Multilayer Feedforward Neural Network Model Research Articles

Research on weathering recognition model of ancient glass based on three-layer neural network

In this paper, a three-layer neural network-based weathering recognition model for ancient glass is proposed to cope with the limitations of traditional weathering analysis methods in dealing with complexity and large-scale data. By constructing a multi-layer feed-forward neural network model containing input, hidden and output layers, this paper achieves high-precision classification of the weathering state of ancient glass. The experimental results show that the accuracy of the model reaches 98.0% and 100% on the training set and test set, respectively, which proves the effectiveness and efficiency of the method in the recognition of ancient glass weathering. This study demonstrates the potential of neural network application in cultural relics conservation and provides a theoretical basis for further optimisation of cultural relics identification techniques.

Boosted multilayer feedforward neural network with multiple output layers

This research introduces the Boosted Ensemble deep Multi-Layer Layer Perceptron (EdMLP) architecture with multiple output layers, a novel enhancement for the traditional Multi-Layer Perceptron (MLP). By adopting a layer-wise training approach, EdMLP enables the integration of boosting techniques within a single model, treating each layer as a weak learner, resulting in substantial performance gains. Additionally, the inclusion of layer-wise hyperparameter tuning allows optimization of individual layers thereby reducing the tuning time. Furthermore, the ensemble deep architecture’s versatility can be extended to other neural network-based models, such as the Self Normalized Network (SNN) where experiments demonstrate substantial performance enhancements yielded by the EdSNN compared to the standard original SNN model. This research underscores the potential of the EdMLP, and the Ed architecture in general as a powerful tool for improving the performance of various multilayer feedforward neural network models. The source code of this work is publicly accessible from the authors GitHub.

Integrated analysis of wireline logs analysis, seismic interpretation, and machine learning for reservoir characterisation: Insights from the late Eocene McKee Formation, onshore Taranaki Basin, New Zealand

This study aims to integrate seismic and well data for re-assessment and delineating the geological and petrophysical characteristics of the Late Eocene McKee Formation. Despite the McKee Formation being a hydrocarbon-producing field, there have been no previous integrated studies conducted to characterize its deposits in the Taranaki Basin. To execute these tasks, the process involves the interpretation of seismic data, wireline analysis, and 3D porosity volume prediction using multi-layered feed-forward neural network models. The seismic data was interpreted and found to contain low displacement faults due to intricate geometric complexities, the contour map that was generated shows lateral extension and thickness variation, and the region of direct hydrocarbon indicators was detected using the root-mean-square. Additionally, the seismic inversion provides a comprehensive understanding of the acoustic impedance variations from low to high value, which signifies the presence of mudstone clasts content. Furthermore, in the wireline log analysis, the reservoir was found to be a Silici-clastic and carbonate field owing to the presence of dolomite, calcite, and quartz. After applying cutoffs, petrophysical parameters were evaluated. The estimated values for effective porosity, reservoir thickness, shale volume, and water and hydrocarbon saturation were 15.04 % to 17.29 %, 85.44 m to 122.41 m, 8.99 to 14.87 % and 12.97–27.10 %, and 72.9 % to 87.03 %, respectively. The outcome of this extensive research aids in accurately characterizing this region for the future prospectively of the field. This integrated study can be applied and adapted to other basins in New Zealand and the southwestern Pacific. The proposed workflow is also suitable for analogous scenarios in various basins worldwide.

Spatial Air Quality Assessment in Peninsular Malaysia: Insights from Environmetric Analyses and Artificial Neural Networks

The research investigated the spatial distribution of air quality in Peninsular Malaysia to offer crucial insights for mitigating air quality degradation. Utilizing historical data obtained from the Malaysian Department of Environment, eleven years' worth (2011-2021) of daily readings of PM10, O3 , SO2 , NO2 , and CO from thirty-four monitoring stations were subjected to environmetric techniques and Artificial Neural Network analysis. HACA successfully classified seven stations into the High Pollution Cluster (HPC), seven stations into the Moderate Pollution Cluster (MPC), and twenty stations into the Low Pollution Cluster (LPC). The discriminant analysis demonstrated a correct assignment rate of 91.87%, suggesting that all five parameters were able to discriminate with a significance level of p<0.0001. Principal Component Analysis (PCA) revealed two Variance Factors (VFs) across all clusters, with cumulative variances of 69.43% (HPC), 82.32% (MPC), and 62.16% (LPC) respectively. Furthermore, employing a Multi-Layer Perceptron Feed-Forward Artificial Neural Network (MLP-FF-ANN) model to predict API readings yielded a significant and strong correlation, as indicated by R2 =0.7774 and RMSE=9.9048. These findings hold significant potential in informing the adoption of effective preventative and management strategies.

