Multi-objective optimization capacitor bank intelligent exchange decision-making system

Abstract

The invention relates to the technical field of exchange decision making, and discloses a multi-objective optimized capacitor bank intelligent exchange decision making system, which comprises a data acquisition and twin synchronization module, a strategy generation module, a risk preview module, an optimization decision making module and a calibration updating module, and is characterized in that real-time state sensing is realized through the data acquisition and twin synchronization module; the strategy generation module formulates a low-oscillation risk exchange strategy, the risk rehearsal module verifies the safety of the strategy, the optimization decision module screens an optimal scheme, and the calibration update module ensures the accuracy of the digital twin and the continuous optimization of the system. By dynamically adjusting the simulation step size and a real-time oscillation detection mechanism, high-frequency numerical oscillation introduced by the power electronic equipment is effectively suppressed, the accuracy of transient process simulation is remarkably improved by combining the regularized admittance matrix and multi-time scale coupling simulation, the recognition capability of a safety verification system on operation risks is ensured, and the safety verification system has a wide application prospect. And mistaken release of a high-risk strategy is avoided.

Description

Technical Field

[0001] The present invention relates to the field of swap decision-making, and more particularly, to a multi-objective optimized intelligent swap decision-making system for capacitor banks. Background Art

[0002] In the context of smart grid construction, capacitor banks are the core equipment for reactive power compensation, and their operating status directly affects the stability and energy efficiency of the grid.

[0003] Traditional switching decisions rely on manual experience and static rules. When the power grid contains a large number of power electronic equipment (such as photovoltaic inverters), the electromagnetic transient simulation of capacitor bank switching decisions causes numerical oscillations due to high-frequency switching interference. The calculated values ​​deviate from the theoretical range and fluctuate to abnormal values, causing the safety verification system to misjudge the actual high-risk strategy and may mistakenly release the switching operation that causes transient overvoltage or resonance. Summary of the Invention

[0004] The present invention provides a multi-objective optimized capacitor bank intelligent swap decision system to solve the technical problem in related technologies that traditional fixed-step simulation cannot adapt to high-frequency interference and multi-time-scale coupling introduced by power electronic equipment.

[0005] The present invention provides a multi-objective optimized capacitor bank intelligent swap decision system, comprising:

[0006] Data acquisition and twin synchronization module: This module collects electrical quantities of the capacitor bank, grid status, and environmental parameters, injects interference signatures, updates the admittance matrix and thermodynamic model, reassesses the health status of the equipment, and encapsulates the twin.

[0007] Strategy Generation Module: Constructs a state vector that incorporates interference features, designs an oscillation-aware reward mechanism, trains federated agents to generate candidate strategy sets, and filters abnormal gradients.

[0008] Risk rehearsal module: loads twins and strategies for transient simulation, dynamically adjusts simulation step size to suppress oscillation, calculates risk entropy, and marks strategy stability;

[0009] Optimization decision module: establishes a multi-objective optimization model and defines stability constraints, initializes a particle swarm to search for a Pareto optimal solution set, selects the best strategy, and generates an anti-oscillation certificate;

[0010] Calibration update module: executes the optimal strategy and collects physical data, quantifies the deviation between simulation and reality, analyzes parameter sensitivity and calibrates the twin model parameters, and reloads the updated twin.

[0011] In a preferred embodiment, the following steps are performed by the data acquisition and twin synchronization module:

[0012] Electrical quantity data acquisition and interference injection: Real-time acquisition of capacitor bank voltage / current data and explicit injection of high-frequency interference characteristics of power electronic equipment;

[0013] Admittance matrix reconstruction and regularization: Based on the grid topology and the electrical quantities after the interference injection, the node admittance matrix is ​​reconstructed and regularization is applied to suppress numerical oscillations;

[0014] Thermodynamic field state update: solve the three-dimensional heat conduction equation and calculate the temperature distribution inside the capacitor bank;

[0015] Health status reassessment: Comprehensively analyze thermal field distribution and electrical stress to update the capacitor bank health index;

[0016] Anti-oscillation twin output: Encapsulates the updated multiphysics state and generates a disturbance-resistant digital twin snapshot.

[0017] In a preferred embodiment, the calculation formula of the reconstructed admittance matrix is ​​as follows:

[0018]

[0019] y ij =G ij +jB ij

[0020] in, is the regularized node admittance matrix, Y bus is the original node admittance matrix, λ is the regularization coefficient, I is the unit matrix, G ij is the real part of the inter-node admittance, B ij is the imaginary part of the inter-node admittance, y ij Represents the element in the i-th row and j-th column of the original node admittance matrix;

[0021] The calculation formula of the thermodynamic model is as follows:

[0022]

[0023] Where T(x,t) is the temperature of the spatial point x at time t, α is the thermal diffusion coefficient, β is the heat generation coefficient, and R equiv is the equivalent resistance of the capacitor, is the Laplace operator, is the capacitive current after the interference is injected, is the partial derivative of temperature with respect to time, which represents the rate of change of temperature with time.

