Gaussian Processes Based Modeling And Control Of Neuromuscular Bolcksde

The main purpose of Neuromuscular blockade (NMB) is to prevent unconscious motion of patients from medical instrument injury. In clinical NMB level control cases, there exist the following problems:traditional NMB drugs administration is based on the experience of anesthetists. There is no uniform quantitative criterion for the administration of NMB drugs, which give rise to the sabotage of system stability; Long time working in a busy and stressful environment will makes anesthetists tired and prone to make mistakes, which brings great risks to the patients. Although with the appearance of sensors and NMB level monitors, at the same time boost the research on automatic drug delivery control, duo to the diversity of individuals and complex nonlinearity of metabolism and all kind of physiological process, difficulties have been added to the design of NMB controller. For these reasons, this study is focusing on the problem of automatic administration of neuromuscular block drugs, using Gauss processes (GPs) model based predictive control technology to realize the automatic drug administration under individual differences.At the beginning, attentions have been paid to the problem of dealing with individual diversity by means of data driven dynamic modeling of NMB, particularly Gaussian processes modeling technique is used. In the process of model update, the inverse computation of covariance matrix is included, which is a big burden for optimization. For this problem, a method of solving the inverse formula is used to solve the problem of calculating the inverse operation, which lessens the computation burdens and improves the speed of optimization. Furthermore, covariance function is the heart of Gaussian processes, which determines the structure and hyperparameters of the model. However there is no uniform principle for the selection of covariance function, then an experiment has been carried out for the selection of a proper covariance function for neuromuscular blockade system. Fixed model is not capable of catching the changes during the procedure of bio-systems. Considering this, an online dynamic modeling method is applied to the modeling of NMB process. Works on the selection of moving window training data size have been done in this research, by the time parameters are updated over time, which assures the local performance of the model and the design of model-based controller. Simulations have been done in this research, it turns out that the proposed method has a high accuracy of modeling and prediction for local dynamics.Gaussian process is a statistical probability modeling method, which mapping the relationship between the input and output without any knowledge about the system structure. Most of the model based controllers design requires the structure information of the system, but there still some exceptions like model predictive control (MPC). MPC is popular for its ability of copping with limitations and without limitations for structure of the predictive model. Considering all the obstruction we have met, GPs based MPC has been used to the administration of NMB drugs. Besides, aiming at the problem of large lag in the process of neuromuscular block, a differential controller has been added to the structure of model predictive control. A taylor expansion approximation has been used to infer the mathematical expression of optimization process and achieved a complete control strategy. Simulation results show that the proposed control scheme shows excellent control performance, at the same time, the mean square tracking error of on-line control algorithm is improved by at least one time than the off-line way, which further shows the advantage of the method.