Island microgrid frequency safety prediction control method based on privacy calculation

Abstract

The present application relates to the technical field of power system automation and information security, and discloses a frequency safety prediction control method for island micro-grid based on privacy calculation. The method comprises the following steps: collecting encrypted load feature vectors of each node and performing secure aggregation based on homomorphic encryption and differential privacy; constructing a dynamic frequency response model and embedding the aggregated disturbance as input; using extended Kalman filtering to estimate the frequency deviation prediction sequence online; establishing a rolling optimization problem with the joint minimization of frequency deviation and regulation cost as the target, and solving it distributedly through the alternating direction multiplier method, only exchanging encrypted intermediate variables; and finally generating power regulation instructions to realize closed-loop control. The present application improves the frequency stability and operation economy of island micro-grid while ensuring the privacy and security of users.

Description

Technical Field

[0001] This invention belongs to the interdisciplinary field of power system automation and information security, specifically relating to a frequency security prediction and control method for isolated microgrids based on privacy computing. Background Technology

[0002] As energy systems rapidly evolve towards distributed and intelligent architectures, islanded microgrids, as a key form for improving power supply reliability and renewable energy absorption capacity, are receiving increasing attention in remote areas, emergency power supply, and critical infrastructure protection scenarios. Islanded microgrids operate independently from the main grid, lacking external inertia support, and their frequency stability is highly dependent on the dynamic balance of internal power sources and loads. Sudden load changes or fluctuations in distributed power output can easily lead to frequency exceedances or even system collapse. Therefore, accurate prediction and rapid coordinated control of frequency security have become crucial for ensuring the stable operation of islanded microgrids.

[0003] Model predictive control (MPC)-based frequency regulation methods have been widely studied due to their forward-looking and optimization capabilities. These methods predict future frequency trajectories by constructing a dynamic system model and solving for the optimal control sequence online to coordinate various frequency regulation resources, including energy storage, diesel generators, and controllable loads. Traditional MPC architectures face the following problems in islanded microgrid applications: model accuracy depends on high-dimensional, high-frequency user electricity consumption data and environmental covariates, which are distributed across multiple user-side nodes; direct centralized modeling would lead to the leakage of sensitive privacy information. Existing predictive control strategies often employ fixed model structures, making it difficult to adapt to the dynamic evolution of user behavior and sudden changes in operating conditions, resulting in response lag, coordination mismatch, and reduced frequency safety margins.

[0004] While existing technologies attempt to improve prediction accuracy by introducing data-driven methods, they neglect the collaborative design of privacy protection and control optimization. Some solutions employ data anonymization or local modeling, sacrificing the predictive performance gained through global information fusion; others focus on encrypted computation but fail to deeply embed privacy mechanisms into the entire process of frequency dynamic modeling and optimization, resulting in high computational complexity and poor real-time performance. In islanded microgrids with multiple stakeholders, how to achieve secure alignment of user behavior characteristics, privacy-preserving approximation of frequency differential equations, and collaborative generation of frequency modulation commands without exchanging raw data remains an unsolved problem. A new paradigm that deeply integrates privacy-preserving computation and predictive control is urgently needed to achieve high-precision prediction of frequency trajectories and efficient collaborative optimization of multiple resources while ensuring data sovereignty and privacy security. Summary of the Invention

[0005] This invention provides a frequency security prediction and control method for islanded microgrids based on privacy computing. By constructing a distributed multi-agent collaborative modeling architecture, it uses homomorphic encryption and differential privacy fusion mechanisms to securely aggregate user-side load characteristics without exchanging original electricity consumption data. Combined with a dynamic frequency response model and rolling optimization strategy, it achieves high-precision prediction of the frequency trajectory of the islanded microgrid and collaborative control of multiple types of distributed resources. This improves system frequency stability and operational economy while ensuring user electricity privacy and security.

