Method for evaluating operation reliability of offshore wind farm by fusing high-dimensional spatiotemporal features of multiple wind turbines and network reconfiguration

Abstract

The application discloses a kind of offshore wind farm operation reliability evaluation methods of fusion multi-wind machine high-dimensional space-time characteristics and network reconstruction.First, extract wind farm historical wind speed and collect power system topological information, fusion as unified condition and gate feature;Second, input to two-stage residual condition diffusion model to generate multi-wind machine wind speed prediction sample;Then, combined with equipment failure rate sampling to construct wind speed-fault joint system state sample;Next, solve the network reconstruction optimization model that meets power system operation constraint, generate switching action and dispatching strategy for spare cable, and calculate actual deliverable power and output loss;Finally, calculate multidimensional operation reliability index hour by hour, and early warning is carried out to operation risk.The application improves the ability of space-time fluctuation characteristics of wind speed, and the evaluation result is more accurate, effectively guides power supply recovery after fault.

Description

Technical Field

[0001] This invention relates to the field of risk assessment technology for offshore wind power systems, and in particular to a method for assessing the operational reliability of offshore wind farms that integrates the high-dimensional spatiotemporal characteristics of multiple wind turbines with network reconstruction. Background Technology

[0002] With the rapid development of the wind power industry, offshore wind power has been developed on a large scale due to its abundant wind energy resources and high power generation efficiency. However, offshore wind farms are usually far from the mainland and are highly susceptible to the complex marine environment, resulting in strong fluctuations and uncertainties in wind power output. Furthermore, equipment such as wind turbines and submarine collection cables experience frequent failures and are difficult to repair. Against this backdrop, accurately assessing the operational risks of offshore wind farms is of significant practical importance for the scientific management and control of these risks.

[0003] The operational reliability assessment of offshore wind farms generally includes three main stages: condition sampling, condition analysis, and index statistics. For offshore wind farms, the validity of the assessment results highly depends on the accuracy of the spatiotemporal correlation modeling of wind speeds across multiple turbines, and the comprehensiveness of the post-failure system condition analysis. However, existing reliability assessment methods still have significant limitations in these two key stages: On the one hand, in the system state sampling stage, traditional research widely uses Copula functions for wind speed correlation modeling. However, with the expansion of wind farm scale, the increase in the number of wind turbines, and the increase in the number of time segments considered, this type of method faces a serious "curse of dimensionality" problem, significantly increasing the modeling difficulty. In recent years, although deep learning methods have made some progress in spatiotemporal feature extraction, they are mostly limited to deterministic predictions and cannot reflect the random fluctuation range of wind speed under the same conditions, failing to provide sufficient random samples of uncertain wind speeds for reliability assessment. At the same time, some emerging generative prediction models often only start from time series evolution or statistical correlation of multi-source variables, without fully considering the profound impact of the internal physical topology of the wind farm on the spatial correlation characteristics of multiple wind turbines, resulting in insufficient spatiotemporal uncertainty representation of wind speeds of multiple wind turbines.

[0004] On the other hand, in the post-fault system state analysis phase, most existing studies only start from the relationship between component fault states and system structure, using analytical models such as multi-state Markov processes, full probability enumeration, or fault trees to analyze the impact of equipment outages on the available power output of wind farms. These methods are mostly limited to the static superposition of component fault states, failing to fully reflect the actual power transfer limitations and dynamic self-healing potential of the system after a local fault occurs. This insufficient consideration of the actual physical power supply recovery capability after a system fault makes it difficult for existing methods to accurately characterize the actual power output loss of the system under disturbed conditions, which often leads to biases in the overall operational reliability risk assessment results. Summary of the Invention

[0005] To address the shortcomings of the existing technologies, the technical problem to be solved by this invention is: how to provide a method for assessing the operational reliability of offshore wind farms that integrates the high-dimensional spatiotemporal characteristics of multiple wind turbines with network reconstruction, so as to solve the technical problem that existing assessment methods are unable to fully characterize the random fluctuation range and physical topological correlation of high-dimensional wind speeds, and fail to effectively quantify the real dynamic self-healing potential after local faults, resulting in inaccurate overall operational risk assessment.

[0006] To solve the above-mentioned technical problems, the present invention adopts the following technical solution: A method for assessing the operational reliability of offshore wind farms that integrates the high-dimensional spatiotemporal characteristics of multiple wind turbines with network reconfiguration includes the following steps: The historical wind speed sequence of the offshore wind farm is obtained for multiple consecutive moments before the evaluation time, and the multidimensional spatiotemporal features of multiple wind turbines in the wind farm are extracted. The multidimensional spatiotemporal features are then processed into unified conditions and gated topology embedding features. The unified conditions and the gated topology embedding features are input into a pre-constructed two-stage residual conditional diffusion model. The trend reasoning module in the two-stage residual conditional diffusion model extracts the center estimate of the wind speed of multiple wind turbines at the next moment. The uncertainty residual estimation module learns the conditional distribution of the difference between the actual wind speed and the center estimate through forward noise addition and reverse noise reduction. The mean prediction value is superimposed with the residual sample to generate a multi-wind turbine wind speed prediction sample with random fluctuation characteristics. By combining the actual failure rate and repair rate parameters of wind turbines and collector cables, Monte Carlo sampling is used to generate the failure state of wind power equipment. The failure state is then combined with the wind speed prediction sample of the multi-wind turbine to construct a wind speed-fault joint system state sample. For each of the generated wind speed-fault joint system state samples, the maximum available output of a single wind turbine is calculated. With minimizing the total output reduction of the wind farm as the optimization objective, and under the preset power system operation constraints, the network reconfiguration optimization model considering the switching of backup cables is solved. Based on this, the switching action and network reconfiguration scheduling strategy for backup cables are generated, and the actual active power that can be delivered and the output loss results of the system after network reconfiguration intervention are calculated. The output loss results under all wind speed-fault combined system state samples are statistically analyzed, and the operational reliability index is calculated hourly according to the evaluation time. Based on the output operational reliability index, the actual operational risks of offshore wind farms are warned.

