Fan fault risk quantification method and system based on multivariate density estimation network

Abstract

The present application belongs to the technical field of fan fault early warning, and particularly relates to a fan fault risk quantification method and system based on a multivariate density estimation network. The method comprises: obtaining multi-source time series monitoring data of fan operation, and pre-processing the data to obtain an input feature sequence; inputting the input feature sequence into a preset multivariate density estimation network for density estimation to obtain various fan fault risk distributions; based on the various fan fault risk distributions, respectively adopting a Monte Carlo repeated sampling method to obtain a set of various risk prediction probabilities, performing dynamic quantification estimation of fan fault risk boundaries based on the set of various risk prediction probabilities, obtaining various fan fault risk boundaries, and determining various risk probability prediction intervals based on the various fan fault risk boundaries. The method realizes probability density quantification and multi-working-condition adaptive fusion of fan fault risk, and provides quantifiable and interpretable uncertainty evaluation basis for operation and maintenance decisions.

Description

Technical Field

[0001] This invention belongs to the field of wind turbine fault early warning technology, specifically involving a method and system for quantifying wind turbine fault risk based on multivariate density estimation network. Background Technology

[0002] Wind turbine generators operate under complex environmental conditions such as random wind conditions and alternating loads for extended periods. Critical components such as the main shaft, gearbox, generator, pitch and yaw systems, and bearings are prone to failures including wear, fatigue, lubrication deterioration, and insulation aging. Failure to identify and quantify these risks in advance can lead to downtime losses, increased maintenance costs, and even cascading damage. Therefore, fault risk early warning and quantitative assessment based on wind turbine operating conditions have significant engineering value.

[0003] Existing fault early warning technologies are mainly divided into three categories: mechanistic models, threshold / rule methods, and data-driven methods. Mechanistic models rely on precise physical modeling, making it difficult to cover the changes in various operating conditions and parameter drift of wind turbines. Threshold / rule methods are simple to implement but heavily rely on expert experience, resulting in high false alarm and false negative rates and alarm lag when facing different turbine models, different wind farm environments, and switching operating conditions. With the development of artificial intelligence technology, data-driven methods can adaptively extract fault features from historical data and optimize model parameters, greatly improving the ability to identify wind turbine fault modes and becoming the current mainstream fault early warning method. However, it still has the following shortcomings in practical engineering applications: First, existing wind turbine failure risk assessment methods directly output discrete classification labels. However, this deterministic output masks the inherent continuity and uncertainty of the risk itself, fails to depict the probability distribution and severity of failure risks, lacks the ability to quantify failure risk density in a probabilistic sense, and cannot reflect the intermediate state and confidence level of risk evolution. This results in opaque model decision-making logic, and risk assessment remains at the qualitative or semi-quantitative level, making it difficult to provide refined support for subsequent graded early warning.

[0004] Secondly, the inherent uncertainty and reliability of risk probability density estimation models are difficult to quantify and assess. Traditional fault warning models output fault categories, while fault probability density estimation models output fault probability distributions. Accuracy metrics cannot be directly used to evaluate model reliability, making it difficult for maintenance personnel to assess the credibility of warning information. They are forced to rely on experience for manual verification, reducing the practicality and operational efficiency of the warning system and hindering the effective implementation of tiered risk-based operational response decisions. Furthermore, the probability density distributions of different fault types differ significantly, requiring model reliability quantification methods to adapt to the probability distributions of different faults. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for quantifying wind turbine failure risk based on multivariate density estimation networks, which solves the problems of existing technologies being unable to output continuous probability density distributions and having difficulty in assessing the uncertainty of the model itself.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: In a first aspect, the present invention provides a method for quantifying wind turbine failure risk based on a multivariate density estimation network, comprising: Acquire multi-source time-series monitoring data of wind turbine operation and preprocess it to obtain the input feature sequence; The input feature sequence is fed into a pre-defined multivariate density estimation network for density estimation to obtain the failure risk distribution of various wind turbines; Based on the risk distribution of various wind turbine failures, the Monte Carlo repeated sampling method is used to obtain the prediction probability set of various risks. Based on the prediction probability set of various risks, the risk boundary of wind turbine failures is dynamically quantified and estimated to obtain the risk boundary of various wind turbine failures. Based on the risk boundary of various wind turbine failures, the prediction interval of various risk probabilities is determined.

[0007] Optionally, the preprocessing specifically includes: Mean fitting and interpolation are performed on missing samples in multi-source time-series monitoring data to obtain continuous and complete time-series data. Scale the continuous and complete time series data to obtain normalized time series data, which is then used as the input feature sequence.

[0008] Optionally, the preset multivariate density estimation network is constructed through the following steps: Based on the input feature sequence, a probability density function of wind turbine failure risk based on Gaussian distribution is constructed; the probability density function of wind turbine failure risk is parameterized to obtain a parameterized probability density function. A multi-branch density estimation network is constructed based on the parameterized probability density function. The multi-branch density estimation network includes a feature transformation layer and several density estimation branches set in parallel. The input feature sequence is fed into the feature transformation layer, and a unified feature representation is obtained through nonlinear mapping and batch normalization. The unified feature representation is then fed into several parallel density estimation branches. Each density estimation branch performs parameterized modeling of different components in the probability density function of wind turbine failure risk for different operating conditions or different implicit operating states. Each density estimation branch outputs its corresponding density estimation result, forming a multi-branch density estimation result. The density estimation branch includes convolution operation units, batch normalization units, nonlinear activation units, and residual connection structures, which are used to extract multi-scale features and enhance the model's expressive power. The multi-branch density estimation results are adaptively weighted and fused to obtain the overall probability density function of wind turbine failure risk. Based on the overall probability density function, a negative log-likelihood loss function is constructed, and the density estimation network is iteratively trained using the maximum likelihood estimation strategy until the loss is minimized to obtain the multivariate density estimation network.