[0006] This invention provides a frequency security predictive control method for islanded microgrids based on privacy computing, comprising: Acquire real-time output status data of each distributed power unit, state of charge and charging / discharging power limits of energy storage devices, and encrypted load feature vectors of each user node; Based on the homomorphic encryption protocol, the encrypted payload feature vector is securely aggregated to generate a global payload perturbation trend representation. A dynamic frequency response differential equation model is constructed that includes system inertia, damping coefficient and primary frequency modulation gain, and the global load disturbance trend characterization is embedded as an external disturbance input term into the dynamic frequency response differential equation model. The extended Kalman filter algorithm is used to estimate the state variables of the dynamic frequency response differential equation model online, and the frequency deviation prediction sequence at the current moment is obtained. Based on the frequency deviation prediction sequence, the state of charge constraint of the energy storage device, and the ramp-up rate limit of the distributed power source, a rolling optimization problem is constructed with the joint objective of minimizing the square integral of frequency deviation and minimizing the control cost. In solving the rolling optimization problem, the Lagrange multiplier method is introduced to handle coupling constraints, and the global optimization problem is decomposed into multiple local subproblems by the alternating direction multiplier method. Each participating entity only needs to exchange encrypted intermediate variables without disclosing the original data. Based on the solution to the rolling optimization problem, active power adjustment commands for each distributed power source and charging / discharging power commands for the energy storage device are generated and sent to the corresponding execution units for closed-loop control.

[0007] Preferably, the encrypted payload feature vector of each user node is obtained, including: Each user node deploys a local edge computing unit to collect its historical electricity load sequence; The historical electricity load sequence is sliced ​​using a sliding window, and four statistical features are extracted from each window: mean, variance, peak-to-valley difference, and rate of change. The four statistical features are concatenated into a 4-dimensional feature vector, and the Paillier homomorphic encryption algorithm is used to encrypt the 4-dimensional feature vector element by element to generate an encrypted payload feature vector. The encrypted load feature vector is uploaded to the microgrid central coordinator.

[0008] Preferably, the secure aggregation of the encrypted payload feature vector based on a homomorphic encryption protocol includes: The microgrid central coordinator receives encrypted load feature vectors uploaded by all user nodes; By utilizing the additive homomorphic property of Paillier homomorphic encryption, the encrypted values ​​of the same feature dimension are multiplied one by one to obtain the encrypted aggregate value of the same feature dimension. Perform the above aggregation operation on each of the four feature dimensions to form a 4-dimensional encrypted aggregation vector; A Laplace noise mechanism is applied to the 4D encrypted aggregation vector, and the noise scale parameter is determined based on the global sensitivity and the preset privacy budget to generate a perturbed encrypted aggregation vector that meets the differential privacy requirements. After decrypting the perturbation encryption aggregation vector, divide it by the total number of user nodes to obtain a noisy global load perturbation trend representation.

[0009] Preferably, the dynamic frequency response differential equation model is expressed as: ; For frequency deviation, Let be the system's equivalent inertia constant. The damping coefficient is... This is the total mechanical power adjustment amount. This represents the total load disturbance. For time, the global load disturbance trend characterization is used to construct The predicted value within the future time window.

[0010] Preferably, the extended Kalman filter algorithm is used to estimate the state variables of the dynamic frequency response differential equation model online, including: Define the state vector as , The rate of change of frequency deviation; The process noise covariance matrix is ​​set to a diagonal matrix.

[0011] Setting observation noise covariance

[0012] With a sampling period of 100 milliseconds, a prediction-correction loop is executed to output the frequency deviation prediction sequence at the current moment.

[0013] Preferably, a rolling optimization problem is constructed with the joint objective of minimizing the integral of the squared frequency deviation and minimizing the control cost, including: Set the prediction time domain ; Set frequency deviation weighting coefficient Adjustment cost weighting coefficient ; Define control input vector Includes the power adjustment of each distributed power source and the charging and discharging power of energy storage; Configure diagonal positive definite weight matrix Its diagonal elements are set according to the unit control cost of each resource.

