Office plug-load dataset for non-intrusive load monitoring in buildings using an open metering platform

Non-intrusive load monitoring (NILM) can support building energy management by estimating appliance-level consumption from aggregate electrical measurements. Yet, most public benchmarks are residential and provide limited evidence for office plug-load monitoring. This paper presents an office plug-load dataset acquired with Open Multi Power Meter v2, an open, low-cost, multi-channel metering platform for time-aligned aggregate and appliance-level monitoring in office and institutional buildings. The dataset was collected under natural operating conditions, without programmed switching sequences or artificial load automation, thereby preserving realistic standby periods, sparse events, low-power operation, and interactions among office devices. The dataset comprises 346,252 time-stamped records from six meters over approximately 10 days of continuous office operation. Active power and complementary electrical and environmental variables were recorded at a 15 s sampling interval and converted to a NILMTK-compatible format. Baseline results show that office plug-load disaggregation is governed by load observability. A Factorial Hidden Markov Model achieved the best overall balance between accuracy and execution time, whereas the two lowest-power devices yielded zero F1-scores despite low absolute power errors. This indicates that some office loads can be measured accurately at the submeter level but remain weakly disaggregable from aggregate active power alone. The results highlight practical limits for device-level NILM in offices and support monitoring strategies based on device grouping, selective submetering, or complementary features.