In my research, I focus on the critical intersection of distributed cooperative control and information security within modern power systems. The rapid integration of renewable energy sources has profoundly transformed the architecture of conventional power grids, introducing unprecedented complexities in system operation, stability management, and cybersecurity. Among the various emerging technologies, the energy storage system has become a cornerstone for ensuring grid flexibility and reliability. Simultaneously, direct current (DC) microgrids have attracted significant research attention due to their high efficiency, seamless integration with renewable sources, and superior power quality. In this context, I concentrate on two fundamental problems: power allocation within distributed energy storage systems, and secondary voltage control in DC microgrids, with a particular emphasis on developing privacy-preserving mechanisms that can protect sensitive information without compromising control performance.
The motivation for my work stems from the inherent vulnerability of distributed control algorithms to cyber threats. In contrast to centralized control architectures where communication with a central controller can be physically safeguarded, distributed control systems necessarily rely on peer-to-peer communication networks, which expand the attack surface substantially. Both internal adversaries (curious but non-malicious nodes) and external adversaries (eavesdroppers) can intercept communication links to extract sensitive operational data, such as state-of-charge (SoC) information of batteries, output currents of converters, and load parameters. This information, if compromised, could be exploited for malicious purposes, including market manipulation, physical system damage, or targeted cyber-physical attacks. Recognizing these challenges, I develop three distinct privacy-preserving mechanisms, each tailored to address specific control objectives within the domain of energy storage systems and DC microgrids.
Problem Formulation and Preliminaries
I consider a distributed energy storage system comprising \(n\) battery units interconnected through a communication network. Each battery unit is equipped with a bidirectional Buck converter and adopts a dual closed-loop strategy incorporating an inner current loop and an outer voltage loop. The state-of-charge dynamics of the \(i\)-th battery can be represented as follows:
\[
\text{SoC}_i(t) = \text{SoC}_i(0) – \frac{\rho_i}{Q_i} \int_0^t I_i(\tau) d\tau, \quad i = 1, 2, \ldots, n,
\]
where \(Q_i\) denotes the battery capacity, \(\rho_i\) represents the coulombic efficiency, and \(I_i(t)\) is the output current of the \(i\)-th battery. In discrete-time form, this dynamics can be rewritten as:
\[
\text{SoC}_i((k+1)T) = \text{SoC}_i(kT) – \frac{\rho_i T}{Q_i V_i} P_i(kT), \quad i = 1, 2, \ldots, n.
\]
For DC microgrids, I consider \(n\) distributed generators (DGs) connected to a common DC bus through power electronic converters. The bus voltage dynamics is described by:
\[
V_B(t) = V_i^{ref}(t) – R_i I_i(t),
\]
where \(V_B(t)\) is the bus voltage, \(R_i\) denotes the line resistance, \(I_i(t)\) is the output current, and \(V_i^{ref}(t) = V^* – k_i I_i(t)\) is the reference voltage with \(k_i\) being the droop coefficient and \(V^*\) the nominal bus voltage. To compensate the voltage deviation caused by the droop control, I introduce a secondary control term \(u_i(t)\):
\[
V_B(t) = V_i^{ref}(t) – R_i I_i(t) + u_i(t).
\]
The primary challenge lies in designing the secondary controller \(u_i(t)\) to achieve the following objectives: (1) voltage restoration in finite time, (2) accurate current sharing proportional to the droop coefficients, and (3) privacy preservation of the current information against internal and external adversaries.
Minimum-Time Cooperative Control for Energy Storage Systems with Non-Broadcast Random Number Privacy Mechanism
My first contribution addresses the power allocation problem in distributed battery energy storage systems. Traditional power allocation strategies based on SoC consensus often cause certain batteries to be over-discharged or over-charged before reaching a balanced state. To overcome this limitation, I propose a minimum-time power allocation strategy based on the consensus of relative SoC variation rates. This approach fundamentally differs from conventional methods by ensuring that all batteries simultaneously reach their full-charge or full-discharge states, thereby eliminating the phenomenon of unnecessary energy circulation between batteries.
In the discharge mode, I redefine variables to simplify the system model. The concept of relative SoC variation rate is a crucial innovation in my study. For each battery, I consider the dynamics during the last sampling interval before full discharge. The essential insight is that if the relative SoC variation rates of all batteries are equalized, then all batteries will reach zero SoC simultaneously. To achieve this goal, each battery requires knowledge of the average battery state, which is a global information that must be estimated in a distributed manner.
