Present State

1. Smart grids use old and antiquated methods to control load leading to millions in lost revenue due to them having to start expensive peeker plants

2. Control not taking into account stochastic nature of energy demand

3. Utility companies have to charge outrageous prices leading to unhappy customers

Desired State

1. Smart grids manage load and don't have to fire up peeker plants

2. Control based off bayesian statistics and machine learning

3. Utility companies maintain their revenue while customers experience a great quality of service

Kepner Tregoe Situation Analysis

Timing: With the advent of electric vehicle and various distributed energy resources, it is now more than ever important to find sound and novel ways to improve the load management of our energy grid.

Trend: Customers are interested in adopting better, cleaner sources of energy and beginning to accept the installation of data generating smart meters

Impact: Customers will be able to have a good quality of experience while utility companies will maintain their profits. This will lead to a cleaner and less wasteful energy grid for the planet.

Kepner Tregoe Problem Analysis

What: Current control methods for the smart grid are based on legacy control theory and optimization. We want to utilize machine learning as our control solution

When: The work by Geoffrey Hinton, Yoshua Bengio and Yann LeCunn has allowed the machine learning community to mature.

Who: Power companies and ratepayers

Where: United States, Europe

Why: Significant losses are incurred by applying a typical legacy control approach as opposed to a dynamic learning based approach

How: Utilizing Tensorflow, Keras, OpenAI Gym, Python, Pandas