Publications
Title: Enhancing Robot Navigation: Integrating Safety Zones into Path Planning
Authors: Evis Plaku, Arben Çela and Erion Plaku
Journal: Lecture Notes in Electrical Engineering Series by Springer
Abstract:
Path planning is essential for guiding a robot to its destination while avoiding obstacles. In practical scenarios, the robot is often required to remain within predefined safe areas during navigation. This allows the robot to divert from its main path during emergencies and follow alternative routes to a safety center.
This paper introduces a novel method to incorporate safety zones into path planning. Each zone is defined by a central point and a radius. Our approach efficiently plans paths to the goal, ensuring that the robot can reach a safety center without having to travel more than the radius of the safety zone. Using sampling, our approach constructs a roadmap with navigation routes and identifies safe locations that satisfy the distance requirements for reaching a safety center. The safe portion of the roadmap is then searched to find a path to the goal.
We demonstrate the effectiveness of our approach through simulated experiments in obstacle-rich 2D and 3D environments, utilizing car and blimp robot models.
To appear
Title: A novel approach to electricity demand forecasting: an optimized Kalman filter-based RBF model
Authors: Agresa Qosja, Didier Georges, Eralda Gjika, Ligor Nikolla, Arben Cela
Journal: International Journal of Dynamics and Control
Publisher: Springer Berlin Heidelberg, Springer Nature
DOI: https://doi.org/10.1007/s40435-025-01638-1
Abstract
This study explores next-day electricity demand forecasting using the Kalman filter for parameter learning in a Radial Basis Function(RBF)NARMAXmodel.Thisapproachiscomparedtobatchregression-basedRBFNARMAXmodels,Feedforward Neural Networks (FFNN), and Recurrent Neural Networks (RNN), incorporating temperature as an exogenous variable. The models are trained using a novel combination of step-forward validation and grid search for parameter selection, which enhances computational efficiency, accuracy, and model complexity. Mean Squared Error (MSE) is employed to determine the optimal parameters. Results indicate that the Kalman filter-based RBF model achieves the highest accuracy with the given dataset, outperforming neural networks in predictive performance. It proves computationally efficient and effectively captures seasonal patterns in the time series. Model performance is evaluated using MAPE, MSE, MAE, and R2, achieving over 90% accuracy on both the training and testing sets for the recommended models
Conference: 2024 IEEE 20th International Conference on Intelligent Computer Communication
Jiang You Université Paris-Est Créteil, FR | Arben Cela UMT-AL, LISSI, ESIEE Paris, FR | René Natowicz ESIEE Paris, FR | Jacob Ouanounou HN-Services, FR | Patrick Siarry Université de Paris Créteil, FR
Abstract: Time series forecasting task predicts future trends based on historical information. Transformer-based U-Net architectures, despite their success in medical image segmentation, have limitations in both expressiveness and computation efficiency in time series forecasting as evidenced in YFormer. To tackle these challenges, we introduce Kernel-U-Net, a flexible and kernel-customizable U-shape neural network architecture. The kernel-U-Net encoder compresses the input series into latent vectors, and its symmetric decoder subsequently expands these vectors into output series. Specifically, Kernel-U-Net separates the procedure of partitioning input time series into patches from kernel manipulation, thereby providing the convenience of customized executing kernels. Our method offers two primary advantages:
1) Flexibility in kernel customization to adapt to specific datasets; and 2) Enhanced computational efficiency, with the complexity of the Transformer layer reduced to linear. Experiments on seven real-world datasets, demonstrate that Kernel-U-Net’s performance either exceeds or meets that of the existing state-of-the-art model in the majority of cases in channel-independent settings. The source code for Kernel-U-Net will be made publicly available for further research and application.
Conference INISTA/IEEE: https://dcti.ucv.ro/inista2024/
Conference: 2024 IEEE 20th International Conference on Intelligent Computer Communication
Jiang You Université Paris-Est Créteil, FR | Arben Cela UMT-AL, LISSI, ESIEE Paris, FR | René Natowicz ESIEE Paris, FR | Jacob Ouanounou HN-Services, FR | Patrick Siarry Université de Paris Créteil, FR
Abstract: Anomaly detection in time series data is a critical challenge across various domains. Traditional methods typically focus on identifying anomalies in immediate subsequent steps, often underestimating the significance of temporal dynamics such as delay time and horizons of anomalies, which generally require extensive post-analysis. This paper introduces a novel approach for time series anomaly prediction, incorporating temporal information directly into the prediction results. We propose a new dataset specifically designed to evaluate this approach and conduct comprehensive experiments using several state-of-the-art methods. Our results demonstrate the efficacy of our approach in providing timely and accurate anomaly predictions, setting a new benchmark for future research in this field.
