Sustainable Clustering Neural Networks

In the modern field of power and renewable energy, advancements in data analysis technology are key to driving innovation and sustainable development. However, existing clustering algorithms and neural networks are inefficient and inaccurate when processing large-scale, high-dimensional data, and they consume significant computational resources. To address these issues, this paper proposes a Sustainable Clustering Neural Network (SCNN). By introducing Gaussian units based on joint Gaussian distribution and combining them with a max-pooling layer, we propose a universal clustering module with a competitive update mechanism. This module includes two sets of trainable parameters, allowing for flexible adjustment of the shape and position of the Gaussian distribution to fit different data distribution characteristics. During clustering, feature selection and dimensionality reduction are achieved through the max-pooling layer, effectively enhancing the model's clustering performance. The competitive update mechanism further promotes competition among Gaussian units, enabling each unit to focus more on specific clusters, thereby improving the accuracy and stability of the clustering results. Experiments on the MNIST and Fashion-MNIST datasets achieved clustering accuracies of 93.38% and 72.83%, respectively, demonstrating strong competitiveness compared to existing methods. The code is available at: https://github.com/ZS520L/HGL-CAE.