Resilience optimization in adversarial Multi-Agent Mission Systems (MAMSs) requires more than resilience measurement: it requires a decision mechanism whose objective is explicitly aligned with the resilience metric itself. This paper proposes a net-effectiveness-based evaluation–optimization framework for bilateral temporal adversarial operational-loop networks (BTAR), turning resilience evaluation into a reusable world model and reward source for decision-making. We model bilateral confrontation as a heterogeneous temporal network and quantify resilience through the integral of net effectiveness, which captures temporal and adversarial dynamics beyond static topological indicators. We then formulate resilience enhancement as a scale-invariant, cost-aware Markov decision process and solve it with a hierarchically decoupled reinforcement learning architecture: an upper-level PPO policy learns when and for which chains to trigger selective replanning, while a lower-level greedy routine selects concrete replacement nodes. In this way, the optimization objective remains formally aligned with the evaluation metric. Experiments on a reproducible multi-scale, multi-style scenario pool show that the learned policy improves Rtotal from 0.3673 to 1.1696, reaching 96.8% of an idealized oracle benchmark (1.2082), while slightly outperforming the always-replan rule (1.1482) with a lower trigger rate. Under nonzero replanning cost, retraining yields a cost-aware policy whose trigger rate drops to 21.2% and whose cost-adjusted resilience at the target cost setting increases by +0.15, surpassing the greedy oracle benchmark in the high-cost regime. These results show that coupling net-effectiveness evaluation with hierarchical RL yields a deployable, scale-invariant, and cost-aware resilience optimizer.

Battery management systems increasingly require accurate battery health prognostics under strict on-device constraints. This paper presents DLNet, a practical framework with dual-stage distillation of liquid neural networks that turns a high-capacity model into compact and edge-deployable models for battery health prediction. DLNet first applies Euler discretization to reformulate liquid dynamics for embedded compatibility. It then performs dual-stage knowledge distillation to transfer the teacher model's temporal behavior and recover it after further compression. Pareto-guided selection under joint error-cost objectives retains student models that balance accuracy and efficiency. We evaluate DLNet on a widely used dataset and validate real-device feasibility on an Arduino Nano 33 BLE Sense using int8 deployment. The final deployed student achieves a low error of 0.0066 when predicting battery health over the next 100 cycles, which is 15.4% lower than the teacher model. It reduces the model size from 616 kB to 94 kB with 84.7% reduction and takes 21 ms per inference on the device. These results support a practical smaller wins observation that a small model can match or exceed a large teacher for edge-based prognostics with proper supervision and selection. Beyond batteries, the DLNet framework can extend to other industrial analytics tasks with strict hardware constraints.

Resilience optimization in adversarial Multi-Agent Mission Systems (MAMSs) requires a decision mechanism whose objective is explicitly aligned with the resilience metric, not merely resilience measurement. This paper proposes a net-effectiveness-based evaluation–optimization framework for bilateral temporal adversarial operational-loop networks (BTAR), turning resilience evaluation into a reusable world model and reward source for decision-making. We model bilateral confrontation as a heterogeneous temporal network and quantify resilience through the integral of net effectiveness, which captures temporal and adversarial dynamics beyond static topological indicators. We then formulate resilience enhancement as a fixed-dimensional, size-agnostic, cost-aware Markov decision process and solve it with a hierarchically decoupled reinforcement learning architecture: an upper-level PPO policy learns when and for which chains to trigger selective replanning, while a lower-level greedy routine selects concrete replacement nodes. In this way, the optimization objective remains aligned with the evaluation metric through a metric-aligned reward. Experiments on a reproducible multi-scale, multi-style scenario pool show that the learned policy improves $R_{\mathrm{total}}$ from $0.3673$ to $1.1696$, reaching $96.8\%$ of an idealized oracle benchmark ($1.2082$), while slightly outperforming the always-replan rule ($1.1482$) with a lower trigger rate. Under nonzero replanning cost, retraining yields a cost-aware policy whose trigger rate drops to $21.2\%$ and whose cost-adjusted resilience at the target cost setting increases by $+0.15$, surpassing the greedy oracle benchmark in the high-cost regime. These results show that coupling net-effectiveness evaluation with hierarchical RL yields a deployable, size-agnostic, and cost-aware resilience optimizer.