Article Preview
TopIntroduction
As global climate change continues to escalate, carbon neutrality has become an important strategy for reducing greenhouse gas emissions and addressing climate change (Zhao et al., 2022). It is achieved by capturing and storing carbon dioxide (CO2) from the atmosphere or by reducing emissions through the use of renewable energy sources, thereby achieving a carbon balance (Waheed et al., 2019). However, in practice, we face a key challenge: how to detect and address anomalies in the carbon sequestration process. Carbon consumption anomalies typically refer to unexpected events that occur in carbon capture and storage systems, which may result in CO2 leakage, emissions, or system failures, thereby jeopardizing the overall efficiency and feasibility of carbon neutrality (Sun & Ren, 2021). The difficulty in detecting carbon consumption anomalies lies in their various forms and the difficulty of predicting them in advance. For example, system failures may result from equipment damage, operator errors, or natural disasters, while leakage events may be caused by pipeline ruptures, poor seals, or operational mistakes (Somu et al., 2021). These abnormal events not only pose environmental risks but can also lead to high maintenance and cleaning costs, and even damage a company’s reputation (Amasyali & El-Gohary, 2018). Another complexity is that carbon neutrality systems typically involve a large number of sensors and monitoring devices that generate a vast amount of time-series data. This data contains information about parameters, such as temperature, pressure, flow rate, and CO2 concentration, which may change when abnormal events occur. Therefore, effectively monitoring and analyzing this time-series data to identify abnormal events is crucial for the success of carbon neutrality (R. Li et al., 2021).
To address these issues, people have started applying deep learning technology to enhance the detection and prediction of anomalies in the carbon neutrality process (Anthony et al., 2020). Deep learning is a powerful machine learning approach that has achieved remarkable success in various fields. In the field of carbon neutrality, deep learning is widely used to tackle the challenges of anomaly detection. By leveraging large amounts of data and powerful neural network models, deep learning can identify and predict anomalies related to carbon neutrality, improving system stability and efficiency (Amasyali & El-Gohary, 2018).
In recent years, the application of time series forecasting in deep learning has been crucial for carbon neutrality research (Liu, Wang et al. 2023). Time series data contain information that varies over time, and these data are essential for detecting abnormal events and predicting the performance of future carbon neutrality processes (Wang, Sun et al. 2021). For instance, in the context of carbon consumption monitoring for running activities, by building deep neural network time series forecasting models, we can learn personalized patterns from the data, including a runner’s exercise habits, breathing patterns, and heart rate variations. These models can be used to estimate carbon consumption in real time, providing runners with advice on how to exercise more environmentally friendly, such as adjusting their stride or exercise intensity to minimize carbon emissions (Liu et al., 2023). Additionally, this task can provide valuable insights into individual health and lifestyle (Zhang et al., 2023). Runners can understand how much carbon they are consuming during their workouts, leading to a better awareness of their carbon footprint and encouraging more environmentally friendly lifestyles (Yang et al., 2022).
Below, we introduce recent relevant work in this area: