A Stochastic Version of the Dynamic Linear Programming Model MESSAGE III

Golodnikov, A., Gritsevskii, A., & Messner, S. (1995). A Stochastic Version of the Dynamic Linear Programming Model MESSAGE III. IIASA Working Paper. IIASA, Laxenburg, Austria: WP-95-094

Preview

Text WP-95-094.pdf Download (576kB) | Preview

Abstract

The ECS Project at IIASA is using a dynamic linear programming model, MESSAGE III, for the analysis of long-term energy strategies to mitigate climate change. These model analyses utilize information on potential future technology characteristics, energy services demands and resource availability to investigate paths into a sustainable future energy system during the next century.

One major shortcoming of conventional energy optimization models is the requirement to use point estimates for the technology characteristics and other important system parameters. This paper introduces a new approach to overcome this problem by introducing distribution functions for technology parameters into model formulation. The stochastic version of MESSAGE captures the risk of underestimating future technology costs. Computational overhead for applying the approach is very low compared to the original model; for the runs investigated here, no increase in CPU time was detected.

Another problem often encountered with deterministic models, in particular linear programming models, is their sensitivity to input parameters. Linear programming models are, by definition, worst in this respect because they tend to favor single solutions and extreme developments instead of mixing various technologies or strategies. In such cases modelers develop robust scenarios by parameter variations and delimiting the solution space to the area that yields acceptable and robust results. Clearly, this approach is rather labor intensive and requires experience on the modeler's side. The important parameters for variation have to be found, and model runs with combinations of such parameter variations have to be performed. Such investigations can result in a large number of model runs or, more probably, in model outcomes that are not robust with respect to small changes in model parameters. In any case additional constraints introduced to stabilize model results reflect experts expectations, which are always subjective to a certain degree, representing a certain risk of over- or underestimating the parameters.

The stochastic approach presented here helps to overcome these problems by explicitly introducing the uncertainties concerning expert's opinions of future investment costs of technologies. The data used for the distribution function of investment costs were derived from the greenhouse gas mitigation technology inventory, called C02DB, developed by ECS over the last five years.

The strategies derived with the stochastic approach possess the required technological diversity without exogenous flexibility constraints. They also have a more robust structure with respect to present uncertainties concerning future parameters. Thirdly, the strategies derived with the stochastic model extension are less costly than strategies obtained on the basis of a purely deterministic model.

Actions (login required)

View Item