[0009] Optionally, the input feature sequence is input into a pre-defined multivariate density estimation network for density estimation to obtain the failure risk distribution of various wind turbines, specifically including: The input feature sequence is input into a preset multivariate density estimation network, where it undergoes nonlinear mapping and batch normalization processes in sequence to obtain the output features. Based on the output features, parallel density estimation is performed on the density components corresponding to different operating conditions and different implicit health states, resulting in several density estimation results. These several density estimation results constitute a multi-branch density estimation result. Extract the global descriptive features of each branch from the multi-branch density estimation results; compress the global descriptive features of each branch by global average pooling to obtain several channel descriptors; combine several channel descriptors to obtain a descriptive vector. The description vector is sequentially processed through compression, non-linear activation function, dimension restoration, and normalization to obtain the weight vector; The weight vector and the multi-branch density estimation results are weighted branch by branch to obtain several adaptive weighted fusion features. These adaptive weighted fusion features are then combined and classified through a classification layer to obtain the risk distribution of various types of wind turbine failures.

[0010] Optionally, the step of obtaining prediction probability sets for various risks based on the risk distribution of different types of wind turbines using the Monte Carlo repeated sampling method, dynamically quantifying and estimating the risk boundaries of wind turbine failures based on these risk prediction probability sets, obtaining risk boundaries for various types of wind turbine failures, and determining prediction intervals for various risk probabilities based on these risk boundaries specifically includes: Based on the risk distribution of various wind turbine failures output by the density estimation network, M Monte Carlo repeated samplings are performed on the risk distribution of various wind turbine failures to obtain the prediction probability set of various risks. Based on the empirical quantiles of various risk prediction probability sets and the pre-set confidence level α, prediction intervals are constructed to obtain the risk boundaries of various wind turbine failures. Obtain different preset confidence levels, calculate the risk boundaries of various wind turbine failures under different confidence levels, determine the probability prediction intervals of various risks based on the risk boundaries of various wind turbine failures under different confidence levels, and output the probability prediction intervals of various risks in descending order.

[0011] Optionally, it also includes quantitative calculations of various wind turbine failure risks based on various risk probability prediction intervals, to obtain the coverage rate of various wind turbine failure risks and the breadth of various wind turbine failure risk intervals, specifically:

[0012]

[0013] In the formula, This indicates the coverage rate of wind turbine failure risks; To assess the sample size; For sample index, =1,2,…, ; For the first The upper bound of the prediction interval for each sample; For the first The lower bound of the prediction interval for each sample; To cover indicator variables; Indicates the breadth of the wind turbine failure risk range; For the first The true observed value or true label value corresponding to each sample This refers to the range of values ​​for the target value or the normalization scaling parameter.

[0014] In a second aspect, the present invention provides a wind turbine fault risk quantification system based on a multivariate density estimation network, characterized in that it comprises: The acquisition unit is used to acquire multi-source time-series monitoring data of wind turbine operation and perform preprocessing to obtain the input feature sequence; The density estimation unit is used to input the input feature sequence into a preset multivariate density estimation network for density estimation, so as to obtain the failure risk distribution of various wind turbines; The interval prediction unit is used to obtain the prediction probability set of each type of wind turbine based on the risk distribution of each type of wind turbine failure using the Monte Carlo repeated sampling method. Based on the prediction probability set of each type of wind turbine failure, the wind turbine failure risk boundary is dynamically quantified and estimated to obtain the risk boundary of each type of wind turbine failure. Based on the risk boundary of each type of wind turbine failure, the prediction interval of each type of risk probability is determined.

[0015] Optionally, it also includes: a quantitative calculation unit, used to perform quantitative calculations of various wind turbine failure risks based on various risk probability prediction intervals, to obtain the coverage rate of various wind turbine failure risks and the width of various wind turbine failure risk intervals.

[0016] In a third aspect, the present invention provides an electronic device including a processor and a memory, the processor being configured to execute a computer program stored in the memory to implement the wind turbine fault risk quantification method based on a multivariate density estimation network as described in any one of the preceding claims.

[0017] In a fourth aspect, the present invention provides a computer-readable storage medium storing at least one instruction that, when executed by a processor, implements the wind turbine fault risk quantification method based on a multivariate density estimation network as described above.

[0018] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention acquires multi-source time-series monitoring data and inputs it into a multivariate density estimation network to directly output the probability density distribution of wind turbine failure risk, rather than a single point prediction value as in traditional methods. This allows for a complete characterization of the probability variation of failure risk across different value ranges. Based on this, a Monte Carlo repeated sampling method is used to perform multiple predictions on the same input sample, dynamically estimating the risk boundary and generating prediction intervals with clear confidence levels. This enables maintenance personnel to intuitively understand the possible fluctuation range and uncertainty of failure risk. Therefore, this invention not only achieves probability density quantification and multi-condition adaptive fusion of wind turbine failure risk but also provides quantifiable and interpretable uncertainty evaluation criteria for maintenance decisions, significantly improving the precision and reliability of failure early warning.

[0019] Furthermore, by constructing a parameterized probability density function based on a Gaussian distribution and employing a multi-branch parallel density estimation network, each branch can independently model different operating conditions or implicit operating states, effectively capturing the multi-peak distribution and complex modal characteristics of wind turbine failure risks. The feature transformation layer and residual connection structure enhance the model's ability to extract multi-scale time-series information, while the adaptive weighted fusion mechanism dynamically integrates the outputs of each branch, improving the fitting accuracy and generalization ability of the overall probability density function under different operating conditions. Finally, a training strategy based on negative log-likelihood loss and maximum likelihood estimation enables the network to optimize parameters in a data-driven manner, accurately approximating the true risk distribution. These features collectively achieve high-precision, multi-modal modeling of the probability density of wind turbine failure risks, providing a reliable foundation for subsequent uncertainty quantification and risk boundary estimation.

