03.Forum of Artificial Intelligence and Its Applications

Permanent URI for this collectionhttps://dspace.univ-eloued.dz/handle/123456789/10731

Browse

Now showing 1 - 20 of 79

Applying Artificial intelligence techniques for predicting amount of CO2 emissions from calcined cement raw materials
(University of Eloued جامعة الوادي, 2022-01-24)
This paper aims to predict the amount of carbon dioxide CO2 emissions from raw material used in cement clinker production during calcination process. The amount of CO2 emissions is mainly from the decarbonisation thermal process that is directly related to chemical composition, distribution of particle size and time exposed at high temperature. These influencing factors interact with each other, making the calculation of the amount of CO2 emissions with conventional techniques more difficult. For this reason, several artificial intelligence techniques are applied to predict the amount of CO2 emissions. The key advantage of the proposed techniques is its ability to learn and to generalise without any prior knowledge of an explicit relationship between target and its influencing parameters. The intelligence techniques applied are deep neural network (DNN), artificial neural networks (ANN) optimised using ant colony optimization (ACO-ANN) and genetic algorithm (GA-ANN). The results obtained are promising and show that all intelligence techniques can provide excellent accuracy with high R2 and low error. DNN predicts the amount of CO2 emissions very accurately when comparing to other techniques. Overall, the performance accuracy of ACO-ANN technique is higher than the GA-ANN. According to R2 values, there are more than 99%, 98.5% and 98% of experimental data in testing phases can be explained by DNN, ACO-ANN and GA-ANN respectively with average relative error less than 1.04%. As conclusion, all intelligence techniques can be employed as an excellent alternative to predict the amount of CO2 emissions.
Artificial intelligent in upstream oil and gas industry: a review of applications, challenges and perspectives
(University of Eloued جامعة الوادي, 2022-01-24) Kenioua, Abdelhamid; Touati Brahim, Ammar; Kenioua, Laid
In the last two decades, oil and gas (O&G) industries are facing several challenges and issues in different levels; from the decrease in commodity prices to the dynamic and unexpected environment. There has been a constant urge to maximize benefits and attain values from limited resources. Traditional empirical and numerical simulation techniques have failed to provide comprehensive optimized solutions in little time due to the Immense amount of data generated on daily basis with various formats, techniques and process. The proper technical analysis of this “explosion of data” is to be carried out to improve performance of O&G industries. Artificial intelligence (AI) has found extensive usage in simplifying complex decision-making procedures in practically every competitive market field, and O&G industry is not an exception. This paper provides a comprehensive stateof- art review in the field of machine learning and artificial intelligence to solve O&G industry problems. We focus on the upstream segment as the most capital- intensive part of oil and gas and the segment of enormous uncertainties to tackle. Based on a summary of various researchers work on machine learning and AI applications, we outline the most recent trends in developing AI-based tools and identify their effects on accelerating the process in the industry. This paper discusses also the main challenges related to non-technical factors that prevent the intensive application of AI in the upstream O&G industry.
Assessing Combinations of Artificial Neural Networks Input Meteorological Parameters to Improve Daily Runoff Simulation
(University of Eloued جامعة الوادي, 2022-01-24) Aoulmi, Yamina; Marouf, Nadir; Amireche, Mohamed
Hydrological models are one of the tools used for reconstruction or simulation, forecasting for the anticipation of future changes in the flow of a river, which allows better management of water resources during low flow periods and anticipation of flood risks during high water periods. Using artificial neural network (ANN) in the fields of hydrology and water resources, became advocated due to its capability of tackling, modeling and forecasting the problems that are nonlinear or stochastic within the Rainfall-Runoff (R-R) system. This research aims to test the practicability of using ANNs with two input configurations; to model the R-R relationship in two stations Mirebek and Ain Berda located in the Seybouse basin in Algeria. The 1st (ANN IP) considers only precipitation as an input variable for the daily flow simulation. The 2nd (ANN IIP,T,Hu) considers a combination of the Temperature and Humidity with precipitation at the model input. The results of the two models were compared through performance metrics, viz., Root Mean Square Error (RMSE),mean absolute error (MAE), Pearson's correlation coefficient (R), Nash Sutcliffe Efficiency coefficient (NSE), and through graphical interpretation (scatter plots and time series). Better flow simulations were provided by the three-input model for the two stations; where R=0.90, NSE=80.5% for Mirebek station and R=0.85, NSE=72% for Ain Berda. This result has confirmed that as much input variables are numerous, as more the model of ANN is efficient. The finding of this study indicates that the developed ANN models could be considered as a powerful tool for predicting Runoff.
