Comparing artificial neural network algorithms for prediction of higher heating value for different types of biomass

The paper presents the comparative qualitative and quantitative analysis of twelve algorithms for training artificial neural networks (ANN) which predict the higher heating value (HHV) of biomass based on the proximate analysis (fixed carbon, volatile matter, and ash percentage). The twelve networks, with the same structure but different training algorithm, were fed with 318 experimental data triplets from literature for different biomass species and trained with 318 corresponding HHV values used for the supervised learning. Our comparative analysis showed that several algorithms resulted in ANNs generating outputs well correlated with the true measured values of the biomass HHV. Of those, Levenberg-Marquardt algorithm gives the best results in terms of mean squared error calculated on the training set data, while Bayesian regularization gives the best results in terms of regression. When applied to new datasets, unknown to the ANNs trained here, the highest accuracy of the HHV prediction was obtained by Conjugate Gradient, Powell/Beale Restarts training function. It ensured prediction based on the unknown datasets better than Levenberg-Marquardt algorithm. The described approach can be used for predicting the calorific values of different biomass species, including newly proposed ones, as well as for optimizing the HHV for both pure biomass and biomass mixtures.

Evaluasi Prediksi Higher Heating Value (HHV) Biomassa Berdasarkan Analisis Proksimat

Abstrak – Biomassa merupakan salah satu energi terbarukan yang sangat mudah ditemui, ramah lingkungan dan cukup ekonomis. Keberadaan biomassa dapat dimaanfaatkan sebagai pengganti bahan bakar fosil, baik itu minyak bumi, gas alam maupun batu bara. Analisi diperlukan sebagai dasar biomassa sebagai energi seperti proksimat dan kalor. Analisis terpenting untuk menilai biomassa sebagai bahan bakar adalah nilai kalori atau higher heating value (HHV). HHV secara eksperimen diukur menggunakan bomb calorimeter, namun pengukuran ini kurang efektif, karena memerlukan waktu serta biaya yang tinggi. Penelitian mengenai prediksi HHV berdasarkan analisis proksimat telah dilakukan sehingga dapat mempermudah dan menghemat biaya yang diperlukan peneliti. Dalam makalah ini dibahas evaluasi persamaan untuk memprediksi HHV berdasarkan analisis proksimat pada biomassa berdasarkan data dari penelitian sebelumnya. Prediksi nilai HHV menggunakan lima persamaan yang dievaluasi dengan 25 data proksimat biomassa dari penelitian sebelumnya, kemudian dibandingkan berdasarkan nilai error untuk mendapatkan prediksi terbaik. Hasil analisis menunjukan, persamaan A terbaik di 7 biomassa, B di 6 biomassa, C di 6 biomassa, D di 5 biomassa dan E di 1 biomassa.Kata kunci: bahan bakar, biomassa, higher heating value, nilai error, proksimat Abstract – Biomass is a renewable energy that is very easy to find, environmentally friendly, and quite economical. The existence of biomass can be used as a substitute for fossil fuels, both oil, natural gas, and coal. Analyzes are needed as a basis for biomass as energy such as proximate and heat. The most critical analysis to assess biomass as fuel is the calorific value or higher heating value (HHV). HHV is experimentally measured using a bomb calorimeter, but this measurement is less effective because it requires time and high costs. Research on the prediction of HHV based on proximate analysis has been carried out so that it can simplify and save costs needed by researchers. In this paper, the evaluation of equations is discussed to predict HHV based on proximate analysis on biomass-based on data from previous studies. HHV prediction values using five equations were evaluated with 25 proximate biomass data from previous studies, then compared based on error value to get the best predictions. The analysis shows that Equation A predicts best in 7 biomass, B in 6 biomass, C in 6 biomass, D in 5 biomass, and E in 1 biomass. Key words: fuel, biomass, higher heating value, error value, proximate

Biomass higher heating value prediction analysis by ANFIS, PSO-ANFIS and GA-ANFIS models

<p>In this study, a new model for biomass higher heating value (HHV) prediction based on the Adaptive Neuro-Fuzzy Inference System (ANFIS) approach was proposed. Proximate analysis (volatile matter, fixed carbon and ash content) data for a wide range of various biomass types from the literature were used as input in model studies. Optimization of ANFIS parameters and formation of the model structure were performed by genetic algorithm (GA) and particle swarm optimization (PSO) algorithm in order to achieve optimum prediction capability. The best-fitting model was selected using statistical analysis tools. According to the analysis, PSO-ANFIS model showed a superior prediction capability over ANFIS and GA optimized ANFIS model. The Root Mean Square Error (RMSE), Mean Absolute Error (MAE), Mean Bias Error (MBE) and coefficient of determination (R2) for PSO-ANFIS were determined as 0.3138, 0.2545, -0.00129 and 0.9791 in the training phase and 0.3287, 0.2748, 0.00120 and 0.9759 in the testing phase, respectively. As a result, it can be concluded that the proposed PSO-ANFIS model is an efficient technique and has potential to calculate biomass HHV prediction with high accuracy.</p>