[0024] In a preferred embodiment, the anti-oscillation twin is as follows:

[0025]

[0026] in, is the anti-interference digital twin at time t, is the regularized node admittance matrix, HI i is the health index of the i-th capacitor, T i max is the maximum temperature of the i-th capacitor, N is the total number of capacitors, A k is the amplitude of the kth interference component, f k is the frequency of the kth interference component, and K is the total number of interference components.

[0027] In a preferred embodiment, the strategy generation module performs the following steps:

[0028] Interference-enhanced state space construction: integrating equipment status, grid operation, environmental parameters, and power electronics interference characteristics to construct a high-dimensional state vector;

[0029] Oscillation-aware reward function design: Define a three-dimensional reward mechanism that includes device balancing, network loss suppression, and oscillation risk penalty;

[0030] Local federated agent training: Each substation trains a deep reinforcement learning policy network based on local data to generate candidate actions;

[0031] Security gradient aggregation: Abnormal gradients are filtered in the cloud and aggregated to generate a global anti-interference strategy model;

[0032] Anti-interference strategy output: Filter strategies with oscillation scores below the threshold to generate a safe candidate set.

[0033] In a preferred embodiment, the calculation formula of the anti-interference strategy is as follows:

[0034]

[0035] in, is the anti-interference strategy set, a k is the candidate strategy, osc_score is the oscillation scoring function, T is the evaluation time window, is the predicted voltage at time τ, and ∈ is the oscillation threshold.

[0036] In a preferred embodiment, the risk pre-test module performs the following steps:

[0037] Twin environment initialization: load the anti-interference digital twin and candidate strategies, and configure initial simulation parameters;

[0038] Adaptive step size control: Dynamically adjust the simulation step size based on the voltage change rate to suppress numerical oscillations;

[0039] L4 norm risk entropy calculation: Statistical overvoltage event probability distribution, combined with transient spike energy to calculate risk entropy;

[0040] Stability label generation: Based on the oscillation detection results and risk entropy, the stability level of the strategy is marked;

[0041] Rehearsal result packaging: Outputs a strategy set and waveform analysis report with stability labels.

[0042] In a preferred embodiment, the following steps are performed by the optimization decision module:

[0043] Multi-objective problem modeling: Define three objective functions that need to be optimized simultaneously and build a mathematical optimization model with stability as a hard constraint;

[0044] Chaotic particle swarm initialization: randomly generate particle swarms in the strategy space and inject chaotic sequences to enhance diversity;

[0045] Constraint violation calculation: quantify the degree to which particles violate constraints to guide feasibility search;

[0046] Chaotic particle swarm update: update particle positions according to multi-objective fitness, and introduce chaos terms to avoid local optimality;

[0047] Pareto front generation: Use non-dominated sorting and crowding calculation to screen the Pareto optimal solution set;

[0048] Optimal strategy selection: Select the strategy closest to the ideal point from the Pareto front based on Chebyshev distance;

[0049] Anti-oscillation certificate generation: Outputs the stability certificate and key parameters of the optimal strategy.

[0050] In a preferred embodiment, the calculation formula of the optimal strategy is as follows:

[0051]

[0052] Among them, a * is the optimal control strategy, w m is the weight of the mth target, is the mth component of the ideal point, z ideal is the ideal point vector, representing the ideal optimal value of each objective function, f m is the mth objective function vector, is the Pareto optimal solution set, In order to obtain the maximum value among the three objectives, maxHI_balance is the ideal maximum value of health balance.

[0053] In a preferred embodiment, the calibration update module performs the following steps:

[0054] Strategy execution and physical monitoring: Execute the optimal strategy in the physical system and simultaneously collect the measured waveforms of key electrical and mechanical quantities;

[0055] Digital-physical deviation quantification: Calculate the time-domain deviation energy ratio between the twin prediction value and the twin actual value;

[0056] Parameter sensitivity matrix construction: Identify key parameters in the twin that are sensitive to deviations and calculate the Jacobian matrix;

[0057] Online gradient descent calibration: iteratively update sensitive parameters along the negative gradient direction to minimize the deviation energy ratio;

[0058] Twin dynamic reloading: inject calibrated parameters into the twin, generate an updated version and verify convergence;

[0059] Calibration certificate generation: output parameter adjustment records and calibration effect evaluation report.

[0060] The beneficial effects of the present invention are:

[0061] The present invention effectively suppresses high-frequency numerical oscillations introduced by power electronic equipment by dynamically adjusting the simulation step size and real-time oscillation detection mechanism. Combined with the regularized admittance matrix and multi-time scale coupling simulation, it significantly improves the accuracy of transient process simulation, ensures the safety verification system's ability to identify operational risks, avoids the erroneous release of high-risk strategies, and realizes the improvement of the safety and robustness of capacitor bank adjustment decisions in complex power grid environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] Figure 1 This is a structural block diagram of the multi-objective optimized capacitor bank intelligent swap decision system proposed in the present invention.

[0063] In the figure: 101, data collection and twin synchronization module; 102, strategy generation module; 103, risk rehearsal module; 104, optimization decision module; 105, calibration update module. DETAILED DESCRIPTION

[0064] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed solely to enable those skilled in the art to better understand and implement the subject matter described herein, and that the functions and arrangements of the elements discussed may be varied without departing from the scope of this specification. Various examples may omit, substitute, or add various processes or components as needed. In addition, features described with respect to some examples may also be combined in other examples.

