New energy station equipment multi-source data fusion diagnosis method and system

Abstract

The invention provides a multi-source data fusion diagnosis method and system for new energy station equipment, and relates to the technical field of intelligent operation and maintenance of power equipment, and the method comprises the steps: collecting equipment operation data, and carrying out the preprocessing to obtain standardized data; transient features and drift features are extracted, a transient fault mode and a degradation fault mode are identified, and a dual-frequency-domain fault feature library is established; calculating a power transmission fluctuation value between devices, analyzing energy transmission efficiency, positioning a fault source device and identifying a fault propagation path; establishing a fault diagnosis evaluation index for grading diagnosis, and evaluating a fault development trend; and generating and executing a maintenance decision scheme. According to the invention, through multi-source data fusion analysis, accurate diagnosis and early warning of equipment faults of the new energy station are realized, and the equipment operation and maintenance efficiency is improved.

Description

Technical Field

[0001] The present invention relates to intelligent operation and maintenance technology for electric power equipment, and in particular to a multi-source data fusion diagnosis method and system for new energy station equipment. Background Art

[0002] There are many types of equipment in new energy stations, including wind turbines, photovoltaic modules, energy storage systems and converters. These devices are affected by environmental changes, operating load fluctuations and equipment aging during long-term operation, and may have various types of faults, such as transient faults, degradation faults and systemic faults. At present, traditional equipment monitoring methods mostly rely on a single data source, such as single variable monitoring such as voltage, current or temperature, which makes it difficult to accurately identify complex fault modes. At the same time, existing fault diagnosis methods are mainly based on single device analysis, lack of in-depth research on the energy transfer relationship between devices, and it is difficult to effectively identify the fault propagation path and potential impact range, resulting in delayed fault handling and affecting the overall operation efficiency of the station.

[0003] With the expansion of the scale of new energy stations and the increase in the demand for intelligence, equipment fault diagnosis methods based on multi-source data fusion have gradually become a research hotspot. By fusing multiple sensor data, such as electrical signals, mechanical vibrations, temperature changes, etc., the equipment operation status information can be fully extracted, and combined with machine learning, frequency domain analysis and other methods, accurate identification of fault modes can be achieved. In addition, by analyzing the energy transfer relationship between devices, the source and propagation path of the fault can be identified, providing a more accurate basis for early warning and maintenance. However, there is currently a lack of a systematic multi-source data fusion method that can effectively integrate multi-dimensional data, build a dual-frequency domain fault feature library, and combine historical data for trend prediction to improve the level of equipment health management.

[0004] The efficient operation of new energy stations depends on accurate equipment fault diagnosis and maintenance decisions. Traditional fault monitoring methods are difficult to meet the needs of multi-device collaborative diagnosis. Therefore, there is an urgent need for a new energy station equipment diagnosis method based on multi-source data fusion, which can comprehensively analyze the equipment operation status, identify fault characteristics, track fault propagation paths, and provide intelligent maintenance decisions to improve the reliability and operation and maintenance efficiency of new energy station equipment. Summary of the invention

[0005] The embodiments of the present invention provide a multi-source data fusion diagnosis method and system for new energy station equipment, which can solve the problems in the prior art.

[0006] According to a first aspect of the embodiments of the present invention, A multi-source data fusion diagnosis method for new energy station equipment is provided, comprising: Collect and pre-process the operating data of new energy station equipment to obtain standardized operating data; Extract transient features and drift features from standardized operation data, identify equipment transient fault modes based on transient features, identify equipment degradation fault modes based on drift features, and establish a dual-frequency domain fault feature library based on transient fault modes and degradation fault modes; Calculate the power transfer fluctuation value between devices and analyze the energy transfer efficiency of the device group. When the energy transfer efficiency is lower than the preset transfer threshold, locate the fault source device based on the device fault mode library, identify the fault propagation path based on the fault source device, and generate a fault impact link diagram. Establish fault diagnosis evaluation indicators based on the fault impact link diagram and dual-frequency domain fault feature library, perform graded diagnosis on the equipment based on the fault diagnosis evaluation indicators, and evaluate the fault development trend in combination with the equipment's historical fault data, output fault warning information and equipment health status assessment report; Generate maintenance decision plans based on fault warning information and equipment health status assessment reports, execute maintenance decision plans, and record maintenance effect data.

[0007] In an optional embodiment, Extract transient features and drift features from standardized operation data, identify equipment transient fault modes based on transient features, identify equipment degradation fault modes based on drift features, and establish a dual-frequency domain fault feature library based on transient fault modes and degradation fault modes, including: Calculating the data volatility of the standardized operation data, constructing an adaptive harmonic basis function group based on the data volatility, and decomposing the standardized operation data to obtain a transient component and a drift component; Extract mutation characteristic parameters of transient component and trend characteristic parameters of drift component respectively; A transient feature matrix is ​​constructed based on mutation feature parameters, the transient feature matrix is ​​matched with a preset transient fault template, a multi-dimensional cosine similarity is calculated to obtain a transient fault similarity matrix, and a device transient fault mode is determined according to a fault type corresponding to a maximum similarity in the transient fault similarity matrix; A drift feature matrix is ​​constructed based on trend feature parameters, the drift feature matrix is ​​matched with a preset degradation fault template, a multi-dimensional Euclidean distance is calculated to obtain a degradation fault similarity matrix, and a device degradation fault mode is determined according to a fault type corresponding to a minimum distance in the degradation fault similarity matrix; The instantaneous fault similarity matrix and the degradation fault similarity matrix are weightedly fused to obtain a comprehensive fault similarity matrix. The instantaneous fault mode and the degradation fault mode are credibility evaluated according to the comprehensive fault similarity matrix. The fault modes that pass the credibility evaluation and the corresponding feature matrices are combined to establish a dual-frequency domain fault feature library.

[0008] In an optional embodiment, The credibility evaluation of the transient fault mode and the degradation fault mode according to the comprehensive fault similarity matrix includes: Collecting the working condition state vector including load power, ambient wind speed, ambient temperature, vibration frequency characteristics and operation efficiency index, and performing maximum and minimum value normalization processing on the working condition state vector to obtain normalized working condition characteristics; Performing cluster analysis based on the normalized working condition characteristics to obtain an initial working condition boundary value, calculating the working condition distribution of the fault mode in the comprehensive fault similarity matrix to obtain a boundary loss function, performing a weighted combination of the boundary loss function and the standard deviation of the normalized working condition characteristics to obtain a boundary update amount, and dynamically optimizing the initial working condition boundary value based on the boundary update amount to obtain an optimized working condition boundary value; Calculating the matching degree of the fault mode in the comprehensive fault similarity matrix under the current working condition to obtain the local fitness, performing exponential smoothing on the local fitness within a specified time window to obtain the time series fitness, and adaptively weighting the local fitness and the time series fitness to obtain the comprehensive working condition fitness; The comprehensive fault similarity matrix and the comprehensive operating condition fitness are weightedly mapped to obtain a confidence reference value, the confidence reference value is multiplied by the constraint function of the optimized operating condition boundary value to obtain a modified confidence, the time-varying correction factor is calculated according to the time series standard deviation of the comprehensive operating condition fitness and the sliding window mean, and the modified confidence and the time-varying correction factor are weightedly fused according to the dynamic weight coefficient to obtain the credibility evaluation results of the transient fault mode and the degradation fault mode.

[0009] In an optional embodiment, Calculate the power transfer fluctuation value between devices and analyze the energy transfer efficiency of the device group. When the energy transfer efficiency is lower than the preset transfer threshold, locate the fault source device in combination with the device fault mode library, identify the fault propagation path based on the fault source device, and generate the fault impact link diagram including: Based on the device output power and the adjacent device input power within the sliding time window, the power difference squared at each sampling moment is multiplied by the corresponding time weight coefficient and then accumulated to obtain the power transfer fluctuation value between devices; The physical connection relationship between the collected devices is used to construct a topological connection matrix, a device transfer fluctuation matrix is ​​constructed based on the power transfer fluctuation value and the topological connection matrix, and the energy transfer efficiency of the device group is obtained according to the ratio of the device transfer fluctuation matrix to a preset reference fluctuation matrix; When the energy transfer efficiency of the device group is lower than the preset transfer threshold, the fault feature vector of each device is calculated based on the device transfer fluctuation matrix, and the fault feature vector is matched with the feature vector in the dual-frequency domain fault feature library for similarity to determine the fault source device; Taking the fault source device as the starting point, a fault propagation influence matrix is ​​constructed based on the device transfer fluctuation matrix and the topological connection matrix, wherein the propagation influence value of a directly connected device pair is its power transfer fluctuation value, and the propagation influence value of an indirectly connected device pair is the multiplication of the power transfer fluctuation value on the path divided by the power of the shortest path length; The priority path of fault propagation is determined based on the size of each element in the fault propagation influence matrix, the fault propagation probability of the priority path is calculated in combination with historical fault data, and a fault influence link diagram including the fault propagation path and propagation probability is generated.

