Efficiency Of Artificial Neural Networks Research Articles

Artificial neural network in optimization of bioactive compound extraction: recent trends and performance comparison with response surface methodology.

Plant products and its by-products are rich source of bioactive compounds like antioxidants, flavonoids, phenolics, pigments and phytochemicals. Bioactive compound's health-promoting properties are well studied. However, optimal extraction of bioactive compounds is a complex, labour- and time-intensive process. It is also highly sensitive to experimental variables. Predicting output variables can reduce the experimental work and has positive environmental impact. Various tools such as Response Surface Methodology (RSM), Mathematical modelling have been commonly used for optimization and predictive modelling of the extraction process. Although mathematical modelling and RSM are efficient, recent studies have used Artificial Neural Network (ANN) which is more efficient and accurate and can perform extensive predictions with high accuracy. The manuscript focuses on current trends of ANN application in optimizing the extraction of bioactive compounds. In this study, ANN and RSM have been compared in terms of their performances in optimizing and modelling the extraction of bioactive compounds from herbs, medicinal plants, fruit, vegetables, and their by-products. The findings from the literature indicate that efficiency of ANN was superior to RSM. Future researches can focus on use of ANN in industrial optimization experiments.

RETRACTED: Relationships among fall cone flow index and consistency identifiers of fine-grained soils with emphasis on toughness limit

It is well known that plasticity characteristics are one of the most important factors controlling the engineering behavior of fine-grained soils. In this regard, Atterberg limits are used for evaluation of plasticity and strength behavior of soils. Of all the plasticity identifiers, fall cone flow index, which is the slope of the flow curve is a descriptor of plasticity and undrained shear strength characteristics of cohesive soils. On the other hand, toughness limit is also efficient in prediction of many index properties of soils including plasticity characteristics, compression index as well as residual friction angle. This study is an attempt to establish relationships among fall cone flow index and consistency identifiers of fine-grained soils. Emphasis is given on toughness limit, which is believed to be a practical and efficient approach in assessment of fall cone flow index. Regression-based equations establishing relationships between fall cone flow index, plasticity index, plasticity ratio, toughness limit and liquid limit were presented. Besides, the efficiency of artificial neural networks in prediction of toughness limit and fall cone flow index was investigated. It should be noted that established relationships using regression equations are based on regional data, which come up with coefficient of determination (R2) values ranging between 0.706 and 0.978. On the other hand, ANN models were used to model global data, R2 values were obtained to be between 0.586-0.972 and 0.522-0.889 for training and testing phases, respectively. Relationships concerning local and global data were presented, along with analysis of effectiveness of application of relationships established for analysis of global data.

Evaluating the efficiency of artificial neural networks and tree-based techniques for forecasting the flexural strength of concrete using waste foundry sand

Evaluating the efficiency of artificial neural networks and tree-based techniques for forecasting the flexural strength of concrete using waste foundry sand

Use of artificial neural networks with the physiological principles to predict growth model

Process-based models have become essential tools for forest planning and management due to their ability to assess the impacts of climate and operational changes on productivity. Physiological Principles Predicting Growth (3-PG) is a process-based/hybrid model widely used, associated with allometric models to estimate dendrometric variables. However, the accuracy of the model outputs still needs to be improved, especially those of interest to forest management. In this study, we tested the efficiency of artificial neural networks (ANNs) to estimate the average diameter at 1.30 m height (dbh), total height, stand volume, and root, stem, and foliage biomass. We also evaluated the prediction of two parameters that estimate the proportion of biomass allocated aboveground (ap and np). A database from published 3-PG parameterizations was generated and used to train the ANNs. Two networks were validated against observed data from two municipalities in the MG state of Brazil. Potential productivity maps at six and seven years were generated for Brazil, to evaluate the performance of the 3-PG + ANN model. All ANNs presented satisfactory statistical results in predicting the ap and np parameters, foliage biomass, dbh, total height and stand volume. The use of artificial neural networks integrated with the 3-PG model or another process-based model is a promising alternative to improve the accuracy of estimates, especially those of interest for forest management. Artificial neural networks offer a simple, flexible way to insert important variables into a process-based model. The parameters and 3-PG outputs can be estimated efficiently using artificial neural networks.