[0007] As an optimization, the steps to process the multidimensional spatiotemporal features into unified conditions and gated topological embedding features are as follows: The historical wind speed sequence is extracted using a sliding time window, and the hourly information and daily sequence information within the year are extracted from the timestamp. A periodic time feature is constructed using sine-cosine coding. The historical wind speed sequence and the periodic time feature are then concatenated to form the unified condition. The topology of the wind farm's power collection system is abstracted into a graph structure. Based on the upstream and downstream connection relationships of the wind turbines in the line and the undirected adjacency relationship, normalized adjacency matrices for upstream association, downstream association and undirected adjacency are constructed respectively. The current wind speed vector is used to weight and aggregate the states of nearby wind turbines to generate spatial correlation features; Extract the branch number of the wind turbine and its hierarchical position in that branch. After normalization, combine it with the wind speed at the current moment, the wind speed at the previous moment, and the difference term to generate static topological features. The spatial correlation features and the static topological features are combined as basic features. A gating mechanism including a topological feature projection network and a gating mapping function is used to adaptively filter and weight the basic features to generate the gated topological embedding features.

[0008] As a further optimization, the steps for generating multi-wind turbine wind speed prediction samples using the two-stage residual conditional diffusion model are as follows: The trend reasoning module performs a linear mapping on the unified conditions and performs a weighted fusion with the gated topology embedding features to obtain a hidden input feature vector. Based on the hidden input feature vector, it outputs the center estimate of the wind speed of multiple wind turbines at the next moment and defines the difference between the actual wind speed and the center estimate as a random residual. During the forward diffusion process, the uncertainty residual estimation module adds Gaussian noise to the random residual step by step to obtain a noisy residual. During the reverse denoising process, the residual denoising branch estimates the noise term and recovers the corresponding residual estimate based on the noisy residual, diffusion step information, uniformity conditions and the center estimate. Multiple residual samples are obtained by repeated backsampling, and the residual samples are superimposed with the central estimate to generate the multi-fan wind speed prediction sample.

[0009] As a further optimization, the steps for generating fault states of wind turbines using Monte Carlo sampling include: A two-state Markov model is used to represent the random outage of wind power equipment. The forced outage rate of wind turbines and collector cables is calculated based on the mean time between failures and mean time to repair of wind power equipment. Based on the forced outage rate, the operating or outage status of each wind turbine node and cable branch at the evaluation time is determined by Monte Carlo random sampling.

[0010] As a further optimization, the preset power system operation constraints consist of the following set of conditions: Branch opening state constraint: Based on the directed branch state variable, constrain the branches connected to the faulty cable and the faulty fan to be disconnected, constrain the original fault-free branches that are still connected to the substation after the fault to remain closed, and constrain the activation state of the candidate backup cable. Topological constraints: After reconstruction using directed tree constraints, the network becomes radial, and potential loops during the reconstruction process are eliminated; Root node constraint: The offshore substation is constrained to act as the root node for system power balance, and its voltage is fixed at the reference voltage. Branch capacity constraint: The active and reactive power transmitted by a branch in operation is limited to not exceeding its apparent power capacity limit. Node and branch voltage constraints: Based on the linearized distflow power flow model, node voltages are constrained within the safe operating range, and the relationship of branch voltage drops is described; Fan output constraint: The output of a faulty fan is constrained to zero, and the actual active output of a normally connected fan does not exceed the maximum available output determined by the wind speed, and the reactive output does not exceed the rated reactive power injection value.

[0011] As a further optimization, the operational reliability indicators include at least: Wind power output loss probability distribution: Calculate the probability that the wind farm output loss falls into multiple preset different loss intervals at the evaluation time; Expected power output loss: The average active power output loss of all wind speed-fault combined system state samples at the evaluation time; Average reconfiguration gain: The difference between the average output loss after network reconfiguration optimization and the average output loss without network reconfiguration at the evaluation time, used to quantify the average effective power recovered by the standby cable switching operation; Average reconstruction recovery rate: The average of the single-sample recovery rates of all wind speed-fault joint system state samples at the evaluation time is used to characterize the average level of the recovery ratio of each state sample. Overall reconfiguration recovery rate: The ratio of the total output recovered by all wind speed-fault joint system state samples through network reconfiguration at the evaluation time to the total output loss without considering network reconfiguration, used to characterize the degree of recovery of the overall output loss of all system state samples.

[0012] In summary, the present invention has the following advantages and beneficial effects: 1. Improved accuracy and realism of probabilistic prediction of high-dimensional spatiotemporal wind speed. This invention constructs a two-stage residual conditional diffusion model, which integrates historical wind speed, periodic time characteristics, and topological correlation characteristics of the power collection system into the conditional input. This model can simultaneously characterize the temporal continuity, spatial correlation, and random fluctuation characteristics of wind speeds from multiple wind turbines. Compared with existing conventional models, the method of this invention has significant advantages in both center prediction accuracy and interval prediction performance under different theoretical coverage rates, providing high-confidence wind speed state samples that better reflect actual operating characteristics for reliability assessment.

[0013] 2. This invention achieves precise quantification of the physical power supply recovery capability after a system failure. In the state analysis phase, it establishes a network reconfiguration optimization model that considers the switching of backup cables, fully taking into account power system operating constraints such as branch disconnection states, radial operating topology, linearized power flow, and node voltages. This model can accurately capture the dynamic power transfer and self-healing potential brought about by network reconfiguration methods under fault scenarios, effectively overcoming the limitations of traditional static evaluation methods.

[0014] 3. A multi-dimensional reliability closed-loop evaluation system was constructed to guide actual dispatching. This invention proposes multi-dimensional customized indicators, including expected output loss (EPL), average reconfiguration gain (ARB), average reconfiguration recovery rate (RRR-A), and overall reconfiguration recovery rate (RRR-T). Through precise evaluation, it can not only adapt to the time-varying characteristics of offshore wind farm operation, but also significantly guide the power grid to reduce the probability of high-loss scenarios, thereby triggering the switching action and dispatching strategy for backup cables, forming a complete safety defense closed loop from theoretical early warning to physical entity control. Attached Figure Description

[0015] Figure 1 This is a flowchart for the reliability assessment of offshore wind farm operations.

[0016] Figure 2 This explains the working principle of the residual diffusion prediction model.

[0017] Figure 3 This is the power characteristic curve of the wind turbine.

[0018] Figure 4 This represents the topology of an offshore wind farm.

[0019] Figure 5 The results show the probability prediction of wind speed at two consecutive time points.

[0020] Figure 6 The absolute error of the wind speed probability prediction mean at two consecutive time points.

[0021] Figure 7 The relative error of the wind speed probability prediction mean at two consecutive time points.

[0022] Figure 8 A comparison of prediction error indices for different methods.

[0023] Figure 9 Let PLDP be the output loss distribution probability.

[0024] Figure 10 The expected output loss (EPL) of wind power at 24 time points.

[0025] Figure 11 The average reconstruction gain (ARB) is calculated over 24 time points.

[0026] Figure 12 The average recovery rate and overall recovery rate are calculated over 24 time points. Detailed Implementation

[0027] The present invention will be further described in detail below with reference to the embodiments.

