复杂工业过程因涉及多种物理/化学反应,其质量指标或环保指标等难测参数的精确数学模型难以构建.常用的基于神经网络的数据驱动建模方法存在可解释性差、样本需求量大等缺点.针对上述问题,提出了一种非神经网络模式的深度集成森林回归(d...复杂工业过程因涉及多种物理/化学反应,其质量指标或环保指标等难测参数的精确数学模型难以构建.常用的基于神经网络的数据驱动建模方法存在可解释性差、样本需求量大等缺点.针对上述问题,提出了一种非神经网络模式的深度集成森林回归(deep ensemble forest regression,DEFR)建模方法.首先,基于样本空间和特征空间的随机采样策略获得训练子集后构建T个基于决策树(decision trees,DT)的子森林模型,将采用K最近邻(K-nearest neighbor,KNN)准则选取的层回归向量与原始特征组合获得的增强层回归向量作为输入层森林模型的输出;然后,采用相同方式构建包含若干预设层数的中间层森林模型;最后,基于上层增强层回归向量构建输出层的子森林模型,通过对其T个输出值的加权获得DEFR模型的预测值.采用加州大学欧文分校(University of California Irvine,UCI)平台混凝土抗压强度数据和城市固废焚烧过程的二口恶英排放质量浓度数据仿真验证了所提方法的有效性.展开更多
针对递归RBF神经网络结构难以自适应问题,提出一种基于递归正交最小二乘(recursive orthogonal least squares,ROLS)算法的结构设计方法。首先,利用ROLS算法来计算隐含层神经元的独立贡献度和损失函数,以此判断增加或归为不活跃组的神经...针对递归RBF神经网络结构难以自适应问题,提出一种基于递归正交最小二乘(recursive orthogonal least squares,ROLS)算法的结构设计方法。首先,利用ROLS算法来计算隐含层神经元的独立贡献度和损失函数,以此判断增加或归为不活跃组的神经元,同时调整神经网络的拓扑结构,并且利用奇异值分解(singular value decomposition,SVD)决定最佳的隐含层神经元个数,以此来删除不活跃组中相对不活跃的神经元,有效地解决了递归RBF神经网络结构冗余和难以自适应问题。其次,利用梯度下降算法更新递归RBF神经网络的参数来保证神经网络的精度。最后,通过对Mackey-Glass时间序列预测、非线性系统辨识和污水处理过程中关键水质参数动态建模,证明了该结构设计方法的可行性和有效性。展开更多
针对小脑模型神经网络(cerebellar model neural network,CMNN)中泛化能力与存储空间容量之间的冲突这一关键问题,提出了一种改进的小脑模型神经网络——模糊隶属度小脑模型神经网络(fuzzy membership cerebellar model neural network,...针对小脑模型神经网络(cerebellar model neural network,CMNN)中泛化能力与存储空间容量之间的冲突这一关键问题,提出了一种改进的小脑模型神经网络——模糊隶属度小脑模型神经网络(fuzzy membership cerebellar model neural network,FM-CMNN),用于解决非线性动态系统的时间序列预测问题.首先,FM-CMNN在保留原始CMNN输入变量的地址映射方式的情况下,在CMNN存储空间中引入铃型模糊隶属度函数,从而保证在不需增加量化级数的情况下提高网络的泛化能力.然后,使用梯度下降算法对网络权值进行更新,提高网络的逼近强度.最后,通过非线性时间序列预测基准实验和污水处理中水质参数预测实验,验证了FM-CMNN性能的可靠性.展开更多
Smoke and fire recognition are of great importance on foreseeing fire disasters and preventing environmental pollution by monitoring the burning process of objects(e.g., straw, fuels). However, since fire images suffe...Smoke and fire recognition are of great importance on foreseeing fire disasters and preventing environmental pollution by monitoring the burning process of objects(e.g., straw, fuels). However, since fire images suffer from problems like the variability of the features, complexity of scenarios, interference from background, changeable weather conditions as well as image quality problems, identifying smoke and fire accurately and promptly from a given image still remains a substantial challenge. Automatically learning the features of smoke images by CNNs has improved the target recognition ability compared to traditional approaches,nonetheless, convolutions and pooling operations in CNNs may cause severe information loss which may lead to misjudgment.To tackle the above problems, this paper proposed a hybrid attention model based on the characteristics of smoke images. This model adopted multiple optimized attention mechanism in several stages to quickly and precisely capture the important features,achieving state-of-the-art performance on smoke and fire recognition in terms of accuracy and speed. Our proposed module mainly consists of two stages: pooling and attention. In the first stage, we conducted several newly proposed first-order pooling methods. Through traversing the data space in a larger scope, features are better reserved, thus constructing a more intact feature space of smoke and fire in an image. In the second stage, feature maps are aggregated together to perform channel and spatial attention. The channel and spatial dependencies allow us to quickly catch the important features presented in an image. By fully exploring the feature space and prominent salient features, characteristics of smoke and fire are better presented so as to obtain better smoke and fire detection results. Experiments have been conducted on public smoke detection dataset and new proposed fine-grained smoke and fire detection database. Experimental results revealed that the proposed method outperformed popular deep CNNs and existing prevalent attention models for smoke and fire detection problems.展开更多
