Algorithm 4: GreenShield Framework Operation

Network traffic stream $\mathcal{S}$, Model parameters $\theta$, energy budget $\mathrm{B}$ Threat classifications, encrypted communications, updated models Initialization: Load compressed student model $\mathcal{S}$ with parameters ${\theta }_{\mathrm{S}}$ Initialize ASCON cipher with session keys Start carbon intensity monitoring Traffic Processing at Edge: Extract features: $\mathrm{x}\leftarrow$ FeatureExtract(packet) Assess threat level: $\tau \left(\mathrm{t}\right)\leftarrow$ (4) Select quantization: ${\mathrm{q}}_{\mathrm{b}}\leftarrow$ (5) Classify: $\widehat{\mathrm{y}}\leftarrow {\mathcal{S}}_{{\mathrm{q}}_{\mathrm{b}}}\left(\mathrm{x};{\theta }_{\mathrm{S}}\right)$ Encrypt alert: $\mathrm{c}\leftarrow \mathrm{ASCON}.\mathrm{Enc}\left(\mathrm{k},\mathrm{alert}\right)$ Forward to fog node for detailed analysis Federated Learning Update (periodic): Compute local gradients: $\nabla {\mathrm{F}}_{\mathrm{k}}\leftarrow {\nabla }_{\theta }{\mathcal{L}}_{\mathrm{K}\mathrm{D}}$ Compress gradients: $\tilde{\nabla }\leftarrow \mathrm{Compress}\left(\nabla {\mathrm{F}}_{\mathrm{k}}\right)$ Transmit to fog node Receive aggregated update from (15) Update local model: ${\theta }_{\mathrm{S}}\leftarrow {\theta }_{\mathrm{S}}-\eta \tilde{\nabla }$ Carbon-Aware Scheduling: Update carbon forecasts via (25) Schedule pending jobs via Algorithm 3 Monitor energy consumption via (6)