Increased focus on sustainability and energy decentralization has positively impacted the adoption of nanogrid (NG). Despite its perceived advantages, NG’s high load volatility poses risk not only to its stability but also to the associated network. This is further exacerbated by the user’s transient moods and the probabilistic nature of metrological events. Moreover, unoptimized operation of electrical appliances in a NG, especially in peak hours, raises the energy cost substantially. Thus, for risk aversion and cost savings, day-ahead demand-side optimization is highly required, especially for shiftable loads. Moreover, most load centers, aspiring to be NGs, have either limited or no data about their shiftable loads’ operations. This makes NG’s shiftable load optimization a daunting task. Existing demand-side management models, predominantly focused on lesser volatile loads, do not cater to the challenges associated with NG, hence are deemed unsuitable. Additionally, comfort levels are also compromised in these models. Thus, to address the challenges, a comprehensive predictive demand-side management (PDSM) approach is developed in this paper. It comprises two components: 1) Integrated machine learning-based shiftable load forecasting and 2) user-centric multi-objective optimization through load shifting. To predict the day-ahead shiftable load, the Stacked-Long Short-Term Memory (SLTSM), Artificial Neural Network (ANN), and Shiftable Equipment Matrices (SEM) modules are integrated. SLSTM module predicts day-ahead load variations (%) using time series data, segregated by the percentile-based method, with a lag of 24 hours (t-24). ANN module, with dynamic feature selection using k-means, predicts day-ahead load forecasting (kW) based on meteorological and load data. Furthermore, a user-centric Mixed Integer Quadratic Programming optimization model is developed that shifts the predicted shiftable load with minimum energy cost and the least discomfort to the user. Results show that the SLSTM predicts variations with R2 of 97.6%, MAPE of 9.7%, and RMSE of 0.165%. ANN predicts load with R2 of 99.5%., Mean Average Percentage Error (MAPE) of 9% and RMSE of 0.165kW and the integrated component predicts shiftable load with R2 of 97.1%. Moreover, the developed optimization model can save up to 5.17% of daily energy costs while incorporating users’ comfort levels.

Library of Congress Subject Headings

Smart power grids; Electric power consumption--Forecasting; Electric power systems--Management; Mathematical optimization--Data processing

Publication Date


Document Type


Student Type


Degree Name

Electrical Engineering (MS)

Department, Program, or Center

Department of Electrical and Microelectronic Engineering (KGCOE)


Bing Yan

Advisor/Committee Member

Sohail A. Dianat

Advisor/Committee Member

Santosh Kurinec


This thesis has been embargoed. The full-text will be available on or around 5/16/2024.


RIT – Main Campus

Plan Codes


Available for download on Wednesday, May 15, 2024