[0014] Preferably, the conditions of the coupling constraint include: The energy storage's state of charge is within a safe range of 20% to 80%. The power regulation rate of distributed power sources shall not exceed 10% of the rated power per minute; The total regulation power meets the regulation requirements of the frequency deviation prediction sequence; The power output of each resource does not exceed its physical limit.

[0015] Preferably, in solving the rolling optimization problem, by alternating direction multiplication... The method decomposes the global optimization problem into multiple local subproblems, including: Set a penalty factor; Set the maximum number of iterations; Set a convergence threshold; The national cryptographic algorithm SM4 is used to encrypt the transmission of the exchanged Lagrange multipliers and dual variables during the iteration process.

[0016] Preferably, the active power adjustment command for each distributed power source and the discharge power command for the energy storage device are generated and sent to the corresponding execution unit for closed-loop control, including: The central coordinator uses a private key to digitally sign the instruction content and generate a verification code. Each execution unit executes the instruction after verifying its legitimacy using the corresponding public key.

[0017] Preferably, the system equivalent inertia constant in the dynamic frequency response differential equation model is... With damping coefficient The frequency response data is updated online using the least squares method.

[0018] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention constructs a privacy computing framework based on the fusion of homomorphic encryption and differential privacy, which realizes the secure aggregation of user load characteristics, avoids the direct exposure of raw electricity consumption data, and solves the risk of data privacy leakage in multi-subject collaborative modeling.

[0019] 2. This invention uses the aggregated global load disturbance trend as input to the dynamic frequency response model, and combines it with extended Kalman filtering for state estimation, thereby improving the accuracy and timeliness of frequency trajectory prediction.

[0020] 3. The distributed solution mechanism combining rolling optimization and alternating direction multiplier method designed in this invention achieves coordinated optimization and control of multiple types of distributed resources under the premise of satisfying the energy storage state of charge constraints and power supply ramp-up rate limits, overcoming the single-point failure risk and communication problems of traditional centralized control.

[0021] 4. While ensuring the privacy and security of user data, this invention achieves accurate prediction and efficient control of frequency security in isolated microgrids, thereby improving the system's operational stability, security, and economy. Attached Figure Description

[0022] Figure 1 This is a schematic diagram of the overall technical solution architecture of the present invention; Figure 2 This is a schematic diagram of the core principle framework of secure aggregation of user load features under the homomorphic encryption and differential privacy fusion mechanism in this invention; Figure 3 This is a flowchart illustrating the logical flow of dynamic frequency response modeling and extended Kalman filtering joint prediction in this invention. Figure 4 This is a logical flow diagram of the rolling optimization and distributed resource collaborative regulation in this invention; Figure 5 This is a schematic diagram of the interactive relationship and data flow of the distributed solution of the alternating direction multiplier method involving multiple entities in this invention; Figure 6 This is a schematic diagram of the end-to-end privacy protection closed-loop control data flow from edge acquisition to instruction execution in this invention. Detailed Implementation

[0023] refer to Figures 1 to 6 This invention provides a frequency security prediction and control method for isolated microgrids based on privacy computing. It constructs a distributed multi-agent collaborative modeling architecture and, without exchanging original electricity consumption data, securely aggregates user-side load characteristics through a fusion mechanism of homomorphic encryption and differential privacy. Combined with a dynamic frequency response model and rolling optimization strategy, it achieves high-precision prediction of the frequency trajectory of the isolated microgrid and collaborative control of multiple types of distributed resources. This frequency security prediction and control method for isolated microgrids improves system frequency stability and operational economy while ensuring user electricity consumption privacy.

