Analyse und Prognose elektrischer Lastgangzeitreihen (English shop)

Preview

Table of Contents, Datei (43 KB)
Extract, Datei (110 KB)

In order to minimise the costs of covering demand for grid-bound energy carriers, energy supply and trading companies plan their energy provision in detail. As a result of the liberalisation of the energy markets and the consequent increase in the number of market players and transactions, it is now necessary to produce separate forecasts for sales and trading companies as well as for the electricity trading tasks of the network operator, with in part completely different requirements regarding methods and systems. This planning task will become increasingly important in the future and is decisive for the economic success of the companies. In operational practice, a large number of different approaches and procedures are used for load forecasting. At present, the method of multiple regression is the most widespread.

Within the scope of this work, forecasting models were developed on the basis of load profile data from different supply collectives. First, statistical investigations were carried out between the load profile data and the associated exogenous variables. Based on the insights gained from this, load forecasting models were developed using methods of classical time series analysis, methods of stochastic processes, and a combination of deterministic and stochastic approaches. Furthermore, the technique of neural networks was applied to load forecasting, since these exhibit a high learning capability and can also easily be trained for other fields of application. The type and topology as well as the number of neural networks were optimised as a function of forecast reliability and tested by means of case studies.