Forecasting

Forecasting is a key part of revenue management systems. You need to know how many customers of each type you should expect, so you can tailor the set of products being offered to maximize revenue.

In PassengerSim, forecasting is included as a step within an RM system, typically after untruncation and before any optimization.

E.py

from passengersim.rm.emsr import ExpectedMarginalSeatRevenue
from passengersim.rm.standard_forecasting import StandardLegForecast
from passengersim.rm.systems import RmSys, RmSysOption, register_rm_system
from passengersim.rm.untruncation import LegUntruncation

@register_rm_system
class E(RmSys):
    actions = [
        LegUntruncation,
        StandardLegForecast.configure(  #(1)!
            algorithm=RmSysOption("forecast_algorithm", default="additive_pickup"),  #(2)!
            alpha=RmSysOption(
                "exp_smoothing_alpha", expected_type=float, default=0.15
            ),
        ),
        ExpectedMarginalSeatRevenue.configure(
            variant=RmSysOption("emsr_variant", default="b"),
        ),
    ]

1. The forecaster (in this example, a standard leg forecast) is included as a step here. Selected options are passed to it via the configure method, which allows options to be set from the RM system configuration.
2. The configurable option name for the RM system is specified here as "forecast_algorithm", to clarify its purpose. The attribute being controlled on the StandardLegForecast is simply algorithm, which is clear within the context of forecast step alone, but potentially ambiguous in the context of the RM system as a whole.

StandardLegForecast

Bases: RmAction

Standard leg-level demand forecasting tool.

algorithm instance-attribute

algorithm = algorithm

Forecasting algorithm.

There are several available forecasting algorithms:

additive_pickup is an additive pickup model, which 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 can safely be omitted when using this algorithm.

exp_smoothing is an exponential smoothing model. This model uses the alpha parameter to control the amount of smoothing applied. It does not (currently) incorporate trend effects or seasonality.

multiplicative_pickup is a multiplicative pickup model. This model is in development.

alpha instance-attribute

alpha = alpha

Exponential smoothing factor.

This setting is ignored if the forecast algorithm is not "exp_smoothing".

StandardPathForecast

Bases: RmAction

Standard path-level demand forecasting tool.

algorithm instance-attribute

algorithm = algorithm

Forecasting algorithm.

There are several available forecasting algorithms:

additive_pickup is an additive pickup model, which 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 can safely be omitted when using this algorithm.

exp_smoothing is an exponential smoothing model. This model uses the alpha parameter to control the amount of smoothing applied. It does not (currently) incorporate trend effects or seasonality.

multiplicative_pickup is a multiplicative pickup model. This model is in development.

alpha instance-attribute

alpha = alpha

Exponential smoothing factor.

This setting is ignored if the forecast algorithm is not "exp_smoothing".