Research Theme 3
Improve the system's endurance and flexibility when it comes to the infrastructure network and similar industries with the use of AI.
Did You Know?
We can make more electricity than the grid was designed to handle!
3.1
Enhance sturdiness and flexibility of the energy infrastructure.
3.2
Create a plan for uncertainties using AI in condition based maintenance (CBM) planning.
3.3
Find the best way to grant alternative energy resources along with any emergency planning.
Main Goals
Future Objectives
Construct
a framework to improve sturdiness and flexibility.
Design
software to discern different patterns.
Create
plans to improve the system.
Build
optimization hardware and software.
Produce
models using weather as impact data for CBM planning.
Generate
algorithmic differentiation and machine learning tools for visualization.
Approach
Building resilience is important for overcoming impacts and failures, so naturally we want our system to build resilience too. This includes things like system reliability and mitigation/restoration planning. This is important for our jurisdictions, as they are in a weak spot in the power grid, so if serious weather were to take out the power, lives could be lost. To enhance the system we have been building to be resilient, we are going to introduce something called Proactive Resilience Improvement over three phases, pre-disaster, during-disaster, and post-disaster.
Proactive Resilience Improvement is all about making sure energy related infrastructure and industries build that flexibility and durability. This will be done with our AI-based health monitoring system, dynamic maintenance, logistic planning, and with the use of alternative energy resources.
In the pre-disaster phase maintenance costs will be reduced due to the health monitoring system assessing the health and preventing failures.
​
In the during and post-disaster phases we can try to restore the state of the system with CBM planning and alternative energy resources.
Did You Know?
The #1 cause of power outages is weather!
Methodology
For the system to be able to make effective decisions, it has to understand the effects of each decision it generates. A tool that can iteratively learn from data to understand effects of actions is known as Deep Reinforcement Learning (DRL). This will allow us to create a system that can make strategic decisions to maintenance and resource allocation issues regarding economic and social impacts.
By integrating this technique with our AI system, we will have a system that looks at the health and predicted health of important nodes and related networks and chooses the most optimal solution.
The decision-making strategy will be based on mathematical formulas with a combination of DRL and optimization methods. This will be able to distribute the right amount of alternative energy to certain locations to maintain the networks.
​
We will accomplish going over the hurdles of interpreting sparse and heterogeneous data for decision-making purposes by integrating bus-branch algorithms with embedded parameter tuning. The steps we will take to do this are:
1) We will create Graph Convolutional Network
(GCN) models that show the spatial
associations in power systems topology.
2) Then we will formulate ways to model
uncertainties with Bayesian machine
learning.
3) We will develop a real-time co-simulation of
this AI-based energy management system
that can be distributed commercially.
Did You Know?
Bayesian machine learning is based off of Bayes' theorem, and is really useful for predictions!
Additional Steps
1) Be able to scale the computational algorithms while stabilizing speed and precision.
2) Be able to tell the difference between structural uncertainty and model uncertainty.
3) Be sure to embrace expert prior knowledge in optimization and action evolution.