A general transfer learning-based framework for thermal load prediction in regional energy system

Accurate and reliable thermal load prediction is of great significance for predictive control and optimal dispatch of regional energy systems. Data-driven approach has more advantages in mining actual load's characteristics and improving prediction accuracy, but it requires significant quantities of historical data to train the models. In practice, there always exist conditions of limited data due to lack of monitoring system or time of data accumulation. This paper, therefore, proposes a general transfer learning-based framework to predict thermal load with limited data. In this framework, similarity measurement index (SMI) is first defined and used to select the optimum source prediction task (buildings with sufficient data), followed by a model-based transfer learning method used to facilitate the modeling of target prediction task (buildings with limited data) with the knowledge learned from source task. Validity of this framework is confirmed by practical cases and data, which suggested that the optimum source task could be selected from 55 source tasks by using SMI. Under different conditions of limited data, the proposed framework could achieve the best prediction stability and reduce the prediction errors by 0.6%-15.26% compared with direct learning and 1.81%-5.65% compared with transfer learning without the selection of source tasks. (C) 2020 Elsevier Ltd. All rights reserved.

Automated summary

This paper presents a transfer learning-based framework for predicting thermal load with limited data. By selecting the optimum source prediction task and utilizing a model-based transfer learning method, the framework achieves the best prediction stability and reduces prediction errors.