Efficient greenhouse gas prediction using IoT data streams and a CNN-BiLSTM-KAN model

In response to the challenge of accurately predicting greenhouse gas emissions in the context of climate change, this study introduces a novel hybrid deep learning model, CNN-BiLSTM-KAN, which integrates Convolutional Neural Networks (CNN), Bidirectional Long Short-Term Memory (BiLSTM) networks, and Kolmogorov–Arnold Networks (KAN). The proposed model utilizes CNN for local feature extraction from time series data, while BiLSTM captures bidirectional temporal dependencies. KAN, with its differentiable computational structure, effectively overcomes the limitations of traditional Multilayer Perceptrons (MLPs) in handling complex time series data. Experimental evaluations on two public datasets, GHGRP and EDGAR, demonstrate the superior performance of the CNN-BiLSTM-KAN model, particularly in terms of key metrics such as Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), and the coefficient of determination ( R 2 ). The model’s ability to manage nonlinear features in complex time series data is a key advantage. Furthermore, ablation experiments highlight the importance of each model component, with KAN playing a crucial role in capturing intricate nonlinear relationships. This research not only improves the accuracy of greenhouse gas emission predictions but also establishes a robust framework for modeling complex time series data.