The study compared Gauss–Legendre, Midpoint, Trapezoidal and Simpson numerical integral equations for CUSUM ARL/ATS under a seasonal long-memory FISMAX model with exponential white noise.
Key findings
- Simpson yielded the smallest ARL/EARL values, followed by Midpoint, Trapezoidal and Gauss–Legendre. Midpoint had the lowest computational ATS/EATS and was preferred for the investigated scenarios.
Why this matters globally
The study informs computation for seasonal long-memory process monitoring by making value-runtime trade-offs visible.
Thai researcher contribution
A KMUTNB researcher developed numerical methods for process control and anomaly signaling.
Limitations to consider
Grid size, tolerance, benchmark error, hardware and full parameters are absent. Ordering estimates does not establish accuracy, and no industrial dataset was tested.