Research Article
A Time Series Forecasting Method Using DBN and Adam Optimization
@INPROCEEDINGS{10.1007/978-3-031-29126-5_8, author={Takashi Kuremoto and Masafumi Furuya and Shingo Mabu and Kunikazu Kobayashi}, title={A Time Series Forecasting Method Using DBN and Adam Optimization}, proceedings={Artificial Intelligence for Communications and Networks. 4th EAI International Conference, AICON 2022, Hiroshima, Japan, November 30 - December 1, 2022, Proceedings}, proceedings_a={AICON}, year={2023}, month={3}, keywords={Time series forecasting Deep learning Deep Belief Net Error Back-Propagation Adam learning optimization}, doi={10.1007/978-3-031-29126-5_8} }- Takashi Kuremoto
Masafumi Furuya
Shingo Mabu
Kunikazu Kobayashi
Year: 2023
A Time Series Forecasting Method Using DBN and Adam Optimization
AICON
Springer
DOI: 10.1007/978-3-031-29126-5_8
Abstract
Deep Belief Net (DBN) was applied to the field of time series forecasting in our early works. In this paper, we propose to adopt Adaptive Moment Estimation (Adam) optimization method to the fine-tuning process of DBN instead of the conventional Error Back-Propagation (BP) method. Meta parameters, such as the number of layers of Restricted Boltzmann Machine (RBM), the number of units in each layer, the learning rate, are optimized by Random Search (RS) or Particle Swarm Optimization (PSO). Comparison experiments showed the priority of the proposed method in both cases of a benchmark dataset CATS which is an artificial time series data used in competitions for long-term forecasting, and Lorenz chaos for short-term forecasting in the sense not only prediction precision but also learning performance.
