摘要
研究提出了一种基于深度学习的湿法冶金全流程分层优化模型。模型采用分层设计,包括工序层优化和过程层优化。在工序层创新引入Transformer模型,建立物耗与质量指标的非线性映射模型,以最小物耗为目标优化各工序参数。在过程层创新引入DDPG算法调整操作变量,实现全流程经济效益最大化。此外,引入生成对抗网络(GAN)生成优化样本,进一步提升优化模型的鲁棒性。基于Transformer的优化模型实时调整氰化钠和锌粉添加量后,金产量明显提升,物料消耗成本显著降低;优化试验与基准试验相比,经济效益提高15%~20%,浸出率和置换率分别提高2%~4%。该模型在经济效益、工艺稳定性及不确定性适应能力方面均优于传统深度学习方法,可为湿法冶金行业的智能化生产提供一种新途径。
A layered optimization model of hydrometallurgy process based on deep learning is proposed.The model adopts hierarchical design,including process level optimization and process level optimization.At the process level,Transformer model is innovatively introduced to establish a nonlinear mapping model of material consumption and quality index,and optimize each process parameter with minimum material consumption as the goal.In the process layer,the innovative DDPG algorithm is introduced to adjust the operating variables and maximize the economic benefits of the whole process.In addition,generative adversarial network(GAN)is introduced to generate optimization samples to further improve the robustness of the optimization model.After the sodium cyanide and zinc powder additions are adjusted in real time by the Transformer optimization model,the gold production can be significantly increased and the material consumption cost can be significantly reduced.Compared with the benchmark test,the economic benefit increased by 15%-20%,the leaching rate and the replacement rate increased by 2%-4%respectively.The model is superior to the traditional deep learning method in terms of economic benefit,process stability and uncertainty adaptability,and can provide a new way for intelligent production in hydrometallurgical industry.
作者
何丹
褚新建
马晓晓
贾钰峰
HE Dan;CHU Xinjian;MA Xiaoxiao;JIA Yufeng(College of Information Science and Engineering,Xinjiang University of Science and Technology,Korla 841000,China)
出处
《湿法冶金》
北大核心
2025年第6期824-832,共9页
Hydrometallurgy of China
基金
新疆维吾尔自治区自然科学基金资助项目(2023D01B23)。