[0010] In an optional embodiment, Determining a priority path for fault propagation based on the size of each element in the fault propagation influence matrix, and calculating the fault propagation probability of the priority path in combination with historical fault data includes: Constructing a propagation path candidate set based on the fault propagation impact matrix, wherein each propagation path includes a sequence of propagation impact values ​​of adjacent device pairs; Performing wavelet decomposition on the propagation impact value sequence to obtain impact value components of different frequency bands, taking the components whose impact value components are less than a preset component threshold as the propagation baseline sequence, and taking the components whose impact value components are greater than the preset component threshold as the fluctuation sequence; The path stability is obtained by dividing the mean of the propagation baseline sequence by the standard deviation of the fluctuation sequence, and the overlap degree between the peak-to-valley distribution characteristics of the fluctuation sequence and the historical fault time is used as the path credibility. The propagation path candidate set is screened according to the path stability; For the selected propagation paths, the sample entropy of the propagation impact value sequence is calculated, the size of the sample entropy is used as the path complexity evaluation index, and the propagation path with the smallest path complexity is selected as the fault propagation priority path; Based on historical fault data, the fault occurrence time sequence correlation of adjacent devices on the propagation priority path is calculated, the fault occurrence time sequence correlation is constructed as a priori probability distribution in the form of a Gaussian distribution, the current propagation impact value sequence is subjected to wavelet transform, the wavelet coefficients at different scales are extracted, the amplitude and phase characteristics of the wavelet coefficients are combined into an observation vector, and the prediction probability at the current moment is calculated based on the posterior probability at the previous moment; A likelihood function is calculated according to the observation vector, the predicted probability is multiplied by the likelihood function and normalized to obtain the posterior probability at the current moment, iteration is repeated until the posterior probability converges, and the converged posterior probability is used as the fault propagation probability of the fault propagation priority path.

[0011] In an optional embodiment, Establish fault diagnosis evaluation indicators based on the fault impact link diagram and dual-frequency domain fault feature library, perform graded diagnosis on the equipment based on the fault diagnosis evaluation indicators, and evaluate the fault development trend in combination with the equipment's historical fault data, output fault warning information and equipment health status evaluation reports including: Extracting common nodes of intersecting links from the fault-affected link graph, calculating the state coordination degree of the common nodes to construct a link coupling matrix, screening features in a dual-frequency domain fault feature library according to the link coupling matrix, calculating the matching degree between the features and the link coupling matrix, and selecting features with high matching degree to construct a candidate feature set; Extract the change sequence of the candidate feature set in the historical fault data, analyze the causal relationship of the change sequence to construct a feature propagation network, identify the strongest causal path from the feature propagation network as a feature diagnosis chain, calculate the propagation strength of each node in the feature diagnosis chain, take the product of the propagation strength and the feature matching degree as the feature weight, and perform weighted combination of the candidate feature set according to the feature weight to obtain the fault diagnosis evaluation index; The probability distribution of fault diagnosis evaluation indicators in historical fault data is statistically analyzed, and the health level division threshold is determined according to the probability distribution. The interval to which the current fault diagnosis evaluation indicator belongs is determined as the equipment health level. The fault development trend is predicted according to the state transition law of the characteristic diagnosis chain, and the fault warning information and health status assessment report are output in combination with the equipment health level.

[0012] In an optional embodiment, Statistically analyzing the probability distribution of fault diagnosis evaluation indicators in historical fault data, and determining the health level classification threshold according to the probability distribution includes: Calculate the time series fluctuation characteristics of the fault diagnosis evaluation index in the historical fault data to obtain a local stability index, dynamically adjust the resampling disturbance intensity according to the local stability index, and use the disturbance intensity to resample the fault diagnosis evaluation index to generate multiple groups of sample data; The inter-class distribution distance, intra-class distribution distance and time series continuity constraints of multiple groups of sample data are combined to construct a multi-objective optimization function, and multiple groups of initial thresholds are obtained through iterative optimization. The kernel density estimation method was used to analyze the distribution characteristics of multiple groups of initial thresholds, and the dynamic intervals of health grade division were constructed based on quantiles at different confidence levels. The distance distribution from the current fault diagnosis evaluation index to the dynamic interval is calculated, the distance distribution is converted into a probability distribution, the information entropy is calculated based on the probability distribution and converted into a credibility index, and the dynamic interval is optimized in combination with the credibility index to obtain the final health level division threshold.

[0013] According to a second aspect of the embodiments of the present invention, Provided is a multi-source data fusion diagnosis system for new energy station equipment, including: The first unit is used to collect and pre-process the operating data of new energy station equipment to obtain standardized operating data; The second unit is used to extract transient features and drift features from the standardized operation data, identify the transient fault mode of the equipment based on the transient features, identify the degradation fault mode of the equipment based on the drift features, and establish a dual-frequency domain fault feature library based on the transient fault mode and the degradation fault mode; The third unit is used to calculate the power transfer fluctuation value between devices, analyze the energy transfer efficiency of the device group, and when the energy transfer efficiency is lower than the preset transfer threshold, locate the fault source device in combination with the device fault mode library, identify the fault propagation path based on the fault source device, and generate a fault impact link diagram; The fourth unit is used to establish fault diagnosis evaluation indicators based on the fault impact link diagram and the dual-frequency domain fault feature library, perform hierarchical diagnosis on the equipment based on the fault diagnosis evaluation indicators, and evaluate the fault development trend in combination with the equipment's historical fault data, and output fault warning information and equipment health status assessment report; The fifth unit is used to generate a maintenance decision plan based on the fault warning information and the equipment health status assessment report, execute the maintenance decision plan, and record the maintenance effect data.

[0014] According to a third aspect of the embodiments of the present invention, An electronic device is provided, comprising: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0015] A fourth aspect of the embodiments of the present invention is: A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the aforementioned method is implemented.

[0016] In this embodiment, accurate fault diagnosis and intelligent maintenance of new energy station equipment can be achieved, and the stability and reliability of equipment operation can be improved. Through multi-source data fusion and standardized preprocessing, noise interference can be effectively reduced, data consistency can be ensured, and high-quality input can be provided for subsequent analysis. The extraction of transient features and drift features can comprehensively capture the short-term abnormalities and long-term degradation trends of equipment, thereby improving the accuracy of fault identification, and constructing a dual-frequency domain fault feature library to provide a basis for the analysis of complex fault modes. By calculating the power transfer fluctuation value between devices and analyzing the energy transfer efficiency, the energy flow anomaly between devices in the new energy station can be effectively identified, the fault source device can be accurately located, and a fault impact link diagram can be generated based on the fault propagation path, so as to quickly determine the impact range of the fault and avoid systemic failure due to single point failure. Based on the fault diagnosis evaluation index, the health status of the equipment can be graded, the fault development trend can be predicted in combination with historical data, and early warning information can be output in advance, so that the operation and maintenance personnel can take preventive measures before the fault occurs, and improve the safety and economy of the station operation. Intelligent maintenance decisions can be made based on fault warning information and health assessment reports, maintenance plans can be optimized, and unnecessary maintenance costs can be reduced. At the same time, maintenance effect data can be recorded to provide feedback for subsequent optimization, thereby achieving full life cycle health management of new energy station equipment, improving equipment utilization, reducing operation and maintenance costs, and improving the overall operating efficiency of new energy stations. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 It is a flow chart of a multi-source data fusion diagnosis method for new energy station equipment according to an embodiment of the present invention; Figure 2 This is a diagram showing the effect of fault detection delay time and correction factor impact analysis according to an embodiment of the present invention; Figure 3 A visualized network diagram of a fault propagation priority path according to an embodiment of the present invention; Figure 4 This is a health level threshold distribution curve diagram of an embodiment of the present invention; Figure 5 It is a structural schematic diagram of a multi-source data fusion diagnosis system for new energy station equipment according to an embodiment of the present invention. DETAILED DESCRIPTION

[0018] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0019] The technical solution of the present invention is described in detail with specific embodiments below. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0020] Figure 1 FIG. 1 is a flow chart of a multi-source data fusion diagnosis method for new energy station equipment according to an embodiment of the present invention. Figure 1 As shown, the method includes: Collect and pre-process the operating data of new energy station equipment to obtain standardized operating data; Extract transient features and drift features from standardized operation data, identify equipment transient fault modes based on transient features, identify equipment degradation fault modes based on drift features, and establish a dual-frequency domain fault feature library based on transient fault modes and degradation fault modes; Calculate the power transfer fluctuation value between devices and analyze the energy transfer efficiency of the device group. When the energy transfer efficiency is lower than the preset transfer threshold, locate the fault source device based on the device fault mode library, identify the fault propagation path based on the fault source device, and generate a fault impact link diagram. Establish fault diagnosis evaluation indicators based on the fault impact link diagram and dual-frequency domain fault feature library, perform graded diagnosis on the equipment based on the fault diagnosis evaluation indicators, and evaluate the fault development trend in combination with the equipment's historical fault data, output fault warning information and equipment health status assessment report; Generate maintenance decision plans based on fault warning information and equipment health status assessment reports, execute maintenance decision plans, and record maintenance effect data.

[0021] In an optional embodiment, Extract transient features and drift features from standardized operation data, identify equipment transient fault modes based on transient features, identify equipment degradation fault modes based on drift features, and establish a dual-frequency domain fault feature library based on transient fault modes and degradation fault modes, including: Calculating the data volatility of the standardized operation data, constructing an adaptive harmonic basis function group based on the data volatility, and decomposing the standardized operation data to obtain a transient component and a drift component; Extract mutation characteristic parameters of transient component and trend characteristic parameters of drift component respectively; A transient feature matrix is ​​constructed based on mutation feature parameters, the transient feature matrix is ​​matched with a preset transient fault template, a multi-dimensional cosine similarity is calculated to obtain a transient fault similarity matrix, and a device transient fault mode is determined according to a fault type corresponding to a maximum similarity in the transient fault similarity matrix; A drift feature matrix is ​​constructed based on trend feature parameters, the drift feature matrix is ​​matched with a preset degradation fault template, a multi-dimensional Euclidean distance is calculated to obtain a degradation fault similarity matrix, and a device degradation fault mode is determined according to a fault type corresponding to a minimum distance in the degradation fault similarity matrix; The instantaneous fault similarity matrix and the degradation fault similarity matrix are weightedly fused to obtain a comprehensive fault similarity matrix. The instantaneous fault mode and the degradation fault mode are credibility evaluated according to the comprehensive fault similarity matrix. The fault modes that pass the credibility evaluation and the corresponding feature matrices are combined to establish a dual-frequency domain fault feature library.

