Warped matrix-variate Gaussian processes with structured kernels for multi-step-ahead driving behaviour prediction
The increased integration of renewable energy, which is known to fluctuate owing to weather conditions, has necessitated power supply-demand adjustments using virtual power plants (VPPs) that utilize vehicle batteries. Predictions of the available adjustment capacity of a VPP can prevent excessive p...
Saved in:
| Main Authors: | , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Taylor & Francis Group
2025-12-01
|
| Series: | SICE Journal of Control, Measurement, and System Integration |
| Subjects: | |
| Online Access: | http://dx.doi.org/10.1080/18824889.2025.2455222 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | The increased integration of renewable energy, which is known to fluctuate owing to weather conditions, has necessitated power supply-demand adjustments using virtual power plants (VPPs) that utilize vehicle batteries. Predictions of the available adjustment capacity of a VPP can prevent excessive power generation. Therefore, the state of charge (SOC) of individual vehicles participating in the supply-demand adjustment market must be predicted up to one week in advance. Because vehicle owners' actions determine SOC behaviour, considering their periodicity and uncertainty is essential. This study proposes a novel probabilistic model called the matrix-variate Gaussian process (GP) with structured kernels while focusing on the periodicity. To achieve high accuracy through recursive multi-step-ahead prediction using GP, large computational resources are required. However, the proposed method can handle a matrix with rows for the time of day and columns for the day of the week, which facilitates the production of a one-week-ahead predictive distribution through a single calculation. Furthermore, the proposed method directly captures the periodicity using separate kernels to express the characteristics of rows and columns, making the parameters easier to interpret. Additionally, from a practical standpoint, the proposed model is extended to constrain the support of the predictive distribution to a finite interval. Finally, we evaluated the performance of the proposed method using real data from 100 vehicles over up to nine months. |
|---|---|
| ISSN: | 1884-9970 |