Bidirectional Multiscale Efficient Dilated Convolutional Recurrent Neural Network Improved by Swarm Intelligence...

Area of Science:

• Artificial Intelligence
• Machine Learning
• Deep Learning

Background:

• Bidirectional convolutional recurrent neural networks (RNNs) show promise for time series and prediction tasks.
• Model performance heavily relies on optimal hyperparameter selection, which is often challenging and inefficient.

Purpose of the Study:

• To develop an automatic hyperparameter optimization method for bidirectional convolutional RNNs.
• To improve regression prediction accuracy using an enhanced model and swarm intelligence optimization.

Main Methods:

• A novel Parallel Multiscale Dilated Convolution (PMDC) module was designed to capture local and global spatial correlations.
• Bidirectional Gated Recurrent Units (BGRUs) were employed to extract temporal information from convolutional features.
• A pretrained Sparrow Search Algorithm (SSA) was integrated for automatic hyperparameter optimization of the PMDC-BGRU model.

Main Results:

• The proposed PMDC-BGRU model integrated with SSA demonstrated superior performance in regression prediction tasks.
• Experimental results on multiple datasets validated the effectiveness of the automated hyperparameter optimization approach.
• The study highlighted the flexibility of intelligent optimization algorithms in solving complex model parameter optimization problems.

Conclusions:

• The automated hyperparameter optimization using SSA significantly enhances the performance of bidirectional convolutional RNNs for prediction tasks.
• The PMDC-BGRU model offers an effective architecture for capturing complex spatial and temporal dependencies in time series data.
• Intelligent optimization algorithms provide a flexible and efficient solution for tuning deep learning models.