[0022] Exemplarily, the data volatility of the standardized operating data is calculated to measure the degree of change of the data over time. The data volatility indicates the stability of the operating state of the equipment and is usually used to analyze the short-term fluctuation amplitude and long-term stability of time series data. The process first smoothes the time series data collected by the equipment to remove high-frequency noise while ensuring that the fault characteristics in the original signal are not weakened. Then, the degree of fluctuation of the data is measured by calculating the range of change of the data within a specific time window. The size of the data volatility is used to judge the applicability of subsequent signal decomposition methods. For example, data with larger volatility usually contains stronger transient characteristics, while data with smaller volatility tends to show long-term trend changes of the equipment.

[0023] An adaptive harmonic basis function group is constructed based on the data volatility, and the basis function group is used to decompose the standardized operation data to obtain transient components and drift components. The adaptive harmonic basis function group is a set of functions that can dynamically adjust parameters. Its decomposition scale is adjusted according to the data volatility to adapt to different types of equipment operation data. The decomposition process of standardized operation data is to split the complex signal into multiple components with different frequency characteristics, among which the transient component mainly contains high-frequency signals, indicating short-term mutations during the operation of the equipment, such as current shocks, mechanical vibrations, etc., while the drift component mainly contains low-frequency signals, reflecting the trend changes in the long-term operation of the equipment, such as component aging, efficiency reduction, etc.

[0024] The mutation characteristic parameters of the transient component and the trend characteristic parameters of the drift component are extracted respectively. The mutation characteristic parameters are used to describe the sudden change characteristics in the equipment operation data, usually including the peak value, mutation time point, change rate, rise and fall time, etc. For example, in the current signal, if the current value at a certain moment suddenly increases and returns to the normal level in a short time, it can be considered that there is a transient fault at that moment. The drift characteristic parameters are used to characterize the slow change trend of the equipment during the long-term operation, usually including the average value change rate, signal drift direction, slope, long-term variance, etc. For example, if the temperature value of a certain device continues to rise over a period of time, it may indicate that there is poor heat dissipation or other potential degradation problems.

[0025] A transient feature matrix is ​​constructed based on mutation feature parameters, and the transient feature matrix is ​​matched with the preset transient fault template. The transient feature matrix is ​​a multidimensional array that stores mutation feature parameters at different time points, such as the mutation amplitude of the device's current, power, voltage and other signals and their corresponding timestamps. The preset transient fault template is a set of standard fault modes trained based on historical operation data, and each mode corresponds to a specific mutation feature. The matching process uses multidimensional cosine similarity calculation, that is, by calculating the angular similarity between the transient feature matrix and each transient fault template, the transient fault mode of the current device is determined.

[0026] A drift feature matrix is ​​constructed based on trend feature parameters, and the drift feature matrix is ​​matched with the preset degradation fault template. The drift feature matrix is ​​used to store the long-term trend characteristics of the equipment, such as mean shift, long-term variance, power factor change, etc. The preset degradation fault template contains the degradation modes that may occur during the long-term operation of the equipment, such as the degree of wear of certain key components, the aging trend of the line, etc. Multi-dimensional Euclidean distance calculation is used in the matching process, that is, the degree of degradation of the equipment is judged by calculating the distance between the drift feature matrix and each degradation fault template. The smaller the Euclidean distance, the closer the current equipment operation state is to a certain degradation mode, thereby determining the degradation fault mode that may occur in the equipment.

[0027] The transient fault similarity matrix and the degradation fault similarity matrix are weightedly fused to obtain a comprehensive fault similarity matrix, and the credibility of the transient fault mode and the degradation fault mode is evaluated based on the comprehensive fault similarity matrix. Weighted fusion is to adjust the importance of transient faults and degradation faults according to the operating status and historical fault conditions of the equipment. For example, for high-load operating equipment, transient characteristics may be more important, while for long-term operating equipment, drift characteristics have a greater impact. Credibility assessment is a reliability analysis of the final fault diagnosis results to ensure that the identified fault mode is consistent with the actual operating status of the equipment. Credibility assessment methods are usually trained based on historical data to improve diagnostic accuracy.

[0028] The fault modes evaluated by credibility and the corresponding feature matrix are combined to establish a dual-frequency domain fault feature library. The dual-frequency domain feature library is usually composed of two frequency domain features: one is the low-frequency domain feature, which is usually related to the overall state change of the equipment; the other is the high-frequency domain feature, which is usually related to the subtle changes of the internal faults of the equipment. The dual-frequency domain fault feature library is used to store the instantaneous fault mode and degradation fault mode of the equipment, and also includes the corresponding feature parameters, fault impact range, fault development trend and other information. The feature library can be continuously updated to adapt to changes in equipment operating conditions and improve the accuracy of fault diagnosis. The establishment of the feature library is conducive to the intelligent operation and maintenance of new energy sites, and provides data support for future equipment health management by continuously accumulating equipment failure data.

[0029] In this embodiment, accurate monitoring and intelligent diagnosis of the operating status of new energy station equipment can be achieved, and the accuracy and reliability of fault identification can be improved. Through the signal decomposition method of the adaptive harmonic basis function group, the transient characteristics and drift characteristics of the equipment operation data are effectively extracted, so that the system can quickly identify short-term abnormalities and long-term degradation trends. The matching analysis of the transient feature matrix and the preset transient fault template can efficiently detect sudden equipment failures, while the matching method of the drift feature matrix and the degradation fault template can accurately determine the long-term performance degradation of the equipment. The weighted fusion of the transient fault similarity matrix and the degradation fault similarity matrix improves the credibility of the diagnosis results and reduces the misjudgment rate, thereby achieving more reliable fault assessment. The construction of the dual-frequency domain fault feature library provides data support for equipment health assessment and predictive maintenance, so that the new energy station has long-term optimization and adaptive adjustment capabilities. Through the accurate identification and trend prediction of equipment faults, potential risks can be discovered in advance, sudden failures can be reduced, maintenance costs can be reduced, the service life of equipment can be extended, and the overall operation stability and new energy utilization efficiency of the station can be improved.

[0030] In an optional embodiment, The credibility evaluation of the transient fault mode and the degradation fault mode according to the comprehensive fault similarity matrix includes: Collecting the working condition state vector including load power, ambient wind speed, ambient temperature, vibration frequency characteristics and operation efficiency index, and performing maximum and minimum value normalization processing on the working condition state vector to obtain normalized working condition characteristics; Performing cluster analysis based on the normalized working condition characteristics to obtain an initial working condition boundary value, calculating the working condition distribution of the fault mode in the comprehensive fault similarity matrix to obtain a boundary loss function, performing a weighted combination of the boundary loss function and the standard deviation of the normalized working condition characteristics to obtain a boundary update amount, and dynamically optimizing the initial working condition boundary value based on the boundary update amount to obtain an optimized working condition boundary value; Calculating the matching degree of the fault mode in the comprehensive fault similarity matrix under the current working condition to obtain the local fitness, performing exponential smoothing on the local fitness within a specified time window to obtain the time series fitness, and adaptively weighting the local fitness and the time series fitness to obtain the comprehensive working condition fitness; The comprehensive fault similarity matrix and the comprehensive operating condition fitness are weightedly mapped to obtain a confidence reference value, the confidence reference value is multiplied by the constraint function of the optimized operating condition boundary value to obtain a modified confidence, the time-varying correction factor is calculated according to the time series standard deviation of the comprehensive operating condition fitness and the sliding window mean, and the modified confidence and the time-varying correction factor are weightedly fused according to the dynamic weight coefficient to obtain the credibility evaluation results of the transient fault mode and the degradation fault mode.

[0031] For example, the collection of the operating state vector involves a variety of key characteristic data, including load power, ambient wind speed, ambient temperature, vibration frequency characteristics and operating efficiency indicators. These data come from equipment sensors, external environment monitoring systems and operating status recording systems. To ensure the accuracy of the data, data cleaning, synchronization and preprocessing are required.

[0032] First, the load power data is provided by the power monitoring module inside the device. The acquisition frequency is set according to the characteristics of the device. For example, high-speed devices can use millisecond sampling, and low-speed devices can use second sampling. Since short-term power fluctuations may occur during the operation of the device, directly using the original data may lead to unstable analysis results. Therefore, a sliding window smoothing method is required to smooth the power data to reduce high-frequency noise interference.

[0033] Ambient wind speed and ambient temperature data are collected by external sensors. Since wind speed and temperature are greatly affected by external conditions, sudden changes or abnormal values ​​may occur. Therefore, time synchronization technology is required during the collection process to ensure that the data is consistent with the operating status of the equipment, and statistical methods are used for abnormal detection. For example, based on the wind speed or temperature change trend over a period of time, the threshold setting method can be used to screen out sudden changes in data to avoid interference with equipment status assessment caused by sudden environmental changes.

