Development, validation, and testing of algorithms for artificial pancreas (AP) systems require mathematical models for the glucose–insulin dynamics inside the body. These physiological models have been extensively studied over the past decades. Two broad types of models are available in diabetic research, each with its own unique purpose: (i) minimal models, which are relatively simple but still manages to capture the macroscopic behavior of the glucose–insulin dynamics of the body, and (ii) high-fidelity models, which are complex and precisely describe the internal dynamics of the glucose–insulin interaction in the body. The minimal models are primarily utilized for control algorithm synthesis, whereas the high-fidelity models are used as platforms for testing and validating AP systems. The most well-known variants of these physiological models are discussed in detail. In addition to these systems, data-driven models such as the auto-regressive moving average with exogenous inputs (ARMAX) models are also used widely in control algorithm synthesis for AP systems. High-fidelity models are utilized for simulating virtual diabetic patients for in silico testing and validation of artificial pancreas systems. Two currently available high-fidelity models are reviewed in this paper for completeness, including the Type-1 diabetes mellitus (T1DM) simulator approved by the food and drug administration of USA. Models accounting for exercise and also glucagon infusion (for dual-hormone AP systems) are also included, which are essential in developing control algorithms with better autonomy and minimal risk.

人工胰腺 (AP) 系统算法的开发、验证和测试需要体内葡萄糖-胰岛素动力学的数学模型。在过去的几十年里，这些生理模型得到了广泛的研究。糖尿病研究中有两种广泛类型的模型，每种都有其独特的目的：(i)最小模型，相对简单但仍设法捕捉身体葡萄糖-胰岛素动力学的宏观行为，以及 (ii)高保真模型，它们很复杂并且精确地描述了体内葡萄糖 - 胰岛素相互作用的内部动力学。最小模型主要用于控制算法综合，而高保真模型用作测试和验证 AP 系统的平台。详细讨论了这些生理模型最著名的变体。除了这些系统之外，数据驱动模型，例如具有外生输入的自回归移动平均 (ARMAX) 模型，也广泛用于 AP 系统的控制算法综合。高保真模型用于模拟虚拟糖尿病患者，用于人工胰腺系统的计算机测试和验证。目前可用的两种高保真模型在本文中审查了完整性，包括美国食品和药物管理局批准的 1 型糖尿病 (T1DM) 模拟器。还包括考虑运动和胰高血糖素输注（用于双激素 AP 系统）的模型，这对于开发具有更好自主性和最小风险的控制算法至关重要。