Prediction and Optimization of the Thermal Properties of TiO2/Water Nanofluids in the Framework of a Machine Learning Approach

In this study, comparing multiple models of machine learning, a multiple linear regression (MLP), multilayer feed-forward artificial neural network (BP) model, and a radial-basis feed-forward artificial neural network (RBF-BP) model are selected for the optimization of the thermal properties of TiO₂/water nanofluids. In particular, the least squares support vector machine (LS-SVM) method and radial basis support vector machine (RB-SVM) method are implemented. First, curve fitting is performed by means of multiple linear regression in order to obtain bivariate correlation functions for thermal conductivity and viscosity of the nanofluid. Then the aforementioned models are used for a predictive analysis of the dependence of its thermal conductivity and viscosity on temperature and volume fraction. The results show that the least squares support vector machine (LS-SVM) has a prediction accuracy higher than the other models. The model predicts the thermal conductivity of TiO₂/water MSE = 1.0853 × 10⁻⁶, R²= 0.99864, MAE = 0.00092, RMSE = 0.00104, and the viscosity of TiO₂/water MSE = 8.1397 × 10⁻⁶, R²= 0.99995, MAE = 0.00074, RMSE = 0.0009.

Multi-workflow scheduling and resource provisioning in Mobile Edge Computing using opposition-based Marine-Predator Algorithm

Workflow Scheduling in Mobile Edge Computing (MEC) tries to allocate the best possible set of resources for the workflows, considering objectives such as deadline, cost, energy, Quality of Service (QoS), and so on. However, MEC may be under different workloads from the IoT and this may not have the required amount of resources to efficiently handle the workflows. To mitigate this problem, in this paper, we use proactive resource provisioning and workload prediction methods. For this purpose, we present a workload prediction method using a multilayer feed-forward Artificial Neural Network (ANN) model and apply its results for resource provisioning. Afterward, we present an opposition-based version of the Marine-Predator Algorithm (MPA) algorithm, denoted as OMPA. In this algorithm, we present a probabilistic opposition-based learning (OBL) method, which benefits from the OBL, Quasi OBL, and dynamic OBL methods. Afterward, the OMPA algorithm is used for training the multi-layer feed-forward ANN model and multi-workflow scheduling by taking into account factors such as the makespan and number of Virtual Machines (VMs). Extensive experiments conducted in the iFogSim simulator and on the NASA and Saskatchewan datasets indicate that the proposed scheme can achieve better results compared to other metaheuristic algorithms and scheduling schemes.

Prediction of the Compressive Strength of Waste-Based Concretes Using Artificial Neural Network.

In the 21st century, numerous numerical calculation techniques have been discovered and used in several fields of science and technology. The purpose of this study was to use an artificial neural network (ANN) to forecast the compressive strength of waste-based concretes. The specimens studied include different kinds of mineral additions: metakaolin, silica fume, fly ash, limestone filler, marble waste, recycled aggregates, and ground granulated blast furnace slag. This method is based on the experimental results available for 1303 different mixtures gathered from 22 bibliographic sources for the ANN learning process. Based on a multilayer feedforward neural network model, the data were arranged and prepared to train and test the model. The model consists of 18 inputs following the type of cement, water content, water to binder ratio, replacement ratio, the quantity of superplasticizer, etc. The ANN model was built and applied with MATLAB software using the neural network module. According to the results forecast by the proposed neural network model, the ANN shows a strong capacity for predicting the compressive strength of concrete and is particularly precise with satisfactory accuracy (R² = 0.9888, MAPE = 2.87%).

Development and comparative selection of surrogate models using artificial neural network for an integrated regenerative transcritical cycle

Surrogate models are becoming increasingly important in replacing the computationally-expensive physics-based simulation models in many applications, such as system optimization, sensitivity analysis and design space exploration. As one of the fastest-growing field, machine learning, specifically artificial neural networks (ANN) have been adapted to model various energy systems. In the present study, five ANN-based surrogate models are developed in replacing the physics-based model of a novel regenerative transcritical power cycle using methanol as the working fluid that is integrated with a small modular reactor. The input layer of the surrogate models consists of the seven design parameters of the cycle, and the output layer returns the 1st-law efficiency, levelized cost of energy and penalty. The evaluation results show that all five candidate surrogate models have demonstrated high R2 score, low relative absolute errors (RAE) and low L1 losses, with the separate multi-layer feed-forward (MLF) neural network model outperforming the others. Once coupled with global optimization, the surrogate model is expected to find the optimal design parameters in order to minimize levelized cost of energy (LCOE) and penalty value in the system.

