Input Dataset Research Articles

Multi-load short-term prediction of an integrated energy system based on GAN-LSTM

An integrated energy system’s load consists of diverse forms, including electric load, cooling load, heat load, and more. Due to its strong randomness and volatility, traditional load forecasting methods are not suitable for the integrated energy system. To address the issue of short-term forecasting for multiple loads of the integrated energy system, the proposed paper utilizes generating adversarial networks (GAN). The proposed method analyzes the characteristics of comprehensive energy multi-load and combines meteorological factors with multi-load historical data to form the input dataset for the prediction model. To further improve the accuracy of the short-term forecasting of comprehensive energy multiple loads, the model employs a generator and a discriminator constructed based on the cyclic neural network of long- and short-term memory (LSTM). The example results demonstrate that the proposed method is effective in predicting multiple loads, with significantly better accuracy than traditional methods.

TManual: Assistant for manually measuring length development in structures built by animals.

Structures built by animals are extended phenotypes, and animal behavior can be better understood by recording the temporal development of structure construction. For most subterranean and wood-boring animals, these structures consist of gallery systems, such as burrows made by mice, tunnel foraging by termites, and nest excavation in ants. Measurement of the length development of such structures is often performed manually. However, it is time-consuming and cognitively costly to track length development in nested branching structures, hindering the quantitative determination of temporal development. Here, I introduce TManual, which aids the manual measurement of structure length development using a number of images. TManual provides a user interface to draw gallery structures and take over all other processes handling input datasets (e.g., zero-adjustment, scaling the units, measuring the length, assigning gallery identities, and extracting network structures). Thus, users can focus on the measuring process without interruptions. As examples, I provide the results of the analysis of a dataset of tunnel construction by three termite species after successfully processing 1125 images in ~3 h. The output datasets clearly visualized the interspecific variation in tunneling speed and branching structures. Furthermore, I applied TManual to a complex gallery system by another termite species and extracted network structures. Measuring the lengths of objects from images is an essential task in biological observation. TManual helps users handle many images in a realistic time scale, enabling a comparative analysis across a wide array of species. TManual does not require programming skills and outputs a tidy data frame in CSV format. Therefore, it is suitable for any user who wants to perform image analysis for length measurements.

The Development of a Consensus Machine Learning Model for Hurricane Rapid Intensification Forecasts with Hurricane Weather Research and Forecasting (HWRF) Data

Abstract This study focused on developing a consensus machine learning (CML) model for tropical cyclone (TC) intensity-change forecasting, especially for rapid intensification (RI). This CML model was built upon selected classical machine learning models with the input data extracted from a high-resolution hurricane model, the Hurricane Weather Research and Forecasting (HWRF) system. The input data contained 21 or 34 RI-related predictors extracted from the 2018 version of HWRF (H218). This study found that TC inner-core predictors can be critical for improving RI predictions, especially the inner-core relative humidity. Moreover, this study emphasized the importance of performing resampling on an imbalanced input dataset. Edited nearest-neighbor and synthetic minority oversampling techniques improved the probability of detection (POD) by ∼10% for the RI class. This study also showed that the CML model has satisfactory performance on RI predictions compared to the operational models. CML reached 56% POD and 46% false alarm ratio (FAR), while the operational models had only 10%–30% POD but 50%–60% FAR. The CML performance on the non-RI classes was comparable to the operational models. The results indicated that, with proper and sufficient training data and RI-related predictors, CML has the potential to provide reliable probabilistic RI forecasts during a hurricane season.

Explainable Artificial Intelligence to Investigate the Contribution of Design Variables to the Static Characteristics of Bistable Composite Laminates.

Material properties, geometrical dimensions, and environmental conditions can greatly influence the characteristics of bistable composite laminates. In the current work, to understand how each input feature contributes to the curvatures of the stable equilibrium shapes of bistable laminates and the snap-through force to change these configurations, the correlation between these inputs and outputs is studied using a novel explainable artificial intelligence (XAI) approach called SHapley Additive exPlanations (SHAP). SHAP is employed to explain the contribution and importance of the features influencing the curvatures and the snap-through force since XAI models change the data into a form that is more convenient for users to understand and interpret. The principle of minimum energy and the Rayleigh-Ritz method is applied to obtain the responses of the bistable laminates used as the input datasets in SHAP. SHAP effectively evaluates the importance of the input variables to the parameters. The results show that the transverse thermal expansion coefficient and moisture variation have the most impact on the model's output for the transverse curvatures and snap-through force. The eXtreme Gradient Boosting (XGBoost) and Finite Element (FM) methods are also employed to identify the feature importance and validate the theoretical approach, respectively.

