Inputs Of Artificial Neural Network Research Articles

Neural Network-Based Body Weight Prediction in Pelibuey Sheep through Biometric Measurements

This paper presents an intelligent system for the dynamic estimation of sheep body weight (BW). The methodology used to estimate body weight is based on measuring seven biometric parameters: height at withers, rump height, body length, body diagonal length, total body length, semicircumference of the abdomen, and semicircumference of the girth. A biometric parameter acquisition system was developed using a Kinect as a sensor. The results were contrasted with measurements obtained manually with a flexometer. The comparison gives an average root mean square error (RMSE) of 9.91 and a mean R2 of 0.81. Subsequently, the parameters were used as input in a back-propagation artificial neural network. Performance tests were performed with different combinations to make the best choice of architecture. In this way, an intelligent body weight estimation system was obtained from biometric parameters, with a 5.8% RMSE in the weight estimations for the best architecture. This approach represents an innovative, feasible, and economical alternative to contribute to decision-making in livestock production systems.

Artificial Neural Network Approach for Assessing Mechanical Properties and Impact Performance of Natural-Fiber Composites Exposed to UV Radiation.

Amidst escalating environmental concerns, short natural-fiber thermoplastic (SNFT) biocomposites have emerged as sustainable materials for the eco-friendly production of mechanical components. However, their limited durability has prompted research into the experimental evaluation of the deterioration of the mechanical characteristics of SNFT biocomposites, particularly under the influence of ultraviolet rays. However, conducting tests to evaluate the mechanical properties can be time-consuming and expensive. In this study, an artificial neural network (ANN) model was employed to predict the mechanical properties (tensile strength) and the impact performance (resistance and absorbed energy) of polypropylene reinforced with 30 wt.% short flax or wood pine fibers (referred to as PP30-F or PP30-P, respectively). Eight parameters were collected from experimental studies. The ANN input parameters comprised nondestructive test results, including mass, hardness, roughness, and natural frequencies, while the output parameters were the tensile strength, the maximum impact load, and absorbed energy. The model was developed using the ANN toolbox in MATLAB. The linear coefficient of correlation and mean squared error were selected as the metrics for evaluating the performance function and accuracy of the ANN model. They calculate the relationship and the average squared difference between the predicted and actual values. The data analysis conducted by the models demonstrated exceptional predictive capability, achieving an accuracy rate exceeding 96%, which was deemed satisfactory. For both the PP30-F and PP30-P biocomposites, the ANN predictions deviated from the experimental data by 3, 5, and 6% with regard to the impact load, absorbed energy, and tensile strength, respectively.

Predicting indoor 3D airflow distribution using artificial neural networks with two different architectures

To meet the thermal comfort requirements of room occupants, a fast and accurate method for predicting indoor high-resolution 3D airflow distribution is necessary, which can be combined with heating, ventilation, and air conditioning (HVAC) systems to adjust the indoor environment. Artificial neural networks (ANN) can establish complex mappings between variables with nonlinear relationships. The aim of this study was to verify the feasibility of an ANN for the fast and accurate prediction of indoor 3D airflow distribution. Considering two prediction strategies, ANN A and B were constructed. The inputs of both ANNs were the air supply velocity, air supply direction, and section location. ANN A output the airflow distribution of the entire room directly, whereas ANN B output the airflow distribution of the characteristic sections first and then the airflow distribution of other sections according to the contribution of the characteristic sections. The two ANNs’ performances on training and testing datasets were compared. Both ANNs performed well in the training and interpolation cases. However, their performances decreased rapidly in the extrapolation cases. In the case of a large gap between the input and output dimensions of the ANNs, the two-step prediction strategy improved the prediction accuracy, accelerated convergence, and alleviated data imbalance problems. The prediction time of ANN A and B was 0.89 s and 12.38 s, respectively. The ANN can predict indoor high-resolution 3D airflow distribution quickly and accurately, and can add functions that include identifying internal heat sources and air supply inlet position changes. ANN combined with HVAC could allow real-time control to regulate indoor airflow.

