Global climate change driven by carbon dioxide (CO2) emissions necessitates accurate forecasting for effective mitigation strategies. Traditional forecasting models, such as ARIMA and statistical methods, often struggle to handle the non-linear and long-term temporal dependencies in CO2 emission data. This paper evaluates the performance of advanced Recurrent Neural Network (RNN) architectures, specifically Long Short-Term Memory (LSTM), Stacked LSTM, Bidirectional LSTM (Bi-LSTM), and Gated Recurrent Unit (GRU) with attention mechanisms, in forecasting long-term CO2 emissions. These models are compared against traditional machine learning algorithms like Random Forest and Gradient Boosted Trees (GBT). The Random Forest and Gradient Boosted Trees, implemented using PySpark framework for distributed computing, are leveraged for efficient big data handling, enabling scalable solutions for CO2 emission forecasting. Our experimental results, using metrics such as Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), and Mean Absolute Percentage Error (MAPE), show that the Stacked LSTM model achieved the lowest MAE of 25.795, RMSE of 80.678, and MAPE of 1.072, outperforming Random Forest (MAE: 72.4, RMSE: 310.5, MAPE: 1.32) and GBT (MAE: 75.6, RMSE: 356.8, MAPE: 1.57). The findings demonstrate that advanced RNN models, particularly the Stacked LSTM and Bi-LSTM with attention, provide superior performance over traditional models, effectively capturing complex temporal patterns in the data. This study offers valuable insights into CO2 emission forecasting approaches, contributing to the development of accurate tools for climate policy and strategy formulation.

Climate change and global warming demand immediate global action to protect future generations. Although deep learning is crucial for solving complex tasks, its high energy consumption and carbon emissions from extensive training processes necessitate improvements in efficiency to reduce its environmental impact. This research addresses the need for accurate carbon emission tracking to mitigate climate change and promote sustainable development. We evaluated deep learning models—specifically, the Temporal Fusion Transformer (TFT), Attention-based LSTM (Att-LSTM), and standard LSTM—in predicting CO2 emissions using datasets from World Bank Indicators (WBI) and Our World in Data (OWID). The results highlight TFT’s superiority, achieving the lowest mean squared error (MSE) in validation (0.0029) compared to Att-LSTM (0.0070) and LSTM (0.0072). Additionally, TFT demonstrates a lower root mean squared error (RMSE) in validation (0.0539) than both Att-LSTM (0.0712) and LSTM (0.0909), showcasing its enhanced predictive accuracy across evaluation tools like CarbonTracker, eco2AI, and CodeCarbon. Furthermore, TFT consistently outperforms LSTM in computational efficiency and carbon emissions, despite LSTM’s lower power usage but higher emissions in specific scenarios. These findings underscore TFT's pivotal role in formulating effective climate policies and promoting renewable energy sources, emphasizing its potential to significantly contribute to global warming mitigation and sustainability.

Rapid technical advancements in recent times have made it possible to produce a class of affordable unmanned aerial vehicles (UAVs), which can serve a wide range of public and private users for different industrial or non-industrial requirements. It is very complex to set up a communication network among the UAVs, which is very essential for completing any task like collecting data in extreme weather conditions. For UAV placement, we have suggested a distributed approach using PPO-based reinforcement learning; here, UAVs are utilized as aerial access points to create a mesh network that connects to ground nodes positioned in a designated area and can collect data from ground nodes to predict weather conditions. Our objective is to fulfill each ground node’s data rate requirements while deploying the fewest number of UAVs possible to cover it. The convex hull formation of swarm uavs satisfy our objective, which arranges the n uavs on a convex hull’s vertex. In this study, we present a deep reinforcement learning-based method for building swarm uavs with convex hull patterns in the euclidean plane. The existing research on convex hull pattern generation uses heuristic and combinatorial optimization techniques and evaluates performance in terms of time and space consumed. Convex hull patterns increase the mutual visibility of swarm UAVs, which helps in the cooperation of UAVs. In our research for the first time, convex hull pattern formation around a given target point is accomplished using PPO-based Deep Reinforcement Learning (DRL).

