This project finds the optimal path to defend from random attackers in a pasture using Differential Evoulution.

The project is divided into three modules:

• pasture.py (Client) defines the geometrical constraints on the pasture.
• simulation.py(PatSi) defines the simulation and evaluation of a given path.
• main.py (PatOp) optimizes a path based on pasture.py and simulation.py using DE from scipy's implementation with best1bin recombination scheme.

Each path is modelled from a float vector as a Bezier curve, sampled uniformly and passed in to the simulation as a discrete sequence of points in a 2D space, bound by the constraints.

To use the project one must first define the in_pasture method in pasture.py and simulation in simulation.py and then run main.py. The warning about linear constraints can be ignored. If the message returned is Max iteration exceeded, the success: False can also be ignored. If there is a different message and success: False, something went wrong.

x is the vector of points defining the bezier curve in format (x1, x2, ..., xD, y1, y2, ...,yD) corresponding to fun, the optimal fitness reached during the run.

Tested on Ryzen 5 4500U (w/ 6 cores, 8GB RAM) which is weaker than i5-4590S, an average CPU 8 years ago (when averaging its higher clock speed together with its lower core count). If client uses multithreading, it might be more helpful.

• NP = 4: 9-11 (async, sync), 15% less computation time when parallelized.
• NP = 15: 19-34, 44% decrease.
• NP = 150: 180-300, 40% decrease.
• NP = 500: 1195-1601, 25% decrease.

• NP = 15: 557-2305 (9 minutes, 38 minutes), 75% decrease.

All benchmarks yielded fitnesses of around 1e-14, while the minimum is 0.