Forecasting Power Consumption of Home Appliances using LSTM

As the consumption of energy in the world rises in means of both households, business organizations and industries, it is becoming paramount for having sophisticated and efficient sustainable energy management practices. In these recent years it has been uplifting in grabbing the attention towards the exploitation of machine learning techniques which can precisely prognosticate the consumption of power, with the aim of minimizing the usage of energy, cost reduction and develop sustainability. While antecedent studies have laid the foundation, this research distinctively approaches the challenge by aggregating insights gleaned from prior investigations. In alignment with the continuum of prior research, the proposed forecasting approach culminates in a novel perspective. This paper aims to address this challenge by exploring the use of sensors to collect user-specific data for developing customized models that can provide accurate power consumption forecasts. The dataset for the study is developed by assimilating Split Core Transformer (SCT-013) and Zero Mode Plate Transformer (ZMPT-101B) sensors with the Micro Controller Unit (Node MCU ESP32) obtaining the power consumption values. The sensor-based forecasting approach was carried out and the proposed Long Short-Term Memory (LSTM) model achieved a Mean Squared Error (MSE) value of 5.81e-08 and a Mean Absolute Error (MAE) value of 0.000197, indicating its effectiveness in accurately forecasting power consumption. Additionally, the high R-squared score of 0.98 recommends a well defended association between the prognosticated and actual power consumption data. In essence, this study reframes the trajectory of power consumption forecasting. It amalgamates previous insights into a novel framework that harnesses sensor-derived data, poised to redefine the landscape of energy management with tailored accuracy and unwavering reliability.