My MATLAB SA-GA reproduction pipeline

My current VRP reproduction work uses MATLAB because the original experiment logic, data processing, and result reproduction are close to the academic workflow I am rebuilding.

The pipeline is simple in structure but strict in checking:

• Load instance data
• Build depot, customer, and charging-station sets
• Initialize a feasible or repairable solution
• Apply SA, GA, or hybrid search components
• Repair or reject infeasible structures
• Decode the final route
• Calculate objective and constraint-related outputs
• Export route map, convergence curve, and summary table
• Compare the output with the paper-level expectation
• Record discrepancy instead of hiding it

Current status

The code can run and produce feasible solutions, but paper alignment is not something I assume automatically. When the result does not match the reported table, I treat it as an audit problem rather than immediately tuning parameters.

Likely causes of discrepancy usually include:

• different initialization logic
• different penalty scaling
• ambiguous EV charging sequence
• hidden repair behavior
• inconsistent fitness-function interpretation
• result table not matching the plotted solution

The hard part is not drawing the final route. The hard part is making sure the route, cost, constraints, curve, and table are all describing the same solution.