The Autoencoder (AE) network is an effective tool for feature extraction, extracting latent features from raw data by minimizing reconstruction error. Over the past few years, many AE-based deep clustering models have been developed to optimize deep neural networks and perform unsupervised clustering simultaneously. Nevertheless, these models directly conduct unsupervised clustering on the latent features extracted by AE, neglecting the potential conflict between clustering and representation learning. The goal of clustering is to improve data aggregation to facilitate the clustering process, while representation learning aims to ensure that the latent features effectively capture the manifold structure of the original data. To address this conflict, this paper proposes a novel expanded deep embedding clustering model. The AE network is employed for representation learning, while an innovative residual expansion module is incorporated to help build a new extended feature space that facilitates the clustering process. The experimental results of the proposed model on commonly used datasets demonstrate its efficiency compared to state-of-the-art methods.