.png){kind=icon}
.png){kind=icon}
Return
Distributed hyperparameter optimization based multivariate time series forecasting
.png){kind=icon}
PRE
AI
DOI：10.1007/s11042-023-15456-0.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Various domains generate continuous data in the form of multivariate time series (MTS), including electricity and stock trading. It is difficult to maintain a low variance in forecasting error while forecasting multi-step MTS. Deep learning models can forecast large and complex multivariate time series with high accuracy, but simulating the hyperparameters for a model to achieve high accuracy is time-consuming. Figuring out the appropriate hyperparameters becomes increasingly stagnant and resource-intensive as the dataset size increases. In this study, we propose a distributed random parameter optimization model to deal with short-term multivariate time series forecasting problems. It utilizes the deep neural network for forecasting in synchronization with the hyperparameter tuning optimization in a distributed manner to exploit the resources and speed up the computation. The proposed model has been validated against a real-world multivariate dataset of electricity consumption and weather conditions for the Canadian province of Ontario from 2010 to 2019. Extensive experiments on the Ontario region demonstrate that the proposed model is scalable and outperforms traditional models such as vector autoregression(VAR), Multilayer Perceptron(MLP), Hidden Markov Model (HMM), and Long Short Term Memory (LSTM). Furthermore, the model outperformed all others in terms of accuracy on a benchmark Spanish electricity dataset.
Keywords:
Recurrent neural network
Information retrieval
LSTM
HMM
Hyperparameter optimization
Journal
IF:
3
Papers:
1.9W
Citations:
3.2W