[0020] Furthermore, by employing nonlinear mapping and batch normalization, the stability and expressive power of the feature representation are improved. Multi-component density estimation based on parallel branches enables separate modeling for different operating conditions and implicit health states, enhancing the model's adaptability to complex operating conditions. Global descriptive features of each branch are further extracted, and a compression-activation mechanism is used to dynamically generate weight vectors. The multi-branch density estimation results are then adaptively weighted and fused branch by branch, allowing the network to automatically adjust the contribution of each component based on the input features, effectively avoiding biases caused by fixed weights or simple averaging. This mechanism significantly improves the fitting accuracy and robustness of the overall probability density function under different operating scenarios, providing more accurate risk distribution input for subsequent Monte Carlo sampling and risk boundary estimation, thereby enhancing the reliability of fault early warning.

[0021] Furthermore, by performing multiple Monte Carlo resampling operations on the same input sample, the inherent randomness and model uncertainty of the wind turbine failure risk distribution are fully captured, forming a risk prediction sample set. Based on the empirical quantiles and pre-set confidence levels of this set, prediction intervals are constructed, enabling dynamic estimation of the fluctuation boundaries of failure risk and avoiding misjudgments caused by single prediction values. By further adjusting the confidence level parameter, risk boundaries at different confidence levels can be obtained, providing multi-level reliability reference intervals for operation and maintenance decisions. This method achieves explicit quantification of failure risk uncertainty, ensuring that early warning results are no longer isolated but accompanied by a confidence range, significantly improving the robustness of failure early warnings and the completeness of decision-making basis.

[0022] Furthermore, based on the prediction interval, risk coverage and interval width can be calculated: coverage is used to verify how well the prediction interval includes the actual risk value, while interval width reflects the accuracy and stability of the prediction. These two indicators together constitute a post-event quantitative evaluation system for the reliability of the early warning results, avoiding the drawback of traditional methods that only provide early warnings without explaining their credibility. Attached Figure Description

[0023] The accompanying drawings, which form part of this application, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an undue limitation of the invention. In the drawings: Figure 1 This is a schematic diagram of the method flow according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the multi-branch density estimation network structure according to an embodiment of the present invention; Figure 3 This is a schematic diagram of the multivariate Gaussian mixture distribution parameters according to an embodiment of the present invention; Figure 4 This is a diagram of the dynamic correction weight generation network structure according to an embodiment of the present invention; Figure 5This is a graph showing the probability quantification results of wind turbine failure risk in an embodiment of the present invention. Figure 6 This is a frequency distribution diagram of wind turbine failure risk according to an embodiment of the present invention; Figure 7 This is a system structure block diagram according to an embodiment of the present invention; Figure 8 This is a structural block diagram of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0024] The present invention will now be described in detail with reference to the accompanying drawings and embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other.

[0025] The following detailed description is exemplary and intended to provide further detailed explanation of the invention. Unless otherwise specified, all technical terms used in this invention have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains. The terminology used in this invention is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention.

[0026] See Figure 1 This application discloses a method for quantifying wind turbine failure risk based on a multivariate density estimation network, characterized by comprising: Acquire multi-source time-series monitoring data of wind turbine operation and preprocess it to obtain the input feature sequence; The input feature sequence is fed into a pre-defined multivariate density estimation network for density estimation to obtain the failure risk distribution of various wind turbines; Based on the risk distribution of various wind turbine failures, a Monte Carlo repeated sampling method is used to obtain the predicted probability sets of various risks. Dynamic quantification estimation of wind turbine failure risk boundaries is then performed based on these risk prediction probability sets to obtain the risk boundaries of various wind turbine failures. Finally, the prediction intervals for various risk probabilities are determined based on these risk boundaries. This invention addresses the problems of existing wind turbine failure early warning methods, such as difficulty in quantifying risk probability density and lack of model uncertainty assessment, by proposing a wind turbine failure risk quantification method and system based on a multivariate density estimation network. This invention acquires and preprocesses multi-source time-series monitoring data to screen key variables; constructs a density estimation network and uses a multivariate Gaussian mixture distribution to parameterize the wind turbine fault risk density function; performs multi-branch parallel density estimation and introduces dynamic correction weights for adaptive weighted fusion to output the wind turbine fault risk density function; constructs a negative log-likelihood loss based on maximum likelihood estimation for training and optimization, and uses Monte Carlo sampling and quantile estimation to obtain fault risk prediction intervals under different confidence levels; this invention achieves probability density quantification of wind turbine fault risk, adaptive fusion of multiple operating conditions, and interpretability assessment of early warning results, significantly improving the precision of fault early warning and the credibility of operation and maintenance decisions.

[0027] In some embodiments, the preprocessing specifically includes: Mean fitting and interpolation are performed on missing samples in multi-source time-series monitoring data to obtain continuous and complete time-series data. Scale the continuous and complete time series data to obtain normalized time series data, which is then used as the input feature sequence.

[0028] In some embodiments, the preset multivariate density estimation network is constructed through the following steps: Based on the input feature sequence, a probability density function of wind turbine failure risk based on Gaussian distribution is constructed; the probability density function of wind turbine failure risk is parameterized to obtain a parameterized probability density function. A multi-branch density estimation network is constructed based on the parameterized probability density function. The multi-branch density estimation network includes a feature transformation layer and several density estimation branches set in parallel. The input feature sequence is fed into the feature transformation layer, and a unified feature representation is obtained through nonlinear mapping and batch normalization. The unified feature representation is then fed into several parallel density estimation branches. Each density estimation branch performs parameterized modeling of different components in the probability density function of wind turbine failure risk for different operating conditions or different implicit operating states. Each density estimation branch outputs its corresponding density estimation result, forming a multi-branch density estimation result. The density estimation branch includes convolution operation units, batch normalization units, nonlinear activation units, and residual connection structures, which are used to extract multi-scale features and enhance the model's expressive power. The multi-branch density estimation results are adaptively weighted and fused to obtain the overall probability density function of wind turbine failure risk. Based on the overall probability density function, a negative log-likelihood loss function is constructed, and the density estimation network is iteratively trained using the maximum likelihood estimation strategy until the loss is minimized to obtain the multivariate density estimation network.

