--- title: "Iterative process of model calibration" subject: "Simulation and Modeling" chapter: 7 key: "semester_8/simulation_and_modeling:calibration-iterative-process" source: https://www.bctengineer.com/q/semester_8/simulation_and_modeling/calibration-iterative-process site: BCT Engineering --- # Iterative process of model calibration Simulation and Modeling · Chapter 7: Verification and validation of simulation models - Asked in: 2073 Magh, 5 marks; 2071 Bhadra, 5 marks; 2070 Magh, 5 marks; 2070 Bhadra, 3 marks (asked 4 times) - Answer checked by Claude Opus 5.5 (high effort) on 2026-10-14 - Page: https://www.bctengineer.com/q/semester_8/simulation_and_modeling/calibration-iterative-process - Whole chapter: https://www.bctengineer.com/md/semester_8/simulation_and_modeling/chapter-7 ## Question Explain the iterative process of calibrating a model with example. ## Short answer Calibration iteratively compares model output with real system, identifies discrepancies, and adjusts parameters or distributions until the difference is acceptable. For example, a bank model initially underestimated wait; changing the service time distribution and adding a loan desk brought wait within tolerance. ## Answer **Calibration** is the process of comparing the model with the real system, then adjusting the model to remove the differences, and repeating until the model is accepted as accurate enough. It is an iterative "compare, revise, compare again" cycle (Banks, Carson, Nelson & Nicol). ```text Real system Initial model | | +----> compare <-----+ | discrepancies found? |yes |no revise the model accept model | (first revision) -> compare again | (second revision) -> compare again ... ``` **Steps** 1. Build the first version of the model from observations and assumptions. 2. Compare the model's behaviour with the real system, using subjective tests (people who know the system review the output) and objective tests (statistical comparison of outputs, e.g. t-test or confidence intervals on means). 3. Find the discrepancies and their causes (wrong distribution, missing detail, wrong parameter). 4. Revise the model: change the input distributions, parameters, or logic. 5. Compare the revised model with the real system again. 6. Repeat until the discrepancy is within the accepted limit. Stop when the cost of further revision is greater than the benefit. **Example: bank with two tellers.** The first model assumes exponential service time with mean 4 minutes. The real bank shows an average customer wait of 6 minutes, but the model gives 3.5 minutes. On inspection, real service times vary less than exponential, and some customers use a special loan desk. In the second version, service time is changed to a normal distribution (mean 4, standard deviation 1), and the loan desk is added. The model now gives a wait of 5.4 minutes. After the arrival rate during lunch hour is corrected in the third version, the model output is 5.9 minutes, which is within the 10% tolerance of the real value. Calibration stops here. Calibration should be followed by validation on a separate data set that was not used in calibration. ## Check yourself The answer's key points as multiple-choice questions, with the correct option and why. 1. What is the basic cycle of calibrating a model? A. Build once, then validate on the same data B. Compare with the real system, find discrepancies, revise, compare again C. Debug the code, then run more replications D. Collect more data, then discard the model Answer: B. Calibration is an iterative compare-revise-compare loop that continues until the model is accurate enough. (Key point: Calibration cycle) 2. When does the calibration iteration stop? A. After exactly three revisions B. When the program has no coding errors C. When the discrepancy is within the accepted limit, or further revision costs more than it gains D. When the model output equals the real value exactly Answer: C. Calibration stops when agreement is within tolerance or when the cost of more revision exceeds the benefit. (Key point: Stopping rule) 3. Which pair describes the tests used to compare the model with the real system? A. Subjective review by system experts and objective statistical comparison (t-test, confidence intervals) B. Only subjective review by the programmer C. Only chi-square tests on the code D. Compiler checks and trace output Answer: A. Both subjective tests by knowledgeable people and objective statistical tests on outputs are used. (Key point: Comparison tests) 4. In the bank example the model gave 3.5 min against a real 6 min. What was done in the second version? A. Number of tellers was doubled B. The run length was increased C. The random seeds were changed D. Service time was changed to normal (mean 4, SD 1) and a loan desk was added Answer: D. Inspection showed less variable service times and a special loan desk, so the distribution and logic were revised. (Key point: Bank example) ## Related in Simulation and Modeling - [Compare verification, validation, calibration](https://www.bctengineer.com/question.md?key=semester_8%2Fsimulation_and_modeling%3Averification-validation-calibration-compare) — Simulation and Modeling, chapter 7 ## Related in other subjects - [Iterative and incremental model](https://www.bctengineer.com/question.md?key=semester_7%2Fict_project_management%3Aiterative-incremental-model) — ICT Project Management, chapter 4 - [Book store process model](https://www.bctengineer.com/question.md?key=semester_5%2Fsoftware_engineering%3Abook-store-process-model) — Software Engineering, chapter 2 - [Steady 2D model: seepage under a dam](https://www.bctengineer.com/question.md?key=semester_8%2Fcomputational_techniques_in_civil_engineering%3Aseepage-under-dam-2d-model) — Computational Techniques in Civil Engineering, chapter 7, asked 3× - [Steps in the design and synthesis process](https://www.bctengineer.com/question.md?key=semester_6%2Fmachine_design_i%3Adesign-process-steps) — Machine Design I, chapter 1