Data driven load modeling and customer behavior change detection
Nurmiranta, Mikko (2017)
Nurmiranta, Mikko
2017
Automaatiotekniikka
Teknisten tieteiden tiedekunta - Faculty of Engineering Sciences
This publication is copyrighted. You may download, display and print it for Your own personal use. Commercial use is prohibited.
Hyväksymispäivämäärä
2017-06-07
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:tty-201705261533
https://urn.fi/URN:NBN:fi:tty-201705261533
Tiivistelmä
The strive for more efficient delivery of electricity and management of production resources in order to control costs and emissions has lead to innovations in power engineering and combined with input from other fields smart grids were developed. More variety and flexibility in production, better customer engagement and improvements in delivery reliability are just some perks offered by smart grids. In particular the vast amount of timely data collected from the customers enables massively improved network state analysis which is necessary for cost-effective distribution. The accuracy needs to be maintained and the detection of changes in consuming habits is useful for this purpose.
In this thesis a method for detecting changes in consumption patterns is proposed. By building a forecasting model for the consumer and comparing the load forecasts with load measurements changes not caused by external factors such as outdoor temperature should be possible to be detected. Two models, a periodic autoregressive model and an artificial neural network are used, with the former coming out as more accurate and simpler to use. The relationship between model forecasts and load measurements is monitored with cumulative sum and Pearson divergence and based on test cases the changes in load shapes can be detected and especially the method based on Pearson divergence achieves promising results. However forecasting the load of individual customers appears to challenging and better models are needed, even if the actual goal of change detection is accomplished.
In this thesis a method for detecting changes in consumption patterns is proposed. By building a forecasting model for the consumer and comparing the load forecasts with load measurements changes not caused by external factors such as outdoor temperature should be possible to be detected. Two models, a periodic autoregressive model and an artificial neural network are used, with the former coming out as more accurate and simpler to use. The relationship between model forecasts and load measurements is monitored with cumulative sum and Pearson divergence and based on test cases the changes in load shapes can be detected and especially the method based on Pearson divergence achieves promising results. However forecasting the load of individual customers appears to challenging and better models are needed, even if the actual goal of change detection is accomplished.