Soft sensor with shape descriptors for flame quality prediction based on LSTM regression

Effectively predicting flame shape and rationally developing corresponding safety measures have important guiding significance for improving combustion quality. Aimed at the problem that the traditional flame shape prediction method fails to predict the combustion quality in the next time period, this chapter proposes a long-term and short-term memory (LSTM) cyclic neural network prediction method. The LSTM method is based on actual monitoring data, including model building, structural design, model training, model prediction, and model optimization. The number of layers and batch size are used as parameters for the model. Experiments show that the LSTM model can effectively predict combustion quality and flue gas emissions in the next time period as compared to a recurrent neural network (RNN). It has higher applicability and reliability in flame shape-based combustion quality estimation using time series prediction. In addition to this model, image processing and artificial intelligence (AI) systems of the IoT can be used for evaluation in the field of energy production. These can help us to monitor and ultimately reduce emission of harmful gases to the atmosphere. Our proposed monitoring system is not only efficient but also cost-effective, and can help reduce the disastrous effects of greenhouse gas emissions and global warming.