Conditional random field model-based deep and far sea wind power converter station electrical equipment reliability analysis method

Abstract

The invention relates to a deep sea wind power converter station electrical equipment reliability analysis method based on a conditional random field model. The method comprises the following steps: data collection: obtaining historical operation data and environmental factor data of deep sea wind power converter station electrical equipment; model construction: considering the mutual dependency relationship between the equipment, constructing a conditional random field model, taking each operation parameter of the electrical equipment as a random variable, and taking an environmental factor as a conditional variable; model training: training a conditional random field model by using historical data, and optimizing model parameters; and reliability analysis: based on the trained model, predicting and analyzing the reliability of the electrical equipment. Compared with the prior art, the method fully considers the complexity and uncertainty of the deep and far sea environment, can more accurately evaluate and predict the reliability of the electrical equipment, and provides powerful support for the safe operation of the deep and far sea wind power converter station.

Description

Technical Field

[0001] The present invention relates to the technical field of deep - sea and far - sea wind power, and in particular to a reliability analysis method for electrical equipment of deep - sea and far - sea wind power converter stations based on a conditional random field model. Background Art

[0002] With the continuous growth of the demand for renewable energy, the development of deep - sea and far - sea wind power has become an important part of the global energy strategy. Deep - sea and far - sea wind power has advantages such as rich wind energy resources and no occupation of land space, but at the same time, it also faces severe technical challenges. Among them, as a key node connecting offshore wind farms and onshore power grids, the reliability of deep - sea and far - sea wind power converter stations directly affects the safety and efficiency of the entire wind power system.

[0003] Currently, the reliability analysis of electrical equipment in deep - sea and far - sea wind power converter stations mainly uses traditional statistical methods and probability models. These methods include Weibull distribution analysis, Markov chain models, etc. However, these methods have limitations in dealing with the complexity of the deep - sea environment conditions and the correlation of multiple variables, such as it is difficult to fully consider the complexity and dynamic change characteristics of the deep - sea environment; the ability to model the complex interaction between various environmental factors and equipment parameters is insufficient; the lack of effective integration and utilization of historical data and real - time monitoring data; the prediction accuracy is relatively low, and it is difficult to meet the high - reliability requirements of deep - sea and far - sea wind power projects.

[0004] In addition, existing reliability analysis methods often ignore the interdependence between electrical equipment in deep - sea and far - sea wind power converter stations, resulting in inaccurate reliability assessment at the system level. At the same time, these methods have low computational efficiency when dealing with large - scale and high - dimensional data, and it is difficult to meet the needs of real - time monitoring and rapid decision - making.

[0005] Therefore, there is an urgent need to develop a new reliability analysis method that can comprehensively consider the complexity of the deep - sea environment, the diversity of equipment parameters, and the dynamic characteristics of the system, improve the accuracy of reliability assessment and prediction, and provide technical support for the safe operation and maintenance decision - making of deep - sea and far - sea wind power converter stations. Summary of the Invention

[0006] The purpose of the present invention is to provide a reliability analysis method for electrical equipment of deep - sea and far - sea wind power converter stations based on a conditional random field model, so as to solve the problems in the prior art that it is difficult to fully consider the complexity of the deep - sea environment, the diversity of equipment parameters, and the dynamic characteristics of the system, improve the accuracy of reliability assessment and prediction, and provide technical support for the safe operation and maintenance decision - making of deep - sea and far - sea wind power converter stations.

[0007] The purpose of the present invention can be achieved by the following technical solutions:

[0008] A method for analyzing the reliability of electrical equipment in deep sea wind power converter stations based on a conditional random field model comprises the following steps:

[0009] S1, data collection: obtaining historical operating data and environmental factor data of electrical equipment in deep-sea wind power converter stations;

[0010] S2, model construction: Considering the interdependence between devices, a conditional random field model is constructed, with the operating parameters of the electrical equipment as random variables and environmental factors as conditional variables;

[0011] S3, model training: use historical data to train the conditional random field model and optimize model parameters;

[0012] S4, Reliability Analysis: Based on the trained model, the reliability of electrical equipment is predicted and analyzed.

[0013] The historical operation data includes: the operation time of electrical equipment, equipment failure records, equipment performance parameters and maintenance records;

[0014] The environmental factor data include: marine meteorological data, sea state data and geographic location data, wherein the marine meteorological data include wind speed, wind direction, temperature and humidity, the sea state data include wave height, wave direction and water temperature, and the geographic location data include longitude and latitude and water depth.

[0015] The conditional random field model is defined as follows:

[0016] Let X = X 1 ,X 2 ,...,X n represents a set of n random variables, corresponding to various operating parameters of electrical equipment;

[0017] Let Y = Y 1 ,Y 2 ,...,Y m represents a set of m conditional variables, corresponding to environmental factors;

[0018] Construct an initial conditional random field model, whose probability distribution is defined as follows:

[0019]

[0020] Among them, Z(Y) is the normalization factor, f k and g k are node characteristic function and edge characteristic function, respectively, k and μ k is the corresponding weight parameter.