Identifying Prominent Environmental Covariates Using Variable Selection Methodologies for Digital Soil Mapping of Tamil Nadu, India

High dimensional datasets that depict intricate spatial variations are necessary to predict complex landscape structures and the corresponding soil properties taking into account the size of the research region in addition to the data attributes. The number and quality of the input datasets taken into consideration essentially determine the quantity and quality of the soil properties that may be predicted thanks to data-driven learning algorithms. The use of variable selection strategies both before and after the prediction can have a significant impact on the outcome and can lower the related computing load. The majority of commonly used variable selection techniques such as correlation analysis, stepwise regression and recursive feature elimination, among others perform recursive statistical/mathematical comparison to identify the significant covariates that improve the effectiveness of the algorithm proposed. In order to identify the effective environmental variables in predicting the soil attribute, this article investigated a widely used recursive ranking method called recursive feature elimination. The covariate layer that produced the lowest RMSE was placed first according to the rankings of the covariates provided by recursive feature elimination. The findings showed that among other factors physiography, mean rainfall, rock outcrop difference ratio, elevation and mean temperature will be effective in predicting the soil properties required for digital soil mapping.

Reconstruction of Missing Well-Logs Using Facies-Informed Discrete Wavelet Transform and Time Series Regression

Summary Geophysical logging is widely used in lithofacies identification, reservoir parameter prediction, and geological modeling. However, it is common to have well-log sections with low-quality and/or missing segments. Repeating the well-log measurements is not only expensive but might also be impossible depending on the condition of the borehole walls. In these situations, reliable and accurate well-log prediction is, therefore, necessary in different stages of the geomodeling workflow. In this study, we propose a time series regression model to predict missing well-log data, incorporating facies information as an additional geological input and using discrete wavelet transform (DWT) to denoise the input data set. The main contributions of this work are threefold: (i) We jointly use facies information with well logs as the input data set; (ii) we use DWT to denoise the input data and consequently improve the signal-to-noise ratio of the input data; and (iii) we regard the depth domain as the time domain and use a time series regression algorithm for log reconstruction modeling. We show a real application example in two distinct scenarios. In the first, we predict missing well-log intervals. In the second, we predict complete well logs. The experimental results show the ability of the proposed prediction model to recover missing well-log data with high accuracy levels.

Solution Space Management to Enable Data Farming in Strategic Network Design

During strategic network design, not only strategic but also operational decisions must be made long before a production network is put into operation. These include determining the location and size of inventories within the production network and setting operational parameters for production lines, such as the shift model. However, the large solution space comprising a high number of highly uncertain design parameters makes these decisions challenging without decision support. Therefore, data farming offers a potential solution, as synthetic data can be generated via the execution of multiple simulation experiments spanning the solution space and then analyzed using data mining techniques to provide data-based decision support. However, data farming has not yet been applied to strategic network design due to the lack of adequate solution space management. To address this shortcoming, this paper presents a structured solution space management approach that integrates production network-specific requirements and Design of Experiment (DoE) methods. The approach enables converting the solution space in strategic network design into individual input data sets for simulation experiments, generating a comprehensive database that can be mined for data-based decision support. The applicability and validity of the comprehensive approach are ensured via a case study in the automotive industry.

Studies on crystallization process for pharmaceutical compounds using ANN modeling and model based control

