Co-gasification is a process that converts coal and biomass into useful products, such as syngas. Analytical and numerical approaches for modeling co-gasification process either require enormous amount of time or make a lot of assumptions which reduce consistency of the models in practical applications. Artificial Intelligence based modeling methods are used to simulate and to make predictions of outcomes of the co-gasification process. Even though previous studies result in successful modelling for specific cases, limited selection of methods and lack of implementation of cross-validation techniques causes insufficiency to explain unbiased performance evaluations and up-scale usability of the methods. In this paper, six different regression methods are employed to predict outputs of co-gasification process using a dataset containing 56 observations. Moreover, the original dataset is randomly resampled so that each model’s generalization ability is further assessed. The prediction performance of the proposed techniques on both datasets is evaluated and practical usability is discussed.
30 June 2020 in International Journal of Advances in Engineering and Pure Sciences
Prediction of higher heating value (HHV) using proximity and ultimate analysis is an important procedure for understanding the characteristic attribute of a fuel. Researches put effort to come up with equations to explain the relationship between the HHV value and those analyses. But conducted methods usually included only simple statistical analysis, thus they were partially effective to use in a practical manner. In this paper we approach this prediction problem from the machine learning perspective, we employ four machine learning methods, i.e. linear regression, polynomial regression, decision tree regression and support vector regression to predict HHV using proximity and ultimate analysis of different type of materials. Data set used is collected from literature and is categorized, where the resulting categories are used as features to be fed to the machine learning models to create prediction models as accurate as possible. Performances of the proposed methods are evaluated with k-fold cross-validation technique and each method’s pros and cons are discussed for both prediction accuracy and computational complexity.
31 March 2020 in International Journal of Advances in Engineering and Pure Sciences
Dry reforming of methane is a promising method to reduce the emission of CO2 and to use it in various type of Fischer–Tropsch synthesis and production of syngas. In order to obtain desirable products efficiently, the effect of reactants on the products must be known precisely. For this purpose, several studies have published for modeling the dry reforming of methane process with artificial intelligence-based data-driven prediction models. Due to lack of investigating overfitting problem and deficient and/or biased performance evaluations, actual potential of proposed methods have not been revealed for predicting certain outputs of the process. In this paper, we employed three regression methods and developed prediction models using a dataset with 57 observations. Performance evaluations of the models are performed with 10-fold cross-validation to ensure unbiased results. Proposed methods’ both training and testing performances are separately investigated, further applicability is discussed.
Epilepsy is a frequently seen neurological disorder manifested by repeating seizures. EEG signals of epilepsy patients are able to depict these seizures due to their high temporal resolution. However, it is generally challenging to differentiate these seizures by manual observation. Furthermore, Fourier based signal processing methods are unable to sufficiently analyze EEG signals as they are nonlinear and nonstationary by nature. Therefore, methods such as empirical mode decomposition (EMD) are exploited when working on epileptic EEG signals with the intention of detecting epileptic seizures. In this paper, we propose a framework in order to extract features from healthy, interictal and ictal EEG signals decomposed via EMD and Hilbert vibration decomposition (HVD), and then classify these signals with a convolutional neural network (CNN). Then, we evaluate the performance of both decomposition methods in detecting epileptic seizures. The obtained features are used for 10-fold cross validation with a CNN. The study was conducted on a benchmark dataset, where the EMD yielded 95.11% classification accuracy while HVD method achieved 100% accuracy. The overall performance of the HVD was found better compared to the EMD.