In the discharge mode, the centralized controller is designed as:
\[
P_i(kT) = P_{set} \cdot \frac{\varphi_{di}(kT)}{\sum_{i=1}^n \varphi_{di}(kT)}, \quad i = 1, 2, \ldots, n,
\]
where \(\varphi_{di}(kT) = \frac{Q_i V_i}{\rho_i} \cdot \text{SoC}_i(kT)\). This controller guarantees that all batteries reach complete discharge simultaneously. However, the centralized structure requires global information, which motivates my distributed implementation. I rewrite the controller as:
\[
P_i(kT) = \frac{P_{set}}{n} \cdot \frac{\varphi_{di}(kT)}{\bar{\varphi}_d(kT)},
\]
where \(\bar{\varphi}_d(kT) = \frac{1}{n}\sum_{i=1}^n \varphi_{di}(kT)\) is the global average battery state. This average value must be estimated in a distributed manner, which forms the core of my algorithm design.
For the distributed estimation of the global average battery state, I propose a privacy-preserving minimum-time discrete average consensus algorithm. Table 1 summarizes the key parameters of my proposed algorithm.
| Parameter | Symbol | Value/Range |
|---|---|---|
| Number of batteries | \(n\) | 10 |
| Set power per battery | \(P_{set}/n\) | 20000 kW |
| Algorithm gain | \(\gamma\) | 100000 |
| Control gain | \(\varepsilon\) | 0.25 |
| Sampling period (estimator) | \(T_e\) | 0.03 s |
| Sampling period (controller) | \(T\) | 0.75 s |
A key innovation of my privacy-preserving consensus algorithm is the introduction of edge-based random numbers. In the initialization phase, each node generates a random number and sends it to its neighbors. These random numbers are designed to cancel in the aggregate, thus preserving the sum of initial states while simultaneously encrypting individual initial states from the perspective of adversaries. To achieve minimum-time convergence, I employ a Hankel matrix approach together with the determinantal test for detecting the convergence time.
The overall distributed control strategy is based on a multi-rate sampling mechanism. The estimator operates at a sampling period \(T_e\), while the controller operates at a higher rate. Specifically, I set \(T = (2+2n)T_e\) to ensure that the estimator has sufficient time to converge before the controller updates. The distributed controller for the \(i\)-th battery is designed as:
\[
P_i(t) = \begin{cases}
\frac{P_{set}}{n}, & kT \leq t < kT + (2+2n)T_e, \\
P_{set} \cdot \frac{\varphi_{di}(kT)}{n \cdot \hat{x}_{di}(kT + (2+2n)T_e)}, & kT + (2+2n)T_e \leq t < (k+1)T + (2+2n)T_e,
\end{cases}
\]
where \(\hat{x}_{di}\) is the estimated average battery state obtained from the privacy-preserving consensus algorithm. I prove in my theoretical analysis that this distributed control strategy achieves both control objectives within minimum time while simultaneously protecting privacy.
Convergence Analysis
My convergence analysis establishes the fundamental property that the algorithm achieves accurate convergence within a bounded number of iterations. The central theorem of this analysis asserts that the system state converges to the average of its initial values within \(2n+1\) steps. This result is achieved through a careful decomposition of the convergence process, where I first verify that the privacy-preserving mechanism does not alter the sum of initial values. The proof leverages the inherent structure of the consensus algorithm combined with the properties of Hankel matrices. Through the implementation of the multi-rate sampling mechanism, I establish that when the controller sampling period \(T\) is set to \((2+2n)T_e\), the estimated value is precisely available at the moment when the controller requires it.
Privacy Performance Analysis
In my analysis of privacy preservation, I quantify the estimation error that adversaries can achieve. Through rigorous mathematical analysis, I demonstrate that if a target node has at least one neighbor that does not collude with the curious node, the estimation range of the curious node for the target node’s initial state extends to the entire real line \((-\infty, +\infty)\). For the external eavesdropper, I prove that the inability to observe the algorithm parameter \(\gamma\) makes it impossible to estimate the private information accurately. The simulation results presented in my work confirm these theoretical findings, demonstrating that the estimation error of adversaries is arbitrarily large.
