Scholarlink

<p>Accurate quality prediction is crucial for the optimization of process industry production. However, the multi-scale data imbalance problem impedes the efficacy of data driven quality prediction model. This paper introduces a multi-model adversarial and collaborative forecasting network (MACFN) to solve this challenge. The proposed model features a Multi-scale Time Series Generative Adversarial Network (MSTSGAN) and a predictor. The MSTSGAN consists of a generator and a shared discriminator, while the predictor is built upon a Convolutional Neural Network cascaded with a Bidirectional Long Short-Term Memory network (CNN-Bi-LSTM). They form two dual-model systems named adversarial learning generation system (ALGS) and collaborative prediction system (CPS), respectively. In ALGS, generator and discriminator learn features of the actual production data in game style and then produce a large number of pseudo samples to enhance the actual imbalanced data. In CPS, leveraging multi-scale data information from the discriminator, predictor identifies the functional relationships between production variables and quality indices in the enhanced data space, thereby improving its forecasting performance. The proposed model’s efficacy is validated by quality prediction experiments implemented by actual production data from a cement manufacturing enterprise, which exemplifies the potential for quality prediction under multi-scale data imbalance problem.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Multi-model adversarial and collaborative forecasting network for quality prediction modeling of high dimensional imbalanced sequences in process industry

  • 期刊论文

  • 发表日期：

    2026年4月14日

查看全文
  • Gaolu Huang,
  • Xin Shi,
  • Xiaochen Hao,
  • Junze Jiao,
  • Xiaodie Ren

摘要

Accurate quality prediction is crucial for the optimization of process industry production. However, the multi-scale data imbalance problem impedes the efficacy of data driven quality prediction model. This paper introduces a multi-model adversarial and collaborative forecasting network (MACFN) to solve this challenge. The proposed model features a Multi-scale Time Series Generative Adversarial Network (MSTSGAN) and a predictor. The MSTSGAN consists of a generator and a shared discriminator, while the predictor is built upon a Convolutional Neural Network cascaded with a Bidirectional Long Short-Term Memory network (CNN-Bi-LSTM). They form two dual-model systems named adversarial learning generation system (ALGS) and collaborative prediction system (CPS), respectively. In ALGS, generator and discriminator learn features of the actual production data in game style and then produce a large number of pseudo samples to enhance the actual imbalanced data. In CPS, leveraging multi-scale data information from the discriminator, predictor identifies the functional relationships between production variables and quality indices in the enhanced data space, thereby improving its forecasting performance. The proposed model’s efficacy is validated by quality prediction experiments implemented by actual production data from a cement manufacturing enterprise, which exemplifies the potential for quality prediction under multi-scale data imbalance problem.