Soft sensor aplicado a plantas de processamento de gás natural baseado em redes neurais artificiais

Abstract

In face of an increasingly competitive market, producing efficiently and effectively is essential for a positive economic balance. Reducing costs, optimizing processes and offering even better products are factors that directly influence the economy of any industry. In this view, techniques that can improve and / or guarantee optimization of processes, such as the monitoring of product quality or advanced and intelligent control become fundamental for the industry as a whole. In case of Natural Gas Processing Units (NGPUs), monitoring the quality of the product is intrinsic to a satisfactory production, and quality control has been done such as in most chemical processes, through the chemical composition of the products. However, even when chromatographs are used for chemical analysis of the components, the analytical process is slow and long measurement intervals are observed. This hampers real-time product monitoring or control techniques from being established to obtain better process performance.Among these products, the most important, economically speaking, is LPG (Liquefied Petroleum Gas) composed of propane, butane and contaminats such as ethane and pentane. In this work, a system called soft sensor that makes the inference of the main components of GLP based on artificial neural network is proposed. Then, the real-time monitoring of the quality of the produced LPG becomes possible, since the measurement of the composition of the LPG will not be obtained through the slow analytical process. Thus, the quality of the process, the LPG itself, consequently, its profitability are improved.In the development of this work, a simulated GNPU has been used in HYSYS, consisting of a deethanizing column in series with a debutanizer column. In the instrumentation of the plant, there are some PID controllers. The virtual sensor is based on process variables of these controllers. In this work, a real-time error correction module of the softsensor is also proposed, based on the measuruments of the LPG composition made by the chromatographs present in the process. The results are promising, attesting the adequate behavior of the soft sensor.