Predicting the Compressibility Factor of Natural Gas by Using Statistical Modeling and Neural Network

Simulating the phase behavior of a reservoir fluid requires the determination of many parameters, such as gas–oil ratio and formation volume factor. The determination of such parameters requires knowledge of the critical properties and compressibility factor (Z factor). There are many techniques to determine the compressibility factor, such as experimental pressure, volume, and temperature (PVT) tests, empirical correlations, and artificial intelligence approaches. In this work, two different models based on statistical regression and multi-layer-feedforward neural network (MLFN) were developed to predict the Z factor of natural gas by utilizing the experimental data of 1079 samples with a wide range of pseudo-reduced pressure (0.12–25.8) and pseudo reduced temperature (1.3–2.4). The statistical regression model was proposed and trained in R using the “rjags” package and Markov chain Monte Carlo simulation, while the multi-layer-feedforward neural network model was postulated and trained using the “neural net” package. The neural network consists of one input layer with two anodes, three hidden layers, and one output layer. The input parameters are the ratio of pseudo-reduced pressure and the pseudo-reduced temperature of the natural hydrocarbon gas, while the output is the Z factor. The proposed statistical and MLFN models showed a positive correlation between the actual and predicted values of the Z factor, with a correlation coefficient of 0.967 and 0.979, respectively. The results from the present study show that the MLFN can lead to accurate and reliable prediction of the natural gas compressibility factor.

Design of a NiTiHf shape memory alloy with an austenite finish temperature beyond 400 °C utilizing artificial intelligence

This paper details the design process of a ternary NiTiHf shape memory alloy (SMA) with an austenite finish temperature (Af) beyond 400 °C. Specifically, available experimental data on the ternary NiTiHf SMA system was utilized to construct a database, which was employed to train and test a machine learning (ML) algorithm to predict the ideal NiTiHf SMA composition to exhibit an Af beyond 400 °C and a relatively smaller hysteresis. For this purpose, a multi-layer feedforward neural network (MLFFNN) model was proposed, trained, and tested. Consequently, the Ni49.7Ti26.6Hf23.7 and Ni50Ti27Hf23 alloys predicted by this ML algorithm were selected for validation experiments to assess the accuracy of the ML model’s predictions. As a result, the Ni49.7Ti26.6Hf23.7 alloy with an Af temperature of 403.5 °C and remarkable cyclic stability was established as a new NiTiHf SMA composition, which can be utilized in applications demanding reversible austenite-to-martensite phase transformation beyond 400 °C.

Stability of biochar derived from banana peel through pyrolysis as alternative source of nutrient in soil: feedforward neural network modelling study.

Biochar derived from banana peels can be used as an alternative nutrient in the soil that can promote crop growth while reducing fertiliser usage. Biochar stability has proportional relationship to biochar residence time in the soil and potassium is one of the vital nutrients needed for plant growth. This research aims at providing optimum pyrolysis operating conditions like temperature, residence time, and heating rate using banana peels as feedstock. An electrical tubular furnace was used to conduct the pyrolysis process to convert banana peels into biochar. The elemental compositions of biochar are potassium, oxygen (O), and carbon (C) content. The O:C ratio was used as the biochar stability indicator. Analysis of results showed that operating temperature has the most remarkable effect on biochar yield, biochar stability, and biochar's potassium content. In addition, a multilayer feedforward artificial neural network model was developed for the pyrolysis process. Eleven training algorithms were selected to model the multi-input multi-output neural network (MIMO). The most suitable training algorithm was identified through four performance criterions which are root mean square error (RMSE), mean absolute error (MSE), mean absolute percentage error (MAPE), and regression (R2). The results show that the Levenberg-Marquardt backpropagation training algorithm has the lowest error. From the chosen training algorithm, neural network was trained, and optimum operating parameters for banana peel were predicted at 490°C, 110min, and 11°C/min with a high yield of 47.78%, O/C ratio of 0.2393, and 14.04 wt. % of potassium.