[0021] Construct a device dependency graph G = (V, E), where V represents the device set and E represents the dependency relationship between devices;

[0022] Introduce the joint probability distribution between devices into the initial conditional random field model, and obtain:

[0023]

[0024] where φ i is the node potential function, and ψ ij is the edge potential function.

[0025] The training of the conditional random field model adopts the maximum likelihood estimation method, and the objective function is:

[0026]

[0027] where θ represents the model parameters, N is the number of training samples, X i , Y i is the i-th training sample.

[0028] In step S3, the improved L-BFGS algorithm is used to optimize the objective function and update the model parameters:

[0029]

[0030] where α t is the learning rate, H t is the approximation of the Hessian matrix, and ▽L(θ t ) is the gradient of the objective function.

[0031] Step S4 includes the following sub-steps:

[0032] S41, Using the trained conditional random field model, calculate the reliability function of the electrical equipment under the given environmental conditions:

[0033] R(t|Y) = P(T > t|Y) = 1 - F(t|Y)

[0034] where R(t|Y) is the reliability function under the given environmental condition Y, T is the lifetime random variable of the equipment, and F(t|Y) is the failure distribution function under the given environmental condition Y;

[0035] S42, Calculate the average failure rate function of the electrical equipment:

[0036]

[0037] where f(t|Y) is the failure density function under the given environmental condition Y;

[0038] S43, Estimate the average remaining life of the electrical equipment:

[0039]

[0040] S44, Calculate the mean time between failures:

[0041]

[0042] S45, Calculate the availability:

[0043]

[0044] where MTTR is the mean time to repair;

[0045] S46, Calculate the system reliability: Assume that the converter station contains q key electrical equipment, and their reliabilities are R 1 , R 2 ,..., R q , then the system reliability is:

[0046]

[0047] The step S4 further includes a sensitivity analysis for evaluating the influence degree of different environmental factors on the reliability of electrical equipment:

[0048] Calculate the partial derivatives of the reliability function with respect to each environmental factor:

[0049]

[0050] where Y i is the i-th environmental factor;

[0051] Calculate the normalized sensitivity index based on the partial derivatives of the reliability function with respect to each environmental factor:

[0052]

[0053] where, is the normalized sensitivity index corresponding to the i-th environmental factor.

[0054] The step S4 further includes a multi-scale analysis for evaluating the reliability of electrical equipment on different time scales:

[0055] Short-term reliability prediction: Based on hourly or daily data, predict the equipment reliability within the next 24 hours to 7 days;

[0056] Medium-term reliability assessment: Based on weekly or monthly data, evaluate the equipment reliability trend within the next 1 month to 6 months;

[0057] Long-term reliability planning: Based on annual data, conduct long-term reliability planning for 5 to 10 years;

[0058] For analysis at different scales, corresponding time series preprocessing methods are adopted:

[0059] X t = T t + S t + R t

[0060] Wherein, X t is the original time series, T t is the trend term, S t is the seasonal term, and R t is the random term.

[0061] The method further includes the following steps:

[0062] S5. Optimization suggestions: According to the analysis results, put forward equipment maintenance and optimization suggestions, wherein the optimization suggestions are based on the following criteria:

[0063] Equipment replacement criterion: When the average remaining life of the equipment is lower than the predetermined threshold T threshold , it is recommended to replace the equipment:

[0064] MRL(t|Y) < T threshold

[0065] Preventive maintenance criterion: When the average failure rate of the equipment exceeds the predetermined threshold λ threshold , it is recommended to perform preventive maintenance:

[0066] λ(t|Y) > λ threshold

[0067] Environmental adaptability optimization criterion: Based on the sensitivity analysis results, put forward equipment adaptability optimization suggestions for the environmental factors with greater influence.

[0068] The step S5 further includes dynamic optimization: Based on the real-time monitoring data, periodically update the conditional random field model and dynamically adjust the maintenance strategy:

[0069] Define the model update period;

[0070] Within each update period, collect new operation data and environmental data;

[0071] Use the incremental learning method to update the model parameters:

[0072] θ new = θ old + η·▽L new (θ old )

[0073] Wherein, η is the learning rate, L new is the objective function based on the new data, and θ oldis the old model parameter, θ new is the new model parameter, ▽L new is the gradient of the objective function based on new data;

[0074] Based on the updated model, re-evaluate the equipment reliability and adjust the maintenance strategy.

[0075] Compared with the prior art, the present invention has the following beneficial effects:

[0076] (1) Comprehensively consider the complexity of the deep and far - sea environment: By introducing the conditional random field model, the present invention fully captures the complex relationship between the deep and far - sea environmental factors and the reliability of electrical equipment, improving the accuracy of reliability assessment.

