Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Bayesian Uncertainty Quantification with Anchored Ensembles for Robust EV Power Consumption Prediction

Ghazal Farhani, Taufiq Rahman, Kieran Humphries

TL;DR

This work tackles reliable EV power prediction by quantifying both epistemic and aleatoric uncertainty. It introduces Bayesian anchored ensembles extended to LSTMs with a heavy-tailed Student-$t$ output, delivering calibrated predictive intervals without test-time sampling. The method achieves strong accuracy and well-calibrated uncertainty across controlled dyno and real-world highway data, outperforming or matching baselines such as MC dropout and quantile regression in calibration and sharpness. The approach offers a deployment-friendly, low-latency solution for uncertainty-aware range estimation and energy management in production EV systems. Overall, the anchored LSTM with a $t$-head provides a principled, efficient framework for uncertainty quantification in sequential EV power prediction.

Abstract

Accurate EV power estimation underpins range prediction and energy management, yet practitioners need both point accuracy and trustworthy uncertainty. We propose an anchored-ensemble Long Short-Term Memory (LSTM) with a Student-t likelihood that jointly captures epistemic (model) and aleatoric (data) uncertainty. Anchoring imposes a Gaussian weight prior (MAP training), yielding posterior-like diversity without test-time sampling, while the t-head provides heavy-tailed robustness and closed-form prediction intervals. Using vehicle-kinematic time series (e.g., speed, motor RPM), our model attains strong accuracy: RMSE 3.36 +/- 1.10, MAE 2.21 +/- 0.89, R-squared = 0.93 +/- 0.02, explained variance 0.93 +/- 0.02, and delivers well-calibrated uncertainty bands with near-nominal coverage. Against competitive baselines (Student-t MC dropout; quantile regression with/without anchoring), our method matches or improves log-scores while producing sharper intervals at the same coverage. Crucially for real-time deployment, inference is a single deterministic pass per ensemble member (or a weight-averaged collapse), eliminating Monte Carlo latency. The result is a compact, theoretically grounded estimator that couples accuracy, calibration, and systems efficiency, enabling reliable range estimation and decision-making for production EV energy management.

Bayesian Uncertainty Quantification with Anchored Ensembles for Robust EV Power Consumption Prediction

TL;DR

This work tackles reliable EV power prediction by quantifying both epistemic and aleatoric uncertainty. It introduces Bayesian anchored ensembles extended to LSTMs with a heavy-tailed Student- output, delivering calibrated predictive intervals without test-time sampling. The method achieves strong accuracy and well-calibrated uncertainty across controlled dyno and real-world highway data, outperforming or matching baselines such as MC dropout and quantile regression in calibration and sharpness. The approach offers a deployment-friendly, low-latency solution for uncertainty-aware range estimation and energy management in production EV systems. Overall, the anchored LSTM with a -head provides a principled, efficient framework for uncertainty quantification in sequential EV power prediction.

Abstract

Accurate EV power estimation underpins range prediction and energy management, yet practitioners need both point accuracy and trustworthy uncertainty. We propose an anchored-ensemble Long Short-Term Memory (LSTM) with a Student-t likelihood that jointly captures epistemic (model) and aleatoric (data) uncertainty. Anchoring imposes a Gaussian weight prior (MAP training), yielding posterior-like diversity without test-time sampling, while the t-head provides heavy-tailed robustness and closed-form prediction intervals. Using vehicle-kinematic time series (e.g., speed, motor RPM), our model attains strong accuracy: RMSE 3.36 +/- 1.10, MAE 2.21 +/- 0.89, R-squared = 0.93 +/- 0.02, explained variance 0.93 +/- 0.02, and delivers well-calibrated uncertainty bands with near-nominal coverage. Against competitive baselines (Student-t MC dropout; quantile regression with/without anchoring), our method matches or improves log-scores while producing sharper intervals at the same coverage. Crucially for real-time deployment, inference is a single deterministic pass per ensemble member (or a weight-averaged collapse), eliminating Monte Carlo latency. The result is a compact, theoretically grounded estimator that couples accuracy, calibration, and systems efficiency, enabling reliable range estimation and decision-making for production EV energy management.

Paper Structure

This paper contains 19 sections, 13 equations, 3 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Methodology: Bayesian Anchored Ensembles for LSTMs
  4. Bayesian Anchored Ensemble for Epistemic Uncertainty
  5. Extension to LSTMs
  6. Aleatoric Uncertainty: Heteroscedastic Gaussian Negative Log-Likelihood
  7. Aleatoric Uncertainty and Combined Objective
  8. Evaluation metrics
  9. Data Collection and Model Architecture
  10. Data Collection
  11. Model Architecture Selection
  12. Feature Selection
  13. Result
  14. The Anchored LSTM Model
  15. Comparative Study
  16. ...and 4 more sections

Figures (3)

  • Figure 1: Estimated power consumption for the sedan (solid dark red) and the truck (solid violet); ground truth shown as dashed orange and dashed blue, respectively. Confidence intervals are depicted by the shaded yellow and green bands.
  • Figure 2: Left Panel: t–NLL Anchor (dark red) and t–NLL Dropout (violet), with their confidence intervals (yellow and light blue bands) Right Panel: Quantile–Anchor (dark blue) and Quantile–Dropout (light blue) with corresponding bands (light pink and light green)
  • Figure 3: Left Panel: t–NLL Anchor (dark red) and t–NLL Dropout (violet), with their confidence intervals (yellow and light blue bands) Right Panel: Quantile–Anchor (dark blue) and Quantile–Dropout (light blue) with corresponding bands (light pink and light green)