Prediction of floodwater impacts on vehicle blockages at bridges using artificial neural network

During extreme flood events, various debris like floating vehicles can block the bridges in urban rivers and floodplains. Blockage of vehicles can influence the floodwater hydrodynamics and potentially on the flood risk implications. Such obstructions often raise upstream water levels with back water effects, causing more water to be redirected into nearby metropolitan areas. This study attempts at evaluating artificial neural network (ANN) model in predicting the variations in floodwater depths and velocities along the channel centreline based on the changes in flowrate and distances from the inlet. The floodwater depth and velocity variations were obtained for three different types of bridges at specified sites along the channel centreline with three incoming discharges. A multilayer feedforward neural network (FFNN) model was used to investigate the effects of discharge (Q) and distance, on depth variation rate (D) or velocity (V). Additionally, a comparison study was done between 2 input 1 output and 2 input 2 output i.e. single output (depth variation rate (D) or velocity (V) versus multi-output depth variation rate (D) and velocity (V) for all the three models of bridges that are blocked by vehicles. The study has applied 12 training algorithms (TA) in the ANN modelling to optimize the TA that is most suitable for the dataset of three different bridges. The optimization is based on the performance criterion namely regression (R), mean squared error (MSE), mean absolute error (MAE), mean absolute percentage (MAPE), accuracy and coefficient of determinant (R2). Bayesian regularization backpropagation (BR) training algorithm gives a highest accuracy when compared in all three bridges. The scenario 2 input 2 output gave greatest accuracy results compared to 2 input 1 output. The findings showed a reliable estimation of significant impacts on the flow propagations and the hydrodynamic processes along rivers and floodplains. This study can help the decision makers in effective river and floodplain management practices.

Estimation of cutting forces and tool tip temperature in turning operation with help of artificial neural network

Estimation of cutting forces for different metal cutting processes are very important to get the idea about specific cutting power required for the machining and for the design of a machine-fixture-tool system. The main objective of this work is to estimate the machining variables and study the effect of cutting parameters on them. In this work cutting forces as main cutting force, feed force, thrust force and tool tip temperature have been selected as machining variables and their values have been estimated under different combination of cutting parameters such as tool geometry rake angle, entering angle and cutting speed. To study the effect of process parameters initially linear regression equations have been generated and then the effect of process parameters on selected machining variables have been shown with help of 3-D surface plots. For the estimation of machining variables as cutting forces and tool tip temperature, different models have been created with help of an expert system Artificial Neural Network (ANN). In this work different models with combinations of transfer functions, number of neurons, number of epochs have been taken ad best possible model has been found out. Then to train, test the different established Multilayer Feed-Forward Neural Network (MLFFNN) models and for validation the experimental data have been used. The value estimated with help of different ANN models have been collected and find out the coefficient of regression R2, Root Mean Square Error RMSE and Mean absolute percentage error MAPE with respect to actual or experimental values. Based on the higher value of R2, lower value of RMSE and MAPE best model has been selected. With help of selected model machining variables have been estimated and compared with actual values. From this research work it has been pointed out that the model with Bayesian Regularization transfer function, 30 number of neurons, 300 number of epochs performed well with R2 = 0.9996, RMSE = 6.161 and MAPE = 1.67 %. This model provide better performance as compared to other ANN models have been trained with Levenberg-Marquardt and Conjugate Gradient training function.

FORECASTING ELECTRICITY SUPPLIED IN TURKEY USING HOLT-WINTERS’ MULTIPLICATIVE METHOD AND ARTIFICIAL NEURAL NETWORK (ANN) MODELS

Electricity is one of the most essential necessities in today’s world and has an important role for the development of societies and economics. The need for electricity is expanding continuously due to increasing population, urbanization and industrialization. Hence, the purpose of this study was to develop the best model for forecasting electricity supplied in Turkey by applying the multiplicative Holt-Winters method and multilayer feed-forward neural network model. The monthly electricity supplied in Turkey from January 2000 until December 2019 were obtained from monthly electricity statistics report presented by the International Energy Agency (EIA). The data were divided into two sets comprising in-sample data from January 2000 until December 2015 and out-sample data from January 2016 to December 2019. The multiplicative Holt-Winters was used since the electricity supplied in Turkey exhibit trend and seasonal gave the out-sample forecast of 3.6990. The best multilayer feed forward neural network (MFFNN) model with three input lag variable, one hidden node, one output node, sigmoid transfer function in hidden layer and linear transfer function in output layer gave the out-sample forecast of 2.1483. Hence it can be concluded that, the multilayer feed-forward neural network model is more accurate than multiplicative Holt-Winters method to forecast the electricity supplied in Turkey.

