Enhancing time series prediction with Hybrid AFSA-TCN: A unified approach to temporal data and optimization
Time series data, with its sequential dependencies presents a unique challenge for traditional machine learning methods such as Random Forest (RF), Support Vector Machines (SVM), and Decision Trees (DT), which often struggle to capture temporal patterns effectively. In contrast, Temporal Convolution...
Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Language: | en |
| Published: |
Arqii Publication
2025
|
| Subjects: | |
| Online Access: | https://umpir.ump.edu.my/id/eprint/47223/1/Enhancing%20time%20series%20prediction%20with%20Hybrid%20AFSA-TCN.pdf https://umpir.ump.edu.my/id/eprint/47223/ http://arqiipubl.com/ojs/index.php/AMS_Journal/article/view/868 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Time series data, with its sequential dependencies presents a unique challenge for traditional machine learning methods such as Random Forest (RF), Support Vector Machines (SVM), and Decision Trees (DT), which often struggle to capture temporal patterns effectively. In contrast, Temporal Convolutional Networks (TCNs) have proven to be highly accurate in addressing these challenges. This study focuses on forecasting the Remaining Useful Life (RUL) of batteries, a critical task for applications such as electric vehicles, renewable energy, and battery management. The dataset used in this study is a battery RUL dataset retrieved from an open-source platform Kaggle, which consists of more than 15,000 rows of time series data. The study introduces a hybrid model that integrates TCN with Artificial Fish Swarm Algorithm (AFSA), a bio-inspired optimization technique designed to fine-tune TCN parameters. The results show that AFSA-TCN achieves outstanding performance with an R-squared (R2) score of 0.9992, Mean Absolute Error (MAE) of 4.8200, Root Mean Squared Error (RMSE) of 9.0840 and Mean Absolute Percentage Error (MAPE) of 5.48%. Further comparisons with other AFSA-based hybrid models, including AFSA-SVM, AFSA-Gated Recurrent Unit (GRU), and AFSA-Artificial Neural Network (ANN), reveal that AFSA-TCN offers superior prediction accuracy, efficient tuning, and adaptability to the battery RUL dataset. The results highlight the robustness and effectiveness of the AFSA-TCN hybrid model in addressing complex challenges in time series forecasting and optimization, particularly for RUL estimation. |
|---|