Due to the large uncertainty in the municipal solid waste incineration(MSWI)process,the furnace temperature of the MSWI process is difficult to control and the controller is updated frequently.To improve the accuracy ...Due to the large uncertainty in the municipal solid waste incineration(MSWI)process,the furnace temperature of the MSWI process is difficult to control and the controller is updated frequently.To improve the accuracy and reduce the number of controller updates,a novel event-triggered control method based correntropy self-organizing TS fuzzy neural network(ETCSTSFNN)is proposed.Firstly,the neurons of the rule layer are grown or pruned adaptively based on activation intensity and control error to meet the dynamic change of the actual operating condition.Meanwhile,the performance index is designed based on the correntropy of tracking errors,and the parameters of the controller are adjusted by gradient descent algorithm.Secondly,a fixed threshold event-triggered condition is designed to determine whether the current controller is updated or not.The stability of the control system is proved based on the Lyapunov stability theory.Finally,the furnace temperature control experiments are conducted based on the actual data of a municipal solid waste incineration plant in Beijing.The experimental results show that the proposed ET-CSTSFNN controller shows a better control performance,which can reduce the number of the controller update significantly while achieving accurate furnace temperature control compared with other traditional control methods.展开更多
Because of coupling,nonlinearity,and uncertainty in a municipal solid waste incineration(MSWI)process,a suitable multivariable controller is difficult to establish under strong disturbance.Additionally,the problems of...Because of coupling,nonlinearity,and uncertainty in a municipal solid waste incineration(MSWI)process,a suitable multivariable controller is difficult to establish under strong disturbance.Additionally,the problems of reducing mechanical wear and energy consumption in the control process should also be considered.To solve these problems,an event-triggered fuzzy neural multivariable controller is proposed in this paper.First,the MSWI object model based on the multiinput multioutput TakagiSugeno fuzzy neural network is established using a data-driven method.Second,a fuzzy neural multivariable controller is designed to control the furnace temperature and flue gas oxygen content synchronously under external disturbance.Third,an event-triggered mechanism is constructed to update the control rate online while ensuring control effects.Then,the stability of the proposed control strategy is proven through the LyapunovⅡtheorem to guide its practical application.Finally,the effectiveness of the controller is verified using the real industrial data of an MSWI factory in Beijing,China.The experimental results show that the proposed control strategy greatly improves the control efficiency,reduces energy consumption by 66.23%,and improves the multivariable tracking control accuracy.展开更多
In this article, the dissolved oxygen(DO) concentration control problem in wastewater treatment process(WWTP) is studied.Unlike existing control strategies that control DO concentration at a fixed value, here we devel...In this article, the dissolved oxygen(DO) concentration control problem in wastewater treatment process(WWTP) is studied.Unlike existing control strategies that control DO concentration at a fixed value, here we develop a different control framework.Under the proposed control framework, an intelligent control method of DO concentration based on reinforcement learning(RL)algorithm is presented to resolve the DO concentration control problem. By using the deep deterministic policy gradient(DDPG)algorithm, the DO concentration of the fifth tank in the activated sludge reactor can be adjusted dynamically. In addition, by designing two different reward functions and by analysing the relationships among effluent quality, energy consumption, and DO concentration, the target of energy-saving and emission-reducing is achieved. The simulation results indicate that the designed control method can reduce energy consumption while ensuring that the effluent quality meet the specified standards.展开更多
文摘针对污水处理复杂系统中关键水质参数生化需氧量(biochemical oxygen demand,BOD)难以准确实时预测的问题,在分析污水处理过程相关影响因素的基础上,提出一种基于敏感度分析法的自组织随机权神经网络(selforganizing neural network with random weights,SONNRW)软测量方法.该方法首先通过机理分析选取原始辅助变量,经过数据预处理,之后采用主元分析法对辅助变量进行精选,作为SONNRW的输入变量进行污水处理关键水质参数BOD的预测.SONNRW算法利用隐含层节点输出及其权值向量计算该隐含层节点对于残差的敏感度,根据敏感度大小对网络隐含层节点进行排序,删除敏感度较低的隐含层节点即冗余点.仿真结果表明:该软测量方法对水质参数BOD的预测精度高、实时性好、模型结构稳定,能够用于污水水质的在线预测.
