EPE Association: EPE 2025 - DS2g: Use of AI in Power-Electronics Applications

EPE 2025 - DS2g: Use of AI in Power-Electronics Applications
You are here: EPE Documents > 01 - EPE & EPE ECCE Conference Proceedings > EPE 2025 - Conference > EPE 2025 - Topic 04: Digitalization: the Powerful Fusion of AI and IoT for Sustainability > EPE 2025 - DS2g: Use of AI in Power-Electronics Applications

   [return to parent folder]

   Harmony Search-Assisted Perturb-and-Observe MPPT Algorithm for PV Systems
By Geon Hee LEE, Seungho CHOI, Zong Woo GEEM, Jehyuk WON [View]
[Download]

Abstract: Nonuniform irradiation in photovoltaic (PV) systems can cause multiple power peaks, where traditional maximum power point tracking (MPPT) methods like perturb-and-observe (P&O) are only effective under uniform conditions. Accurate global MPPT is crucial under partial shading conditions (PSCs). This paper proposes a hybrid MPPT algorithm integrating harmony search (HS) with P&O. This approach reduces power loss during searches and avoids getting stuck at local maxima under PSCs, a common issue with standard P&O methods. A parameter-setting-free HS (PSF-HS) is introduced to enhance search efficiency, followed by variable-step-size P&O to fine-tune the GMPP location. Experiments on a 250 W PV system under various PSCs using a PV emulator demonstrate the algorithm's effectiveness.

   LTO BATTERY USEFUL LIFE PREDICTION FOR ALWAYS ON EDGE AIoT BASED STRUCTURAL HEALTH MONITORING
By Ivan ARAKISTAIN, Diego ZAMORA, David GARCIA-SANCHEZ, Alberto ARMIJO [View]
[Download]

Abstract: Although many high-sampling sensor systems tend to be power-hungry, critical monitoring applications require reliable battery-powered sensor nodes that can retrieve and compute data on the edge for years. With the advent of Tiny Machine Learning (TinyML), it is becoming increasingly feasible to deploy always-on inference Machine Learning models on constrained battery-powered microcontroller-based nodes. However, owing to unpredictable and dynamic energy harvesting availability conditions and the limitations of battery technology, long-term operation is still challenging. In this paper, we present a hardware and software solution for long term continuous solar operation of power-hungry wireless sensor nodes with Lithium titanate oxide (LTO) batteries.

EPE 2025 - DS2g: Use of AI in Power-Electronics Applications
You are here: EPE Documents > 01 - EPE & EPE ECCE Conference Proceedings > EPE 2025 - Conference > EPE 2025 - Topic 04: Digitalization: the Powerful Fusion of AI and IoT for Sustainability > EPE 2025 - DS2g: Use of AI in Power-Electronics Applications
	[return to parent folder]

	Harmony Search-Assisted Perturb-and-Observe MPPT Algorithm for PV Systems By Geon Hee LEE, Seungho CHOI, Zong Woo GEEM, Jehyuk WON	[View] [Download]
Abstract: Nonuniform irradiation in photovoltaic (PV) systems can cause multiple power peaks, where traditional maximum power point tracking (MPPT) methods like perturb-and-observe (P&O) are only effective under uniform conditions. Accurate global MPPT is crucial under partial shading conditions (PSCs). This paper proposes a hybrid MPPT algorithm integrating harmony search (HS) with P&O. This approach reduces power loss during searches and avoids getting stuck at local maxima under PSCs, a common issue with standard P&O methods. A parameter-setting-free HS (PSF-HS) is introduced to enhance search efficiency, followed by variable-step-size P&O to fine-tune the GMPP location. Experiments on a 250 W PV system under various PSCs using a PV emulator demonstrate the algorithm's effectiveness.

	LTO BATTERY USEFUL LIFE PREDICTION FOR ALWAYS ON EDGE AIoT BASED STRUCTURAL HEALTH MONITORING By Ivan ARAKISTAIN, Diego ZAMORA, David GARCIA-SANCHEZ, Alberto ARMIJO	[View] [Download]
Abstract: Although many high-sampling sensor systems tend to be power-hungry, critical monitoring applications require reliable battery-powered sensor nodes that can retrieve and compute data on the edge for years. With the advent of Tiny Machine Learning (TinyML), it is becoming increasingly feasible to deploy always-on inference Machine Learning models on constrained battery-powered microcontroller-based nodes. However, owing to unpredictable and dynamic energy harvesting availability conditions and the limitations of battery technology, long-term operation is still challenging. In this paper, we present a hardware and software solution for long term continuous solar operation of power-hungry wireless sensor nodes with Lithium titanate oxide (LTO) batteries.