Published: 03-08-2020| Version 1 | DOI: 10.17632/hsmmhc997z.1
Wanderley Celeste,
Victor Pereira Firmes,
Leonardo José Silvestre,
Daniel José Custódio Coura,
Luís Otávio Rigo Júnior,
Helder Rocha


Each dataset has contained in a zipped folder with samples of electrical voltage and current demanded by devices of an electrical installation. The samples were acquired using the NI 9220 module from National Instrument, at a sampling rate of 100 kilo-samples per second and a 16-bit sample resolution. The installation's devices are limited to a maximum of four, resulting in 16 distinct loads. Samples of each resulting load have acquired and arranged in the following columns: L0 (no device on), L1 (device 1 on), L2 (device 2 on), L3 (devices 1 and 2 on), L4 (device 3 on), L5 (devices 3 and 1 on), L6 (devices 3 and 2 on), L7 (devices 3, 2 and 1 on), L8 (device 4 on), L9 (devices 4 and 1 on), L10 (devices 4 and 2 on), L11 (devices 4, 2 and 1 on), L12 (devices 4 and 3 on), L13 (devices 4, 3 and 1 on), L14 (devices 4, 3 and 2 on), and L15 (all four devices on). As devices, we firstly used four technically identical fluorescent lamps - from the same manufacturer, with 15 W of power, 127 V, and power factor of 0.5. We applied acquisition windows of 80 seconds to obtain samples of each of the sixteen loads resulting from the combination of fluorescent lamps. They are in zipped folder A. The folders A0, A1, ... A15 contain 600 cases per class in each have derived samples of the dataset stored in A, with the respective processing performed as described in the sequel: - A0, A1, A2, and A3 are composed of 100%, 50%, 25%, and 12.5% of consecutive samples per case of the dataset A; - A4, A5, and A6 have built by getting samples spaced every 2, 4, and 8 samples from A until reaching 600 cases with 1666 samples. Each case represents 2, 4, and 8 60 Hz cycles in each secondary dataset, respectively, that is a subsampling process (a division of the original sampling rate by the respective spacing factors); - A7, A8... A13 and A14 are built from 600 first consecutive cases of the dataset A, employing sample quantization in 15, 14, and 8 bits, respectively, this is a reduction of sample resolution. We also use as devices, four microcomputers Dell Optiplex 960 Core 2 Duo 3 GHz, 4 GB of RAM, and 150 GB HD, even technically identical. Ten acquisition windows, each lasting 1 second, randomly spaced in time, were performed to obtain samples for each of the sixteen loads resulting from the combination of microcomputers. They are in zipped folder B.