Prediction of the compressive strength of lightweight concrete containing industrial and waste steel fibers using a multilayer synthetic neural network

The use of waste and industrial steel fibers as part of the materials used in concrete can increase resistance and reduce cost and air pollution. It also saves energy. One of the important measures for technical inspections and assessment of the existing condition of structures, especially bridges, which is the most important communication factor, is to check the compressive strength. Considering that the calculation of compressive strength in the laboratory is done with the intervention of human power and is undoubtedly affected by human error, we decided to use it through.Predicting the mechanical properties of concrete reinforced with steel fibers based on artificial neural network models without the need to conduct any laboratory studies will save money in construction projects. Unlike classical methods in statistical theories, neural networks do not require any specific model or function along with limiting assumptions to linearize problems.For this purpose, this research was done with the aim of compensating this problem and with the aim of building a neural network with high accuracy that can make the desired predictions with the least error. In this research, this modeling was done using artificial neural networks (ANN) and Levenberg algorithm. The data used to train the neural network was collected from 45 different mixing schemes. Then the compressive strength of the sample is determined experimentally. The parameters considered for the ANN inputs are the values of steel fiber, water, water-cement ratio, cement and superlubricant. The objective data of this study included the compressive strength of each of these mixing designs at the ages of 7, 28 and 60 days. Then, to design the neural network, 75% of the data were considered as training data, 15% as target data and 15% as validation data. The compressive strength of concrete samples made from waste steel fibers increases. One of the reasons for this result is the placement and uniform distribution of fibers in the cement matrix, or in other words, the optimal amount of desired fibers in concrete. For experimental information and data, results can be seen with the help of neural network in data analysis. It was observed that the validation is correlated with a correlation coefficient above 99% and the constructed neural network has sufficient accuracy and validity.

Artificial Neural Network Technique For Seismic Fragility Analysis Of A Storage Tank Supported By Multi-Storey Frame

Abstract Fragility function, which defines the conditional probability of exceeding a limit state given an intensity measure (IM) of the earthquake, is an essential ingredient of modern approaches like the performance-based earthquake engineering methodology. However, the generation of such curves generally entails a high computational effort to account for epistemic and aleatory uncertainties associated with structural analysis and seismic load. Moreover, a certain probability function, such as the log-normal distribution, is usually assumed in order to carry out the conditional probability of failure of a structure, without any prior information on the correct probability distribution. In this paper, an artificial neural network (ANN) model is proposed to carry out fragility curves in order to avoid the aforementioned problems. In this respect, this paper investigates the following aspects: (i) implementation of an efficient algorithm to select proper seismic intensity measures as inputs for ANN, (ii) derivation of surrogate models by using the ANN techniques, (iii) computation of fragility curves by means Monte Carlo Simulations and (iv) validation phase.

Neural networks based linear (PCA) and nonlinear (ISOMAP) feature extraction for soil swelling pressure prediction (North East Algeria)

The swelling pressure (SP) of expansive soils is crucial for both geotechnical studies as well as practitioners. Multiple attempts have been made to correlate the SP with the properties of soil due to the difficulty of determining it in the laboratory. However, the large number of environmental and physical governing parameters makes accurate SP predictions difficult. In this paper, Artificial Neural Networks (ANNs) are used to assess accurate prediction of SP of soil. Dimension reduction techniques are intensely required for ANNs inputs. Feature extraction (FE) based dimension reduction (DR) methods map original multidimensional space into a space of reduced dimensionality. This paper presents a comparative study of linear FE using Principal Component Analysis (PCA) and nonlinear FE using ISOmetric MAPping (ISOMAP) for feed forward neural models to predict SP. Results showed that FE technique improves ANNs models compared to multiple linear regression (MLR) and ANNs model without DR. Moreover, nonlinear ISOMAP based DR technique has proven its effectiveness regarding performance metrics for five dimensions inputs (Dims), Determination coefficient (R2 = 0.923), Mean absolute percentage error (MAPE = 0.072), and Root mean square error (RMSE = 54.937) and Root relative squared error (RRSE = 0.383). Therefore, ISOMAP-ANN models can be adopted to solve geotechnical problems specially those of expansive soils which have a very complex and nonlinear structure.

Modeling and optimization of CO2 capture into mixed MEA-PZ amine solutions using machine learning based on ANN and RSM models

Carbon dioxide (CO2) sequestration by chemical absorption is widely regarded as the most effective method for its reduction in natural gas streams or flue gases from fossil fuel power plants. In this article, modeling and optimization of CO2 mass transfer flux (NCO₂) are investigated. A combination of Piperazine (PZ) and Monoethanolamine (MEA) amines has been used for CO2 absorption. Artificial neural networks (ANN) and Response Surface Methodology (RSM) were used to achieve goals. The dimensionless numbers for the input of ANNs and RSM was obtained using Buckingham's Pi theory. The resulting models can provide acceptable results in an effect of independent variables and the interaction between them by the impact on the objective function, to optimize the process of CO2 capture. In the RSM approach, the quadratic model is used. Optimized ANNs and a structure with the least error and the most matching with the experimental data were obtained. Both ANNs and RSM models showed acceptable prediction of experimental data with maximum R2 value of 0.9974 and 0.9723 respectively. Due to the mean squared error of5.2 × 10−4, the ANN is recommended for the development of absorption simulation models.

Gender Estimation with Parameters Obtained From the Upper Dental Arcade by Using Machine Learning Algorithms and Artificial Neural Networks

Objective: The aim of this study is to estimate gender with parameters obtained from the upper dental arcade by using machine learning algorithms and artificial neural networks. Methods: The study was conducted on cone-beamed computed tomography images of 176 individuals between the ages of 18 and 55 who did not have any pathologies or surgical interventions in their upper dental arcade. The images obtained were transferred to RadiAnt DICOM Viewer program in Digital Imaging and Communications in Medicine format and all images were brought to orthogonal plane by applying 3D Curved Multiplanar Reconstruction. Length and curvature length measurements were performed on these images brought to orthogonal plane. The data obtained were used in machine learning algorithms (ML) and artificial neural networks input and rates of gender estimation were shown. Results: In the study, an accuracy ratio of 0.86 was found with ML models linear discriminant analysis (LDA), quadratic discriminant analysis (QDA), logistic regression (LR) algorithm and an accuracy ratio of 0.86 was found with random forest (RF) algorithm. It was found with SHAP analyser of RF algorithm that the parameter of width at the level of 3rd molar teeth contributed the most to gender. An accuracy rate of 0.92 was found as a result of training for 500 times with multilayer perceptron classifier (MLCP), which is an artificial neural network (ANN) model. Conclusion: As a result of our study, it was found that the parameters obtained from the upper dental arcade showed a high accuracy in estimation of gender. In this context, we believe that the present study will make important contributions to forensic sciences.

Estimating 3D ground reaction forces in running using three inertial measurement units.