Computational intelligence modelling of methylene blue adsorption by metal-organic frameworks

ABSTRACT Computational intelligence models for wastewater treatment utilise mathematical algorithms to simulate and optimise the processes involved in the removal of pollutants from wastewater, aiding in the design and operation of the wastewater treatment plant. The aim of this work was to compare the efficiency of computational models in modelling the dynamic adsorption of dyes. Metal-organic frameworks (MOFs) are considered as adsorbents because of their excellent properties. Further, the dataset used for this study was sourced from literature pertaining to the adsorption of methylene blue (MB), which is one of the commonly found dyes in wastewater. Adsorption parameters considered for the study are initial concentration, bed height, flow rate, pH and time, for the continuous adsorption process. The efficacy of the selected computational models was compared by employing various statistical metrics. Moreover, the results show the efficiency of artificial neural network and radial basis function neural network models is superior to other models based on R2 and mean square error, which were in the range of 0.95–0.999 and 10–3–10–5, respectively. In summary, the computational intelligence models serve as the best tools for the prediction of the adsorption performance of MOFs for MB.

Spider chart, greenness and whiteness assessment of experimentally designed multivariate models for simultaneous determination of three drugs used as a combinatory antibiotic regimen in critical care units: Comparative study

In this study, five earth-friendly spectrophotometric methods using multivariate techniques were developed to analyze levofloxacin, linezolid, and meropenem, which are utilized in critical care units as combination therapies. These techniques were used to determine the mentioned medications in laboratory-prepared mixtures, pharmaceutical products and spiked human plasma that had not been separated before handling. These methods were named classical least squares (CLS), principal component regression (PCR), partial least squares (PLS), genetic algorithm partial least squares (GA-PLS), and artificial neural network (ANN). The methods used a five-level, three-factor experimental design to make different concentrations of the antibiotics mentioned (based on how much of them are found in the plasma of critical care patients and their linearity ranges). The approaches used for levofloxacin, linezolid, and meropenem were in the ranges of 3–15, 8–20, and 5–25 µg/mL, respectively. Several analytical tools were used to test the proposed methods' performance. These included the root mean square error of prediction, the root mean square error of cross-validation, percentage recoveries, standard deviations, and correlation coefficients. The outcome was highly satisfactory. The study found that the root mean square errors of prediction for levofloxacin were 0.090, 0.079, 0.065, 0.027, and 0.001 for the CLS, PCR, PLS, GA-PLS, and ANN models, respectively. The corresponding values for linezolid were 0.127, 0.122, 0.108, 0.05, and 0.114, respectively. For meropenem, the values were 0.230, 0.222, 0.179, 0.097, and 0.099 for the same models, respectively. These results indicate that the developed models were highly accurate and precise. This study compared the efficiency of artificial neural networks and classical chemometric models in enhancing spectral data selectivity for quickly identifying three antimicrobials. The results from these five models were subjected to statistical analysis and compared with each other and with the previously published ones. Finally, the whiteness of the methods was assessed by the recently published white analytical chemistry (WAC) RGB 12, and the greenness of the proposed methods was assessed using AGREE, GAPI, NEMI, Raynie and Driver, and eco-scale, which showed that the suggested approaches had the least negative environmental impact. Furthermore, to demonstrate solvent sustainability, a greenness index using a spider chart methodology was employed.

