Using Machine Learning and Artificial Neural Networks to Improve System-Wide Energy Efficiency
Annual AIChE Student Conference
2020
2020 Virtual Annual Student Conference
Annual Student Conference
Undergraduate Student Poster Session: Computing and Process Control
Monday, November 16, 2020 - 10:00am to 12:30pm
This work uses Artificial Neural Networks (ANNs) and Machine Learning (ML) to predict; (1) energy and exergy flow characterizations, and (2) the thermal response of an electric vehicleâs powertrain governed by nonlinear battery chemistry. The powertrain response is largely impacted by the variable operating conditions and drive cycle, further dictated by the non-linear dissimilar energy flows governing energy production and delivery. Powertrain performance may be degraded because of the dissimilar energy flow interactions, as such energy and exergy flow. The dissimilar energy flows are highly nonlinear, so accurate representation of the complex phenomena may be developed using advanced interdisciplinary mechanics. However, such systems may be stiff with minimal compliance and are not suitable for real-time control and optimization. Due to recent advancements in data and computational sciences, ANNs and ML based algorithms have become effective tools that could be used to characterize the flow of energy. These algorithms can easily identify nonlinear patterns and handle multiple arrays of variables or inputs for given output yielding extensible and adaptable predictive based algorithms. However, one potential downside of developing and deploying ML based algorithms is that it requires a large diverse training dataset, which in turn means the training process is computationally expensive and time consuming. To offset these disadvantages, a set of ANNs and Long-Short-Term Memory (LSTM) neural networks were developed and trained. The ANN and LSTM were used on different sensor-based components to approximate energy flow and predict the future flow. Our future work, which further examines efficiency in complex energy system, will shift focus to analyze more complex energy system that include the complex thermal-chemical-fluid interactions of an internal combustion engine, and to leverage the prediction-based algorithms to develop and deploy Model Predictive Control (MPC) to improve energy efficiency system wide.