[0077] (2) Adapt to multi - variable and high - dimensional data: The conditional random field model constructed by the present invention can effectively process various environmental factors and equipment parameters, adapting to the complex operating environment of deep and far - sea wind power converter stations.

[0078] (3) Dynamic optimization and real - time update: Through the incremental learning method, the model can continuously absorb new operation data and environmental data, achieving dynamic optimization and improving the adaptability and prediction accuracy of the model.

[0079] (4) Multi - scale analysis ability: The present invention supports short - term, medium - term and long - term reliability analysis, meeting the decision - making needs of different time scales.

[0080] (5) Consider the dependencies between devices: By introducing the device dependency graph and joint probability distribution, the present invention more accurately reflects the reliability characteristics at the system level.

[0081] (6) Strong scalability: The present invention can be extended to the reliability assessment of wind farm groups and intelligent operation and maintenance decision - making support, with broad application prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0082] Figure 1 is the schematic flow chart of the method of the present invention;

[0083] Figure 2 is the schematic diagram of the construction and training process of the conditional random field model of the present invention;

[0084] Figure 3 is the schematic diagram of multi - scale analysis of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0085] The present invention will be described in detail below with reference to the drawings and specific embodiments. This embodiment is implemented on the premise of the technical solution of the present invention, and gives the detailed implementation method and specific operation process, but the protection scope of the present invention is not limited to the following embodiments.

[0086] This embodiment provides a method for analyzing the reliability of electrical equipment in a deep - sea and far - sea wind power converter station based on a conditional random field model, as follows: Figure 1 shown, including the following steps:

[0087] S1, Data collection: Obtain the historical operation data and environmental factor data of the electrical equipment in the deep - sea and far - sea wind power converter station.

[0088] Among them, the historical operation data is collected in real - time by sensors installed on each electrical equipment in the converter station, including: equipment ID, operation time of the electrical equipment, operation status, voltage, current, power, temperature, equipment failure records, equipment performance parameters, maintenance records, etc.

[0089] The environmental factor data is obtained through meteorological stations, buoys, satellite remote sensing, etc., including: marine meteorological data, sea condition data, and geographical location data, etc. Among them, the marine meteorological data includes wind speed, wind direction, air temperature, humidity, and air pressure, etc., the sea condition data includes wave height, wave direction, water temperature, and salinity, etc., and the geographical location data includes longitude, latitude, and water depth.

[0090] After pre - processing the collected raw data, such as cleaning, standardization, and feature extraction, it is used as the variable of the constructed conditional random field model.

[0091] 1) Data cleaning: Remove outliers and missing values. For outliers, the 3σ principle is used for identification; for missing values, interpolation or deletion is selected according to the data characteristics.

[0092] 2) Data standardization: Standardize data with different dimensions, using the Z - score standardization method:

[0093]

[0094] where \(x\) is the raw data, \(\mu\) is the mean, and \(\sigma\) is the standard deviation.

[0095] 3) Feature extraction: Extract features from the raw data to reduce the data dimension and extract effective information. The following methods are mainly used:

[0096] Statistical features: Calculate statistics such as mean, variance, skewness, and kurtosis.

[0097] Time - frequency domain features: Use the fast Fourier transform (FFT) to extract frequency - domain features.

[0098] Wavelet transform: Use the discrete wavelet transform to extract time - frequency domain features.

[0099] S2, Model construction: Considering the mutual dependence relationship between equipment, construct a conditional random field model, taking the operating parameters of electrical equipment as random variables and environmental factors as conditional variables.

[0100] As shown Figure 2 below, it specifically includes the following sub-steps:

[0101] S21, let \(X = X_{1}, X_{2},..., X_{n}\) 1 , \(X_{1}\) 2 ,..., \(X_{n}\) n represent a set of \(n\) random variables corresponding to the operating parameters of the electrical equipment.

[0102] S22, let \(Y = Y_{1}, Y_{2},..., Y_{m}\) 1 , \(Y_{1}\) 2 ,..., \(Y_{m}\) m represent a set of \(m\) conditional variables corresponding to environmental factors.

[0103] S23, construct an initial conditional random field model, and its probability distribution is defined as follows:

[0104]

[0105] where \(Z(Y)\) is the normalization factor, \(f\) k and \(g\) k are the node feature function and the edge feature function respectively, and \(\lambda\) k and \(\mu\) k are the corresponding weight parameters.

[0106] The node feature function is: \(f(X_{i}, Y)=\exp(\beta\cdot g(X_{i}, Y))\) k (\(X_{i}\) i , \(Y\)) \(=\exp(\beta\) k \(\cdot g\) k (\(X_{i}\) i , \(Y\))

[0107] The edge feature function is: \(g(X_{i}, X_{j}, Y)=\exp(\gamma\cdot h(X_{i}, X_{j}, Y))\) k (\(X_{i}\) i , \(X_{j}\) j , \(Y\)) \(=\exp(\gamma\) k \(\cdot h\) k (\(X_{i}\) i , \(X_{j}\) j , \(Y\))

[0108] where \(g\) k and \(h\) k are user-defined feature functions, and \(\beta\) k and \(\gamma\) k are parameters to be learned.

