Curious about my electric bill, I put current transducers on the three main power circuits in my house, and recorded the current measurements twice a second for about 3 weeks. It's amazing how much one can see in these simple numbers.
AC Currents were measured using three Hawkeye H-922 Analog Current Sensors, $84.61 each from Sentinel Power, www.sentinelpower.com. These marvelous gadgets clip around the conductor (which means they can be installed without interrupting service), and produce an output voltage, in the range 0 - 5 V, proportional to the 60 Hz AC current in the conductor. They use transformer magic to couple to the current being measured, and draw power in the same way, so no battery or other power source is needed.
Analog to digital conversion was performed by an Electronic Energy Control ADC-16 dual 8-channel 8-bit analog-to-digital converter board. I bought this board in the early 1990s, boxed it, and fitted it with screw terminal strips and a wall-wart power supply. It is controlled through an RS-232 serial port.
The software that drives the ADC through the serial port and records the measurements was written in Python, using the PosixSerial package for easy access to the serial port. Grant Edwards, the author of PosixSerial, says that a module named pyserial was built on top of PosixSerial and works on non-Unix platforms.
Graphs of current as a function of time (sample number, actually) were produced by various Python programs that read the (huge) data files and write text files that are then plotted using the excellent, if ancient, xplot program.
I would be happy to share any of this software.
Measurements for Tuesday, 2005-10-04, are here:
Observations:
Enlarging a quiet part of the image, covering the hours of midnight to 6 AM, reveals more:
A recurring theme is visible: a huge current spike, followed by a period of constant current consumption, followed by an abrupt drop. This is the signature of an electric motor, which draws a large current for a second or two while starting.
The forest of spikes is the sum of two roughly periodic populations, one with taller spikes and a period of about 8000 samples, and one with shorter spikes and a period of about 3000 samples. Since there are two refrigerators in this house, it's easy to conclude that these are they.
At breakfast time, we see another trick for distinguishing loads:
When red and blue loads switch on and off simultaneously, that's an indication of a 220-volt load, in this case the electric range heating coffee. Shortly after sample 48500, I probably turned the knob from HI to LOW, since we see the duty cycle fall dramatically.
Finally, an observation on the pool pumps:
Since red and blue jump together, this is a 220-volt pump. The trace starts with the familiar electric-motor spike, then falls back to a more modest current, but the current gradually grows over the next 130 samples, before reaching a plateau. This is the priming period, during which the air-locked pump is spinning almost uselessly, until water reaches it. There's a second pump whose characteristics are quite similar.
To improve readability, I wrote a Python program to pick out familiar structures and move them to the top of the picture. Now the eye can more easily survey what's left:
The red "h" shapes across the top of the plot are the two refrigerators, with their initial spikes, moved 200 and 170 units in the Y direction to separate them from everything else. (My software didn't recognize all of the h's, so there are some gaps.) The broad, black h's are the two pool pumps. Something else occurred periodically, and I moved it up to Y = 160, but I don't know what it is.
Summaries of the information gleaned are embedded as comments in the xplot files:
; channel 0 residue: mean: 0.01 watts, 1 segment ; channel 1 residue: mean: 118.51 watts, 1 segment ; channel 2 residue: mean: 439.76 watts, 1 segment ; fridge 1: mean: 58.02 watts, 50 segments ; fridge 2: mean: 129.88 watts, 20 segments ; pump 1 circuit 1: mean: 189.75 watts, 1 segment ; pump 1 circuit 2: mean: 189.75 watts, 1 segment ; pump 2 circuit 1: mean: 116.66 watts, 1 segment ; pump 2 circuit 2: mean: 116.66 watts, 1 segment ; square 2: mean: 13.61 watts, 44 segments
I looked at 18 single-day plots, took the average Fridge-1 power value from the days when most of the Fridge-1 events were correctly recognized, and adjusted each power value upward in proportion to the apparent number of overlooked events. I then did the same for Fridge 2 and the pool pumps, with the following bottom line:
| watts | marginal $/month | |
| fridge 1 | 62 | 9 |
| fridge 2: | 140 | 20 |
| pump 1: | 187 * 2 = 374 | 55 |
| pump 2: | 117 * 2 = 234 | 34 |
| channel 1 residue: | 203 | 30 |
| channel 2 residue: | 438 | 64 |
| sum | 1451 | 211 |
2006-04-21
My email address:
