دانلود کتاب یادگیری ماشین در پایتون برای سیستم‌های فرآیند پویا

Model predictive control (MPC) and real-time optimization (RTO) are among the most critical technologies that drive the process industry. Any experienced process engineer would vouch that the success of MPCs and RTOs depend heavily on the accuracy of the underlying process models. Similarly, the key requirement for other important process technologies like dynamic data reconciliation, process monitoring, etc. is availability of accurate process models. Modeling efforts during commissioning of these tools can easily consume up to 90% of the project cost and time. Modern data revolution, whereby abundant amount of process data are easily available, and practical difficulties in building high-fidelity first-principles dynamic models for complex industrial processes have popularized the usage of empirical data-driven/machine learning (ML) models. Building dynamic models using process data is called system identification (SysID) and it’s a very mature field with extensive literature. However, it is also very easy for a process data scientist (PDS) new to this field to get overwhelmed with the SysID mathematics and ‘drowned’ in the sea of SysID terminology. Several noteworthy ML books have been written on time-series analysis; however, in process industry, input-output models are of greater import. Unfortunately, there aren’t many books that cater to the needs of modern PDSs interested in dynamic process modeling (DPM) without weighing them down with too much mathematical details and therein lies our motivation for authoring this book: specifically, a reader-friendly and easy to understand book that provides a comprehensive coverage of ML techniques that have proven useful for building dynamic process models with focus on practical implementations.   

It would be clear to you by now that this book is designed to teach working process engineers and budding PDSs about DPM. While doing so, this book attempts to avoid a pitfall that several generic ML books fall into: overemphasis on ‘modern’ and complex ML techniques such as artificial neural networks (ANNs) and undertreatment of classical DPM methods. Classical  techniques like FIR and ARX still dominate the dynamic modeling solutions offered by commercial vendors of industrial solutions and are no less ‘machine-learning’ than the ANNs. These two along with other classical techniques (such as OE, ARIMAX, CVA) predate the ANN-craze era and have stood the test of time in providing equal (if not superior) performance compared to ANNs. Correspondingly, along with modern ML techniques like RNNs, considerable portion of the book is devoted to classical dynamic models with the emphasis on understanding the implications of modeling decisions such as the impact of implicit noise model assumption with ARX model, the implication of differencing data, etc.