Design of an Improved Model for Smart Grid Pricing Using ST-GNN-PNet and MAD-RL-StackelNet

¹ Assistant Professor, Department of Electrical Engineering, P. R. Pote Patil College of Engineering & Management, Amravati, Maharashtra
² Professor, Department of Electrical Engineering, P. R. Pote Patil College of Engineering & Management, Amravati, Maharashtra
³ Assistant Professor, Department of Electrical Engineering, P. R. Pote Patil College of Engineering & Management, Amravati, Maharashtra
⁴ Assistant Professor, Department of Electrical Engineering, P. R. Pote Patil College of Engineering & Management, Amravati, Maharashtra

^* Corresponding author: sajalit@prpotepatilengg.ac.in

Published online: 18 June 2025

Abstract

The shift to decentralized smart grids requires dynamic pricing based on demand, supported by advanced technology to adapt to behavioral changes. However, current pricing models fail to capture spatio-temporal load behavior, consumer heterogeneity, and externalities like emissions. Privacy constraints also hinder granular data collection, causing revenue loss. To address these issues, this proposal introduces the Topo-Behavioral Hybrid Learning Model (TBHLM) for dynamic pricing in smart grids. TBHLM has five key modules, ST-GNN-PNet: Uses temporal graph convolutions to forecast loads, congestion, and locational marginal prices (LMPs) with <3.5% MAPE and <3s latency. FBEM-Net: Applies federated learning for privacy-preserving elasticity modeling, achieving ~92% behavioral prediction accuracy and a 15% increase in demand response participation. MAD-RL-StackelNet: Uses multi-agent reinforcement learning for equilibrium pricing, leading to 18-22% peak shaving and a 30% rise in pricing stability. RBEIO-Opt: Integrates carbon penalties into economic dispatch, reducing emissions by 12.6% and improving welfare by 6.1%. PIDE-Engine: Uses inverse optimization for utility estimation with a privacy breach probability of <0.01%. TBHLM provides an adaptive, secure, and consumer-centric framework for real-time pricing, enhancing efficiency, sustainability, and grid intelligence sets.