Research
Profiles
Research
Profiles
Embedded Files
Thesis
Thesis
Inverse Design of Plasmonic Sensors: Optimization and Performance Enhancement Using Deep Learning
Undergrad Thesis (Aug 2024 – Sept 2025)
Supervisor: Prof. Dr. Rakibul Hasan Sagor
Abstract:
Abstract:
This research presents a reinforcement learning (RL)–based inverse design framework developed to enhance the performance of plasmonic refractive index nanosensors by optimizing their geometric and topological parameters. Plasmonic refractive index sensors operate by exploiting the interaction between light and the metal–insulator interface, where surface plasmon polaritons (SPPs) enable highly sensitive, label-free detection of refractive index variations in the surrounding medium. In this work, the inverse design framework was implemented using the Lumerical scripting language integrated with Python, enabling automated, iterative refinement of sensor geometry toward predefined targets of sensitivity and figure of merit (FOM). The optimization process employed a Deep Q-Network (DQN)–based reinforcement learning algorithm, wherein the agent dynamically adjusted key design parameters such as resonator radius, width, thickness, and nanorod dimensions. Beginning with a pentagonal ring resonator and an initial target sensitivity of 2000 nm/RIU, the RL agent progressively evolved the geometry into an octagonal configuration. Upon incorporating FOM into the reward function, the optimized design achieved a sensitivity of 2638.15 nm/RIU and an FOM of 10.71 RIU⁻¹. The results demonstrate that coupling reinforcement learning with plasmonic inverse design significantly accelerates the discovery of high-performance sensor geometries, outperforming conventional manual and brute-force optimization methods, and highlight the potential of RL-driven optimization as a transformative approach for the intelligent design of plasmonic sensors with enhanced sensitivity, compactness, and operational efficiency.
Publications
Publications
Multi-Year Dataset on Daily Electricity Demand, Generation, Load Shedding, and External Conditions in Bangladesh
Md. Ikrama Hossain, Tasnia Nafs, Sakif Yeaser, and Asif Newaz
Data in Brief- Volume 62, October 2025, 112014
Abstract:
Abstract:
This dataset compiles daily electricity statistics for Bangladesh across national and divisional levels. The data were programmatically scraped from the Bangladesh Power Development Board’s (BPDB) digital archive and processed into a structured, machine-readable format using a custom Python pipeline. The dataset consists of 1867 daily reports, spanning from November 21, 2019, to December 30, 2024. Each record includes key variables such as electricity demand, generation, load shedding, temperature, and supply limitations due to gas shortages, coal availability, and low water levels. The dataset was curated through multiple stages, which include manual verification, holiday classification, missing value imputation, and outlier correction. Five sequential versions are provided, which reflect progressive enhancements from raw extraction to modeling readiness. The data can be used in time series analysis, load forecasting, energy policy research, and machine learning applications in resource-constrained settings. Additionally, the collection spans the COVID-19 pandemic period, offering unique opportunities for studying the impact of external factors on national energy systems.
Comparative Analysis of Deep Learning Models for Long-Term Electricity Demand Forecasting in Bangladesh Using Web-Scraped Data
Sakif Yeaser, Tasnia Nafs and Md. Ikrama Hossain
2025 17th International Conference on Electronics, Computers and Artificial Intelligence (ECAI), Târgoviste, Romania
Abstract:
Abstract:
As energy demand continues to rise in Bangladesh, there is a growing need for more accurate forecasting methods to improve the balance between electricity supply and consumption. Despite increased generation capacity, the country still experiences frequent disruptions due to limitations in prediction accuracy and structural inefficiencies within the power system. This research carries out an evaluative comparison of notable deep learning (DL) frameworks, including Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), Bidirectional LSTM (BiLSTM), and Bidirectional GRU (BiGRU), for forecasting daily peak electricity demand at both the national level and across the country's eight divisions. The dataset was compiled from the Bangladesh Power Development Board (BPDB) using an automated web scraping pipeline. All models were trained on four years of historical data and evaluated using a one-year testing set. Among the models assessed, the BiGRU architecture outperformed others, achieving the lowest testing Mean Absolute Percentage Error (MAPE) value of 4.75%. The BiGRU model was also employed for division-wise forecasting, effectively capturing regional demand variations. Additionally, it was applied to unseen future dates, which are not included in the dataset, where it recursively predicted energy demand one day at a time and achieved an MAPE of 7.3%, demonstrating strong generalization capability. These results signify the aptitude of deep learning-based methodologies for enabling resilient and scalable energy consumption modeling.
Page updated
Google Sites
Report abuse