I also conduct a comparative analysis in my research, evaluating my method against differential privacy and state decomposition approaches. The differential privacy mechanism introduces noise that compromises convergence precision, whereas my method achieves both precise consensus and minimum-step convergence. More importantly, the state decomposition method requires secret communication between nodes to adjust initial states, which is impractical for large-scale distributed systems. My non-broadcast random number mechanism does not require such secret communication, making it more applicable to practical energy storage system configurations.
Finite-Time Secondary Control for DC Microgrids with Node Decomposition Privacy Mechanism
My second contribution focuses on the secondary control of DC microgrids, where I address the dual challenges of finite-time convergence and privacy preservation in the presence of constant power loads (CPLs). In my second research thrust, I focus on the critical challenge of voltage restoration in DC microgrids with CPLs. The presence of CPLs introduces nonlinear negative impedance characteristics that significantly degrade system stability margin. I design a node decomposition privacy-preserving mechanism where each node divides itself into multiple sub-nodes, with the number of sub-nodes serving as a secret key embedded in the control law. This decomposition must be compensated in the secondary controller because it artificially inflates the topology size that appears in the consensus calculations.
Node Decomposition Privacy-Preserving Mechanism
In the node decomposition mechanism, the original communication topology \(\mathcal{G}\) is expanded into a new topology \(\tilde{\mathcal{G}}\) by decomposing each node \(v_i\) into \(n_i\) sub-nodes \(\{v_{i1}, v_{i2}, \ldots, v_{in_i}\}\), where \(N = \sum_{i=1}^n n_i\) denotes the total number of sub-nodes. I treat each sub-node of the original communication graph as a generator node. The primary challenge introduced by this decomposition is twofold: (1) each node must maintain the internal states of all its sub-nodes consistently, and (2) the controller must compensate for the fact that the topology size is now \(N\) rather than \(n\). To address these challenges, I design two distributed estimation algorithms to estimate the size of the original topology \(\mathcal{G}\) and the decomposed topology \(\tilde{\mathcal{G}}\).
Building on these estimates, I design a finite-time secondary controller based on a dynamic average consensus algorithm. The key challenge is to ensure that the privacy protection provided by node decomposition does not compromise the control accuracy. I achieve this by incorporating the estimated topology sizes as correction factors in the controller design.
Finite-Time Consensus Algorithm
For the distributed estimation of topology size, I design Algorithm 3.1, which leverages the property that the convergence time of a consensus algorithm on an undirected connected graph is related to the graph structure. By initializing the algorithm with random values and monitoring the rank of the Hankel matrix, each node can determine the iteration count at which consensus is achieved. This information, combined with a consensus-based maximization step, allows all nodes to estimate the topology size in finite time. My first algorithm uses the following iteration:
\[
\theta_i(k+1) = \theta_i(k) – \varepsilon \sum_{j \in \mathcal{N}_i} (\theta_i(k) – \theta_j(k)),
\]
with the Hankel matrix test
\[
\Gamma(\Theta_i[0,1,\ldots,k]) = \begin{bmatrix}
\theta_i(1) – \theta_i(0) & \cdots & \theta_i(k) – \theta_i(k-1) \\
\theta_i(2) – \theta_i(1) & \cdots & \theta_i(k+1) – \theta_i(k) \\
\vdots & & \vdots \\
\theta_i(k) – \theta_i(k-1) & \cdots & \theta_i(2k) – \theta_i(2k-1)
\end{bmatrix}.
\]
When the rank of this Hankel matrix becomes deficient, the node records the current iteration count. By exchanging these counts through a max-consensus procedure, all nodes can estimate \(\hat{n}\) in at most \(3n-1\) iterations. For the second algorithm addressing the decomposed topology, I incorporate a primality testing mechanism to verify the correctness of the estimation. A key novel element is the inclusion of a prime number \(p\) in the initialization, which enables a distributed primality test. Based on the theoretical analysis, I prove that for the decomposed topology \(\tilde{\mathcal{G}}\), this method can identify a node with the desired spectral property and estimate the total number of sub-nodes \(N\) in finite time. The estimated value satisfies \(\xi_{ih}(k)/2 = N\) for all \(k \geq 3n_{max} \cdot n – 1\).
