The construction industry has experienced important changes in recent years due to advancements in digital, artificial intelligence, and construction technologies, as well as the sector's on-going development and the advancement of science and technology. The creative growth of building industry, creative creation of architectural forms are partially supported technically by sophisticated parametric design apparatuses, the potent computing benefits of computer technology. In this manuscript, Exploration of Natural Element Form Optimization Algorithm using Spatial-Temporal Multi-Scale Alignment Graph Neural Network in Architectural Design Based on Morphological Theory (ENEF-OA-ADMT) is proposed. The STMSA-GNN and the Chaotic Coyote Algorithm (CCA) are two tools used by the proposed ENEF-OA-ADMT approach to improve architectural design based on morphological theory. The ST-MSA GNN's ability to capture intricate interactions and dependencies between several components in both space and time allows it to perform a comprehensive study of the morphological aspects of architectural designs. This graph neural network's integration of spatial and temporal dimensions enables a deeper understanding of how the architectural structural form design changes over time. The CCA optimized the ST-MSA-GNN to enhance the architectural structural form design. The proposed ENEF-OA-ADMT methodology skill fully combines these methodologies, creating a strong framework that allows architects and designers to work together to explore, refine, and create architectural structural design forms. The framework provided serves as a spur for further research, encouraging a more complete integration of technology and environment in the architectural domain. The effectiveness of proposed method is executed in python, evaluated through performance metrics encompassing accuracy, precision, specificity, Recall, computational time, F1 score, population diversification, randomness. Proposed ENEF-OA-ADMT method 34.56%, 28.63% and 21.89% higher accuracy, 34.97%, 32.13% and 21.89% higher precision and 34.68%, 20.84% and 29.76% higher randomness when compared with the existing methods such as Study of Morphological Design of Architecture from Geometric Logic Perspective (SOT-MDA-GLP), learning deep morphological networks by neural architecture search (LD-MN-NAS) and identifying degrees of deprivation from space utilizing deep learning with morphological spatial analysis of deprived urban areas (IDDS-DLMSA-DUA) respectively.