Solvent selection and Controlling of operating parameters play a crucial role in batch cooling crystallization process. Choosing a best solvent for crystallization process involves more experimentation and time. To overcome this problem, an Artificial Neural Network (ANN) model technique is used to predict the carbamazepine form Ⅲ solubility by considering the thermodynamic properties of different solvents i.e. critical temperature, critical pressure, temperature, molecular weight, and acentric factor. The ANN model was trained and evaluated for solubility at various input data sets using experimental solubility data available in the literature. The ANN model with 20 hidden neurons has given the R2 value of 0.9943 which shows that the developed ANN model can be used for the selection of best solvent for batch crystallization process. Further, to determine the optimal cooling profile of batch cooling crystallization process, a multi-objective optimization problem is formulated by considering objectives as minimizing the coefficient of variation (CV) and maximizing the Number mean size (NMS) of crystals subjected to population balance equations using “method of moments” technique. Two types of temperature strategies i.e., piece-wise constant and piece-wise linear are developed and solved using NSGA-Ⅱ dynamic optimization procedure. The optimal NMS value attained through piece-wise linear strategy was 197.1 µm. This value has been increased by 28.3 µm from the nominal case (without optimization) and the coefficient of variation has decreased from 0.951 to 0.76. Further, optimal NMS value attained through piece-wise constant strategy was 205 µm. The value has been increased by 36.2 µm and the coefficient of variation has decreased from 0.951 to 0.73. This proves that the crystal attributes can be improved by optimal cooling temperature profile obtained by multi-objective optimization framework. For implementing the optimal cooling profile an advanced model-based control, i.e., Generic Model Control (GMC) was developed. It was observed that the GMC controller has the good tracking profile with no offset with/without disturbances and small value of root mean square error (RMSE) of 0.0016 using piece-wise constant as set point temperature. Using piece-wise linear as set point temperature, the RMSE value was 0.0018. In particular, it is advantageous to operate the batch cooling crystallization process with piece-wise linear strategy for set point trajectory tracking problems.

Energy processes prediction by a convolutional radial basis function network

If an approach based on the gradient steepest descent is utilized to adapt the parameters of a radial basis function network, then it requires dimensionality reduction of the input dataset for the complexity reduction and efficiency improvement, resulting in a more precise energy processes prediction. The convolution operation could provide one way to perform dimensionality reduction of the input dataset. In this research, the convolutional radial basis function network is utilized for the energy processes prediction. The advances are exposed as follows: (1) the convolutional radial basis function network containing a convolution part, a hidden part, and an output part is utilized for the energy processes prediction, (2) the convolution operation is utilized in the convolution part to perform dimensionality reduction of the input dataset, and to change the magnitude of the input dataset for the complexity reduction, (3) the gradient steepest descent is utilized to adapt the parameters in the hidden part and output part for the efficiency improvement. The convolutional radial basis function network is compared against the radial basis function network, the feedforward neural network, and the neuro fuzzy system for the hourly electrical power demand prediction and for the chiller prediction.

Deep Learning-Assisted Citation Recommendation System Using Multi- cell RNN Approach

A process of retrieving relevant or suitable papers that are mostly related to the research topic while carrying research knowing the behavior and patterns of citation for mapping the disciplines and subjects. This is used increasingly for judging usually takes a huge amount of effort and time. The citation recommendation is used frequently for solving this issue on suggesting the list of candidate papers automatically which may match the references of user. On applying Deep learning (DL) models in the citation recommendation system, the performance enhancement is greatly visible in various recent research studies. Hence, in this work, a deep learning-based model is employed for the citation recommendation system. Primarily, the input dataset is taken and the function words are removed using stochastic matching pattern scheme at which the library is created. Then, the canonicalization of document is carried by means of probabilistic min-match algorithm. The data is classified finally using Multi-cell RNN (Recurrent neural network) approach after which the rank is set for citation. At last, the performance is assessed for various metrics like MAP, MRR, precision, recall, and F1-measure. The obtained results are then compared with traditional models to validate and show the enhancement of proposed model over existing schemes. It is revealed from the analysis that the proposed scheme shows improvement in results.

An optimized Extreme Learning Machine by Evolutionary Computation for River Flow Prediction and Simulation of Water Pollution in Colorado River Basin, USA

Humanity is witnessing scientific advances and since one of the vital concerns has been accessing water for drinking and agriculture, scientific innovations have been used to solve water challenges. Recently, Machine Learning (ML) is one of the most promising developments, therefore, for utilization of this innovation in simulating the Water Quality-Quantity Assessment (WQA) issues, the author has developed the Extreme Learning Machine (ELM) as the stand-alone model and its combination with evolutionary algorithms (EA) to optimize the modeling of the WQA parameters. So, the Particle Swarm Optimization (PSO), Genetic Algorithm (GA), and Imperialist Competitive Algorithm (ICA) are implemented to improve the performance of WQA parameters prediction. Then, the three novel models are developed for this purpose simultaneously with the stand-alone ML model. The novel models are GAELM, PSELM, and ICELM. The study area was the Colorado River basin in the USA, which was the input dataset extracted from the US Geological Survey. The considered parameters were Power of hydrogen (pH), river flow or Debi, Electrical Conductivity (EC), and Dissolved Oxygen (DO). To over- and under-fitting ML models, the input dataset was detrended and randomized by K-fold cross-validation. Furthermore, this study used seven evaluation methods for determining the models' performances. Based on the evaluation metrics, the ICELM model was the best in the EC, DO, and pH modeling. Also, the ELM model was superior in Debi prediction. Additionally, the discrepancy percent charts for all models in simulations were drawn. Moreover, the error percent plots also were drawn for all modeling. Finally, the Wilson method analyzed the prediction errors and related uncertainties.