Secondary Controller Design
The secondary controller incorporating the privacy-preserving mechanism is designed in Algorithm 3.3. For each sub-node \(v_{ih}\), the virtual voltage drop is defined as \(\varphi_{ih}(t)\). The key innovation here is the introduction of a non-broadcast continuous smooth signal chosen randomly by each sub-node, which adds an additional layer of privacy protection. The controller structure involves auxiliary variables \(z_{ih}\), \(x_{ih}\), and \(\psi_{ih}\), with the following update laws:
\[
\begin{aligned}
\dot{z}_{ih}(t) &= \alpha \gamma \sum_{v_{il} \in \tilde{\mathcal{N}}_{ih}} \left[ \text{sgn}(x_{il}(t) – x_{ih}(t)) + \text{sgn}(x_{il}(t) – x_{ih}(t)) \right], \\
\dot{\psi}_{ih}(t) &= \sum_{v_{il} \in \tilde{\mathcal{N}}_{ih}} r_{ih,il}(t) + \varphi_{ih}(t) + \sum_{v_{il} \in \tilde{\mathcal{N}}_{ih}} (r_{ih,il}(t) – r_{il,ih}(t)), \\
x_{ih}(t) &= z_{ih}(t) + \psi_{ih}(t).
\end{aligned}
\]
Where \(r_{ih,jm}(t)\) represents a randomly selected non-broadcast continuous smooth signal generated by node \(v_{ih}\) for the communication link with node \(v_{jm}\). This signal is designed to cancel when aggregated across all nodes, thereby not affecting the final consensus value while significantly enhancing the privacy protection against malicious internal nodes.
In my theoretical analysis, I prove that under the condition \(2\gamma – 2N\sigma – 2\gamma\varepsilon \geq \alpha\lambda\), the bus voltage converges to the nominal value in finite time:
\[
t^* = \frac{1}{2\gamma} \ln\left( \sqrt{2\gamma} \|e(0)\| + 1 \right),
\]
where \(e(0) = x(0) – \psi(0)\mathbf{1}_N\). My Lyapunov-based proof demonstrates that \(V(t) = \frac{1}{2}\|e(t)\|^2\) satisfies the differential inequality \(\dot{V}(t) \leq -2\gamma V(t) – \sqrt{2\gamma}\sqrt{V(t)}\), which yields the finite-time convergence result.
The privacy protection of my method is based on a critical insight: the number of sub-nodes \(n_i\) created by each node is the secret parameter that cannot be accurately estimated by adversaries. My proofs demonstrate that both malicious internal nodes and external eavesdroppers lack the necessary information to determine \(n_i\), as shown in Table 2.
| Mechanism | Convergence Speed | Privacy | Communication Overhead | Topology Requirement |
|---|---|---|---|---|
| Gaussian Noise | Asymptotic | Guaranteed | Low | None |
| State Decomposition | Asymptotic | Requires secret comm. | Low | Loose |
| Differential Privacy | Randomized | Guaranteed | Low | None |
| Node Decomposition | Finite-time | Guaranteed | Medium | None |
| Non-broadcast Random Number | Minimum-time | Guaranteed | Medium | Loose |
Adaptive Cooperative Control for DC Microgrids with Gaussian White Noise Privacy Mechanism
My third major contribution addresses the stability of DC microgrids with constant power loads. Unlike the previous finite-time approach, this method focuses on ensuring stability under various load conditions while preserving privacy. The central challenge is that CPLs exhibit negative incremental impedance, which can destabilize the microgrid if the total load exceeds a critical threshold. My approach is to design an adaptive secondary controller that can estimate the load parameters in real-time and adjust the control strategy accordingly, thereby ensuring stability even under varying load conditions. For the distributed current sum estimation, I leverage a new insight: a Hankel matrix-based termination test. In my fourth algorithm, the estimators for the unknown parameters \(\theta_i(k) = [C \quad Y_L(k-1) \quad P(k-1)]^T\) are updated using the recursive least squares method with a forgetting factor.
System Model and Problem Formulation
I consider a DC microgrid where multiple DC voltage sources and converters are connected to a common DC bus. The mathematical model of the microgrid with constant power loads can be expressed as a set of nonlinear differential equations. To stabilize this nonlinear system, I first design a centralized adaptive secondary controller that requires global information about the load:
\[
V_i^{ref}(t) = V^* + a_m L_i k_i \left( V^* Y_L + P \right) – R_i I_i(t) – a_m L_i I_i(t),
\]
where \(a_m\) is the control gain, \(Y_L\) is the load admittance, \(P\) is the constant power load rating, \(R_i\) the line resistance, \(L_i\) the line inductance, and \(I_i(t)\) the output current. The stability analysis of this centralized controller is presented. I analyze the equilibrium point of the \((n+1)\)-dimensional system and prove that the controller guarantees stability under specific conditions. When the CPL power is relatively small relative to the resistive load, the sufficient condition becomes simple: the nominal voltage squared times the load admittance should exceed the CPL power. In this case, selecting a positive control gain \(a_m\) ensures stability. When the CPL power is larger, a more restrictive condition emerges involving the control gain and the induction parameters.
