A demonstration of how machine-learning can be used to predict total electricity output from a supplier's solar cell capacity when predicted rainfall and maximum temperature for a particular day are available.
To calculate the daily global solar exposure (in MJ/m2) at a particular place when predicted rainfall and maximum temperature data for a particular day are available. Once the solar exposure information is calculated, since values of total kW of solar capacity installed, angle of tilt of the solar panel and efficiency of conversion are available with the supplier, the total electricity output for that day can be calculated.
All the data used in this task are downloaded from Australian Bureau of Meteorology http://www.bom.gov.au/climate/data/ Data pertains to Canberra ACT (35.30S, 149.13E) as recorded at station 070351 Canberra Airport, NSW for the period 19/9/2008 - 2/5/2017
Ideally, I would have loved to build this model using data for number hours of sunshine on a particular day, or, cloud cover in a particular day along with maximum temperature reached during the day. However, data related to number of hours of sunshine/cloud cover was available only for 13 months, which would not have been sufficient to develop the model. Other option was to make a special request to get the data from Bureau of Meteorology which would have cost and time associated with it. I have used rainfall as a measure of lack of sunlight over the course of the day. Admittedly, even if rainfall is zero, but there is cloud cover, less power would be generated by the solar cells, which, the model I have developed would not be able to capture.
- Variables used are: ** Solar exposure (dependant variable) ** Rainfall (independent variable) ** Maximum Temperature (independent variable)
- Execute multiple linear regression using numpy
- Future values of solar exposure are predicted for a period of 7 days from 5/5/2017 to 11/05/2017.
- Graph is plotted using pylab.
Results of the exercise show that there is statistically significant relationship between the variables being analysed, which can be used for future predictions of solar exposure, given predicted maximum temperature and predicted rainfall data.