Recently, there has been a surge of interest in approximating the model predictive control (MPC) law using expert supervised learning techniques, such as deep neural networks (DNN). Approximating the MPC control policy requires labeled training data sets, which is typically obtained by sampling the state-space and evaluating the control law by solving the numerical optimization problem offline for each sample. The accuracy of the MPC policy approximation is dependent on the availability of large training data set sampled across the entire state space. Although the resulting approximate MPC law can be cheaply evaluated online, generating large training samples to learn the MPC control law can be time consuming and prohibitively expensive. This paper aims to address this issue, and proposes the use of NLP sensitivities in order to cheaply generate additional training samples in the neighborhood of the existing samples.

中文翻译：

---

用于学习近似模型预测控制器的基于灵敏度的数据增强

最近，人们对使用专家监督学习技术（例如深度神经网络 (DNN)）逼近模型预测控制 (MPC) 法则的兴趣激增。逼近 MPC 控制策略需要标记的训练数据集，这通常是通过对状态空间进行采样并通过对每个样本离线求解数值优化问题来评估控制律来获得的。MPC 策略近似的准确性取决于在整个状态空间中采样的大型训练数据集的可用性。虽然由此产生的近似 MPC 法则可以在线进行廉价评估，但生成大量训练样本来学习 MPC 控制法则可能非常耗时且成本高得令人望而却步。本文旨在解决这个问题，