[0109] S24, construct an equipment dependency graph \(G=(V, E)\), where \(V\) represents the set of equipment and \(E\) represents the dependency relationship between equipment.

[0110] 1) The node set \(V\) represents all key electrical equipment in the converter station.

[0111] 2) The edge set \(E\) is determined by the following method:

[0112] Determine the physical connection based on the system topology structure.

[0113] Use mutual information analysis to analyze the statistical correlation between device parameters.

[0114] Combine expert knowledge to determine the functional dependency relationship.

[0115] 3) Assign a weight w to each edge (i, j) ∈ E ij , representing the dependency strength:

[0116] w ij = α·MI(X i , X j ) + β·Topo ij + γ·Expert ij

[0117] where MI(X i , X j ) is the mutual information value, Topo ij is the topological correlation degree, Expert ij is the expert score, and α, β, γ are weight coefficients.

[0118] S25. Introduce the joint probability distribution between devices in the initial conditional random field model to obtain:

[0119]

[0120] where φ i is the node potential function, and ψ ij is the edge potential function.

[0121] In specific implementation, define the node potential function and the edge potential function as follows:

[0122] Node potential function:

[0123] Edge potential function:

[0124] where and are custom feature functions, and are the corresponding weight parameters.

[0125] By introducing these potential functions, the mutual influence between devices can be modeled more accurately, thereby improving the accuracy of the reliability analysis of the entire system.

[0126] S3. Model training: Use historical data to train the conditional random field model and optimize the model parameters.

[0127] In S31, the maximum likelihood estimation method is used for training the conditional random field model, and the objective function is constructed as follows:

[0128]

[0129] Among them, θ represents the model parameters, N is the number of training samples, and X i , Y i is the i-th training sample.

[0130] In S32, the improved L-BFGS algorithm is used to optimize the objective function and update the model parameters:

[0131]

[0132] Among them, α t is the learning rate, H t is the approximation of the Hessian matrix, and ▽L(θ t ) is the gradient of the objective function.

[0133] To improve the training efficiency, the following strategies are adopted:

[0134] 1) Batch training: The training data is divided into multiple small batches, and the data of one batch is used for each update.

[0135] 2) Learning rate decay: As the number of training rounds increases, the learning rate is gradually decreased:

[0136] η t = η 0 ·(1 + α·t) -β

[0137] Among them, η t is the learning rate of the t-th round, η 0 is the initial learning rate, and α and β are decay parameters.

[0138] 3) Regularization: An L2 regularization term is added to the objective function to prevent overfitting:

[0139]

[0140] Among them, λ is the regularization coefficient.

[0141] In S4, reliability analysis: Based on the trained model, the reliability of electrical equipment is predicted and analyzed.

[0142] In S41, using the trained conditional random field model, calculate the reliability function of electrical equipment under given environmental conditions. The specific steps are as follows:

[0143] 1) For the given environmental condition Y, use the conditional random field model to generate the probability distribution of the device parameter X.

[0144] 2) Generate a large number of samples through the Monte Carlo simulation method and calculate the lifetime of each sample.

[0145] 3) Based on the generated samples, estimate the reliability function R(t|Y):

[0146] R(t|Y) = P(T > t|Y) = 1 - F(t|Y)

[0147] where R(t|Y) is the reliability function under the given environmental condition Y, T is the lifetime random variable of the device, and F(t|Y) is the failure distribution function under the given environmental condition Y.

[0148] S42. Calculate the average failure rate function of the electrical equipment:

[0149]

[0150] where f(t|Y) is the failure density function under the given environmental condition Y, and approximate the failure density function f(t|Y) using numerical methods:

[0151]

[0152] where Δt is a small time increment.

[0153] S43. Estimate the average remaining lifetime of the electrical equipment:

[0154]

[0155] In actual calculation, use the numerical integration method for estimation:

[0156]

[0157] where, [t, t max is divided into N small intervals, t i is the midpoint of the i-th interval, and Δt i is the interval length.

[0158] S44. Calculate the mean time to failure (MTTF):

[0159]

[0160] S45. Calculate the availability:

[0161]

[0162] where MTTR is the mean time to repair.

[0163] S46. Calculate the reliability of the computing system: Assume that the converter station contains q key electrical devices, and their reliabilities are R 1 , R 2 ,..., R q , then the system reliability is:

[0164]

[0165] S47. Sensitivity analysis is used to evaluate the influence degree of different environmental factors on the reliability of electrical equipment:

[0166] S471. Calculate the partial derivatives of the reliability function with respect to each environmental factor:

[0167]

[0168] where Y i is the i-th environmental factor.

[0169] To calculate the partial derivatives of the reliability function with respect to each environmental factor, the central difference method is used for estimation, that is:

[0170]

[0171] where ΔY i is a small perturbation of the environmental factor Y i .