Distributed Parameter Estimation
To implement the control in a distributed manner, I design a finite-time distributed current summation consensus algorithm that can estimate the total current in the system while preserving privacy. The key idea is to add Gaussian white noise to the current measurements before broadcasting them to neighbors, preventing sensitive information leakage. The convergence of this summation algorithm is established through matrix analysis. I prove that with the right gain matrix, the algorithm converges to the correct sum value within a finite number of iterations.
Building on this summation result, I adopt a distributed recursive least squares method with a forgetting factor for load parameter identification. The estimation algorithm takes place in each iteration \(k\). The forgetting factor allows the algorithm to track time-varying load parameters, which is critical for practical scenarios where loads may change. The main iteration of the recursive least squares method used in my work is:
\[
\hat{\theta}_i(kT) = \hat{\theta}_i((k-1)T) + K_i(kT) \left( z_i(kT) – h_i^T(kT)\hat{\theta}_i((k-1)T) \right),
\]
with the gain and covariance updates:
\[
K_i(kT) = \frac{Q_i(k-1)T h_i(kT)}{\Lambda(kT) + h_i^T(kT) Q_i((k-1)T) h_i(kT)}, \quad Q_i(kT) = \left( I – K_i(kT) h_i^T(kT) \right) Q_i((k-1)T),
\]
I prove that this distributed identification algorithm converges to the true load parameters in the sense that the estimation error decays asymptotically. This result is significant because it shows that even with Gaussian white noise added for privacy purposes, the identification algorithm can still accurately estimate the load parameters.
Stability Analysis of the Distributed Adaptive Controller
Based on these estimates, I design a distributed adaptive secondary controller that stabilizes the DC microgrid with CPLs. The controller structure is analogous to the centralized case, but with estimated load parameters replacing true values. For the closed-loop system under the distributed adaptive controller, I explore a condition. The stability criterion I derive involves only local information and the estimated load parameters. First, I analyze the case where the estimated CPL power is at most equal to the nominal voltage squared times the estimated load admittance. In this case, the system is stable for any positive control gain. For the complementary case where the estimated CPL power is larger, I derive a more restrictive condition involving the control gain:
\[
\frac{1}{V^{*2}} \left( \hat{P} – V^{*2} \hat{Y}_L \right) \cdot \frac{L}{a_m} < \frac{1}{C} \sum_{i=1}^n \frac{1}{L_i} < \frac{1}{V^{*2}} \left( \hat{P} – V^{*2} \hat{Y}_L \right) \cdot \frac{C}{a_m}.
\]
This condition establishes the distributed stability criterion for DC microgrids with CPLs. The interpretation of this condition is that the combined effect of line inductances and capacitance must fall within a certain range to guarantee stability. This is the first distributed stability criterion for such systems, allowing each generator to independently verify whether the current load satisfies the stability condition using only local information and estimated parameters.
| Parameter | Symbol | Value |
|---|---|---|
| DC voltage source | \(V_{DCi}\) | 100 V |
| Inductance | \(L_i\) | 10 mH |
| Capacitance | \(C_i\) | 200 μF |
| Line resistance | \(R_i\) | 0.1 Ω |
| Control gain | \(a_m\) | 1 |
| Resistive load | \(R_L\) | 10 Ω (initial) |
| CPL power | \(P\) | 500 W (initial) |
| Nominal voltage | \(V^*\) | 50 V |
| Droop coefficient | \(k_i\) | [0.05, 0.15, 0.20, 0.25, 0.35] |
In my privacy-preserving approach, the current information is protected by adding Gaussian white noise before transmission. I prove that both internal and external adversaries cannot accurately estimate the true current information through the communication channels. The variance of the noise serves as a measure of privacy protection, and larger variances provide stronger protection. However, I note that there is a trade-off between privacy protection and control performance, since excessive noise can slow down the convergence of the parameter identification algorithm.
