QR Code
  • Institute: Institute of Information And Communication Technology
  • Symposium: Environmental and Cultural Sustainability
  • Day: 1 , Session: 2 , Location: Conference Hall 1
  • Session Type: Medium 15 min (Presentation of ongoing research developments and findings) , Start: 12:15 , End: 12:35

Abstract

Metering water consumption is a complex problem. Apparent losses in water consumption are a major cost to water supply companies that are caused by water leaks, theft and even water meter inaccuracies. Water meters can become very inaccurate over their lifetime and under read the actual consumption of a household. The age of the water meter, as well as water consumption patterns can affect how inaccurate the readings can become. These phenomena make data pre-processing a very important step to understand water meter consumption.

Water meter readings are often described as a timeseries. Actual water consumption has different seasonality patterns: daily; weekly; and even yearly. Individual consumers can also have a trend, such an increase in water consumption, as well an amount of randomness, or noise, in the amount consumed. The time series can be decomposed into these components.

Further data pre-processing includes outlier detection and data engineering which are useful for a variety of statistical methods. Outlier detection, for example, can look at the noise component of the decomposed timeseries to detect points with exceptionally large noise. Additionally, data pre-processing can also benefit from creation of new data attributes from existing data.

An example of data pre-processing will be demonstrated using R.