Extended Deep Learning Model to Predict the Electric Vehicle Motor Operating Point University of Michigan

Format:

Book

Conference/Event

Author/Creator:

Kolachalama, Srikanth, author.

Contributor:

Malik, Hafiz

Conference Name:

WCX SAE World Congress Experience (2024-04-16 : Detroit, Michigan, United States)

Language:

English

Physical Description:

1 online resource cm

Place of Publication:

Warrendale, PA SAE International 2024

Summary:

The transition from combustion engines to electric propulsion is accelerating in every coordinate of the globe. The engineers had strived hard to augment the engine performance for more than eight decades, and a similar challenge had emerged again for electric vehicles. To analyze the performance of the engine, the vector engine operating point (EOP) is defined, which is common industry practice, and the performance vector electric vehicle motor operating point (EVMOP) is not explored in the existing literature. In an analogous sense, electric vehicles are embedded with three primary components, e.g., Battery, Inverter, Motor, and in this article, the EVMOP is defined using the parameters [motor torque, motor speed, motor current]. As a second aspect of this research, deep learning models are developed to predict the EVMOP by mapping the parameters representing the dynamic state of the system in real-time. The required data is obtained by the testing of a 2023 Cadillac Lyriq electric vehicle (single motor) driven on a specified road segment. The trained functions are developed utilizing the integrated functions of MATLAB, which include the machine learning methods including non-linear autoregressive (NARX), long short-term memory (LSTM), and neural net fit (NNF), suiting vehicle data in the time domain. The performance of the methods is validated by estimating the Error vector representing the conformance of actual and predicted values of randomly selected road snippet data for ten seconds. The RMSE values of the Error, and its first-order derivative (dError) are analyzed as a metric of performance, and thus the best suitable method for modeling the electric vehicle data is identified. It is observed that the NARX method in conjunction with the scaled conjugate gradient (SCG) solver outperformed with the least computational time (< 66.48 s) and lower RMSE, e.g., Error < 3.64 and dError < 4.94, values in all scenarios

Notes:

Vendor supplied data

Publisher Number:

2024-01-2551

Access Restriction:

Restricted for use by site license