Prediction of vitrinite reflectance values using machine learning techniques: a new approach
- PDF / 3,021,670 Bytes
- 21 Pages / 595.276 x 790.866 pts Page_size
- 68 Downloads / 170 Views
ORIGINAL PAPER-EXPLORATION GEOLOGY
Prediction of vitrinite reflectance values using machine learning techniques: a new approach Zahra Sadeghtabaghi1 · Mohsen Talebkeikhah1 · Ahmad Reza Rabbani1 Received: 9 June 2020 / Accepted: 6 November 2020 © The Author(s) 2020
Abstract Vitrinite reflectance (VR) is considered the most used maturity indicator of source rocks. Although vitrinite reflectance is an acceptable parameter for maturity and is widely used, it is sometimes difficult to measure. Furthermore, Rock-Eval pyrolysis is a current technique for geochemical investigations and evaluating source rock by their quality and quantity of organic matter, which provide low cost, quick, and valid information. Predicting vitrinite reflectance by using a quick and straightforward method like Rock-Eval pyrolysis results in determining accurate and reliable values of VR with consuming low cost and time. Previous studies used empirical equations for vitrinite reflectance prediction by the Tmax data, which was accompanied by poor results. Therefore, finding a way for precise vitrinite reflectance prediction by Rock-Eval data seems useful. For this aim, vitrinite reflectance values are predicted by 15 distinct machine learning models of the decision tree, random forest, support vector machine, group method of data handling, radial basis function, multilayer perceptron, adaptive neuro-fuzzy inference system, and multilayer perceptron and adaptive neuro-fuzzy inference system, which are coupled with evolutionary optimization methods such as grasshopper optimization algorithm, bat algorithm, particle swarm optimization, and genetic algorithm, with four inputs of Rock-Eval pyrolysis parameters of Tmax, S1/TOC, HI, and depth for the first time. Statistical evaluations indicate that the decision tree is the most precise model for VR prediction, which can estimate vitrinite reflectance precisely. The comparison between the decision tree and previous proposed empirical equations indicates that the machine learning method performs much more accurately. Keywords Vitrinite reflectance · Rock-eval pyrolysis · Maturity · Depth · Decision tree
Introduction Rock-Eval pyrolysis is considered one of the most powerful and influential geochemical techniques which can provide valuable information about organic matter sediments promptly (Espitalié et al. 1977; Behar et al. 2001). This technique is also used extensively in petroleum source rock evaluation through the determination of organic matter types, generation potential, and level of maturity (Dembicki 2016). S1 (formerly generated hydrocarbon), S2 (remain potential of hydrocarbon generation), TOC (total organic carbon), Tmax (temperature of S2 maximum), HI (hydrogen index), and OI (oxygen index) are some parameters which are obtained by Rock-Eval pyrolysis. Despite all Rock-Eval advantages, this * Zahra Sadeghtabaghi [email protected] 1
Department of Petroleum Engineering, Amir Kabir University of Technology, Tehran, Iran
method is associated with some obstacles for maturit
Data Loading...