[0028] The reliability assessment of offshore wind farms does not focus on the reliability of traditional load power supply, but primarily on the system's effective power transmission capacity to the main grid and its fluctuation characteristics. For offshore wind farms under fault disturbances, turbine or collector cable failures may lead to collector line interruptions or partial feeder islanding, resulting in a loss of system-deliverable power. Therefore, system operational reliability depends not only on the degree of power loss caused by equipment failures, but also on the level of power transmission capacity recovery after network reconfiguration measures. To evaluate the power transmission capacity and reconfiguration recovery effect of offshore wind farms under fault conditions, this embodiment constructs a set of operational reliability index systems. The focus is on three aspects: output loss probability distribution, expected loss level, and reconfiguration recovery effect. Specifically, the output loss probability distribution describes the probability characteristics of system output loss falling into different loss ranges under fault disturbances; the expected loss level reflects the average degree of power loss in the system's deliverable capacity; and the reconfiguration recovery effect measures the level of power transmission capacity recovery achieved by network reconfiguration. Considering the changes in wind speed and equipment failure status over time, the operational reliability indices constructed in this embodiment are calculated hourly at the evaluation time to characterize the time-varying pattern of reliability levels during the continuous operation of offshore wind farms. Therefore, this embodiment mainly constructs the following indices: power loss distribution probability (PLDP), expected power loss (EPL), average reconfiguration benefit (ARB), average reconfiguration recovery ratio (RRR-A), and total reconfiguration recovery ratio (RRR-T).

[0029] To facilitate the definition of the constructed indicator system, let the evaluation time be t = t0, t0+1, ..., T. For any evaluation time t, N indicators are generated through state sampling. s A number of wind speed-fault joint system state samples are provided, each containing the wind speed of each wind turbine in the wind farm and the operating status of the turbines and cables. For the s-th (s=1,2,...,N) s By performing state analysis on a sample of the system state, the upper limit of the total available power output of the wind farm at time t can be obtained. and actual output power . and The maximum available output of each wind turbine at time t and state s is respectively determined by the maximum available output of each wind turbine. and actual available output Summing yields the result. The corresponding sample output loss caused by faults, etc., can be expressed as: and The difference. Among them, The calculation can be divided into two cases: considering network reconstruction and not considering it.

[0030] Based on this, the reliability index constructed in this embodiment is defined as follows.

[0031] 1) Wind power output loss distribution probability: Divide the output loss range into K intervals, and denote the k-th loss interval as... Then, at time t, the probability that the output loss falls into the k-th loss interval can be defined as: (1) In the formula: This is a conditional function; it takes the value 1 if the condition within the parentheses is true, and 0 otherwise.

[0032] 2) Expected output loss: At time t, the average output loss of all system state samples can be defined as: (2) 3) Average Reconstruction Gain: At time t, the average recovery power brought about by network reconstruction across all system state samples can be defined as: (3) In the formula: and Let represent the output loss of the s-th system state sample without considering network reconstruction methods and with consideration, respectively.

[0033] 4) Average reconstruction recovery rate and overall reconstruction recovery rate The average reconstruction recovery rate focuses on reflecting the average level of recovery proportions for samples in each state, while the overall reconstruction recovery rate focuses on reflecting the degree of recovery of the overall output loss for all samples in all states. The index can be defined as: (4) (5) The wind farm operation reliability assessment framework proposed in this embodiment is as follows: Figure 1 As shown, it mainly includes three stages: system state generation, system state analysis, and reliability index calculation.

[0034] This framework is based on the probability prediction results of the combined wind speed of multiple wind turbines at the evaluation time, combined with the sampling of fault states of wind turbines and collector cables, to construct a wind speed-fault combined system state. Then, under each combined system state, a post-fault state analysis model considering network reconfiguration is solved to obtain the system output loss and recovery results. Finally, the analysis results of all combined system states at the evaluation time are statistically analyzed to obtain the operational reliability index of the offshore wind farm. The specific steps are as follows: 1) System state generation: After the evaluation time t0 and the maximum number of states are determined, the wind speed state of multiple wind turbines at the evaluation time is generated first based on the historical wind speed information of the L times before t0 using the topology-aware conditional diffusion model. At the same time, based on the failure rate and repair rate of wind turbines and cables, the component two-state model and Monte Carlo sampling are used to generate the component failure state, and the wind speed-fault joint system state is further constructed.

[0035] 2) System state analysis: For the state of each joint system at the evaluation time t=t0, first calculate the available output of the wind turbine based on the wind speed-power relationship, then solve the state analysis model considering network reconfiguration, and perform convergence verification on the state analysis results.

[0036] 3) Calculation of reliability indicators: Based on the analysis results of the status of each joint system, the reliability indicators are calculated according to the definition.

[0037] Construction of Spatiotemporal Correlation Features of Wind Speed ​​for Multiple Wind Turbines: Wind speed sequences of multiple wind turbines within a wind farm exhibit significant temporal and spatial correlations. Temporal correlation manifests as the continuous evolution and periodic variation of wind speed over short timescales; spatial correlation reflects the correlation characteristics formed by the spatial distribution of wind turbines at different locations, the connection relationships of power collection lines, and the hierarchy of branch lines. Therefore, this embodiment constructs the wind speed prediction input for multiple wind turbines from two aspects: temporal correlation features and spatial structural correlation features.

[0038] 1) Construction of time-dependent features: Suppose there are N wind turbines in the wind farm, and the wind speed vector of the multiple turbines at time t is denoted as: (6) In the formula: v i,t denoted as the actual wind speed value of the i-th wind turbine at time t; N represents the total number of wind turbines in the wind farm.

[0039] Considering that wind speed typically exhibits strong continuity over very short timescales, this embodiment employs a sliding time window of length L to construct historical wind speed input samples, namely: (7) In the formula: X t This represents the historical wind speed window up to the predicted time t+1, where L represents the length of the time window.

[0040] The input X is constructed using historical wind speed information from time t and L times prior to it. t Then, the model's prediction objective can be represented as the multi-turbine wind speed vector at the next time step. .

[0041] To further enhance the model's ability to characterize short-term trends, difference features between adjacent time points are constructed. Meanwhile, wind speed changes typically exhibit certain periodic characteristics. To characterize the recurring patterns of wind speed changes on intraday and intrayear scales, this embodiment extracts hourly information and intrayear daily sequence information from the original timestamps, and constructs periodic time features using sine-cosine coding: (8) In the formula: τ t h is the periodic time eigenvector; t r represents the number of hours in a day corresponding to time t; t This indicates the day of the year corresponding to time t.

