1.2 Why Simulate?

Imagine trying to analyze the following situation. Patients arrive at an emergency room. The arrival of the patients to the emergency department occurs randomly and may vary with the day of the week and even the hour of the day. The hospital has a triage station, where the arriving patient’s condition is monitored. If the patient’s condition warrants immediate attention, the patient is expedited to an emergency room bed to be attended by a doctor and a nurse. In this case, the patient’s admitting information may be obtained from a relative. If the patient does not require immediate attention, the patient goes through the admitting process, where the patient’s information is obtained. The patient is then directed to the waiting room, to wait for allocation to a room, a doctor, and a nurse. The doctors and nurses within the emergency department must monitor the health of the patients by performing tests and diagnosing the patient’s symptoms. This occurs on a periodic basis. As the patient receives care, the patient may be moved to and require other facilities (MRI, X-ray, etc.). Eventually, the patient is either discharged after receiving care or admitted to the main hospital. The hospital is interested in conducting a study of the emergency department in order to improve the care of the patients while better utilizing the available resources. To investigate this situation, you might need to understand the behavior of certain measures of performance:

• The average number of patients that are waiting.

• The average waiting time of the patients and their average total time in the emergency department.

• The average number rooms required per hour.

• The average utilization of the doctors and nurses (and other equipment).

Because of the importance of emergency department operations, the hospital has historical records available on the operation of the department through its patient tracking system. With these records, you might be able to estimate the current performance of the emergency department. Despite the availability of this information, when conducting a study of the emergency department you might want to propose changes to how the department will operate (e.g. staffing levels) in the future. Thus, you are faced with trying to predict the future behavior of the system and its performance when making changes to the system. In this situation, you cannot realistically experiment with the actual system without possibly endangering the lives or care of the patients. Thus, it would be better to model the system and to test the effect of changes on the model. If the model has acceptable fidelity, then you can infer how the changes will affect the real system. This is where simulation techniques can be utilized.

If you are familiar with operations research and industrial engineering techniques, you may be thinking that the emergency department can be analyzed by using queueing models. Later chapters of this book will present more about queueing models; however, for the present situation, the application of queueing models will most likely be inadequate due to the complex policies for allocating nurses, doctors, and beds to the patients. In addition, the dynamic nature of this system (the non-stationary arrivals, changing staffing levels, etc.) cannot be well modeled with current analytical queueing models. Queueing models might be used to analyze portions of the system, but a total analysis of the dynamic behavior of the entire system is beyond the capability of these types of models. But, a total analysis of the system is not beyond simulation modeling.

Simulation may be the preferred modeling methodology if the understanding gained from developing and using a simulation model is worth the time and cost associated with developing and using the model. Good uses of simulation include:

• Understanding how complex interactions in the system effect performance.

• Understanding how randomness effects performance.

• Comparing a fixed set of design alternatives to determine which design meets the performance goals under which conditions

• Training people to prepare them for dealing with events that may be disruptive to the actual system.

• The model will be used repeatedly for decision making.

• When the decision associated with the problem has a high cost so that the cost of building the model and evaluating the design is worth its development.

• When the current system does not yet exist and you need to ensure that the chosen design will meet specifications.

Simulation modeling activities encapsulate all three major modeling methods of data analytics: descriptive, predictive, and prescriptive. Descriptive modeling uses historical data to describe what happened in order to understand past behavior of a system. Predictive modeling uses historical data to develop models that help us understand future behavior in order the answer what may happen. Descriptive modeling summarizes past data for understanding. Predictive modeling uses past data to predict future behavior. Prescriptive modeling indicates what should be done and is integral to answering questions involving system design.

A simulation model is both a descriptive and predictive model. In addition, when coupled with stochastic optimization methods or a rigorous design process that evaluates and recommends designs, a simulation model becomes an integral part of the prescriptive modeling process. A simulation model describes how a system works by encapsulating that description within the operating runs/constructs of the model. A simulation model uses descriptive models (input models, summary statistics). A simulation model predicts future system response. A simulation model can be used to predict future behavior through running what-if scenarios. Simulation is inherently a predictive modeling methodology. Unlike other predictive modeling techniques found in data analytics, such as regression, neural networks, random forests, etc., simulation explicitly incorporates domain knowledge into the modeling activity by incorporating system operating behavior. The behavior is modeled through the physical and logical rules that apply to the relationships between the components of the system. Unlike pure statistical predictive models, simulation has the advantage of explicitly representing relationships rather than just relying on discovering relationships. A key advantage of simulation modeling is that it has the capability of modeling the entire system and its complex inter-relationships. The representational power of simulation provides the flexible modeling that is required for capturing complex processes. As a result, all the important interactions among the different components of the system can be accounted for within the model. The modeling of these interactions is inherent in simulation modeling because simulation imitates the behavior of the real system (as closely as necessary). The prediction of the future behavior of the system is then achieved by monitoring the behavior of different modeling scenarios as a function of simulated time.

Real world systems are often too complex for analytical models and often too expensive to experiment with directly. Simulation models allow the modeling of this complexity and enable low cost experimentation to make inferences about how the actual system might behave.