Automation of the reading and interpretation of the OTDR plot
(University of Eloued جامعة الوادي, 2022-01-24) MAZOUZI, Amine; ABOUN ABID, Miloud
Unlike conventional networks, optical fiber transmission systems allow optical signals to be transported from the transmitter to the receiver. They have become the adaptable solution for transmissions of all types of data. By taking into account the problems and the external constraints which degrade the quality of the transmitted signal, the monitoring of the optical fibers is carried out by measurement tools making it possible to detect the various events and the defects along the fiber and this in order to do the maintenance. Necessary to provide better transmission on the medium. Currently the operation of monitoring an optical transmission link is carried out by an agent who constantly monitors the various changes in events that have occurred in the layout of the transmission line using OTDR equipments.
Ball bearing monitoring using decision-tree and adaptive neuro-fuzzy inference system
(University of Eloued جامعة الوادي, 2022-01-24) Euldji, Riadh; Boumahdi, Mouloud; Bachene, Mourad; Euldji, Rafik
This study aims to provide a methodology that relies on the combination of the following approaches: the decision tree, the neural network, and the fuzzy logic to monitor the evolution of bearing degradation. Data collected from the vibratory signals generated from the tests carried out on ball bearings mounted in an experimental fatigue platform, are used. The decision tree method is applied to select the most relevant monitoring indicator, which will be used to develop an Adaptive Neuro-Fuzzy Inference System (ANFIS). The training and test data required for model development have been classified according to the following states: normal, abnormal, and dangerous. These were defined from two thresholds: alert threshold and danger threshold. Then, the ANFIS model is trained from the indicators selected by the decision tree to predict the behaviour of the bearing in operation. The results confirm the effectiveness of the proposed approach for monitoring the health of ball bearing
Big Data Veracity: Methods and challenges
(University of Eloued جامعة الوادي, 2022-01-24) benabderrahmane, MOUTASSEM; laouni, DJAFRI; GAAFOUR, Abdel-Kader
Today, the Internet has become the main source of information, a place where there are no restrictions on who can share information .This latter can play an important role in prediction, estimation and decision making processes. But, this role will not only be achieved through abundance, it will also be the result of data quality. Veracity refers to the assurance of quality or credibility of the data collected. The data can be incomplete, biased, vague or wrong. For this reason, automatic filtering mechanism has been developed. Moreover, due to the increasing velocity of information spread, manual assessment of information veracity became hard, a time consuming process and even the already existing automatic filtering mechanisms has to be improved to cope with the speed of information spread. In this paper, a literature review is established to highlight the recent methods and techniques which are exploited in computerized veracity assessment. The challenges and limitations of existing works will be discussed, and future research directions will be proposed to address critical issues of data veracity in the era of big data.
A CBR approach based on ontology to supplier selection
(University of Eloued جامعة الوادي, 2022-01-24) BEKKAOUI, Mokhtaria; KARRAY, Mohamed Hedi; MELIANI, Sidi Mohammed
Most previous research categorizes supply chain management (SCM) into the following three major parts: purchasing, manufacturing, and distribution. In the purchasing process, supplier selection represents a crucial step for enhancing firms’ competitiveness.An increasing number of researches have been devoted to the development of different methodologies to cope with this problem.The majority of research explores the difference in the set of criteria for supplier selection. They used famous methods for selection such as AHP (Analytic Hierarchy Process); ANP (Analytic Network Process); DEA (Data Envelopment Analysis);TOPSIS (Technique for Order Preference by Similarity to Ideal Solution). Thus, this paper presents a CBR (Case Based Reasoning) approach in the context of supplier selection decision making.