[0024] The method includes the following steps: Acquire real-time output status data of each distributed power unit, state of charge and charging / discharging power limits of energy storage devices, and encrypted load feature vectors of each user node; Based on the homomorphic encryption protocol, the encrypted payload feature vector is securely aggregated to generate a global payload perturbation trend representation. A dynamic frequency response differential equation model is constructed that includes system inertia, damping coefficient and primary frequency modulation gain, and the global load disturbance trend characterization is embedded as an external disturbance input term into the dynamic frequency response differential equation model. The extended Kalman filter algorithm is used to estimate the state variables of the dynamic frequency response differential equation model online, and the frequency deviation prediction sequence at the current moment is obtained. Based on the frequency deviation prediction sequence, the state of charge constraint of the energy storage device, and the ramp-up rate limit of the distributed power source, a rolling optimization problem is constructed with the joint objective of minimizing the square integral of frequency deviation and minimizing the control cost. In solving the rolling optimization problem, the Lagrange multiplier method is introduced to handle coupling constraints, and the global optimization problem is decomposed into multiple local subproblems by the alternating direction multiplier method. Each participating entity only needs to exchange encrypted intermediate variables without disclosing the original data. Based on the solution to the rolling optimization problem, active power adjustment commands for each distributed power source and charging / discharging power commands for the energy storage device are generated and sent to the corresponding execution units for closed-loop control.

[0025] The system acquires real-time output status data for each distributed power unit, the state of charge (SOC) and charge / discharge power limits of the energy storage device, and encrypted load feature vectors for each user node. Each distributed power unit includes, but is not limited to, photovoltaic inverters, wind power converters, and diesel generators. Its real-time output status data is uploaded every 100 milliseconds via a built-in communication interface. The data includes current active power output, reactive power output, voltage amplitude, current phase angle, and device operating status flags. The energy storage device includes lithium battery packs, supercapacitors, or hybrid energy storage systems. Its SOC is monitored in real-time by a battery management system with a sampling period of 100 milliseconds. The system also records the current maximum allowable charging power and maximum allowable discharging power, with limits determined by battery health, temperature, and remaining capacity.

[0026] Each user node deploys a local edge computing unit, which collects its historical electricity load sequence at a sampling frequency of 10 times per second, forming continuous time-series data. The historical electricity load sequence is sliced ​​into sliding windows with a window length of 60 seconds and a step size of 10 seconds. Within each window, four statistical features are extracted: mean, variance, peak-to-valley difference, and rate of change. The mean is the arithmetic mean of all sampling points within the window; the variance is the average of the squares of the differences between each sampling point and the mean; the peak-to-valley difference is the difference between the maximum and minimum values ​​within the window; and the rate of change is the ratio of the load increment to the time interval in the last 10 seconds of the window. These four statistical features are concatenated into a 4-dimensional feature vector, which is then encrypted element-by-element using the Paillier homomorphic encryption algorithm to generate an encrypted load feature vector. The public key for the Paillier homomorphic encryption algorithm is uniformly generated and distributed to each user node by the microgrid central coordinator, while the private key is securely stored by the central coordinator. The encrypted 4-dimensional feature vector is uploaded to the microgrid central coordinator via a secure communication channel at a frequency of once per minute.

[0027] Based on a homomorphic encryption protocol, the encrypted load feature vectors are securely aggregated to generate a global load disturbance trend representation. The microgrid central coordinator receives the encrypted load feature vectors uploaded by all user nodes, assuming there are... For each user node, each feature dimension corresponds to The first encrypted value. Utilizing the additive homomorphic property of Paillier homomorphic encryption, all encrypted values ​​of the same feature dimension are multiplied term-by-term to obtain the encrypted aggregate value of that dimension. For the first... Each feature dimension, its encrypted aggregate value This is the product of encrypted values ​​for the same feature dimension uploaded by all user nodes. The above aggregation operation is performed on each of the four feature dimensions to form a 4-dimensional encrypted aggregate vector. A Laplacian noise mechanism is then applied to the 4-dimensional encrypted aggregate vector, with the noise scale parameter... Based on global sensitivity With preset privacy budget Confirmed, satisfied Global sensitivity Defined as the maximum change in the aggregation result caused by a change in the feature vector of a single user node, the global sensitivity is calculated for the four features—mean, variance, peak-to-valley difference, and rate of change—as the maximum load value, the square of the maximum load value, twice the maximum load value, and the maximum load rate of change, respectively. Preset privacy budget. The value is set to 0.5 to balance the strength of privacy protection with data availability.

