As an example, I will use my system as an example. The PVI-5000-OUTD-AU efficiency plot is given below.

Aurora Efficiency Plot

Panel: TH175M24
Pmax: 175watts
Vmp: 36.2V
Imp: 4.85A
Voc: 43.9V
Isc: 5.30A

It can be seen from the inverter efficiency diagram that the sweet spot is at 50% full power and 345V string voltage. Note that it has 2 MPPTs. These may be independently specified.

Now the 28 panels for 4.9kW need to be split between the MPPTs such that the Pmax string voltages are near the optimal 345V. This gives us 2 optimal and a symterical choices:

1. 2 x 9 Panels in parallel + 1 x 10 panels. This gives string voltages of 325.8V and 362V.
2. 2 x 10 panels in parallel + 1 x 8 panels. This gives string voltages of 362V and 289.6V.
3. 2 x 7 panels in parallel + 2 x 7 panels in parallel. This gives string voltages of 253.4V

The open circuit string voltage for this inverter must be less than 600V. This is 439V for 10 panels, so this parameter is ok.

By inspection of the efficiency plot, configuration 1 is the best. In fact is it 0.3% better than the symmetrical configuration, which would result in about $10 per year more income/power at 68c/kWh.

The MPPT which has 2 strings in parallel has a greater weight attributed when choosing panel configuration as it supplies more power.

Refs:
Aurora PVI-5000-OUTD-AU Datasheet
TH175M24 specs

All things going well, we should be generating power and income in June some time, fingers crossed. Hopefully AGL wont stuff us around like others I know when we apply for the premium rate tarrif too.

4.9kW PV Panel Install with 28 x 175W Mono

Researching udev a bit, I found /lib/udev/rules.d/60-persistent-serial.rules (mine is a debian system). This file shows how the standard symlinks are done. Since each USB port is unique, I should be able to use that uniqueness to map another symlink to the device.

Firstly plug in the device in the chosen USB port and issue
udevadm info --query=all --name=/dev/ttyUSB1

This shows a heap of stuff but mainly we are interested in
P: /devices/pci0000:00/0000:00:1a.0/usb2/2-1/2-1:1.0/ttyUSB1/tty/ttyUSB1

Create a file in /etc/udev/rules.d/70-persistent-serial.rules which contains

#see /lib/udev/rules.d/60-persistent-serial.rules

ACTION!=”add|change”, GOTO=”persistent_serial_end”
SUBSYSTEM!=”tty”, GOTO=”persistent_serial_end”
KERNEL!=”ttyUSB[0-9]*|ttyACM[0-9]*”, GOTO=”persistent_serial_end”

IMPORT=”usb_id –export %p”
#IMPORT=”path_id %p”

ENV{ID_SERIAL}==”", GOTO=”persistent_serial_end”

# usb nearest ethernet connector
ENV{DEVPATH}==”*usb2/2-2/2-2:1.0*”, SYMLINK+=”serial/by-name/envi”
#bottom front connector
ENV{DEVPATH}==”*usb7/7-1/7-1:1.0*”, SYMLINK+=”serial/by-name/rs485″
# usb below nearest ethernet connector
ENV{DEVPATH}==”*usb2/2-1/2-1:1.0*”, SYMLINK+=”serial/by-name/rs485″

LABEL=”persistent_serial_end”

Replug and voila you get /dev/serial/by-name/rs485 which will always be the correct device.

Now I just have to run the wire to where the GCI will be installed.

A better but not intuitive explanation is that the mechanical vibration of each unit alters the behaviour of the other, increasing the total VA (since that’s what the current sensor measures). The 2 units are located next to each other so are mechanically coupled through the floor.

Of course this explanation could be wrong too but I can’t think of another ATM.

I had read about the 144 Bug. I suspect it is related to this.

This has to do with the way that the transmitter samples the current sensor. I have noticed that the resolution is much better at lower power than higher power. I suspect this is because the ADC reference in the PIC is connected to a PWM output (smoothed of course) to provide a variable dynamic range sampling system. As you would expect, there would be quite a few range change points in the sampling algorithm. The 144 bug details one at 3kW. I suspect there is also one at somewhere between 450W and 600W which is not calibrated correctly. There are probably others too.

If these range change points are known, it would be easy enough to add fudge factors during datalogging to correct for this. Alternatively I could disassemble the transmitter PIC code and just fix it.

No time for this now.

Another possibility is that the sampling system is non-linear, increasing apparent measurements. Not enough calibrated sample points to prove. Anyhow the former is more likely.

Fridge + Freezer Power Usage

As can be seen, the base load is about 320W which consists of 2 PCs and incidental standbys. The fridge cycles more frequently than the freezer. The fridge power alone is about 75W and the freezer alone is about 100W. You would thus expect their combined power to be 175W. However, this is not so! Measured combined power is about 250W.

Given that the mains power monitor is power factor insensitive, and the compressors of both devices have inductive power factor, this result is unexpected. I’m also assuming that the power factor of the baseload is 1.0, which I think is reasonable as using another power meter showed a power factor of 1 for the PCs and these predominate the baseload.

I would expect the vector sum of fridge+freeze+baseload to be less than the sum of their magnitudes.

I had a play with the linux web software that others had done before me.

I implemented a MQTT server and used a modified perl script to write to it first, as others had done in the UK, but these people were mostly IBM employees and thought it was a good idea to have a middle man feeding data to multiple consumers. It occurred to me that this may be a bit of an overkill. All the consumers also used rrdtool in some way to present the data.

Then I decided to get Danny Tsang’s energy@home project and after a few mods, got it going here.

Please don’t sit on it or I’ll have to password protect the page.

The benefits as I see them are:

Nicer looking graph

data is sent raw and the graphing is done locally

possibility to feed new samples without uploading the whole graph each refresh period

less load in the server

There are some things I still want to change, including experimenting with rrdtool, but displaying using jquery/flot which generates some very nice looking graphs.

I also need to daemonize the data acquisition bit, probably using daemon and some logging to wrap the datalogger.

There were a couple of omissions in the install details as I had to get some jgrid files to make it all work.

The temperature shows what it is at the server/inside as that is where the monitor it.

The transmitter is quite powerful and can blast over 10m from within full a metal enclosed fusebox. I suspect the 433MHz RF is coupling on the house wiring, otherwise it probably wouldn’t transmit very well. I originally extended the transmitter antenna, expecting it not to work very well, but found that it was unnecessary.

Fairly basic display. Very easy to setup.

I measured the power consumption of the display with my plug in power meter from jaycar and it read 0W. Which means the display is fairly well designed.

Haven’t got the USB cable yet, so I haven’t implemented logging, which is why I got it.

It looks like the dynamic range is variable with more sensitivity at lower power levels than higher ones. Because the transmitter uses a PIC 16F689, which has a 10bit A/D, it has only so much accuracy. So what they must have done is to use one of the PWM outputs to generate the reference, and vary this to provide the dynamic ranging capability. This is all done in the transmitter. It also means that the 288W bug mentioned in http://blog.technicaone.net/post/2010/01/26/Current-Cost-5c-Watts-Clever-The-144-Bug.aspx is also fixable in the transmitter.

Since there is a programming header on the transmitter (visible from the battery compartment), I wonder if they forgot to blow the security bits, allowing the program to be read? But that’s for later.

The pinouts are

Pin 1 +V (3V)
Pin 6 GND
Pin 7 TTL OUT (3V)


Refs:
http://mungbean.org/blog/?p=477
http://e.inste.in/2008/06/15/interfacing-the-currentcost-meter-to-your-pc/