Application of Artificial Neural Networks to Predict Genotypic Values of Soybean Derived from Wide and Restricted Crosses for Relative Maturity Groups

The primary objective of soybean-breeding programs is to develop cultivars that offer both high grain yield and a maturity cycle tailored to the specific soil and climatic conditions of their cultivation. Therefore, predicting the genetic value is essential for selecting and advancing promising genotypes. Among the various analytical approaches available, deep machine learning emerges as a promising choice due to its capability to predict the genetic component of phenotypes assessed under field conditions, thereby enhancing the precision of breeding decisions. This study aimed to determine the efficiency of artificial neural networks (ANNs) in predicting the genetic values of soybean genotypes belonging to populations derived from crosses between parents of different relative maturity groups (RMGs). We characterized populations with broad and restricted genetic bases for RMG traits. Data from three soybean populations, evaluated over three different agricultural years, were used. Genetic values were predicted using the multilayer perceptron (MLP) artificial neural network and compared to those obtained using the best unbiased linear prediction from variance components using restricted maximum likelihood (RR-BLUP). The MLP neural network efficiently predicted genetic values for the relative maturity group trait for genotypes belonging to populations of broad and restricted crosses, with an R2 of 0.999 and root-mean-square error (RMSE) of 0.241, and for grain yield, there was an R2 of 0.999 and an RMSE of 0.076. While the percentage of coincident superior genotypes remained relatively consistent, a significant difference was observed in their ranking order. The genetic gain with selection estimated using MLP was higher by 30–110% compared to RR-BLUP for the relative maturity group trait and 90–500% for grain yield. Artificial neural networks (ANNs) showed higher efficiency than RR-BLUP in predicting the genetic values of the soybean population. Local selection at intermediate latitudes is conducive to developing lines adaptable for regions at higher and lower latitudes.

Estimated volume of eucalyptus plantations through ALOS satellite images

Estimates of wood volume, either in even or uneven-aged forests, is necessary for planning forest management, studies concerning carbon sequestration and natural resources conservation, as well as energy planning in regions where wood is the primary fuel for power generation. Such purposes mainly involve analysis of large areas, which leads to challenges that can be solved through remote sensing. Using multispectral and microwave data, respectively, from the AVNIR-2 and PALSAR sensors, both on board in the ALOS satellite, the wood volume of a commercial eucalyptus plantation located in the state of Minas Gerais, Brazil was estimated using artificial neural. Neural networks emerged as a powerful machine learning technique, capable of modeling intrinsically nonlinear relationships that are present in the data. Thus, the obtained estimates for wood volume in the eucalyptus plantations showed a correlation coefficient with the corresponding observed volumes (forest inventory) of 0.99, a square root of the quadratic error (RQEM) of 0.3% and errors with amplitude between -1 to 1%. The efficiency of artificial neural networks for estimating wood volume in homogeneous plantations has been proven. The multilayer perceptron neural network was efficient in modeling the complex relationships that exist between multispectral, microwave and wood volume data, producing highly accurate estimates. Only a small amount of plots was sufficient to train the artificial neural networks in order to estimate volume. In addition, the fast convergence obtained by the BFGS algorithm allowed a comprehensive data analysis.

Application of artificial neural network and dynamic adsorption models to predict humic substances extraction from municipal solid waste leachate

Sustainable municipal solid waste leachate (MSWL) management requires a paradigm shift from removing contaminants to effectively recovering resources and decreasing contaminants simultaneously. In this study, two types of humic substances, fulvic acid (FA) and humic acid (HA) were extracted from MSWL. HA was extracted using HCl and NaOH solution, followed by FA using a column bed under diversified operations such as flow rate, input concentration, and bed height. Also, this work aims to evaluate efficiency of Artificial Neural Network (ANN) and Dynamic adsorption models in predicting FA. With the flow rate of 0.3 mL/min, bed height of 15.5 cm, and input concentration of 4.27 g/mL, the maximum capacity of FA was obtained at 23.03 mg/g. FTIR analysis in HA and FA revealed several oxygen-containing functional groups including carboxylic, phenolic, aliphatic, and ketone. The high correlation coefficient value (R2) and a lower mean squared error value (MSE) were obtained using the ANN, indicating the superior ability of ANN to predict adsorption capacity compared to traditional modeling.

