Search

Porosity is one of the most petrophysical parameters, which has profound impact on reservoir characterization, reserves estimation, and production forecasting. This parameter is determined from experimental implementation in laboratory core analysis, as well as interpretation of porosity log. Since there exists no reliable source of porosity data (core data and porosity log data) before of drilling a well, developing a reliable model for estimation of porosity from seismic attributes is highly valuable. In current study, a merged model is proposed for identifying formulation between porosity and seismic attributes in a field which is located in the Persian Gulf. In the first step, suitable seismic attributes which have prominent influence on porosity are extracted through forward stepwise selection variable method and considered as input parameters of the model. Secondly, input variables are transformed into higher correlated data space by virtue of nonparametric method so-called alternating conditional expectation (ACE). In third step, making quantitative correlation between ACE transformed of input parameters and porosity through improved intelligence model, including optimized neural network (ONN), optimized support vector regression (OSVR), and optimized fuzzy logic (OFL) are achieved. Optimization method which embedded in intelligence models formulation for ameliorating those performances is bat-inspired algorithm (BA). In the last step, outputs of improved models are combined through a committee machine (CM) in the sake of enhancing in the prediction accuracy. Optimal contribution of improved models in overall estimation is computed by mean of the BA method. Comparison between the individual models and the CM model shows that integrating models with the CM produce results with lowest error and highest correlation and consequently is superior. The results prove that proposed strategy in this study is reliable alternative way for mapping functional dependency between porosity and seismic attributes. [ABSTRACT FROM AUTHOR]

Free fluid porosity and rock permeability, undoubtedly the most critical parameters of hydrocarbon reservoir, could be obtained by processing of nuclear magnetic resonance (NMR) log. Despite conventional well logs (CWLs), NMR logging is very expensive and time-consuming. Therefore, idea of synthesizing NMR log from CWLs would be of a great appeal among reservoir engineers. For this purpose, three optimization strategies are followed. Firstly, artificial neural network (ANN) is optimized by virtue of hybrid genetic algorithm-pattern search (GA-PS) technique, then fuzzy logic (FL) is optimized by means of GA-PS, and eventually an alternative condition expectation (ACE) model is constructed using the concept of committee machine to combine outputs of optimized and non-optimized FL and ANN models. Results indicated that optimization of traditional ANN and FL model using GA-PS technique significantly enhances their performances. Furthermore, the ACE committee of aforementioned models produces more accurate and reliable results compared with a singular model performing alone. [ABSTRACT FROM AUTHOR]

Precipitation and deposition of asphaltene during different stages of petroleum production is recognized as problematic in oil industry because of the increase in production cost and the inhibition of a consistent flow of crude oil in different medium. Numerous correlations have been developed to determine asphaltene stability in crude oil. In this study, a novel ONN method was used to estimate difference index from SARA fraction data for rapid, accurate, and cost-effective determination of asphaltene stability. Neural networks are highly in danger of trapping in local minima. To eliminate this flaw, a hybrid genetic algorithm-pattern search technique was used instead of common back-propagation algorithm for training the employed neural network. A comparison between neural network and optimized neural network indicated superiority of optimized neural network. [ABSTRACT FROM PUBLISHER]

Oil-CO2minimum miscible pressure (MMP) has significance in selecting appropriate reservoir for miscible gas injection and greatly governs performance of local displacement. Accurate determination of MMP is very expensive, time-consuming, and labor intensive. Therefore, the quest for a method to determine MMP accurately and save time and money is necessary. This study held a competition between neural network and support vector regression models and assessed their performance in prediction of MMP for both pure and impure miscible CO2injection. Subsequently, a committee machine was constructed based on divide and conquer principle to reap benefits of both model and increases the precision of final prediction. Results indicated committee machine performed more satisfyingly compared with individual intelligent models performing alone. [ABSTRACT FROM PUBLISHER]

Highlights: [•] In this study, precipitated asphaltene amount is predicted from titration data. [•] Neural network (NN), fuzzy logic (FL) were used as modeling tools. [•] NN and FL were then optimized by GA-PS to enhance prediction accuracy. [•] A CM was then constructed to integrate aforementioned models and enhances precision. [•] Implementation of the proposed strategy considerably saves time and reduces costs. [Copyright &y& Elsevier]

Environmental pollution has mainly been attributed to urbanization and industrial developments across the globe. Air pollution has been marked as one of the major problems of metropolitan areas around the world, especially in Tehran, the capital of Iran, where its administrators and residents have long been struggling with air pollution damage such as the health issues of its citizens. As far as the study area of this research is concerned, a considerable proportion of Tehran air pollution is attributed to PM10 and PM2.5 pollutants. Therefore, the present study was conducted to determine the prediction models to determine air pollutions based on PM10 and PM2.5 pollution concentrations in Tehran. To predict the air-pollution, the data related to day of week, month of year, topography, meteorology, and pollutant rate of two nearest neighbors as the input parameters and machine learning methods were used. These methods include a regression support vector machine, geographically weighted regression, artificial neural network and auto-regressive nonlinear neural network with an external input as the machine learning method for the air pollution prediction. A prediction model was then proposed to improve the afore-mentioned methods, by which the error percentage has been reduced and improved by 57%, 47%, 47% and 94%, respectively. The most reliable algorithm for the prediction of air pollution was autoregressive nonlinear neural network with external input using the proposed prediction model, where its one-day prediction error reached 1.79 µg/m3. Finally, using genetic algorithm, data for day of week, month of year, topography, wind direction, maximum temperature and pollutant rate of the two nearest neighbors were identified as the most effective parameters in the prediction of air pollution. [ABSTRACT FROM AUTHOR]