Prof. Asoc. Dr Lekë Pepkolaj nga Universiteti Metropolitan Tirana, në bashkëpunim me Qendrën e Fizikës së Aplikuar, Universiteti i Salentos Itali; Institutin Kombëtar të Fizikës Bërthamore Lecce, Itali; Departamentin e Shkencave dhe Teknologjive Biologjike dhe Mjedisore, Universiteti i Salentos, Itali dhe Departamentin e Gjeografisë, Universiteti i Tiranës vjen me studimin me temë: “Comparative Analysis of Airborne Bacterial and Fungal Communities in South-Eastern Italy and in Albania Using the Compositional Analysis of 16S and ITS rRNA Gene Sequencing Datasets”.
Një studim krahasues mjaft i rendësishëm në lidhje me ndotjen e ajrit, konsiderata me vlerë të madhe për shendetin publik. Në analizën e të dhënave janë përdorur metodologji dhe teknika të avancuara të kohëve të fundit. Artikulli është i publikuar në revistën Atmosphere e indeksuar Q2 në Scopus. Studime të tjera në këtë fushë të rendësishme për vendin tonë janë në vazhdim.
Për më tepër shikoni linkun: https://www.mdpi.com/2073-4433/15/10/1155
Agresa Qosja | Didier Georges | Eralda Gjika | Ligor Nikolla | Arben Cela
Abstract
Electricity consumption forecasting stands as acritical research domain within electrical engineering, with myr-iad of traditional forecasting models and artificial intelligence techniques undergoing rigorous examination. This paper is
devoted to a comparison of three Machine Learning approaches to surrogate modelling and forecasting of the daily electricity consumption in Tirana, Albania: A Radial Basis Function (RBF) approach, a feed forward Neural Network approach and a Recurrent Neural Network approach. Through meticulous
experimentation across four distinct scenarios encompassing variations in training/testing splits, historical data utilization, and hyper parameter optimization, we thoroughly evaluate the performance of each model. Comparative analysis is con- ducted based on model fit, computational efficiency, and error measurement metrics. Our findings highlight the remarkable
performance of the RBF ARX approach, underscoring its effectiveness in accurately forecasting electricity consumption.
Key-words: Electricity consumption forecasting, neural net- work modelling, Radial Basis Function modelling
Evis Plaku | Klei Jahaj | Arben Cela | Nikolla Civici
Abstract
Automated news article classification is a method of categorizing textual data into predefined classes. Addressing this problem finds applications in diverse domains, including information retrieval, topic modeling, sentiment analysis and content recommendation systems. In Albanian, though there is a rapid increase of digital content, there is limited availability of text corpora, presenting significant obstacles for advancement of natural language processing research and applications.
The contribution of this paper is twofold. First, we introduce a dataset consisting of 9600 news article titles spanning across various categories. Second, we utilize this dataset to assess the effectiveness of several machine learning algorithms for topic classification. Experimental results demonstrate the efficacy of recurrent neural networks in comparison to simpler classifiers and ensemble methods.
Agresa Qosja | Eralda Gjika | Didier Georges | Ligor Nikolla | Arben Cela
Abstract
Forecasting electricity consumption and production remains essential, necessitating the application of recent, hybrid, and classical models. Traditional approaches often fall short in capturing the intricacies of nonlinear and irregular data patterns. This prompts the need for new or adaptable models capable of achieving high accuracy. This paper compares three distinct Machine Learning approaches for surrogate modeling and forecasting of daily electricity consumption across five regions in Albania: a Radial Basis Function (RBF) approach, a feedforward Neural Network approach, and a Recurrent Neural Network approach.
Our aim is to develop a generalized model for these stations and provide highly accurate one-day ahead predictions. We conduct a comparative analysis based on model fit, computational efficiency, and error measurement metrics. Our findings underscore the exceptional performance of the RBF ARX approach, showcasing its efficacy in accurately forecasting electricity consumption based on daily data. Additionally, we identify daily mean temperature as a key factor influencing model performance.
Elton Domnori | Donald Elmazi | Gledi Tace
Abstract
The Internet of Things (IoT), has significantly impacted several industry sectors in recent years, particularly those that deal with air, soil, and water quality. With the growing diversity of sensors and their processing and communication capabilities, a number of methods have been put up for creating systems that continuously gather environmental data. These data can be further processed for long-term analysis or quick response. Every solution that has been put forth presents a static system, meaning that the sensors are positioned in fixed, designer-determined locations. Our work presents a dynamic and autonomous system in which sensors are free to travel within a given area and, by utilizing intelligent algorithms and distributed systems techniques, they determine the optimal way to employ in order to monitor the target environment by combining intelligent algorithms and distributed systems techniques.
