PSO-MISMO modeling strategy for multistep-ahead time series prediction

Area of Science:

• Artificial Intelligence
• Machine Learning
• Data Science

Background:

• Multistep-ahead time series prediction is a complex challenge in time series modeling.
• The multiple-input several multiple-outputs (MISMO) strategy offers advantages over iterated and direct methods.
• Existing MISMO strategies often use rigid divisions for prediction horizons.

Purpose of the Study:

• To propose a novel particle swarm optimization-based MISMO (PSO-MISMO) strategy.
• To enable self-adaptive determination of sub-models and flexible prediction horizons.
• To enhance the flexibility and performance of multistep-ahead time series prediction models.

Main Methods:

• Implementation of a particle swarm optimization (PSO) algorithm for MISMO modeling.
• Development of a heuristic for creating flexible, varying-sized prediction horizon divides.
• Integration of neural networks within the proposed PSO-MISMO framework.

Main Results:

• The PSO-MISMO strategy demonstrated self-adaptive capability in determining sub-model numbers.
• Flexible division of prediction horizons with varying sizes was successfully implemented.
• Validation with simulated and real-world datasets confirmed the strategy's effectiveness.

Conclusions:

• The proposed PSO-MISMO strategy provides significant flexibility in time series model construction.
• This approach offers a promising advancement for challenging multistep-ahead prediction tasks.
• The method's adaptability makes it suitable for diverse time series prediction applications.