[0172] S472. To better display the relative importance of each environmental factor, normalization processing is introduced, and the normalized sensitivity index is calculated based on the partial derivatives of the reliability function with respect to each environmental factor:

[0173]

[0174] where NSI i (t) is the normalized sensitivity index corresponding to the i-th environmental factor.

[0175] S48. Multi-scale analysis is used to evaluate the reliability of electrical equipment on different time scales.

[0176] As Figure 3 shown, it specifically includes:

[0177] S481. Short-term reliability prediction: Based on hourly or daily data, predict the equipment reliability within the next 24 hours to 7 days;

[0178] S482. Medium-term reliability assessment: Based on weekly or monthly data, evaluate the equipment reliability trend within the next 1 month to 6 months;

[0179] S483, Long-term Reliability Planning: Based on grade data, conduct long-term reliability planning for 5 to 10 years. To achieve multi-scale analysis, first decompose the time series data:

[0180] X t = T t + S t + R t

[0181] where X t is the original time series, T t is the trend term, S t is the seasonal term, and R t is the random term.

[0182] The specific implementation uses the STL (Seasonal and Trend decomposition using Loess) method:

[0183] 1) Use Loess smoothing to estimate the trend term T t .

[0184] 2) Remove the trend from the original series and calculate the seasonal term S t .

[0185] 3) The remaining term R t is regarded as random fluctuations.

[0186] Based on the decomposed time series, different methods are used for multi-scale prediction:

[0187] 1) Short-term prediction (24 hours to 7 days): Use the ARIMA model to predict the trend term T t ; Use the periodic pattern to predict the seasonal term S t ; Use the conditional random field model to predict the random term R t .

[0188] 2) Medium-term prediction (1 month to 6 months): Use exponential smoothing to predict the trend term T t ; Use Fourier analysis to predict the seasonal term S t ; Use the conditional random field model to predict the random term R t .

[0189] 3) Long-term prediction (5 years to 10 years): Use regression analysis to predict the trend term T t ; Assume that the seasonal term S t remains stable; Use Monte Carlo simulation to generate the random term R t .

[0190] The final multi-scale prediction result is obtained by synthesizing the predicted values of each component:

[0191]

[0192] S5, Optimization Suggestions: Based on the analysis results, propose suggestions for equipment maintenance and optimization.

[0193] S51, Equipment Replacement Criterion: When the average remaining life of the equipment is lower than the predetermined threshold T threshold it is recommended to replace the equipment:

[0194] MRL(t|Y) < T threshold

[0195] In this embodiment, a fuzzy decision-making method is adopted to generate equipment replacement suggestions:

[0196]

[0197] When ReplaceScore > 0.7, the system will generate equipment replacement suggestions.

[0198] S52, Preventive Maintenance Criterion: When the average failure rate of the equipment exceeds the predetermined threshold λ threshold it is recommended to perform preventive maintenance:

[0199] λ(t|Y) > λ threshold

[0200] In this embodiment, a similar fuzzy decision-making method is adopted to generate preventive maintenance suggestions:

[0201]

[0202] When MaintenanceScore > 0.6, the system will generate preventive maintenance suggestions.

[0203] S53, Environmental Adaptability Optimization Criterion: Based on the sensitivity analysis results, propose equipment adaptability optimization suggestions for the environmental factors with greater influence. The specific steps are as follows:

[0204] 1) Select the environmental factors with NSI i * (t) > 0.1 as the key objects of concern.

[0205] 2) For each key environmental factor, analyze its historical change trend and predict future changes.

[0206] 3) Combine the technical parameters and operating characteristics of the equipment to generate targeted optimization suggestions, such as improving the equipment sealing, enhancing the corrosion resistance, and optimizing the cooling system.

[0207] S54, Dynamic Optimization: Based on the real-time monitoring data, periodically update the conditional random field model and dynamically adjust the maintenance strategy:

[0208] S541. Define the model update period T update ; usually set to 1 week or 1 month.

[0209] S542. During each update period, collect new operation data and environmental data, denoted as D new .

[0210] S543. Combine the historical data D old and the new data D new , and construct a weighted likelihood function:

[0211] L weighted (θ) = α · L old (θ) + (1 - α) · L new (θ)

[0212] where α is the weight of historical data, usually set to 0.7 - 0.9.

[0213] Update the model parameters using the stochastic gradient descent method:

[0214] θ t+1 = θ t - η t · ▽L weighted (θ t )

[0215] where η t is the adaptive learning rate, which can be adjusted using the Adam optimizer.

[0216] S544. Based on the updated model, re - evaluate the equipment reliability and adjust the maintenance strategy.

[0217] To ensure the continuous effectiveness of the model, introduce a model evaluation and selection mechanism:

[0218] 1) Define evaluation metrics, such as log - likelihood, mean squared error (MSE), and mean absolute error (MAE).

