Non-Intrusive Load Monitoring (Energy Disaggregation) Using Sequence-to-Point Deep Learning

A smart meter measures only the total power of a house. Non-intrusive load monitoring (NILM) estimates from that single signal how much each appliance group uses - for energy feedback, audits and demand response. This project disaggregates the three sub-metered circuits of a REAL house (kitchen, laundry, electric water heater + air conditioner) from its aggregate active power: UCI Machine Learning Repository dataset 235 "Individual household electric power consumption" (Hebrail & Berard, DOI 10.24432/C58K54, CC BY 4.0), one house in Sceaux near Paris, 2,075,259 one-minute rows from December 2006 to November 2010. The classical combinatorial optimisation (CO, Hart 1992) with k-means power states is compared with LightGBM on window features and a sequence-to-point 1-D CNN (Zhang et al. 2018) on a leakage-safe time split (train Dec 2006 - May 2009, validation Jun 2009 - May 2010, test Jun - Nov 2010; the same 122,265 test samples for every method). Mean absolute error over the three circuits on the test months: mean predictor 216.2 W, CO 194.9 W, LightGBM 66.4 W, LightGBM+ (with reactive power, voltage, current and time of day) 57.4 W, seq2point CNN 63.9 W (deployed - chosen on the validation year). For the water heater + AC the seq2point energy error is 2.1 % with on/off F1 0.92; kitchen and laundry remain hard. GNU Octave reproduces the classical part exactly (run), a complete MATLAB version and a NILMTK experiment are written (not run), and a web application with five databases turns the estimates into kWh per circuit, a two-rate tariff split and load-shifting advice. Everything runs on a standard laptop CPU (8 GB RAM, no GPU).

Category: Electrical and Electronics Engineering (Smart Grid, Smart Metering, Energy Management, Non-Intrusive Load Monitoring, Energy Disaggregation, Demand Response, Time-of-Use Tariffs, Combinatorial Optimisation, Deep Learning, Sequence-to-Point Convolutional Neural Network, LightGBM, PyTorch, NILMTK, GNU Octave, MATLAB, Python)

Electrical & Electronics Engineering Project EE15 from ProTechSC Academy, with graphical abstract, video demonstration and project guidance.