Fault detection in district heating substations

a cluster-based and an instance-based approach

Abstract

In district heating or collective heating substations, components can fail or can be inappropriately installed or configured (e.g. valves get broken, heat exchangers get fouled, controller parameters are inappropriately chosen, heat exchanger wrongly connected, internal heating system problems, etc.). The result of these faults is a reduced cooling of the supply water, and as such higher than necessary return temperatures to the grid and higher volume flows (to deliver the same needed power) occur, leading to higher OPEX for all stakeholders. In this work, two approaches for a fault detection routine for district heating substations are introduced, based solely on the energy meter data, with an application on a real-life district heating network in Sweden. The first approach is a cluster-based approach in which substations within the district heating are compared to each other using the overflow method and performance signatures to flag substations with sub-optimal performance compared to other substations in the network. The second method is an instance-based approach using a black-box model to predict the behaviour of the substation using an extended set of input variables and comparing the predictions to the measurements. The results from the two fault detection approaches show that both the cluster-based and the instance-based methods can detect deviating behaviours in DH customer installations.