wiki:public/20122011

Project: Quasi Real-Time Individual Customer Based Forecasting of Energy Load Demand Using In Memory Computing

Team: Prof. Witold Abramowicz, Dr. Monika Kaczmarek, Dr. Tomasz Rudny, Wioletta Sokołowska

Research institution: Department of Information Systems, Poznan University of Economics, Poland

Abstract: The European energy markets are facing many challenges, e.g., usage of renewable energy sources, application of smart metering and, consequently, the emergence of new players in the market (e.g., of prosumers). However, one of the biggest challenges for the energy sector is still how to accurately predict a short- and long-term value of energy demand, as well as the level of energy production from different sources. In our project, we focused on a short term prognosis of energy demand. The quality of the demand forecast methods depends first of all, on the availability of historical consumption data, as well as on the knowledge on the important influence variables and their (forecasted) values. Another very important factor is the application of an appropriate forecasting tool and methods taking into account the data that we have, as well as existing limitations (e.g., computational).

Most of the existing approaches to energy demand prognosis focus on characterising aggregated electricity system demand load profile. However, the total energy load forecasting can be done more accurately, if forecasts are calculated on the lowest level, i.e., for all customers separately and then combined via bottom-up strategy to produce the total load forecast. Also, additional improvement is expected, if different models and parameters can be tested quickly (including the prognosis of the prices for the energy balancing market) to identify the best-fitted model for an individual time series. This quasi real time load forecasting could lead to substantial savings for companies, up till now however, it was computationally difficult, as for a typical energy seller the number of residential customers exceeds hundreds of thousands.

Within the conducted project, we focus on a time series approach in order to predict the short-term energy demand value. We apply the time series methods at an individual dwelling level. We formulate a hypothesis: forecasting accuracy (which is directly related to costs) can be substantially increased by the application of various individual forecasting methods on the level of each customer performed in quasi real-time using SAP HANA parallelism, and new insights into the analysed data can be gained. The additional goal of the project is to evaluate efficiency and performance of other computational strategies. Also, the possibility of adding different exogenous variables to the models and being able to quickly analyse their significance opens up new applications of energy load forecasting.

Last modified 6 years ago Last modified on Apr 9, 2013 11:08:14 AM

Attachments (1)

Download all attachments as: .zip