Naran M. Pindoriya

This article presents a multi-stage deep reinforcement learning framework for volt-var or volt-watt optimization (VVO/VWO) in solar photovoltaic-rich active distribution networks (ADNs) to enhance voltage regulation and minimize network losses. The framework introduces a centralized–decentralized training and execution structure for a multi-agent environment that segregates the objectives—voltage regulation and loss minimization into decentralized and centralized tasks, respectively. This architecture leverages independent learning to reduce training iterations and the computation overheads, particularly in large-scale agent deployments, while sharing limited information during centralized training. A novel contribution factor method based on reward backstepping is proposed to address the non-stationarity caused by independent learning, which enables accurate credit assignment and stable learning. The proposed framework has been evaluated and validated on a practical ADN, a modified IEEE 33 bus, and an IEEE 119 bus system that are segmented into different areas of control (AoC), where each consumer within the AoC comprises local loads and an individual PV-fed inverter, which serves as the primary controlling device for executing VVO/VWO. The results indicate effective reduction of voltage violations and network losses, improved training convergence, and decreased computational requirements compared to existing frameworks.

The cost of unserved energy is a parameter used to calculate the worth of electricity that is not served by a power system due to planned and unplanned outages. This figure is influenced by several factors such as affected group, duration of interruption, time of interception, etc. This paper studies the impact of interruption duration on the cost of unserved energy experienced by the consumers in the commercial sector. The study is based on a consumer survey conducted to gather the data required from the consumers belong to 15 subcategories. Using the collected data, average and weighted average cost of unserved energy due to expected and unexpected outages corresponding to 5 outage durations are computed. Results are analyzed and identified patterns in variations are discussed. The analysis reveals that the cost of unserved energy due to unexpected outages of most subcategories is high for short outage durations, when compared to that of long outage durations. Whereas, the cost of unserved energy due to expected outages gradually increases with the interruption duration. Thus, this study reveals that the cost of unserved energy shows different variations with the interruption duration for expected and unexpected outages.