Optimal feature selection and detection of sickle cell anemia detection using enhanced whale optimization with clustering based boosted C5.0 algorithm for tribes of Nilgiris

Degradation of red blood cells (RBC) causes many diseases, like sickle cell anemia. Diagnosing this disease takes more time because peripheral blood samples must be examined under the microscope. Since the isolated RBC observation is subjective and high error rate leads to reduce the accuracy, the technology is needed to perform this approach. To fulfil these requirements, optimal feature selection and detection of sickle cell anemia using enhanced whale optimization with clustering based boosted C5.0 algorithm in tribes of Nilgiris is proposed in this manuscript. The input dataset is taken from real data set via non-government organization named NAWA (situated in Kotagiri). The reason behind of Nilgiri region is considered in this work is a collection of sickle cell anemia test results of the tribe people who lives in different areas of Nilgiri region. These images are pre-processed using wavelet packet transform cochlear filter bank method to eradicate the noises and to improve the superiority of image. After that, the features are extracted using force-invariant improved feature extraction method. To select the optimal features, enhanced whale optimization (EWO) algorithm is used. These optimal features are classified utilizing clustering based boosted C5.0 algorithm. The proposed method is activated on PYTHON. The proposed method shows 52.32%, 43.78%, 32.78% and 45.90% higher accuracy compared with the existing methods, such as RGSA-MLP-SCA, CRFA-SVM-SCA, AO-LSTM-SCA and BOA-CNN-SCA.

Study on Software Defect Prediction Based on SVM and Decision Tree Algorithm

Software Defect Prediction is a process of identifying the potential defects in software systems before they occur. In this approach, the dataset containing information about software attributes is used as input, and the output is the prediction of whether the software is defective or not. The input dataset is generally in the form of a CSV file, which contains various software attributes such as cyclomatic complexity, essential complexity, Design Complexity, etc. The output of the defect prediction is binary classification. It is done by using SVM (Support Vector Machine) and a decision tree algorithm. This approach can help software developers identify their systems' defects before they cause any harm or affect the system’s performance.

Heat Transfer Optimization Studies of Semi-Spherical Protrusions on a Concave Surface during Jet Impingement

Much of the literature available on heat transfer enhancement from rough surfaces concerns flat plates or channels. Studies on protruding concave surfaces are surprisingly rare though they have the potential to enhance heat transfer. Therefore, three-dimensional steady-state simulations and corresponding experimental validation is carried out for an air jet impinging on a concave surface projected with semi-spherical protrusions. By providing multiple protrusions, various geometrical configurations are studied by considering the dimensions, angular positions, and pitch between the protrusions. Each individual configuration has its own unique flow pattern which makes prediction of heat transfer complex. This motivates present study to consider entire geometrical parameters affecting heat transfer characteristics with the help of networking tool. Past investigations on protruded flat surfaces concerns random allocation of protrusions. Hence artificial neural network is employed in this study to easily predict the corresponding heat transfer value with set of few input data. Simulation results give detailed flow physics which helps in understanding the heat transfer behavior. It is observed that each protrusion is marked with high-pressure region around its uphill and negative pressure around the downhill. As the flow moves downstream, the protrusions obstruct the flow path and flow separates away as soon as it reaches center of the protrusion. A maximum heat transfer enhancement of 11% is obtained over a corresponding smooth surface.

