SOM-Based Selection of Monitored Consumers for Demand Prediction

Abstract

Modern utility companies manage extensive distribution systems to provide multiple consumers with water, heat and electrical power. At the same time significant savings can be received from a combination of monitoring systems and modelling applications used to optimize the distribution systems. Thus, in case of pipeline systems, the problem of identifying key hydraulic control and monitoring points has been extensively studied.

Our recent research shows that by investigating the data acquired from heat meters in a district heating company, the accuracy of demand prediction can be significantly improved. The proposed methods rely on the availability of a system monitoring selected heat meters in an on-line manner. Thus, the purpose of this study is to develop and evaluate different methods of selecting heat meters to be monitored in order to provide data for prediction models. Self-organising maps have been applied to identify groups of consumers. The optimal number of monitored consumers in every group and the strategies of selecting the consumers to be monitored are searched for.