An artificial neural network module for simulation of the energy storage system of a hybrid electric vehicle

Abstract

A hybrid electric vehicle (HEV) is a complex system integrating interactive subsystems of disparate degrees of complexity. The simulation of an HEV thus poses a challenge. An accurate simulation requires highly accurate models of each subsystem. Without these, the system has a poor overall performance. Typically, modelling problems are not amenable to physical solutions without simplifying assumptions that impair their accuracy. Conventional empirical models, on the other hand, are time consuming and data intensive and falter where extensive non-linearity is encountered. An artificial neural network (ANN) approach to simulation of an HEV is presented in this paper. An ANN model of the energy storage system (ESS) of an HEV was deployed in the ADVISOR simulation software developed by the National Renewable Energy Laboratories (NREL) of the US Department of Energy. The ANN model mapped the state of charge (SOC) and the power requirement of the vehicle to the voltage and current at the ESS output An ANN model was able accurately to capture the complex, non-linear phenomena underlying the ESS.

A novel performance-enhancing technique for design of ANN training data, Smart Select, is described here. It resulted in a model of 0.9978 correlation (R2 error) with data. ANNs can be data hungry. The issue of knowledge sharing between ANN models to save development time and effort is also addressed in this paper. The model transfer technique presents a way of levering the expertise of one ANN into the development of another for a similar modelling task. Lastly, integration of the ANN model of the ESS into the ADVISOR software, on the MATLAB software platform, is described.

Keywords

artificial neural network energy storage hybrid electric vehicle

Get full access to this article

View all access options for this article.

References

Haykin

Neural Networks: A Comprehensive Foundation, 1994 (Macmillan College Publishing Company, New York).

Rumelhart

Hinton

Williams

Learning internal representation by error propagation. In Parallel Distributed Processing, 1986 (MIT Press, Cambridge, Massachusetts).

Bishop

Neural Networks for Pattern Recognition, 1995 (Oxford).

Hornik

Stichcombe

White

Multilayer feedforward neural networks are universal approximators. Neural Networks, 1989, 2, 359–366.

Smith

Neural Networks for Statistical Modeling, 1993 (Van Nostrand Reinhold, New York).

Werbos

Beyond regression: new tools for prediction and analysis in the behavioral sciences. PhD thesis, Harvard University, Boston, Massachusetts, 1974.

Marwah

Neural network modeling techniques for electronics manufacturing processes. MS thesis, University of Colorado at Boulder, Colorado, 1993.

Marwah

Mahajan

Integrated neural network modeling for electronic manufacturing. J. Electronics Mfg, June 1996, 6(2), 79–91.

10.

Kelkar

Mahajan

Sani

Real-time physiconeural solutions for MOCVD. Trans. ASME, J. Heat Transfer, 1996, 118(8), 814.

11.

Mahajan

Wang

Neural network models for thermally based microelectronic manufacturing processes. J. Electrochem. Soc., August 1993, 140(8), 2287–2293.

12.

Marwah

Mahajan

Building neural network equipment models using model modifier techniques. IEEE Trans. Semiconductor Mfg, August 1999, 12(3), 377–381.

13.

Bhatikar

Artificial neural network based energy storage system modeling for hybrid electric vehicles. MS thesis, University of Colorado at Boulder, Colorado, 1999.