Clustering Educational Items from Response Data using Penalized Pearson coefficient and deep autoencoders
(University of Eloued جامعة الوادي, 2022-01-24) Harbouche, Khadidja; Smaani, Nassima; Zenbout, Imene
Educational data mining techniques are very useful to analyze learner performance in purpose to optimize the approach of item-to-skill mapping. Therefore computing a degree of similarity between items using different measures based on the performance of the learner toward items, enhance the clustering of different items into knowledge components. This paper proposes a computational framework to group the elements of the corresponding knowledge component. The first phase of the framework represents a variation of Pearson coefficient to measure item similarity by applying a penalty score that is calculated from the number of hints taken by the learner during solving two items. The second phase applies a dimensionality reduction using deep auto encoders to improve the clustering accuracy. The experimental results show that clustering based on the penalized Pearson coefficient and the deep dimensionality reduction (PPC+DDR) outperforms basic clustering based on different similarity methods , with approximately +0.2 in Mean silhouette coefficient.
A CNN approach for the identification of dorsal veins of the hand
(University of Eloued جامعة الوادي, 2022-01-24) Benaouda, Abdelkarim; Aymen, Haouari Mustapha; Benziane, Sarâh
In this paper, we proposed a dorsal hand vein recognition method based on Convolutional Neural Network (CNN). Firstly, implementations of raw images the region of interest (ROI) of dorsal hand vein images was extracted, and then contrast limited adaptive histogram equalization (CLAHE) and were used to preprocess the images. Next, the extraction of information using the Sato filter and the Otsu thresholding algorithm to create a new database containing only the processed images. Finally, CNN was applied for identification. The experimental results was has been optimized with Hyperparameter Optimization. The dorsal hand vein recognition rate reaches 99%.
A Comparative Study between the Two Applications of the Neural Network and Space Vector PWM for Direct Torque Control of a DSIM Fed by Multi-Level Inverters
(University of Eloued جامعة الوادي, 2022-01-24) Benaouda, O. F.; Mezaache, M.; Abdelkader, R.
Nowadays, thanks to the development of control and power electronics, the dual stator induction machine DSIM has become among the most important multi-phase machines included in industrial application, this is due to its positive features among them is its high reliability and reduce both losses and rotor torque ripple. This paper aims to apply both techniques of artificial intelligence represented by the neural network algorithm NNA and the Space Vector PWM SVM for direct torque control DTC of the DSIM to improve the machine performance and control algorithms DTNC and DTC-SVM. Generalization capacity, the parallelism of operation, computational speed, and learning capacity all these features made it possible to exploit the neural network algorithm to control the machine. Fixed switching frequency obtained, dispensed with the vector selection table and the hysteresis controller, the three pros allowed the inclusion of SVM technique in DTC strategy. Three-level NPC inverters are included to feed the DSIM. A several of the results obtained prove the two applied techniques (NNA, SVPWM) in improving the quality of both electromagnetic torque and flux and the dynamic responses of the DSIM.
A Comparative Study Of Road Traffic Forecasting Models
(University of Eloued جامعة الوادي, 2022-01-24) Benabdallah Benarmas, Redouane; Beghdad, Kadda
In the context of Intelligent Transport Systems (ITS), the behaviour of road traffic has been the subject of many theoretical and experimental researches. In the last decade, road prediction is placed as the first line of research in this field. The problem has been solved with a variety of models to assist the traffic control, this includes, improving the efficiency of transport, guidance in the road, and smart coordination signals. This paper tries to synthesize the carried out, on three main approaches, namely based on statistical methods, time series and deep learning.A comparatives synthesis in terms quantitative and qualitative index of is provided in order to evaluate the performance and potential of the three forecasting approaches.
Comparing NoSQL Databases with YCSB Standard Benchmark
(University of Eloued جامعة الوادي, 2022-01-24) Ben Seghier, Nadia; Kazar, Okba
NoSQL data-stores are commonly used to provide flexibility and availability for Big Data handling. The important companies in the IT sector find these NoSQL systems, new solutions to respond to scalability needs. These databases can be broadly classified as key value stores, column based, document stores and graph database depending on their mechanism of data storage and other features. NoSQL databases assert that their performance is better than legacy Relational Database systems for higher workloads, particularly common in Big Data and Cloud Computing applications. Multiple open-source and proprietary models of NoSQL are available on the market. Because of the large number and diversity of existing solutions, it is difficult to select an appropriate solution for a specific problem. In this paper, we develop a comparative study about the performance of three solutions widely employed: Redis 3.0.504, MongoDB 4.4.0 and Cassandra 3.11.1, and tests the runtime for different proportions of read, update, scan, readmodify- write and insert operations using six workloads by YCSB 0.17.0 tool on Windows OS. The purpose of our comparative study is to provide assistance and support to actors interested of Big Data and Cloud Computing for eventual decisions for the choice of solutions to be adopted.
