超低能耗铝电解过程动态演化建模与决策参数稳健优化

The issue of energy consumption plays a key role in aluminum electrolytic production process. An effective way named as new type cathode technology for reducing the energy consumption is to lower the velocity of liquid metal and weaken the surface oscillation of liquid metal. Similarly, another technology called perforated anode can diminish the gas bubbles, by which the carbon dioxide gas at the bottom of the anode can be discharged through some designed hole. It means that the energy consumption can be decreased. Our research group deduces that the superposition effect of above two new technologies can allow about six percents of energy efficiency rate. However,there is a phenomenon that the system becomes very senstitive to uncetainties from system disturbance and state change and its interaction. In this case, new technologies don't match three field model including the thermal field, electromagnetic field and flow field of liquid metal,two phase of gas-liquid flow patterns, solubility and feeding style of alumina when the cell voltage is reduced to the certain limit in the experiment. In the project, the dynamic evolution mechanism of process model would be thoroughly studied under the conditions of multiple uncetainties from system disturbance and state change and its interaction. dyamic response charactersistics and coupling analysis method at low-temperature and low-voltage state is employeed to reveal the law of dynamic evolution, and visual decision variable selection algorithm based on the rate of contribution to the stability is proposed. Adaptive dynamic evolution modeling method with model error in critical condition is proposed after analysis of the impact of the process noise and measurement noise on the dynamic model error.In order to achieve the purpose that target performance is insensitive to the change of decision parameters, multi-objective robust optimization algorithm is proposed after calculating robust optimization target and operating feasibility determination and soft constraints adjustment. .This project can provide new method to solve control problems for ultra-low energy consumption under multiple uncertain conditions in aluminum electrolysis industrial.

铝电解生产耗能巨大。异型阴极和穿孔阳极分别通过减薄铝液波动层和气泡层厚度降低槽电压，节能效果显著。理论计算两项新技术的叠加效应可使工艺能量利用率提高6%左右。但试验发现当极距与磁流体稳定性处于临界状态时，槽内三场分布模型、阳极气液两相流型态与新工艺严重失配，难以实现超低能耗目标。课题以两项新技术联用的预焙槽为研究对象，系统深入地研究受环境扰动与状态的耦合变化带来的多重不确定性影响的系统模型动态演化规律问题。研究低温低电压状态观测值动态响应及耦合性特性，考察工艺参数对各稳定性的贡献率；分析过程噪声和观测噪声对动态模型误差的影响，建立临界状态下带模型误差的动态演化模型；为使得目标性能对决策参数的变差不灵敏，通过稳健优化目标计算、可行性判定和软约束调整，实施多目标决策参数稳健优化。本课题提出的动态演化建模与决策参数优化方法，可为解决超低能耗铝电解生产中多重不确定性条件下优化控制问题提供新的思路。

铝电解生产耗能巨大。异型阴极和穿孔阳极分别通过减薄铝液波动层和气泡层厚度降低槽电压，节能效果显著。课题以两项新技术联用的预焙槽为研究对象，系统深入地研究受环境扰动与状态的耦合变化带来的多重不确定性影响的系统模型动态演化规律问题。.本项目在以下四个方面取得了进展：（1）在过程建模方面：提出强跟踪平方根无迹Kalman 神经网络(STR-UKFNN), 并用其建立铝电解槽工艺能耗的动态演化模型. 该方法利用误差协方差矩阵的平方根代替UKFNN 算法中的协方差阵, 避免误差协方差矩阵可能出现负定而导致滤波发散, 并在UKFNN 算法中引入渐消因子和弱化因子, 实时调整滤波增益, 提高模型收敛速度和其对突变状态的跟踪能力。（2）在多参数相关性分析方面：提出一种基于混合Copula 模型的铝电解槽多参数相关性分析方法。在铝电解槽参数分布类型未知的情况下，首先利用非参数核密度函数估计建立变量的边缘密度函数；再构建基于混合Copula 模型的多变量联合分布函数，并通过权重参数调节不同类型Copula 函数的贡献比重；最后利用极大似然法对模型参数进行估计。（3）在多目标优化方面：提出一种基于拥挤距离排序的多目标细菌觅食算法。方法在保证铝电解槽平稳运行的基础上，建立电流效率最大和温室气体排放量最小的多目标优化模型；利用拥挤距离更新外部档案及对菌群步长进行自适应动态调整，以改进种群的收敛性和多样性，最后对优化模型求解。（4）在槽况诊断方面：出一种基于特征子空间虚假邻点判别的槽况诊断方法，首先考察各原始变量置零前后在核空间主元投影上的相似度，根据其对槽况的解释能力进行原始变量选择；再将约简后的原始变量输入概率神经网络，对各类异常槽况进行诊断。.本课题提出的动态演化建模与决策参数优化方法，可为解决超低能耗铝电解生产中多重不确定性条件下优化控制问题提供新的思路。