[0065] like Figure 1As shown in the figure, the multi-objective optimization capacitor bank intelligent swap decision system includes the following modules:

[0066] Data acquisition and twin synchronization module 101: collects capacitor bank electrical quantities, grid status, and environmental parameters, injects interference signatures, updates the admittance matrix and thermodynamic model, reassesses the equipment health status, and encapsulates the twin.

[0067] The following steps are performed through this module:

[0068] Collect electrical quantity data of capacitor banks and inject harmonic interference characteristics, reconstruct the admittance matrix and apply regularization to suppress oscillations, solve the heat conduction equation to update the temperature field, re-evaluate the health status of the equipment, and encapsulate the multi-physics field state to generate an anti-interference twin;

[0069] In one embodiment of the present invention, the following steps are specifically included:

[0070] S110, electrical quantity data acquisition and interference injection: real-time acquisition of capacitor bank voltage / current data and display of high-frequency interference characteristics injected into power electronic equipment;

[0071] formula:

[0072]

[0073] in, is the capacitive current after the interference is injected, is the original capacitor current, A k is the amplitude of the kth interference component, f k is the frequency of the kth interference component, φ k is the phase of the kth interference component, K is the total number of interference components;

[0074] It should be noted that high-frequency interference characteristics include but are not limited to switching frequency harmonic components (2-20kHz), resonant frequency components (0.5-5kHz), high-frequency oscillation waveforms (>20kHz), voltage / current spikes, and high-frequency noise (white noise).

[0075] S120, Admittance Matrix Reconstruction and Regularization: Based on the grid topology and the electrical quantities after the disturbance is injected, the node admittance matrix is ​​reconstructed and regularization is applied to suppress numerical oscillations.

[0076] formula:

[0077]

[0078] y ij =G ij +jB ij

[0079] in, is the regularized node admittance matrix, Y bus is the original node admittance matrix, λ is the regularization coefficient, I is the unit matrix, G ij is the real part of the inter-node admittance, B ij is the imaginary part of the inter-node admittance, y ij Represents the element in the i-th row and j-th column of the original node admittance matrix;

[0080] S130, thermodynamic field state update: solve the three-dimensional heat conduction equation and calculate the temperature distribution inside the capacitor bank;

[0081] formula:

[0082]

[0083] Where T(x,t) is the temperature of the spatial point x at time t, α is the thermal diffusion coefficient, β is the heat generation coefficient, and R equiv is the equivalent resistance of the capacitor, is the Laplace operator, is the capacitive current after the interference is injected, is the partial derivative of temperature with respect to time, which indicates the rate of change of temperature with time;

[0084] S140, Health Status (HI) reassessment: Comprehensively analyzes thermal field distribution and electrical stress to update the capacitor bank health index;

[0085] formula:

[0086]

[0087]

[0088] in, is the health index of the i-th capacitor at time t, is the health index of the i-th capacitor at time t-1, Δt is the time step, k1, k2 are the health attenuation coefficients, ΔT i The temperature exceeds the limit. is the capacitor voltage after the interference is injected, V rated is the rated voltage of the capacitor, n is the voltage stress index, T ref is the reference temperature, Ω i is the spatial domain of the i-th capacitor, T(x,t) is the temperature of the spatial point x at time t;

[0089] S150, anti-oscillation twin output: encapsulates the updated multiphysics state and generates an anti-interference digital twin snapshot;

[0090] Output structure:

[0091]

[0092] in, is the anti-interference digital twin at time t, is the regularized node admittance matrix, HI i is the health index of the i-th capacitor, T i max is the maximum temperature of the i-th capacitor, N is the total number of capacitors, A k is the amplitude of the kth interference component, f k is the frequency of the kth interference component, K is the total number of interference components;

[0093] It should be noted that Used to describe the characteristics of the power grid, Used to describe the state characteristics of capacitor equipment, Used to describe the interference characteristics in the system;

[0094] Strategy Generation Module 102: Constructs a state vector that incorporates interference features, designs an oscillation-aware reward mechanism, trains federated agents to generate candidate strategy sets, and filters abnormal gradients.

[0095] The following steps are performed through this module:

[0096] Construct a state vector that incorporates interference features, design a reward function with oscillation penalties, train local federated agents to generate candidate strategies, filter abnormal gradients, aggregate the global model, and output a set of strategies with low oscillation risk.

[0097] In one embodiment of the present invention, the following steps are specifically included:

[0098] S210, Interference Enhanced State Space Construction: Constructing a high-dimensional state vector by integrating device status, grid operation, environmental parameters, and power electronics interference characteristics;

[0099] formula:

[0100]

[0101] Among them, s t is the current state vector, is the mean of health index, σ HI is the health index standard deviation (used to calculate the normalized standard deviation ), is the reactive power gradient, V min is the minimum voltage value, THD v is the voltage total harmonic distortion rate, A k is the amplitude of the kth interference component, f k is the frequency of the kth interference component, is the maximum switching frequency, K is the total number of interference components;

[0102] It should be noted that Indicates the equipment and grid status parameters, [V min ,THD v ] represents the power quality parameters, represents interference characteristic parameters;

[0103] S220, Oscillation-aware reward function design: Define a three-dimensional reward mechanism that includes device balancing, network loss suppression, and oscillation risk penalty;

[0104] formula:

[0105]

[0106] Among them, η is the equilibrium reward coefficient, μ is the network loss penalty coefficient, ξ is the oscillation penalty coefficient, is the standard deviation of the normalized health index, is the upper limit of the standard deviation of the health index, P loss is the network loss, To predict the peak voltage, s t is the current state vector, σ HI is the standard deviation of the health index, represents the expectation operator, sim_osc_index(·) represents the oscillation index calculation function, inputs the state vector, and outputs the predicted voltage oscillation peak value, R t Represents the total reward value at time t, which is used to evaluate the comprehensive performance of the control strategy;

[0107] It should be noted that Represents the balanced reward item, which is used to reward the balance of the equipment health status, μ·P loss Represents the network loss penalty item, which is used to punish network loss. Represents the oscillation penalty term, which is used to punish the case where the voltage oscillation exceeds the threshold;

[0108] S230, local federated agent training: each substation trains a deep reinforcement learning policy network based on local data to generate candidate actions;

[0109] formula:

[0110]

[0111] in, is the action of the ith agent at time t, is the policy network of the i-th agent, θ i is the policy network parameter of the ith agent, α is the learning rate, γ is the discount factor, is the local dataset of the ith agent, is the action value function of the i-th agent, ∑ is the exploration noise covariance matrix, R t represents the total reward value at time t, s t is the current state vector, s t+1 is the state vector of the next state, represents the temporal difference loss function, which is used to evaluate the prediction error. Represents the parameter θ i The gradient operator, represents Gaussian noise with mean 0 and covariance ∑, a′ is the candidate action at the next moment, is the local dataset of the i-th agent, containing the transition samples of state-action-reward-next state, represents the expectation operator;

[0112] S240, Security Gradient Aggregation: Filters abnormal gradients in the cloud and aggregates them to generate a global anti-interference strategy model;

[0113] formula:

[0114]

[0115] in, is the global model parameter at time t+1, is the global model parameter at time t, is the policy network parameter of the i-th agent at time t, w i is the weight of the i-th agent, τ is the gradient deviation threshold, N is the total number of agents, is the local dataset of the ith agent, is the size of the local dataset of the ith agent, is the parameter gradient of the ith agent, is the parameter gradient of the global model at time t, Represents an indicator function, which is 1 when the condition is met and 0 otherwise;

[0116] S250, anti-interference strategy output: screening strategies with oscillation scores below the threshold to generate a safe candidate set;

[0117] formula:

[0118]

[0119] in, is the anti-interference strategy set, a k is the candidate strategy, osc_score is the oscillation scoring function, T is the evaluation time window, is the predicted voltage at time τ, ∈ is the oscillation threshold;

[0120] Risk rehearsal module 103: loads the twin and strategy for transient simulation, dynamically adjusts the simulation step size to suppress oscillation, calculates risk entropy and marks the strategy stability;

[0121] The following steps are performed through this module:

[0122] Load the twin and strategy to initialize the simulation environment, dynamically adjust the step size to suppress numerical oscillations, detect the third-order derivative of voltage in real time to trigger step size compression, calculate the L4 norm risk entropy, mark the strategy stability, and package the results;

[0123] In one embodiment of the present invention, the following steps are specifically included:

[0124] S310, Twin environment initialization: Loading the anti-interference digital twin DT robust and candidate strategy a k , configure the initial simulation parameters;

[0125] formula:

[0126] Env sim = <DT robust ,a k ,t start =0,Δt0=10μs>

[0127] Among them, DT robust For anti-interference digital twins, a k is the candidate strategy, t start is the initial time, Δt0 is the initial simulation step, Env sim It is a simulation environment object, which contains the complete configuration of twins, strategies and simulation parameters;

[0128] S320, Adaptive Step Size Control: Dynamically adjusts the simulation step size based on the voltage change rate to suppress numerical oscillations;

[0129] formula:

[0130]

[0131] Where Δt n+1 is the next step length, Δt max is the maximum allowed step size, ∈ tol is the error tolerance, is the voltage change of the current step, ||·|| ∞ is the infinite norm, where 0.8 is the safety factor and 1 / 5 is the empirical exponent used to adjust the sensitivity of the step size to voltage changes;

[0132] S330, third-order derivative oscillation detection: calculates the third-order derivative of voltage in real time, detects high-frequency oscillations and triggers emergency step compression;

[0133] formula:

[0134]

[0135] Among them, γ n is the third-order derivative norm of voltage, V C is the capacitor voltage, γ th is the oscillation threshold, Δt n+1 For the next step;

[0136] S340, L4 norm risk entropy calculation: Statistical overvoltage event probability distribution, combined with transient spike energy to calculate risk entropy;

[0137] formula:

[0138]

[0139] in, is the risk entropy of the k-th strategy, p j is the probability of the jth overvoltage event, Δv j is the voltage deviation, V rated is the rated voltage of the capacitor, ΔT is the integration time window, M is the total number of events, V C (t) is the capacitor voltage at time t;