[0042] 2) Spatial Relevance Feature Construction: Considering the characteristics of offshore wind farms where wind turbines are arranged in series along feeders and different feeders form branch structures, the wind farm power collection system is abstracted as a graph structure. Based on the upstream and downstream connection relationships of the wind turbines in the lines, three types of adjacency matrices A are constructed. (1) A (2) A (3) , representing upstream association, downstream association, and undirected adjacency relationships, respectively. To avoid the impact of differences in the number of adjacencies between different nodes on feature aggregation, this embodiment uses row normalization to process the adjacency matrix, that is, weighted aggregation of adjacent wind turbine information based on the number of adjacent nodes of each node: (9) (10) In the formula: D represents the normalized adjacency matrix; k Adjacency matrix The corresponding row-degree matrix has its main diagonal elements representing the number of adjacent nodes for each node under this connection relationship, and all other elements are 0; when k=3, D3 is the matrix after adding the self-loop. The corresponding row degree matrix; I is the identity matrix. For boundary wind turbines that do not have corresponding adjacent nodes, their normalized denominator is set to 1, and the corresponding aggregation result is a zero vector.

[0043] After obtaining the normalized adjacency matrix, the current wind speed vector can be used to aggregate the states of neighboring wind turbines to obtain different types of spatial correlation features: (11) In the formula: This represents the spatial features formed by aggregating information from neighboring wind turbines under the connection relationships described by the k-th type of adjacency matrix.

[0044] In addition to the adjacency relationship, a static position code for the wind turbine in the collector line is further introduced. Let the branch number where the i-th wind turbine is located be l. i The hierarchical position in this branch is d. iAfter normalization, the static topological location features are obtained. and By arranging the normalized branch numbers and hierarchical positions of all wind turbines according to their turbine numbers, the overall field position vector can be obtained. and By combining the current wind speed, the previous wind speed, and the difference term, we can further obtain the fundamental quantities used for topological coding, namely: (12) Since different types of topological features contribute differently to the spatial correlation of wind speed at different operating times, this paper employs a gating mechanism to adaptively filter and weight the basic features. The gated topological embedding can be represented as: (13) In the formula: This is a topological feature projection network used to map basic topological features to a hidden feature space; This is a gated mapping function used to generate gated weights with the same dimension as the projected features; The Sigmoid activation function constrains the gate weights to between 0 and 1. Represents element-wise product; Representation layer normalization operation; g t Let be the gated topological embedding feature at time t.

[0045] Based on this, historical wind speed windows and periodic time characteristics are used to form a unified input condition for multi-turbine wind speed prediction at time t. t : (14) A diffusion probability prediction model based on trend inference and uncertainty residual estimation is constructed: Offshore wind farm wind speeds are influenced by factors such as the marine environment, spatial location, and collector topology, exhibiting significant uncertainty and random fluctuation characteristics. Under the same historical conditions, the wind speed of multiple turbines at the next moment is not uniquely determined but presents multiple possible values, thus posing a challenge to the joint wind speed probability prediction of multiple turbines. To address this problem, this embodiment proposes a two-stage residual diffusion probability prediction model. This model uses conditional input c... tBased on this, the central variation law of the wind speed of multiple wind turbines at the next moment is first extracted through the trend inference module; then, the deviation between the actual wind speed and the estimated central value is used as the random residual, and its conditional distribution is learned using the uncertainty residual estimation module. Since the random residual can reflect the wind speed fluctuation component outside the mean trend, its distribution usually has nonlinear and uncertain characteristics. Therefore, this embodiment introduces a diffusion probability model for generative modeling. The diffusion probability model is a type of generative model based on variational inference, which can essentially be represented as a parameterized Markov chain, capable of learning and sampling the distribution of complex random variables through stepwise denoising. The working principle of the residual diffusion prediction model proposed in this embodiment is as follows: Figure 2 As shown.

[0046] Figure 2 In this system, the trend inference module extracts the central trend of wind speed variation for multiple wind turbines at the next moment from the input conditions, representing the deterministic evolutionary component of wind speed; the uncertainty residual estimation module learns the residual distribution between the actual wind speed and the central estimate, characterizing random fluctuations beyond the mean trend. The two modules are then combined to generate a multi-wind turbine wind speed prediction sample with uncertainty.

[0047] 1) Trend Inference Module: The trend inference module is used to characterize the predictable trend of the wind speed of multiple wind turbines in the next moment. Let the normalized true wind speed of multiple wind turbines in the next moment be y. t+1 Conditional input c t The topological embedding feature is g t First, the conditional input is linearly mapped and then weighted and fused with the topological embedding features to obtain the implicit input feature vector of the trend inference module: (15) Based on this, the central variation law of the wind speed of multiple wind turbines at the next moment can be extracted using the trend prediction function, which can be expressed as: (16) In the formula: and These are the weight matrix and bias vector of the conditional input linear mapping, respectively; R t H represents the implicit features resulting from the fusion of historical time-series information, periodic time information, and topological association information; H is a trainable topological feature fusion coefficient used to adjust the contribution of topological embedding features in the trend inference process. Indicates the branch of the mean forecast; For network parameters; This is the estimated center value of the wind speed of the multiple wind turbines at the next moment.

[0048] After obtaining the center estimate, the difference between the actual wind speed and the center estimate is defined as the random residual, as follows: (17) 2) Uncertainty Residual Estimation Module: The uncertainty residual estimation module uses the random residuals obtained from equation (17) as the modeling object to characterize the uncertain fluctuations in wind speed under given conditions. During the forward diffusion process, the residual e... t+1 According to the diffusion step By gradually adding Gaussian noise, we obtain the noisy residual: (18) In the formula: For the first Step-diffusion cumulative coefficient; represents Gaussian noise. I represents the identity matrix with the same dimension as the input sample. This indicates that the noise follows a multivariate Gaussian distribution with zero mean and unit covariance matrix.

[0049] In the reverse denoising process, the residual denoising branch estimates the noise term based on the noisy residual, diffusion step information, conditional input, and mean branch output. ,Right now: (19) In the formula: Represents the residual denoising branch; Its network parameters.

[0050] Furthermore, the corresponding residual estimates can be recovered based on the noise estimation results: (20) In the formula: This represents the residual estimate recovered by the inverse denoising process.

[0051] During the prediction phase, the nth residual sample can be obtained by repeated backsampling. Finally, the residual samples are superimposed with the mean branch output to obtain the multi-fan wind speed sample for the next time step corresponding to the nth sampling: (twenty one) In the formula: N represents the number of repeated samplings.