Accurate forecasting of critical pollutants such as NO2and PM2.5 is essential in the ongoing battle against air pollution. This paper introduces ConvTrans, a novel hybrid architecture that integrates convolutional and Transformer technologies to enhance the precision of PM2.5 and NO2predictions. Our evaluations demonstrate that ConvTrans improves PM2.5 prediction accuracy, increasing the R-squared value by 3.6%, reducing RMSE by 5.4%, MSE by 7.5%, and MAE by 17.3%. Similarly, for NO2predictions, the model shows a 4.2% increase in R-squared, along with a 5.4% reduction in RMSE, a 13.7% decrease in MSE, and a 5.6% reduction in MAE. These results highlight the model’s effectiveness in short-term forecasting, addressing key challenges in environmental management.

Air pollution presents serious threats to public health and environmental sustainability, especially when it comes to concentrations of ozone (O3). Even though they are accurate, traditional monitoring methods have spatial limitations that cause gaps in the coverage of data. In order to tackle this issue, we provide a hybrid deep learning model that predicts O3levels in different parts of India by integrating Transformer and Convolutional Neural Networks (CNN) using satellite and meteorological data. While the Transformer model uses meteorological data to identify temporal trends, the CNN component analyzes satellite pictures to extract spatial information. Multiple assessment criteria, including R2, MSE, RMSE, and MAE, show that we enhance prediction accuracy by merging both models. Our model performs better in both short- and long-term predictions than other deep learning techniques and RNN, such as CNN, GRU, and LSTM models. The results suggest that the hybrid CNN-Transformer model is a promising solution for accurate air pollution forecasting, offering critical insights for timely interventions and policy development.

Air quality is a critical factor influencing public health, with pollutants such as particulate matter (PM10, PM2.5), ground-level ozone (O3), carbon monoxide (CO), sulfur dioxide (SO2), and nitrogen dioxide (NO2) posing significant health risks. Understanding the relationship between these pollutants is essential for effective air quality management. In this study, we apply a comprehensive range of statistical models, including Multivariate Analysis of Variance (MANOVA), Canonical Correlation Analysis (CCA), Factor Analysis, post-hoc tests, linear regression, and other techniques to evaluate the correlation between PM2.5 and key air pollutants. By employing these methods, we aim to identify the most significant contributors to PM2.5 levels and provide insights to inform air quality control strategies. Our findings offer a robust statistical framework for predicting PM2.5 concentrations, enhancing the ability of policymakers and environmental agencies to mitigate pollution-related health risks.

The rapid growth of mobile data and computing demands has strained resource-constrained edge devices, particularly in supporting IoT applications. Mobile Edge Computing (MEC) offloading alleviates these challenges by shifting complex tasks to edge-cloud servers, reducing computational burdens and enhancing efficiency. The integration of 5G and 6G technologies further enhances MEC by providing ultra-low latency and high-bandwidth connections. Despite the use of deep learning methods in task offloading, current approaches struggle with slow learning and adaptability issues. To address these challenges, we introduce a Deep Meta Reinforcement Learning based Offloading (Deep Meta-RL) Framework. Formulating the task offloading problem as a Markov Decision Process (MDP) enables us to leverage the Deep Meta-RL algorithm for precise offloading decisions, reinforcement learning’s decision-making, and meta-learning’s adaptability. Simulation results demonstrate that Deep Meta-RL significantly outperforms traditional DQN algorithms, achieving a 16.75% improvement in rewards and reducing latency by 20%.

This research work introduces a new strategy to enhance computation offloading in wireless-powered Edge-Enabled Mobile Computing (EEMC) networks by utilizing Dueling Double Deep Q-Networks with Attention (Dueling DDQN-A). EEMC facilitates the offloading of computational tasks from mobile devices to local edge servers, leading to decreased latency and improved energy efficiency. However, the dynamic and stochastic nature of wireless environments presents significant challenges for real-time task offloading decisions. To address this, we enhance the Dueling DDQN framework by integrating an attention mechanism to prioritize the most relevant features in the state space. The proposed Dueling DDQN-A model enhances decision-making by separating state value estimation from action advantage estimation. Additionally, the model uses attention to prioritize key state features, leading to improved adaptability in dynamic wireless channel and network environments. We frame the offloading decision problem as a Markov Decision Process (MDP) and apply deep reinforcement learning to optimize both the computation rate and energy efficiency. We compare and evaluate the model through comprehensive simulations using the latest techniques like Simple DQN, Double DQN, and Dueling DDQN. Our results demonstrate that Dueling DDQN-A achieves a 59.12% improvement in average computation rate and a 16.19% reduction in energy consumption over the baseline models. Additionally, it significantly outperforms the baselines in terms of training loss and convergence speed. These findings suggest that Dueling DDQN-A offers a robust and highly efficient solution for real-time computation offloading in wireless-powered EEMC networks, making it a strong candidate for deployment in real-world scenarios.

