Page 157 - Data Science Algorithms in a Week

Simulation Optimization Using a Hybrid Scheme … 141

5,000 units
5,000 units
5,000 units
5,000 units

2,000 units
2,000 units
2,000 units
2,000 units
0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30
Time (week)
Finished Goods Inventory:-15% units
Finished Goods Inventory:-10% units
Finished Goods Inventory:-5% units
Finished Goods Inventory:+10% units

Figure 10. Behavior of Finished Goods Inv. due to changes in customer orders.

Stability returns approximately 10 weeks and 16 weeks after the system was
disturbed (response time) for -10% and -15% decrease in customer orders respectively.
Amplifications are on the order of 1% under the EPs for both -10% and -15% decrease in
customer orders.

CONCLUSION

We propose a hybrid algorithm to obtain a quick convergence of the ADE. This
algorithm is based on a search engine that combines the advantage of PSO optimization
to determine the most promising regions of the search space and the properties of PHC
algorithm to accelerate locating the optimum that makes the ADE to convergence.
Although it is not required to find the global optimum to obtain a satisfactory reduction in
instability, our hybrid algorithm provides solutions that escape local convergence and
lead to stabilization polices with few oscillations and fast stability. This broader search to
find more effective stabilization policies is also possible due to the fact that we
incorporate a theorem that allows finding the best equilibrium levels that minimize the
ADE.
We conclude that the convergence of the ADE generates stabilization policies that
are robust. To test robustness on these policies we produced a perturbation in the stable
system by changing the value of an exogenous variable. The results show that the
variables of interest reach new equilibrium points after a period of adaptation to the
alteration of the system. Moreover, perturbations generated by sudden changes produce