[0029] In some embodiments, the input feature sequence is input into a preset multivariate density estimation network for density estimation to obtain the failure risk distribution of various wind turbines, specifically including: The input feature sequence is input into a preset multivariate density estimation network, where it undergoes nonlinear mapping and batch normalization processes in sequence to obtain the output features. Based on the output features, parallel density estimation is performed on the density components corresponding to different operating conditions and different implicit health states, resulting in several density estimation results. These several density estimation results constitute a multi-branch density estimation result. Extract the global descriptive features of each branch from the multi-branch density estimation results; compress the global descriptive features of each branch by global average pooling to obtain several channel descriptors; combine several channel descriptors to obtain a descriptive vector. The description vector is sequentially processed through compression, non-linear activation function, dimension restoration, and normalization to obtain the weight vector; The weight vector and the multi-branch density estimation results are weighted branch by branch to obtain several adaptive weighted fusion features. These adaptive weighted fusion features are then combined and classified through a classification layer to obtain the risk distribution of various types of wind turbine failures.

[0030] In some embodiments, the step of obtaining various risk prediction probability sets based on the risk distribution of different types of wind turbines using the Monte Carlo repeated sampling method, dynamically quantifying and estimating the risk boundaries of wind turbine failures based on these risk prediction probability sets, obtaining risk boundaries for various types of wind turbine failures, and determining the probability prediction intervals for various types of risks based on these risk boundaries specifically includes: Based on the risk distribution of various wind turbine failures output by the density estimation network, M Monte Carlo repeated samplings are performed on the risk distribution of various wind turbine failures to obtain the prediction probability set of various risks. Based on the empirical quantiles of various risk prediction probability sets and the pre-set confidence level α, prediction intervals are constructed to obtain the risk boundaries of various wind turbine failures. Obtain different preset confidence levels, calculate the risk boundaries of various wind turbine failures under different confidence levels, determine the probability prediction intervals of various risks based on the risk boundaries of various wind turbine failures under different confidence levels, and output the probability prediction intervals of various risks in descending order.

[0031] In some embodiments, the step of obtaining a risk prediction sample set based on the wind turbine failure risk distribution using the Monte Carlo repeated sampling method, dynamically estimating the wind turbine failure risk boundary based on the obtained risk prediction sample set to obtain risk prediction samples, and calculating the prediction interval based on the risk prediction samples specifically includes: Based on the wind turbine failure risk distribution output by the density estimation network, M Monte Carlo resampling operations are performed on the same input sample to obtain a risk prediction sample set. Let the pre-set confidence level be α. Based on the empirical quantiles of the risk prediction sample set obtained from the Monte Carlo repeated sampling, a prediction interval is constructed. Then, the lower bound and upper bound of the prediction interval for the i-th sample are expressed as follows: [Formula omitted for brevity]. upper bound of the prediction interval They are represented as follows:

[0032] in, This represents the interval quantile.

[0033] Furthermore, the single prediction of the Monte Carlo sampling is represented as:

[0034] in, This represents the predicted risk value of the i-th sample in the m-th sampling. For the density estimation network, the prediction mapping function is used. This represents the network random parameter state or random perturbation state corresponding to the m-th sampling. Furthermore, the upper and lower bounds of the wind turbine failure risk prediction interval are repeatedly calculated based on different preset confidence levels to obtain the wind turbine failure risk boundary under different confidence levels, thereby characterizing the fluctuation range and uncertainty of wind turbine failure risk, and providing a basis for subsequent reliability evaluation and risk quantification based on the prediction interval.

[0035] In some embodiments, the method further includes quantifying the risk of various wind turbine failures based on various risk probability prediction intervals to obtain the risk coverage rate of various wind turbine failures and the width of various wind turbine failure risk intervals, specifically:

[0036]

[0037] In the formula, This indicates the coverage rate of wind turbine failure risks; To assess the sample size; For sample index, =1,2,…, ; For the first The upper bound of the prediction interval for each sample; For the first The lower bound of the prediction interval for each sample; To cover indicator variables; Indicates the breadth of the wind turbine failure risk range; For the first The true observed value or true label value corresponding to each sample This refers to the range of values ​​for the target value or the normalization scaling parameter.

[0038] Example 1 See Figures 2-6 A method for quantifying wind turbine failure risk based on multivariate density estimation networks includes: Acquire multi-source time-series monitoring data of wind turbine operation, and perform missing value imputation and normalization on the data to generate an input feature sequence; Based on the input feature sequence, a probability density function of wind turbine failure risk is constructed under the Gaussian distribution assumption, and the probability density function is parameterized and modeled. A multi-branch density estimation network is constructed based on the parameterized probability density function, and the multi-branch structure is used to perform multivariate parallel estimation of the wind turbine failure risk probability density function. The probability density estimation results of each branch are adaptively weighted and fused to obtain the overall probability density function of wind turbine failure risk; Based on the probability density function, a negative log-likelihood loss function is constructed, and the density estimation network is trained using the maximum likelihood estimation strategy. During the inference phase, risk prediction samples are generated using the Monte Carlo sampling method based on the probability density function, and the risk prediction interval is obtained by quantile calculation. Based on the risk prediction interval, the prediction interval coverage probability and the normalized average width of the prediction interval are calculated to quantitatively characterize the wind turbine failure risk.