To understand the mechanisms causing running injuries, it is crucial to get insights into biomechanical loading in the runners' environment. Ground reaction forces (GRFs) describe the external forces on the body during running, however, measuring these forces is usually only possible in a gait laboratory. Previous studies show that it is possible to use inertial measurement units (IMUs) to estimate vertical forces, however, forces in anterior-posterior direction play an important role in the push-off. Furthermore, to perform an inverse dynamics approach, for modelling tissue specific loads, 3D GRFs are needed as input. Therefore, the goal of this work was to estimate 3D GRFs using three inertial measurement units. Twelve rear foot strike runners did nine trials at three different velocities (10, 12 and 14 km/h) and three stride frequencies (preferred and preferred ± 10%) on an instrumented treadmill. Then, data from IMUs placed on the pelvis and lower legs were used as input for artificial neural networks (ANNs) to estimate 3D GRFs. Additionally, estimated vertical GRF from a physical model was used as input to create a hybrid machine learning model. Using different splits in validation and training data, different ANNs were fitted and assembled into an ensemble model. Leave-one-subject-out cross-validation was used to validate the models. Performance of the machine learning, hybrid machine learning and a physical model were compared. The estimated vs. measured GRF for the hybrid model had a RMSE normalized over the full range of values of 10.8, 7.8 and 6.8% and a Pearson correlation coefficient of 0.58, 0.91, 0.97 for the mediolateral direction, posterior-anterior and vertical direction respectively. Performance for the three compared models was similar. The ensemble models showed higher model accuracy compared to the ensemble-members. This study is the first to estimate 3D GRF during continuous running from IMUs and shows that it is possible to estimate GRF in posterior-anterior and vertical direction, making it possible to estimate these forces in the outdoor setting. This step towards quantification of biomechanical load in the runners' environment is helpful to gain a better understanding of the development of running injuries.

Memristive hyperchaotic system-based complex-valued artificial neural synchronization for secured communication in Industrial Internet of Things

In this article, a memristive hyperchaotic system-based complex-valued Artificial Neural Network (ANN) synchronization for secured communication between Industrial Internet of Things (IIoT) nodes is proposed. The IIoT, an extension of the Internet of Things (IoT), has significant promise for resource use in the industrial sector and intelligent transformation. Security and privacy issues exist when industrial data is sent over an open network via sensor devices. For this, a secure key exchange protocol is necessary. Synchronized key exchange using an ANN is one of the solutions. Additionally, there is a dearth of research on the reciprocating training of ANNs and the use of a robust Pseudo-Random Number Generator (PRNG) to generate a shared input. This study explains how to use a hyperchaotic environment to swiftly assess how well ANNs finalize their coordination and synchronize for session key switch over. Reciprocating learning is used to synchronize two neural networks and transfer the ANN’s output across a public channel. When the ANNs have produced the same outputs in previous iterations, coordination is tested using that function. The proposed technique has lot of benefits, like: (1) using a hyperchaotic-guided PRNG to produce the ANN input sequences; (2) generating the session key via the public network using a reciprocal neuronal alignment of complex ANNs; and (3) allowing two communicating partners to identify complete synchronization more quickly when compared to earlier methodologies. The tested proposed technique outperforms comparable strategies in the literature, according to the results.

Noise Estimation and Type Identification in Natural Scene and Medical Images using Deep Learning Approaches

The image enhancement for the natural images is the vast field where the quality of the images degrades based on the capturing and processing methods employed by the capturing devices. Based on noise type and estimation of noise, filter need to be adopted for enhancing the quality of the image. In the same manner, the medical field also needs some filtering mechanism to reduce the noise and detection of the disease based on the clarity of the image captured; in accordance with it, the preprocessing steps play a vital role to reduce the burden on the radiologist to make the decision on presence of disease. Based on the estimated noise and its type, the filters are selected to delete the unwanted signals from the image. Hence, identifying noise types and denoising play an important role in image analysis. The proposed framework addresses the noise estimation and filtering process to obtain the enhanced images. This paper estimates and detects the noise types, namely Gaussian, motion artifacts, Poisson, salt-andpepper, and speckle noises. Noise is estimated by using discrete wavelet transformation (DWT). This separates the image into quadruple sub-bands. Noise and HH sub-band are high-frequency components. HH sub-band also has vertical edges. These vertical edges are removed by performing Hadamard operation on downsampled Sobel edge-detected image and HH sub-band. Using HH sub-band after removing vertical edges is considered for estimating the noise. The Rician energy equation is used to estimate the noise. This is given as input for Artificial Neural Network to improve the estimated noise level. For identifying noise type, CNN is used. After removing vertical edges, the HH sub-band is given to the CNN model for classification. The classification accuracy results of identifying noise type are 100% on natural images and 96.3% on medical images.

