Forecasting vehicle accelerations using LSTM
The purpose of this paper is to forecast vehicle accelerations by using a Long Short Term Memory (LSTM) approach. Such a predictive capability can be particularly helpful in the case of medical emergency vehicles. During emergency transport, a patient is likely to experience vehicle accelerations in...
Guardado en:
| Autores principales: | , , |
|---|---|
| Formato: | article |
| Lenguaje: | EN |
| Publicado: |
The Japan Society of Mechanical Engineers
2021
|
| Materias: | |
| Acceso en línea: | https://doaj.org/article/7d5811d17264484ebc7e02705896877e |
| Etiquetas: |
Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
| Sumario: | The purpose of this paper is to forecast vehicle accelerations by using a Long Short Term Memory (LSTM) approach. Such a predictive capability can be particularly helpful in the case of medical emergency vehicles. During emergency transport, a patient is likely to experience vehicle accelerations in both longitudinal and lateral directions. Although the effect of these accelerations acting on a patient can be reduced by actively controlling the attitude of the ambulance bed, there is inevitably a delay from the time the acceleration is measured until the attitude of the bed reaches its target state. In our approach, we forecast future accelerations from past time series data by using LSTM, a recurrent neural network architecture utilized in deep learning, to predict future accelerations at each time step. Using various driving scenarios, the LSTM is trained with different training data and different numbers of hidden layers, units, and epochs. To evaluate the performance and usefulness of the approach, real-time simulations are conducted using measured longitudinal and lateral vehicle acceleration data. Forecast accuracy is assessed for the trained LSTM with different parameters, and the results show the capability of producing accurate real-time forecasts for certain parameter settings. Comparisons of the LSTM’s forecast results with the results of an autoregressive integrated moving average model shows the advantages of the LSTM approach especially for unsteady time series data that includes such elements as sudden or large acceleration changes. |
|---|