[0052] Wind farm equipment reliability and available output model: In reliability analysis, a two-state Markov model is often used to represent random equipment outages. Based on this, the forced outage rate K of the equipment can be expressed by the following formula: (twenty two) In the formula: λ and μ represent the equipment failure rate and repair rate, respectively; t MTTR and t MTTF These represent the mean time to repair and the mean time between failures (MTBF), respectively. Where λ = 8760 / tMTTF μ = 8760 / t MTTR .

[0053] On the other hand, the output of wind turbines is closely related to wind speed and their own power characteristics. Their output characteristics are as follows: Figure 3 As shown.

[0054] The conversion relationship between wind power output and wind speed is as follows: (twenty three) In the formula: v represents the maximum available active power output of wind turbine i in the s-th system state sample at time t; i,s (t) represents the wind speed of fan i in the s-th system state sample at time t; v ci v co v r These represent the cut-in wind speed, cut-out wind speed, and rated wind speed, respectively; P r The rated power of the fan is represented by A, B, and C, which are the fitting coefficients of the quadratic polynomial.

[0055] To accurately quantify the loss of deliverable power in offshore wind farms under fault conditions and reflect the impact of system recovery measures on power loss, this embodiment establishes a minimum turbine-cutting optimization model considering the switching of backup cables for each wind speed-fault combined scenario. This model adds a backup cable at the end of the original collector line and improves the system's reliability under fault scenarios by optimizing the decision branch disconnection state and the actual turbine output power.

[0056] For ease of subsequent description, we first define the concept of a branch set. Let the root node represent the offshore substation, and the remaining nodes correspond to the wind turbine nodes. For the s-th system state at time t, the branches in the original collector lines and candidate backup cables that can participate in the state analysis are collectively denoted as... Based on this, and according to the equipment fault status and the network connectivity after the fault, The branch roads are further divided into three categories: This indicates the set of branches that cannot participate in operation under the current condition, including faulty cables and branches directly connected to faulty fans; This refers to the original collection of power lines that remain connected to the offshore substation after a fault and have not experienced a fault; these branches maintain their original closed state. This represents the set of candidate branches that can be used to restore power after a fault, including backup cables and the original lines that are still intact within the island. Based on this, the model is described in detail below: 1) Optimization Objective: For the s-th system state at time t, the network reconfiguration optimization objective is to minimize the total power output reduction of the wind turbines, i.e.: (twenty four) In the formula: Let be the actual active power delivered by wind turbine i after reconfiguration and optimization at time t and state s. A collection of wind turbines.

[0057] 2) Constraints: This embodiment uses a directed tree to describe the reconstructed radial network, and the branch opening and closing directions are characterized by directed branch state variables. The relevant constraints are as follows: (1) Branch Opening State Constraints: Equations (25)-(27) are used to determine the opening state of each branch after a fault. Equation (25) indicates that branches connected to the faulty cable and the faulty fan must be disconnected; Equation (26) indicates that the fault-free original branches still connected to the substation after the fault remain closed; Equation (27) indicates that the backup cable and other candidate recovery branches can be activated by the optimization model. The mathematical expression is: (25) (26) (27) In the formula: a ij,s (t) is a binary variable used to represent the on / off state of branch (i,j) in the s-th system state at time t. When a ij,s When (t)=1, it indicates that branch ij is closed and connected to the radial network with i as upstream and j as downstream; when a ij,s When (t)=0, it indicates that branch ij is disconnected. Since the same branch can only connect to the network in one direction, therefore a ij,s (t) and a ji,s (t) cannot be 1 at the same time.

[0058] (2) Topological constraints: Equations (28)-(32) are used to ensure that the reconstructed network satisfies the radial connection relationship. Among them, equations (28)-(29) indicate that if a branch ij is put into operation, its two end nodes must be connected to the root node. Equation (30) is used to determine the connectivity status of wind turbine nodes: when it is connected to the network connected to the root node through an upstream branch, the node is connected; if there is no upstream branch, the node is not connected; Equation (31) indicates that after a system fault event occurs, the fault-free wind turbine nodes that are still connected to the substation should maintain the connectivity status; Equation (32) is used to eliminate potential loops in the reconstructing process. The mathematical expression is: (28) (29) (30) (31) (32) In the formula: r i,s (t) is a binary variable representing whether node i is connected to the root node; N represents the set of normal wind turbines that remain connected to the root node at time t and state s; loop l represents the number of nodes corresponding to a potential loop. loop This represents the set of branches in a potential cycle. For a cycle, the number of nodes must equal the number of branches.

[0059] (3) Root Node Constraints: Equations (33)-(35) are used to ensure that the offshore substation always acts as the root node during the reconfiguration process. Equation (33) indicates that the offshore substation, as the root node, remains connected throughout the reconfiguration process; Equation (34) indicates that the root node is not allowed to have any upstream branches; Equation (35) indicates that the root node acts as the balancing node of the offshore wind farm system, and its voltage is fixed as the reference voltage, used as the benchmark for calculating the voltage of the entire network. The mathematical expression is: (33) (34) (35) In the formula: Node 1 represents the number of the offshore substation; U 1,s (t) represents the square of the root node voltage under state s at time t; This is the squared value of the reference voltage.

[0060] (4) Branch capacity constraints: Equations (36)-(37) are used to describe the branch transmission power that should meet its capacity limit: (36) (37) In the formula: P ij,s (t) and Q ij,s (t) represent the active and reactive power on the directed branch ij, respectively; Let a be the upper limit of the apparent power capacity of branch ij. ij,s When (t)=0, the branch is not in operation, and its power is limited to 0; when a ij,s When (t)=1, the branch is put into operation, and the branch power must meet the upper limit constraint of capacity.

[0061] (5) Node voltage and branch voltage constraints: Equation (38) is used to limit the operating range of voltage at each node of the offshore wind farm system; Equation (39) is the linearized distflow power flow model, used to describe the voltage drop relationship of branch ij. When the branch is disconnected, this equation is relaxed by the big M-term method. The mathematical expression is: (38) (39) In the formula: U i,s (t) represents the squared voltage value at node i; U max and U min These are the upper and lower limits of the square of the node voltage, respectively; R ij and X ij Let be the resistance and reactance of branch ij, respectively; M is a sufficiently large constant.

[0062] (6) Fan output constraints: Equation (40) indicates that a faulty fan cannot participate in power output; Equation (41) indicates that the actual output of a normal fan cannot exceed the maximum available output determined by the wind speed, and can only be output when the node is connected to the root node; Equation (42) indicates that reactive power can only be output when the fan is fault-free and connected to the root node. The mathematical expression is: (40) (41) (42) In the formula: Let be the set of faulty fans in state s at time t; Let Q be the set of fault-free wind turbines at time t and state s. i,s (t) represents the actual reactive power output of node i; This represents the reactive power injection value of the wind turbine corresponding to node i under normal operating conditions.

[0063] (7) Node power balance constraint: Equations (43)-(44) are used to describe the node power balance relationship of the wind farm system, that is, the power flowing into each node is equal to the power flowing out of the same node. The mathematical expression is: (43) (44) To verify the effectiveness of the proposed operational reliability assessment method, this embodiment uses an offshore wind farm with a total installed capacity of 400MW, located 100km offshore, as an example for case analysis. The wind farm has a total of 80 wind turbines, each with a capacity of 5MW, with 8 turbines forming a feeder unit. The system adopts a radial wiring configuration, and its topology is as follows: Figure 4As shown in the figure, the dashed line indicates the location of the backup cable. The spacing between wind turbines in each string is 1 km, and the spacing between strings is 800 m. Reliability parameters of key equipment in the wind farm are shown in Table 1. Wind turbine power characteristic parameters are shown in Table 2. The cable capacity is 1.2 times the sum of the rated capacities of the wind turbines in a single feeder line.

[0064] Table 1 Reliability parameters of key equipment The unavailability rate of the fan and collector cable can be obtained from the parameters shown in Table 1 according to formula (22).

[0065] Table 2 Wind Turbine Power Characteristic Parameters Based on the parameters shown in Table 2, the output of each fan under different system states s can be calculated using formula (23).

[0066] This embodiment will be analyzed from the following three aspects.

[0067] 1) Analyze the multi-wind turbine joint wind speed probability prediction results of the proposed topological sensing conditional diffusion model to verify the model's ability to characterize wind speed change trends and uncertainty fluctuations. 2) By comparing with traditional methods, the advantages of the proposed method in terms of prediction accuracy and interval prediction performance are verified; 3) Generate wind speed-fault joint system state samples based on wind speed probability prediction results to verify the effect of network reconstruction on improving the reliability index of offshore wind farm operation.

[0068] Validation of the effectiveness of multi-wind turbine joint wind speed probability prediction results: In the example, the historical window length is L=3, corresponding to the wind speed sequence of 80 wind turbines in the most recent 3 moments; at the same time, sine and cosine encoding of intraday hours and intra-year daily sequence are used to construct 4-dimensional periodic time features, so the conditional input dimension of a single sample is 80. 3+4=244; In the model, the hidden layer dimension of the trend inference module is set to 384; the hidden layer dimension of the uncertainty residual estimation module is 256, the diffusion step is set to 200, and the number of training rounds is set to 200. The batch sample size for both modules during training is set to 256, and the learning rate is 1×10⁻⁶ for both. -4 .

[0069] Based on the above settings, complete the wind speed probability prediction for 24 consecutive time periods, and display the prediction results for two consecutive time periods, such as... Figure 5As shown in the figure, the predicted wind speed of each wind turbine over two consecutive time periods maintains a consistent overall trend with the actual wind speed, and the peak and valley positions match well. This indicates that the proposed model can effectively characterize the spatial distribution characteristics of the combined wind speed of a multi-wind turbine system. To further quantify the prediction bias, Figure 6 and Figure 7 The absolute and relative errors of the wind speed probability prediction mean for each wind turbine at two consecutive time points are given. Figure 6 It can be seen that the absolute error of most wind turbine nodes remains below 0.4 m / s, with only a few nodes showing prediction deviations exceeding 0.6 m / s; further... Figure 7 It can be seen that the relative error of most wind turbines remains within 3%, and the overall relative error level is low. Although Figure 6 and Figure 7 This indicates that there are still some individual wind turbine nodes with relatively high prediction mean errors, but combined with... Figure 5 As can be seen from the predicted fluctuation range, most of the actual wind speeds still fall within the predicted fluctuation range. This indicates that although the proposed method still has room for improvement in the central prediction accuracy of local nodes, it can generally reflect the wind speed change trend well and effectively characterize the random fluctuation range and uncertainty of wind speeds across multiple wind turbines.

[0070] Performance advantage analysis of the proposed probabilistic prediction method: To verify the performance advantage of the proposed method in multi-wind turbine wind speed probabilistic prediction, this embodiment compares the calculation results of the following methods: M1: The improved diffusion model proposed in this embodiment; M2: Basic conditional diffusion model, which uses the same historical wind speed window and periodic time features as the model in this embodiment as the conditional input, directly performs diffusion modeling on the 80-dimensional multi-wind turbine wind speed vector at the next moment, but does not introduce topology-aware coding, trend reasoning module and uncertainty residual estimation module. M3: Copula-Markov model, as a contrast to existing typical traditional probabilistic modeling methods; M4: CNN-LSTM-MC model, used as a comparison with existing typical deep learning spatiotemporal prediction methods.

[0071] When evaluating the deviation between the predicted mean and the actual wind speed from different methods (i.e., the accuracy of center prediction), this embodiment uses the Mean Absolute Error (MAE) and Root Mean Square Error (RMSE) as evaluation metrics, and their calculation formulas are as follows: (45) (46) In the formula: N is the number of fans; and Let represent the actual wind speed and the predicted mean wind speed of the i-th wind turbine at the t-th prediction time, respectively.

[0072] In evaluating the interval prediction performance of different probability prediction methods, this embodiment sets multiple theoretical coverage levels and obtains the wind speed interval prediction results of each method under the corresponding theoretical coverage requirements. Theoretical coverage represents the proportion of the predicted interval expected to cover the actual wind speed. The higher the theoretical coverage to be achieved, the wider the required prediction interval is usually. However, selecting an excessively wide interval will reduce the accuracy of the prediction results in characterizing the wind speed fluctuation range. Therefore, this embodiment introduces an Interval Score (IS) to comprehensively evaluate the performance of the obtained prediction interval from two aspects: "prediction interval width" and "probability of the actual value falling into the interval". Under the same theoretical coverage, the smaller the interval score, the smaller the interval width of the prediction interval while covering the actual value, the more concentrated the prediction range, and the better the interval prediction performance.

[0073] The comparison results of MAE and RMSE obtained by each method are as follows: Figure 8 As shown in Table 3, the interval scores of each method under different theoretical coverage are compared.

[0074] Depend on Figure 8 As can be seen, the proposed model M1 in this embodiment achieves the minimum values ​​for both MAE and RMSE, at 0.312 m / s and 0.487 m / s respectively, which are reduced by approximately 82.0% and 77.3% compared to the ordinary conditional diffusion model M2. This indicates that the model's prediction accuracy is significantly improved after introducing topology-aware encoding, trend inference, and uncertainty residual estimation modules. Furthermore, compared to the Copula-Markov model M3 and the CNN-LSTM-MC model M4, the proposed model reduces MAE and RMSE by 83.1% and 80.9%, and 78.7% and 75.8% respectively. This demonstrates that compared to traditional typical probabilistic modeling methods and deep learning methods, the diffusion model can more effectively extract the spatiotemporal correlation features of wind speeds from multiple wind turbines, thus making the probabilistic prediction mean closer to the actual wind speed.

[0075] Table 3. Comparison of interval scores for different methods under different theoretical coverage rates As shown in Table 3, when the theoretical coverage requirement varies from 10% to 90%, the interval scores of the proposed model M1 in this embodiment range from 0.7258 to 4.5839, which are significantly lower than the interval scores of M2 to M4. For example, when the theoretical coverage is 90%, the interval scores of the proposed model M1 are reduced by approximately 48.1%, 55.0%, and 72.4% compared to M2 to M4, respectively. This indicates that the proposed model can maintain a significant advantage in interval scores even with a high theoretical coverage, and can avoid overly wide prediction intervals while covering the actual wind speed as much as possible, making the prediction results more concentrated and thus more reasonably representing the random fluctuation range of wind speed.

[0076] Analysis of Reliability Assessment Results for Offshore Wind Farms Considering Network Reconfiguration: This embodiment conducts a 24-time-period reliability assessment of the offshore wind farm system, using the Monte Carlo method for statistical analysis of indicators. Each time-period considers 10,000 wind speed-fault joint system state samples. Based on this, the reliability assessment results are compared and analyzed with and without network reconfiguration as a recovery method during system state analysis. The assessment indicators used include output loss distribution, expected output loss (EPL), average reconfiguration gain (ARB), average reconfiguration recovery rate (RRR-A), and overall reconfiguration recovery rate (RRR-T). The output loss distribution probability curve is shown below. Figure 9 As shown, the variation curves of EPL, ARB, RRR-A, and RRR-T at 24 consecutive time points are respectively as follows: Figures 10-12 As shown.

[0077] Depend on Figure 9 It can be seen that, without considering reconfiguration, the system output loss is mainly concentrated in the range of 40-80 MW, with a corresponding probability of 43.89%. Meanwhile, high-loss scenarios with output losses exceeding 100 MW still account for 12.29%, indicating that the system faces a certain risk of high output loss after a fault. In contrast, if network reconfiguration is considered as a recovery method, the main distribution range of system output loss can be reduced to 20-50 MW, with a corresponding probability of 43.21%; the probability of high-loss scenarios with output losses exceeding 100 MW decreases from 12.29% to 2.24%. These results demonstrate that network reconfiguration can significantly reduce output loss under fault scenarios and effectively reduce the probability of high-loss scenarios.

[0078] Depend on Figure 10 It can be seen that, at all evaluation points, the EPL (Energy Performance Loss) when network reconfiguration is considered is lower than the corresponding result when reconfiguration is not considered. Specifically, when reconfiguration is not considered, the EPL ranges from 42 to 72 MW; after considering reconfiguration, its range drops to 17 to 52 MW. These results indicate that network reconfiguration can effectively reduce the expected power output loss of the system at all evaluation points, demonstrating its effectiveness as a post-fault recovery method.

[0079] Depend on Figure 11 It can be seen that the average reconstruction gain (ARB) at each time point is greater than 0, and its value remains within the range of 20-29 MW, indicating that network reconstruction can effectively reduce the output loss caused by system faults and generate a stable positive recovery gain throughout the entire evaluation period. Based on this, Figure 12 The changes in the average reconfiguration recovery rate (RRR-A) and the overall reconfiguration recovery rate (RRR-T) at 24 time points are presented. As shown in the figure, RRR-A generally remained between 30% and 61%, exceeding 40% at 18 evaluation time points; RRR-T generally remained between 28% and 61%, exceeding 40% at 17 evaluation time points. These results indicate that network reconfiguration has a significant fault recovery capability and can effectively improve the transmittable power level after a fault.

[0080] In summary, the evaluation results of the multi-dimensional indicators consistently show that network reconfiguration can significantly enhance the fault recovery capability of offshore wind farms, reduce power loss, and improve the power transmission level after a fault, making it an effective means to improve the reliability of system operation.

[0081] In summary, addressing the issues of insufficient wind speed uncertainty representation and inadequate post-fault recovery capability in the reliability assessment of offshore wind farm operations, this embodiment proposes an operational reliability assessment method that integrates the high-dimensional spatiotemporal characteristics of multiple wind turbines with network reconfiguration. This method generates system state samples that reflect the spatiotemporal correlation and random fluctuation characteristics of wind speed across multiple turbines through probabilistic prediction, and combines this with network reconfiguration analysis to assess power output recovery capability after a fault. The main conclusions are as follows: 1) A conditional diffusion probability prediction model integrating topology-aware coding, trend inference, and uncertainty residual estimation was constructed. This model incorporates historical wind speed, periodic time characteristics, and topological correlation characteristics of the power collection system into the conditional inputs, simultaneously characterizing the temporal continuity, spatial correlation, and random fluctuation characteristics of wind speeds from multiple wind turbines. This provides wind speed state samples that better reflect actual operating characteristics for reliability assessment. Numerical examples demonstrate that the proposed model can accurately track the overall trend of wind speed changes from multiple wind turbines and effectively characterize its uncertainty. Compared with several existing methods, the proposed model has advantages in both center prediction accuracy and interval prediction performance, validating its superiority in high-dimensional wind speed probability prediction for multiple wind turbines.

[0082] 2) Evaluation results under continuous 24-hour operation show that, considering network reconfiguration, the expected output loss at each time point is reduced, while the average reconfiguration gain and reconfiguration recovery rate remain generally stable. This indicates that the proposed evaluation method can adapt to the time-varying characteristics of offshore wind farm operation, and network reconfiguration can effectively restore the output power after a fault, thereby improving the operational reliability of offshore wind farms.

[0083] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for assessing the operational reliability of offshore wind farms that integrates the high-dimensional spatiotemporal characteristics of multiple wind turbines with network reconstruction, characterized in that, Includes the following steps: The historical wind speed sequence of the offshore wind farm is obtained for multiple consecutive moments before the evaluation time, and the multidimensional spatiotemporal features of multiple wind turbines in the wind farm are extracted. The multidimensional spatiotemporal features are then processed into unified conditions and gated topology embedding features. The unified conditions and the gated topology embedding features are input into a pre-constructed two-stage residual conditional diffusion model. The trend reasoning module in the two-stage residual conditional diffusion model extracts the center estimate of the wind speed of multiple wind turbines at the next moment. The uncertainty residual estimation module learns the conditional distribution of the difference between the actual wind speed and the center estimate through forward noise addition and reverse noise reduction. The mean prediction value is superimposed with the residual sample to generate a multi-wind turbine wind speed prediction sample with random fluctuation characteristics. By combining the actual failure rate and repair rate parameters of wind turbines and collector cables, Monte Carlo sampling is used to generate the failure state of wind power equipment. The failure state is then combined with the wind speed prediction sample of the multi-wind turbine to construct a wind speed-fault joint system state sample. For each of the generated wind speed-fault joint system state samples, the maximum available output of a single wind turbine is calculated. With minimizing the total output reduction of the wind farm as the optimization objective, and under the preset power system operation constraints, the network reconfiguration optimization model considering the switching of backup cables is solved. The switching actions and network reconfiguration scheduling strategies for backup cables are generated, and the actual active power that can be delivered and the output loss results of the system after network reconfiguration intervention are calculated. The output loss results under all wind speed-fault combined system state samples are statistically analyzed, and the operational reliability index is calculated hourly according to the evaluation time. Based on the output operational reliability index, the actual operational risks of offshore wind farms are warned.

2. The offshore wind farm operation reliability assessment method based on the integration of high-dimensional spatiotemporal characteristics of multiple wind turbines and network reconstruction as described in claim 1, characterized in that, The steps for processing the multidimensional spatiotemporal features into unified conditions and gated topological embedding features are as follows: The historical wind speed sequence is extracted using a sliding time window, and hourly information and annual day sequence information are extracted from the timestamps. Sine-cosine coding is then used to construct periodic time features. The historical wind speed sequence is concatenated with the periodic time feature to form the unified condition; The topology of the wind farm's power collection system is abstracted into a graph structure. Based on the upstream and downstream connection relationships of the wind turbines in the line and the undirected adjacency relationship, normalized adjacency matrices for upstream association, downstream association and undirected adjacency are constructed respectively. The current wind speed vector is used to weight and aggregate the states of nearby wind turbines to generate spatial correlation features; Extract the branch number of the wind turbine and its hierarchical position in that branch. After normalization, combine it with the wind speed at the current moment, the wind speed at the previous moment, and the difference term to generate static topological features. The spatial correlation features and the static topological features are combined as basic features. A gating mechanism including a topological feature projection network and a gating mapping function is used to adaptively filter and weight the basic features to generate the gated topological embedding features.

3. The offshore wind farm operation reliability assessment method based on the integration of high-dimensional spatiotemporal characteristics of multiple wind turbines and network reconstruction as described in claim 1, characterized in that, The steps for generating multi-fan wind speed prediction samples using the two-stage residual conditional diffusion model are as follows: The trend reasoning module performs a linear mapping on the unified conditions and performs a weighted fusion with the gated topology embedding features to obtain a hidden input feature vector. Based on the hidden input feature vector, it outputs the center estimate of the wind speed of multiple wind turbines at the next moment and defines the difference between the actual wind speed and the center estimate as a random residual. During the forward diffusion process, the uncertainty residual estimation module adds Gaussian noise to the random residual step by step to obtain a noisy residual. During the reverse denoising process, the residual denoising branch estimates the noise term and recovers the corresponding residual estimate based on the noisy residual, diffusion step information, uniformity conditions and the center estimate. Multiple residual samples are obtained by repeated backsampling, and the residual samples are superimposed with the central estimate to generate the multi-fan wind speed prediction sample.

4. The offshore wind farm operation reliability assessment method based on the integration of high-dimensional spatiotemporal characteristics of multiple wind turbines and network reconstruction as described in claim 1, characterized in that, The steps for generating fault states of wind power equipment using Monte Carlo sampling include: A two-state Markov model is used to represent the random outage of wind power equipment. The forced outage rate of wind turbines and collector cables is calculated based on the mean time between failures and mean time to repair of wind power equipment. Based on the forced outage rate, the operating or outage status of each wind turbine node and cable branch at the evaluation time is determined by Monte Carlo random sampling.

5. The offshore wind farm operation reliability assessment method based on the integration of high-dimensional spatiotemporal characteristics of multiple wind turbines and network reconstruction as described in claim 1, characterized in that, The preset power system operation constraints consist of the following set of conditions: Branch opening state constraint: Based on the directed branch state variable, constrain the branches connected to the faulty cable and the faulty fan to be disconnected, constrain the original fault-free branches that are still connected to the substation after the fault to remain closed, and constrain the activation state of the candidate backup cable. Topological constraints: After reconstruction using directed tree constraints, the network becomes radial, and potential loops during the reconstruction process are eliminated; Root node constraint: The offshore substation is constrained to act as the root node for system power balance, and its voltage is fixed at the reference voltage. Branch capacity constraint: The active and reactive power transmitted by a branch in operation is limited to not exceeding its apparent power capacity limit. Node and branch voltage constraints: Based on the linearized distflow power flow model, node voltages are constrained within the safe operating range, and the relationship of branch voltage drops is described; Fan output constraint: The output of a faulty fan is constrained to zero, and the actual active output of a normally connected fan does not exceed the maximum available output determined by the wind speed, and the reactive output does not exceed the rated reactive power injection value.

6. The offshore wind farm operation reliability assessment method based on the integration of high-dimensional spatiotemporal characteristics of multiple wind turbines and network reconstruction as described in claim 1, characterized in that, The operational reliability indicators include at least: Wind power output loss probability distribution: Calculate the probability that the wind farm output loss falls into multiple preset different loss intervals at the evaluation time; Expected power output loss: The average active power output loss of all wind speed-fault combined system state samples at the evaluation time; Average reconfiguration gain: The difference between the average output loss after network reconfiguration optimization and the average output loss without network reconfiguration at the evaluation time, used to quantify the average effective power recovered by the standby cable switching operation; Average reconstruction recovery rate: The average of the single-sample recovery rates of all wind speed-fault joint system state samples at the evaluation time is used to characterize the average level of the recovery ratio of each state sample. Overall reconfiguration recovery rate: The ratio of the total output recovered by all wind speed-fault joint system state samples through network reconfiguration at the evaluation time to the total output loss without considering network reconfiguration, used to characterize the degree of recovery of the overall output loss of all system state samples.

Images