[0034] The vibration frequency characteristic data is provided by an accelerometer or a laser vibrometer, and usually requires time domain or frequency domain conversion to extract key characteristic values. Specifically, the root mean square value, peak value, kurtosis and other characteristics can be extracted based on time domain analysis, or the frequency spectrum characteristics of the vibration signal can be analyzed using fast Fourier transform to extract information such as the main frequency component and harmonic distribution. In order to reduce the interference of environmental noise, bandpass filtering technology can be combined in the signal processing process to remove interference signals in irrelevant frequency bands to improve the representativeness of the characteristics.

[0035] The calculation of operating efficiency index is based on the ratio of input power to output power of the equipment, or other parameters that measure the health status of the equipment, such as thermal efficiency, conversion efficiency, etc. Since efficiency calculation depends on multiple variable inputs, outliers need to be eliminated to prevent a single abnormal data from affecting the overall trend analysis. For example, the historical mean method can be used to calculate the efficiency range within a certain period of time, and the abnormal points that exceed the set deviation range can be eliminated.

[0036] After completing data collection, the data needs to be normalized to eliminate the impact between different feature dimensions and improve the comparability of the data. The normalization method uses maximum and minimum normalization, that is, the maximum and minimum values ​​of each feature are calculated and mapped to a unified interval using linear transformation. The normalized data is the normalized condition feature, which ensures that different features have the same numerical range, thereby avoiding the uneven impact of individual features on subsequent analysis due to excessively large or small values.

[0037] Based on the normalized working condition features, cluster analysis can be performed to determine the initial working condition boundary values. The purpose of cluster analysis is to cluster similar working conditions together in order to classify different operating condition categories. Specifically, a density-based method can be used to divide the data set into several working condition categories by calculating the similarity between features, and each category corresponds to an independent working condition. The similarity calculation can be based on Euclidean distance or cosine similarity to measure the relative proximity of different working conditions. After clustering is completed, the feature range of each category can be obtained through statistical analysis and used as the initial working condition boundary value.

[0038] In order to further optimize the operating condition boundary value, it is necessary to calculate the operating condition distribution of each fault mode in the comprehensive fault similarity matrix. The comprehensive fault similarity matrix is ​​used to store the matching degree between different fault modes and each operating condition, and its calculation is based on the correspondence between historical fault data and operating condition characteristics. By statistically analyzing the fault distribution under different operating conditions, the probability distribution of each fault mode in each operating condition category can be obtained. If a certain fault mode is widely distributed in a certain operating condition category or there is an obvious abnormality, it means that the initial operating condition boundary value may need to be adjusted.

[0039] Based on the operating condition distribution of the failure mode, the boundary loss function can be calculated. The boundary loss function is used to measure the deviation between the initial operating condition boundary value and the failure mode distribution. Its essence is to measure whether the boundary is reasonable. If the boundary of a certain operating condition category causes the failure mode distribution to be too concentrated or mutated, the boundary may need to be adjusted. The calculation of the boundary loss function can be combined with statistical analysis and empirical data. By comparing the failure distribution under different boundary setting methods, the optimal boundary adjustment scheme can be selected.

[0040] The implementation of boundary optimization relies on the weighted combination of the boundary loss function and the standard deviation of the normalized working condition characteristics to calculate the boundary update amount. The standard deviation of the normalized working condition characteristics indicates the degree of characteristic change within the working condition category. If the standard deviation of a working condition category is too large, it indicates that its internal characteristics change more drastically and the boundary may need to be adjusted. By weighted combination of boundary loss function and standard deviation, the boundary update amount can be obtained, and the initial working condition boundary value is adjusted based on the update amount to optimize the rationality of boundary division.

[0041] Based on the optimized working condition boundary values, the matching degree of each fault mode in the comprehensive fault similarity matrix under the current working condition can be calculated to obtain the local fitness. The local fitness is used to measure the degree of conformity of a certain fault mode under the current working condition. Its calculation is based on the similarity matching between the fault mode and the working condition characteristics. In order to reduce the impact of short-term fluctuations on the matching results, the local fitness can be exponentially smoothed within the specified time window to obtain the time series fitness. The exponential smoothing method gives different weights to historical data, making the latest data have a greater impact on the results, thereby reducing the interference of short-term anomalies on the matching results.

[0042] The local fitness and the time series fitness are adaptively weighted and fused to obtain the comprehensive working condition fitness. The core of adaptive weighted fusion is to dynamically adjust the weights according to the changes in the data to ensure that the fused results can accurately reflect the current working condition status. Specifically, a dynamic weight adjustment strategy can be set according to the changing trends of the local fitness and the time series fitness, so that the final working condition fitness calculation results are more robust.

[0043] After obtaining the comprehensive working condition fitness, it can be weighted mapped with the comprehensive fault similarity matrix to obtain the confidence reference value. The confidence reference value is used to represent the credibility of different fault modes under the current working condition, and its calculation is based on the joint influence of fault similarity and working condition fitness. Furthermore, the confidence reference value can be multiplied by the constraint function of the optimized working condition boundary value to obtain the modified confidence. The constraint function is used to adjust the confidence calculation rules under different working condition categories to ensure that the final confidence evaluation conforms to the actual situation of the working condition state.

[0044] The time-varying correction factor can be calculated based on the time series standard deviation and sliding window mean of the comprehensive working condition fitness. The role of the time-varying correction factor is to dynamically adjust the confidence calculation to adapt to the changes in the working condition. Finally, the corrected confidence and the time-varying correction factor are weighted and fused according to the dynamic weight coefficient to obtain the credibility assessment results of the transient fault mode and the degradation fault mode. The transient fault mode assessment is used to detect sudden faults in a short period of time, while the degradation fault mode assessment is used to identify long-term trend faults, thereby providing accurate data support for the health status diagnosis of the equipment.

[0045] Figure 2 This is a diagram showing the effect of fault detection delay time and correction factor on the analysis of the embodiment of the present invention. Figure 2 As shown in the figure, the influence of the correction factor value on the fault detection delay time is analyzed, and the performance difference between the present technical solution and the fixed correction factor method and the method without correction factor is compared. The horizontal axis represents the correction factor value (0.1-0.9), and the vertical axis represents the fault detection delay time (seconds). It can be seen that the detection delay time of the present technical solution shows a "U"-shaped trend of first decreasing and then increasing with the change of the correction factor value, and reaches the optimal value of 1.3 seconds when the correction factor value is 0.5 (circled in the figure). In contrast, the detection delay time of the fixed correction factor method increases linearly with the increase of the correction factor value, from 4.2 seconds (correction factor 0.1) to 8.0 seconds (correction factor 0.9). The method without correction factor is not affected by the change of the correction factor value and always maintains a high delay level of 9.1 seconds. This result shows that the present technical solution uses the time-varying correction factor calculated by the time series standard deviation of the comprehensive working condition fitness and the sliding window mean to enable the system to adaptively adjust the detection sensitivity according to the dynamic changes of the working condition state, effectively reducing the detection delay while ensuring the detection accuracy, and improving the real-time response capability of the system.

[0046] In this embodiment, the credibility of the instantaneous fault mode and the degradation fault mode is evaluated by the comprehensive fault similarity matrix, which can significantly improve the accuracy and stability of the equipment fault diagnosis. First, a variety of key feature data, such as load power, ambient wind speed, temperature, vibration frequency and operating efficiency indicators, are collected to provide comprehensive working condition status information. The existing technology often relies on only a small amount of sensor data, which may not be able to fully reflect the working state of the equipment. The present technical solution improves the comprehensiveness and accuracy of the diagnosis by integrating multi-dimensional data. The maximum and minimum value normalization processing is adopted to solve the problem of dimensional differences between different features and enhance the comparability of data processing. In the prior art, when the feature data is not normalized, it is easy to have an unbalanced impact on the analysis results due to a large or small feature value. This solution avoids this problem through normalization and improves the stability of data analysis. In terms of working condition boundary optimization, through the introduction of cluster analysis and boundary loss function, the initial working condition boundary can be accurately determined and dynamically adjusted. Existing technologies mostly rely on fixed boundaries or make adjustments based on simple statistical methods, which easily ignores subtle changes in operating condition characteristics. However, this solution dynamically optimizes the operating condition boundary through a weighted combination of boundary loss function and standard deviation of operating condition characteristics, and can adapt to complex and changeable working environments, thereby improving the accuracy and responsiveness of fault mode recognition. For the evaluation of fault modes, this technical solution utilizes a combination of local fitness and time series fitness, and overcomes the impact of short-term fluctuations on diagnostic results through exponential smoothing and adaptive weighted fusion. Existing technologies usually rely only on static models or simple time window methods, which may lead to slow response to sudden faults or unclear identification of degenerative faults. However, this solution improves the adaptability of fault modes under different working conditions through dynamic adjustment of time series fitness, and can reflect changes in the health status of equipment in real time. A calculation method for time-varying correction factors is introduced, and the correction confidence is dynamically adjusted based on the time series standard deviation and sliding window mean of the comprehensive working condition fitness. The existing technology often fails to fully consider the time-varying factors, and the calculation of the corrected confidence is relatively fixed. However, this technical solution introduces a dynamic correction factor to make the confidence evaluation result more consistent with the actual working condition of the equipment.

[0047] Through these technical improvements, this solution can provide a more accurate, real-time and stable credibility assessment for the equipment's failure mode, significantly improve the reliability and accuracy of equipment health status diagnosis, meet the needs of efficient prevention and timely maintenance, and is especially suitable for equipment monitoring and fault diagnosis under complex working conditions.