Detect and Evaluate Visual Pollution on Street Imagery Taken of a Moving Vehicle

Visual pollution is a growing problem in urban areas. It is important for environmental management to identify, formalize, measure and evaluate visual pollution. This paper presents a study on the development of an automated system for visual pollution classification using street images taken from a moving vehicle. The proposed system uses convolutional neural networks to classify different types of visual pollutants such as graffiti, faded signage, potholes, litter, construction zones, broken signage, poor street lighting, poor billboards, road sand, sidewalk clutter, and unmaintained facades.In this study, we utilized a large dataset of raw sensor camera inputs gathered from a fleet of multiple vehicles in a specific geographical area. Our aim was to develop convolutional neural networks that simulate human learning to classify visual pollutants from these images. The successful implementation of this system would be a significant contribution to the development of urban planning and the strengthening of communities worldwide. Additionally, it could lead to the creation of a "visual pollution score/index" for urban areas, which could serve as a new metric for urban environmental management. Our findings, which we present in this paper, will be a valuable addition to the academic community and the field of computer vision for environmental management applications.

Leveraging augmentation techniques for tasks with unbalancedness within the financial domain: a two-level ensemble approach

Modern financial markets produce massive datasets that need to be analysed using new modelling techniques like those from (deep) Machine Learning and Artificial Intelligence. The common goal of these techniques is to forecast the behaviour of the market, which can be translated into various classification tasks, such as, for instance, predicting the likelihood of companies’ bankruptcy or in fraud detection systems. However, it is often the case that real-world financial data are unbalanced, meaning that the classes’ distribution is not equally represented in such datasets. This gives the main issue since any Machine Learning model is trained according to the majority class mainly, leading to inaccurate predictions. In this paper, we explore different data augmentation techniques to deal with very unbalanced financial data. We consider a number of publicly available datasets, then apply state-of-the-art augmentation strategies to them, and finally evaluate the results for several Machine Learning models trained on the sampled data. The performance of the various approaches is evaluated according to their accuracy, micro, and macro F1 score, and finally by analyzing the precision and recall over the minority class. We show that a consistent and accurate improvement is achieved when data augmentation is employed. The obtained classification results look promising and indicate the efficiency of augmentation strategies on financial tasks. On the basis of these results, we present an approach focused on classification tasks within the financial domain that takes a dataset as input, identifies what kind of augmentation technique to use, and then applies an ensemble of all the augmentation techniques of the identified type to the input dataset along with an ensemble of different methods to tackle the underlying classification.

Analysis of forest tree dieback using UltraCam and UAV imagery

ABSTRACT In recent years, increasing tree diebacks and mortality in some forests, particularly in forest parks, created a need amongst forest managers to find effective methods to gather information about the rate of dieback and mortality and their reasons. High-quality air and space-born remote sensing data has established as an alternative to field surveys for certain inventory tasks. This study used high-quality UltraCam-Xp and UAV drone images from 2016 and 2021 to map tree dieback and mortality in Daland Forest Park, Golestan Province, Iran. High-quality ortho mosaics and Digital Surface Models (DSMs) were generated from UltraCam (2016) and UAV (2021) images. The images were then classified through object-based classification by Nearest Neighbor (NN), Support Vector Machine (SVM), and Bayes algorithms using various input data sets including spectral bands, Canopy Height Model (CHM), vegetation indices, and texture analysis features. Our results indicate that the Bayes algorithm is more precise in mapping tree dieback for the two time steps compared to other algorithms. The best tree dieback map on UltraCam images was obtained using the spectral bands with CHM, texture analysis features, and vegetation indices. This combination resulted in an overall accuracy of 91.20% and a Kappa coefficient of 0.88. It was also found that combining the UAV main bands with CHM and texture features did produce a high-accuracy map with an overall accuracy of 88.46% and a Kappa coefficient of 0.84. Change detection analysis of tree dieback showed that between 2016 and 2021, the number of healthy trees decreased, and the number of gaps and open areas increased in the study area. We conclude that UltraCam and UAV photographs can serve to identify and map tree dieback and dead trees with good accuracies and can hence support forest health monitoring.

