This paper provides a comprehensive review of deep learning models for time series forecasting (TSF), highlighting recent developments and challenges. The authors introduce the latest trends in TSF, propose a new taxonomy of deep neural network models, and organize commonly used experimental evaluation metrics and datasets. They discuss the limitations of existing solutions and suggest future research directions. The paper covers four main categories of models: RNN-based, Transformer-based, CNN-based, and MLP-based. Each category is detailed with representative models, their innovations, and trends over the past five years. The authors also address common issues in TSF, such as handling large-scale data, achieving longer forecasting ranges, and reducing computational complexity. The paper aims to serve as a practical guide for researchers and practitioners in the field of TSF, providing insights into the latest advancements and potential future directions.This paper provides a comprehensive review of deep learning models for time series forecasting (TSF), highlighting recent developments and challenges. The authors introduce the latest trends in TSF, propose a new taxonomy of deep neural network models, and organize commonly used experimental evaluation metrics and datasets. They discuss the limitations of existing solutions and suggest future research directions. The paper covers four main categories of models: RNN-based, Transformer-based, CNN-based, and MLP-based. Each category is detailed with representative models, their innovations, and trends over the past five years. The authors also address common issues in TSF, such as handling large-scale data, achieving longer forecasting ranges, and reducing computational complexity. The paper aims to serve as a practical guide for researchers and practitioners in the field of TSF, providing insights into the latest advancements and potential future directions.