A Comprehensive Study of multicast Routing Protocols in the Internet of Things
(University of Eloued جامعة الوادي, 2022-01-24) Lakhlef, Issam Eddine; Djamaa, Badis; Senouci, Mustapha Reda
IP multicast is a desired communication feature in the Internet of Things (IoT) as it provides noticeable resource savings, especially for Low-power and Lossy Networks (LLNs). Indeed, multicast allows cost-, energy-, and time-efficient networking for a multitude of LLN applications ranging from over-the-air programming and information sharing to device configuration and resource discovery. In this context, several multicast routing protocols have been recently proposed for LLNs including Stateless multicast RPL Forwarding (SMRF), Enhanced SMRF (ESMRF), Bi-Directional multicast Forwarding Algorithm (BMFA), and multicast Protocol for LLNs (MPL). Nevertheless, each protocol has been evaluated under different conditions, topologies, and traffic flow, which prevents making comprehensive comparisons of their characteristics and performance. In this paper, we provide an overview of recent LLN multicast protocols followed by a multidimensional performance evaluation of the most popular ones to extract their advantages and drawbacks under different traffic conditions, routing scenarios, and network topologies. Obtained results from extensive realistic simulations using Cooja show that, although each protocol is dominant under specific conditions, MPL remains the best in terms of packet delivery ratio in all scenarios at the expense of extra energy consumption, which requires new resourcesaware multicast solutions for the IoT.
Data-Intensive Scientific Workflow Scheduling Based on Genetic Algorithm in Cloud Computing
(University of Eloued جامعة الوادي, 2022-01-24) Kouidri, Siham; Kouidri, Chaimaa
Cloud Computing is increasingly recognized as a new way to use ondemand, computing, storage and network services in a transparent and efficient way. Cloud Computing environment consists of large customers requesting for cloud resources. Nowadays, task scheduling problem and data placement are the current research topic in cloud computing. In this work, a new technique for task scheduling and data placement are proposed based on genetic algorithm to fulfill a final goal such as minimizing total workflow response time. the scheduling of scientific workflows is considered to be an NP-complete problem, i.e. a problem not solvable within polynomial time with current resources The performance of this proposed algorithm has been evaluated using CloudSim toolkit, Simulation results show the effectiveness of the proposed algorithm in comparison with well-known algorithms such as genetic algorithm with Random data placement.
Deep approach based on user’s profile analysis for capturing user’s interests
(University of Eloued جامعة الوادي, 2022-01-24) Besnkhelifa, Randa; Bouhyaoui, Nasria
Capturing user’s interests and preferences by analyzing and interpreting the daily-shared contents in online social networks offer a unique information source for several domains such as business, marketing and politics. User’s profile describes its owner’s characteristics, where it contains several important personal information such as (age, sex, job title, level of education, etc.), which can help to improve the process of user’s interests identification. This information can typically represent a range of values representing only one user profile. Hence, the shared posts, the reactions on other posts and their circle of friends can help to reflect their interests. However, exploiting all this information through the analysis of user profiles can help to enhance user’s interests identification performances. In this paper, we propose a deep learning user’s profile analysis based approach that relies on users’ personal information and textual content for detecting user’s interests and preferences. We experimented our approach using a large Facebook dataset, and show how the deep learning approach perform significantly better than the classical algorithms such as SVM.
Deep learning for seismic data semantic segmentation
(University of Eloued جامعة الوادي, 2022-01-24) Naoui, Mohammed Anouar; Lejdel, Brahim; Kazar, Okba; Berrehouma, Ridha
Drilling for oil and gas is an expensive and time-consuming process. Companies in the oil and gas industry invest millions of dollars in an effort to improve their understanding of subsurface components, and using traditional workflows for interpreting large volumes of seismic data is an important part of this effort. Manually defining links between geological characteristics and seismic patterns is required by geoscientists. As a result, geologists and oil and gas industry businesses resorted to a seismic survey, in which seismic waves provide a wealth of information about what is inside the earth without the need to dig. The main of this paper concerns the identification of salt layers of a seismic image by a computer which often coexist with gas and oil under the ground by proposing a deep Learning for seismic analysis.We propose U-net architecture to discover seismic data. Moreover, we study the data augmentation with U-net architecture. The result of data augmentation can perform 10 % the U-net architecture model.
