Resilient Hierarchical Power Control

• RHPC is a hierarchical control framework that combines distributed local controllers with centralized optimization to restore frequency and enable proportional power sharing, even during cyber-attacks.
• It leverages inter-layer cooperation, saturation-aware algorithms, and adaptive consensus mechanisms to enforce operational constraints and coordinate economic dispatch in microgrids.
• Empirical validations indicate RHPC can restore nominal operation within seconds, reduce power-sharing errors by up to 78%, and maintain performance under severe disturbances.

Resilient Hierarchical Power Control (RHPC) defines a class of hierarchical, multi-layer control strategies for microgrids and distributed power systems engineered for robust performance and explicit resilience against bounded or unbounded cyber-physical attacks, component failures, and severe disturbances. RHPC synthesizes multiple distributed and centralized control principles, combining inter-layer cooperation, saturation-aware optimization, and condition-adaptive algorithms. Core objectives are: (i) precise frequency restoration, (ii) proportional active- and reactive-power sharing, (iii) operational feasibility under physical constraints, (iv) optimal economic dispatch, and (v) detection, isolation, and mitigation of cyber-attacks including stealthy false-data injection (FDI), packet loss (PL), and dynamic saturation. RHPC frameworks accommodate heterogeneous device types (e.g., grid-forming/grid-following inverters), forecast uncertainties, and ensure provable system-theoretic stability and performance guarantees under both nominal and adverse conditions (Vaishnav et al., 2023, Pratap et al., 9 Dec 2025, Ding et al., 20 Jan 2026).

1. Multi-Layered RHPC Architectures

RHPC schemes operate over two or more tightly-coupled control layers, aligned to system timescales and functional scope:

• Primary Layer: Local, near-instantaneous controllers (e.g., enhanced droop control) enforce frequency–power coupling, voltage–reactive power regulation, and strict actuator or storage saturations. These run on milliseconds–seconds timescales and guarantee instantaneous power balance as well as hard constraints protection (Pratap et al., 9 Dec 2025, Ding et al., 20 Jan 2026).
  • In GFM inverters:

  $$\dot{\theta}_i = \omega_i = \omega^* - m_i P_i,\qquad |V_i| = |V^*| - n_i Q_i$$ - In GFL inverters: tracking of current reference via PLLs.

• Secondary Layer: Distributed consensus or optimization protocol for frequency restoration and power/proportional sharing. Controllers exchange local states (e.g., frequency $\omega_i$, scaled power $m_{P_i}P_i$) over a sparse communication graph and/or utilize auxiliary “virtual” network layers for resilience and detectability (Vaishnav et al., 2023, Ding et al., 20 Jan 2026).
• Tertiary/Central Layer: Economic dispatch via centralized Energy Management Systems (EMS) or SCADA, solving robust unit commitment or Optimal Power Flow (OPF) problems incorporating forecast uncertainty over slower (minutes–hours) timescales (Pratap et al., 9 Dec 2025, Ding et al., 20 Jan 2026).
• Auxiliary (Virtual) Layer (Editor’s term): An internal, non-networked mirror of the physical layer, used for control augmentation and attack detection (see Section 3).

The following table summarizes representative RHPC architectures:

Layer	Function	Example References
Primary	Fast droop / local saturation & limits	(Pratap et al., 9 Dec 2025, Ding et al., 20 Jan 2026)
Secondary	Distributed consensus/robust sharing	(Vaishnav et al., 2023, Ding et al., 20 Jan 2026)
Tertiary	EMS / economic setpoint scheduling	(Pratap et al., 9 Dec 2025, Ding et al., 20 Jan 2026)
Auxiliary	Virtual (hidden) agents for resilience	(Vaishnav et al., 2023)

2. Mathematical Models and Distributed Control Laws

Within RHPC, distributed controllers are rigorously coupled via algebraic and dynamical relationships:

2.1 Frequency and Power Sharing Control

The attack-resilient consensus in (Vaishnav et al., 2023) is governed by the coupled Σ/Π (physical/auxiliary) layer equations: $$\begin{aligned} \dot e_{\omega} &= A\,e_{\omega} + \beta A z + L d_{\omega} \ \dot z &= A\,z - \beta A e_{\omega} \end{aligned}$$ with $A = -(L+G)$, $e_{\omega} = \omega - \omega^* 1_n$, $z$ auxiliary states, and $d_{\omega}$ FDI attacks.

