Research Theme 2
Develop a monitoring and predicting system to keep track in real-time of the health of the infrastructure networks with the use of AI.
Did You Know?
Smart grids are armed with different smart technologies to allow for the transfer of data along with electricity!
Main Goals
Classify the smart grid and pipeline networks.
2.1
Deep dive into interdependencies at an embedded system level, and be able to distinguish malware to improve security for IoT devices.
2.2
Identify
the smart grid nodes with phasor measurement units (PMUs) and edge cases with data from smart inverters.
Classify
data from the pipeline networks when major natural events occur.
Examine
the interdependencies at an embedded system's level among the pipeline and the power grid.
Distinguish
malware to improve security for IoT devices.
Future Objectives
Did You Know?
Smart inverters may be used with smart grids to integrate different types of distributed energy resources (DERs) into the grid. (They are also able to make decisions to keep the grid stable!)
Approach
Fist we will take the the vulnerable clusters we found in Research Theme 1 and add real-time prediction to it to be able to find out when and why a failure happened. To be able to achieve this we will integrate AI in smart sensing devices that will be immersed in different infrastructure networks. For sensors we will be utilizing Internet of Thing (IoT) devices to help minimize the stress being put onto the centralized cloud.
In order to be able to collect the network topology that suggests failure, we will also be using different deep learning methods. Putting everything together, we will end up with a decision-theoretic framework that assesses the health of the energy network and possible failures by looking into our heterogeneous smart sensing data .
​
Being that consecutive power outages are triggered by line events, we need to be able to detect these events before any serious consequences arise. For this we will look into ways to partition the network into zones based on sensitivity. From there we will use graph theoretic deep learning methods to PMU data to find these events automatically.
​
By using the failure model from Research Theme 1, Objective 1, we can figure out relevant failure states for line events under a variety of failure conditions. We will be using additional synthetic data from the OPAL-RT real-time simulator at UARK's National Center for Reliable Electric Power Transmission to support the potential exclusions of serious situations. Once developed, the methods will be tested with real PMU measures from emergency situations through UARK's GRAPES member companies and Montana-Dakota Utilities.
​
When it comes to pipeline networks, there are at least five heterogeneous datasets:
1) Real-time Natural Force Data
2) Real-time Pipeline Condition/Operation Data
3) Historical Pipeline Inspection and
Maintenance Data
4) Public Report Data
5) Failure State Data From Pipeline System
Simulation Models
We will gather this data from various resources and map them to datasets for later classification.
Methodology
To create our decentralized AI-based IoT system we will have three different tiers that will each consist of a different level. The bottom tier will be the Local Decision-Making tier that will focus on the physical level of things. The next tier up is the Central Decision-Making Tier which will include the network level, and the last tier will be the Strategic Decision-Making tier with the operating level.
​
The physical level will be in charge of the state of the physical system, formulating local decisions, and answering to the system at higher levels. It will consist of the physical constituent parts and be integrated with the framework created from Research Theme 1. The data will be then used for to expose failures, predict them, and have a sense of awareness. The integration will also provide us with more analytical data from the centralized cloud. This will help with harmonizing of related nodes found from different interdependencies.
​
The network layer will be keeping track of any variations found in the health of the vulnerable nodes over a period of time.
​
The operating level will be the what focuses on implementing different deep learning algorithms to be able to predict failure, maintain a specified quality within the energy network, and will be the one giving responses in emergency situations. Failure will be determined by certain network topologies that do not match the usual topologies.
​
Lastly, to maintain security with the system and prevent false data from IoT devices, we will be creating methods to be able to detect malware on the system. This will be done on a Network File System (NFS) distributed server, General Parallel File System (GPFS) storage system, and with a OpenStack virtualization cluster at NDSU's High Performing Computer Center-CCAST.
​
Putting everything together, we will end up with a decision-theoretic framework that assesses the health of the energy network and possible failures by looking into our heterogeneous smart sensing data .
Did You Know?
There are 3 power grids in the US:
Western Interconnection
Eastern Interconnection
Texas Interconnected System