Work

LSTM-based Flow Prediction

Year: 2023

Type: conference-paper

Abstract: This paper introduces an innovative approach for predicting continuous time series variables in the context of production or flow. The proposed method employs an LSTM algorithm enhanced through multivariate tuning. This algorithm refines the conventional LSTM by transforming time series data into supervised learning sequences, tailored to the distinctive features of industrial data. The primary novelty lies in incorporating the concepts of periodic measurement and time window into industrial prediction, particularly when dealing with time series characteristics in industrial data. Experimental results using real-world datasets demonstrate a notable improvement in prediction accuracy, achieving a 54.05% enhancement over the traditional LSTM algorithm.

Source: International Conference on Big Data and Information Analytics (BigDIA)

DOI: 10.1109/bigdia60676.2023.10429740