SO-Attn-ALSTN: A Deep Learning and Seeker Optimization-Based Framework for Low-Voltage Distribution Network Planning

Abstract

Rising electricity demand, the proliferation of distributed energy resources (DERs), and the complexity of modern urban infrastructure pose significant challenges for low-voltage distribution network (LVDN) This research proposes an intelligent optimization approach for LVDN planning using Deep learning (DL) to address limitations in traditional methods, addressing dynamic load patterns and renewable energy integration. The goal is to reduce power losses and infrastructure costs while maintaining voltage stability and load balancing across the network. A comprehensive low-voltage smart grid planning dataset was sourced from an open-access platform, Kaggle. To assure data quality, normalization and outlier reduction were performed during preprocessing. Fast Fourier Transform (FFT) was used to extract features and uncover hidden patterns in load demand and energy flows. This research proposes a Seeker Optimized Attention with Adjustable Long Short-Term Network (SO-Attn-ALSTN) model, which combines an attention-enhanced ALSTN for spatiotemporal load forecasting with a Seeker Optimization Algorithm (SOA) for efficient planning. Attention enhances ALSTN performance by focusing on temporal inputs, while SOA ensures robust parameter tuning and faster convergence. Forecasted loads optimize cable routing, transformer sizing, and DER allocation. Experimental results validate the model's superiority: the proposed SO-Attn-ALSTN achieved a MAPE of 6.53%, RMSE of 1.14%, MAE of 0.99%, and APE of 2.01%. Comparative convergence time analysis shows a 30–40% improvement over existing methods, LMBP and IGWO-SVM, with a convergence time of 2.708 seconds at an error threshold of 0.01. Thus, the hybrid SO-Attn-ALSTN framework presents an intelligent, adaptive, and computationally efficient solution for modern LVDN planning.