[0028] Laplace noise is implemented by randomly perturbing the aggregated value in the encrypted domain. Specifically, the encrypted aggregated value is multiplied by an encrypted Laplace noise sample, which is generated by the central coordinator before encryption and added after encryption. After generating a perturbed encrypted aggregated vector that meets differential privacy requirements, the central coordinator decrypts the perturbed encrypted aggregated vector using its private key, obtaining a noisy 4-dimensional plaintext aggregated vector. Dividing this perturbed encrypted aggregated vector by the total number of user nodes N yields a noisy global load perturbation trend representation. This global load perturbation trend representation includes four dimensions: the mean trend of the global load, fluctuation intensity, peak-valley characteristics, and rate of change, which are used as perturbation inputs for subsequent frequency prediction models.

[0029] A dynamic frequency response differential equation model is constructed, incorporating system inertia, damping coefficient, and primary frequency modulation gain. The global load disturbance trend is then embedded as an external disturbance input term into this dynamic frequency response differential equation model. The dynamic frequency response differential equation model is expressed as follows: ; Frequency deviation is defined as the difference between the actual frequency and the rated frequency of 50 Hz. The system's equivalent inertia constant, expressed in seconds, reflects the system's ability to resist frequency changes. The damping coefficient, measured in megawatts per hertz, characterizes the natural response of a load to frequency variations. The total mechanical power regulation is provided by the primary frequency regulation action of the distributed power source and the rapid response of the energy storage device. This represents the total load disturbance, caused by random fluctuations in user load. For time. The global load disturbance trend characterization is used to construct The predicted value within the future time window. Specifically, the mean trend component in the global load disturbance trend representation is used as... The baseline value, fluctuation intensity component is used to define the variance range of the disturbance, peak-to-valley difference component is used to identify potential large disturbance events, and change rate component is used to extrapolate the load change trend over the next 5 seconds. A method combining linear extrapolation and Gaussian process regression is used to generate one sampling point every 100 milliseconds for the next five seconds. The predicted sequence is used as an external perturbation input to the dynamic frequency response model.

[0030] The extended Kalman filter algorithm is used to estimate the state variables of the dynamic frequency response differential equation model online, obtaining the frequency deviation prediction sequence at the current time. The state vector of the extended Kalman filter algorithm is defined as follows: , Let be the rate of change of frequency deviation. The state transition equation is obtained by discretizing the dynamic frequency response differential equation using the first-order Euler method, with a discretization time step of 100 milliseconds. The process noise covariance matrix is ​​set as a diagonal matrix, with its diagonal elements being... and , corresponding to the model uncertainties of frequency deviation and frequency change rate, respectively. The observation equation is a direct measurement of frequency deviation, and the observation noise covariance is set to . This reflects the accuracy of the frequency measurement device. The extended Kalman filter algorithm executes once every 100 milliseconds, and its inputs include the current frequency measurement value, the state estimate value from the previous moment, the process noise covariance, the observation noise covariance, and the values ​​for the next 5 seconds. Predicted Sequence. Through a predict-correction loop, the optimal state estimate for the current moment is output, and the frequency deviation sequence for the next 5 seconds is predicted forward based on the state transition equation, forming the frequency deviation prediction sequence.

[0031] Based on the frequency deviation prediction sequence, the state-of-charge constraints of the energy storage device, and the ramp-up rate limit of the distributed power source, a rolling optimization problem is constructed with the joint objectives of minimizing the square integral of the frequency deviation and minimizing the control cost. The objective function of the rolling optimization problem is... for: ; For a moment frequency deviation, For the time domain of the prediction, the value is 5 seconds; This is the frequency deviation weighting coefficient, with a value of 10. To adjust the cost weighting coefficient, a value of 1 is set. The control input vector includes the power adjustment of each distributed power source and the energy storage charging and discharging power; The weighted matrix is ​​a diagonal positive definite matrix, with its diagonal elements set according to the unit control cost of each resource. For example, the unit cost of a photovoltaic inverter is 0.05, the unit cost of an energy storage device is 0.2, and the unit cost of a diesel generator is 0.5. Constraints include: the energy storage state of charge must be within the safe range of 20% to 80%; the power adjustment rate of distributed power sources must not exceed 10% of the rated power per minute; the total regulated power must meet the regulation requirements of the frequency deviation prediction sequence; and the power output of each resource must not exceed its physical limit. This rolling optimization problem is reconstructed and solved in each control cycle, executing the control command only in the first time step to achieve rolling optimization.