[0039] Furthermore, the missing value imputation and normalization processing of the multi-source time-series monitoring data of wind turbine operation to generate an input feature sequence specifically includes: A missing value imputation model is constructed to perform mean fitting and interpolation on missing samples in the time series data to obtain continuous and complete time series data; the mean fitting and interpolation adopts the following formula:

[0040] in, This is the fill value for the i-th monitored variable at time t. Let be the actual observed value of the i-th monitored variable at time s. This is the set of neighborhood time indices used to calculate the mean. The size of the neighborhood set; Furthermore, a data normalization model is constructed to unify the scale of the time-series data after outlier processing, mapping each monitored variable to a preset numerical range or satisfying a preset distribution scale, thereby forming normalized time-series data that can be used as input for subsequent models; the data normalization model uses min-max normalization to map each monitored variable to the [0,1] interval, and its expression is:

[0041] in, Let be the normalized value of the i-th monitored variable at time t. Let be the original observation value of the i-th monitored variable at time t. Let i be the minimum value of the i-th monitored variable in the historical data. This represents the maximum value of the i-th monitored variable in the historical data.

[0042] Furthermore, the step of constructing a probability density function for wind turbine failure risk based on the input feature sequence under the Gaussian distribution assumption, and then parameterizing and modeling the probability density function, specifically includes: Based on the statistical characteristics of wind turbine monitoring data being real-valued data, under the Gaussian distribution assumption, the conditional probability of the observed value under given distribution parameters is characterized by a Gaussian likelihood function, and this Gaussian likelihood function is used as the base model for density function modeling; the form of the Gaussian likelihood function is:

[0043] in, For given parameters and The Gaussian likelihood function value of the observed value z under the given conditions; z is the input real-valued observation data, representing the value of the monitoring quantity collected by the wind turbine at time t or the characteristic value composed of the monitoring quantity; is the mean parameter of the Gaussian distribution; Let be the standard deviation parameter of the Gaussian distribution, and .

[0044] Furthermore, based on the input feature sequence, a probability density function for wind turbine failure risk is constructed under the Gaussian distribution assumption, and the probability density function is parameterized and modeled. The mean parameter and standard deviation parameter are parameterized and output by the density estimation network on the latent variable representation. The parameterization method includes at least the following forms:

[0045] Let represent the latent variable representation of the density estimation network at time t for the i-th monitoring channel (or the i-th feature dimension); i is the monitoring channel number or feature dimension number; t is the time index. These are the learnable weight vector and bias corresponding to the mean parameterization branch, respectively; These are the learnable weight vector and bias corresponding to the standard deviation parameterization branch, respectively; This is a Softplus nonlinear function used to ensure the standard deviation parameter. Constantly positive and improve training stability.

[0046] Furthermore, a multi-branch density estimation network model is constructed to perform multivariate parallel estimation of the wind turbine failure risk density function, specifically as follows: The feature data output by the above-mentioned automated time series data processing framework is input into the feature transformation layer of the density estimation network. The feature transformation layer first performs nonlinear mapping on the input features through the GELU activation function to improve the model's ability to express nonlinear operating condition changes and fault evolution features. Subsequently, batch normalization is applied to the features after nonlinear mapping. Scaling and translation factors are used to scale and correct the bias of the extracted features, suppressing distribution drift between different batches and accelerating network convergence. The mathematical expression for batch normalization is:

[0047] Where y is the output feature after batch normalization; x is the input feature value before batch normalization; The input feature x is the mean of the current batch; The variance of the input feature x within the current batch; This is a smoothing term used to avoid zero denominators and improve numerical stability. is the scaling factor, a learnable parameter used to scale and adjust the amplitude of the normalized features; is the translation factor, which is a learnable parameter used to perform bias correction and distribution shifting on the normalized features.

[0048] After completing the feature transformation layer processing, multiple parallel density estimation modules are constructed to form a multi-branch density estimation layer. Each branch of the multi-branch density estimation layer independently parameterizes the wind turbine fault risk density function to achieve parallel modeling of density components corresponding to different operating conditions and different implicit health states. The outputs of each parallel branch are not merged, but are used as inputs for subsequent density mixing, risk quantification or early warning decisions, in order to maintain the distinguishability and interpretability of the estimation results of each branch. Furthermore, each parallel branch contains at least one set of convolutional / transformation units and residual connection structures to enhance deep feature extraction capabilities while preserving the original feature information, and to alleviate gradient vanishing and model degradation problems through residual connections.

[0049] Furthermore, the probability density estimation results from each branch are adaptively weighted and fused to obtain the overall probability density function of wind turbine failure risk. First, a weighting function based on dynamically corrected weights is established to adaptively weight and fuse the parallelized density estimation results, specifically including: The density estimation results of each branch are compressed using global average pooling, and the global descriptive features of each branch are extracted. The feature map output by the multi-branch density estimation layer, containing C branches, is denoted as...

[0050] in, Let H represent the output feature map of the c-th branch, where H and W are the height and width of the feature map, respectively. The feature map of each branch is compressed into a real number, i.e., a channel descriptor, through a global average pooling operation. The calculation formula is as follows:

[0051] in, Output the feature map of the c-th branch in spatial location The value on, This refers to the global descriptive feature of the c-th branch, which embeds the response distribution information of that branch in the global spatial dimension.

[0052] Combine the channel descriptors of all branches to form a description vector, denoted as:

[0053] Furthermore, the global descriptive features are input into a weight generation network consisting of a dimensionality reduction layer, an activation function, and a dimensionality increase layer. Dynamically corrected weights for each branch are learned. The obtained descriptive vector S is input into a weight generation network consisting of at least two fully connected layers. This network first compresses S through a dimensionality reduction layer, then passes it through a non-linear activation function, and finally restores the dimension to the original number of branches C through a dimensionality increase layer. The final weight vector is then obtained through a normalization function. This process is represented as:

[0054] in, The input description vector; These are the weight matrix and bias vector of the first fully connected layer, respectively, used to reduce the dimensionality of the input vector, where r is the dimensionality reduction ratio; This is the first activation function, used to introduce a nonlinear transformation; These are the weight matrix and bias vector of the second fully connected layer, respectively, used to increase the dimensionality of the activated features and restore them to the original number of channels; This is the second activation function, used to normalize the output weight values ​​so that their values ​​are within the range of [0,1]. The final dynamically corrected weight vector is generated, where That is, the weight coefficient corresponding to the c-th branch.

