Sparse Data-Driven Model for Monitoring of Industrial Process Based on Distributed Feature Extraction Neural Network

Modern industrial processes store a large amount of process data in the integration of subsystems or subprocesses, which creates conditions for data-driven models. However, due to a few features in various tasks possessing a direct correlation, there are many variables and complex relationships, which may result in incomplete input information. To address this problem, this paper proposes a task decomposition and feature integration-based distributed process monitoring model. Firstly, a sparse subspace clustering algorithm is introduced for task decomposition. This algorithm divides the original space into several interactive feature subspaces and allocates weights to quantify the contribution of the subtask simultaneously. Secondly, based on the divided features, a distributed framework for spatial feature integration is proposed. The framework constructs a differentiated parallel coding network by designing a structural self-organization mechanism, which achieves feature extraction and fusion of each subspace. Finally, a collaborative optimization algorithm is proposed to optimize the network parameters of each sub-model at the same time to ensure the accuracy of the model. To demonstrate the effectiveness of this data modeling method, we tested it in several benchmark data sets and a high-dimensional nonlinear system. The experimental results show that the model has better performance in data dimensionality reduction.