Real Time Control Application of the Robotic Arm Using Neural Network Based Inverse Kinematics Solution

Robotic arms are widely used in many industrial applications at present. The control of robotic arms involves position coordination Cartesian space by a forward/inverse kinematics solution method. The inverse kinematics is difficult for real-time control applications, computational requirements are intensive and the run-time is high. The traditional solution methods used geometric, algebraic, and numerical iterative techniques are inadequate and slow in the inverse kinematics solution. Recently, alternative solution methods based on artificial intelligence techniques have been developed to solve the inverse kinematics problem. In this study, a multi-layered feed-forward Artificial Neural Network model was developed to solve the inverse kinematics of the 5 degrees of freedom robotic arm. Using the Proportional-Integral control algorithm combined with this Artificial Neural Network model, the real-time position control of the robotic arm was accomplished. The obtained results were compared with the PI control supported by an analytical inverse kinematics solution in real-time. The results showed that the PI control combined with Artificial Neural Network has superior tracking ability, smaller control error, and better absolute fit to the reference.

Prediction of velocity, gas content from neural network modeling and estimation of coal bed permeability from image log in coal bed methane reservoirs: Case study of South Karanpura Coalfield, India

P-wave velocity and gas content of major coal seams are predicted from three wells from the South Karanpura coalfield, India. Multilayered feed-forward neural network (MLFN) model is developed for prediction P-wave velocity from two wells using three input log parameters: gamma ray, resistivity and bulk density. The model is tested on coal seams: Sayal, Balkudra, Banasgarah and Argada of three wells namely; A, B and C with satisfactory goodness of fit (R2 = 0.77). The ratio of P-wave and S-wave velocities (Vp/Vs) ranges from 1.68 to 3.10 in the coal seams of this field. Cleat density using image log is obtained from four major coal seams namely; Balkudra, Kurse, Hathidari, Banasgarah varying from 1/m to 11/m. Ash content varies from 15 to 51% whereas gas content of major coal seams varies from 2.30 to 4.4 10−3 kg/m3. There is a good correlation between ash content and Vp/Vs of coal seams under the study area. MLFN model for prediction of gas content of the above-mentioned coal seams from two well is trained with input parameters such as: Vp/Vs, bulk density, cleat density and ash content. The model estimates gas content of other coal seams: Argada from well A, Hathidari from well B and Balkudra, Kurse, Banasgarah from well C with satisfactory R2 = 0.77. Maximum horizontal stress orientation is observed from azimuthal shear wave anisotropy from cross multipole array acoustic (XMAC) log for a well A. It varies from N110° to N115° in these coal seams. Permeability of coal seams are computed from X-tended Range Microresistivity Imager (XRMI) tool from a well C. Permeability of coal is varying from 0.5 md in Saunda seam to 17.29 md in Banasgarah seam. The estimated permeability matches well with the pre-fracture permeability of these seams. It is observed that increase in cleat density enhances coal seam fracture permeability in coal seam.

Classification of Cellulosic Insulation State Based on Smart Life Prediction Approach (SLPA)

The state of cellulosic solid kraft paper (CSKP) insulation, to a large extent, is an indication of a transformer’s health. It not only reflects the condition of transformer but also diagnose its residual life. The quantity of 2-furfuraldehyde (2-FAL), carbon dioxide (CO2), and carbon monoxide (CO) dissolved in the transformer oil are useful diagnostic indicators to predict the state of the CSKP insulation. In this work, the current physical state of the CSKP is determined with the help of easily measurable parameters, like temperature, moisture, and the aging time. Here, the degree of deterioration of CSKP insulation has been determined using an integrated insulation health assessment system. This technique integrates a two-stage system comprising of a neural network (NN) model followed by a Smart Life Prediction Approach (SLPA). A thermo-moisture-aging multi-layer feed-forward NN model has been developed to predict the concentrations of 2-FAL, CO2, and CO, which are further correlated to estimate the Degree of Polymerization (DP) values adopting an SLPA. The advantage of the proposed integrated system is that it provides an alternative means of paper health assessment based on Dissolved Gas Analysis (DGA) without estimating dissolved gas concentrations in oil, thereby avoiding the use of sophisticated measuring instruments. The optimal configuration of the NN model has been achieved at minimum iterations with an average cross-validation mean square error of 3.78 × 10−7. The proposed system thereby avoids destructive and offline measurement of DP and facilitates real-time condition monitoring of oil-immersed transformers. The test results of the developed system show considerable reliability in determining insulation health using easily measurable parameters. Furthermore, the system’s performance is compared with reported work and has been found to provide encouraging outcomes.