[0048] In an optional embodiment, Calculate the power transfer fluctuation value between devices and analyze the energy transfer efficiency of the device group. When the energy transfer efficiency is lower than the preset transfer threshold, locate the fault source device in combination with the device fault mode library, identify the fault propagation path based on the fault source device, and generate the fault impact link diagram including: Based on the device output power and the adjacent device input power within the sliding time window, the power difference squared at each sampling moment is multiplied by the corresponding time weight coefficient and then accumulated to obtain the power transfer fluctuation value between devices; The physical connection relationship between the collected devices is used to construct a topological connection matrix, a device transfer fluctuation matrix is ​​constructed based on the power transfer fluctuation value and the topological connection matrix, and the energy transfer efficiency of the device group is obtained according to the ratio of the device transfer fluctuation matrix to a preset reference fluctuation matrix; When the energy transfer efficiency of the device group is lower than the preset transfer threshold, the fault feature vector of each device is calculated based on the device transfer fluctuation matrix, and the fault feature vector is matched with the feature vector in the dual-frequency domain fault feature library for similarity to determine the fault source device; Taking the fault source device as the starting point, a fault propagation influence matrix is ​​constructed based on the device transfer fluctuation matrix and the topological connection matrix, wherein the propagation influence value of a directly connected device pair is its power transfer fluctuation value, and the propagation influence value of an indirectly connected device pair is the multiplication of the power transfer fluctuation value on the path divided by the power of the shortest path length; The priority path of fault propagation is determined based on the size of each element in the fault propagation influence matrix, the fault propagation probability of the priority path is calculated in combination with historical fault data, and a fault influence link diagram including the fault propagation path and propagation probability is generated.

[0049] Exemplarily, first, collecting power data between devices is the starting point of the implementation process. The output power of the device and the input power of the adjacent device must be collected at predetermined time intervals. In order to eliminate short-term interference caused by equipment power fluctuations, it is necessary to calculate the difference between the output power and the input power of the adjacent device at each time point. The square of the power difference will be multiplied by a time weight coefficient, which is used to express the different degrees of overall impact of the power difference in different time periods. The length of the time window selection can be determined according to the working cycle of the equipment. Short-term fluctuations are suitable for smaller window lengths, while longer trends are suitable for larger time windows. The calculated power difference and the time-weighted result need to be accumulated hour by hour to finally obtain the power transfer fluctuation value between devices in each time period.

[0050] Next, we need to construct a topological connection matrix between devices. This matrix describes the physical connection relationship between devices. Each row and column represents a device, and the elements of the matrix indicate whether there is a direct connection between devices. If device A is connected to device B, the value of the corresponding position in the matrix is ​​1, otherwise it is 0. By analyzing these connection relationships, we can obtain the structural information of the device group, which helps with subsequent fault propagation analysis.

[0051] After constructing the topology connection matrix, it is necessary to use the power transfer fluctuation values ​​between devices and the topology connection matrix to create the device transfer fluctuation matrix. The device transfer fluctuation matrix reflects the power fluctuation experienced by each device in the entire device group. This can be done by applying the power difference value to the topology connection matrix to evaluate the transfer of fluctuations between devices. Each row represents the output fluctuation of a device, and each column represents the fluctuations transferred by the device between other devices.

[0052] In order to analyze the energy transfer efficiency of the device group, it is also necessary to compare the device transfer fluctuation matrix with the preset benchmark fluctuation matrix. The benchmark fluctuation matrix is ​​usually constructed based on historical data or expert experience, and represents the energy transfer fluctuation under ideal conditions. By comparing the two, the energy transfer efficiency of the device group can be obtained. If the transfer efficiency is lower than the preset threshold, it means that there may be a fault in the device group.

[0053] When the energy transfer efficiency of the device group is lower than the threshold, the next step is to locate the fault source device. This step requires calculating the fault feature vector of each device first. The fault feature vector can be calculated by collecting the vibration data, power fluctuation data and other operating parameters of the device, using specific algorithms such as principal component analysis (PCA) or weighted average method. After obtaining these feature vectors, the device closest to the feature vector can be determined as the fault source device by similarity matching with the dual-frequency domain features in the fault feature library. The dual-frequency domain fault feature library contains features of various known fault modes, and the matching algorithm can identify the most consistent fault source device by calculating the distance or similarity between feature vectors.

[0054] Once the fault source device is located, the next step is to construct the fault propagation impact matrix. This step relies on the power transfer fluctuation matrix and the topological connection matrix between devices. First, the impact value of the directly connected device pair in the propagation impact matrix is ​​equal to the power transfer fluctuation value between them. For the indirectly connected device pair, the propagation impact value is calculated by multiplying the power transfer fluctuation value on the path divided by the power of the shortest path length. This method takes into account multiple transmission paths between devices, and the impact of propagation gradually weakens as the number of paths increases.

[0055] By constructing a fault propagation impact matrix, the system can determine the priority path for fault propagation based on the size of the impact value. Priority paths refer to those paths with larger propagation impact values, which will be affected by the fault first. After determining the priority path, combined with historical fault data, a fault propagation probability can be assigned to each path. This probability is based on the frequency or possibility of fault propagation on this path under similar fault conditions in history.

[0056] Finally, the propagation path is combined with the fault propagation probability to generate a fault impact link diagram. This link diagram shows the propagation path from the fault source device to other devices, and is accompanied by the fault propagation probability of each path. Through such a link diagram, the propagation process of the fault in the device group can be clearly displayed, which helps the subsequent fault detection and maintenance decision-making.

[0057] Assume that there is a device group consisting of four devices, device A, device B, device C and device D, which form a topological structure through physical connection. The output power of device A is 100W, the input power of device B is 95W, the input power of device C is 90W, and the input power of device D is 85W. Based on this set of power data, the power difference between each pair of adjacent devices is first calculated, and a time weight coefficient is assigned to each difference. Then, the weighted power difference in all time periods is accumulated to obtain the power transfer fluctuation value. By analyzing these power fluctuation data and the topological relationship between the devices, the device transfer fluctuation matrix is ​​constructed. If device A and device B are directly connected, their fluctuation transfer influence matrix values ​​will be calculated based on the power difference. If device C and device D are indirectly connected through device B, the propagation influence between them will be calculated based on the power fluctuation multiplication on the path. Subsequently, the energy transfer efficiency of the device group is calculated. The obtained device transfer fluctuation matrix is ​​compared with the preset benchmark fluctuation matrix, and it is found that the transfer efficiency is lower than the preset threshold, so it is necessary to locate the fault source device. By calculating the fault feature vector of each device and matching it with the feature vector in the dual-frequency domain fault feature library, device A is determined to be the fault source device. Once the fault source device is located, the fault propagation impact matrix is ​​constructed. The fault of device A will directly affect device B, while the fault propagation of devices C and D will pass through device B. Based on the power fluctuation data, the propagation impact value is obtained and the priority path of propagation is calculated. Combined with historical data, it is determined that the probability of the fault propagating from device A to device C is higher. Finally, a fault impact link diagram is generated, showing the path of the fault propagation from device A to devices C and D, with the fault propagation probability of each path. This link diagram can help operators quickly identify the fault propagation process and provide decision support for fault repair.

[0058] In this embodiment, by analyzing the energy transfer fluctuations of the device group and identifying the fault propagation path, the accuracy and timeliness of device fault detection can be effectively improved. By calculating the power transfer fluctuation value and energy transfer efficiency between devices, the energy flow between devices can be monitored in real time, abnormal fluctuations in the system can be quickly discovered, and potential fault sources can be identified. When the energy transfer efficiency is lower than the preset threshold, the fault source is located based on the fault feature library and the device transfer fluctuation matrix, and the fault propagation path is accurately identified. This solution constructs a fault propagation impact matrix, calculates the fault propagation probability in combination with historical fault data, and generates a detailed fault impact link diagram to help technicians determine the impact range of the fault and the priority repair path. Compared with traditional fault detection technology, this solution makes fault detection and positioning more accurate and real-time by dynamically optimizing the working condition boundary, introducing timing fitness and time-varying correction factors, which can greatly improve the operation and maintenance efficiency and fault response speed of the device group and reduce the loss of system failures.

[0059] In an optional embodiment, Determining a priority path for fault propagation based on the size of each element in the fault propagation influence matrix, and calculating the fault propagation probability of the priority path in combination with historical fault data includes: Constructing a propagation path candidate set based on the fault propagation impact matrix, wherein each propagation path includes a sequence of propagation impact values ​​of adjacent device pairs; Performing wavelet decomposition on the propagation impact value sequence to obtain impact value components of different frequency bands, taking the components whose impact value components are less than a preset component threshold as the propagation baseline sequence, and taking the components whose impact value components are greater than the preset component threshold as the fluctuation sequence; The path stability is obtained by dividing the mean of the propagation baseline sequence by the standard deviation of the fluctuation sequence, and the overlap degree between the peak-to-valley distribution characteristics of the fluctuation sequence and the historical fault time is used as the path credibility. The propagation path candidate set is screened according to the path stability; For the selected propagation paths, the sample entropy of the propagation impact value sequence is calculated, the size of the sample entropy is used as the path complexity evaluation index, and the propagation path with the smallest path complexity is selected as the fault propagation priority path; Based on historical fault data, the fault occurrence time sequence correlation of adjacent devices on the propagation priority path is calculated, the fault occurrence time sequence correlation is constructed as a priori probability distribution in the form of a Gaussian distribution, the current propagation impact value sequence is subjected to wavelet transform, the wavelet coefficients at different scales are extracted, the amplitude and phase characteristics of the wavelet coefficients are combined into an observation vector, and the prediction probability at the current moment is calculated based on the posterior probability at the previous moment; A likelihood function is calculated according to the observation vector, the predicted probability is multiplied by the likelihood function and normalized to obtain the posterior probability at the current moment, iteration is repeated until the posterior probability converges, and the converged posterior probability is used as the fault propagation probability of the fault propagation priority path.

