نشریه توسعه پایدار و ساختمان هوشمند
پیاپی 4 (زمستان 1403)

The fusion layer has been widely used for image classification with more than a decade of deep learning. Many models using the fusion layer have tried to overcome the disadvantages of the traditional algorithm, which depends on the hand-drawn features.In general, deep learning models often contain trainable parameters and require a large number of labeled examples to achieve optimal performance. This paper focuses on optimal parameters and related previous work. Various advanced approaches have been evaluated through several experiments. The evaluation results on the CIFAR-10dataset showed that thefastest and best method is the computationally simplest method, which is the Maximum Fusion method. After the Maximum method, the Lip Fusion method has the best results, but it increases the detection time by almost fifty percent. This increase in time isdue to its logarithmic and exponential computational complexities. The hybrid method could not perform better than the maximum pooling in terms of classification criteria in experiments. However, this method has the lowest error rate in thetraining phase.

Imbalanced datasets are very common in the real world and pose a significant challenge in classification and clustering tasks. Imbalanced datasets are defined as those where the instances of one or more classes significantly outnumber the instances of other classes. This research aims to present a mechanism for balancing imbalanced datasets by integrating undersampling and oversampling techniques. Overall, the goal of this study is to propose a hybrid undersampling-oversampling method that can efficiently and effectively balance imbalanced datasets. The proposed approach focuses on minority and majority data points rather than entire minority and majority classes. To identify these data points, the Birch clustering algorithm is employed. Consequently, the proposed method is named Birch-Resampling. This method consists of six stages. The MATLAB software was used for the implementation of this algorithm, and for evaluation, 11two-class and multi-class datasets with varying sizes and imbalance ratios were utilized to comprehensively assess the performance of the proposed system. The results obtained from the proposed algorithm werecompared with five other methods. The outcomes of these comparisons demonstrate that the proposed algorithm exhibits high accuracy in data classification.

Considering the vast amount of available information, the issue of improving information retrieval has always been one of themost important factors in increasing the speed of information retrieval. The prohibited search algorithm is a meta-heuristic method that utilizes memory components to explore and exploit the search space. This research presents a new retrieval method based on the whale optimization algorithm and the prohibited search method. This approach uses a selection list to record the best candidate methods. These methods are employed to guide the members in order to prevent them from getting trapped in local optima. The mentioned method enhances the diversity of the group members using a merging operator inspired by genetic algorithms, which increases the ability to achieve a faster convergence rate. Additionally, it provides the possibility of obtaining high-quality solutions through the integration of methods among members and the selection list. Evaluation results indicate that the proposed method outperforms the compared methods across all evaluation criteria.
 

Given the importance of seismic resilience in the design and performance evaluation of structures under earthquakes, implementing strategies to enhance this resilience has become a necessity in structural engineering. Among these strategies, the use of viscous dampers has gained attention as an effective method for controlling structural seismic responses. However, a quantitative investigation into the effect of increased effective damping on the seismic resilience index (particularly in reinforced concretestructures) has not yet been thoroughly conducted. This research investigates the impact of increased effective damping on the seismic resilience of a reinforced concrete moment frame. To achieve this, a four-story, three-bay reinforced concrete moment frame was modeled in the OpenSees software, and viscous dampers with a uniform distribution were installed on all floors. The damping coefficients of the dampers were adjusted in two different scenarios to achieve effective damping ratios of 15%and 35%for the frame, respectively. Nonlinear dynamic analyses were performed on the model, and the seismic resilience index of the frame was calculated for each damping level. The results indicate that the seismic resilience index of the frame increases with increasing effective damping. This suggests that the use of viscous dampers can significantly improve the seismic performance of structures. Increasing damping not only reduces the seismic responses of thestructure, including story drifts and floor accelerations, but also helps improve the seismic resilience of the structure against earthquakes. The findings of this research can serve as a scientific basis for the optimal design of viscous damper systems in reinforced concrete structures.

The Energy sector is one of the most important economic infrastructure of the economy. Investment in the energy sector can have wide-ranging effects on macroeconomic variables and play a key role in the economic and social development of countries. Energy policy goals and economic goals cannot be achieved without the development of the energy sector. In this study, a general equilibrium model is used to study the effects of the economic performance of Iran's energy sector investment growth.Three scenarios regarding the increase of investment with the rates of 10, 20and 30percent have been considered. According to the Social Accounting Matrix 2013, the changes of some macroeconomic variables under these scenarios are analyzed. The results show that the growth of GDP up to about 4%, and export growth up to about 12%is possible. Also, household income and consumption will improve significantly. In the study of sectoral effects, the highest changes of production and export isrelated to the energy and industry sector, which is related to the improvement of access to energy and the reduction of its costs. It is worth mentioning that the research results show the increase in carbon emissions and can point to the need to focus on clean energy in the investment process in the energy sector so that this development to be environmentally sustainable.

With the growing use of encryption in network traffic, especially through protocols like TLS and SSL, the effectiveness of traditional application identification methods based on packet content analysis has significantly declined. In modern high-bandwidth networks, real-time and accurate application identification faces challenges not only in terms of speed, volume, and traffic diversity but also due to high resource costs. This study presents an adaptive approach for traffic stream classification that dynamically and continuously updates the classification model by leveraging Deep Packet Inspection (DPI) as a reference source. At the core of the proposed classifier, a Deep Neural Network (DNN) learning algorithm is employed, capable ofaccurately identifying complex and hidden patterns in encrypted traffic. The proposed method is empirically evaluated on two public datasets, ISCX VPN-nonVPN and ISCX Tor-nonTor, and the results indicate that DNN performs better in certain scenarios—particularly in classifying more complex flows or when real-time labeled data is scarce—compared to traditional classifiers.