- Tensorflow v1.3
- http://tflearn.org/installation/
Produce a temporal prediction of overall substation power generation & consumption.
- Perform literature review
- Be able to predict the next hour's consumption/generation with >90% accuracy
- Be able to predict up to 12 hours from current period
- Missing Data? Use a polynomial fit of other points to fill it in
- Submodels for discrete power sources. (e.g. unique submodel for three different types of solar panels)
Report for each substation:
- Generation (single number in KW/h)
- Consumption (single number in KW/h)
- Accuracy
- Error
- Deviation and other statistics that we may find useful for validation
- Freq of 10 m, for one year
- wind direction & amplitude (No direct measurement from wind turbines)
- diesel generator output can be considered the demand, little buffer
- temperature of environment (affects power generation from solar and power consumption from populace)
- school days (consumes more power than usual)
- build dynamically so that more sources of energy can be added to model
- Pull any additional data we may need
- Angle of the sun
- Utilize momentum in calculations (SMA for the previous hours leading up to the estimation time)
- For lstm model: https://github.com/RobRomijnders/LSTM_tsc