[0055] Furthermore, the learned dynamic correction weights are weighted branch-by-branch with the original density estimation results of each branch to achieve adaptive weighted fusion. The generated dynamic correction weight vector is then... Each weight coefficient in The original density estimation results of the corresponding branch in step 1 Perform element-wise multiplication to obtain a weighted new feature map. The calculation formula is:

[0056] in, This is the original density estimate for the c-th branch. To assign a weight coefficient to the c-th branch, This is the output of the c-th branch after weighted fusion. Perform this operation on all C branches to obtain the overall density estimate output after weighted fusion. .

[0057] The weighted fusion results from each branch are combined to form the final fault risk density function output. Furthermore, the dynamic correction weights are adaptively updated with the input conditions during model training or online inference, enabling the Gaussian component outputs to be weighted according to their contribution to the current risk density estimate under different wind speeds, loads, and environmental conditions. This improves the robustness and generalization ability of the fault risk density function fusion result.

[0058] Furthermore, based on the probability density function, a negative log-likelihood loss function is constructed, and the density estimation network is trained using a maximum likelihood estimation strategy, including: A log-likelihood function of a multivariate Gaussian mixture distribution is constructed in the wind turbine fault risk density estimation network to calculate the probability that the sample is generated by the multivariate Gaussian mixture distribution, given the sample observations and their corresponding network latent variables. The log-likelihood objective function is expressed as:

[0059] in, These are the actual observations or actual label values, used to match the observations corresponding to the model's output distribution parameters; Furthermore, based on the learning objective of maximizing log-likelihood, a negative log-likelihood loss is constructed for the wind turbine fault risk density estimation network, which is used to transform the likelihood maximization problem into a loss function optimization problem that can be directly minimized. When the density estimation network uses a single Gaussian distribution or expands the mixed distribution components into single-component forms for loss calculation, the final loss function is as follows:

[0060] in, is the loss function, used to measure the model's overall explanatory power for the observed samples; i is the sample sequence index; N is the total number of sample sequences; t is the time index; T is the starting time point for prediction or modeling; T is the ending time point for prediction or modeling. Let be the observation value of the i-th time series sample at time t, representing the value of the wind turbine monitoring quantity or the characteristic value composed of the monitoring quantity; An abstract distribution for density estimation networks; This is a parameter mapping function used to map the network's latent variable representation to a set of distributed parameters; Furthermore, the loss quantization method minimizes the loss during training. To optimize the objective, the parameters of the density estimation network are updated to maximize the log-likelihood of the probability density estimation results for the wind turbine under normal and fault evolution states, thereby improving the accuracy and robustness of fault risk density modeling. During training, the Adam optimizer is used to update the network parameters. The Adam optimizer uses the momentum method to accelerate the gradient descent process, while the adaptive learning rate improves the convergence speed and accuracy of the neural network model. A bias correction method is used to adjust the momentum at each order to balance the gradient proportion at each iteration of network training. The calculation formula is as follows:

[0061] in, First-order momentum It is second momentum; and The exponential decay rate; and These are deviation corrections for first-order momentum and second-order momentum, respectively. It is a relatively small constant that ensures numerical stability during training; The learning rate; These are the model parameters.

[0062] Furthermore, risk prediction samples are generated using the Monte Carlo sampling method based on the probability density function, and the risk prediction interval is obtained through quantile calculation, specifically including: Based on the wind turbine failure risk distribution output by the density estimation network, a Monte Carlo repeated sampling strategy is adopted to capture the overall characteristics and fluctuation patterns of the failure risk distribution through multiple random sampling operations.

[0063] Furthermore, a dynamic wind turbine failure risk boundary based on quantiles is constructed, and the boundary is accurately estimated under a pre-set confidence level to quantify the range of risk fluctuation. Based on massive sample data obtained from Monte Carlo repeated sampling, the upper and lower bounds of the failure risk prediction interval corresponding to pre-set confidence levels (such as 95% and 99%) are screened and calculated to clarify the range of wind turbine failure risk fluctuation under different confidence levels.

[0064] With the preset reliability as At the same time, for the same sample Perform M Monte Carlo sampling operations to obtain the risk prediction sample set for the i-th sample. And based on empirical quantiles, the upper and lower bounds of the prediction interval are calculated as follows:

[0065] The single prediction of the Monte Carlo sampling can be expressed as: ; This represents the predicted risk value of the i-th sample in the m-th sampling. For the density estimation network, the prediction mapping function is used. This represents the network random parameter state or random perturbation state corresponding to the m-th sampling. Furthermore, a wind turbine risk early warning strategy based on fault risk boundaries is adopted, which directly maps the predicted interval to an early warning credibility index. For example, "narrow risk interval under high confidence" corresponds to a high credibility early warning, while "wide interval under low confidence" triggers secondary verification. The transparency and quantifiable interpretation of the early warning results are achieved through the correlation between interval width and confidence level.

[0066] With the preset reliability as At the same time, for the same sample Perform M Monte Carlo sampling operations to obtain the risk prediction sample set for the i-th sample. And based on empirical quantiles, the upper and lower bounds of the prediction interval are calculated as follows:

[0067] Furthermore, methods for quantitatively characterizing the reliability and uncertainty of the wind turbine failure risk density estimation network model specifically include: The reliability of the prediction interval is evaluated based on the Probability of Predicted Interval Coverage (PICP), which measures the proportion of actual observations falling within the prediction interval. Its expression is:

[0068] The uncertainty of the prediction interval is evaluated based on the normalized average width (PINAW) of the prediction interval. PINAW measures the width of the prediction interval, and its expression is:

[0069] Where n is the number of evaluation samples; i is the sample index, i=1,2,…,n; This is the upper bound of the prediction interval for the i-th sample; This is the lower bound of the prediction interval for the i-th sample; Let be the prediction interval for the i-th sample under the preset confidence level; For the i-th sample, it represents the true observation or true label value. ; To cover indicator variables, used to represent the true value Whether it falls within the prediction interval; R is the range of the target value or the normalization scale parameter, used to normalize the width of the prediction interval; Furthermore, the quantitative characterization method repeatedly calculates PICP and PINAW at different confidence levels to obtain a quantitative relationship between reliability and uncertainty as confidence changes, and uses the upper and lower bounds of the prediction interval and its width as the interpretability output of the early warning result, enabling operation and maintenance personnel to make credible decisions on the wind turbine failure risk early warning result based on the trade-off relationship between "coverage degree - interval width".

