Review of Biomass Gasification and Pyrolysis Process Based on Long Short-Term Memory Network

Abstract

Abstract. With the global energy crisis become escalated day by day, finding an effective way of exploring the renewable energy has become an important direction as its widely resources, comparatively low environment impact. This paper has concluded the biomass pyrolysis and gasification process based on long- and short-term memory (LSTM) method, including the LSTM principle, current application, faced challenges and future development. LSTM as a kind of special recurrent neural network, which could effectively solve the gradient vanish and explosion problem in time-series data. Research shows that, LSTM has an excellent performance in terms of reaction condition and optimization, which could be able to capture the relationship between complex variables, the pyrolysis efficiency could be improved, and prediction accuracy could over 90%. During the data preprocessing and model training stage, it is vital to make sure the quality of data, including data normalisation, outlier handling and time series splitting, etc. Through these steps, a better prediction accuracy could be achieved by LSTM. At the same time, the real application cases of LSTM used in biomass pyrolysis process have been discussed, it is shown that the combination of LSTM and other machine learning techniques could obviously improve the system stability and product quality. In addition, this paper has also looked forward to the enhancement of model explainability and the prospect of uncertainty quantification, which is helpful to improve the LSTM application and practical value.