[0140] S350, stability label generation: based on the oscillation detection results and risk entropy, the stability level of the strategy is marked;

[0141] formula:

[0142]

[0143] Among them, Stable k is the stability label of the k-th strategy, is the risk entropy of the k-th strategy, γ n is the third-order derivative norm of voltage, γ th is the oscillation threshold;

[0144] S360, preview result packaging: output strategy sets and waveform analysis reports with stability labels;

[0145] Output structure:

[0146]

[0147] Among them, a k For candidate strategies, is the risk entropy value of the k-th strategy, Stable k is the stability label of the k-th strategy, γ max is the maximum oscillation value, i.e. γ nThe maximum value, Δt min is the minimum step size, corresponding to Δt n+1 The minimum value of Represents the set of preview results, which contains complete information of strategy evaluation;

[0148] Optimization decision module 104: establishes a multi-objective optimization model and defines stability constraints, initializes a particle swarm to search for a Pareto optimal solution set, selects the best strategy and generates an anti-oscillation certificate;

[0149] The following steps are performed through this module:

[0150] Establish a multi-objective optimization model and define stability constraints, initialize a chaotic particle swarm to inject random perturbations, quantify constraint violations, update particle positions to generate a Pareto front, and select the optimal strategy based on Chebyshev distance;

[0151] In one embodiment of the present invention, the following steps are specifically included:

[0152] S410, Multi-Objective Problem Modeling: Define three objective functions to be optimized simultaneously and construct a mathematical optimization model with stability as a hard constraint;

[0153] formula:

[0154]

[0155] stg1(a)=Stable(a)=True

[0156] g2(a)=V min (a)≥0.95

[0157]

[0158] Among them, f(a) is the objective function vector, f1(a) is the health balance target (negative value), f2(a) is the network loss target, f3(a) is the risk entropy target, g1, g2, g3 are constraints, V min is the minimum voltage value, γ th is the oscillation threshold, V C is the capacitor voltage, a is the decision variable vector, HI_balance(a) is the health balance function, P loss (a) is the network loss function, H risk (a) is the risk entropy function, Stable(a) represents the stability judgment function, represents the third derivative of the capacitor voltage;

[0159] S420, Chaotic Particle Swarm Initialization: In Strategy Space Randomly generate particle swarms and inject chaotic sequences to enhance diversity;

[0160] formula:

[0161]

[0162] c i =chaos_map(seed=i),chaos_map:c j+1 =4c j (1-c j )

[0163] in, is the initial position of the i-th particle, a k is the candidate strategy, β is the chaos intensity coefficient, c i is a chaotic sequence, is the initial velocity of the i-th particle, c j is the chaos value of the jth iteration;

[0164] S430, constraint violation calculation: quantifying the degree to which particles violate constraints, to guide feasibility search;

[0165] formula:

[0166]

[0167] Where CV(x) is the constraint violation degree (the cumulative value of all constraint violations), is the normalized constraint function, is the stability indicator function (stable is 1, unstable is 0), V min is the minimum voltage value, γ th is the oscillation threshold, V C is the capacitor voltage;

[0168] S440, Chaotic Particle Swarm Update: Updates particle positions based on multi-objective fitness and introduces chaotic terms to avoid local optimality.

[0169] formula:

[0170]

[0171] Among them, ω is the inertia weight (control speed inheritance), c1, c2 are acceleration coefficients (control the influence of individual and global optimality), r1, r2 are random numbers (uniformly distributed in [0, 1]), p i is the individual optimal position, g is the global optimal position, c i is the chaotic sequence, β is the chaos intensity coefficient, is the velocity vector of the i-th particle at time t+1, is the velocity vector of the i-th particle at time t, is the position vector of the i-th particle at time t+1, is the position vector of the i-th particle at time t, i is the particle number, and t is the number of iterations;

[0172] S450, Pareto front generation: using non-dominated sorting and crowding calculation to screen the Pareto optimal solution set;

[0173] formula:

[0174]

[0175]

[0176] in, is the Pareto optimal solution set, x is a decision variable vector, y is another decision variable vector, f(y)<f(x) means y dominates x (all objectives are not worse than x and at least one objective is better than x), is the maximum / minimum value of the mth target, x next ,x prev is the adjacent solution of x after sorting on the target m, f m is the mth objective function vector, crowding_dist(x) represents the crowding distance of the x solution;

[0177] S460, Optimal Strategy Selection: Selecting the strategy closest to the ideal point from the Pareto front based on Chebyshev distance;

[0178] formula:

[0179]

[0180] z ideal =[maxHI_balance,0,0] T

[0181] Among them, a * is the optimal control strategy, w m is the weight of the mth target (reflecting the importance of the target), is the mth component of the ideal point, z ideal is the ideal point vector, representing the ideal optimal value of each objective function, f m is the mth objective function vector, is the Pareto optimal solution set, To obtain the maximum value among the three objectives, maxHI_balance is the ideal maximum value of health balance;

[0182] S470, Anti-oscillation certificate generation: Outputs the stability certificate and key parameters of the optimal strategy;

[0183] Output structure:

[0184]

[0185] Among them, a * is the optimal control strategy, is the maximum third-order derivative value in the simulation (used to verify the oscillation constraint), is the minimum time step used (to ensure numerical stability), ERR v is the voltage fitting error (to verify the simulation accuracy), Cert osc A certificate of resistance to oscillation, containing a complete set of parameters verifying the stability of the strategy;

[0186] Calibration and update module 105: executes the optimal strategy and collects physical data, quantifies the deviation between simulation and reality, analyzes parameter sensitivity and calibrates the twin model parameters, and reloads the updated twin;

[0187] The following steps are performed through this module:

[0188] Execute the optimal strategy and simultaneously collect physical data, calculate the energy ratio of the simulated and actual waveform deviations, construct the parameter sensitivity matrix, calibrate key parameters along the negative gradient direction, reload the twin and verify convergence;

[0189] In one embodiment of the present invention, the following steps are specifically included:

[0190] S510, Strategy Execution and Physical Monitoring: Execute the optimal strategy in the physical system * , synchronously collect the measured waveforms of key electrical quantities (voltage, current) and mechanical quantities (temperature, vibration);

[0191] formula:

[0192]

[0193] in, is the measured capacitor voltage, is the measured capacitor current, T actual (t) is the measured temperature, Vib actual (t) is the measured vibration, t end is the measurement end time, D actual is the actual value of the twin;

[0194] S520, Quantifying Digital-Physical Deviation: Calculating Twin Prediction Value D sim and the twin actual value D actual The time domain deviation energy ratio;

[0195] formula:

[0196]

[0197] Where ERR is the deviation energy ratio, N is the number of sampling points, is the simulation value of the kth sampling point, is the actual twin value of the k-th sampling point, is the L2 norm, t end is the measurement end time;

[0198] S530, parameter sensitivity matrix construction: identifying key parameters in the twin that are sensitive to deviations and calculating the Jacobian matrix;

[0199] formula:

[0200]

[0201] Among them, λ is the regularization coefficient, β is the heat generation coefficient, k1, k2 are the health attenuation coefficients, n is the voltage stress index, γ th is the oscillation threshold, δ is the parameter perturbation amount, ERR is the deviation energy ratio, and J represents;

[0202] S540, online gradient descent calibration: iteratively update sensitive parameters along the negative gradient direction to minimize the deviation energy ratio;

[0203] formula:

[0204]

[0205] in, is the updated parameter value, is the parameter value before update, η is the learning rate, ∈ clip is the gradient clipping threshold, ERR is the deviation energy ratio;

[0206] S550, Twin Dynamic Reload: Inject calibrated parameters into the twin, generate an updated version and verify convergence;

[0207] formula:

[0208]

[0209] Verification conditions:

[0210] ||ERR new -ERR old ||<∈ conv

[0211] in, is the updated digital twin, Θ calib The parameter set after calibration includes all parameters defined in S530. new is the updated deviation, ERR old is the deviation before updating, ∈ conv is the convergence threshold;

[0212] S560, calibration certificate generation: output parameter adjustment records and calibration effect evaluation report;

[0213] Output structure:

[0214] Calib cert = * ,Θ calib ,ΔERR,J top >

[0215] Among them, a * is the optimal control strategy, Θ calib is the calibrated parameter set, ΔERR is the deviation improvement, J top As the main sensitive parameter set, Calib cert Indicates a calibration certificate, including parameter adjustment records and calibration effect evaluation.

[0216] Based on the above multi-objective optimization capacitor bank intelligent swap decision system, an application of the intelligent capacitor bank swap system in an industrial park power grid is given:

[0217] 1. Data acquisition and twin synchronization module:

[0218] The voltage and current data of the capacitor bank in the industrial park are collected in real time, and simulated high-frequency harmonic interference (such as 5kHz resonant frequency components and 15kHz switching frequency harmonics) is injected.

[0219] According to the collected data, the admittance matrix of the power grid nodes is reconstructed, and the numerical oscillation is suppressed through regularization technology.

[0220] Calculate the temperature distribution inside the capacitor and, combined with the electrical stress, re-evaluate the health index of each capacitor.

[0221] Encapsulate the current power grid and equipment status and generate interference-resistant digital twin snapshots.

[0222] The execution results under this module are:

[0223] A digital twin including the electrical status, temperature field distribution and health index of the capacitor bank was obtained.

[0224] It was identified that the health index of individual capacitors dropped to 0.85 (the health threshold was set at 0.9) due to high temperature rise, indicating potential risks.

[0225] The admittance matrix of the power grid nodes is regularized, the numerical stability is enhanced, and the oscillation problem is alleviated.

[0226] 2. Strategy generation module:

[0227] ​Utilize the state data in the digital twin to construct a high-dimensional state vector that includes equipment health, grid operating parameters, and interference characteristics.

[0228] A reward function is designed that comprehensively considers device health balance, network loss suppression, and oscillation risk penalty, and trains the local federation agent to generate multiple sets of candidate swap strategies.

[0229] The safety gradient aggregation technology is used to filter abnormal gradients, integrate the strategies of each substation, and generate a global low oscillation risk strategy set.

[0230] The execution results under this module are:

[0231] Ten sets of candidate swap strategies were generated, all of which had oscillation scores below the preset threshold, indicating high stability of the strategy set.