[0070] Example 2 like Figure 7 As shown, based on the same inventive concept as the above embodiments, the present invention also provides a wind turbine fault risk quantification system based on a multivariate density estimation network, comprising: The acquisition unit is used to acquire multi-source time-series monitoring data of wind turbine operation and perform preprocessing to obtain the input feature sequence; The density estimation unit is used to input the input feature sequence into a preset multivariate density estimation network for density estimation to obtain the wind turbine failure risk distribution; and to classify the wind turbine failure risk distribution to obtain the failure risk distribution of each type of wind turbine. The interval prediction unit is used to obtain various risk prediction sample sets based on the risk distribution of various wind turbines using the Monte Carlo repeated sampling method, to dynamically estimate the risk boundary of wind turbines based on the risk prediction sample sets, to obtain various risk prediction samples, and to determine various risk prediction intervals based on the various risk prediction samples.

[0071] In some embodiments, a quantification calculation unit is also included, which is used to perform quantification calculation of wind turbine failure risk based on various risk prediction intervals to obtain the coverage rate of various wind turbine failure risks and the width of various wind turbine failure risk intervals.

[0072] In some embodiments, the density estimation unit inputs the input feature sequence into a preset multivariate density estimation network for density estimation to obtain the wind turbine failure risk distribution, specifically including: The input feature sequence is input into a preset multivariate density estimation network, where it undergoes nonlinear mapping and batch normalization processes in sequence to obtain the output features. Based on the output features, parallel density estimation is performed on the density components corresponding to different operating conditions and different implicit health states, resulting in several density estimation results. These several density estimation results constitute a multi-branch density estimation result. Extract the global descriptive features of each branch from the multi-branch density estimation results; compress the global descriptive features of each branch by global average pooling to obtain several channel descriptors; combine several channel descriptors to obtain a descriptive vector. The description vector is sequentially processed through compression, non-linear activation function, dimension restoration, and normalization to obtain the weight vector; The weight vector and the multi-branch density estimation results are weighted branch by branch to obtain several adaptive weighted fusion features. These adaptive weighted fusion features are then combined to obtain the wind turbine failure risk distribution.

[0073] Example 3 like Figure 8 As shown, the present invention also provides an electronic device 100 for implementing a wind turbine fault risk quantification method and system based on a multivariate density estimation network; The electronic device 100 includes a memory 101, at least one processor 102, a computer program 103 stored in the memory 101 and executable on at least one processor 102, and at least one communication bus 104.

[0074] The memory 101 can be used to store the computer program 103. The processor 102 implements a method and system steps for quantifying wind turbine failure risk based on a multivariate density estimation network by running or executing the computer program stored in the memory 101 and calling the data stored in the memory 101.

[0075] The memory 101 may primarily include a program storage area and a data storage area. The program storage area may store the operating system, application programs required for at least one function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created based on the use of the electronic device 100 (such as audio data), etc. In addition, the memory 101 may include non-volatile memory, such as hard disk, RAM, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other non-volatile solid-state storage device.

[0076] At least one processor 102 may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Processor 102 may be a microprocessor or any conventional processor. Processor 102 is the control center of electronic device 100, connecting various parts of electronic device 100 via various interfaces and lines.

[0077] The memory 101 in the electronic device 100 stores multiple instructions to implement a wind turbine fault risk quantification method and system based on a multivariate density estimation network, and the processor 102 can execute multiple instructions to achieve the following: Acquire multi-source time-series monitoring data of wind turbine operation and preprocess it to obtain the input feature sequence; The input feature sequence is fed into a pre-defined multivariate density estimation network for density estimation to obtain the wind turbine failure risk distribution; Based on the distribution of wind turbine failure risks, the Monte Carlo repeated sampling method is used to obtain a risk prediction sample set. Based on the obtained risk prediction sample set, the wind turbine failure risk boundary is dynamically estimated to obtain risk prediction samples. Based on the risk prediction samples, the prediction interval is calculated. Based on the prediction interval, the wind turbine failure risk is quantitatively calculated to obtain the wind turbine failure risk coverage rate and the width of the wind turbine failure risk interval.

[0078] Example 4 If the modules / units integrated in the electronic device 100 are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, and read-only memory (ROM).

[0079] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0080] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0081] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0082] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0083] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0084] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for quantifying wind turbine failure risk based on multivariate density estimation networks, characterized in that, include: Acquire multi-source time-series monitoring data of wind turbine operation and preprocess it to obtain the input feature sequence; The input feature sequence is fed into a pre-defined multivariate density estimation network for density estimation to obtain the failure risk distribution of various wind turbines; Based on the risk distribution of various wind turbine failures, the Monte Carlo repeated sampling method is used to obtain the prediction probability set of various risks. Based on the prediction probability set of various risks, the risk boundary of wind turbine failures is dynamically quantified and estimated to obtain the risk boundary of various wind turbine failures. Based on the risk boundary of various wind turbine failures, the prediction interval of various risk probabilities is determined.

2. The wind turbine failure risk quantification method based on multivariate density estimation network according to claim 1, characterized in that, The preprocessing specifically includes: Mean fitting and interpolation are performed on missing samples in multi-source time-series monitoring data to obtain continuous and complete time-series data. Scale the continuous and complete time series data to obtain normalized time series data, which is then used as the input feature sequence.

3. The wind turbine failure risk quantification method based on multivariate density estimation network according to claim 1, characterized in that, The preset multivariate density estimation network is constructed through the following steps: Based on the input feature sequence, a probability density function of wind turbine failure risk based on Gaussian distribution is constructed; the probability density function of wind turbine failure risk is parameterized to obtain a parameterized probability density function. A multi-branch density estimation network is constructed based on the parameterized probability density function. The multi-branch density estimation network includes a feature transformation layer and several density estimation branches set in parallel. The input feature sequence is input into the feature transformation layer, and a unified feature representation is obtained through nonlinear mapping and batch normalization. The unified feature representation is then input into several parallel density estimation branches. Each density estimation branch performs parameterized modeling of different components in the probability density function of wind turbine failure risk for different operating conditions or different implicit operating states. Each density estimation branch outputs its corresponding density estimation result, forming a multi-branch density estimation result. The density estimation branch includes a convolution operation unit, a batch normalization unit, a nonlinear activation unit, and a residual connection structure, which are used to extract multi-scale features and enhance the model's expressive power. The multi-branch density estimation results are adaptively weighted and fused to obtain the overall probability density function of wind turbine failure risk. Based on the overall probability density function, a negative log-likelihood loss function is constructed, and the density estimation network is iteratively trained using the maximum likelihood estimation strategy until the loss is minimized to obtain the multivariate density estimation network.

4. The wind turbine failure risk quantification method based on multivariate density estimation network according to claim 1, characterized in that, The input feature sequence is fed into a pre-defined multivariate density estimation network for density estimation, resulting in the distribution of fault risks across various wind turbine types, specifically including: The input feature sequence is input into a preset multivariate density estimation network, where it undergoes nonlinear mapping and batch normalization processes in sequence to obtain the output features. Based on the output features, parallel density estimation is performed on the density components corresponding to different operating conditions and different implicit health states, resulting in several density estimation results. These several density estimation results constitute a multi-branch density estimation result. Extract the global descriptive features of each branch from the multi-branch density estimation results; compress the global descriptive features of each branch by global average pooling to obtain several channel descriptors; combine several channel descriptors to obtain a descriptive vector. The description vector is sequentially processed through compression, non-linear activation function, dimension restoration, and normalization to obtain the weight vector; The weight vector and the multi-branch density estimation results are weighted branch by branch to obtain several adaptive weighted fusion features. These adaptive weighted fusion features are then combined and classified through a classification layer to obtain the risk distribution of various types of wind turbine failures.

5. The wind turbine failure risk quantification method based on multivariate density estimation network according to claim 1, characterized in that, Based on the risk distribution of various wind turbine failures, the Monte Carlo repeated sampling method is used to obtain the predicted probability sets of various risks. Based on these risk prediction probability sets, dynamic quantitative estimation of wind turbine failure risk boundaries is performed to obtain the risk boundaries of various wind turbine failures. Based on these risk boundaries, the predicted probability intervals of various risks are determined. Specifically, this includes: Based on the risk distribution of various wind turbine failures output by the density estimation network, M Monte Carlo repeated samplings are performed on the risk distribution of various wind turbine failures to obtain the prediction probability set of various risks. Based on the empirical quantiles of various risk prediction probability sets and the pre-set confidence level α, prediction intervals are constructed to obtain the risk boundaries of various wind turbine failures. Obtain different preset confidence levels, calculate the risk boundaries of various wind turbine failures under different confidence levels, determine the probability prediction intervals of various risks based on the risk boundaries of various wind turbine failures under different confidence levels, and output the probability prediction intervals of various risks in descending order.

6. The wind turbine failure risk quantification method based on multivariate density estimation network according to claim 1, characterized in that, It also includes quantitative calculations of various wind turbine failure risks based on various risk probability prediction intervals, to obtain the coverage rate of various wind turbine failure risks and the width of various wind turbine failure risk intervals, specifically: In the formula, This indicates the coverage rate of wind turbine failure risks; To assess the sample size; For sample index, =1,2,…, ; For the first The upper bound of the prediction interval for each sample; For the first The lower bound of the prediction interval for each sample; To cover indicator variables; This indicates the breadth of the wind turbine failure risk range; For the first The true observed value or true label value corresponding to each sample This refers to the range of values ​​for the target value or the normalization scaling parameter.

7. A wind turbine failure risk quantification system based on multivariate density estimation network, characterized in that, include: The acquisition unit is used to acquire multi-source time-series monitoring data of wind turbine operation and perform preprocessing to obtain the input feature sequence; The density estimation unit is used to input the input feature sequence into a preset multivariate density estimation network for density estimation, so as to obtain the failure risk distribution of various wind turbines; The interval prediction unit is used to obtain the prediction probability set of each type of wind turbine based on the risk distribution of each type of wind turbine failure using the Monte Carlo repeated sampling method. Based on the prediction probability set of each type of wind turbine failure, the wind turbine failure risk boundary is dynamically quantified and estimated to obtain the risk boundary of each type of wind turbine failure. Based on the risk boundary of each type of wind turbine failure, the prediction interval of each type of risk probability is determined.

8. The wind turbine failure risk quantification system based on a multivariate density estimation network according to claim 7, characterized in that, It also includes a quantitative calculation unit, which is used to perform quantitative calculations of various wind turbine failure risks based on various risk probability prediction intervals, and to obtain the coverage rate of various wind turbine failure risks and the width of various wind turbine failure risk intervals.

9. An electronic device, characterized in that, It includes a processor and a memory, the processor being used to execute a computer program stored in the memory to implement the wind turbine failure risk quantification method based on a multivariate density estimation network as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores at least one instruction, which, when executed by a processor, implements the wind turbine fault risk quantification method based on a multivariate density estimation network as described in any one of claims 1 to 7.

Images