New energy base networking capability assessment method based on multi-dimensional indexes

Abstract

The invention relates to the field of new energy power systems, and discloses a multi-dimensional index-based new energy base network construction capability evaluation method, which comprises the following steps of: constructing five-dimensional dynamic tensor fusion multi-source data, optimizing low-rank decomposition by combining a quantum annealing algorithm, and embedding a power grid topology hole feature constraint; adopting a federated learning framework to realize distributed parameter optimization, and introducing a quantum noise gradient to eliminate resolution difference; based on a causal graph model and back door adjustment stripping hybrid bias, generating an unbiased evaluation index; and dynamically correcting quantum annealing parameters through a closed-loop feedback mechanism. According to the method, multi-dimensional data fusion, topological vulnerability perception, privacy security calculation, unbiased evaluation and adaptive optimization are realized, and the precision and reliability of new energy base networking capability evaluation are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of new energy power systems, and in particular to a method for evaluating the networking capability of new energy bases based on multi-dimensional indicators. Background Art

[0002] With the rapid increase in the proportion of renewable energy power generation, the grid-forming capability assessment of large-scale renewable energy bases has become a key technology to ensure the stable operation of the power system. Traditional assessment methods are mostly based on single-dimensional or static indicator systems, which are difficult to adapt to the complex scenarios of renewable energy output volatility, dynamic evolution of grid topology, and coordinated control of multi-source equipment. Existing technologies usually use simplified mathematical models or empirical thresholds to perform linear weighted evaluation of grid-forming capability. Such methods have significant limitations in data representation: on the one hand, the high-dimensional nonlinear correlation between the spatiotemporal dynamic data of renewable energy bases (such as second-level power fluctuations, minute-level meteorological changes, and equipment control parameters) and the grid topology structure is difficult to be effectively modeled; on the other hand, the assessment process is easily affected by the inherent topological vulnerability of the grid (such as power congestion caused by the ring structure) and the confounding factors of environmental variables, resulting in the deviation of the assessment results from the actual operation risks.

[0003] At the distributed computing level, existing methods often rely on centralized data processing frameworks, making it difficult to meet the privacy protection and cross-regional collaborative optimization needs of massive heterogeneous data in new energy bases. Furthermore, traditional assessment models generally lack dynamic feedback mechanisms, and the coupling between assessment results and the real-time operating status of the power grid is insufficient, making it impossible to provide closed-loop decision support for the online regulation of new energy bases. These shortcomings make it difficult for existing technologies to meet the power system's needs for accurate network construction capability assessment and risk warning under high-proportion renewable energy access, hindering the improvement of the security and flexibility of new power systems.

[0004] Therefore, the present invention proposes a new energy base networking capability evaluation method based on multi-dimensional indicators to address the shortcomings of the existing technology. Summary of the Invention

[0005] In order to address the shortcomings of existing network capability assessment methods, such as insufficient fusion of multi-source heterogeneous data, lack of quantification of power grid topology vulnerability, interference of the assessment process by mixed bias, difficulty in balancing privacy security and calculation accuracy, and the inability of static models to dynamically adapt to operating states, the present invention provides a dynamic assessment method for the network capability of new energy bases that integrates quantum-topological collaborative optimization and causal closed-loop feedback.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solutions: a method for evaluating the network construction capability of a new energy base based on multi-dimensional indicators, comprising the following steps:

[0007] S1. Multidimensional Dynamic Tensor Modeling and Decomposition: A five-dimensional dynamic tensor is constructed based on the spatiotemporal data of the new energy base. This dynamic tensor is then subjected to a low-rank decomposition using a quantum annealing algorithm, where the quantum annealing process is dynamically constrained by topological hole characteristics.

[0008] S2. Grid topology feature extraction: Perform continuous coherent analysis of the grid topology to extract hole features that characterize grid vulnerability, and input the hole features into the quantum annealing process in step S1;

[0009] S3. Cross-scale federated parameter optimization: Based on the federated learning framework, distributed parameter updates are performed on the dynamic tensor decomposed in step S1. The update process introduces quantum noise gradients to eliminate data resolution differences.

[0010] S4. Causal-driven evaluation calculation: Based on the decomposition results of step S1 and the topological features of step S2, a causal graph model is constructed and backdoor adjustments are performed to eliminate the confounding bias of the evaluation indicators to generate unbiased evaluation indicators.

[0011] S5. Closed-loop feedback iterative optimization: Feedback the evaluation indicators generated in step S4 to the quantum annealing process in step S1, dynamically adjust the rank constraint parameters of the tensor decomposition, and form an evaluation-feedback closed-loop optimization mechanism.

[0012] Preferably, in S1, the dimensions of the five-dimensional dynamic tensor include time (t∈T), spatial position (s∈S), device type (d∈D), control parameters (c∈C) and environmental variables (e∈E);

[0013] The low-rank decomposition is achieved by minimizing the objective function:

[0014]

[0015] in: The original dynamic tensor; The decomposed low-rank tensor; Rank constraint weight coefficient.

[0016] Preferably, the Hamiltonian of the quantum annealing algorithm is designed as follows:

[0017]

[0018] in, The coupling strength between qubits i and j; The Pauli operator of qubit i; Time-dependent transverse magnetic field strength; The topological hole features extracted in step S2.

[0019] Preferably, the power grid topology feature extraction includes the following steps:

[0020] The simplicial chain complex is generated by constructing the Vietoris-Rips complex of the power grid topology, which is defined as:

[0021]

[0022] Where V={v1,v2,…,v n}: grid node set; distance threshold;

[0023] Calculate the 0-dimensional and 1-dimensional Betti numbers β0, β1;

[0024] Among them, β0=dimH0(VR(V,∈)): the number of connected components; β1=dimH1(VR(V,∈)): the number of ring structures.

[0025] Preferably, the hole feature H1(s) is generated by integral mapping:

[0026]

[0027] Among them, ∈′∈[∈ min ,∈ max ]: dynamic distance threshold variable;

[0028] Satisfy 0<∈ min <∈ max ≤max u,v∈V ||uv||;

[0029] β1(s,∈′): 1-dimensional Betti number at spatial position s and distance threshold ∈′.

[0030] Preferably, the steps of cross-scale federated parameter optimization are:

[0031] The client performs a local dynamic tensor T (k) Perform slice dimensionality reduction to generate core tensors:

[0032]

[0033] in:

[0034] n-th dimension factor matrix (n=1,2,…,5n=1,2,…,5)

[0035] Rank parameter after dimension reduction of the nth dimension;

[0036] The server aggregates the core tensors of each client and uses orthogonal constraints Realize cross-dimensional parameter fusion.

[0037] Preferably, the parameter update adopts the gradient rule of quantum noise injection

[0038]

[0039] in:

[0040] Model parameter vector for the kth iteration

[0041] Learning rate

[0042] Quantum bit XOR operation

[0043] Q(x)=x+N(0,σ 2 ): quantum noise channel function, is Gaussian noise.

[0044] Preferably, the causal drive evaluation calculation comprises the following steps:

[0045] The controlled variables X include short-circuit capacity ratio and voltage deviation rate;

[0046] The confounding factor Z = (H0(s), H1(s)) is the joint topological feature, where:

[0047] H0(s) = β0(s): connectivity feature at spatial position s, H1(s): hole feature as defined in claim 5;

[0048] Evaluation indicator Y∈[0,1]: Network building capability score.

[0049] Preferably, the backdoor adjustment formula is:

[0050] P(Y|do(X))=∑ z∈Z P(Y|X,z)P(z);

[0051] Where P(Y|X,z) is the conditional probability of Y given the control variable X and the confounding factor z; P(z) is the joint distribution probability of the confounding factor z.

[0052] Preferably, the closed-loop feedback iterative optimization includes the following:

[0053] The dynamic evaluation matrix M is composed of the core tensor T core Calculated with backdoor adjustment results:

[0054] M=T core ×diag(P(Y|do(X)));

[0055] Where, diag(·): converts the probability vector into a diagonal matrix;

[0056] The feedback mechanism is realized by adjusting the Hamiltonian:

[0057]

[0058] in, Feedback gain coefficient; ||M|| F : Frobenius norm of the matrix M.

[0059] The present invention provides a method for evaluating the network construction capability of new energy bases based on multi-dimensional indicators. It has the following beneficial effects:

[0060] 1. The present invention realizes the efficient fusion and unified characterization of multi-source heterogeneous data in new energy bases by constructing a five-dimensional dynamic tensor that includes time, space, equipment type, control parameters and environmental variables. The tensor model performs low-rank decomposition through the quantum annealing algorithm, which effectively suppresses the interference of noise and redundant information while retaining key spatiotemporal evolution characteristics. Among them, the topological hole feature constraints in the quantum annealing process enable the decomposition results to adaptively focus on vulnerable areas of the power grid, significantly improving the model's ability to analyze complex operating states. Compared with traditional matrix or low-dimensional tensor methods, the present invention has significant advantages in data representation dimensions and dynamic correlation modeling, providing a high-fidelity data foundation for subsequent cross-scale analysis.

[0061] 2. Based on the theory of persistent homology, the present invention accurately extracts the hole characteristics (H1(s)) in the power grid topology by constructing a multi-scale Vietoris-Rips complex. This feature quantifies the stability of the ring structure under different distance thresholds through dynamic integral mapping, overcoming the limitations of single-scale analysis. Furthermore, H1(s) is embedded in the quantum annealing Hamiltonian to form a closed-loop mechanism of "topological fragility-quantum constraint tensor rank optimization", so that the model dynamically adjusts the rank constraint weights of different regions during the decomposition process. This collaborative perception mechanism not only enhances the model's sensitivity to power grid chain failures, but also provides a quantitative basis for redundant design and weak link reinforcement in power grid planning.

[0062] 3. The present invention proposes a federated learning framework based on orthogonal constraints and quantum noise gradients, which effectively resolves the contradiction between distributed data heterogeneity and privacy protection. The client reduces the dimension by slicing the core tensor, mapping the high-dimensional features of the local data to a low-rank space, and achieving feature sharing while protecting the privacy of the original data; the server fuses global parameters through orthogonal projection to ensure the geometric structure consistency of multi-source data. In addition, quantum noise gradient injection destroys the gradient direction deviation of the low-resolution client through bit-level perturbations, significantly improving the robustness of cross-scale data training. This method achieves collaborative optimization and precision balance of multi-client models while ensuring data security.

[0063] 4. By constructing a causal graph model containing control variables, confounding factors and evaluation indicators, the present invention uses a backdoor adjustment formula to remove the implicit interference of the inherent topological structure of the power grid on the evaluation results. Based on non-parametric conditional probability estimation and kernel density distribution calculation, the causal effects of control variables such as short-circuit capacity ratio and voltage deviation rate are accurately quantified to generate an unbiased network construction capability score. The Bayesian network dynamic update mechanism is further introduced. When the quantum annealing parameters or topological characteristics change, the causal model parameters are adjusted in real time to ensure that the evaluation results are strictly synchronized with the actual state of the power grid. This method breaks through the limitations of traditional statistical correlation analysis and provides causal-level decision support for the precise regulation of new energy bases.

[0064] 5. The present invention encodes the causal reasoning results into the feedback term of the quantum annealing Hamiltonian through a dynamic evaluation matrix (M), forming a closed-loop iterative mechanism of "evaluation feedback-optimization". The matrix fuses the low-rank characteristics of the core tensor and the unbiased evaluation index through tensor multiplication, and its Frobenius norm directly reflects the spatiotemporal distribution intensity of the network-building capability. The feedback mechanism dynamically optimizes the rank constraint parameters of the tensor decomposition by adjusting the longitudinal magnetic field term in the Hamiltonian, so that the model continues to focus on high-value spatiotemporal regions. This closed-loop system breaks through the rigidity of the traditional static evaluation model, realizes the dynamic adaptation of the evaluation accuracy and the operating status, and provides self-evolution capabilities for the long-term reliable operation of the new energy base. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] Figure 1 This is a flow chart of the new energy base network capability assessment method based on multi-dimensional indicators. DETAILED DESCRIPTION

[0066] The following will clearly and completely describe the technical solution of the present invention in conjunction with the accompanying drawings. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0067] Please see the attached Figure 1 The embodiment of the present invention provides a method for evaluating the networking capability of a new energy base based on multi-dimensional indicators. The following describes in detail each step of the method of the present invention.

[0068] A method for evaluating the network construction capability of a new energy base based on multi-dimensional indicators includes the following steps:

[0069] S1. Multidimensional Dynamic Tensor Modeling and Decomposition: A five-dimensional dynamic tensor is constructed based on the spatiotemporal data of the new energy base. This dynamic tensor is then subjected to a low-rank decomposition using a quantum annealing algorithm, where the quantum annealing process is dynamically constrained by topological hole characteristics.

[0070] In this embodiment, step S1 constructs a dynamic tensor model by fusing multi-source heterogeneous data, and implements low-rank decomposition optimization of the tensor based on the quantum annealing algorithm. The specific implementation process is as follows:

[0071] Data collection and preprocessing:

[0072] This example first collects spatiotemporal multi-source data from the new energy base, including time series data, spatial topology data, equipment parameters, and environmental variables. Time series data includes new energy unit output, energy storage charging and discharging power, and meteorological monitoring data. Spatial data includes grid node coordinates and equipment layout topology. Equipment parameters include control variables such as inverter modulation ratio and virtual synchronous generator damping coefficient. Environmental variables include external factors such as terrain elevation and vegetation cover index.

[0073] When preprocessing the above data, the timestamps of multi-source data are aligned using the dynamic time warping algorithm to solve the problem of time axis inconsistency caused by differences in sampling frequency. For spatial data, the Kriging interpolation algorithm is used to convert discrete node data into a continuous spatial field to ensure the integrity of the spatial dimension. At the same time, based on the 3σ criterion, abnormal data points are identified and eliminated, and linear interpolation of adjacent data is used to fill in the gaps to ensure data continuity.

[0074] Five-dimensional dynamic tensor construction:

[0075] After completing data preprocessing, construct a five-dimensional dynamic tensor The dimensions are defined as follows:

[0076] Time dimension T: covers time series data from minute to year in the form of a sliding window, and the window length is dynamically adjusted according to evaluation requirements;

[0077] Spatial dimension S: divides the geographical area into equally spaced grids or directly maps to grid node numbers, with each spatial unit associated with latitude and longitude coordinates;

[0078] Device type dimension D: uses classification codes to identify different devices (e.g., photovoltaic = 1, wind power = 2, energy storage = 3) to support collaborative analysis of multiple devices;

[0079] Control parameter dimension C: Normalizes the inverter modulation ratio, damping coefficient and other parameters and maps them to the [0, 1] interval;

[0080] Environmental variable dimension E: z-score normalization is performed on environmental data such as temperature and irradiance to eliminate dimensional differences.

[0081] Each element of the dynamic tensor, T(t,s,d,c,e), represents the combined state of a control parameter c and an environmental variable e at time t, spatial location s, and device type d. This method maps the complex operating state of a new energy base into a high-order tensor structure, providing a unified data source for subsequent quantitative analysis.

[0082] Quantum Dynamic Tensor Decomposition:

[0083] In order to extract the core features in the tensor and reduce the computational complexity, this embodiment uses a quantum annealing algorithm to perform low-rank decomposition on the dynamic tensor.

[0084] The optimization goal is to minimize the following function:

[0085]

[0086] in, is the decomposed low-rank tensor, and λ is the rank constraint weight coefficient, which is used to balance the reconstruction error and model complexity.

[0087] To achieve the above goals, this embodiment designs a quantum annealing Hamiltonian that integrates topological features:

[0088]

[0089] Where, J ij The coupling strength between qubits is set based on the correlation between tensor dimensions (e.g., temporally adjacent dimensions give higher coupling strength). and is the Pauli operator, which characterizes the spin state of the quantum bit and the transverse magnetic field effect respectively; Γ(t) is the transverse magnetic field intensity that decays with time, which is used to control the quantum tunneling effect; H1(s) is the topological hole feature extracted in step S2. By dynamically constraining the distribution of the tensor rank, the decomposition result prioritizes retaining the refined information of the vulnerable areas of the power grid.

[0090] Preferably, an adaptive parameter update strategy is adopted in the quantum annealing process: the transverse magnetic field strength Γ(t) decays exponentially with the number of iterations, the initial value is set according to the tensor scale, and the decay rate is related to the data dynamics; the coupling strength J ij The Pearson correlation coefficient of each dimension of the tensor is dynamically adjusted to reflect the correlation strength of multidimensional data.

[0091] Through the above method, this embodiment embeds the topological characteristics of the power grid into the quantum annealing process, so that the tensor decomposition not only reduces the data dimension, but also retains the spatial vulnerability information that is crucial for the assessment of network capability, laying the foundation for subsequent cross-scale analysis and causal reasoning.

[0092] S2. Grid topology feature extraction: Perform continuous coherent analysis of the grid topology to extract hole features that characterize grid vulnerability, and input the hole features into the quantum annealing process in step S1;

[0093] In this embodiment, step S2 extracts the hole characteristics of the power grid topology through continuous coherence analysis to provide dynamic constraints for the quantum annealing process. The specific implementation process is as follows:

[0094] Power grid topology data modeling

[0095] This embodiment first obtains the grid topology data of the new energy base, including the coordinate set of all nodes V = {v1, v2, ..., v n}, where each node v i Contains longitude and latitude coordinate information. Calculates the Euclidean distance between nodes based on node coordinates and constructs a distance matrix Matrix element D ij =||v i -v j || represents the spatial connection relationship between nodes.

[0096] Vietoris-Rips complex construction:

[0097] In order to analyze the multi-scale structural characteristics of the power grid topology, this embodiment uses the Vietoris-Rips complex to perform geometric abstraction on the node set. min ,∈ max ], generates the corresponding simplicial complex:

[0098]

[0099] Among them, ∈ min and ∈ max are the preset minimum and maximum distance thresholds respectively.

[0100] Preferably, ∈ min Set to 10% of the diameter of the grid node set,∈ max It is set to 50% of the node set diameter to ensure that the complete evolution of the grid topology from sparse to dense connections is covered.

[0101] Calculation of persistent homology and Betti number:

[0102] For each distance threshold ∈′, the global and local structural features of the power grid topology are extracted by calculating the homology group of the simplicial complex:

[0103] Boundary matrix generation: Construct the boundary operator matrix of the fire-dimensional chain group (k=0,1), where the 0-dimensional chain group represents the node connected component and the 1-dimensional chain group represents the closed loop;

[0104] Homology group solution: By calculating the matrix rank, we can get the k-dimensional closed chain group Z k With edge chain group B k , and then obtain the homology group H k =Z k / B k ;

[0105] Betti number extraction: Calculate the dimension of the homology group, that is, the Betti number:

[0106] β k =dimH k ;

[0107] Among them, β0 represents the number of connected components, and β1 represents the number of ring structures (holes).

[0108] Hole feature map:

[0109] To quantify the vulnerability of the power grid topology, this embodiment performs a multi-scale integration on the ring structure at the spatial position s:

[0110]

[0111] Where β1(s,∈′) represents the 1-dimensional Betti number near location s at a distance threshold ∈′, and its value is obtained through persistent homology analysis of the local node set. The integral result, H1(s), reflects the stability of the ring structure at that location at different connection scales. Higher values ​​of H1(s) indicate greater vulnerability due to the ring topology in that region, making it more susceptible to power congestion or cascading failures.

[0112] Coupling of topological features and quantum annealing:

[0113] This embodiment uses the hole signature H1(s) as a dynamic constraint input into the quantum annealing process in step S1, specifically as an additional term λ·H1(s) in the Hamiltonian. Through this coupling mechanism, the quantum annealing algorithm prioritizes the high-rank characteristics of vulnerable areas in the power grid when optimizing tensor decomposition, thereby improving the assessment model's ability to detect key risk areas.

[0114] S3. Cross-scale federated parameter optimization: Based on the federated learning framework, distributed parameter updates are performed on the dynamic tensor decomposed in step S1. The update process introduces quantum noise gradients to eliminate data resolution differences.

[0115] In this embodiment, step S3 implements distributed parameter optimization of dynamic tensors through a cross-scale federated learning framework, and introduces quantum noise gradient to eliminate data resolution differences. The specific implementation process is as follows:

[0116] Client-side local tensor processing:

[0117] In this embodiment, the low-rank tensor λ decomposed in step S1 is first distributed to each federated learning client. The client performs slicing and dimensionality reduction on the dynamic tensor based on the local data distribution characteristics. Specifically, client k divides its local tensor T (k) Perform multimodal factorization:

[0118]

[0119] in, The factor matrix generated by quantum annealing optimization in step S1 satisfies the orthogonality constraint Ensure the independence of parameters in each dimension; R n is the rank parameter after the n-th dimension reduction, which is adaptively adjusted according to the tensor modal variance contribution rate.

[0120] Preferably, higher rank parameters are retained for high variance dimensions (such as the time dimension T) to capture dynamic evolution details; and rank parameters are reduced for low variance dimensions (such as the device type dimension D) to suppress redundant information.

[0121] Federated parameter aggregation and fusion:

[0122] Each client will reduce the core tensor Upload to the federated learning server. The server uses the orthogonal projection algorithm to aggregate multi-client parameters, including:

[0123] Core tensor alignment: align the core tensors of each client based on the spatiotemporal dimensions to fill in missing units caused by differences in data distribution;

[0124] Factor matrix fusion: factor matrix U n Perform singular value decomposition (SVD) to extract the principal component directions and update the global factor matrix;

[0125] Parameter broadcast: Distribute the fused global factor matrix to each client for the next round of local model training.

[0126] Quantum noise gradient update: To eliminate the interference of data resolution differences between clients on parameter updates, this embodiment injects quantum noise into the gradient descent process. Specifically, the server calculates the global model parameter gradient Then, the following noise injection rules are executed:

[0127]

[0128] Where w k is the model parameter vector of the kth iteration, η is the learning rate, sgn(·) is the sign function, represents the qubit XOR operation, Q(x) is the quantum noise channel function, which is defined as:

[0129] Q(x)=x+N(0,σ 2 );

[0130] Among them, N(0,σ 2 ) is Gaussian noise, the noise intensity σ 2 Positively correlated with the gradient amplitude.

[0131] Preferably, set is the noise gain factor.

[0132] Through the above operations, quantum noise is mixed with classical gradients at the binary bit level, destroying the local gradient direction consistency of low-resolution clients, thereby suppressing their parameter update deviations.

[0133] Preferably, to enhance the data security and credibility of the federated learning framework, the communication between the client and the server can use blockchain technology to achieve data verification and traceability. Specifically, it includes:

[0134] Data signing and encryption: The client uses an asymmetric encryption algorithm (such as RSA) to sign the local core tensor. Encrypt and attach a digital signature based on a private key to ensure data integrity and source traceability;

[0135] Consortium chain storage and verification: Encrypted data and its hash value are uploaded to a consortium chain jointly maintained by multiple parties in the new energy base (such as power generation companies and grid operators). Before aggregating parameters, the server verifies the legitimacy of the client's public key signature through a smart contract and compares the hash value on the chain with the hash value of the received data, aggregating only the data that passes the verification.

[0136] Access control: Smart contracts define data access rules. Authorization servers can only obtain encrypted data after obtaining permission from the client, and all data operation logs are recorded in the blockchain, making the entire process auditable.

[0137] This solution, as a security-enhancing extension of federated learning, does not affect the core parameter optimization process and is compatible with orthogonal projection and quantum noise gradient mechanisms.

[0138] S4. Causal-driven evaluation calculation: Based on the decomposition results of step S1 and the topological features of step S2, a causal graph model is constructed and backdoor adjustments are performed to eliminate the confounding bias of the evaluation indicators to generate unbiased evaluation indicators.

[0139] In this embodiment, step S4 eliminates the mixed bias of the evaluation indicators through the causal graph model and backdoor adjustment to generate an unbiased network construction capability evaluation result. The specific implementation process is as follows:

[0140] Causal graph model construction:

[0141] This embodiment constructs a causal graph model based on the low-rank tensor λ decomposed in step S1 and the topological hole feature H1(s) extracted in step S2 to quantify the impact of control variables on networking capabilities.

[0142] The causal graph model contains three types of nodes:

[0143] Control variables X: These include the short-circuit capacity ratio (SCR) and the voltage deviation rate (VDR), both of which directly affect the grid-building capability of the new energy base.

[0144] Confounding factor Z: It is composed of topological features H0(s) (connectivity features) and H1(s) (hole features), reflecting the potential interference of the inherent properties of the power grid structure on the evaluation indicators;

[0145] Evaluation indicator Y: Network construction capability score, obtained by multi-objective fusion calculation of voltage stability, frequency response and fault ride-through capability through fuzzy logic synthesis algorithm, with a value range of [0,1].

[0146] In a causal diagram, there is a direct causal path X→Y between the control variable X and the evaluation indicator Y, while the confounding factor Z affects both X and Y, forming a backdoor path X←Z→Y. To eliminate confounding bias, the backdoor path must be blocked to accurately estimate the causal effect of X on Y.

[0147] Backdoor adjustment calculation: To block the interference of confounding factors, this embodiment uses the backdoor adjustment formula to calculate the intervention effect:

[0148] P(Y|do(X))=∑ z∈Z P(Y|X,z)P(z);

[0149] Where P(Y|X,z) is the conditional probability of Y given the control variable X and the confounding factor z, and P(z) is the joint distribution probability of the confounding factor.

[0150] The specific implementation includes the following steps:

[0151] Conditional probability estimation: based on the core tensor decomposed in step S1 Count the number of data units that meet the conditions X = x and Z = z, and calculate the non-parametric conditional probability:

[0152]

[0153] Where δ(x,z) is an indicator function, which takes the value 1 when the control variable X=x and the confounding factor Z=z, and takes the value 0 otherwise.

[0154] Joint distribution calculation:

[0155] The joint probability distribution of the confounding factor Z is calculated by kernel density estimation (KDE):

[0156]

[0157] Where N is the total number of data samples, K(·) is the Gaussian kernel function, and h is the bandwidth parameter.

[0158] Preferably, the Silverman criterion is adaptively set in is the sample standard deviation.

[0159] Intervention effect score:

[0160] Integrate all possible confounding factor values ​​z∈Z and calculate the unbiased evaluation index:

[0161] Y unbiased =∫ z P(Y|X=x,z)P(z)dz;

[0162] Preferably, Monte Carlo sampling is used to approximate the integral and reduce the computational complexity by randomly extracting z samples.

[0163] Dynamic causal graph update

[0164] To adapt to the dynamic operation state of the new energy base, this embodiment introduces a Bayesian network update mechanism:

[0165] When the quantum annealing process in step S1 adjusts the tensor decomposition rank parameter, the core tensor It is then updated, triggering a re-estimation of the conditional probability P(Y|X,z);

[0166] When the topological feature H1(s) extracted in step S2 changes, the joint distribution P(z) of the confounding factor Z is updated synchronously to ensure the real-time consistency between the causal model and the actual state of the power grid.

[0167] S5. Closed-loop feedback iterative optimization: Feedback the evaluation indicators generated in step S4 to the quantum annealing process in step S1, dynamically adjust the rank constraint parameters of the tensor decomposition, and form an evaluation-feedback closed-loop optimization mechanism.

[0168] In this embodiment, step S5 dynamically embeds the evaluation results into the quantum annealing optimization process through a closed-loop feedback mechanism, thereby achieving collaborative iterative optimization of network building capability evaluation and tensor decomposition parameters. The specific implementation process is as follows:

[0169] Dynamic evaluation matrix generation:

[0170] Based on the unbiased evaluation index P(Y|do(X)) calculated in step S4, this embodiment constructs a dynamic evaluation matrix M to quantify the comprehensive score of network building capabilities under different spatiotemporal conditions. The construction method is:

[0171] M=T core ×diag(P(Y|do(X)));

[0172] Where, T core is the core tensor decomposed in step S1. diag(·) converts the probability vector obtained from the backdoor adjustment into a diagonal matrix. Each element of the matrix M(t,s,d,c,e) represents the network capability score after eliminating confounding bias at time t, spatial location s, device type d, control parameter c, and environmental variable e. Through tensor multiplication, the causal inference results are coupled with the low-rank features of the core tensor to form a temporally and spatially resolved evaluation matrix.

[0173] Quantum annealing Hamiltonian feedback adjustment:

[0174] In order to feed the evaluation results back to the annealing process, this embodiment dynamically modifies the Hamiltonian in step S1:

[0175]

[0176] Where H is the original Hamiltonian, μ is the feedback gain coefficient, ||M|| F is the Frobenius norm of the evaluation matrix, is the qubit spin operator. This correction term converts the global amplitude information of the evaluation matrix into the longitudinal magnetic field strength of the qubit, thereby adjusting the rank constraint weight of the tensor decomposition. Preferably, the feedback gain coefficient μ is positively correlated with the initial rank parameter of the core tensor to ensure that the feedback strength matches the model complexity.

[0177] Dynamic optimization of rank constraint parameters

[0178] According to the modified Hamiltonian H new , re-solve the tensor low-rank decomposition problem:

[0179]

[0180] Through iterative optimization of quantum annealing algorithm, the rank constraint parameter λ is dynamically adjusted so that the core tensor T core While retaining the detailed features of high-scoring areas, redundant information in low-scoring areas is suppressed.

[0181] Preferably, an adaptive simulated annealing strategy is used: when evaluating the matrix norm ||M|| F As it rises, we increase λ to strengthen the low-rank constraint; when ||M|| FWhen decreases, λ is reduced to improve the reconstruction accuracy.

[0182] Closed-loop iterative mechanism:

[0183] This embodiment establishes a closed-loop optimization process of evaluation and feedback

[0184] Initial iteration: Perform tensor decomposition based on the initial Hamiltonian H to generate the core tensor T core ;

[0185] Causal evaluation: Calculate the unbiased evaluation index through step S4 and construct the dynamic evaluation matrix M;

[0186] Feedback correction: Update Hamiltonian to H according to M new , and re-optimize the tensor decomposition parameters;

[0187] Termination condition: When the Frobenius norm change rate of the evaluation matrix is ​​lower than the preset threshold, the iteration is terminated and the final evaluation result is output.

[0188] In specific implementation scenarios, the dynamic adjustment of quantum annealing parameters can be further combined with deep reinforcement learning algorithms. Specifically,

[0189] State space definition: The real-time characteristics of the five-dimensional dynamic tensor T, the topological hole characteristics H1(s) and the unbiased evaluation index Y unbiased As the state input of the Deep Q Network (DQN);

[0190] Action space design: The output action is the quantum annealing parameter adjustment, including the amplitude correction of the transverse magnetic field intensity Γ(t), the coupling intensity J ij The weight distribution of

[0191] Reward function construction: To evaluate the Frobenius norm of the matrix ||M|| F The improvement of is used as a reward signal to drive DQN to learn the optimal parameter adjustment strategy;

[0192] Multi-scenario training: By simulating different renewable energy output combinations (PV / wind power ratios), grid topologies (radial / ring), and load fluctuation scenarios, the DQN model's generalization capabilities are trained to achieve adaptive parameter optimization under complex operating conditions. This extended implementation enhances the intelligent level of parameter adjustment through the synergy of reinforcement learning and quantum annealing.

[0193] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A method for evaluating the network construction capability of a new energy base based on multi-dimensional indicators, characterized in that: The following steps are involved: S1. Multidimensional Dynamic Tensor Modeling and Decomposition: A five-dimensional dynamic tensor is constructed based on the spatiotemporal data of the new energy base. This dynamic tensor is then subjected to a low-rank decomposition using a quantum annealing algorithm, where the quantum annealing process is dynamically constrained by topological hole characteristics. S2. Grid topology feature extraction: Perform continuous coherent analysis of the grid topology to extract hole features that characterize grid vulnerability, and input the hole features into the quantum annealing process in step S1; S3. Cross-scale federated parameter optimization: Based on the federated learning framework, distributed parameter updates are performed on the dynamic tensor decomposed in step S1. The update process introduces quantum noise gradients to eliminate data resolution differences. S4. Causal-driven evaluation calculation: Based on the decomposition results of step S1 and the topological features of step S2, a causal graph model is constructed and backdoor adjustments are performed to eliminate the confounding bias of the evaluation indicators to generate unbiased evaluation indicators. S5. Closed-loop feedback iterative optimization: Feedback the evaluation indicators generated in step S4 to the quantum annealing process in step S1, dynamically adjust the rank constraint parameters of the tensor decomposition, and form an evaluation-feedback closed-loop optimization mechanism.

2. The method for evaluating the network construction capability of a new energy base based on multi-dimensional indicators according to claim 1 is characterized in that: In S1, the dimensions of the five-dimensional dynamic tensor include time (t∈T), spatial position (s∈S), device type (d∈D), control parameters (c∈C) and environmental variables (e∈E); The low-rank decomposition is achieved by minimizing the objective function: in: The original dynamic tensor; The decomposed low-rank tensor; Rank constraint weight coefficient.

3. The method for evaluating the network construction capability of a new energy base based on multi-dimensional indicators according to claim 1 is characterized in that: The Hamiltonian of the quantum annealing algorithm is designed as: in, The coupling strength between qubits i and j; The Pauli operator of qubit i; Time-dependent transverse magnetic field strength; The topological hole features extracted in step S2.

4. The method for evaluating the network construction capability of a new energy base based on multi-dimensional indicators according to claim 1 is characterized in that: The grid topology feature extraction comprises the following steps: The simplicial chain complex is generated by constructing the Vietoris-Rips complex of the power grid topology, which is defined as: Where V={v1,v2,…,v n }: grid node set; distance threshold; Calculate the 0-dimensional and 1-dimensional Betti numbers β0, β1; Among them, β0=dimH0(VR(V,∈)): the number of connected components; β1=dimH1(VR(V,∈)): the number of ring structures.

5. The method for evaluating the network construction capability of a new energy base based on multi-dimensional indicators according to claim 1 is characterized in that: The hole feature H1(s) is generated by integral mapping: Among them, ∈′∈[∈ min ,∈ max ]: dynamic distance threshold variable; ∈ min ,∈ max ∈R + :Satisfies 0<∈ min <∈ max ≤max u,v∈V ||uv||; β1(s,∈′): 1-dimensional Betti number at spatial position s and distance threshold ∈′.

6. The method for evaluating the network construction capability of a new energy base based on multi-dimensional indicators according to claim 1 is characterized in that: The steps of cross-scale federated parameter optimization are: The client performs a local dynamic tensor T (k) Perform slice dimensionality reduction to generate core tensors: in: n-th dimension factor matrix (n=1,2,…,5n=1,2,…,5) Rank parameter after dimension reduction of the nth dimension; The server aggregates the core tensors of each client and uses orthogonal constraints Realize cross-dimensional parameter fusion.

7. The method for evaluating the network construction capability of a new energy base based on multi-dimensional indicators according to claim 1 is characterized in that: The parameter update adopts the gradient rule of quantum noise injection in: Model parameter vector for the kth iteration Learning rate Quantum bit XOR operation Q(x)=x+N(0,σ 2 ): quantum noise channel function, is Gaussian noise.

8. The method for evaluating the network construction capability of a new energy base based on multi-dimensional indicators according to claim 1 is characterized in that: The causal drive evaluation calculation includes the following steps: The controlled variables X include short-circuit capacity ratio and voltage deviation rate; The confounding factor Z = (H0(s), H1(s)) is the joint topological feature, where: H0(s) = β0(s): connectivity feature at spatial position s, H1(s): hole feature as defined in claim 5; Evaluation indicator Y∈[0,1]: Network building capability score.

9. The method for evaluating the network construction capability of a new energy base based on multi-dimensional indicators according to claim 1 is characterized in that: The backdoor adjustment formula is: P(Y∣do(X))=∑ z∈Z P(Y∣X,z)P(z); Where P(Y|X,z) is the conditional probability of Y given the control variable X and the confounding factor z; P(z) is the joint distribution probability of the confounding factor z.

10. The method for evaluating the network construction capability of a new energy base based on multi-dimensional indicators according to claim 1 is characterized in that: The closed-loop feedback iterative optimization includes the following: The dynamic evaluation matrix M is composed of the core tensor T core Calculated with backdoor adjustment results: M=T core ×diag(P(Y∣do(X))); Where, diag(·): converts the probability vector into a diagonal matrix; The feedback mechanism is realized by adjusting the Hamiltonian: in, Feedback gain coefficient; ||M|| F : Frobenius norm of the matrix M.

Images