Gas Pipeline Flow Prediction Model Based on LSTM with Grid Search Parameter Optimization

Abstract

Due to the abundant operation data (e.g., pressure, flow rate, and temperature) for natural gas (NG) gathering pipelines provided by the supervisory control and data acquisition (SCADA) system, the machine-learning-based real-time flow prediction has become a critical solution to enable the identification of the abnormality of pipelines, further to guarantee the safe operation of the pipelines. However, traditional machine-learning-based methods cannot always function well due to the temporal characteristics of the SCADA data often being ignored, resulting from a lack of time memory capability. Therefore, this paper proposes a method to automatically perform the feature mining of flow time series by considering the correlation of flow data at both ends of the pipeline, combined with the long short-term memory (LSTM) network. The current and historical data at both pipeline ends are used as input vectors of the LSTM network to predict the terminal output flow at the next moment. Furthermore, to solve the problem that the parameters of the LSTM model are configured with manual experience, a grid search algorithm (GSA) is introduced to optimize the parameters of the LSTM. Consequently, the effectiveness and superiority of the proposed method are carried out in a real-world NG gathering pipeline.