The power-sharing analog (leaderless) is: $$\begin{aligned} \dot e_P &= -L e_P + \beta L z + L d_P \ \dot z &= -L z - \beta L e_P \end{aligned}$$ where $e_P = m_P P - \Delta_P 1_n$.

2.2 Saturation-Aware Distributed Optimization

In hybrid GFL/GFM microgrids (Ding et al., 20 Jan 2026), the secondary layer uses “standardized power increment” consensus, with projection operators and dynamic activation: $$\bm{u}_i(k) = \mathscr{P}_{\mathcal{X}_i}\biggl[ (1-c)\bm{x}_i(k) + c\sum_{j\in\mathcal{N}_i} a_{ij}(k)\tilde{\bm{x}}_{ij}(k-\tau_{ij}) \biggr]$$ for GFL nodes, where $\mathcal{X}_i$ defines saturation bounds and $\mathscr{P}$ is the projection onto feasible states.

2.3 Economic and Robust Scheduling

EMS/SCADA solve robust unit commitment or OPF problems typically formulated as: $$\min_{\delta}\max_{w\in\Xi} \sum_{j=1}^{N_p} \ell(p(k+j),\delta(k+j),x(k+j-1))$$ subject to algebraic balance, droop-saturation, storage dynamics, and $w\in\Xi$ forecast intervals (Pratap et al., 9 Dec 2025, Ding et al., 20 Jan 2026).

3. Cyber-Physical Attack Detection and Tolerance

A central RHPC tenet is explicit resilience to cyber-attacks, especially FDI:

• Bounded, State-Dependent FDI Attacks (Vaishnav et al., 2023): System robustness is ensured even if every communication channel is corrupted by bounded, dynamic FDI modeled as $d_{\omega}(t), d_P(t)$ with $\lVert d_{\omega}(t)\rVert\leq\bar D_{\omega}$, $\lVert d_P(t)\rVert\leq\bar D_P$.
• Auxiliary Layer Aided Attack Identification: The auxiliary layer Π in (Vaishnav et al., 2023) enables side-channel estimation:

  $$\hat\omega_{ij} = \frac{1}{\beta}(\bar z_{ij} - \bar \omega_{ij}) = \omega_j$$

  By comparing $\hat\omega_{ij}$ with the possibly corrupted $\check\omega_{ij}$, the system isolates attacked links for mitigation.

• Multi-Scale Attention and LSTM Resilience (Ding et al., 20 Jan 2026): RHPC employs an attention mechanism weighting incoming data streams and LSTM-based predictors to reconstruct lost/corrupted packets, ensuring bounded residuals even under unbounded (stealthy) attacks and PL.

4. Physical and Economic Constraint Handling

RHPC frameworks address operational boundary enforcement and economic coordination:

• Saturation-Based Droop (Pratap et al., 9 Dec 2025, Ding et al., 20 Jan 2026): Nonlinear droop laws with explicit power (and state of charge for storage) bounds via

$$P_{\text{out}} = \operatorname{sat}_{[P_{\min}, P_{\max}]}\left(P_{\text{ref}} - K_d(V_{\text{meas}} - V_{\text{ref}})\right)$$

guarantee that individual units never violate actuator or storage resource constraints, irrespective of network oscillations or scheduling errors.

• Dynamic Activation/Projection in Mixed Inverter Types (Ding et al., 20 Jan 2026): In hybrid GFL/GFM microgrids, projection operators enforce GFL saturation, while an activation matrix automatically isolates saturated (constraint-bound) GFLs from secondary consensus, offloading imbalances onto the GFM backbone, thus preventing integrator windup and preserving stability of the unsaturated subnetwork.
• Economic Setpoint and Power Increment Coordination (Ding et al., 20 Jan 2026): Tertiary OPF setpoints ($P_{DG,i}$) define proportional sharing baselines; RHPC constrains secondary increments to strictly match economic ratios, regardless of saturation events or cyber disturbances.