Modeling of phenol removal from water by NiFe2O4 nanocomposite using response surface methodology and artificial neural network techniques

This study demonstrates the synthesis, characterization, and modeling of nickel ferrite nanocomposite, NiFe2O4 (NFC) as an adsorbent for the phenol contaminated aqueous environment. The characterization of the prepared NFC was performed with X-ray diffraction spectroscopy (XRD), scanning electron microscopy (SEM), transmission electron microscopy (TEM) and vibrating sample magnetometer (VSM) techniques. The optimization and modeling of phenol removal using NFC was done through central composite design (CCD) and effective parameters of CCD were measured as input variables including the amount of NFC, pH, contact time and initial phenol concentration. The predicted results showed that the adsorption process using NFC as adsorbent had the maximum phenol removal (~99%) under predicted optimal conditions (pH = 7.67, NFC dosage = 0.15 g at room temperature), which also corresponded to the experimental values. In addition, a multilayer feed-forward artificial neural network (ANN) model was used to obtain a speculative phenol removal model. The network was trained for six replications after selection of the best neuron number for hidden layer. The value of MSE trained network was found to be 6.01718e-3 along with regression coefficient (R2 = 0.9934) that indicated satisfactory relationship. Isothermal modeling of phenol adsorption onto NFC was performed using well-known Temkin, Freundlich and Langmuir models and it was clear from the higher R2 value of 0.961 that the Langmuir model was significantly followed by experimental data. The maximum Langmuir adsorption capacity was found to be 274.72 mg/g at the optimal conditions. The obtained results prove that NFC could be an effective adsorbent for elimination of phenol contaminant from aqueous environment.

Retraction Note to: Artificial neural networks application to predict the compressive damage of lightweight geopolymer

In this work, compressive strength of lightweight geopolymers produced by fine fly ash and rice husk–bark ash together with palm oil clinker (POC) aggregates has been investigated experimentally and modeled based on artificial neural networks. Different specimens made from a mixture of fine fly ash and rice husk–bark ash with and without POC were subjected to compressive strength tests at 2, 7, and 28 days of curing. A model based on artificial neural networks for predicting the compressive strength of the specimens has been presented. To build the model, training and testing using experimental results from 144 specimens were conducted. The data used in the multilayer feed-forward neural networks models are arranged in a format of six input parameters that cover the quantity of fine POC particles, the quantity of coarse POC particles, the quantity of FA + RHBA mixture, the ratio of alkali activator to ashes mixture, the age of curing and the test trial number. According to these input parameters, in the neural networks model, the compressive strength of each specimen was predicted. The training and testing results in the neural networks model have shown a strong potential for predicting the compressive strength of the geopolymer specimens in the considered range.

Water quality and macrophytes in the Danube River: Artificial neural network modelling

Ecological assessment of large rivers such as the Danube is a challenging task. Eutrophication was reported as one of the main drivers that structure aquatic communities in the Danube basin. Due to their sedentary nature, relatively slow growth/ long life spans, and engineering role in aquatic ecosystems, macrophytes are widely used in the detection of nutrient enrichment. In this study, macrophyte presence-absence data within the 3 km long reaches obtained from the Joint Danube Survey (JDS3) were used to predict the water quality of the Danube river and its main tributaries. For each water quality variable (dissolved oxygen, nitrate-nitrogen, and orthophosphates), a multi-layer feed-forward artificial neural network model (ANN) was constructed using the macrophytes as explanatory variables. Despite the limited number of samples (123) along the wide trophic gradient of the Danube, the model showed good predictive performances for the main river channel. The highest discrepancy between observed and predicted water quality was obtained for the samples collected in the tributaries or downstream from the tributaries' mouth, where the model predicted better trophic conditions compared to measured ones. From 64 analysed macrophyte species, 28 were selected by sensitivity analysis as key water quality indicators (KIS) for at least one environmental variable. KIS mainly belonged to the eutrophic tolerant submerged or emerged species with broad ecological amplitude, which reflects the significance of the developed model for use on rivers subjected to nutrient pollution. However, the use of the developed predictive model is restricted to the river sections having a water velocity suitable for macrophytes growth. The developed ANN architecture represents the modelling approach which could be applied to other lotic systems and biological quality elements.