The assessment and diagnosis of controller performance for model‐based closed‐loop control systems has received considerable attention in recent years. A recognized factor dilapidating the controller performance is mismatching the true dynamical model of the plant and the mathematic model employed in the controller. In order to reduce the heavy effort to re‐identify the entire model, a large amount of recent works have been focusing on locating the mismatch. To further improve the mathematic model as well as controller performance, in this article, we provide a novel prediction‐based plant‐model mismatch quantification approach, which belongs to the class of moment match methodology. In particular, two cases of output prediction are considered: one‐step ahead prediction and multistep ahead prediction. Compared with existing efforts along the same line of mismatch quantification, when using the one‐step ahead prediction, our method shows an advantage of light computational complexity. On the other hand, when utilizing multistep predictions, it shows better convergence and robustness than the former, especially in the case of structural mismatches, and the long‐term prediction capability of the model is employed. With both predictions and consideration of the existence of unknown noise models in practice, we show that the plant‐model mismatch and unknown parameters of noise models can be quantified as the solution of a minimization issue that penalizes the discrepancy between the sample of plant outputs and the predictions calculated by considering the plant‐model mismatch and noise model parameters. Several examples are provided to demonstrate the effectiveness of the proposed methods, respectively.

中文翻译：

---

基于输出预测的闭环系统中数据驱动的工厂模型失配量化

近年来，基于模型的闭环控制系统的控制器性能评估和诊断受到了广泛关注。一个公认的破坏控制器性能的因素是对象的真实动态模型与控制器中采用的数学模型不匹配。为了减少重新识别整个模型的繁重工作，最近的大量工作一直专注于定位不匹配。为了进一步改进数学模型和控制器性能，在本文中，我们提供了一种新颖的基于预测的被控对象模型失配量化方法，该方法属于矩匹配方法类别。特别地，考虑输出预测的两种情况：一步提前预测和多步提前预测。与现有的失配量化方面的努力相比，当使用一步提前预测时，我们的方法显示出计算复杂度较低的优势。另一方面，当利用多步预测时，它比前者表现出更好的收敛性和鲁棒性，特别是在结构不匹配的情况下，并且利用了模型的长期预测能力。通过预测和考虑实践中未知噪声模型的存在，我们表明，工厂模型失配和噪声模型的未知参数可以量化为最小化问题的解决方案，该问题惩罚工厂输出样本与样本之间的差异。通过考虑对象模型失配和噪声模型参数来计算预测。分别提供了几个例子来证明所提出方法的有效性。