My simulation results demonstrate the effectiveness of the proposed approach. The parameter identification algorithm accurately estimates load parameters despite the presence of Gaussian white noise, as shown in Figures (not reproduced here). The bus voltage converges to the nominal value quickly and precisely, while the current sharing is realized according to the droop coefficients. The privacy protection performance is validated by showing that adversaries cannot extract the true current information from the noisy measurements. The step load tests reveal that the aging factor enables the distributed recursive least squares algorithm to track time-varying loads effectively, a significant advantage over fixed-load assumptions. This adaptability is critical for practical DC microgrids where loads routinely fluctuate. The hot-swap tests further demonstrate the robustness of the method to communication topology changes, confirming that it does not require dedicated communication feedback from the DC bus voltage.
Comparison of Privacy-Preserving Approaches
In my research, I systematically compare the three privacy-preserving mechanisms I developed, as summarized in the following table. Each method exhibits distinct characteristics regarding convergence speed, privacy guarantees, and applicability:
| Feature | Non-broadcast Random Numbers | Node Decomposition | Gaussian White Noise |
|---|---|---|---|
| Control Scenario | Battery Energy Storage | DC Microgrid Secondary | DC Microgrid Adaptive |
| Convergence Speed | Minimum-time | Finite-time | Asymptotic |
| Privacy Guarantee | Internal & External | Internal & External | Internal & External |
| Communication Overhead | Medium | Medium | Low |
| Topology Requirements | Loose | None | None |
| Impact on Accuracy | None | None | None |
The key insight gained from this comprehensive study is that the choice of privacy-preserving mechanism must be tailored to the specific requirements of the control application. For power allocation in energy storage systems, where fast convergence is critical, the non-broadcast random number mechanism provides the optimal balance between speed and privacy. For voltage control in DC microgrids with strict time constraints, the node decomposition mechanism offers finite-time convergence without compromising accuracy. For adaptive control scenarios with time-varying loads, the Gaussian white noise mechanism provides a simpler implementation with sufficient asymptotic convergence, while the noise variance can be tuned to achieve the desired privacy level.
Conclusions and Future Work
In conclusion, my research addresses the critical intersection of distributed cooperative control and privacy preservation in energy storage systems and DC microgrids. Through the development of three distinct privacy-preserving mechanisms, I demonstrate that it is possible to achieve both high-performance control and strong privacy guarantees simultaneously. My minimum-time power allocation strategy for battery energy storage systems achieves precise power distribution without causing over-discharging of individual batteries, while the privacy-preserving mechanism ensures that sensitive SoC information remains confidential. This approach not only validates the feasibility of such a mechanism in a realistic storage system context but also highlights its superiority over conventional differential privacy and state decomposition approaches. For DC microgrid secondary control, my finite-time converging controller successfully balances the inherent conflict between precise voltage regulation and privacy preservation. I establish that through careful design of the node decomposition mechanism and finite-time topology estimation, accurate voltage compensation can be achieved in finite time without disclosing any sensitive current information to adversaries.
Looking ahead, I identify several promising directions for future research. First, the development of privacy-preserving mechanisms that can withstand super adversaries possessing both internal and external capabilities remains an open challenge. Existing methods typically rely on hiding certain algorithm parameters, which can potentially be estimated by such powerful adversaries. Therefore, exploring methods based on independent noise injection may offer a more robust solution. Second, the standardization of privacy performance metrics across different mechanisms is essential to facilitate systematic comparison and practical deployment in industrial settings. The research community would greatly benefit from unified evaluation standards for privacy-preserving distributed control algorithms. Third, extending the stability analysis to include large-signal stability for DC microgrids with CPLs would significantly enhance the practical applicability of the proposed methods. While small-signal analysis provides local stability guarantees, large-signal stability analysis would address global stability concerns that are critical for real-world applications involving large disturbances.
In summary, my work contributes to the foundations of privacy-preserving distributed control for energy storage systems and DC microgrids. As the transformation towards renewable energy accelerates, the importance of secure and reliable control systems cannot be overstated. I believe that the mechanisms developed in my research will serve as valuable building blocks for the next generation of smart grid control systems, particularly in the context of large-scale battery energy storage system deployment and distributed real-time optimization of power networks. The integration of privacy-preserving primitives into mainstream power electronic control architectures will be crucial for the transition to fully decentralized smart grid architectures, ensuring that the benefits of distributed control can be fully realized without compromising the safety and security of the overall electrical infrastructure.