[Paper Review] Data-Efficient Modeling for Precise Power Consumption Estimation of Quadrotor Operations Using Ensemble Learning

Wei Dai, Mingcheng Zhang+1 more|arXiv (Cornell University)|May 23, 2022

Air Traffic Management and Optimization4 citations

TL;DR

This paper proposes a data-efficient stacking ensemble model using random forest and XGBoost as base learners with linear regression as meta-learner to estimate quadrotor power consumption with high precision. The model achieves less than 0.5% error in total flight power prediction with over 80% confidence and outperforms deep learning baselines using only ~2,000 flight samples, enabling scalable deployment across diverse eVTOL aircraft types.

ABSTRACT

Electric Take-Off and Landing (eVTOL) aircraft is considered as the major aircraft type in the emerging urban air mobility. Accurate power consumption estimation is crucial to eVTOL, supporting advanced power management strategies and improving the efficiency and safety performance of flight operations. In this study, a framework for power consumption modeling of eVTOL aircraft was established. We employed an ensemble learning method, namely stacking, to develop a data-driven model using flight records of three different types of quadrotors. Random forest and extreme gradient boosting, showing advantages in prediction, were chosen as base-models, and a linear regression model was used as the meta-model. The established stacking model can accurately estimate the power of a quadrotor. Error analysis shows that about 80% prediction errors fall within one standard deviation interval and less than 0.5% error in the prediction for an entire flight can be expected with a confidence of more than 80%. Our model outperforms the existing models in two aspects: firstly, our model has a better prediction performance, and secondly, our model is more data-efficient, requiring a much smaller dataset. Our model provides a powerful tool for operators of eVTOL aircraft in mission management and contributes to promoting safe and energy-efficient urban air traffic.

Motivation & Objective

• To develop a highly accurate and data-efficient power consumption model for eVTOL aircraft to support safe and energy-efficient urban air mobility operations.
• To address the limitations of existing models that require large datasets and lack robustness across diverse aircraft types.
• To improve flight mission planning and energy management by enabling precise, real-time power estimation using minimal flight test data.
• To establish a scalable framework applicable to new eVTOL platforms without extensive retraining or data collection.
• To outperform state-of-the-art deep learning models in prediction accuracy while requiring significantly fewer training samples.

Proposed method

• Employed a stacking ensemble learning framework using random forest and XGBoost as base models, and linear regression as the meta-model to aggregate predictions.
• Trained the model on flight test data from three quadrotor types, including two custom datasets and one open-source dataset, totaling approximately 2,000 flights.
• Used feature engineering to include flight dynamics such as throttle, pitch, roll, yaw, and altitude to capture power variation during maneuvers.
• Optimized hyperparameters using cross-validation and evaluated model performance using R², MSE, and MAPE metrics on a held-out test set.
• Conducted error analysis to assess prediction reliability, including distribution of instantaneous and cumulative power errors across entire missions.
• Compared model performance against two state-of-the-art models: a 35-layer MLP and a temporal convolutional network (TCN) with residual connections.

Experimental results

Research questions

• RQ1 Can an ensemble learning model achieve higher accuracy than existing deep learning models in power consumption estimation with significantly less training data?
• RQ2 To what extent does the stacking model reduce prediction error in both instantaneous and cumulative power estimation across diverse quadrotor operations?
• RQ3 How does the model’s performance scale across different aircraft types with minimal retraining or data collection?
• RQ4 What is the confidence level of the model in predicting total flight power within a 0.5% error margin?
• RQ5 Can the model maintain high accuracy when applied to new eVTOL platforms with limited flight data?

Key findings

• The stacking model achieved an R² of 0.998, MSE of 1.2×10⁴, and MAPE of 1.8% on the test set, outperforming both the MLP and TCN baselines in all three metrics.
• Approximately 80% of instantaneous power prediction errors fell within one standard deviation of the true values, indicating strong prediction consistency.
• For entire flight missions, less than 0.5% error in total power consumption prediction was achievable with over 80% confidence, which is critical for mission safety.
• The model required only ~2,000 flight samples for training, demonstrating high data efficiency compared to deep learning models that require tens of thousands of samples.
• The model showed superior ability to capture high-frequency power variations during complex maneuvers, outperforming both MLP and TCN in curve approximation.
• The model’s performance was robust across different quadrotor types, suggesting strong generalization and scalability for new eVTOL platforms.