5. Stability, Convergence, and Performance Guarantees

RHPC designs are analytically shown to achieve closed-loop boundedness and performance metrics:

• Lyapunov and Spectral Analysis (Vaishnav et al., 2023, Ding et al., 20 Jan 2026): Stability proofs use Hurwitz block matrices ($\mathcal{K}$) and quadratic Lyapunov candidates. Uniform ultimate boundedness (UUB) is demonstrated for the global error $\bm{E}(k)$:

$$\|\bm{E}(k+1)\|\leq\rho \|\bm{E}(k)\|+\Xi \implies \limsup_{k\to\infty}\|\bm{E}(k)\|\leq \frac{\Xi}{1-\rho}$$

• Resilience Scalings: Frequency and power-sharing errors are $O(1/\beta)$ in (Vaishnav et al., 2023); in practice, $\beta=2$–$5$ suffices for $\leq 1\%$ deviations. Frequency restoration and proportional sharing are rigorously bounded by graph connectivities ($\lambda_{2}(L)$) and network parameters.
• No Conservatism in Resource Utilization: The RHPC of (Pratap et al., 9 Dec 2025) achieves near-prescient operational cost (regret $\lesssim$5%), dramatically less conservative than minimax MPC in scenarios with moderate uncertainty, while never violating operational feasibility (hard constraints and storage limits satisfied under all $w\in\Xi$).

6. Case Studies and Empirical Validation

Key results from the principal works demonstrate the practical effectiveness and quantitative resilience of RHPC:

• Attack-Resilient Islanded Microgrid (Vaishnav et al., 2023): In a four-DG, 38 kW microgrid, unmitigated FDI attacks skewed frequency by up to 19 rad/s (without auxiliary layer). RHPC restored frequency to within $\pm 0.5$ rad/s in under 6 s after attack, held power-sharing error $<$3%, and detected/isolated attacked links within 0.1 s, preserving network spanning tree and nominal operation.
• Hybrid GFL/GFM Microgrid under Mixed Attacks (Ding et al., 20 Jan 2026): On the IEEE 33-bus test system subjected to unbounded FDI, PL, and delay, RHPC reduced active-increment error by 78% (from 0.41 to 0.09 grid-connected, and 0.68 to 0.15 islanded), controlled GFM overshoot (0.15 MW vs. 0.57 MW), and achieved reactive-power errors under 0.036 vs. 0.73 for traditional schemes.
• Optimal Operation under Uncertainty (Pratap et al., 9 Dec 2025): Over one week and eleven interpolated renewable/load scenarios, RHPC matched the prescient (optimistic) operational cost within 5% and outperformed minimax MPC by up to 30% in medium-uncertainty cases, with hard constraint satisfaction throughout.

7. Design and Implementation Considerations

Quantitative design guidelines for RHPC emphasize tuning of inter-layer gain ($\beta$), pinning matrix $G$, and communication graph connectivity (maximizing $\lambda_2(L)$):

• Increasing $\beta$ tightens steady-state bounds, but practical gains plateau beyond $\beta\sim5$ (Vaishnav et al., 2023).
• Ensuring high $\lambda_{\min}(L+G)$ (strong pinning/leaders) and large algebraic connectivity $\lambda_2(L)$ is mandatory for tolerance and fast convergence.
• Selection of activation/projection thresholds for saturation handling must reflect the worst expected load and forecast error envelope to guarantee persistent feasibility (Pratap et al., 9 Dec 2025, Ding et al., 20 Jan 2026).

A plausible implication is that the RHPC paradigm can be generalized to a range of distributed control networks in power and beyond, given its blend of provable constraint enforcement, cyber-resilience, and real-time feasibility. However, effectiveness is contingent on careful inter-layer design, communication topology robustness, and system identification of physical constraint sets.