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!
|
| _version_ | 1832584705991507968 |
|---|---|
| author | Seiya Takano Tomohiko Jimbo |
| author_facet | Seiya Takano Tomohiko Jimbo |
| author_sort | Seiya Takano |
| collection | DOAJ |
| description | 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. |
| format | Article |
| id | doaj-art-184f811f48894b56b3b66211cc5cb534 |
| institution | Kabale University |
| issn | 1884-9970 |
| language | English |
| publishDate | 2025-12-01 |
| publisher | Taylor & Francis Group |
| record_format | Article |
| series | SICE Journal of Control, Measurement, and System Integration |
| spelling | doaj-art-184f811f48894b56b3b66211cc5cb5342025-01-27T11:59:10ZengTaylor & Francis GroupSICE Journal of Control, Measurement, and System Integration1884-99702025-12-0118110.1080/18824889.2025.24552222455222Warped matrix-variate Gaussian processes with structured kernels for multi-step-ahead driving behaviour predictionSeiya Takano0Tomohiko Jimbo1Toyota Motor CorporationToyota Central R&D Labs., IncThe 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.http://dx.doi.org/10.1080/18824889.2025.2455222matrix-variate gaussian processkernel designinterpretabilitydriving behaviourstate of charge |
| spellingShingle | Seiya Takano Tomohiko Jimbo Warped matrix-variate Gaussian processes with structured kernels for multi-step-ahead driving behaviour prediction SICE Journal of Control, Measurement, and System Integration matrix-variate gaussian process kernel design interpretability driving behaviour state of charge |
| title | Warped matrix-variate Gaussian processes with structured kernels for multi-step-ahead driving behaviour prediction |
| title_full | Warped matrix-variate Gaussian processes with structured kernels for multi-step-ahead driving behaviour prediction |
| title_fullStr | Warped matrix-variate Gaussian processes with structured kernels for multi-step-ahead driving behaviour prediction |
| title_full_unstemmed | Warped matrix-variate Gaussian processes with structured kernels for multi-step-ahead driving behaviour prediction |
| title_short | Warped matrix-variate Gaussian processes with structured kernels for multi-step-ahead driving behaviour prediction |
| title_sort | warped matrix variate gaussian processes with structured kernels for multi step ahead driving behaviour prediction |
| topic | matrix-variate gaussian process kernel design interpretability driving behaviour state of charge |
| url | http://dx.doi.org/10.1080/18824889.2025.2455222 |
| work_keys_str_mv | AT seiyatakano warpedmatrixvariategaussianprocesseswithstructuredkernelsformultistepaheaddrivingbehaviourprediction AT tomohikojimbo warpedmatrixvariategaussianprocesseswithstructuredkernelsformultistepaheaddrivingbehaviourprediction |