Abdusamed Kura | Donald Elmazi
Abstract
Machine learning models are essential instruments in modern data analytics, propelling progress in a multitude of fields. However, the effectiveness of these approaches depends on careful model selection and data preprocessing. Consequently, it is imperative to devise an enhanced model builder that facilitate data preprocessing and improve the creation of machine learning models. Understanding the critical role that machine learning models play in contemporary data analytics is essential to this effort. On the other hand, careful data preprocessing and wise model selection are necessary for these models to function effectively. Our study presents a solution that automates data preparation activities by addressing data irregularities such as null values, incorrect inputs, and redundant entries. The program then creates a variety of models on its own, increasing the adaptability and effectiveness of machine learning applications in order to solve these issues. This work has produced software that is a substantial development in terms of both speedy preprocessing and data cleansing, as well as its ability to construct several machine learning models. The program employs a strict assessment methodology to choose and maintain the most effective model by looking at a wide range of performance measures. Our software framework’s versatility and effectiveness are demonstrated through experimental validation, allaying worries about inflexible code. Rather, the program is a flexible framework that can easily learn from and adjust to a variety of datasets. In conclusion, by presenting a novel software solution designed for effective data preprocessing and model building, this work makes a substantial contribution to the development of machine learning approaches. Our software can help advance machine learning research and applications by automating important operations and making the process of selecting the best models easier.
Evis Plaku | Arben Cela | Erion Plaku
Abstract
Path planning is essential for guiding a robot to its destination while avoiding obstacles. In practical scenarios, the robot is often required to remain within predefined safe areas during navigation. This allows the robot to divert from its main path during emergencies and follow alternative routes to a safety center. This paper introduces a novel method to incorporate safety zones into path planning. Each zone is defined by a central point and a radius. Our approach efficiently plans paths to the goal, ensuring that the robot can reach a safety center without having to travel more than the radius of the safety zone. Using sampling, our approach constructs a roadmap with navigation routes and identifies safe locations that satisfy the distance requirements for reaching a safety center. The safe portion of the roadmap is then searched to find a path to the goal. We demonstrate the effectiveness of our approach through simulated experiments in obstacle-rich 2D and 3D environments, utilizing car and blimp robot models.
Fabjan Pjetri | Fjorda Prifti | Rajs Kastrati
Abstract
The paper presented by the authors aims to research the role of Microfinance in Albania over the years, especially in improving living conditions for the less privileged groups. The authors have tried to give a summarized panorama regarding the progress and evolution of Microfinance in the state of Albania. Portraying the latter as a good alternative solution to the shortcomings of the Traditional Banking System, to address the needs of certain segments of the population. The analysis used by the authors focuses on the major and current challenges that the Microfinance sector in Albania is facing, it also examines the perspectives and potential future opportunities of this developing sector in our country. Since most of the economic entities that operate in our country are small and medium-sized businesses, where family businesses account for a significant number of the total number of economic entities, the authors have considered it important to research this segment of the economy. Keywords: Microfinance, Financial System, Financial Supervision Authority, Digitization of financial services in relation to market demands.
Evis Plaku | Erion Plaku | Patricio Simari
Abstract
This paper presents an approach that integrates the geometric notion of clearance (distance to the closest obstacle) into sampling-based motion planning to enable a robot to safely navigate in challenging environments. To reach the goal destination, the robot must obey geometric and differential constraints that arise from the underlying motion dynamics and the characteristics of the environment. To produce safe paths, the proposed approach expands a motion tree of collision-free and dynamically feasible motions while maintaining locally maximal clearance. In distinction from related work, rather than explicitly constructing the medial axis, the proposed approach imposes a grid or a triangular tessellation over the free space and uses the clearance information to construct a weighted graph where edges that connect regions with low clearance have high cost. Minimum-cost paths over this graph produce high-clearance routes that tend to follow the medial axis without requiring its explicit construction. A key aspect of the proposed approach is a route-following component which efficiently expands the motion tree to closely follow such high-clearance routes. When expansion along the current route becomes difficult, edges in the tessellation are penalized in order to promote motion-tree expansions along alternative high-clearance routes to the goal. Experiments using vehicle models with second-order dynamics demonstrate that the robot is able to successfully navigate in complex environments. Comparisons to the state-of-the-art show computational speedups of one or more orders of magnitude.
Evis Plaku | Erion Plaku | Patricio Simari
Abstract
This paper presents an approach that combines geometry processing with motion planning to enable a robot to efficiently navigate in unstructured environments. The proposed approach relies on a novel oversegmentation method to produce a decomposition of the free space into a set of connected regions. This provides a general and simplified planning layer with navigational routes along which sampling-based motion planning expands a tree of collision-free and dynamically feasible motions to reach the goal. Experiments using robot models with nonlinear dynamics operating in complex environments show significant speedups over related work.