Optimizing the Treatment of Recycled Aggregate (>4 mm), Artificial Intelligence and Analytical Approaches.

Attached, old mortar removal methods are evolving to improve recycled aggregate quality. Despite the improved quality of recycled aggregate, treatment of recycled aggregate at the required level cannot be obtained and predicted well. In the present study, an analytical approach was developed and proposed to use the Ball Mill Method smartly. As a result, more interesting and unique results were found. One of the interesting results was the abrasion coefficient which was composed according to experimental test results; and the Abrasion Coefficient enables quick decision-making to get the best results for recycled aggregate before the Ball mill method application on recycled aggregate. The proposed approach provided an adjustment in water absorption of recycled aggregate, and the required reduction level in water absorption of recycled aggregate was easily achieved by accurately composing Ball Mill Method combinations (drum rotation-steel ball). In addition, artificial neural network models were built for the Ball Mill Method The artificial neural network input parameters were Ball Mill Method drum rotations, steel ball numbers and/or Abrasion Coefficient, and the output parameter was the water absorption of recycled aggregate. Training and testing processes were conducted using the Ball Mill Method results, and the results were compared with test data. Eventually, the developed approach gave the Ball Mill Method more ability and more effectiveness. Also, the predicted results of the proposed Abrasion Coefficient were found close to the experimental and literature data. Besides, an artificial neural network was found to be a useful tool for the prediction of water absorption of processed recycled aggregate.

Short–long-term streamflow forecasting using a coupled wavelet transform–artificial neural network (WT–ANN) model at the Gilgit River Basin, Pakistan

Abstract Streamflow forecasting is highly crucial in the domain of water resources. For this study, we coupled the Wavelet Transform (WT) and Artificial Neural Network (ANN) to forecast Gilgit streamflow at short-term (T0.33 and T0.66), intermediate-term (T1), and long-term (T2, T4, and T8) monthly intervals. Streamflow forecasts are uncertain due to stochastic disturbances caused by variations in snow-melting routines and local orography. To remedy this situation, decomposition by WT was undertaken to enhance the associative relation between the input and target sets for ANN to process. For ANN modeling, cross-correlation was used to guide input selection. Corresponding to six intervals, nine configurations were developed. Short-term intervals performed best, especially for T0.33; intermediate intervals showed decreasing performance. However, interestingly, performance regains back to a decent level for long-term forecasting. Almost all the models underestimate high flows and slightly overestimate low- to intermediate-flow conditions. At last, inference implicitly implies that shorter forecasting benefits from extrapolated trends, while the good results of long-term forecasting is associated to a larger recurrent pattern of the Gilgit River. In this way, weak performance for intermediate forecasting could be attributed to the insufficient ability of the model to capture either one of these patterns.

A new intelligent decision-maker method determining the optimal connection point and operating conditions of hydrogen energy-based DGs to the main grid

This study presents a new two-step intelligent decision-maker method using hydrogen energy-based distributed generators (HEDGs) to contribute to the reliability, durability, and stability of power transmission system in Bursa. In the first stage, the proposed method uses the power flow parameters evaluation (PFPE) algorithm to define the possible appropriate connection point of HEDGs by determining the electrical parameters. Then, to determine the conditions in which the HEDGs connected to the grid should be switched on, the power flow data such as load status, bus bar powers, and, line capacities are evaluated with the artificial neural network (ANN)-based method with a scaled conjugate gradient (SCG) algorithm. With the proposed intelligent two-step decision-maker method, HEDGs are connected to the points determined using the PFPE algorithm, and then the appropriate operating conditions for which HEDGs should be enabled are determined by the ANN with SCG. Different combinations of load status, bus bar powers, and line capacities values are applied to the ANN input and important features are determined. The ANN with SCG can predict the operating conditions of HEDGs with 96.8% accuracy in the test set and, 98.4% accuracy in the validation set. Thanks to the developed holistic PFPE&ANN approach, optimum connection points and suitable operating conditions can be determined, which ensures reliability and safety for HEDGs in overload and/or failure conditions.