[0032] In solving the rolling optimization problem, the Lagrange multiplier method is introduced to handle coupling constraints, and the global optimization problem is decomposed into multiple local subproblems using the alternating direction multiplier method. Each participating entity only needs to exchange encrypted intermediate variables without disclosing the original data. The coupling constraint mainly refers to the requirement that the total adjustment power must equal the adjustment amount required by the frequency deviation prediction sequence. This coupling constraint couples the decision variables of each local resource together. The Lagrange multiplier λ is introduced to handle this coupling constraint, forming an augmented Lagrange function. The alternating direction multiplier method decomposes the original problem into multiple local subproblems, each solved independently by the corresponding resource controller, requiring only knowledge of the local cost function, local constraints, and the current Lagrange multipliers and dual variables. After solving the local subproblem in each iteration, each participating entity encrypts the updated local power adjustment amount and sends it to the central coordinator. The central coordinator aggregates all encrypted local variables, calculates the global residual, updates the Lagrange multipliers and dual variables, and then encrypts the updated multipliers and distributes them back to each participating entity. In the iterative update rule of the alternating direction multiplier method, the penalty factor is set to 0.5, the maximum number of iterations is set to 20, and the convergence threshold is set to 10. -4 The encrypted communication interface uses the national cryptographic algorithm SM4 to encrypt the transmission of Lagrange multipliers and dual variables exchanged during the iteration process, ensuring the confidentiality of intermediate variables.

[0033] Based on the solution to the rolling optimization problem, active power adjustment commands for each distributed power source and charging / discharging power commands for the energy storage device are generated and sent to the corresponding execution units for closed-loop control. Each power command includes a checksum, which is generated by the central coordinator using its private key to digitally sign the command content. Each execution unit stores the central coordinator's public key to verify the integrity and authenticity of the received power commands. After successful verification, the execution unit parses the command into specific power setpoints and adjusts the device output through the underlying controller to complete the closed-loop control. The entire control cycle is 100 milliseconds, ensuring a rapid response to frequency disturbances.

[0034] The system includes a multi-source data security acquisition module, a privacy protection aggregation module, a dynamic frequency modeling module, a frequency trajectory prediction module, a rolling optimization decision module, a distributed solution module, and an instruction issuance and execution module.

[0035] The multi-source data security acquisition module is used to acquire real-time output status data of each distributed power unit, state of charge and charging / discharging power limits of energy storage devices, and encrypted load feature vectors of each user node. This multi-source data security acquisition module includes edge computing units deployed on each user node. Each edge computing unit is equipped with a load feature extraction unit and a homomorphic encryption unit. The load feature extraction unit extracts four statistical features—mean, variance, peak-to-valley difference, and rate of change—from historical electricity load sequences. The homomorphic encryption unit uses the Paillier homomorphic encryption algorithm to encrypt these four statistical features element by element.

[0036] The privacy-preserving aggregation module is used to securely aggregate the encrypted payload feature vector based on a homomorphic encryption protocol, generating a global payload perturbation trend representation. This privacy-preserving aggregation module includes an encrypted aggregation calculation unit and a differential privacy perturbation unit. The encrypted aggregation calculation unit utilizes the additive homomorphic property of Paillier homomorphic encryption to perform term-by-term multiplication of all encrypted values ​​of the same feature dimension. The differential privacy perturbation unit applies Laplace noise to the aggregated encrypted vector.

