Energy Neural Network

A neural network written in Python, consisting of a single neuron that uses back propagation to learn the influence on weather and occupancy variables on energy consumption. Original code is from Milo Harper's "Simple Neural Network" project that shows how 3 input variables influence a single node.

Dependencies

1. numpy
2. xlrd

Data Sources

1. OARDC Hourly Local Weather History

2. The Ohio State University Academic Calendar

3. (Provided) Hourly Electrical and Steam consumption for 3 campus buildings

Data Organization

Data was organized using Excel to match hourly datasets together. YELLOW = Inputs. BLUE/GRAY = Outputs (note net trained on one output at a time).

Inputs

The Ohio State University Academic Calendar is organized on the left 6 input (YELLOW) columns using several IF statements that check to see if converted Julian dates for each date/time variable falls within regular class session, exams, or holidays/breaks. Many of these are broken down into simple 1's and 0's to be more clear about how their state influences the energy consumption.

OARDC Hourly Local Weather History takes up the remaining Input (YELLOW) columns, matching date/times for output variables.

Outputs

The information provided consisted of hourly electrical and steam meter readings. Below are examples of those readings from one of the buildings:

Data input

The data is imported into python using xlrd. Currently, for the most part it imports, although now just working through syntax and other data formatting nuances.