Alert/status emails

In alerts folder, see alerts/HOWTO.txt for details.

Drying models

Rate of change estimation

In Data\models\predict_roc.

perform_fit.py is the script that performs the fit and gives the constants and some basic stats. It's fairly heavily documented so shouldn't be too hard to figure out. Pretty easy to change the input variables, etc.

output_r_fit_results.py gives the same stats for constants found using R for the same fit. Pretty much the same results. Minor variation due to the specific fitting method used.

Looking at the results, it seems the a lot of the error is focused in a particular set of runs. It would be worth seeing if there's a reason for that, like bad measurements or wildly different weather conditions or such.

seed_results.csv is the result of the fit for the data available, giving the input/output variables, the prediction, and the error.

Drying time estimation

In Data\models\predict_drying_time.

There is one attempt that is essentially the same as for rate of change estimation. In fact, the fitting script is identical except for the input variables, output variables, and fit function. The fit isn't amazing. If you plot time_remaining and prediction from seed_results_remaining.csv you can see that it's sort of generally ok.

The solution I ended up investigating is to build a model of temperature over the course of the day. A sum of temperatures is probably a useful input variable, so using that should tighten up the predictions. However, that does mean that we need to also predict that. Intuitively, given how consistent the conditions in Cebu are, it should be possible to predict the temperature sum for the remainder of a day given the time of day and the temperature sum to that point. daily_cycle.py dumps out the values needed to test this idea and daily_cycle.csv contains the results for nodes 1-40. Note that only days with 100% yield are considered. Plotting a 3D graph as a quick check confirmed my assumption that it would be really hard to predict at the start of the day, and get easier as the day progressed. To demonstrate that a statistical model should work (maybe a multivariate gaussian?), temperature_sum_example.eps shows the remaining temperature sum over the rest of the day for all instances of sample_interval of 200-210 and current_temperature_sum of 8000-9000. The spread is pretty tight and looks reasonably gaussian.

Water evaporation

In Data\models\water_evaporation.

Again, essentially the same as the above but using the water bath weight data. Kinda boring as next rate of change can be predicted almost perfectly from the current rate of change.

what_is_all_this.txt has an overview of the important files.

Analysis

In Data\models\analysis.

A summary of the results from the various models.

Airflow zero point calibration

In Data\models\airflow.

See HOWTO.txt for more details.