[0060] Exemplarily, by collecting the operation data of the equipment, especially the power transfer data between adjacent equipment, the energy transfer fluctuation value between the equipment can be established. These data are usually derived from the input power and output power of the equipment. The power transfer fluctuation value is obtained by squaring the power difference at each sampling moment and multiplying it by the corresponding time weight coefficient and then accumulating it. This step is to quantify the fluctuation of energy transfer between equipment, with the purpose of providing basic data for subsequent fault propagation path analysis.

[0061] Once the power transfer fluctuation value is calculated, the next step is to construct a topological connection matrix based on the physical connection relationship between devices. The topological connection matrix is ​​a matrix that represents the connection relationship between devices, where each matrix element indicates whether there is a physical connection between two devices and the transmission impact between them. This matrix provides structural data for subsequent propagation path analysis, allowing the device transfer fluctuation matrix to be further constructed based on the power transfer fluctuation value between devices. The device transfer fluctuation matrix is ​​derived from the combination of the power transfer fluctuation value and the topological connection matrix, and is used to describe the energy transfer status within the device group.

[0062] After constructing the device transfer fluctuation matrix, the next step is to calculate the energy transfer efficiency of the device group. This calculation is obtained by comparing the device transfer fluctuation matrix with the preset reference fluctuation matrix. The preset reference fluctuation matrix represents the energy transfer state of the device group under ideal or normal conditions, and the energy transfer efficiency of the device group is determined by comparing the difference between the two. When the energy transfer efficiency of the device group is lower than the preset threshold, it means that there is a potential fault or energy loss in the device group, and it is necessary to locate the fault source device.

[0063] The fault source device is located by calculating the fault feature vector. The fault feature vector of each device is calculated based on the power transfer fluctuation value of the device. The fault feature vector can be regarded as the typical performance characteristics of the device under fault conditions. These fault feature vectors are matched with the known feature vectors in the fault mode library for similarity, and the closest device is obtained by calculation to determine the fault source device.

[0064] After the fault source device is determined, the next step is to identify the fault propagation path. By combining the device transfer fluctuation matrix and the topological connection matrix, a fault propagation impact matrix can be constructed. The fault propagation impact matrix is ​​used to quantify the degree of influence from the fault source device to other devices. The propagation impact value of a pair of directly connected devices is equal to the power transfer fluctuation value between them, while the propagation impact value of an indirectly connected pair of devices is obtained by calculating the multiplication of their power transfer fluctuation values ​​and dividing it by the power of the shortest path length. This can more accurately reflect the influence intensity between different devices during the fault propagation process.

[0065] Once the fault propagation impact matrix is ​​established, the priority path for fault propagation can be determined based on the size of each propagation impact value. The priority path refers to the path that is most susceptible to the fault in the device group. In order to screen out the most influential propagation path, it is necessary to perform wavelet decomposition on the propagation impact value sequence. Wavelet decomposition is a signal processing method that can decompose the time domain signal into multiple frequency bands. In this way, components of different frequencies can be extracted from the propagation impact value sequence. The part of the impact value component that is less than the preset component threshold is regarded as the propagation baseline sequence, while the part greater than the threshold is the fluctuation sequence.

[0066] The propagation baseline sequence represents the normal energy transfer pattern in the device group, while the fluctuation sequence may contain potential signs of failure or instability. The path stability can be obtained by dividing the mean of the propagation baseline sequence by the standard deviation of the fluctuation sequence. Path stability reflects the stability of the propagation path, and the higher the value, the more stable the path. Next, evaluate the degree of overlap between the peak and valley distribution characteristics of the fluctuation sequence and the historical failure time to determine the credibility of the path. The higher the degree of overlap of the historical failure time, the higher the probability of failure of the path in the future, so it needs to be paid special attention.

[0067] For the selected propagation paths, its complexity needs to be further calculated. The sample entropy of the propagation impact value sequence is calculated. Sample entropy is an indicator to measure the complexity of the time series. The smaller the value, the lower the complexity of the sequence and the more predictable the behavior of the system. The propagation path with the smallest sample entropy is selected as the priority path for fault propagation.

[0068] Finally, the fault occurrence time sequence correlation of adjacent devices on the propagation priority path is calculated based on historical fault data. The fault occurrence time sequence correlation reflects the time correlation of fault occurrence between devices. A higher correlation indicates that these devices are more likely to fail at the same time. Based on these time sequence correlations, a priori probability distribution in the form of a Gaussian distribution is constructed to predict the probability of fault propagation. Subsequently, the current propagation impact value sequence is subjected to wavelet transform, wavelet coefficients at different scales are extracted, and the fault propagation probability at the current moment is predicted based on the posterior probability at the previous moment. The Bayesian inference method calculates the relationship between the current observation data and the prior probability, continuously updates the posterior probability until convergence, and finally obtains the fault propagation probability of the fault propagation priority path.

[0069] Figure 3 This is a visualized network diagram of the fault propagation priority path according to an embodiment of the present invention, such as Figure 3 As shown in the figure, the network diagram intuitively shows the fault propagation path in a device group consisting of 9 device nodes. Device A (center top) is marked as the fault source, and the other 8 devices (B to I) are distributed around it. The thick solid line in the figure represents the priority path of fault propagation, the black dotted line represents the secondary propagation path, and the value on the connecting line represents the propagation impact value. It can be clearly seen from the figure that the priority path of fault propagation forms a closed loop: from device A→device B→device H→device I→device C→device A, forming a complete fault propagation cycle. The propagation impact values ​​on this priority path are all above 0.78, among which the propagation impact value from device A to device B is the highest, reaching 0.92, the propagation impact value from device B to device H is 0.86, the propagation impact value from device H to device I is 0.78, the propagation impact value from device I to device C is 0.83, and the propagation impact value from device C to device A is 0.89. At the same time, the fault propagation probability is marked on each device node, among which device B has the highest propagation probability (0.86), followed by device C (0.76) and device D (0.72). The propagation impact values ​​on the secondary propagation paths are generally lower than those on the priority paths, and most are between 0.6-0.75. The network diagram intuitively shows how the fault propagates from the source device to other devices, as well as the relative importance and propagation possibility of each propagation path. This visualization form is of great value for equipment operation and maintenance personnel to understand the fault propagation mechanism, predict the fault propagation path, and formulate prevention and response strategies. It can assist operation and maintenance personnel to prioritize monitoring and maintenance of key equipment on the priority path, thereby more effectively preventing fault propagation and chain reactions.

[0070] In this embodiment, efficient fault propagation path identification and prediction of fault propagation probability are achieved through comprehensive analysis of equipment power transfer fluctuations, topological connection relationships and equipment fault characteristics. Traditional technologies mainly rely on simple fault pattern recognition and static equipment monitoring methods, which usually have lag and poor accuracy when dealing with fault propagation in complex systems. Especially when facing the collaborative work of multiple devices, it is difficult to accurately determine the fault source and its propagation path. This solution constructs a power transfer fluctuation matrix and a topological connection matrix, and combines a variety of advanced algorithms such as wavelet decomposition, sample entropy and Bayesian inference to dynamically analyze the energy transfer efficiency and fault propagation path of the equipment group in real time, providing more accurate fault source location and propagation path prediction. Compared with the prior art, this solution introduces wavelet decomposition and sample entropy analysis in the process of fault propagation path identification, thereby improving the sensitivity and accuracy of complex fault signals, and at the same time calculating the probability of fault propagation through Bayesian inference, optimizing the prediction accuracy of fault occurrence. In addition, by analyzing the temporal correlation of fault propagation using the dynamically updated transfer fluctuation matrix and historical data, the fault propagation trend can be identified in a timely manner, effectively reducing the impact of faults on the overall operating efficiency of the equipment group. Through these improvements, this technology can achieve more efficient and accurate fault diagnosis in multiple devices and complex environments, significantly improving the decision-making support capabilities for equipment maintenance.

[0071] In an optional embodiment, Establish fault diagnosis evaluation indicators based on the fault impact link diagram and dual-frequency domain fault feature library, perform graded diagnosis on the equipment based on the fault diagnosis evaluation indicators, and evaluate the fault development trend in combination with the equipment's historical fault data, output fault warning information and equipment health status evaluation reports including: Extracting common nodes of intersecting links from the fault-affected link graph, calculating the state coordination degree of the common nodes to construct a link coupling matrix, screening features in a dual-frequency domain fault feature library according to the link coupling matrix, calculating the matching degree between the features and the link coupling matrix, and selecting features with high matching degree to construct a candidate feature set; Extract the change sequence of the candidate feature set in the historical fault data, analyze the causal relationship of the change sequence to construct a feature propagation network, identify the strongest causal path from the feature propagation network as a feature diagnosis chain, calculate the propagation strength of each node in the feature diagnosis chain, take the product of the propagation strength and the feature matching degree as the feature weight, and perform weighted combination of the candidate feature set according to the feature weight to obtain the fault diagnosis evaluation index; The probability distribution of fault diagnosis evaluation indicators in historical fault data is statistically analyzed, and the health level division threshold is determined according to the probability distribution. The interval to which the current fault diagnosis evaluation indicator belongs is determined as the equipment health level. The fault development trend is predicted according to the state transition law of the characteristic diagnosis chain, and the fault warning information and health status assessment report are output in combination with the equipment health level.