The potential of region-specific machine-learning-based ground motion models: Application to Turkey

Conventional ground motion models have extensively been established worldwide based on classical regression analysis of records. Alternatively, advanced nonparametric machine-learning (ML) algorithms may capture the complex nonlinear behaviour of earthquake motions. This paper investigates the efficiency of artificial neural network (ANN) and extreme gradient boosting (XGBoost) in predicting peak ground acceleration (PGA), peak ground velocity (PGV) and pseudo-spectral acceleration (PSA) (period, T = 0.03–2.0 s) for the Turkish dataset. The dataset involves 1166 records of 383 events with a moment magnitude (Mw) of 4.0–7.6, Joyner and Boore distance (RJB) of 0–200 km, focal depth (FD) less than 35 km, and site condition as the averaged shear wave velocity of the soil on the top 30 m (VS30) of 131–1380 m/s. The performance of the models is compared against empirical models in terms of root-mean-square error (RMSE), coefficient of determination (R2), Pearson correlation coefficient (r), and inter-event and intra-event residuals. To perform residual analysis, a likelihood function is developed. Findings reveal that the XGBoost approach gives an unbiased model with a higher correlation and lower residual than ANN. Finally, an online platform is provided for any interested users.

Improving 3-PG calibration and parameterization using artificial neural networks

Understanding how the physiological processes of trees are affected by the environment or silvicultural practices is important for forest management, which requires process-based models. It enables the evaluation of the growth of a forest under different scenarios. The 3-PG model has been widely used all over the world, justified by its simplicity and efficiency, as it uses a more accessible language and fewer parameters than other process-based models. It is a model of greatest interest for forest management because it enables the use of allometric equations to calculate variables of interest in this area, such as the average diameter at 1.30 m height (DBH), total height and stand volume. The 3-PG parameterization is essential to guarantee the model's good performance; however, in some cases, when observed data are not available, values from the literature is used or calibration is performed. In general, there is a mixture of these alternatives in the same parameterization, but some of the parameters generate greater sensitivity in some outputs or change according to site characteristics. In the present work, we analyzed the efficiency of artificial neural networks to predict some of the parameters pointed out in the literature as being of the greatest importance for 3-PG using climate and process variables as inputs. For this, a simulated database was generated, using 16 parameterizations of 3-PG, for different regions of Brazil. The parameters values of the DBH function (as and ns), minimum and maximum fraction of biomass allocated to the root (ηRn and ηRx), and age at full canopy cover (tc) were associated with this database. The Artificial Neural Networks (ANNs) were trained using the database with parameter repetition over time and with the average condition of each site. In the second case, training was performed using 100% of the data, and validation was performed using a simulated database. The efficiency of neural networks has been proven in predicting the parameters as, ns and ηRx, with validation root mean squared error (RMSE) of 6.9%, 6.9% and 4.8%, in the first training approach, respectively. For training based on sites average condition RMSE was 20.7%, 3.0% and 8.8%, for as, ns and ηRx, respectively. The study showed the need for more scientific investigation for the other parameters, including information and input variables such as soil characteristics. As demonstrated in this study, the possibility of parameterizing 3-PG with ANNs or any machine learning technique may contribute to the broader use of this process-based model. In addition, artificial neural networks have great potential to assist in the calibration process of the 3-PG model, making this process more efficient by integrating environmental conditions and allowing the association between parameters. It is recommended to apply these ANNs for the conditions tested here.

Application of artificial neural network and multi-linear regression techniques in groundwater quality and health risk assessment around Egbema, Southeastern Nigeria

