Backward simulation of temperature changes of District Heating networks for enabling loading history in predictive maintenance

Abstract

District Heating (DH) networks, like most of industries, are in transition to the fourth​ industry age and they are retrofitting themselves with different sensing and inspection technologies to enable cyber connectivity for different purposes, such as system optimization, failure detection, maintenance, etc. Since DH pipes show different ageing behaviour under different conditions and initially the pre-insulated bounded pipes had been designed for a minimum of 30 years life span, a long-term loading history is required for predictive maintenance (PdM) purposes and it is necessary to understand the ageing of the DH pipes. These historical temperature changes of the networks are not available for such a long period and they are usually limited to the past few years. To exploit the available implemented technologies for PdM , the missing data must become available to understand the ageing patterns and expand the ageing model to the pipes in use. In this research, various Machine Learning (ML) techniques such as Support Vector Machine (SVM), Random Forest algorithm (RF), Artificial Neural Networks (ANN) have been tested to train a model and backward simulate the temperature changes of the system based on recorded weather data. Various none-temperature variables have been used to enhance the prediction qualities to the real-world data. The historical temperature changes of the system shall be used for different ageing estimation such as fatigue cycles or remaining useful life of the polyurethane (PUR) foam.