Multivariate techniques for the analysis of partial equilibrium energy models results

The energy sector is of outstanding importance for defining sound planning strategies on supranational, national and local scale, because of its direct and indirect consequences on environment as well as on socioeconomic development. In this framework, a valuable tool is represented by the TIMES energy models generators, developed under the aegis of IEA, that allow a comprehensive representation of complex energy systems based on a linear programming approach, individuating the optimal configuration of energy activities that fulfill the commodities demand at the minimum feasible cost. The solutions provide a large amount of results concerning energy consumption, levels of utilisation of technologies, emissions of main atmospheric pollutants, equilibrium prices of energy resources, etc. Some of the investigated parameters are usually combined in trade-off curves to point out the effects of variations of exogenous boundary conditions on energy systems and to derive the best suited solutions for the achievement of prefixed targets (scenario analysis). However, most of the information contained in the solutions are not completely exploited and there is a need of individuating the parameters with the high informational content, in order to derive "the robust strategies" and the priorities for the implementation of the energy technology roadmap, taking also into account the social acceptance of the envisaged measures. In this context, a powerful method to investigate in depth the data correlation structure is represented by multivariate statistical procedures, currently used in many environmental studies. In fact, the evaluation and characterization of the correlation patterns underlying a multidimensional data set represents an important step for data analysis and their interpretation. These statistical methods, based on the analysis of [samples x descriptors] matrices, allow to reduce the original data set without losing important information (ordination or dimension reduction methods), to point out homogeneous subsets samples or descriptors (variance analysis, discriminant analysis, clustering) and to characterise the data structure (canonical correlation analysis, non parametric multiscaling). Here we present an innovative application of multivariate techniques (Cluster analysis and Principal Component Analysis) for the interpretation of the Business As Usual - BAU scenario results for 28 EU country models, obtained in the NEEDS project (VI Framework Programme, Priority 6.1: Sustainable Energy Systems). In this application, data matrices are made up by electric energy and heat production, fuels import and consumption, renewable use, CO2 emissions by country and by time period (time horizon 2000-2050, subdivided into ten time periods). The aim of the analysis is to derive a general applicable procedure for characterizing information redundancy and data correlation structure, in order to identify sustainability indicators, to support decision making processes for complex energy systems.