[0219] 2) After each update, calculate the performance metrics of the new model on the validation set.

[0220] 3) If the performance of the new model is significantly better than that of the old model (e.g., the performance improvement exceeds 5%), then adopt the new model; otherwise, retain the old model and record the update failure.

[0221] 4) If the update fails continuously for multiple times (e.g., 3 times), trigger the model reconstruction mechanism and retrain the conditional random field model.

[0222] S6. Uncertainty quantification.

[0223] S61, perform posterior sampling on the parameters of the conditional random field model using the Markov Chain Monte Carlo (MCMC) method.

[0224] S62, perform reliability analysis on each set of parameter samples to obtain the result distribution.

[0225] In specific implementation, use the Metropolis-Hastings algorithm for MCMC sampling:

[0226] a) Initialize the parameter θ 0

[0227] b) For t = 1, 2, ..., T:

[0228] Propose a new parameter θ * ~q(θ * |θ t-1 );

[0229] Calculate the acceptance probability:

[0230] Accept θ with probability A * , otherwise keep θ t-1 ;

[0231] where p(θ|D) is the posterior distribution of the parameter, and q(·|·) is the proposal distribution.

[0232] S63, calculate the confidence interval of the reliability index:

[0233]

[0234] where is the reliability estimate, is the standard error, z 1-α / 2 is the critical value of the standard normal distribution.

[0235] Taking the reliability function as an example:

[0236] 1) For a given time point t and environmental condition Y, calculate multiple sample values of the reliability R(t|Y).

[0237] 2) Sort these sample values, and take the 2.5% and 97.5% quantiles as the lower and upper limits of the 95% confidence interval.

[0238] 3) Use the kernel density estimation (KDE) method to estimate the probability density function of the reliability:

[0239]

[0240] where K(·) is the kernel function (such as the Gaussian kernel), and h is the bandwidth parameter.

[0241] In this embodiment, the Monte Carlo method and confidence interval estimation are adopted to provide an uncertainty assessment of the reliability analysis results, providing more comprehensive information support for decision-making.

[0242] S7. Anomaly detection and early warning.

[0243] S71. Calculate the normal probability distribution of device parameters based on the trained conditional random field model;

[0244] S72. Define the anomaly detection threshold τ, usually set as the 1% quantile of the normal sample distribution;

[0245] S73. Perform anomaly detection on the real-time monitoring data x t : If P(x t |Y t ) < τ, then x t is determined to be abnormal;

[0246] To improve the robustness of detection, a sliding window mechanism is introduced to calculate the anomaly probability of the recent w time points:

[0247]

[0248] where I(·) is the indicator function. When P anomaly exceeds the predetermined threshold, an anomaly early warning is triggered.

[0249] S74. When an anomaly is detected, trigger the early warning mechanism and provide preliminary diagnostic suggestions.

[0250] When an anomaly is detected, the system will perform the following steps for fault diagnosis and early warning:

[0251] 1) Based on the FMEA knowledge base, match the most similar fault mode.

[0252] 2) Calculate the probabilities of various possible faults:

[0253]

[0254] where F i represents the i-th fault mode.

[0255] 3) Generate a fault diagnosis report, including possible fault types, fault probabilities, and recommended countermeasures.

[0256] 4) Determine the early warning level (such as general, important, urgent) according to the severity and urgency of the fault.

[0257] 5) Send early warning information through preset communication channels (such as text messages, emails, system alerts).

[0258] S8, Visualization and Decision Support.

[0259] S81, Generate device reliability trend charts, including historical data and prediction results;

[0260] S82, Draw sensitivity analysis heat maps to visually show the influence degree of each environmental factor;

[0261] S83, Construct a multi-dimensional decision matrix, comprehensively considering reliability, economy, and environmental impact;

[0262] S84, Provide an interactive dashboard to support decision-makers for real-time monitoring and strategy adjustment.

[0263] S9, Knowledge Base Construction and Experience Feedback.

[0264] S91, Establish a Failure Mode and Effects Analysis (FMEA) knowledge base to record historical failure cases and solutions;

[0265] S92, Use the analysis results of the conditional random field model to continuously update and improve the knowledge base;

[0266] S93, Establish an experience feedback mechanism to incorporate the actual operation experience of maintenance personnel into the model optimization process.

[0267] S10, Reliability Assessment of Deep Offshore Wind Farm Clusters.

[0268] S101, Extend the reliability analysis method of a single converter station to a wind farm cluster composed of multiple converter stations;

[0269] S102, Consider the network topology structure and power transmission characteristics within the wind farm cluster;

[0270] S103, Introduce the overall reliability index R of the wind farm cluster group :

[0271] R group = f(R system,1 , R system,2 ,..., R system,k , T)

[0272] where R system,i is the system reliability of the i-th converter station, k is the number of converter stations in the wind farm cluster, and T is the network topology structure.

[0273] S11, Intelligent Operation and Maintenance Decision Support.