[0037] The dynamic frequency modeling module is used to construct a dynamic frequency response differential equation model that includes system inertia, damping coefficient, and primary frequency modulation gain. The global load disturbance trend characterization is embedded as an external disturbance input term into this dynamic frequency response differential equation model. This module includes a built-in system parameter identification unit, which updates the system's equivalent inertia constant online using the least squares method based on historical frequency response data. With damping coefficient The estimated value.

[0038] The frequency trajectory prediction module is used to perform online estimation of the state variables of the dynamic frequency response differential equation model using the extended Kalman filter algorithm, so as to obtain the frequency deviation prediction sequence at the current moment.

[0039] The rolling optimization decision module is used to construct a rolling optimization problem with the joint objectives of minimizing the square integral of the frequency deviation and minimizing the control cost, based on the frequency deviation prediction sequence, the state of charge constraint of the energy storage device, and the ramp-up rate limit of the distributed power source. This rolling optimization decision module is equipped with a constraint management unit, which is used to verify in real time whether the energy storage state of charge is within the safe range of 20% to 80%, and to ensure that the power adjustment rate of the distributed power source does not exceed 10% of the rated power per minute.

[0040] The distributed solver module introduces the Lagrange multiplier method to handle coupling constraints during the solution of the rolling optimization problem, and decomposes the global optimization problem into multiple local subproblems using the alternating direction multiplier method. Each participating entity only needs to exchange encrypted intermediate variables without disclosing the original data. This distributed solver module includes an encrypted communication interface, which uses the national cryptographic algorithm SM4 to encrypt the transmission of Lagrange multipliers and dual variables exchanged during the alternating direction multiplier method iteration.

[0041] The instruction issuance and execution module generates active power adjustment instructions for each distributed power source and charging / discharging power instructions for the energy storage device based on the solution results of the rolling optimization problem, and issues these instructions to the corresponding execution units for closed-loop control. This instruction issuance and execution module integrates a digital signature verification unit, which stores the public key of the central coordinator to verify the integrity and authenticity of the received power instructions.

[0042] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0043] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A frequency security predictive control method for islanded microgrids based on privacy computing, characterized in that, include: Acquire real-time output status data of each distributed power unit, state of charge and charging / discharging power limits of energy storage devices, and encrypted load feature vectors of each user node; Based on the homomorphic encryption protocol, the encrypted payload feature vector is securely aggregated to generate a global payload perturbation trend representation. A dynamic frequency response differential equation model is constructed that includes system inertia, damping coefficient and primary frequency modulation gain, and the global load disturbance trend characterization is embedded as an external disturbance input term into the dynamic frequency response differential equation model. The extended Kalman filter algorithm is used to estimate the state variables of the dynamic frequency response differential equation model online, and the frequency deviation prediction sequence at the current moment is obtained. Based on the frequency deviation prediction sequence, the state of charge constraint of the energy storage device, and the ramp-up rate limit of the distributed power source, a rolling optimization problem is constructed with the joint objective of minimizing the square integral of frequency deviation and minimizing the control cost. In solving the rolling optimization problem, the Lagrange multiplier method is introduced to handle coupling constraints, and the global optimization problem is decomposed into multiple local subproblems by the alternating direction multiplier method. Each participating entity only needs to exchange encrypted intermediate variables without disclosing the original data. Based on the solution to the rolling optimization problem, active power adjustment commands for each distributed power source and charging / discharging power commands for the energy storage device are generated and sent to the corresponding execution units for closed-loop control.

2. The frequency security predictive control method for islanded microgrids based on privacy computing according to claim 1, characterized in that, Obtain the cryptographic payload feature vectors for each user node, including: Each user node deploys a local edge computing unit to collect its historical electricity load sequence; The historical electricity load sequence is sliced ​​using a sliding window, and four statistical features are extracted from each window: mean, variance, peak-to-valley difference, and rate of change. The four statistical features are concatenated into a 4-dimensional feature vector, and the Paillier homomorphic encryption algorithm is used to encrypt the 4-dimensional feature vector element by element to generate an encrypted payload feature vector. The encrypted load feature vector is uploaded to the microgrid central coordinator.

