Skip to content
Open access

A Cognitive Radio-Enhanced Smart Grid Architecture with Cloud-Native Hybrid Bi-LSTM–Attention Microservices for City-Scale Load Forecasting over NAN/WAN Hierarchies

Sep 2026 · Journal of Al-Farabi for Engineering Sciences · 0 citations

Abstract

The increasing deployment of distributed sensing and data-driven control in smart grids requires forecasting services that are accurate at the system level and viable over heterogeneous communication infrastructures. This paper proposes a cognitive-radio-enhanced smart-grid architecture in which a Hybrid Bidirectional Long Short-Term Memory (Bi-LSTM)–Multi-Head Attention model is exposed as a cloud-native microservice across Neighbourhood Area Network (NAN) and Wide Area Network (WAN) hierarchies. The architecture maps household- and building-level telemetry to regional megawatt-scale load forecasting through edge aggregation, in-network feature processing, cognitive-radio-assisted NAN communication, and secure cloud-service interfaces. A 24-hour sliding window is used to forecast the subsequent 24 hourly load values from PJM regional time-series data. The forecasting model is evaluated using normalised root mean square error (NRMSE), normalised mean absolute error (NMAE), coefficient of variation of RMSE, normalised mean bias error, mean absolute percentage error (MAPE), and the coefficient of determination (R²). Across the regions reported in the present results table, NRMSE ranges from 0.0125 to 0.0372, MAPE from 1.28% to 4.78%, and R² from 0.9260 to 0.9915. The system-level evaluation further defines an experimental protocol for measuring end-to-end tail latency, NAN throughput, bit-error rate, serialisation overhead, and resource consumption under controlled communication conditions. The proposed framework links predictive performance with network-service viability rather than dealing with load forecasting as an isolated offline learning task. The study achieved a reproducible basis for assessing whether attention-enhanced forecasting can be deployed as a scalable network function for regional smart-grid operation. Also, this study claims concerning communication performance are limited to the conditions and simulations explicitly reported.

Read PDF

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.