文摘复杂工业过程因涉及多种物理/化学反应,其质量指标或环保指标等难测参数的精确数学模型难以构建.常用的基于神经网络的数据驱动建模方法存在可解释性差、样本需求量大等缺点.针对上述问题,提出了一种非神经网络模式的深度集成森林回归(deep ensemble forest regression,DEFR)建模方法.首先,基于样本空间和特征空间的随机采样策略获得训练子集后构建T个基于决策树(decision trees,DT)的子森林模型,将采用K最近邻(K-nearest neighbor,KNN)准则选取的层回归向量与原始特征组合获得的增强层回归向量作为输入层森林模型的输出;然后,采用相同方式构建包含若干预设层数的中间层森林模型;最后,基于上层增强层回归向量构建输出层的子森林模型,通过对其T个输出值的加权获得DEFR模型的预测值.采用加州大学欧文分校(University of California Irvine,UCI)平台混凝土抗压强度数据和城市固废焚烧过程的二口恶英排放质量浓度数据仿真验证了所提方法的有效性.
文摘针对递归RBF神经网络结构难以自适应问题,提出一种基于递归正交最小二乘(recursive orthogonal least squares,ROLS)算法的结构设计方法。首先,利用ROLS算法来计算隐含层神经元的独立贡献度和损失函数,以此判断增加或归为不活跃组的神经元,同时调整神经网络的拓扑结构,并且利用奇异值分解(singular value decomposition,SVD)决定最佳的隐含层神经元个数,以此来删除不活跃组中相对不活跃的神经元,有效地解决了递归RBF神经网络结构冗余和难以自适应问题。其次,利用梯度下降算法更新递归RBF神经网络的参数来保证神经网络的精度。最后,通过对Mackey-Glass时间序列预测、非线性系统辨识和污水处理过程中关键水质参数动态建模,证明了该结构设计方法的可行性和有效性。
文摘针对小脑模型神经网络(cerebellar model neural network,CMNN)中泛化能力与存储空间容量之间的冲突这一关键问题,提出了一种改进的小脑模型神经网络——模糊隶属度小脑模型神经网络(fuzzy membership cerebellar model neural network,FM-CMNN),用于解决非线性动态系统的时间序列预测问题.首先,FM-CMNN在保留原始CMNN输入变量的地址映射方式的情况下,在CMNN存储空间中引入铃型模糊隶属度函数,从而保证在不需增加量化级数的情况下提高网络的泛化能力.然后,使用梯度下降算法对网络权值进行更新,提高网络的逼近强度.最后,通过非线性时间序列预测基准实验和污水处理中水质参数预测实验,验证了FM-CMNN性能的可靠性.
基金supported by the National Key Research and Development Program of China(Grant No. 2021ZD0112302)the National Natural Science Foundation of China(Grant Nos. 62076013, 62021003, 61890935)CAAI-Huawei MindSpore Open Fund(Grant No. CAAIXSJLJJ-2021-016A)。
文摘Smoke and fire recognition are of great importance on foreseeing fire disasters and preventing environmental pollution by monitoring the burning process of objects(e.g., straw, fuels). However, since fire images suffer from problems like the variability of the features, complexity of scenarios, interference from background, changeable weather conditions as well as image quality problems, identifying smoke and fire accurately and promptly from a given image still remains a substantial challenge. Automatically learning the features of smoke images by CNNs has improved the target recognition ability compared to traditional approaches,nonetheless, convolutions and pooling operations in CNNs may cause severe information loss which may lead to misjudgment.To tackle the above problems, this paper proposed a hybrid attention model based on the characteristics of smoke images. This model adopted multiple optimized attention mechanism in several stages to quickly and precisely capture the important features,achieving state-of-the-art performance on smoke and fire recognition in terms of accuracy and speed. Our proposed module mainly consists of two stages: pooling and attention. In the first stage, we conducted several newly proposed first-order pooling methods. Through traversing the data space in a larger scope, features are better reserved, thus constructing a more intact feature space of smoke and fire in an image. In the second stage, feature maps are aggregated together to perform channel and spatial attention. The channel and spatial dependencies allow us to quickly catch the important features presented in an image. By fully exploring the feature space and prominent salient features, characteristics of smoke and fire are better presented so as to obtain better smoke and fire detection results. Experiments have been conducted on public smoke detection dataset and new proposed fine-grained smoke and fire detection database. Experimental results revealed that the proposed method outperformed popular deep CNNs and existing prevalent attention models for smoke and fire detection problems.
