In CPVRP, there are two objectives to be optimized: number of vehicles and total distance covered by all vehicles. The work first conducts an independent multi-objective experiment of either algorithm on a CVRP dataset. However both algorithms are known for their strengths and weaknesses therefore, a new approach that combines both strategies is used to overcome their disadvantages. Using the hybrid approach, a population based search algorithm (Genetic Algorithm (GA)) is used to optimize both objectives using Weighted Sum (WS) fitness measure, by placing a bigger weight on the number of vehicles. This biases the search towards optimizing number of vehicles. Routes for the best-evolved solution produced by GA are further optimized using a local search algorithm (Simulated Annealing (SA)) as a post optimizer to move distance discovered by each vehicle in GA to its local optima. When compared to GA and SA, the hybrid strategy recorded improved solutions however was statistically insignificant using a t-test with confidence level 0.05.
I developed the entire system was developed in Java
see source code on: