System for mapping network construction capability evaluation index and power supply parameter under system perspective

Abstract

The invention relates to the technical field of electric power system operation and control, and discloses a system-perspective networking capability evaluation index and power supply parameter mapping system, which comprises a data acquisition module, a dynamic tensor modeling module, a coupling decomposition module, a distributed optimization module, an index fusion module and an edge cloud collaboration module. The method comprises the steps of constructing a dynamic tensor by collecting power grid data in real time, decomposing and extracting coupling features and generating optimization constraints, realizing parameter cooperative adjustment based on hierarchical multi-objective optimization and quantum particle swarm optimization, and forming a closed-loop optimization process by combining entropy weight-TOPSIS weight distribution and edge federated learning feedback updating. According to the method, the time-space correlation characteristics of the power grid are extracted by improving Tucker tensor decomposition, hierarchical ADMM optimization and quantum particle swarm optimization are combined, an entropy weight-TOPSIS dynamic weighting and federated edge collaboration mechanism is adopted, a power supply parameter-network construction capability index mapping model is constructed, and the multi-target optimization precision, the real-time performance and the data privacy are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of power system operation and control, in particular to a system perspective network construction capability evaluation index and power supply parameter mapping system. BACKGROUND

[0002] Under the background of accelerating the construction of new power systems, the high proportion of new energy grid-connected leads to the increasingly complex dynamic characteristics of the power grid, and the traditional network construction capability evaluation method based on steady-state model has been difficult to meet the real-time control demand.

[0003] The prior art usually uses a matrix model to represent the static mapping relationship between power supply parameters and network construction indexes, but cannot effectively capture the dynamic coupling effect between parameters in multiple time scales. Especially in the power grid scene with close time and space dimension coupling, such method easily ignores the nonlinear influence of transient process on parameter correlation, resulting in insufficient quantitative precision of the evaluation model for key capabilities such as system inertia support and voltage stability.

[0004] In addition, the existing distributed optimization algorithm mostly adopts a fixed weight allocation strategy, which is difficult to respond to random fluctuations on both sides of the source and load in a timely manner, and the centralized data fusion architecture has the risk of privacy leakage, which restricts the actual application performance of cross-regional collaborative optimization. With the rapid evolution of power grid topology and operation mode, it is necessary to build an evaluation system that takes into account dynamic feature analysis, privacy protection and collaborative decision-making to realize accurate mapping and real-time regulation of power supply parameters and network construction capability.

[0005] Therefore, the present application proposes a system perspective network construction capability evaluation index and power supply parameter mapping system to solve the deficiencies of the prior art. SUMMARY

[0006] In view of the deficiencies of the prior art, the present application provides a system perspective network construction capability evaluation index and power supply parameter mapping system, which solves the problems of difficulty in accurately representing multi-dimensional dynamic coupling characteristics, cross-regional collaborative optimization weight allocation mismatch and insufficient data privacy protection.

[0007] To achieve the above purpose, the present application realizes the following technical scheme: a system perspective network construction capability evaluation index and power supply parameter mapping system, comprising:

[0008] A data acquisition module is used to acquire power supply parameters, network construction capability indexes and topology data of the power grid in real time.

[0009] A dynamic tensor modeling module is used to slice the power supply parameters, network construction capability indexes and topology data according to time windows, and construct a three-dimensional dynamic tensor containing parameter dimension, time dimension and space dimension.

[0010] a coupling decomposition module configured to perform tensor decomposition on the dynamic tensor to extract dynamic coupling features among parameters and generate optimization constraints based on the dynamic coupling features;

[0011] a distributed optimization module configured to perform multi-objective collaborative optimization on the power supply parameters through a hierarchical architecture based on the dynamic coupling features and the optimization constraints, wherein the hierarchical architecture comprises an upper-layer global coordination unit and a lower-layer partition optimization unit;

[0012] an index fusion module configured to assign optimization weights according to dynamic changes of multi-time scale network construction capability indexes;

[0013] an edge cloud end collaborative module configured to update a global model through federated aggregation and feed back the updated model to the dynamic tensor modeling module and the distributed optimization module.

[0014] Preferably, the data acquisition module comprises:

[0015] a parameter acquisition unit configured to acquire power supply parameters in real time, wherein the power supply parameters comprise voltage V, frequency f, virtual inertia H v , active power P and reactive power Q;

[0016] a topology acquisition unit configured to acquire a grid node admittance matrix

[0017] wherein N b represents a total number of grid nodes;

[0018] an index acquisition unit configured to acquire network construction capability indexes, including short-circuit capacity S sc , inertia time constant T i and voltage stability margin M v .

[0019] Preferably, the dynamic tensor modeling module comprises:

[0020] a time slicing unit configured to slice data according to a preset time window Δt;

[0021] a tensor construction unit configured to construct a three-dimensional dynamic tensor

[0022] wherein N p represents a number of parameter types; N t represents a number of time slices, which is calculated from Δt and a total duration; and N s represents a number of spatial numbers of grid nodes or partitions.

[0023] Preferably, the coupling decomposition module comprises:

[0024] Tensor decomposition unit, employing improved Tucker decomposition for dynamic tensors Decompose to obtain the core tensor and factor matrix U (p) U (t) U (s) ,satisfy:

[0025] T≈G×1U (p) ×2U (t) ×3U (s) ;

[0026] Coupled computational unit, calculates the dynamic coupling coefficient α between parameters. ij (t), whose expression is:

[0027]

[0028] Where i represents the i-th type of power supply parameter; j represents the j-th type of grid construction capability index; t represents the index of the time slice, corresponding to the division result of the time window Δt; k represents the index of the spatial dimension, representing the number of the grid node or partition; ∈ is a zero-prevention constant; ||·|| F U represents the Frobenius norm; (p) (i,r p ) is the parameter i at the rth p The projected weights on each parameter factor reflect the contribution of the parameter to the factor components; U (t) (t,r t ) represents time slice t at time r t The distribution coefficients over each time factor characterize the time dynamics; U (s) (k,r s ) represents the spatial node / partition k at the rth node. s The distributed weights on each spatial factor describe spatial correlation.

[0029] Preferably, the distributed optimization module includes:

[0030] The global coordination unit uses the ADMM framework to decompose the power grid into P partitions and defines a global optimization problem:

[0031]

[0032] Where, x p Let f be the optimization variable for the P-th partition. p (x p Let A be the partitioning objective function. p This is a dynamic constraint matrix, and its element value a mm The dynamic coupling coefficient α generated by the coupling decomposition module ij (t) is positively correlated, satisfying a mm∝α ij (t);

[0033] The partition optimization unit uses an improved quantum particle swarm optimization algorithm to optimize each partition.

[0034] Preferably, the particle position update of the improved quantum particle swarm optimization algorithm includes a quantum tunneling mechanism, and its position update formula is as follows:

[0035]

[0036] in, This represents the position of the i-th particle in the d-th dimension parameter and the k-th iteration. This represents the historical best position of the i-th particle in the d-th dimension; β is the learning factor; u is a uniformly distributed random number. For the average optimal position, N p The total number of particles in the particle swarm. p represents the optimal position of the i-th particle in the k-th iteration. t The quantum tunneling probability is represented by the deviation of the multi-timescale index; This represents the velocity of the i-th particle in the d-th dimension and the (k+1)-th iteration.

[0037] Preferably, the indicator fusion module includes:

[0038] Entropy weight calculation unit, based on the standardized matrix R = [r st Calculate information entropy e s With weight w s ,satisfy:

[0039]

[0040] Where lnT is the normalization factor, T is the total number of time windows; S is the number of time scales; r st For the first s The standardized value of each indicator in the t-th time window satisfies r st ∈[0,1];

[0041] The weight adjustment unit adjusts the weights using the TOPSIS method.

[0042] Among them, C i For the degree of closeness.

[0043] Preferably, the edge-cloud collaboration module includes:

[0044] Edge computing units, deployed at grid nodes or zones, perform local tensor decomposition and output the core tensor G. p ;

[0045] The federal aggregation unit aggregates each partition core tensor by weighted average to obtain a global core tensor:

[0046]

[0047] Wherein, P is the total number of power grid partitions.

[0048] Preferably, the edge cloud coordination module further comprises:

[0049] The model updating unit updates the local tensor decomposition parameters by delivering the aggregated global core tensor G global to each edge computing unit.

[0050] The application also provides a system perspective network construction capability evaluation index and power supply parameter mapping method, comprising the following steps:

[0051] S1, real-time collection of power supply parameters, network construction capability indexes and topology data of the power grid, the power supply parameters including voltage, frequency, virtual inertia, active power and reactive power, the network construction capability indexes including short-circuit capacity, inertia time constant and voltage stability margin;

[0052] S2, slicing the collected data according to a preset time window to build a three-dimensional dynamic tensor containing parameter dimension, time dimension and space dimension, wherein the space dimension is the number of power grid nodes or partitions;

[0053] S3, tensor decomposition of the dynamic tensor to extract dynamic coupling characteristics between parameters, and generation of optimization constraints based on the dynamic coupling characteristics;

[0054] S4, multi-objective optimization by a hierarchical architecture, the upper layer global coordination decomposes the power grid into multiple partitions and defines a global optimization problem, and the lower layer partition optimization optimizes in each partition by using a quantum particle swarm algorithm, the quantum particle swarm algorithm containing a quantum tunneling mechanism;

[0055] S5, according to the dynamic changes of multi-time scale network construction capability indexes, calculating the initial weight by an entropy weight method, and correcting the weight by a TOPSIS method;

[0056] S6, local tensor decomposition at the edge node, aggregation of core tensors of each partition by weighted average to generate a global core tensor and feedback to the local module to update the dynamic tensor modeling and optimization constraints.

[0057] The application provides a system perspective network construction capability evaluation index and power supply parameter mapping system. The following beneficial effects are achieved:

[0058] 1. The application extracts the spatio-temporal correlation characteristics between parameters from the power grid dynamic tensor data through the improved Tucker tensor decomposition algorithm, and quantifies the dynamic coupling relationship explicitly. Compared with traditional methods, the application can simultaneously capture the nonlinear interaction rules of parameters in time slices, spatial distribution and multi-dimensional indicators, providing high-precision feature representation for network capacity assessment.

[0059] 2. The application is based on the hierarchical architecture of upper global coordination and lower partition optimization. The application ensures the consistency of cross-regional constraints through the ADMM framework, and enhances the local search efficiency by combining the quantum particle swarm algorithm. The contradiction between computational complexity and global convergence in large-scale power grid optimization is effectively solved, and the real-time performance and robustness of multi-objective collaborative optimization are significantly improved.

[0060] 3. The application can dynamically adjust the priority of optimization objectives according to the volatility of index data and the degree of closeness to ideal solution through the weight allocation mechanism of information entropy and TOPSIS method. Compared with the fixed weight strategy, this method can adapt to the time-varying characteristics of power grid operation state, and avoid the optimization deviation caused by fixed weight.

[0061] 4. The application is based on the edge cloud collaborative mechanism of federated learning. The application aggregates local model parameters while preserving the original data on the local node. Through hybrid weight update and model fine-tuning, the balance between global knowledge sharing and local feature preservation is achieved, solving the data privacy leakage risk of traditional centralized training.

[0062] 5. The application establishes an interpretable mapping relationship from power supply parameters to network capacity indicators through dynamic tensor modeling and multi-module collaborative feedback mechanism. The mapping model can track the changes of power grid topology and operation state in real time, providing accurate decision basis for the stable control of high penetration rate of new energy power grid. BRIEF DESCRIPTION OF DRAWINGS

[0063] Figure 1 The system architecture diagram of the application;

[0064] Figure 2 The method flowchart of the application. DETAILED DESCRIPTION

[0065] The technical solutions in the embodiments of the application will be described in detail below with reference to the drawings of the application specification. Obviously, the described embodiments are only a part of the embodiments of the application, not all embodiments. Based on the embodiments in the application, all other embodiments obtained by those of ordinary skill in the art without creative labor are within the scope of protection of the application.

[0066] Please refer to Figure 1The embodiment of the application provides a system perspective network construction capability evaluation index and power supply parameter mapping system, which comprises:

[0067] A data acquisition module is configured to acquire power supply parameters, network construction capability indexes and topology data of the power grid in real time.

[0068] In the embodiment, the data acquisition module is configured to acquire power supply parameters, network construction capability indexes and topology data of the power grid in real time, thereby providing basic data input for subsequent dynamic tensor modeling and optimization.

[0069] A parameter acquisition unit is configured to acquire power supply operation parameters in real time, including voltage V, frequency f, virtual inertia H v , active power P and reactive power Q.

[0070] Voltage and frequency acquisition: through a synchronized phasor measurement device (PMU) arranged at a node of the power grid, voltage amplitude V and frequency f are acquired in real time at a preset sampling period (preferably 20 ms), and harmonic interference is eliminated through Fourier transform.

[0071] Virtual inertia calculation: based on a rotor motion equation of a new energy unit virtual inertia H is calculated in real time v , wherein J is unit rotational inertia, ω0 is rated angular velocity, S B is unit rated capacity.

[0072] Power data acquisition: through an SCADA system interface, active power P and reactive power Q are read, and data is normalized, and the normalization formula is:

[0073]

[0074] P max , P min and Q max , Q min are preset upper and lower limits of power.

[0075] A topology acquisition unit is configured to acquire an admittance matrix Y of the power grid node , wherein N b represents the total number of nodes of the power grid.

[0076] Admittance matrix generation: based on a topology connection relationship of the power grid, an admittance matrix Y is constructed through a node analysis method, and an element Y ij represents an equivalent admittance between node i and node j, and the calculation formula is:

[0077]

[0078] Z ij is a line impedance between nodes, The node pair admittance.

[0079] Dynamic updating mechanism: when the power grid topology changes (such as switch action or fault), the admittance matrix Y is updated in real time through the topology identification algorithm to ensure that it is consistent with the current power grid structure.

[0080] The index acquisition unit is used to obtain the network construction capability evaluation index, including the short-circuit capacity S sc , the inertia time constant T i , and the voltage stability margin M v .

[0081] Short-circuit capacity calculation: based on the node admittance matrix Y, the short-circuit current calculation model The short-circuit capacity S sc of each node is obtained, where V base is the reference voltage.

[0082] Inertia time constant extraction: according to the inertia response characteristics of synchronous machines and virtual synchronous machines, the equivalent inertia time constant T i is calculated by the following formula:

[0083]

[0084] Where Δω is the frequency deviation and ΔP is the power disturbance.

[0085] Voltage stability margin evaluation: the continuous flow method is used to calculate the voltage stability margin M v , which is defined as the power margin percentage from the current operating point to the voltage collapse point, and the calculation formula is:

[0086]

[0087] Where P max is the maximum transmission active power of the current power grid node or partition at the voltage stability limit, which is calculated by the continuous flow method and represents the upper limit of the power that the system can carry before voltage collapse; P current is the actual transmission active power of the current power grid node or partition, which is measured in real time by the data acquisition module or generated by state estimation.

[0088] The dynamic tensor modeling module is used to slice the power supply parameters, network construction capability indicators and topology data according to time windows, and build a three-dimensional dynamic tensor containing parameter dimension, time dimension and space dimension.

[0089] In this embodiment, the dynamic tensor modeling module is used to slice power parameters, network capability indicators and topology data according to time windows to construct a three-dimensional dynamic tensor containing parameter dimension, time dimension and spatial dimension, so as to preserve the spatiotemporal coupling relationship between parameters and provide structured input for subsequent coupling decomposition and optimization.

[0090] The time slicing unit is used to slice continuously collected data according to a preset time window Δt to ensure the continuity of the time dimension and the consistency of the data.

[0091] Time window setting: Based on the dynamic characteristics of the power grid, it is preferable to set Δt to a minute-level time interval (e.g., Δt = 1 min) to balance time resolution and computational complexity.

[0092] Slicing rules:

[0093] Total duration T total (e.g., 24 hours) are divided into There are 1 time slice, and each time slice contains all parameter sampling data within Δt.

[0094] For non-uniformly sampled data (such as SCADA second-level data and PMU millisecond-level data), an interpolation algorithm is used to align it to a unified timestamp. The interpolation formula is as follows:

[0095]

[0096] Where x(t) represents the interpolation result of the parameter at time t, t i and t i+1 These represent adjacent sampling times.

[0097] Tensor construction unit, which constructs a three-dimensional dynamic tensor based on the sliced ​​data. The dimensions are defined as follows:

[0098] Parameter dimension (N) p ):

[0099] Includes power supply parameters (voltage V, frequency f, virtual inertia H) v Active power (P), reactive power (Q) and grid capacity indicators (short-circuit capacity S) sc Inertial time constant T i Voltage stability margin M v N is preferred. p =8.

[0100] The parameters are arranged in order of type to form the first dimension of the tensor, for example:

[0101] T(1,:,:)=V,T(2,:,:)=f,…,T(8,:,:)=Mv ;

[0102] where T(1,:,:) = V is the node real-time voltage amplitude, representing the basic electrical state of the power grid; T(2,:,:) = f is the system real-time frequency, reflecting the balance of active power supply and demand; T(3,:,:) = H v is the equivalent inertia of the new energy unit simulation, supporting frequency stability; T(4,:,:) = P is the node real-time active power, representing the energy transmission demand; T(5,:,:) = Q is the node real-time reactive power, maintaining the voltage level; T(6,:,:) = S sc is the maximum current tolerance capacity of the node at the time of short circuit, reflecting the strength of the power grid; T(7,:,:) = T i is the system equivalent inertia time, quantifying the frequency dynamic response speed; T(8,:,:) = M v is the voltage stability margin, measuring the distance between the operating point and the voltage collapse point.

[0103] Time dimension (N t ):

[0104] determined by the number of time slices , and each slice corresponds to the second dimension of the tensor.

[0105] The time slice data is arranged in chronological order, for example, T(:,1,:) represents the data set from t=0 to t=Δt.

[0106] Space dimension (N s ):

[0107] represents the spatial position number of the grid node or partition, preferably directly using the original grid node number (for example, 1, 2, dots, N s ).

[0108] The data of each node or partition is mapped to the third dimension of the tensor, for example, T(:,:,k) represents the full parameter time series of the kth node.

[0109] Tensor data structure and storage, mathematical representation: dynamic tensor The element T(i,j,k) of the dynamic tensor is defined as the data value of the ith type of parameter, the jth time slice, and the kth node.

[0110] Data storage: stored using multi-dimensional arrays or tensor databases, supporting efficient read-write and subsequent decomposition operations.

[0111] Parameter dimension construction:

[0112] Power supply parameters (voltage V, frequency f, virtual inertia H v , active power P, and reactive power Q) are provided by the data acquisition module, and the network construction capability indicators (short circuit capacity Ssc , inertia time constant T i , voltage stability margin M v ) is generated by a real-time calculation engine.

[0113] Each parameter needs to be normalized before being stored in the tensor, and the normalization formula is:

[0114]

[0115] Where x max , x min are the historical maximum and minimum values of the parameter.

[0116] Time dimension alignment:

[0117] For parameters with different sampling frequencies, interpolation algorithms are used to unify the same timestamp, ensuring the consistency of the time dimension.

[0118] The interpolated data is sliced by time window Δt, and each slice contains sampling points (T sample is the sampling period).

[0119] Spatial dimension mapping:

[0120] If the power grid is partitioned, the spatial dimension can be represented as the partition number, and the partition rule is preferably based on electrical distance or geographical area.

[0121] The data of each node in the partition is aggregated into a partition-level parameter value by weighted average, and the weight is the node load proportion.

[0122] The coupling decomposition module is used to perform tensor decomposition on the dynamic tensor to extract the dynamic coupling features between parameters, and generate optimization constraints based on the dynamic coupling features;

[0123] In this embodiment, the coupling decomposition module is used to perform tensor decomposition on the dynamic tensor, extract the dynamic coupling features between parameters, and generate optimization constraints to support the multi-objective collaborative calculation of the subsequent distributed optimization module.

[0124] The tensor decomposition unit uses an improved Tucker decomposition method to reduce the dimension and extract features of the dynamic tensor , to obtain the core tensor and factor matrix satisfying the following decomposition relationship:

[0125] T≈G×1U (p) ×2U (t) ×3U (s) ;

[0126] Where × nR represents the product of a tensor and the nth modulus of a matrix. p R t R s The number of latent factors in terms of parameters, time, and space dimensions are respectively preferred, satisfying R. p <N p R t <N t R s <N s .

[0127] Improved Tucker decomposition algorithm:

[0128] Objective function: To improve decomposition stability, a regularization term is introduced. The objective function is defined as follows:

[0129]

[0130] Where λ is the regularization coefficient, preferably 0.1 ≤ λ ≤ 1.0, to prevent overfitting; ||·|| F The Frobenius norm is defined as the square root of the sum of squares of the elements of a tensor or matrix, and is used to quantify reconstruction errors and regularization constraints.

[0131] Alternating Least Squares (ALS): Optimizes the core tensor alternately. With factor matrix U (p) U (t) U (s) The solution steps are as follows:

[0132] Fixed factor matrix: Update core tensor

[0133]

[0134] in, This indicates a pseudo-inverse operation.

[0135] Fixed core tensor: Update each factor matrix U sequentially. (p) U (t) U (s) ,For example:

[0136]

[0137] in, For tensor The modulo-1 expansion matrix, This represents the Kronecker product.

[0138] The coupling calculation unit calculates the dynamic coupling coefficient α between parameters based on the decomposition results. ij(t), which is defined as the association strength between the ith power supply parameter and the jth network-forming capability index at time t, and is expressed as:

[0139]

[0140] where i represents the ith power supply parameter; j represents the jth network-forming capability index; t represents the index of the time slice, corresponding to the division result of the time window Δt; k represents the index of the spatial dimension, representing the number of the grid node or partition; ∈ is a constant for preventing zero; ||·||F F represents the Frobenius norm; U (p) (i, r p ) is the projection weight of parameter i on the r p th parameter factor, reflecting the contribution of the parameter to the factor component; U (t) (t, r t ) is the distribution coefficient of time slice t on the r t th time factor, representing the time dynamic characteristic; U (s) (k, r s ) is the distribution weight of spatial node / partition k on the r s th spatial factor, describing the spatial correlation.

[0141] Dynamic constraint condition: generate inequality or equality constraints according to the coupling coefficient α ij (t). For example, when α ij (t) ≥ θ (the threshold value θ is preferably 0.7), add the constraint V i · H v,j ≥ S sc,k , where V i , H v,j , and S sc,k are the voltage, virtual inertia, and short-circuit capacity, respectively.

[0142] Constraint matrix construction: convert the constraint condition into a matrix form A p x p ≤ b p , where the absolute value of the matrix element A p (m, n) is positively correlated with α ij (t).

[0143] A distributed optimization module is configured to perform multi-objective collaborative optimization on the power supply parameters through a hierarchical architecture based on the dynamic coupling characteristics and optimization constraints, where the hierarchical architecture includes an upper-layer global coordination unit and a lower-layer partition optimization unit.

[0144] In the embodiment, the distributed optimization module is used for multi-objective collaborative optimization of power supply parameters based on dynamic coupling characteristics and optimization constraints through a hierarchical architecture, and the core thereof includes a global coordination unit and a partition optimization unit, which respectively realize decomposition of a global optimization target of a power grid and efficient search within a partition.

[0145] The global coordination unit decomposes the power grid into P partitions in an ADMM (Alternating Direction Method of Multipliers) framework and defines a global optimization problem as follows:

[0146]

[0147] wherein, x p is an optimization variable of the Pth partition, f p (x p ) is a partition objective function, A p is a dynamic constraint matrix, and an element value a mm of the dynamic constraint matrix is positively correlated with a dynamic coupling coefficient α ij (t) generated by the coupling decomposition module, and satisfies a mm ∝α ij (t).

[0148] ADMM iteration steps are as follows:

[0149] Original variable update: each partition solves a local problem in parallel:

[0150]

[0151] wherein, ρ>0 is a penalty factor, preferably ρ∈[1.0, 5.0], z k is a global consistency variable, is a dual variable.

[0152] Global consistency update: aggregate results of each partition:

[0153]

[0154] Dual variable update:

[0155]

[0156] Constraint matrix generation rule:

[0157] If a dynamic coupling coefficient α ij (t)≥θ (a threshold value θ is preferably 0.7), an element a p at a corresponding position in the constraint matrix A mn is equal to α ij (t), otherwise it is set to zero.

[0158] The partition optimization unit adopts a reinforcement learning driven improved quantum particle swarm optimization (QPSO) for optimization in each partition, and the core innovation lies in introducing a quantum tunneling mechanism to enhance the global search capability.

[0159] Particle position update formula:

[0160]

[0161] wherein, represents the position of the i th particle in the d th dimension parameter in the k th iteration; represents the historical optimal position of the i th particle in the d th dimension; β is a learning factor; u is a uniformly distributed random number, is the average optimal position, N p is the total number of particles in the particle swarm, is the optimal position of the i th particle in the k th iteration; p t represents the quantum tunneling probability, which is calculated by the multi-time scale index deviation, such as the voltage fluctuation rate and the frequency change rate, and the higher the p t is, the stronger the response ability to sudden working conditions is; represents the speed of the i th particle in the d th dimension in the k+1 th iteration.

[0162] Fitness function design:

[0163] The fitness function F(x p ) comprehensively evaluates the multi-objective optimization effect, and is defined as:

[0164]

[0165] wherein, w1, w2, and w3 are weight coefficients, which are dynamically allocated by an index fusion module; is the lower limit of the voltage allowed; is the maximum and minimum value of the voltage amplitude of the whole network, which is determined by historical data statistics or topology analysis; H v,p is the virtual inertia value of the p th partition, which is calculated by the equivalent inertia calculated by the new energy unit control strategy simulation; is the maximum and minimum value of the virtual inertia; S sc,p is the short-circuit capacity of the p th partition, which is generated by calculating the equivalent impedance matrix of the node; is the allowed range of the short-circuit capacity;

[0166] Reinforcement learning adaptive parameter adjustment:

[0167] State observation: real-time input power grid operating state s = [ΔP, Δf, M v ], including power deviation, frequency fluctuation, and voltage stability margin;

[0168] Power deviation ΔP: Current active power P in the partition current Compared with the planned value P ref The difference, i.e., ΔP = P current -P ref It is used to reflect the impact of fluctuations in new energy output on the power grid.

[0169] Frequency fluctuation Δf: Real-time frequency f current With nominal frequency f nom The deviation, i.e., Δf = |f current -f nom | represents the active power balance state of the power grid.

[0170] Voltage stability margin M v The distance between the current operating point and the voltage collapse point is reflected in real time by the coupled decomposition results of the dynamic tensor modeling module.

[0171] Action space: Output QPSO parameter adjustment action a = [β, p t [w1], dynamically adjust the learning factor, tunneling probability, and voltage weights, including:

[0172] Virtual inertia setpoint H v,set The adjustment amount of the equivalent inertia of the new energy unit, with an action output of ΔH. v ∈[-H step H step ], where H step The maximum adjustment step size is determined by the constraints of the equipment control strategy.

[0173] Active power regulation ΔP adj The command to adjust the active power of a zone has an output of ΔP. adj ∈[ΔP min ,ΔP max Its range is limited by the schedulable capacity of the partition.

[0174] Reactive power regulation ΔQ adj The reactive power compensation command for the partition has an output of ΔQ. adj ∈[ΔQ min ,ΔQ max [It is limited by the reactive power capacity of the inverter.]

[0175] The reward function r t The specific formula is calculated dynamically based on the power grid stability index:

[0176]

[0177] Among them, f this the frequency fluctuation threshold, which is set according to the power grid safety standard; M v,th is the lowest allowable value of voltage stability margin; η 1, η 2, η 3 are positive reward coefficients, satisfying η 1 + η 2 + η 3 = 1 ; γ 1, γ 2 are penalty coefficients, which are dynamically adjusted according to the severity of the limit.

[0178] The agent updates the policy network parameters θ by using a proximal policy optimization (PPO) algorithm, and the specific steps include:

[0179] Interaction data storage: store the state s t , action a t , reward r t , and next state s t+1 into an experience replay pool.

[0180] Advantage function calculation: calculate the advantage estimate A t based on the time difference error δ t = r t + γV(s t+1 ) - V(s t ), where V(s) is the state value output by the value network, and γ is the discount factor.

[0181] Policy gradient update: maximize the objective function L(θ) = E t [min(ρ t (θ)A t , clip(ρ t (θ), 1-ε, 1+ε)A t )], where ρ t (θ) is the probability ratio of the new and old policies, and ε is the clipping coefficient.

[0182] The updated policy network outputs the action a t , and dynamically adjusts the following parameters of the QPSO algorithm:

[0183] Search direction: adjust the individual optimal weight in the particle velocity update formula according to ΔH v ;

[0184] Inertia weight: adjust the balance coefficient between global exploration and local development of the particle swarm based on ΔP adj and ΔQ adj .

