Title:

Recurrent Neural Networks for Prediction of Geothermal Reservoir Performance

Authors:

Anyue JIANG, Zhen QIN, Trenton T. CLADOUHOS, Dave FAULDER, Behnam JAFARPOUR

Key Words:

machine learning, predictive analytics, recurrent neural networks, geothermal reservoirs

Conference:

Stanford Geothermal Workshop

Year:

2021

Session:

Reservoir Engineering

Language:

English

Paper Number:

Jiang

File Size:

1245 KB

View File:

Abstract:

Improving the efficiency of energy production from geothermal reservoirs hinges on accurate prediction models that describe the performance of the geothermal reservoir under alternative development scenarios. Physics-based models offer a comprehensive prediction framework that requires the construction of a reliable reservoir model, which involves complex multi-physics modeling and integration of multiple sources of data. Additionally, the uncertainty in the description of reservoir models as well as physical processes and the related properties, the resulting predictions are subject to a significant level of uncertainty. These limitations, especially the effort required to construct a model, complicate the use of physics-based simulation models for managing daily operations and planning, as well as potentially long-term management of geothermal reservoirs. An efficient alternative to the physics-based model is data-driven approaches that extract statistical patterns and dependencies from various sources of data to develop predictive models. For dynamical systems that involve data sequences, recurrent neural networks (RNN) present an effective tool for capturing the temporal trends in the data. RNN can be viewed as a directed graph with a temporal sequence that can be used to model dynamic data. An important advantage of RNN is its internal state (memory), which is used to process data sequences of variable lengths. We present a robust sequence-to-sequence RNN model for predicting the dynamic response of geothermal reservoirs after training with historical monitoring measurements. The RNN model consists of an encoder that summarizes the dynamics within the historical data and a decoder RNN that predicts the production trends based on the learned trends and the applied control forcing. We develop a labeling scheme that enables the RNN to ignore certain time steps that involves error or missing data. We present the RNN model in detail, including its architecture, training procedure, and demonstrate its prediction performance by applying it to different datasets, including field data from a binary cycle geothermal power plant.


ec2-54-163-221-133.compute-1.amazonaws.com, you have accessed 0 records today.

Press the Back button in your browser, or search again.

Copyright 2021, Stanford Geothermal Program: Readers who download papers from this site should honor the copyright of the original authors and may not copy or distribute the work further without the permission of the original publisher.


Attend the nwxt Stanford Geothermal Workshop, click here for details.

Accessed by: ec2-54-163-221-133.compute-1.amazonaws.com (54.163.221.133)
Accessed: Thursday 28th of March 2024 08:41:29 AM