[0072] Exemplarily, first, common nodes of intersecting links are extracted from the fault impact link graph. The fault impact link graph is a graphical representation of device fault propagation, where each node represents a device or fault mode, and each edge represents the propagation path of the fault between devices. When there are multiple links intersecting, these intersecting nodes represent the intersection points of multiple devices or fault paths, which are usually key nodes for fault propagation. Common nodes refer to the places where multiple links intersect, representing the interaction of faults between devices. Calculating the state coordination of these nodes is the first step in analyzing the interaction of faults between devices. The state coordination reflects the commonality between intersecting nodes under fault conditions, that is, the synchronization of faults of different devices in time and space. If multiple devices have similar faults at the same time, the state coordination of these devices will be higher. Using this information, a link coupling matrix can be constructed to indicate the degree of mutual influence of faults between devices. The link coupling matrix is ​​essentially an association matrix, and each element in the matrix represents the degree of coupling between two devices under fault modes. The higher the coupling degree, the closer the fault propagation of the two devices.

[0073] The fault features in the dual-frequency domain fault feature library are screened according to the link coupling matrix. In order to match the structure of the link coupling matrix, it is necessary to screen out the features most relevant to the fault propagation path by calculating the matching degree between each fault feature and the relevant devices in the link coupling matrix. These features with high matching degree can effectively reflect the mechanism of fault propagation between devices and provide a basis for subsequent fault diagnosis. The candidate feature set obtained through screening will be used as input data for further analysis.

[0074] Next, the change sequence of the candidate feature set in the historical fault data is extracted. Each fault event in the historical fault data has a timestamp, which records the change of the fault state of the equipment over time. By analyzing this data, the change trend of the fault characteristics can be obtained, and the correlation between the characteristics can be identified through causal analysis. The purpose of causal analysis is to reveal which fault characteristics will cause changes in other characteristics and establish a propagation relationship between characteristics. This process can be achieved by constructing a feature propagation network, in which each node represents a fault feature and each edge represents the path of propagation from one fault feature to another. By analyzing the feature propagation network, the strongest causal paths, that is, those fault feature propagation paths with the greatest influence, can be identified. These strong causal paths will be used as feature diagnosis chains.

[0075] Calculate the propagation strength of each node in the feature diagnosis chain. The propagation strength indicates the importance of a feature in the feature propagation chain to the entire fault propagation process. Nodes with greater propagation strength usually play a key role in fault propagation. In order to quantify the strength of feature propagation, it can be determined based on the propagation order, path length, and impact range between features. Nodes with greater propagation strength will be given higher weights in the subsequent fault diagnosis process.

[0076] The product of propagation intensity and feature matching degree is used as the feature weight. This means that the importance of a feature in fault diagnosis evaluation is not only related to its own matching degree, but also to its position and role in the fault propagation chain. Features with larger feature weights will occupy a larger proportion in the comprehensive fault diagnosis evaluation index. According to these weights, the candidate feature sets are weighted and combined to obtain the final fault diagnosis evaluation index. This evaluation index will reflect the fault status of the equipment and is the key basis for judging the health status of the equipment.

[0077] After analyzing the fault diagnosis evaluation indicators, the probability distribution of the diagnostic indicators in the historical fault data is statistically analyzed. The purpose of this step is to establish a probability distribution model of equipment fault indicators based on past fault data. Through statistical data, the threshold for dividing the equipment health level can be determined. This threshold will help determine the current health status of the equipment. If the current fault diagnosis evaluation indicator falls into a certain health level range, it means that the equipment is in a health state of that level.

[0078] Combined with the state transition law of the characteristic diagnosis chain, the trend of fault occurrence can be predicted. The state transition law of the characteristic diagnosis chain refers to the evolution pattern of the equipment fault characteristics over time. For example, some fault modes may gradually worsen over time, while some faults may recover by themselves. By analyzing these laws, the trend of future equipment fault development can be predicted.

[0079] According to the predicted fault development trend and the current equipment health level, fault warning information and health status assessment reports can be output. Fault warning information can promptly notify maintenance personnel to conduct equipment inspections, while health status assessment reports provide a basis for equipment maintenance decisions. By combining historical data and current diagnostic evaluations, this solution can provide more accurate information support for equipment management and fault prediction.

[0080] In this embodiment, by combining the fault impact link diagram with the dual-frequency domain fault feature library, the fault propagation path between devices can be more comprehensively understood, thereby accurately identifying key fault features. This solution not only improves the accuracy of fault diagnosis, but also accurately evaluates the trend of fault development by establishing a feature propagation network and analyzing cause-effect relationships. Compared with traditional methods, existing technologies often rely on the fault data of a single device for judgment, lacking comprehensive consideration of the impact of faults between devices, resulting in low diagnostic accuracy and efficiency. However, this technology can effectively capture the correlation between devices and improve the timeliness and accuracy of fault prediction through multi-level data analysis such as link coupling matrix and fault propagation impact matrix. Ultimately, the solution can identify potential risks in advance, accurately evaluate the health status of equipment, provide forward-looking decision support for equipment management, and greatly reduce the probability of unexpected failures.

[0081] In an optional embodiment, Statistically analyzing the probability distribution of fault diagnosis evaluation indicators in historical fault data, and determining the health level classification threshold according to the probability distribution includes: Calculate the time series fluctuation characteristics of the fault diagnosis evaluation index in the historical fault data to obtain a local stability index, dynamically adjust the resampling disturbance intensity according to the local stability index, and use the disturbance intensity to resample the fault diagnosis evaluation index to generate multiple groups of sample data; The inter-class distribution distance, intra-class distribution distance and time series continuity constraints of multiple groups of sample data are combined to construct a multi-objective optimization function, and multiple groups of initial thresholds are obtained through iterative optimization. The kernel density estimation method was used to analyze the distribution characteristics of multiple groups of initial thresholds, and the dynamic intervals of health grade division were constructed based on quantiles at different confidence levels. The distance distribution from the current fault diagnosis evaluation index to the dynamic interval is calculated, the distance distribution is converted into a probability distribution, the information entropy is calculated based on the probability distribution and converted into a credibility index, and the dynamic interval is optimized in combination with the credibility index to obtain the final health level division threshold.

[0082] Exemplarily, the time series fluctuation characteristics of the fault diagnosis evaluation index in the historical fault data need to be calculated. The purpose of this step is to extract the time series fluctuations in the historical data in order to analyze the stability of the current equipment state. By analyzing these time series data, a local stability index is obtained. The function of the local stability index is to measure the fluctuation of the data sequence within a certain time window, usually by calculating the change amplitude or variance of a certain indicator within a time period. The greater the fluctuation, the worse the stability of the equipment and the higher the probability of failure. The local stability index can reflect whether the equipment is in a relatively stable working state.

[0083] After obtaining the local stability index, the resampling disturbance intensity needs to be adjusted dynamically. Resampling disturbance refers to repeated sampling of historical data through certain disturbance methods to generate different sample data sets. These disturbance methods generally include random disturbance, normal disturbance, etc. The adjustment of disturbance intensity is based on the local stability index: if the device shows large volatility within a certain period of time, the disturbance intensity needs to be increased to generate more diverse sample data; conversely, the disturbance intensity is reduced to maintain the stability of the data. Through this method, it can ensure that the resampled data is more representative, thereby improving the accuracy of the health level classification.

[0084] The multi-objective optimization function is constructed using the multiple sets of sample data generated above. The multi-objective optimization function is used to balance the inter-class distribution distance, intra-class distribution distance, and temporal continuity constraints. In this step, the inter-class distribution distance refers to the distance measurement between different categories, the intra-class distribution distance refers to the data distribution within the same category, and the temporal continuity constraint ensures the consistency of the data in the time dimension. The purpose of the optimization function is to find a balance between the three to ensure the accuracy and rationality of the threshold division. The function will be continuously optimized through an iterative process according to the actual situation of the data, so as to obtain a preliminary health level division threshold.

[0085] The distribution of the initial threshold is analyzed using the kernel density estimation method. Kernel density estimation is a non-parametric estimation method that estimates the probability density function of the data by smoothing the data using a set of kernel functions. In this process, the kernel density estimation method can effectively extract distribution features from the disturbed sample data and reflect the distribution trend of the data in different intervals. By analyzing these distribution features, the dynamic interval of the health level division can be obtained. The kernel density estimation method does not rely on the specific distribution model of the data and can more accurately adapt to complex fault diagnosis and evaluation data.

[0086] After obtaining the dynamic interval, the next step is to calculate the distance distribution between the current fault diagnosis evaluation index and the dynamic interval. This operation is to find the position of the current data in the dynamic interval and convert it into a probability distribution. By calculating the distance of the current data, the distribution of the current fault diagnosis evaluation index in the interval can be obtained, thus providing a basis for the subsequent health level assessment.

[0087] Based on the distance distribution, information entropy is then calculated. Information entropy is a concept in information theory that is used to measure the uncertainty of data. The higher the entropy value of the data, the more uncertain the distribution of the data. In this step, information entropy is used to measure the uncertainty of the current data within the health level range. The lower the entropy value, the more certain the current fault diagnosis evaluation index is and the higher the credibility is. Through the entropy value, a credibility index can be generated to determine the credibility of the current data.

[0088] Finally, the dynamic interval is optimized according to the credibility index to obtain the final health level division threshold. During the optimization process, the changes in credibility will be taken into account and the health level division interval will be adjusted to ensure that the interval can more accurately reflect the health status of the device. The final health level division threshold will be adjusted according to the actual situation of the current data to adapt to the changes in the device status.