This paper examined the efficiency of artificial neural network (ANN) and multivariate linear regression (MLR) models in the prediction of groundwater quality parameters such as ecological risk index (ERI), pollution load index (PLI), metal pollution index (MPI), Nemerow pollution index (NPI), and geoaccumulation index (Igeo). 40 groundwater samples were collected systematically and analyzed for mainly heavy metals. Results revealed that adopting measured parameters is effective in modeling the parameters with high level of accuracy. Contamination factor results reveal that Ni, Zn, Pb, Cd, and Cu have relatively low values < 1 within the region while the Iron values ranged from low contamination to very high contamination (> 6). PLI, MPI, and ERI results indicated low pollution. NPI results indicated that the majority of the samples were heavily polluted. Quantification of Contamination results revealed that most of the sample's quality was geogenically influenced. Igeo results revealed that most of the samples had extreme pollution. The health risk assessment results revealed that children are substantially prone to more health risk more than adults. The ANN and MLR models showed a high effective tendency in the prediction of ERI, PLI, MPI, NPI and Igeo. Principal component analysis results showed appreciable variable loadings while the correlation matrix results reveal that there exists weak and positive correlation amongst elements. Based on the outcome of this study, this research recommends the use of ANN and MLR models in the prediction of groundwater quality parameters as they yielded positive, reliable, acceptable, and appropriate accuracy performances.

Design of a Computational Heuristic to Solve the Nonlinear Li閚ard Differential Model

In this study, the design of a computational heuristic based on the nonlinear Liénard model is presented using the efficiency of artificial neural networks (ANNs) along with the hybridization procedures of global and local search approaches. The global search genetic algorithm (GA) and local search sequential quadratic programming scheme (SQPS) are implemented to solve the nonlinear Liénard model. An objective function using the differential model and boundary conditions is designed and optimized by the hybrid computing strength of the GA-SQPS. The motivation of the ANN procedures along with GA-SQPS comes to present reliable, feasible and precise frameworks to tackle stiff and highly nonlinear differential models. The designed procedures of ANNs along with GA-SQPS are applied for three highly nonlinear differential models. The achieved numerical outcomes on multiple trials using the designed procedures are compared to authenticate the correctness, viability and efficacy. Moreover, statistical performances based on different measures are also provided to check the reliability of the ANN along with GA-SQPS.

Upscaling Shear Strength of Heterogeneous Oil Sands with Interbedded Shales Using Artificial Neural Network

SummaryUnderstanding the shear strength of caprock shale and oil sands is important in risk assessment of slope stability in open-pit mining, caprock integrity of in-situ thermal recovery, and optimization of bitumen production from oil sands. A robust and efficient upscaling technique is essential to model the impact of heterogeneity on the deformation and failure of oil sands and caprock shale. Although conventional analytical and numerical upscaling techniques are available, many of these methods consider oversimplified assumptions and have high computational costs, especially when considering the impact of spatially correlated interbedded shales on the shear strength. A machine learning enhanced upscaling (MLEU) technique that leverages the accuracy of local numerical upscaling and the efficiency of artificial neural network (ANN) is proposed here. MLEU uses a fast and accurate ANN proxy model to predict the anisotropic shear strength of heterogeneous oil sands with interbedded shales. The R2 values of the trained ANN models exceed 0.94 for estimating shear strengths in horizontal and vertical directions. The deviation of upscaled shear strength from numerical upscaled results is improved by 12–76% compared with multivariate regression methods like response surface methodology (RSM) and polynomial chaos expansion (PCE). In terms of computational efficiency, the proposed MLEU method can save computational effort by two orders of magnitude compared with numerical upscaling. MLEU provides a reasonable estimate of anisotropic shear strength while considering uncertainties caused by different distributions of shale beddings. With the increasing demand for regional scale modeling of geomechanical problems, the proposed MLEU technique can be extended to other geological settings, where weak beddings play a significant role and the impact of heterogeneity on shear strength is important.

Ultralow Power Always-On Intelligent and Connected SNN-Based System for Multimedia IoT-Enabled Applications

The recent advances in artificial neural networks (ANNs) have created immense opportunities to achieve excellent results on the Internet of Things (IoT), which serve the uses of various real-time smart applications, such as in computer vision and speech recognition. However, the efficiency of ANNs comes at the expense of a huge number of computational resources. This tends to necessitate larger and wider ANNs unapplicable for embedded systems with limited hardware resources, e.g., mobile and wearable devices. To that end, biologically realistic spiking neural networks (SNNs) are first employed to build an always-on intelligent and connected integrated IoT-enabled system with ultralow power consumption, where we encode the temporal dynamic stimuli into effective, efficient, and reconstructable spike patterns to facilitate the subsequent processing. Herein, neural encoding plays a key role in faithfully describing the temporally rich patterns for downstream cognitive tasks. Therefore, a novel nonlinear piecewise latency coding approach for a fully event-driven SNN system is developed. Moreover, a surrogate postsynaptic potential kernel function is utilized to address the nondifferential nature of the spike generation scheme when using the error backpropagation learning method. The effectiveness of the proposal tandem with SNNs has been corroborated by indicative empirical results on different data sets serving cognitive tasks.

