Advanced machine learning approach with dynamic kernel weighting for accurate electrical load forecasting
This article presents a load forecasting model for commercial buildings with Enhanced Dynamically Weighted Multiple Kernel Support Vector Regression (EDW-MKSVR) and a mini-batch gradient descent method to achieve good regularization along with clustering techniques to segment different times of the...
Saved in:
| Main Authors: | , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
AIP Publishing LLC
2025-01-01
|
| Series: | AIP Advances |
| Online Access: | http://dx.doi.org/10.1063/5.0218832 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1832542793044590592 |
|---|---|
| author | C. Jeevakarunya V. Manikandan |
| author_facet | C. Jeevakarunya V. Manikandan |
| author_sort | C. Jeevakarunya |
| collection | DOAJ |
| description | This article presents a load forecasting model for commercial buildings with Enhanced Dynamically Weighted Multiple Kernel Support Vector Regression (EDW-MKSVR) and a mini-batch gradient descent method to achieve good regularization along with clustering techniques to segment different times of the day. The first step in preparing the dataset is to perform preprocessing, which involves conducting correlation analysis, scaling, and normalization followed by initial hyperparameter tuning using the multiswarm Levy flight particle swarm optimization technique. Compared with traditional methods, EDW-MKSVR offers greater adaptability to shifting load patterns since it makes use of dynamic kernel weight modifications that are dependent on data attributes. This method is used on the Commercial Buildings Energy Consumption Survey dataset, which includes information on the energy use of commercial buildings together with the climatic characteristics that are related to them across a range of periods, including the week, season, hour, and human behavior. This segmentation can be carried out using the clustering technique. In terms of performance, the study assesses the effectiveness of EDW-MKSVR models against those of boosted tree, random forest regression, K-nearest neighbor, support vector regression, and long short term memory. The study’s conclusions imply that EDW-MKSVR outperforms other current methods in terms of accuracy when obtaining complex load patterns. This article demonstrates how reliable and accurate the EDW-MKSVR approach is in predicting energy use. It facilitates the power industry’s ability to make better judgments by providing more precision and flexibility. Furthermore, the robustness of the proposed model is evaluated by predicting the electricity consumption of two additional datasets using the proposed method. Next, we compare these predictions’ performance measures, which has demonstrated the ability to generate highly ranked accuracy metrics. |
| format | Article |
| id | doaj-art-a48abaace7484be3b05e3e9eac870b68 |
| institution | Kabale University |
| issn | 2158-3226 |
| language | English |
| publishDate | 2025-01-01 |
| publisher | AIP Publishing LLC |
| record_format | Article |
| series | AIP Advances |
| spelling | doaj-art-a48abaace7484be3b05e3e9eac870b682025-02-03T16:40:41ZengAIP Publishing LLCAIP Advances2158-32262025-01-01151015011015011-1310.1063/5.0218832Advanced machine learning approach with dynamic kernel weighting for accurate electrical load forecastingC. JeevakarunyaV. ManikandanThis article presents a load forecasting model for commercial buildings with Enhanced Dynamically Weighted Multiple Kernel Support Vector Regression (EDW-MKSVR) and a mini-batch gradient descent method to achieve good regularization along with clustering techniques to segment different times of the day. The first step in preparing the dataset is to perform preprocessing, which involves conducting correlation analysis, scaling, and normalization followed by initial hyperparameter tuning using the multiswarm Levy flight particle swarm optimization technique. Compared with traditional methods, EDW-MKSVR offers greater adaptability to shifting load patterns since it makes use of dynamic kernel weight modifications that are dependent on data attributes. This method is used on the Commercial Buildings Energy Consumption Survey dataset, which includes information on the energy use of commercial buildings together with the climatic characteristics that are related to them across a range of periods, including the week, season, hour, and human behavior. This segmentation can be carried out using the clustering technique. In terms of performance, the study assesses the effectiveness of EDW-MKSVR models against those of boosted tree, random forest regression, K-nearest neighbor, support vector regression, and long short term memory. The study’s conclusions imply that EDW-MKSVR outperforms other current methods in terms of accuracy when obtaining complex load patterns. This article demonstrates how reliable and accurate the EDW-MKSVR approach is in predicting energy use. It facilitates the power industry’s ability to make better judgments by providing more precision and flexibility. Furthermore, the robustness of the proposed model is evaluated by predicting the electricity consumption of two additional datasets using the proposed method. Next, we compare these predictions’ performance measures, which has demonstrated the ability to generate highly ranked accuracy metrics.http://dx.doi.org/10.1063/5.0218832 |
| spellingShingle | C. Jeevakarunya V. Manikandan Advanced machine learning approach with dynamic kernel weighting for accurate electrical load forecasting AIP Advances |
| title | Advanced machine learning approach with dynamic kernel weighting for accurate electrical load forecasting |
| title_full | Advanced machine learning approach with dynamic kernel weighting for accurate electrical load forecasting |
| title_fullStr | Advanced machine learning approach with dynamic kernel weighting for accurate electrical load forecasting |
| title_full_unstemmed | Advanced machine learning approach with dynamic kernel weighting for accurate electrical load forecasting |
| title_short | Advanced machine learning approach with dynamic kernel weighting for accurate electrical load forecasting |
| title_sort | advanced machine learning approach with dynamic kernel weighting for accurate electrical load forecasting |
| url | http://dx.doi.org/10.1063/5.0218832 |
| work_keys_str_mv | AT cjeevakarunya advancedmachinelearningapproachwithdynamickernelweightingforaccurateelectricalloadforecasting AT vmanikandan advancedmachinelearningapproachwithdynamickernelweightingforaccurateelectricalloadforecasting |