Edge Deployment of Behaviour-Based Smart Home Automation System Using LSTM on Raspberry Pi for Real-Time Electrical Load Control

Abstract

This paper presents the edge deployment of a behaviour-based Smart Home Automation System (SHAS) that integrates Internet of Things (IoT) sensors, Long Short-Term Memory (LSTM) deep learning, and real-time control on a Raspberry Pi platform. Unlike cloud-dependent approaches, the proposed system executes all processes at the edge, including data collection, streaming dataset generation, model training, and automatic control execution. A behaviour-oriented LSTM model is developed from three months of historical data and dynamically retrained through an adaptive learning mechanism when power consumption exceeds the overload threshold or user comfort is compromised. Implemented on Node-RED, the system successfully controls smartplug-connected appliances in real time, achieving low-latency communication, stable edge resource utilization, and reliable prediction of ON/OFF appliance states. Experimental results indicate a 29.2% reduction in overload duration, demonstrating the effectiveness of the edge-based deployment for both energy safety and operational efficiency. This work highlights the practicality of running deep learning models for behaviour-based energy management entirely on edge devices, offering enhanced privacy, reduced dependency on internet connectivity, and scalability for real-world smart home environments.