abating. This inadequacy of supply leads to many hours of so-called load shedding every day, causing massive losses to local industry and reduced household utility. Household consumers have typically installed battery-inverter systems to cope with this demand–supply gap, often complementing this by local generation sources such as solar photovoltaic (PV) panels. Besides being an additional financial burden on the consumer, there is little visibility on the adverse consequences of installing these systems. These include electric losses (incurred every time the battery is recharged) which can increase the annual electricity bill for a household by thousands of rupees. The presence of many such battery-inverter or solar PV systems in a neighborhood can also lead to costly grid reinforcement requirements. Intelligent, automatic control using reinforcement learning can reduce these adverse effects by charging batteries only as much as necessary depending on consumer demand, local solar production (if any) and recent electricity load shedding patterns. This reduces the household electricity bill and improves the longevity of both the battery and electric grid. This project will focus on developing a framework which makes all of this possible. The result of the project will be in the form of concrete findings for homeowners as well as for policy makers. Additionally, a control algorithm based on cutting edge reinforcement learning techniques will be developed and tested in simulations on real data. The basic hypothesis underlying the project is twofold. First, it seeks to inform usage of a common commodity in Pakistan (and other developing countries), namely the battery-inverter. Secondly, it intends to investigate the applicability of recent advances in reinforcement learning (RL) for demand response of electricity. The objective of this project is to investigate the potential for demand response in the case of an unstable grid such as in Pakistan using commonly available electric battery-inverter systems and to study the trade-off between occupant comfort and energy efficiency with different control strategies.