Method and apparatus for multivariate lstm prediction of carbon isotope abundance

Abstract

The application discloses a kind of multivariate LSTM method and equipment for predicting carbon isotope abundance, comprising: including: collecting carbon isotope data in air and meteorological factor and calculating relative index, then preprocessing collected data;The collected data is filled with missing value, abnormal value correction and normalization processing, and a multivariate time series dataset is constructed;Multivariate LSTM modeling is carried out based on the improved attention mechanism combined with carbon isotope fractionation constraint;The constructed model is trained and optimized in combination with the physical constraint term of carbon source mixing;Carbon isotope abundance prediction and error correction are carried out by the optimized model.The application realizes the physical consistency modeling of carbon isotope sequence by introducing the multivariate LSTM model of fractionation calculation mechanism and Keeling auxiliary task, and improves the precision and interpretability of carbon isotope abundance prediction.

Images