[0185] Index fusion module, for allocating optimization weights according to the dynamic changes of multi-time scale network construction capability indexes;

[0186] In the embodiment, the index fusion module is used to dynamically allocate optimization weights by entropy weight method and TOPSIS method according to the dynamic changes of multi-time scale network construction capability indexes, so as to comprehensively reflect the evaluation needs of the power grid under different operating states.

[0187] An entropy weight calculation unit, which calculates the entropy weight based on the standardized matrix Calculates the information entropy e s With the initial weight w s , where S is the number of index types, T is the total number of time windows, r st is the standardized value of the s-th index in the t-th time window.

[0188] Standardization process:

[0189] Original data acquisition: Obtain multi-time scale network construction capability index data from the dynamic tensor modeling module, including short circuit capacity S sc , inertia time constant T i , voltage stability margin M v .

[0190] Normalization method: Adopt minimum-maximum normalization to eliminate dimensional differences, and the calculation formula is:

[0191]

[0192] Where d st is the original index value, d min and d max are the historical minimum and maximum values, respectively.

[0193] Information entropy calculation:

[0194] Information entropy definition: The information entropy e s of the s-th index reflects the discrete degree of its data distribution, and is calculated as:

[0195]

[0196] Where lnT is a normalization factor to ensure e s ∈[0,1]; T is the total number of time windows; r st is the standardized value of the s-th index in the t-th time window, satisfying r st ∈[0,1].

[0197] Weight generation: The smaller the information entropy, the higher the discrete degree of the index data, and the greater the weight w s , and the initial weight is calculated as:

[0198]

[0199] Where S is the number of time scales.

[0200] A weight correction unit corrects the initial weights by a TOPSIS (Technique for Order Preference by Similarity to Ideal Solution) method to take into account both dynamic change trends and global optimization requirements.

[0201] TOPSIS correction step:

[0202] Decision matrix construction: based on the normalized matrix R, a decision matrix D = [d st ] is constructed, where d st = w s · r st .

[0203] Positive and negative ideal solution determination:

[0204] Positive ideal solution where

[0205] Negative ideal solution where

[0206] Distance calculation: the Euclidean distance of each time window index value to the positive / negative ideal solution is calculated:

[0207]

[0208] Proximity calculation: the proximity C t of the tth time window is defined as:

[0209]

[0210] Weight correction: the weight is dynamically adjusted according to the proximity, and the correction formula is:

[0211]

[0212] where C i is a proximity state tensor modeling module and a distributed optimization module;

[0213] An edge-cloud collaborative module is configured to update a global model through federated aggregation and feed the updated model back to the dynamic tensor modeling module and the distributed optimization module.

[0214] In this embodiment, the edge-cloud collaborative module is configured to realize dynamic collaborative updating of the global model and the local model through a federated aggregation mechanism, thereby ensuring the real-time performance and consistency of the dynamic tensor modeling and distributed optimization modules.

[0215] An edge computing unit is deployed at a power grid node or a subzone and is responsible for performing local tensor decomposition and outputting a core tensor, and the specific implementation steps are as follows:

[0216] Local tensor decomposition:

[0217] Obtain local 3D dynamic tensors from the dynamic tensor modeling module Where N p N t M s These are parameters, time, and spatial dimensions, respectively.

[0218] An improved Tucker decomposition method was used to analyze T. p Decompose to obtain the local core tensor and factor matrix satisfy:

[0219]

[0220] Data on-chain verification:

[0221] Compute the local core tensor G p hash value A blockchain transaction is generated by combining timestamps, partition identifiers, and device digital signatures, and then written to the private chain node.

[0222] For raw data The data acquisition process is traceable and recorded to ensure that the data source is verifiable and has not been tampered with. Core tensor compression:

[0223] For G p Dimensionality reduction and normalization are performed to reduce the amount of data transmitted, and the Frobenius norm ||G is preferably preserved. p || F As the basis for aggregation weights.

[0224] The federated aggregation unit, deployed in the cloud, aggregates the core tensors of each partition using a weighted average algorithm to generate a global core tensor. The specific process is as follows:

[0225] Blockchain data integrity verification:

[0226] Receive data uploaded by each edge node And its hash value, calling the smart contract for verification. Consistency with on-chain records.

[0227] If the hash verification fails, the data in that partition is marked as untrusted, triggering an alarm and excluding it from the aggregation process. Weight calculation:

[0228] The aggregation weight λ is calculated based on the Frobenius norm of the core tensor of each partition. p This reflects the contribution of data from each partition:

[0229]

[0230] where P is the total number of grid partitions,

[0231] Global core tensor generation: aggregate each partition core tensor by weight:

[0232]

[0233] The formula retains the spatio-temporal feature contribution of each partition by weighted average, avoiding the dominance of single partition data in the global model.

[0234] The model updating unit updates the local model parameters by issuing the aggregated global core tensor to each edge node, and the specific steps include:

[0235] Global model issuance and blockchain record: G global is transmitted to each edge computing unit through an encrypted communication protocol, and the hash value, version number and issuance timestamp of the model update transaction are recorded in the blockchain.

[0236] Local model update:

[0237] The edge node receives G global , combines it with the local factor matrix , and reconstructs the updated local tensor model:

[0238]

[0239] The updated tensor model is input into the coupling decomposition module for dynamic coupling feature extraction and constraint generation.

[0240] Optimization module parameter synchronization: feedback the spatio-temporal feature information of G global to the distributed optimization module to adjust the search space and fitness function weight of the quantum particle swarm algorithm, ensuring the consistency of the optimization target and the global model.

[0241] Please refer to Figure 2 , the application also provides a system perspective network construction capability evaluation index and power supply In the embodiment, the edge-cloud collaborative module is used to realize dynamic collaborative update of the global model and the local model through the federal aggregation mechanism, ensuring the real-time performance and consistency of the dynamic tensor modeling and the distributed optimization module. The parameter mapping method comprises the following steps:

[0242] S1, real-time collection of power supply parameters, network construction capability indicators and topology data in power grid operation, wherein the power supply parameters include instantaneous measurement values of voltage, frequency, virtual inertia, active power and reactive power, the network construction capability indicators cover calculated values of short-circuit capacity, inertia time constant and voltage stability margin, and the topology data includes power grid node connection relationship and partition division information. The data sources include intelligent sensors, phasor measurement units (PMU) and SCADA systems, and data preprocessing is realized through filtering and denoising, anomaly detection and time stamp alignment to ensure the timeliness and reliability of input data;

[0243] S2, the preprocessed data is dynamically sliced according to a preset time window, and a sliding window mechanism is used to process continuous data stream, and the window length is adaptively adjusted according to the time scale of the dynamic process of the power grid. Non-synchronous data is aligned through linear interpolation to construct a three-dimensional dynamic tensor including parameter dimension, time dimension and space dimension, wherein the parameter dimension maps the power supply parameters and the network construction capability indicators, the time dimension corresponds to the time window sequence, and the space dimension represents the physical position number of the power grid node or partition. Dimension differences are eliminated through normalization processing, and missing data is compensated through spatio-temporal Kriging interpolation to form a complete structured tensor;

[0244] S3, the dynamic tensor is subjected to improved Tucker decomposition to obtain a low-dimensional core tensor and factor matrices, the core tensor represents the cross-dimensional coupling relationship between parameters, and the factor matrices reflect the distribution patterns of the characteristics of each dimension. Based on the decomposition results, dynamic coupling coefficients between parameters and network construction capability indicators are calculated to quantify their correlation strength, and preset threshold values are used to generate optimization constraints, including equality constraints of controllable parameters and inequality constraints of state parameters, and the constraint boundaries are dynamically adjusted according to the coupling strength;

[0245] S4, multi-objective optimization is realized by using a hierarchical architecture, the upper layer global coordination unit decomposes the power grid into multiple partitions with low electrical coupling degree based on the ADMM framework, defines a global optimization problem and coordinates the consistency of solutions of each partition; the lower layer partition optimization unit runs an improved quantum particle swarm algorithm in each partition to enhance the global search ability through quantum tunneling mechanism and avoid falling into local optimum. The upper and lower layers iteratively approach the optimal solution by alternately updating the local solution and global variable, and dynamically adjust the penalty factor to balance the convergence speed and accuracy;

[0246] S5, the information entropy of each network construction capability indicator is calculated based on the entropy weight method to evaluate the dispersion degree of its data distribution and generate an initial weight reflecting the volatility of the indicator. Further, the closeness of the indicator to the ideal solution is calculated by the TOPSIS method, and the initial weight is corrected according to the actual optimization demand to enhance the decision-making influence of the high closeness indicator. The time window rolling update and conflict detection mechanism are introduced to dynamically smooth the weight mutation, ensuring the balance and stability of multi-objective optimization;

[0247] S6, performing local tensor decomposition at the edge node, extracting the partition core tensor and encrypting and uploading to the cloud, the cloud aggregating each partition core tensor through weighted average to generate a fusion model representing global features. The aggregated global core tensor is fed back to each edge node to drive local model parameter update, while injecting a dynamic tensor modeling module to optimize the tensor decomposition rank, and synchronizing to a distributed optimization module to adjust the constraint priority. Through the federated learning framework, model co-evolution under privacy protection is realized, and the overall adaptability of the system is improved.

[0248] While embodiments of the present application have been shown and described with reference to particular embodiments thereof, it will be understood by those of ordinary skill in the art that various changes, modifications, replacements, and variations can be made thereto without departing from the principles and spirit of the present application, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A system view network capability evaluation index and power supply parameter mapping system, characterized in that, The system comprises: a data acquisition module for acquiring power supply parameters, network construction capability indicators and topology data of the power grid in real time; a dynamic tensor modeling module for slicing the power supply parameters, network construction capability indicators and topology data according to a time window to construct a three-dimensional dynamic tensor comprising a parameter dimension, a time dimension and a space dimension; a coupling decomposition module for performing tensor decomposition on the dynamic tensor to extract dynamic coupling characteristics between parameters and generating optimization constraints based on the dynamic coupling characteristics; a distributed optimization module for performing multi-objective collaborative optimization on the power supply parameters through a hierarchical architecture based on the dynamic coupling characteristics and optimization constraints, wherein the hierarchical architecture comprises an upper-layer global coordination unit and a lower-layer partition optimization unit; an index fusion module for allocating optimization weights according to dynamic changes of network construction capability indicators of multiple time scales; an edge cloud collaborative module for updating a global model through federated aggregation and feeding back the updated model to the dynamic tensor modeling module and the distributed optimization module.

2. The system of claim 1, wherein the mapping system is configured to map the network capability evaluation index to the power supply parameter based on a system view. The data acquisition module comprises: The parameter acquisition unit is used to acquire power supply parameters in real time, including voltage V, frequency f, and virtual inertia H. v Active power P, reactive power Q; Topology acquisition unit, configured to acquire an admittance matrix of a power grid node where N b is the total number of grid nodes; The index acquisition unit is configured to acquire the network construction capability indexes, including short-circuit capacity S sc , inertia time constant T i , voltage stability margin M v .

3. The system of claim 1, wherein the mapping system is configured to map the network capability evaluation index to the power supply parameter based on a system view. The dynamic tensor modeling module comprises: a time slicing unit for slicing data according to a preset time window Δt; Tensor construction unit, constructs three-dimensional dynamic tensor wherein, N p represents the number of parameter types; N t represents the number of time slices, calculated by Δt and total duration; N s represents the number of spatial numbers of grid nodes or partitions.

4. The system of claim 1, wherein the mapping system is configured to map the network capability evaluation index to the power supply parameter based on a system view. The coupling decomposition module comprises: Tensor decomposition unit, employing a modified Tucker decomposition on dynamic tensors is decomposed to obtain a core tensor and factor matrices U (p) U (t) U (s) satisfying: T ~ G x 1 U (p) x 2 U (t) x 3 U (s) ; a coupling calculation unit for calculating a dynamic coupling coefficient α between the parameters ij (t), whose expression is: where i denotes the i-th type of power supply parameter; j denotes the j-th type of network construction capability index; t denotes the index of time slice, corresponding to the division result of time window Δt; k denotes the index of spatial dimension, representing the number of grid nodes or partitions; ∈ is a constant for preventing zero; ||·||F F denotes the Frobenius norm; U (p) (i, r p ) is the projection weight of parameter i on the r p -th parameter factor, reflecting the contribution of the parameter to the factor component; U (t) (t, r t ) is the distribution coefficient of time slice t on the r t -th time factor, representing the time dynamic characteristic; U (s) (k, r s ) is the distribution weight of spatial node / partition k on the r s -th spatial factor, describing the spatial correlation.

5. The system of claim 1, wherein the mapping system is configured to map the network capability evaluation index to the power supply parameter based on a system view. The distributed optimization module comprises: a global coordination unit for decomposing the power grid into P partitions using an ADMM framework and defining a global optimization problem: Wherein, x p is the optimization variable of the Pth partition, f p (x p ) is the partition objective function, A p is a dynamic constraint matrix, and the element value a mm of the dynamic constraint matrix is positively correlated with the dynamic coupling coefficient α ij (t) generated by the coupling decomposition module, and satisfies a mm ∝α ij (t); a partition optimization unit for optimizing in each partition using a reinforcement learning driven improved quantum particle swarm optimization algorithm, specifically comprising: state observation: using real-time running states of the power grid, including power deviation, frequency fluctuation and voltage stability margin as observation inputs of the reinforcement learning agent; action space: defining power supply parameter adjustment strategies, including virtual inertia setting values and active / reactive power regulation amounts as action outputs of the agent; reward function: dynamically generating a reward signal according to power grid stability indicators, giving a positive reward when the frequency fluctuation is within a preset threshold and the voltage stability margin is improved; otherwise, applying a penalty when the threshold is exceeded or the margin decreases; strategy update: the agent learns the optimal parameter adjustment strategy through interaction with the environment, and updates the search direction and inertia weight of the quantum particle swarm optimization algorithm in real time to drive the multi-objective optimization process.

6. The system of claim 5, wherein the mapping system is configured to map the network capability evaluation index to the power supply parameter based on a system view. The particle position update of the improved quantum particle swarm optimization algorithm comprises a quantum tunneling mechanism, and the position update formula is: wherein, represents the position of the i-th particle in the d-th dimension parameter at the k-th iteration; represents the history optimal position of the i-th particle in the d-th dimension; β is a learning factor; u is a uniformly distributed random number, is the average optimal position, N p is the total number of particles in the particle swarm, is the optimal position of the i-th particle at the k-th iteration; p t represents the quantum tunneling probability, which is calculated from the multi-time scale index deviation; represents the velocity of the i-th particle in the d-th dimension at the k+1-th iteration.

7. The system of claim 1, wherein the mapping system is configured to map the network capability evaluation index to the power supply parameter based on a system view. The index fusion module comprises: An entropy weight calculation unit calculates information entropy e st ] based on the normalized matrix R = [r s and the weight w s , which satisfies: wherein lnT is a normalization factor, T is the total number of time windows; S is the number of time scales; r st is the standardized value of the sthindex in the thtime window, satisfying r st ∈[0,1] ; The weight correction unit corrects the weight by a TOPSIS method where C i is the closeness.

8. The system of claim 1, wherein the mapping system is configured to map the network capability evaluation index to the power supply parameter based on a system view. The edge cloud collaborative module comprises: An edge computing unit is deployed at a grid node or a subzone, performs local tensor decomposition and outputs a core tensor At the same time, the hash value of the grid data and the related metadata (including timestamp, subzone identifier, and device signature) are recorded to the blockchain network. a federated aggregation unit for aggregating core tensors of each partition by weighted averaging and verifying blockchain data integrity to obtain a global core tensor: Wherein, P is the total number of power grid partitions; weight λ p Satisfies the dynamic adjustment by the blockchain verification result:

9. The system of claim 8, wherein the mapping system is configured to map the network capability evaluation index to the power supply parameter based on a system view. The edge cloud collaborative module further comprises: A model updating unit updates the aggregated global core tensor G global The local tensor decomposition parameters are updated, and the hash value of the parameter updating transaction is recorded in the blockchain to ensure that the updating process is traceable and tamper-proof.

10. A system perspective network construction capability evaluation indicator and power supply parameter mapping method, applied to the system of any one of claims 1-9, comprising the following steps: S1, acquiring power supply parameters, network construction capability indicators and topology data of the power grid in real time, wherein the power supply parameters comprise voltage, frequency, virtual inertia, active power and reactive power, and the network construction capability indicators comprise short-circuit capacity, inertia time constant and voltage stability margin; S2, slice the collected data according to a preset time window to construct a three-dimensional dynamic tensor containing parameter dimension, time dimension and space dimension, wherein the space dimension is the number of power grid nodes or partitions; S3, perform tensor decomposition on the dynamic tensor to extract dynamic coupling characteristics between parameters, and generate optimization constraints based on the dynamic coupling characteristics; S4, perform multi-objective optimization through a hierarchical architecture, the upper layer global coordination decomposes the power grid into multiple partitions and defines a global optimization problem; the lower layer partition optimization optimizes in each partition using a quantum particle swarm algorithm, which includes a quantum tunneling mechanism; S5, according to the dynamic changes of the multi-time scale network construction capability index, calculate the initial weight by entropy weight method, and correct the weight by TOPSIS method; S6, perform local tensor decomposition at the edge node, aggregate the core tensors of each partition by weighted average to generate a global core tensor and feedback to the local module to update the dynamic tensor modeling and optimization constraints.

Images