Predicting Solid Particle Levels in Diesel Generators using an Autoformer Neural Network

Abstract

The period between 2010 and 2017 was marked by robust economic growth, which resulted in a notable rise in global energy consumption. Specifically, electricity usage surged by 19.4%, while fossil fuels experienced a 9.1% increase. In this context, the innovation of smart, environmentally-friendly diesel generators, coupled with the implementation of advanced artificial intelligence algorithms for optimizing performance and minimizing emissions, is garnering significant attention within industrial automation sectors. This study introduces the autoformer neural network framework, tailored for predicting solid particle emissions from diesel generators. The prediction model utilized input parameters such as vibration data, acoustic signals, and thermal images of exhaust gases. The generator was tested under varying loads of 0.0 kW, 0.3 kW, 0.6 kW, 1.0 kW, 1.3 kW, 1.6 kW, and 2.0 kW, while measuring exhaust particle sizes of 0.3 μm, 0.5 μm, 1.0 μm, 2.5 μm, 5.0 μm, and 10.0 μm. The autoformer model yielded optimal predictions at a generator load of 1.3 kW, achieving a MAPE of 1.1%, 3.2%, 0.8%, 1.4%, 1%, and 0.9% for particles sizes of 0.3 μm, 0.5 μm, 1.0 μm, 2.5 μm, 5.0 μm, and 10.0 μm, respectively. Although the autoformer model’s accuracy declined under varying load conditions, these findings affirm its potential as an innovative tool for predicting exhaust particle emissions effectively.