This paper provides a comprehensive review of self-supervised learning (SSL) for electroencephalogram (EEG) signals, addressing the challenges of limited labeled data and label shifts in EEG analysis. The authors introduce the concept and theory of SSL, including typical SSL frameworks, and provide a detailed taxonomy of existing EEG-based SSL methods. They categorize these methods into four types: prediction-based, generation-based, contrastive-based, and hybrid SSL methods. The paper also discusses the adaptation of SSL approaches to various downstream tasks and the potential future directions for SSL-EEG research. The authors highlight the importance of SSL in mitigating the issues of inconsistent labeling and improving generalization in EEG analysis. The review covers the mathematical definitions, loss functions, and specific tasks within each category, providing a detailed analysis of the current state and future opportunities in SSL for EEG signals.This paper provides a comprehensive review of self-supervised learning (SSL) for electroencephalogram (EEG) signals, addressing the challenges of limited labeled data and label shifts in EEG analysis. The authors introduce the concept and theory of SSL, including typical SSL frameworks, and provide a detailed taxonomy of existing EEG-based SSL methods. They categorize these methods into four types: prediction-based, generation-based, contrastive-based, and hybrid SSL methods. The paper also discusses the adaptation of SSL approaches to various downstream tasks and the potential future directions for SSL-EEG research. The authors highlight the importance of SSL in mitigating the issues of inconsistent labeling and improving generalization in EEG analysis. The review covers the mathematical definitions, loss functions, and specific tasks within each category, providing a detailed analysis of the current state and future opportunities in SSL for EEG signals.