[0274] S111, Based on the reliability analysis results, combined with the deep reinforcement learning algorithm, construct an intelligent operation and maintenance decision model;

[0275] S112. Define the state space S, action space A, and reward function R:

[0276] R(s,a) = w 1 ·R system +w 2 ·C maintenance +w 3 ·E output

[0277] where w 1 、w 2 、w 3 are weight coefficients, R system is the system reliability, C maintenance is the maintenance cost, and E output is the energy output;

[0278] S113. Use the Deep Q-Network (DQN) or policy gradient method to train the intelligent operation and maintenance decision-making model;

[0279] S114. Implement predictive maintenance to maximize equipment life and system availability.

[0280] This method is applied to the reliability assessment of deep-sea and far-sea wind farm groups, extending the reliability analysis method of a single converter station to a wind farm group composed of multiple converter stations. At the same time, considering the network topology structure and power transmission characteristics within the wind farm group, the overall reliability and output stability of the wind farm group are evaluated.

[0281] In summary, the present invention comprehensively considers the complexity of the deep - sea and far - sea environment. Through the conditional random field model, it effectively captures the complex relationship between deep - sea and far - sea environmental factors and the reliability of electrical equipment, improving the accuracy of reliability assessment; its multi - variable and high - dimensional data processing ability can handle multiple environmental factors and equipment parameters simultaneously, adapting to the complex operating environment of deep - sea and far - sea wind power converter stations; by adopting the incremental learning method, the model can continuously absorb new operating data and environmental data to achieve dynamic optimization, improving the adaptability and prediction accuracy of the model; its multi - scale analysis ability supports short - term, medium - term, and long - term reliability analysis to meet the decision - making needs of different time scales; considering the dependencies between equipment, by introducing equipment dependency graphs and joint probability distributions, it more accurately reflects the reliability characteristics at the system level; using the MCMC method and confidence interval estimation, it provides an uncertainty assessment of the reliability analysis results, providing more comprehensive information support for decision - making; it realizes real - time anomaly detection and early warning, combines with the FMEA knowledge base for fault diagnosis, improving the safety and reliability of the system; by generating suggestions for equipment replacement, preventive maintenance, and environmental adaptability optimization, it provides specific decision - making support for operation and maintenance personnel; by establishing the FMEA knowledge base and experience feedback mechanism, it realizes the accumulation of knowledge and the continuous optimization of the model; this method can be extended and applied to the reliability assessment of wind farm clusters and intelligent operation and maintenance decision - making support, with broad application prospects.

[0282] The method provided by the present invention can significantly improve the accuracy, practicability, and adaptability of the reliability analysis of electrical equipment in deep - sea and far - sea wind power converter stations, providing strong support for the safe operation and economic benefit improvement of deep - sea and far - sea wind power projects.

[0283] The preferred specific embodiments of the present invention have been described in detail above. It should be understood that those of ordinary skill in the art can make many modifications and variations based on the concept of the present invention without creative work. Therefore, all technical solutions that can be obtained by those skilled in the art in the technical field based on the concept of the present invention through logical analysis, reasoning, or limited experiments on the basis of the prior art should fall within the protection scope determined by the claims.

Claims

1. A method for reliability analysis of electrical equipment in deep sea wind power converter stations based on conditional random field model, characterized in that: The following steps are involved: S1, data collection: obtaining historical operating data and environmental factor data of electrical equipment in deep-sea wind power converter stations; S2, model construction: Considering the interdependence between devices, a conditional random field model is constructed, with the operating parameters of the electrical equipment as random variables and environmental factors as conditional variables; S3, model training: use historical data to train the conditional random field model and optimize model parameters; S4, Reliability Analysis: Based on the trained model, the reliability of electrical equipment is predicted and analyzed.

2. The method for analyzing the reliability of electrical equipment in deep sea wind power converter stations based on a conditional random field model according to claim 1 is characterized in that: The historical operation data includes: the operation time of electrical equipment, equipment failure records, equipment performance parameters and maintenance records; The environmental factor data include: marine meteorological data, sea state data and geographic location data, wherein the marine meteorological data include wind speed, wind direction, temperature and humidity, the sea state data include wave height, wave direction and water temperature, and the geographic location data include longitude and latitude and water depth.

3. The method for analyzing the reliability of electrical equipment in deep sea wind power converter stations based on a conditional random field model according to claim 1 is characterized in that: The conditional random field model is defined as follows: Let X = X1, X2, ..., X n represents a set of n random variables, corresponding to various operating parameters of electrical equipment; Let Y = Y1, Y2, ..., Y m represents a set of m conditional variables, corresponding to environmental factors; Construct an initial conditional random field model, whose probability distribution is defined as follows: Among them, Z(Y) is the normalization factor, f k and g k are node characteristic function and edge characteristic function, respectively, k and μ k is the corresponding weight parameter; The node characteristic function is: f k (X i ,Y)=exp(β k ·g k (X i ,Y)) The edge characteristic function is: g k (X i ,X j ,Y)=exp(γ k ·h k (X i ,X j ,Y)) Among them, g k and h k is a custom feature function, β k and γ k are the parameters to be learned; Construct a device dependency graph G = (V, E), where V represents the device set and E represents the dependency relationship between devices; Introducing the joint probability distribution between devices into the initial conditional random field model, we get: Among them, φ i is the node potential function, ψ ij is the edge potential function, The node potential function is: The edge potential function is: in, and is a custom feature function, and is the corresponding weight parameter.

