AI and Power Engineering - Part 4 | DIgSILENT Interface and Project / Protection Audit

DIgSILENT Protection Audit

The other week I showed a simple example of an AI creating a PSCAD model of an AC1A AVR in PSCAD. It was fun, but perhaps not overly practical. This week, I have put together a more interesting real world application of AI, using it to connect to an existing network study carried out in DIgSILENT Powerfactory and analyse the system configuration and protection settings and then to carry out a protection audit. Auditing a project is one of those things that we all know we should do, but can be a bit of a tedious process, which is an ideal task to delegate to an AI. This particular example is from the day job with my Aurora hat on.

This first step of the audit is to carry out a sweep on the system to identify all the cables, transformers, generators and system loads as well as understand the network topology, identify N/O points, parallel redundant systems and any network constraints. This is self explantory but a key step in making sure the AI understands the system it is looking at before offering opinions on the suitability of the protection.

The second stage was to carry out a protection audit on the system. That is I used the AI to sweep through all the protection elements (Relays, CTs, LV trip units, fuses etc.) and then cross check them against the network element they were set to protect. This involves check the overcurrent and earth fault pickup elements against transformer ratings, feeder ratings using industry standards rules. Then to check coordination between the relays for maximum and minimum short circuit levels, and identify any areas where the grading margin is unsuitable.

Audit Output

This is an example of where the AI approach really starts to earn its keep. Protection coordination studies, are a slow, careful and complicated process, that often requires checking, rechecking and back tracking on settings. It is easy to get lost in fields of data on the more complex relays. The protection audit approach allows you to take a step back and really check a project for a final time.
I have included the audit report to the post for interest (with key parts anonymised), to show the kind of things that can be done relatively easily once you understand how to harness an AI and interface it to the analysis software. The audit is by no means perfect and it has missed a few subtleties of the protection scheme, where it necessary to tweak settings to get better overall grading - which is an inherently subjective process. The rules I had developed for the AI and audit parameters are still a work in progress, but still reasonably effective. Download the example protection audit report (PDF)

Summary

The key takeaway from this exercise is to see how an AI, with the correct interface and rule structure can really start to shine at a commercial level. With a bit of care and setup it can automate and review a whole system and check for inconsistencies within protection. It needs a bit of refinement this tool is a serious development, as it allows an engineer to inspect and assess large networks and audit them against pre-defined rulesets and look for weak parts of the system. It does not replace the human, it just helps them spot things that we might otherwise miss.