Disaggregated approach to urban trip distribution: a comparative analysis between artificial neural networks and discrete choice models

Discrete choice models have been used over the years in disaggregated approaches to forecast destination choices. However, there are important constraints in some of these models that pose obstacles to using them, such as the Independence of Irrelevant Alternatives (IIA) property in the Multinomial Logit model, the need to assume specific structures and high calibration times, depending on the complexity of the case being evaluated. However, some of these mentioned constraints could be mitigated using Mixed Models or Nested Logit. Therefore, this paper proposes a comparative analysis between the Artificial Neural Network (ANNs), the Multinomial and Nested Logit models for disaggregated forecasting of urban trip distribution. A case study was conducted in a medium-sized Brazilian city, Santa Maria (RS), Brazil. The data used come from a household survey, prepared for the Urban Mobility Master Plan. For the sake of comparison, hit rates and frequency of trip distribution distances were analyzed, showing that ANNs can be as efficient as the Discrete Choice models for disaggregated forecasting of urban trip destination without, however, assuming some constraints. Finally, based on the results obtained, the efficiency of ANNs is observed for predicting alternatives with a low number of observations. They are important tools for obtaining Origin-Destination matrices from incomplete sample matrices or with a low number of observations. However, it is important to mention that discrete choice models can provide important information for the analyst, such as statistical significance of parameters, elasticities, subjective value of attributes, etc.

Artificial Neural Network for Discrimination and Classification of Tropical Soybean Genotypes of Different Relative Maturity Groups

Soybean has a recognized narrow genetic base that often makes it difficult to visualize available genetic and phenotypic variability and identify superior genotypes during the selection process. However, the phenotypic expression of soybean plants is highly affected by photoperiod and the cultivation of a given variety is performed in the latitude range that presents ideal conditions for its development based on its relative maturity group (RMG) for the optimization of the phenotypic expression of its genotype. Based on the above, this study aimed to evaluate the efficiency of artificial neural networks (ANNs) as a tool for the correct discrimination and classification of tropical soybean genotypes according to their relative maturity group during the population selection process with the aim of optimizing the phenotypic performance of these selected genotypes. For this purpose, three biparental populations were synthesized, one with a wide genetic variability for the RMG character obtained from the hybridization between genitors of maturity groups RMG 5 (Sub-tropical 23° LS) × RMG 9.4 (Tropical 0° LS) and two populations with a narrow variability obtained between genitors RMG 7.3 (Tropical 20° LS) × RMG 9.4 and RMG 5.3 × RMG 6.7, respectively. Criteria for comparing the developed ANN architecture with Fisher’s linear and Anderson’s quadratic parametric discriminant methodologies were applied to the data for the discrimination and classification of the genotypes. ANN showed an apparent error rate of less than 8.16% as well as a low influence of environmental factors, correctly classifying the genotypes in the populations even in cases of reduced genetic variability such as in the RMG 5 × RMG 6 population. In contrast, the discriminant functions were inefficient in correctly classifying the genotypes in the populations with genealogical similarity (RMG 5 × RMG 6) and wide genetic variability, with an error rate of more than 50%. Based on the results of this study, ANN can be used for the discrimination of genotypes in the initial generations of selection in breeding programs for the development of high performance cultivars for wide and reduced photoperiod amplitudes, even with fewer selection environments, more efficiently, and with fewer time and resources applied. As a result of similarity between the parents, ANN can correctly classify genotypes from populations with a narrow genetic base, in addition to pure lines and genotypes with a high degree of inbreeding.

