RM Systems
A revenue management (RM) system is defined by one or more steps. The steps tell the simulation which demand untruncation, demand forecasting, and optimization algorithms to use. These steps also provide information on the kind of forecast and optimization used (i.e., leg or path) and if path-level information should be aggregated to leg-level information before a step is performed. Finally, these steps allow the user to specify algorithm-specific parameters, e.g., when using exponential smoothing the user can specify the smoothing constant, alpha.
Each carrier should have an RM system that it uses. In PassengerSim, users have the ability to create a single RM system and assign it to all carriers, or to create multiple RM systems and assign different RM systems to different carriers.
Below is an example that defines 4 RM systems. It will be helpful to use these examples to understand the different step_types and options associated with each step_type.
rm_systems:
fcfs:
steps: []
rm_no_detruncation:
processes:
DCP:
- step_type: untruncation
name: untruncation
algorithm: none
kind: leg
- step_type: forecast
name: forecast
algorithm: additive_pickup
alpha: 0.1
kind: leg
- step_type: emsr
name: optimization
algorithm: emsrb
kind: leg
rm_em:
steps:
- step_type: untruncation
name: untruncation
algorithm: none
kind: leg
- step_type: forecast
name: forecast
algorithm: additive_pickup
alpha: 0.15
kind: leg
- step_type: fcfs
name: optimization
rm_probp:
steps:
- step_type: untruncation
name: untruncation
algorithm: em
kind: path
- step_type: forecast
name: path_forecast
algorithm: exp_smoothing
alpha: 0.15
kind: path
- step_type: probp
name: optimization
- step_type: aggregation
name: aggregate
- step_type: emsr
name: optimization
algorithm: emsrb
kind: leg
The first RM system is based on a first-come, first-serve approach (named fcfs).
No steps are defined for fcfs as there is no demand detruncation, demand
forecasting, or optimization done with FCFS. If step_types are defined when
fcfs is explicitly specified as the optimization algorithm, they will be ignored.
The second RM system, named rm_no_detruncation, is leg-based and uses an additive_pickup
forecasting model with EMSRb and no demand detrunction. The alpha parameter that
is specified in the forecast step_type will be ignored as it is not used for
the additive pick-up model.
The third RM system, named rm_em, is using the expectation-maximization (EM)
method of detruncation with an exponential smoothing demand forecasting approach
that has a smoothing constant of alpha of 0.15 with and EMSRb optimizer. Both the
forecasts and optimization are done using leg-level inputs.
Finally, the fourth RM system, named rm_probp, is also using the EM method of
detruncation with an exponential smoothing dmeand forecasting approach that has
a smoothing constant or alpha of 0.15. Unlike in rm_em however, the
untruncation and forecasting steps are done at the path level. The optimization
step is based on probp that first finds displacement costs at a path level, then
aggregates them to a leg-level in the aggregation step type, and finally calculates
protection levels using EMSRb with leg-level demand inputs.
Given an overview of how RM systems are constructed, let's now look at each step in detail.
Untruncation
There are three untruncation (also called detruncation) algorithms. The first,
em, is based on the expectation maximization method. The naive1 and naive2
methods are based on Shebelov presentation.
The untruncation steps can be performed at the path or leg level.
Forecast
- step_type: 'forecast'
algorithm: { additive_pickup, exp_smoothing }
alpha: float
kind: { leg, path, hybrid }
There are two forecasting algorithms: exponential smoothing (that uses a smoothing or alpha parameter) and additive pickup model. The exponential smoothing model does not (currently) incorporate trend or seasonality.
The additive pickup model generates a forecast by considering the "pickup", or the number of new sales in a booking class, in each time period (DCP). This model is additive in that the forecast of demand yet to come at given time is computed as the sum of forecast pickups in all future time periods. This forecasting model does not consider the level of demand already accumulated, only the demand expected in the future. The forecast is made considering the results from the prior 26 sample days. The additive pickup model ignores the value of the alpha parameter, and it should be omitted when using this algorithm.
Either forecast algorithm can be performed based on data collected at either the leg or path level. However, if the forecast is at the leg level then detruncation must also have been performed at the same level.
EMSR Optimization
The step_type emsr is used for algorithms based on the expected marginal seat revenue approach and also can be used for fcfs. (although the first RM system defined as fcfs in the example above is a cleaner way to specify the fcfs option).
ProBP Optimization
The step_type probp does just as the name suggests - uses the probabilistic bid price algorithm to determine path-based displacment costs. After these are found, two more steps are needed - the first is to aggregate path-level information to leg-level information and do the probp proration? and the next is to use the leg-level inputs to find protection levels using emsrb.