In English

Streaming-based data validation in Advanced Metering Infrastructures

Johan Swetzén
Göteborg : Chalmers tekniska högskola, 2015. 54 s.
[Examensarbete på avancerad nivå]

In recent years, the development of smart grids has moved forward both technologically and politically. A law (2012:510) introduced in Sweden in 2012, giving consumers the right to have their electricity charged by the hour, brought demands for a more data-intense metering infrastructure. The automatic measuring, often involving wireless communication, introduces errors both in software and during data transmission. Combined with increased data volumes that drive a further increase in errors, the new demands present a challenge to utilities. Measurement errors cannot be allowed to propagate to the data stored by utilities and, in the end, electricity bills. To handle these errors as well as enabling future real-time services, there is a need for validation of very large quantities of metering data with low-latency guarantees. The systems responsible for such validation need to handle the unbounded streams of data generated by an increasing number of meters as well as more ne-grained energy consumption readings in the future. Another important aspect is exibility, as these systems need to account for both current and future errors in the data, adapting to changes in the Advanced Metering Infrastructure (AMI). This thesis aims to address these issues using the data stream processing paradigm, in which complex analysis is performed on unbounded sequences of data. Since this fits the AMI data well, validation rules are built upon this paradigm and then implemented on an actual stream processing engine to assess performance. Furthermore, patterns of common errors are identified, triggering alerts as a first step towards automatic correction of errors. The prototype system is developed in close cooperation with Swedish utility company Goteborg Energi. Data from their current AMI is used both to identify the commonly occurring errors and assess the performance of the system. In evaluating the prototype system, the results showed a processing capacity of 1500 measurements per second at a latency of no more than a few seconds. This corresponds to 5 million smart meters reporting hourly readings or 1.35 million meters operating at 15 minute intervals. Investigation of error frequency based on the validation rules found the single most common meter fault at Goteborg Energi. This error a ects one particular meter model and accounts for 46% of all errors encountered in the AMI. The prototype system developed in this thesis is able to identify 61% of the occurrences of this this error in the examined data set. These 61% conform to a speci c pattern that clearly marks out this particular error. This thesis shows that stream processing can be used to validate large volumes of electricity meter data online with low processing latency, identifying common errors as they appear.



Publikationen registrerades 2015-11-17.

CPL ID: 225872

Detta är en tjänst från Chalmers bibliotek