December 22, 2011
The electrical power network fault response system simulates an electrical grid network. Grid faults are generated through a multimodal stochastic process. Agents respond to these events by diagnosing and repairing the network faults.
- A* search is used by the agents to navigate the simulated world.
- Weighted K-means and Bayesian inference are used to reduce the agent's diagnostic search space.
- Planning is used to generate restoration plans.
- Built a simulation environment to test different response strategies.
- Generated a realistic network topology from data stored in a punch card format.
- Linearized AC power flow and programmed a linear solver to compute the power flow across the network.
- Created a multimodal stochastic process to induce grid faults with pretend weather data, temperature, time of day, ect...
- Built network code to support multiple viewing clients.
Evaluation of Results
Negative results. The fault response system was no better than the naive approach of searching around the most affected area.
I should not build Sim City every time I want to look at some interesting facet. Some aspects of the project are undercompleted.
December 22, 2011 Attach:anthony_vardaro_ai_grid.pdf
Oodles of code: Attach:grid_sim.tgz