Background

The mechanism of glucose regulation in human blood is a nonlinear complicated biological system with uncertain parameters and external disturbances which cannot be imitated accurately by a simple mathematical model. So to achieve an artificial pancreas, a method that does not need a model is necessary.

Methods

In this paper, a model free third order terminal sliding mode controller is developed and applied to blood glucose regulation system. So in this paper, a data driven control method is proposed which doesn't need a pre specified mathematical model of the system. The proposed method uses a third order terminal sliding mode controller to overcome the problem in finite time without chattering. It also uses a disturbance estimation technique to reject external disturbances. The sliding mode algorithm is equipped with a regression algorithm to release its need to model of the system. It is proved theoretically that the method is stable and the error converges to zero. In order to determine the parameters needed in this method, an algorithm is provided.

Results

Simulation studies are carried out with different scenarios and compared with Model Free Adaptive Control method. At the first scenario, the proposed method is applied to a virtual type- 1 diabetic patient without considering of external disturbances. The blood glucose level of 110 mg/dl is considered as the goal and it is illustrated that the desired glucose concentration is obtained. It is illustrated that the proposed method shows better performance against Model Free Adaptive Controller. Then in the next scenario, blood glucose of the patient is controlled in presence of three meal times during a day with different values of carbohydrate. The maximum of the blood glucose in this scenario is obtained as 168.5 mg/dl and the minimum of it stays on 85.5 Mg/dl. So the patient blood glucose level is almost within acceptable range (70–180 mg/dl) unlike the Model Free Adaptive Controller. In the last scenario, 22 tests are done for different patients (by randomly varying simulator parameters in ± 40% range) and the control performance is evaluated by the well-known Control Variability Grid Analysis CVGA. For all of them, the blood glucose remains in the green zone (safe region) of the CVGA .

Conclusion

Simulation results show that the proposed method acts robustly and can overcome uncertainties and external disturbances. The blood glucose level remains in safe region in all case. So the proposed method can be used in an artificial pancreas.

中文翻译：

---

用于血糖控制的无模型滑模控制器：无需系统的数学模型即可转向人造胰腺。

背景

人体血液中葡萄糖的调节机制是一个非线性复杂的生物系统，具有不确定的参数和外部干扰，无法通过简单的数学模型精确地模拟。因此，要获得人造胰腺，就需要一种不需要模型的方法。

方法

本文开发了一种无模型的三阶终端滑模控制器，并将其应用于血糖调节系统。因此，本文提出了一种数据驱动的控制方法，该方法不需要预先定义的系统数学模型。所提出的方法使用三阶终端滑模控制器来在有限时间内克服抖动的问题。它还使用干扰估计技术来拒绝外部干扰。滑模算法配有回归算法，以释放其对系统模型的需求。理论上证明该方法是稳定的，误差收敛到零。为了确定该方法所需的参数，提供了一种算法。

结果

在不同的情况下进行了仿真研究，并与无模型自适应控制方法进行了比较。在第一种情况下，将所提出的方法应用于虚拟1型糖尿病患者，而无需考虑外部干扰。110 mg / dl的血糖水平被认为是目标，可以说明获得了所需的葡萄糖浓度。结果表明，所提出的方法与无模型自适应控制器相比具有更好的性能。然后在下一个场景中，在一天中有3次进餐时间，不同碳水化合物含量的情况下控制患者的血糖。在这种情况下，最高血糖为168.5 mg / dl，最低血糖保持在85.5 Mg / dl。因此，与无模型自适应控制器不同，患者的血糖水平几乎在可接受的范围内（70-180 mg / dl）。在最后一种情况下，针对不同患者进行了22次测试（通过在±40％范围内随机改变模拟器参数），并通过众所周知的“控制变量网格分析” CVGA评估控制性能。对于所有这些人，血糖都保留在CVGA的绿色区域（安全区域）中。

结论

仿真结果表明，该方法鲁棒性强，可以克服不确定性和外部干扰。在所有情况下，血糖水平都保持在安全范围内。因此，该方法可用于人工胰腺。