Deep Neural networks based TensorFlow Model for IoT lightweight cipher attack
(University of Eloued جامعة الوادي, 2022-01-24) TOLBA, Zakaria; DERDOUR, Makhlouf
The internet of Things (IoT) technology is present in all aspects of our modern lives, and its standard usage is increasing remarkably. But their inherent limitations in size, storage memory, and power consumption limit its specific functionality in the secure transmission of sensitive information, where the development of lightweight ciphers responds adequately to these limitations. However, the conventional cryptanalysis of these modern ciphers can be impractical or demonstrate apparent limitations to be generalized. Because they frequently require a large amount of considerable time, known plain texts, and big storage memory, they are typically performed without the restriction of key space, or only the reduced round variants are attacked. This work proposes a deep learning (DL) model-based approach for a successful attack that discovers the plain text from cipher text one, it’s demonstrated that the proposed DL-based cryptanalysis represents a promising step towards a more efficient and automated test to verify the security of emerging lightweight ciphers. We directly attack the encryption independently of the key or the number of rounds using the TensorFlow platform in google collaboratory notebook environment that runs in the cloud and stores the results on Google Drive, the results are communicated to demonstrate precisely the effective performance of the attack, and numerous experiments were performed to confirm the study.
Digital Text Authentication Using Deep Learning: Proposition for the Digital Quranic Text
(University of Eloued جامعة الوادي, 2022-01-24) Touati-Hamad, Zineb; Laouar, Mohamed Ridda; Bendib, Issam
Nowadays, the detection of digital text manipulation is a hot topic in natural language processing and artificial intelligence. This type of text spreads quickly and inexpensively, which can cause great concern due to its negative impact on social life. The text authentication process has gained a great deal of interest. However, the authentication of Arabic texts is still under development. The Quran is one of the Arabic texts sensitive to change and the most vulnerable to falsification. In order to prevent misuse of this type of text, in this research, a deep learning approach based on the LSTM network and the pretrained Word Embeddings has been developed for authentication one of the manipulations types of the Arabic Quranic texts. By building a model that automatically enables Internet users to validate the Quran content's arrangement, the experimental results showed that the proposed approach could improve text classification accuracy and achieve a significant time difference compared to previous works.
Dilated Convolutions based 3D U-Net for Multi-Modal Brain Image Segmentation
(University of Eloued جامعة الوادي, 2022-01-24) Kemassi, Ouissam; Maamri, Oussama; Bouanane, Khadra; Kriker, Ouissal
Several deep learning based medical image segmentation methods use U-Net architecture and its variants as a baseline model. This is because U-Net has been successfully applied to many other tasks. it was noticed that the U-Net-based models are unable to extract features for segmenting small masks or fine edges. To overcome this issue, we propose a new 3D U-Net-based model, baptized Y- Net. In this model, we make use of dilated convolution which has shown its effectiveness in grasping different features at different scales. This allows us to capture more information from small anatomical parts. Our model is assessed on MRbrains13 dataset for brain tissue segmentation task. Compared to the traditional UNet 3D, the obtained results show that the proposed model performs well, especially in segmenting white matter and grey matter tissues.
Do We Need Change Detection for Dynamic Optimization Problems?
(University of Eloued جامعة الوادي, 2022-01-24) Boulesnane, Abdennour; Meshoul, Souham
Solving dynamic optimization problems is more challenging than static ones. When a change in the objective landscape occurs, the search process may not be powerful enough to track new optima. For population based algorithms this is referred to as diversity loss problem. Furthermore, the memory of old optima becomes outdated and if not correctly dealt with, the evolution of the search process may be misguided. Recently, a new interesting trend in dealing with optimization in dynamic environments has emerged toward developing new algorithms that are able to effectively handle changes without using any change detection scheme, and hence no extra computational cost is needed. There exist several works in the literature that attempt to maintain diversity without change detection. However, not that much work has been devoted to studies that investigate the possibility to overcome the outdated memory problem without expensive change detection. This study presents a comprehensive survey of the various change detection based methods. As part of this survey, we include a classification of the change detection schemes and we identify the main features of each method.

Browse

Browsing 03.Forum of Artificial Intelligence and Its Applications by Title