Modeling of annual maximum flows with geographic data components and artificial neural networks

The flow rate at which the instantaneous maximum flow is recorded throughout the year is called the Annual Maximum Flow (AMF). These flow rates often cause disasters such as floods. Snow melts and extreme precipitation associated with temperature fluctuations are the two most important factors that occurred flooding. The deluge that follows kills people and destroys property in communities and agricultural lands. As a result, it's critical to predict the flow that causes flooding and take appropriate precautions to limit the damage. The prediction of the probability of a flood event in advance is very important for the safety of life and property of large masses and agricultural lands. Early warning systems, disaster management plans and minimizing these losses are among the important goals of the country's administration. This study was used in five Current Observation Stations (COS) located in Yeşilırmak Basin in Turkey. By using 8 input data including geographical location, altitude and area information of these stations, AMF data were tried to be estimated for each COS. A total of 240 input data was used in the study. The data period covers the years 1964-2012. Unfortunately, AMF values cannot be monitored for all 5 stations used after 2012. Therefore, the data period was stopped in 2012. In this study, Multilayer Artificial Neural Networks (MANN), Generalized Artificial Neural Networks (GANN), Radial Based Artificial Neural Networks (RBANN) and Multiple Linear Regulation (MLR) methods were used. Input data sets were made into 4 packets and these packages were used respectively in both training and testing stages. In these packages, the AMF data measured for the 5 stations mentioned above between 1965 and 2012 were divided into 4 and used by creating 25% (test) and 75% (training) packages. Root Means Square Error (RMSE), Mean Absolute Error (MAE) and correlation coefficient (R) were used as the comparison criteria. The results are as follow; MANN (RMSE = 119.118, MAE = 93.213, R = 0.808), and RBANN (RMSE = 111.559, MAE = 81.114, R = 0.900). These results show that AMF can be predicted with artificial intelligence techniques and can be used as an alternative method.

Research and Application of Semi-Supervised Category Dictionary Model Based on Transfer Learning

Classification tasks are pivotal across diverse applications, yet the burgeoning amount of data, coupled with complicating factors such as noise, exacerbates the challenge of classifying complex data. For algorithms that require a large amount of data, the annotation work for datasets is also exceptionally complex and tedious. Drawing upon existing research, this paper first introduces a novel semi-supervised category dictionary model based on transfer learning (SSDT). This model is designed to construct a more representative category dictionary and to delineate the associations of information across different domains, utilizing the lens of conditional probability distribution. This approach is particularly apt for semi-supervised transfer learning scenarios. Subsequently, the proposed method is applied to the domain of bearing fault diagnosis. This model is suitable for transfer scenarios; moreover, its semi-supervised characteristic eliminates the need for labeling the entire input dataset, significantly reducing manual workload. Experimental results attest to the model’s practical utility. When benchmarked against other 6 models, the SSDT model demonstrates enhanced generalization performance.

Optimization of painting efficiency applying unique techniques of high-voltage conductors and nitrotherm spray: Developing deep learning models using computational fluid dynamics dataset

The impetus of the current three-dimensional Eulerian–Lagrangian work is to analyze the impact of simultaneously using the inventive high-voltage conductors and Nitrotherm spraying technique for maximizing the industrial painting process efficiency. This investigation employs high-fidelity computational fluid dynamics (CFD) results in deep learning models as an input dataset. The novel conductors are called high-voltage retractable blades (HVRB) and high-voltage adjustable control-ring (HVACR) mounted on the head of the electrostatic rotating bell sprayer. The influence of dominant operational parameters, such as temperature and velocity of injected nitrogen or air, droplets' electric charge values, and their size ranges, and electric field density are examined in the considered database for the Nitrotherm spraying methodology. This broad range of parametric investigation illustrates that the inclusion of shaping nitrogen flow, manipulated electric field density, and droplet charging weights significantly affect the spraying deposition rate. The pressurized clean heated nitrogen flow, which is injected from the nozzles of the atomizers, positively redirects and harmonizes the charged droplets that construct an optimized spray plume pattern with a smaller diameter. Using innovative HVRB and HVACR conductors is manipulated the electric fields and leads to denser distribution, intensifying the acting electric force on the droplets, resulting in higher spraying transfer efficiency (TE) and thicker film formation. Based on the results, employing the introduced conductors in combination with the heated nitrogen instead of air leads to higher TE, rare overspray occurrence, formation of an esthetic paint film, lower paint consumption, and application time. Also, the collected complete database is employed for machine learning investigation to predict flow with high accuracy, aiming to reduce computational time/cost. A convolutional auto-encoder is used to reduce the computational cost with just 10% of the initial CFD computations, with a mean error of 1% on the prediction of the deposited droplet areas of the spray. The analysis revealed that by employing recurrent convolutional layers, superior capturing of the input pattern is obtained, which significantly aids the final prediction.