基金supported by the National National Science Foundation of China(Grant Nos.62073006 and 62021003)Beijing Natural Science Foundation(Grant No.4212032)。
文摘Due to the large uncertainty in the municipal solid waste incineration(MSWI)process,the furnace temperature of the MSWI process is difficult to control and the controller is updated frequently.To improve the accuracy and reduce the number of controller updates,a novel event-triggered control method based correntropy self-organizing TS fuzzy neural network(ETCSTSFNN)is proposed.Firstly,the neurons of the rule layer are grown or pruned adaptively based on activation intensity and control error to meet the dynamic change of the actual operating condition.Meanwhile,the performance index is designed based on the correntropy of tracking errors,and the parameters of the controller are adjusted by gradient descent algorithm.Secondly,a fixed threshold event-triggered condition is designed to determine whether the current controller is updated or not.The stability of the control system is proved based on the Lyapunov stability theory.Finally,the furnace temperature control experiments are conducted based on the actual data of a municipal solid waste incineration plant in Beijing.The experimental results show that the proposed ET-CSTSFNN controller shows a better control performance,which can reduce the number of the controller update significantly while achieving accurate furnace temperature control compared with other traditional control methods.
基金supported by the Science and Technology Innovation 2030-“New Generation Artificial Intelligence”Major Project of China(Grant No.2021ZD0112300)the Innovative Research Group Project of the National Natural Science Foundation of China(Grant No.62021003)+1 种基金the National Natural Science Foundation of China(Grant No.62073006)the Natural Science Foundation of Beijing(Grant Nos.4212032 and4192009)。
文摘Because of coupling,nonlinearity,and uncertainty in a municipal solid waste incineration(MSWI)process,a suitable multivariable controller is difficult to establish under strong disturbance.Additionally,the problems of reducing mechanical wear and energy consumption in the control process should also be considered.To solve these problems,an event-triggered fuzzy neural multivariable controller is proposed in this paper.First,the MSWI object model based on the multiinput multioutput TakagiSugeno fuzzy neural network is established using a data-driven method.Second,a fuzzy neural multivariable controller is designed to control the furnace temperature and flue gas oxygen content synchronously under external disturbance.Third,an event-triggered mechanism is constructed to update the control rate online while ensuring control effects.Then,the stability of the proposed control strategy is proven through the LyapunovⅡtheorem to guide its practical application.Finally,the effectiveness of the controller is verified using the real industrial data of an MSWI factory in Beijing,China.The experimental results show that the proposed control strategy greatly improves the control efficiency,reduces energy consumption by 66.23%,and improves the multivariable tracking control accuracy.
基金supported by the National Natural Science Foundation of China(Grant No.62173009)the National Key Research and Development Program of China(Grant No.2021ZD0112302)。
文摘In this article, the dissolved oxygen(DO) concentration control problem in wastewater treatment process(WWTP) is studied.Unlike existing control strategies that control DO concentration at a fixed value, here we develop a different control framework.Under the proposed control framework, an intelligent control method of DO concentration based on reinforcement learning(RL)algorithm is presented to resolve the DO concentration control problem. By using the deep deterministic policy gradient(DDPG)algorithm, the DO concentration of the fifth tank in the activated sludge reactor can be adjusted dynamically. In addition, by designing two different reward functions and by analysing the relationships among effluent quality, energy consumption, and DO concentration, the target of energy-saving and emission-reducing is achieved. The simulation results indicate that the designed control method can reduce energy consumption while ensuring that the effluent quality meet the specified standards.