4. The method for analyzing the reliability of electrical equipment in deep sea wind power converter stations based on a conditional random field model according to claim 3 is characterized in that: The training of the conditional random field model adopts the maximum likelihood estimation method, and the objective function is: Among them, θ represents the model parameters, N is the number of training samples, and X i ,Y i is the ith training sample.

5. The method for analyzing the reliability of electrical equipment in deep sea wind power converter stations based on a conditional random field model according to claim 4 is characterized in that: In step S3, the improved L-BFGS algorithm is used to optimize the objective function and update the model parameters: where α t is the learning rate, H t is an approximation of the Hessian matrix, is the gradient of the objective function.

6. The method for analyzing the reliability of electrical equipment in deep sea wind power converter stations based on a conditional random field model according to claim 4 is characterized in that: The step S4 comprises the following sub-steps: S41, using the trained conditional random field model, calculate the reliability function of electrical equipment under given environmental conditions: R(t|Y)=P(T>t|Y)=1-F(t|Y) Where R(t|Y) is the reliability function under given environmental conditions Y, T is the life random variable of the equipment, and F(t|Y) is the failure distribution function under given environmental conditions Y; S42, calculate the average failure rate function of electrical equipment: Where f(t|Y) is the failure density function under given environmental conditions Y; S43, Estimation of the average remaining life of electrical equipment: S44, calculate the mean time between failures: S45, calculate availability: Among them, MTTR is the mean time to repair; S46, calculate system reliability: Assume that the converter station contains q key electrical equipment, whose reliability is R1, R2, ..., R q , then the system reliability is:

7. The method for analyzing the reliability of electrical equipment in deep sea wind power converter stations based on a conditional random field model according to claim 6 is characterized in that: The step S4 also includes a sensitivity analysis for evaluating the influence of different environmental factors on the reliability of the electrical equipment: Calculate the partial derivatives of the reliability function with respect to each environmental factor: where Y i is the i-th environmental factor; The normalized sensitivity index is calculated based on the partial derivative of the reliability function to each environmental factor: in, is the normalized sensitivity index corresponding to the i-th environmental factor.

8. The method for analyzing the reliability of electrical equipment in deep sea wind power converter stations based on a conditional random field model according to claim 6 is characterized in that: The step S4 also includes a multi-scale analysis for evaluating the reliability of the electrical equipment at different time scales: Short-term reliability prediction: Based on hourly or daily data, predict the equipment reliability within the next 24 hours to 7 days; Mid-term reliability assessment: Based on weekly or monthly data, evaluate the equipment reliability trend for the next 1 to 6 months; Long-term reliability planning: Based on grade data, long-term reliability planning for 5 to 10 years is carried out; For analysis of different scales, corresponding time series preprocessing methods are used: X t =T t +S t +R t Among them, X t is the original time series, T t is the trend term, S t is the seasonal term, R t is a random item.

9. The method for analyzing the reliability of electrical equipment in deep sea wind power converter stations based on a conditional random field model according to claim 6 is characterized in that: The method further comprises the following steps: S5, Optimization suggestions: Based on the analysis results, equipment maintenance and optimization suggestions are proposed, wherein the optimization suggestions are based on the following criteria: Equipment replacement criteria: When the average remaining life of the equipment is lower than the predetermined threshold T threshold It is recommended to replace the device: MRL(t|Y)<T threshold Preventive maintenance criterion: When the average failure rate of the equipment exceeds the predetermined threshold λ threshold Preventive maintenance is recommended when: λ(t|Y)>λ threshold Environmental adaptability optimization criteria: Based on the results of sensitivity analysis, equipment adaptability optimization suggestions are made for environmental factors.

10. The method for analyzing the reliability of electrical equipment in deep sea wind power converter stations based on a conditional random field model according to claim 9, characterized in that: The step S5 also includes dynamic optimization: based on real-time monitoring data, periodically updating the conditional random field model and dynamically adjusting the maintenance strategy: Define the model update cycle; During each update cycle, new operational and environmental data are collected; Update model parameters using incremental learning method: i new =θ old +η·▽L new (i old ) Among them, η is the learning rate, L new is the objective function based on new data, θ old is the old model parameter, θ new is the new model parameter, ▽L new is the gradient of the objective function based on the new data; Based on the updated model, reassess equipment reliability and adjust maintenance strategies.

Images