UAVs have the potential to enhance wireless systems by improving range and quality, and this can be achieved by utilising the Mobile Edge Computing provided by the unmanned aerial vehicle. In this system, the MEC server mounted on the unmanned aerial vehicle can offer offload services to all the User Equipment in a given space. By offloading some proportion of its tasks to unmanned aerial vehicle for computation, the UE can perform the remaining tasks locally. The objective of this study is to minimize the maximum processing delay in the whole process by optimizing on four parameters, scheduling of the user equipment, portion of the task that has to be offloaded, angle of flight for the UAV, and speed of flight of the UAV, taking into account discrete variables and power constraints. However, due to the nonconvexity, the state space’s high dimension, and action’s space continuous nature of this problem, we are proposing a Proximal Policy optimization algorithm which is based on boosting the policy gradient. Further we will do a comparative analysis of PPO algorithm with other popular Reinforcement Learning algorithms, particularly the Deep Deterministic Policy Gradient algorithm. PPO algorithm can quickly achieve the optimal policy for offloading computation in a dynamic environment. The results obtained implies that the both PPO and DDPG algorithm converges quickly, and the processing delay is minimizd. However, our proposed PPO algorithm has shown significant improvement in minimizing the processing delay. Our model also performs way better compared to basic algorithm such as Deep Q Network (DQN).

Datacenter carbon emissions are rising, which poses a serious issue that must be addressed quickly. We may see differences in emissions when we look at the carbon intensity of the electrical system since various areas have different energy sources. We can time the execution of workloads to occur during periods when the carbon intensity is lower by using this temporal variability. This paper aims to address the challenge of reducing carbon emissions in cloud computing by proposing a framework for workload shifting based on low carbon intensity periods in the power grid. The study focuses on four countries and their carbon production in the year 2022, calculating the carbon intensity for each country. Additionally, the paper identifies different cloud computing workloads and integrates constraints such as power, SLA, carbon emissions, and routing into the framework. The crucial factors considered during workload shifting include geo-distributed load balancing and right-sizing the data center.A simulation is developed to evaluate the proposed framework, simulating scenarios with shiftable workloads. The results are compared and analyzed, assessing the framework’s effectiveness in reducing carbon emissions while meeting the specified constraints. The findings highlight the potential benefits of workload shifting in reducing carbon emissions and improving environmental sustainability. Overall, this research contributes to the advancement of green computing and offers insights into the development of sustainable cloud computing practices.

Climate change and global warming are urgent environmental issues demanding immediate action to safeguard future generations. The major contributor to the greenhouse effect, carbon dioxide $\left(\mathrm{CO}_{2}\right)$, primarily originates from industrial and transportation fossil fuel combustion. International agreements, like the Paris Agreement, call for a 30-35% reduction in CO2emissions compared to 2005 levels. This research aims to predict CO2emissions and raise awareness among SAARC nations and governments about the increasing trend. We introduce a novel predictive framework using Attention-based Long Short-Term Memory (A-LSTM) for CO2emissions analysis. The Attention mechanism assigns variable weights to input data, facilitating indirect connections between LSTM outputs and pertinent inputs. This enhances resource allocation in the A-LSTM model, overcoming computational constraints. We integrate input parameters encompassing CO2emissions from land-use changes, oil, natural gas, and coal combustion to forecast CO2emissions and correlate them with population and per capita GDP. Our comparative analysis conclusively demonstrates the superior performance of A-LSTM models over baseline LSTM models when applied to the CO2emission dataset sourced from Our World in Data (OWID) and World Bank Indicator database. Specifically, the LSTM model registers a MAPE of 24.968 and an RMSE of 0.34, whereas the Attention-based LSTM model showcases a marked improvement of 57% with a considerably lower MAPE of 10.5902 and an RMSE of 0.107.