[0232] The strategy prioritizes a solution with balanced load and low network loss, which is expected to reduce the total system power loss by about 3%.

[0233] The risk of oscillation is significantly reduced, and voltage fluctuations are expected to be controlled within ±2%.

[0234] 3. Risk rehearsal module:

[0235] Load the digital twin and candidate strategies and perform transient simulation in the simulation environment.

[0236] Dynamically adjust the simulation step size to suppress numerical oscillations, and calculate the third-order derivative of voltage in real time to detect oscillation events.

[0237] Calculate the risk entropy of each strategy, combine it with the oscillation detection results, and mark the strategy stability level.

[0238] The execution results under this module are:

[0239] Simulations show that all candidate strategies can effectively suppress voltage transient oscillations, with the maximum oscillation amplitude being less than 2.5%.

[0240] Five strategies were labeled "highly stable", four strategies were labeled "moderately stable", and one strategy was labeled "unstable" due to occasional oscillations.

[0241] Generate a detailed analysis report including strategy oscillation waveform and risk entropy to facilitate subsequent decision-making.

[0242] 4. Optimize the decision-making module:

[0243] Based on the candidate strategies and their stability constraints, a multi-objective optimization model is established, with the objectives including equipment health balance, network loss minimization and risk entropy reduction.

[0244] The chaotic particle swarm optimization algorithm is used to search for the Pareto optimal solution set to avoid falling into the local optimum.

[0245] The optimal strategy closest to the ideal point is selected and an anti-oscillation stability certificate is generated.

[0246] The execution results under this module are:

[0247] After optimization, a set of strategies was finally selected, which achieved the best health balance index, reduced network loss by 5%, and minimized risk entropy.

[0248] The maximum value of the third-order derivative of the oscillation corresponding to this strategy meets the stability constraint, and the simulation stability step size is ≥0.01 seconds.

[0249] An anti-oscillation certificate containing the key parameters and stability verification of the strategy was generated as a basis for subsequent execution.

[0250] 5. Calibration update module:

[0251] Optimization strategies are implemented on capacitor banks in actual industrial parks, collecting physical data such as voltage, current, temperature, and vibration in real time.

[0252] The error energy ratio between the simulation prediction and the actual measured waveform is calculated to identify sensitive model parameters.

[0253] The online gradient descent method is used to adjust the twin model parameters to improve the simulation accuracy of the digital twin.

[0254] Reload the calibrated twin to verify the convergence of simulation and measured data.

[0255] The execution results under this module are:

[0256] The energy ratio of the deviation between physical and simulation data is reduced from the initial 0.12 to 0.03, and the simulation results are closer to reality.

[0257] Key parameters such as heat generation coefficient and health decay coefficient are precisely adjusted, and model stability is significantly improved.

[0258] Output calibration certificate, detailing the parameter adjustment process and effect evaluation, to ensure the system continues to operate accurately.

[0259] The application achieves real-time status perception through data acquisition and twin synchronization modules, the strategy generation module formulates low oscillation risk swap strategies, the risk rehearsal module verifies the strategy safety, the optimization decision module selects the optimal solution, and the calibration update module ensures the accuracy of the digital twin and continuous optimization of the system.

[0260] The above describes the embodiments of the present invention, but the present invention is not limited to the above specific implementation methods. The above specific implementation methods are merely illustrative and not restrictive. Ordinary technicians in this field can also make many forms under the guidance of the present invention, all of which are protected by the present invention.

Claims

1. A multi-objective optimized capacitor bank intelligent swap decision system, characterized by: include: Data acquisition and twin synchronization module: This module collects electrical quantities of the capacitor bank, grid status, and environmental parameters, injects interference signatures, updates the admittance matrix and thermodynamic model, reassesses the health status of the equipment, and encapsulates the twin. Strategy Generation Module: Constructs a state vector that incorporates interference features, designs an oscillation-aware reward mechanism, trains federated agents to generate candidate strategy sets, and filters abnormal gradients. Risk rehearsal module: loads twins and strategies for transient simulation, dynamically adjusts simulation step size to suppress oscillation, calculates risk entropy, and marks strategy stability; Optimization decision module: establishes a multi-objective optimization model and defines stability constraints, initializes a particle swarm to search for a Pareto optimal solution set, selects the best strategy, and generates an anti-oscillation certificate; Calibration update module: executes the optimal strategy and collects physical data, quantifies the deviation between simulation and reality, analyzes parameter sensitivity and calibrates the twin model parameters, and reloads the updated twin.

2. The multi-objective optimization capacitor bank intelligent swap decision system according to claim 1, characterized in that: The following steps are performed through the data acquisition and twin synchronization module: Electrical quantity data acquisition and interference injection: Real-time acquisition of capacitor bank voltage / current data and explicit injection of high-frequency interference characteristics of power electronic equipment; Admittance matrix reconstruction and regularization: Based on the grid topology and the electrical quantities after the interference injection, the node admittance matrix is ​​reconstructed and regularization is applied to suppress numerical oscillations; Thermodynamic field state update: solve the three-dimensional heat conduction equation and calculate the temperature distribution inside the capacitor bank; Health status reassessment: Comprehensively analyze thermal field distribution and electrical stress to update the capacitor bank health index; Anti-oscillation twin output: Encapsulates the updated multiphysics state and generates a disturbance-resistant digital twin snapshot.

