| 引用本文: | 温在鑫,钱斌,胡蓉,金怀平,杨媛媛.基于学习型人工蜂群算法优化双向GRU的乙烯产率预测[J].控制理论与应用,2023,40(10):1746~1756.[点击复制] |
| WEN Zai-xin,QIAN Bin,HU Rong,JIN Huai-ping,YANG Yuan-yuan.Ethylene yield prediction based on bi-directional GRU optimized by learning-based artificial bee colony algorithm[J].Control Theory and Technology,2023,40(10):1746~1756.[点击复制] |
|
| 基于学习型人工蜂群算法优化双向GRU的乙烯产率预测 |
| Ethylene yield prediction based on bi-directional GRU optimized by learning-based artificial bee colony algorithm |
| 摘要点击 645 全文点击 238 投稿时间：2022-03-24 修订日期：2023-09-08 |
| 查看全文 查看/发表评论 下载PDF阅读器 |
| DOI编号 10.7641/CTA.2022.20211 |
| 2023,40(10):1746-1756 |
| 中文关键词 深度强化学习 双向GRU 人工蜂群算法 乙烯裂解炉 生产能力预测 |
| 英文关键词 deep reinforcement learning bi-directional GRU ABC ethylene cracking furnace production capacity prediction |
| 基金项目 国家自然科学基金项目(62173169, 61963022), 云南省基础研究重点项目(202201AS070030) |
|
| 中文摘要 |
| 本文针对以乙烯产率为生产指标的预测问题, 基于双向门控循环单元网络(BGRU)建立乙烯产率预测模型,以最小化模型误差为优化目标并提出一种学习型人工蜂群算法 (LABC) 对预测模型进行优化和设计. 在构建BGRU预测模型时, 先对乙烯裂解炉实际生产过程进行分析, 确定影响产率的关键因素并将其作为模型的输入; 再采用LABC对BGRU网络模型的结构、初始权值和阈值、训练比和动量因子进行全面的优化和设计. 在LABC中, 首先根据人工蜂群算法(ABC)特点构建强化学习(RL)框架下的状态集、动作集、奖励函数和最优混合搜索策略, 在此基础上, 提出一种深度双Q网络(DDQN) 来实现最优混合搜索策略, 通过该策略可智能选择合适的搜索动作来执行针对不同状态的局部搜索. 本文通过在标准数据集和实际生产数据上的测试及算法对比, 验证了所提学习型人工蜂群算法优化的双向GRU 网络(LABC BGRU)模型具有预测精度高、适用性强的特性. |
| 英文摘要 |
| Aiming at prediction problem that takes ethylene yield as production index, this paper establishes the ethylene yield prediction model based on the bi-directional gated recurrent neural network (BGRU), a learning based artificial bee colony algorithm (LABC) is proposed to optimize and design the prediction model with the goal of minimizing model error. When constructing the BGRU prediction model, the actual production process of ethylene cracking furnace is analyzed to determine the key factors that affect the yield and take them as the input of the model. In addition, LABC is designed to comprehensively evolve and design the structure, initial weight and threshold, training ratio and momentum factor of the BGRU model. In LABC, the state set, action set, reward function and optimal hybrid search strategy in reinforcement learning framework are constructed according to the characteristics of artificial bee colony algorithm (ABC), on this basis, a new deep double Q network (DDQN) is proposed to realize the optimal hybrid search strategy. Through this strategy, appropriate search actions can be intelligently selected to perform local search for different states. Results of experiments and comparisons on actual production data and standard data set demonstrate that LABC BGRU model has the characteristics of high prediction accuracy and strong applicability. |
|
|
|
|
|