3. The frequency security predictive control method for islanded microgrids based on privacy computing according to claim 2, characterized in that, Secure aggregation of the encrypted payload feature vector based on a homomorphic encryption protocol includes: The microgrid central coordinator receives encrypted load feature vectors uploaded by all user nodes; By utilizing the additive homomorphic property of Paillier homomorphic encryption, we can perform a term-by-term multiplication operation on all encrypted values ​​of the same feature dimension to obtain the encrypted aggregate value of the same feature dimension. Perform the above aggregation operation on each of the four feature dimensions to form a 4-dimensional encrypted aggregation vector; A Laplace noise mechanism is applied to the 4D encrypted aggregation vector, and the noise scale parameter is determined based on the global sensitivity and the preset privacy budget to generate a perturbed encrypted aggregation vector that meets the differential privacy requirements. After decrypting the perturbation encryption aggregation vector, divide it by the total number of user nodes to obtain a noisy global load perturbation trend representation.

4. The frequency security predictive control method for islanded microgrids based on privacy computing according to claim 3, characterized in that, The dynamic frequency response differential equation model is expressed as: ； For frequency deviation, Let be the system's equivalent inertia constant. The damping coefficient is... This is the total mechanical power adjustment amount. This represents the total load disturbance. For time, the global load disturbance trend characterization is used to construct The predicted value within the future time window.

5. The frequency security predictive control method for islanded microgrids based on privacy computing according to claim 4, characterized in that, The extended Kalman filter algorithm is used to estimate the state variables of the dynamic frequency response differential equation model online, including: Define the state vector as , The rate of change of frequency deviation; The process noise covariance matrix is ​​set to be a diagonal matrix; Define the observation noise covariance; With a sampling period of 100 milliseconds, a prediction-correction loop is executed to output the frequency deviation prediction sequence at the current moment.

6. The frequency security predictive control method for islanded microgrids based on privacy computing according to claim 5, characterized in that, Construct a rolling optimization problem with the joint objectives of minimizing the integral of the squared frequency deviation and minimizing the control cost, including: Set the prediction time domain ; Set frequency deviation weighting coefficient Adjustment cost weighting coefficient ; Define control input vector Includes the power adjustment of each distributed power source and the charging and discharging power of energy storage; Configure diagonal positive definite weight matrix Its diagonal elements are set according to the unit control cost of each resource.

7. The frequency security predictive control method for islanded microgrids based on privacy computing according to claim 6, characterized in that, The conditions of the coupling constraint include: The energy storage's state of charge is within a safe range of 20% to 80%. The power regulation rate of distributed power sources shall not exceed 10% of the rated power per minute; The total regulation power meets the regulation requirements of the frequency deviation prediction sequence; The power output of each resource does not exceed its physical limit.

8. The frequency security predictive control method for islanded microgrids based on privacy computing according to claim 7, characterized in that, In solving the rolling optimization problem, by alternating direction multiplication... The method decomposes the global optimization problem into multiple local subproblems, including: Set a penalty factor; Set the maximum number of iterations; Set a convergence threshold; The national cryptographic algorithm SM4 is used to encrypt the transmission of the exchanged Lagrange multipliers and dual variables during the iteration process.

9. The frequency security predictive control method for islanded microgrids based on privacy computing according to claim 8, characterized in that, Generate active power adjustment commands for each distributed power source and discharge power commands for the energy storage device, and send them to the corresponding execution units for closed-loop control, including: The central coordinator uses a private key to digitally sign the instruction content and generate a verification code. Each execution unit executes the instruction after verifying its legitimacy using the corresponding public key.

10. The frequency security predictive control method for islanded microgrids based on privacy computing according to claim 1, characterized in that, The system equivalent inertia constant in the dynamic frequency response differential equation model With damping coefficient The frequency response data is updated online using the least squares method.

Images