3. The multi-objective optimization capacitor bank intelligent swap decision system according to claim 2, characterized in that: The calculation formula of the reconstructed admittance matrix is ​​as follows: in, is the regularized node admittance matrix, is the original node admittance matrix, is the regularization coefficient, is the identity matrix, is the real part of the inter-node admittance, is the imaginary part of the inter-node admittance, Represents the element in the i-th row and j-th column of the original node admittance matrix; The calculation formula of the thermodynamic model is as follows: in, is the temperature of the spatial point x at time t, is the thermal diffusivity, is the heat generation coefficient, is the equivalent resistance of the capacitor, is the Laplace operator, is the capacitive current after the interference is injected, is the partial derivative of temperature with respect to time, which represents the rate of change of temperature with time.

4. The multi-objective optimization capacitor bank intelligent swap decision system according to claim 3, characterized in that: The anti-oscillation twins are as follows: in, is the anti-interference digital twin at time t, is the regularized node admittance matrix, is the health index of the i-th capacitor, is the maximum temperature of the i-th capacitor, is the total number of capacitors, is the amplitude of the kth interference component, is the frequency of the kth interference component, is the total number of interference components.

5. The multi-objective optimized capacitor bank intelligent swap decision system according to claim 1, characterized in that: The strategy generation module performs the following steps: Interference-enhanced state space construction: integrating equipment status, grid operation, environmental parameters, and power electronics interference characteristics to construct a high-dimensional state vector; Oscillation-aware reward function design: Define a three-dimensional reward mechanism that includes device balancing, network loss suppression, and oscillation risk penalty; Local federated agent training: Each substation trains a deep reinforcement learning policy network based on local data to generate candidate actions; Security gradient aggregation: Abnormal gradients are filtered in the cloud and aggregated to generate a global anti-interference strategy model; Anti-interference strategy output: Filter strategies with oscillation scores below the threshold to generate a safe candidate set.

6. The multi-objective optimized capacitor bank intelligent swap decision system according to claim 5, characterized in that: The calculation formula of the anti-interference strategy is as follows: in, is the anti-interference strategy set, For candidate strategies, is the oscillating scoring function, For the evaluation time window, is the predicted voltage at time τ, is the oscillation threshold.

7. The multi-objective optimized capacitor bank intelligent swap decision system according to claim 1, characterized in that: The following steps are performed through the risk rehearsal module: Twin environment initialization: load the anti-interference digital twin and candidate strategies, and configure initial simulation parameters; Adaptive step size control: Dynamically adjust the simulation step size based on the voltage change rate to suppress numerical oscillations; L4 norm risk entropy calculation: Statistical overvoltage event probability distribution, combined with transient spike energy to calculate risk entropy; Stability label generation: Based on the oscillation detection results and risk entropy, the stability level of the strategy is marked; Rehearsal result packaging: Outputs a strategy set and waveform analysis report with stability labels.

8. The multi-objective optimized capacitor bank intelligent swap decision system according to claim 1, characterized in that: The following steps are performed by optimizing the decision module: Multi-objective problem modeling: Define three objective functions that need to be optimized simultaneously and build a mathematical optimization model with stability as a hard constraint; Chaotic particle swarm initialization: randomly generate particle swarms in the strategy space and inject chaotic sequences to enhance diversity; Constraint violation calculation: quantify the degree to which particles violate constraints to guide feasibility search; Chaotic particle swarm update: update particle positions according to multi-objective fitness, and introduce chaos terms to avoid local optimality; Pareto front generation: Use non-dominated sorting and crowding calculation to screen the Pareto optimal solution set; Optimal strategy selection: Select the strategy closest to the ideal point from the Pareto front based on Chebyshev distance; Anti-oscillation certificate generation: Outputs the stability certificate and key parameters of the optimal strategy.

9. The multi-objective optimized capacitor bank intelligent swap decision system according to claim 8, characterized in that: The calculation formula for the optimal strategy is as follows: in, is the optimal control strategy, is the weight of the mth target, is the mth component of the ideal point, is the ideal point vector, which represents the ideal optimal value of each objective function. is the mth objective function vector, is the Pareto optimal solution set, To obtain the maximum value among the three objectives, It is the ideal maximum value for healthy balance.

10. The multi-objective optimized capacitor bank intelligent swap decision system according to claim 1, characterized in that: To update the module through calibration, perform the following steps: Strategy execution and physical monitoring: Execute the optimal strategy in the physical system and simultaneously collect the measured waveforms of key electrical and mechanical quantities; Digital-physical deviation quantification: Calculate the time-domain deviation energy ratio between the twin prediction value and the twin actual value; Parameter sensitivity matrix construction: Identify key parameters in the twin that are sensitive to deviations and calculate the Jacobian matrix; Online gradient descent calibration: iteratively update sensitive parameters along the negative gradient direction to minimize the deviation energy ratio; Twin dynamic reloading: inject calibrated parameters into the twin, generate an updated version and verify convergence; Calibration certificate generation: output parameter adjustment records and calibration effect evaluation report.

Images