Carbon-NeuGC: Neural Granger Causality Based Attribution Analysis of Power System Carbon Intensity

With the escalating global concern over climate change, accurately attributing carbon emissions within power systems is crucial for formulating effective environmental policies. This paper introduces Carbon-NeuGC, a novel approach leveraging Long Short-Term Memory networks to analyze Granger causality (GC) in carbon intensity which measure the emissions per unit active power consumption. The purposed approach constructs component-wise LSTM models to capture complex, nonlinear dynamics between carbon intensity and influencing factors, such as power generation and demand without pre-defined time lag. Component-wise LSTM models are trained on empirical data, with group lasso regularization applied to improve GC identification. Granger causal relationships are then inferred by analyzing the weight coefficients of the network's first layer. A case study based on the IEEE 14-node system demonstrates the framework's robustness in identifying Granger causal relationships, providing actionable insights for policymakers to mitigate carbon emissions.