Multitask Neural Network for Mapping the Glass Transition and Melting Temperature Space of Homo- and Co-Polyhydroxyalkanoates Using σProfiles Molecular Inputs

Polyhydroxyalkanoates (PHAs) are an emerging type of bioplastic that have the potential to replace petroleum-based plastics. They are biosynthetizable, biodegradable, and economically viable and have a range of tunable properties. Despite their great potential, the structure and properties of PHA remain unexplored due to their theoretically infinite chemical space. Therefore, computational approaches for accurate predictions of their various properties need to be developed to effectively explore this large chemical space. For this purpose, this work presents a multitask artificial neural network (ANN) capable of predicting the glass transition temperature (Tg) and melting temperature (Tm) of PHA homopolymers and copolymers. The ANN inputs included the σProfiles as molecular parameters describing the monomer chemistry and their composition. In contrast, the polymer molecular weight (M) and polydispersity index (PDI) were used to describe the polymer state. The results showed that after optimizing the hyperparameters, the selected ANN architecture was remarkable in predicting the Tg and Tm of PHA with R2 values of 0.979 and 0.986 and average absolute relative deviation (AARD) of 0.476% and 0.520%, respectively. The proposed model represents an initiative to promote the development of robust, open source, and user-friendly models capable of predicting the properties of polymers based solely on molecular parameters (σProfiles), thereby saving time and resources for researchers worldwide. The framework described in this work is flexible so that it can be applied to a larger chemical space and incorporate other properties of polymers.

A combined HT and ANN based early broken bar fault diagnosis approach for IFOC fed induction motor drive

In recent years, fault diagnosis in the Induction Motor Drive (IMD) has been a popular and important field in the motor-drive research area. The development of control circuits for induction motors has prompted the attention of both researchers and industrialists. This paper proposes a broken bar fault diagnosis using Hilbert Transform (HT) and Artificial Neural Networks (ANN), with the drive regulated through the Indirect Field Orientation Control (IFOC). The HT obtains the spectrum of stator current, which is utilized to identify the Broken Rotor Bar (BRB) failure. The magnitude and side-band frequency of the drive are extracted using the Fast Fourier Transform (FFT), and these parameters are fed into the ANN inputs. The fault severity is computed by the ratio of mean side-band frequency amplitude to the main frequency amplitude for finding the impact of failure in the drive. ANN is used to diagnose failure with high accuracy. The tested and training results are used to attain the minimum Mean Square Errors (MSEs). The IFOC is involved in this proposed system to ensure high performance under the variable speed drives. The proposed scheme is validated in both MATLAB/Simulink and experimental platforms.

An artificial neural network classification method employing longitudinally monitored immune biomarkers to predict the clinical outcome of critically ill COVID-19 patients.

The severe form of COVID-19 can cause a dysregulated host immune syndrome that might lead patients to death. To understand the underlying immune mechanisms that contribute to COVID-19 disease we have examined 28 different biomarkers in two cohorts of COVID-19 patients, aiming to systematically capture, quantify, and algorithmize how immune signals might be associated to the clinical outcome of COVID-19 patients. The longitudinal concentration of 28 biomarkers of 95 COVID-19 patients was measured. We performed a dimensionality reduction analysis to determine meaningful biomarkers for explaining the data variability. The biomarkers were used as input of artificial neural network, random forest, classification and regression trees, k-nearest neighbors and support vector machines. Two different clinical cohorts were used to grant validity to the findings. We benchmarked the classification capacity of two COVID-19 clinicals studies with different models and found that artificial neural networks was the best classifier. From it, we could employ different sets of biomarkers to predict the clinical outcome of COVID-19 patients. First, all the biomarkers available yielded a satisfactory classification. Next, we assessed the prediction capacity of each protein separated. With a reduced set of biomarkers, our model presented 94% accuracy, 96.6% precision, 91.6% recall, and 95% of specificity upon the testing data. We used the same model to predict 83% and 87% (recovered and deceased) of unseen data, granting validity to the results obtained. In this work, using state-of-the-art computational techniques, we systematically identified an optimal set of biomarkers that are related to a prediction capacity of COVID-19 patients. The screening of such biomarkers might assist in understanding the underlying immune response towards inflammatory diseases.

Defense Strategy against False Data Injection Attacks in Ship DC Microgrids

False Data Injection Attacks (FDIA) on ship Direct Current (DC) microgrids may result in the priority trip of a large load, a black-out, and serious accidents of ship collisions when maneuvering in the port. The key of the prevention of FDIA is the detection of and countermeasures to false data. In this paper, a defense strategy is developed to detect and mitigate against FDIA on ship DC microgrids. First, a DC bus voltage estimator is trained with an Artificial Neural Network (ANN) model. The error between the estimate value and the measure value is compared with a threshold generated from history data to detect the occurrence of FDIA. Combined with the correlation of artificial neural network inputs, bad data are identified and recovered. The method is tested under six cases with different network and physical disturbances in Matlab/Simulink. The results show that the method can identify and mitigate the FDIA effectively; the error of identifying FDIA by ANN is less than 0.5 V. Therefore, the deviation between the actual bus voltage and the reference voltage is less than 0.5 V.

Comparative Analysis of Artificial Neural Networks and Deep Neural Networks for Detection of Dementia

Dementia is a neurocognitive brain disease that emerged as a worldwide health challenge. Machine learning and deep learning have been effectively applied for the detection of dementia using magnetic resonance imaging. In this work, the performance of both machine learning and deep learning frameworks along with artificial neural networks are assessed for detecting dementia and normal subjects using MRI images. The first-order and second-order hand-crafted features are used as input for machine learning and artificial neural networks. And automatic feature extraction is used in the last framework with the pre-trained networks. The outcomes show that the framework using the deep neural networks performs better contrasted with the first two methodologies used in terms of various performance measures.

Estimation of the storage properties of rapeseeds using an artificial neural network

Rapeseed losses during storage can lead to undesirable difficulties in oil and biodiesel production. In this paper, three artificial neural networks were created to anticipate the main quality parameters of thirteen rapeseed varieties - cultivars and hybrids ( Brassica napus L.) - during drying and storage. The varieties, drying temperature, air velocity and drying time were used as inputs to the artificial neural network model to predict the changes in seed weight and moisture during the drying process. The moisture diffusivity and activation energy of the investigated rapeseed varieties were determined under convective drying. For the experiment, an on-site drying system was used at 40, 60 and 80 °C drying air temperature. The effective diffusivity ranged from: 7.947.10 −10 to 1.459.10 −8 m 2 /s (first drying period) and 4.716.10 −10 to 8.611.10 −9 m 2 /s (second drying period). The predicted Arrhenius constant and activation energy ranged from 17.169 to 42.546 kJ/mol (first drying period) and from 31.261 to 50.474 kJ/mol (second drying period). Seed oil content, free fatty acids and thousand seed weight were determined after drying at different temperatures and after 12 months of storage under the three different storage conditions. To predict these parameters after storage time, a multilayer perceptron model with three layers (input, hidden and output) for three artificial neural networks (ANNs) was used for modelling using the implemented drying parameters (such as: variety, drying temperature, air velocity and drying time, along with initial oil and free fatty acid content and storage type) were used. The prediction of the developed model was accurate enough for the prediction of the output parameters. The coefficients of determination ranged from 0.965 to 0.998 when predicting the weight and moisture of the rapeseed during the drying process and the oil and free fatty acid content and thousand grain weights after the 12 months storage period. • 13 rapeseed hybrids quality parameters were anticipated during drying and storage. • prediction of drying and storage parameters was performed using neural network model. • rapeseed varieties, drying temperature, air velocity and drying time were ANN inputs. • optimization of drying parameters was achieved through multi-object optimization.