Chasing Power Anomalies

Recently, we and a number of our neighbors have been noticing our lights flickering in the evening and early morning. While we have considered it to be mostly an annoyance, this has bothered some of our neighbors enough that they have opened cases with the utility and began raising the issue on our street mailing list.

Pacific Gas and Electric (PG&E) responded to these customers with a visit, and in some cases replaced the service entrance cable to the home. In at least one case PG&E also said they might need to replace the pole transformer, which would take a few months to complete. I have heard no reports that these efforts have made any difference.

This isn’t our first recent challenge with voltage regulation in our neighborhood. Our most recent issue was a longer-term voltage regulation problem that occurred on hot days, apparently due to load from air conditioners and the fact that our neighborhood is fed by older 4-kilovolt service from the substation. This is different, and raised several questions:

• How local are the anomalies? Are neighbors on different parts of the street seeing the same anomalies, or are they localized to particular pole transformers or individual homes?
• What is the duration and nature of the anomalies?
• Are they only happening in the evening and early morning, or do we just notice them at these times?

To try to answer these questions, I found a test rig that I built several years ago when we were noticing some dimming of our lights, apparently due to neighbors’ air conditioners starting on summer evenings. The test rig consists of a pair of filament transformers: 110 volt to 6 volt transformers that were used in equipment with electronic tubes, which typically used 6 volts to heat the tube’s filament. The transformers are connected in cascade to reduce the line voltage to a suitable level for the line-in audio input on a computer. An open-source audio editing program, Audacity, is used to record the line voltage. I often joke that this is a very boring recording: mostly just a continuous 60 hertz tone.

At the same time, I started recording the times our lights flickered (or my uninterruptable power supply clicked, another symptom). I asked my neighbors to record when they see their lights flicker and report that back to me.

I created a collection of 24-hour recordings of the power line, and went looking for the reported power anomalies. It was a bit of a tedious process, because not everyone’s clocks are exactly synchronized. But I was successful in identifying several power anomalies that were observed by neighbors on opposite ends of the street (about three blocks). Here’s a typical example:

As you can see, the problem is very short in duration, about 60 milliseconds or so.

I was getting a lot of flicker reports, and as I mentioned, searching for these anomalies was tedious. So I began looking at the analysis capabilities of Audacity. I noticed a Silence Finder plug-in and attempted to search for the anomalies using that tool. But Silence Finder is designed to find the kind of silence that one might find between tracks on an LP: very quiet for a second or so. Not surprisingly, Silence Finder didn’t find anything for me.

I noticed that Silence Finder is written in a specialized Lisp-like signal processing language known as Nyquist. So I had a look at the source code, which is included with Audacity, and was able to understand quite a bit of what was going on. For efficiency reasons, Silence Finder down-samples the input data so it doesn’t have to deal with as much data. In order to search for shorter anomalies, I needed to change that, as well as the user interface limits on minimum silence duration. Also, the amplitude of the silence was expressed in dB, which makes sense for audio but I needed more sensitivity to subtle changes in amplitude. So I changed the silence amplitude from dB to a linear voltage value.

The result was quite helpful. The modified plug-in, which I called “Glitch Finder”, was able to quite reliably find voltage anomalies. For example:

The label track generated by Glitch Finder points out the location of the anomalies (at 17:05:12, 23:00:12, and 7:17:56 the next morning), although they’re not visible at this scale. Zoom in a few times and they become quite obvious:

Thus far I have reached these tentative conclusions:

• The power problems are primarily common to the neighborhood, and unlikely to be caused by a local load transient such as plugging an electric car in.
• They seem to be concentrated mainly in the evening (4-11 pm) and morning (6-10 am). These seem to be times when power load is changing, due to heating, cooking, lighting, and home solar power systems going off and on at sunset and sunrise.
• The longer term voltage goes up or down a bit at the time of a power anomaly. This requires further investigation, but may be due to switching activity by the utility.

Further work

As usual, a study like this often raises new questions about as quickly as it answers questions. Here are a few that I’m still curious about.

1. What is the actual effect on lights that causes people to notice these anomalies so easily? I currently have an oscilloscope connected to a photoelectric cell, set to trigger when the lights flash. It will be interesting to see how that compares with the magnitude of the anomaly.
2. Do LED lights manifest this more than incandescent bulbs? It seems unlikely that such a short variation would affect the filament temperature of an incandescent bulb significantly.
3. Do the anomalies correlate with any longer-term voltage changes? My test rig measures long-term voltage in an uncalibrated way, but the processing I’m currently doing doesn’t make it easy to look at longer-term voltage changes as well.