DETECTING EPISTATIC EFFECTS UNDER A QUALITATIVE CONTEXT: IMPORTANCE AND QUANTIFICATION

The objectives of this work were to simulate and quantify epistatic effects on oligogenic traits and to verify the efficiency of Artificial Neural Networks (ANN) and Ridge Regression Best Linear Unbiased Predictor (RRBLUP) in the prediction of genetic values of oligogenic traits controlled by epistatic genes. We simulated 10 F2 populations in Hardy-Weinberg equilibrium with 800 individuals, each. The individuals were genotyped with 105 codominant markers equidistantly distributed along 10 chromosomes. Genotypic values were simulated for each individual considering five oligogenic traits controlled by two biallelic loci, according to five different epistatic models: duplicate recessive genes, dominant and recessive interaction, duplicate dominant genes, recessive epistasis, and dominant epistasis. RRBLUP, as well as an ANN, were used to perform genomic selection. The coefficient of determination of the regression model revealed a mean epistatic effect ranging from 13.3% in the duplicate recessive genes to 62.5% in the duplicate dominant genes. The identification of epistatic genes was superior in the ANN model compared with the RRBLUP approach. Our result reinforces the potential of ANN in predicting genetic values in situations where other than linear or quadratic relationships are present. The results presented in this work offer important insights about the exploration of epistasis in the qualitative context. In situations where epistatic effects are completely ignored, they can play a role on genetic values, as seen for the duplicative dominant genes.

Improvement of the five-hole probe calibration using artificial neural networks

In the present study, the artificial neural networks (ANNs) technique was implemented to link non-dimensional pressure coefficients and flow characteristics to calibrate a five-hole probe. The experimental data of this work were obtained from a subsonic open-circuit wind tunnel at the velocity of 10 m/s. Here, the efficiency of ANNs was compared with two conventional data reduction methods, including linear interpolation technique and 5th-order polynomial surface fit algorithm. Based on the statistical parameters of calibration data, it was concluded that the radial basis function (RBF) algorithm was more accurate and had more flexibility compared to the multi-layer perceptron (MLP) regression algorithm, the linear interpolation and 5th-order polynomial methods. In the RBF method, the mean absolute errors of 0.11, 0.64, 0.02 and 0.03 were achieved for α, β, Cpt and Cps , respectively. Furthermore, the effects of training data reduction and data selection on the performance of RBF were studied. The accuracy of the proposed RBF method was analyzed at different α angles and for random test data. Finally, the influence of increasing number of test data on the efficiency of calculated RBF method was evaluated.

Fundamentals and Applications of Artificial Neural Network Modelling of Continuous Bifidobacteria Monoculture at a Low Flow Rate

The application of artificial neural networks (ANNs) to mathematical modelling in microbiology and biotechnology has been a promising and convenient tool for over 30 years because ANNs make it possible to predict complex multiparametric dependencies. This article is devoted to the investigation and justification of ANN choice for modelling the growth of a probiotic strain of Bifidobacterium adolescentis in a continuous monoculture, at low flow rates, under different oligofructose (OF) concentrations, as a preliminary study for a predictive model of the behaviour of intestinal microbiota. We considered the possibility and effectiveness of various classes of ANN. Taking into account the specifics of the experimental data, we proposed two-layer perceptrons as a mathematical modelling tool trained on the basis of the error backpropagation algorithm. We proposed and tested the mechanisms for training, testing and tuning the perceptron on the basis of both the standard ratio between the training and test sample volumes and under the condition of limited training data, due to the high cost, duration and the complexity of the experiments. We developed and tested the specific ANN models (class, structure, training settings, weight coefficients) with new data. The validity of the model was confirmed using RMSE, which was from 4.24 to 980% for different concentrations. The results showed the high efficiency of ANNs in general and bilayer perceptrons in particular in solving modelling tasks in microbiology and biotechnology, making it possible to recommend this tool for further wider applications.