[0089] Figure 4 This is a health level threshold distribution curve diagram of an embodiment of the present invention, such as Figure 4 As shown in the figure, the health level threshold distribution curve obtained by the kernel density estimation method is shown. The horizontal axis represents the fault diagnosis evaluation index value (0-1.0), and the vertical axis represents the kernel density estimation value (0-7.0). The continuous curve in the figure shows the probability density distribution of the fault diagnosis evaluation index. The three vertical dotted lines mark the three health level division thresholds: T1 (0.235), T2 (0.500) and T3 (0.763), which divide the entire interval into four health status areas: healthy (0-0.235), slightly degraded (0.235-0.500), moderately degraded (0.500-0.763) and severely degraded (0.763-1.0). The four rectangular marks in the figure represent the 0.95 quantile of each area, and the corresponding credibility is marked: the credibility of the healthy area is 0.92, the slightly degraded area is 0.94, the moderately degraded area is 0.93, and the severely degraded area is 0.96. The bottom of the figure shows that the information entropy of the overall distribution is 0.37 and the comprehensive credibility is 0.938. From the shape of the distribution curve, four obvious peaks can be observed, which are located at the index values ​​of about 0.15, 0.32, 0.48 and 0.64, corresponding to the typical performance of different health states. The kernel density curve shows low values ​​near the thresholds T1, T2 and T3, indicating that these positions are reasonable state demarcation points. This technical solution not only accurately captures the distribution characteristics of different health states through the kernel density estimation method, but also quantifies the reliability of the division results through quantiles and credibility. Compared with the traditional threshold division method based on experience or simple statistics, this method based on data distribution characteristics can better reflect the natural hierarchical structure of the equipment health state and avoid the misjudgment that may be caused by arbitrary division. At the same time, the low information entropy value of 0.37 and the high comprehensive credibility of 0.938 further prove the effectiveness and reliability of the division method, providing a scientific basis for the assessment of equipment health status.

[0090] In this embodiment, through in-depth analysis of historical fault data, the resampling disturbance intensity can be dynamically adjusted to generate multiple groups of sample data, thereby ensuring the robustness and data representativeness of the fault diagnosis model. Compared with the traditional static sample sampling method, this technical means is more flexible and accurate, can adapt to changes in different equipment states, and improves the accuracy of fault diagnosis. Through the introduction of multi-objective optimization functions, it is possible to effectively balance the inter-class distribution distance, the intra-class distribution distance, and the time series continuity constraints, so as to find a balance point in the threshold division, ensuring that the health level division of the equipment is consistent with the actual distribution of the data and can reflect the long-term stability of the equipment. Compared with the prior art, this method avoids the deviation that may be caused by a single standard or model by comprehensively considering multiple dimensions of data distribution, and improves the accuracy of health level division. The kernel density estimation method is used to analyze the distribution of the initial threshold, and the probability density function of the data can be estimated without parameters to adapt to more complex fault data situations. This is more flexible than the traditional method of assuming a fixed distribution model, and can adapt to the state changes of different equipment in real time, further improving the adaptability and accuracy of the model. By calculating the distance distribution from the current fault diagnosis evaluation index to the dynamic interval and converting it into a probability distribution, the health status of the equipment can be dynamically evaluated. Combining information entropy as a credibility indicator can more accurately reflect the uncertainty of current data and enhance the model's ability to predict equipment failures. Compared with traditional fixed interval divisions or simple rule judgments, this method can dynamically adapt to the development trend of failures, thereby providing more accurate and timely warnings. By optimizing the dynamic interval, the health level division threshold can be more reasonably defined, making the health status assessment of the equipment more scientific and sensitive. Compared with existing static assessment methods, this technical solution has significant advantages in accuracy, flexibility and adaptability, and can more effectively prevent and predict equipment failures and improve the efficiency of equipment management and maintenance.

[0091] Figure 5 FIG. 1 is a schematic diagram of the structure of a multi-source data fusion diagnosis system for new energy station equipment according to an embodiment of the present invention. Figure 5 As shown, the system comprises: The first unit is used to collect and pre-process the operating data of new energy station equipment to obtain standardized operating data; The second unit is used to extract transient features and drift features from the standardized operation data, identify the transient fault mode of the equipment based on the transient features, identify the degradation fault mode of the equipment based on the drift features, and establish a dual-frequency domain fault feature library based on the transient fault mode and the degradation fault mode; The third unit is used to calculate the power transfer fluctuation value between devices, analyze the energy transfer efficiency of the device group, and when the energy transfer efficiency is lower than the preset transfer threshold, locate the fault source device in combination with the device fault mode library, identify the fault propagation path based on the fault source device, and generate a fault impact link diagram; The fourth unit is used to establish fault diagnosis evaluation indicators based on the fault impact link diagram and the dual-frequency domain fault feature library, perform hierarchical diagnosis on the equipment based on the fault diagnosis evaluation indicators, and evaluate the fault development trend in combination with the equipment's historical fault data, and output fault warning information and equipment health status assessment report; The fifth unit is used to generate a maintenance decision plan based on the fault warning information and the equipment health status assessment report, execute the maintenance decision plan, and record the maintenance effect data.

[0092] According to a third aspect of the embodiments of the present invention, An electronic device is provided, comprising: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0093] A fourth aspect of the embodiments of the present invention is: A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the aforementioned method is implemented.

[0094] The present invention may be a method, an apparatus, a system and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for executing various aspects of the present invention.

[0095] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A multi-source data fusion diagnosis method for new energy station equipment, characterized in that: include: Collect and pre-process the operating data of new energy station equipment to obtain standardized operating data; Extract transient features and drift features from standardized operation data, identify equipment transient fault modes based on transient features, identify equipment degradation fault modes based on drift features, and establish a dual-frequency domain fault feature library based on transient fault modes and degradation fault modes; Calculate the power transfer fluctuation value between devices and analyze the energy transfer efficiency of the device group. When the energy transfer efficiency is lower than the preset transfer threshold, locate the fault source device based on the device fault mode library, identify the fault propagation path based on the fault source device, and generate a fault impact link diagram. Establish fault diagnosis evaluation indicators based on the fault impact link diagram and dual-frequency domain fault feature library, perform graded diagnosis on the equipment based on the fault diagnosis evaluation indicators, and evaluate the fault development trend in combination with the equipment's historical fault data, output fault warning information and equipment health status assessment report; Generate maintenance decision plans based on fault warning information and equipment health status assessment reports, execute maintenance decision plans, and record maintenance effect data.

2. The method according to claim 1, characterized in that Extract transient features and drift features from standardized operation data, identify equipment transient fault modes based on transient features, identify equipment degradation fault modes based on drift features, and establish a dual-frequency domain fault feature library based on transient fault modes and degradation fault modes, including: Calculating the data volatility of the standardized operation data, constructing an adaptive harmonic basis function group based on the data volatility, and decomposing the standardized operation data to obtain a transient component and a drift component; Extract mutation characteristic parameters of transient component and trend characteristic parameters of drift component respectively; A transient feature matrix is ​​constructed based on mutation feature parameters, the transient feature matrix is ​​matched with a preset transient fault template, a multi-dimensional cosine similarity is calculated to obtain a transient fault similarity matrix, and a device transient fault mode is determined according to a fault type corresponding to a maximum similarity in the transient fault similarity matrix; A drift feature matrix is ​​constructed based on trend feature parameters, the drift feature matrix is ​​matched with a preset degradation fault template, a multi-dimensional Euclidean distance is calculated to obtain a degradation fault similarity matrix, and a device degradation fault mode is determined according to a fault type corresponding to a minimum distance in the degradation fault similarity matrix; The instantaneous fault similarity matrix and the degradation fault similarity matrix are weightedly fused to obtain a comprehensive fault similarity matrix. The instantaneous fault mode and the degradation fault mode are credibility evaluated according to the comprehensive fault similarity matrix. The fault modes that pass the credibility evaluation and the corresponding feature matrices are combined to establish a dual-frequency domain fault feature library.

3. The method according to claim 2, characterized in that The credibility evaluation of the transient fault mode and the degradation fault mode according to the comprehensive fault similarity matrix includes: Collecting the working condition state vector including load power, ambient wind speed, ambient temperature, vibration frequency characteristics and operation efficiency index, and performing maximum and minimum value normalization processing on the working condition state vector to obtain normalized working condition characteristics; Performing cluster analysis based on the normalized working condition characteristics to obtain an initial working condition boundary value, calculating the working condition distribution of the fault mode in the comprehensive fault similarity matrix to obtain a boundary loss function, performing a weighted combination of the boundary loss function and the standard deviation of the normalized working condition characteristics to obtain a boundary update amount, and dynamically optimizing the initial working condition boundary value based on the boundary update amount to obtain an optimized working condition boundary value; Calculating the matching degree of the fault mode in the comprehensive fault similarity matrix under the current working condition to obtain the local fitness, performing exponential smoothing on the local fitness within a specified time window to obtain the time series fitness, and adaptively weighting the local fitness and the time series fitness to obtain the comprehensive working condition fitness; The comprehensive fault similarity matrix and the comprehensive operating condition fitness are weightedly mapped to obtain a confidence reference value, the confidence reference value is multiplied by the constraint function of the optimized operating condition boundary value to obtain a modified confidence, the time-varying correction factor is calculated according to the time series standard deviation of the comprehensive operating condition fitness and the sliding window mean, and the modified confidence and the time-varying correction factor are weightedly fused according to the dynamic weight coefficient to obtain the credibility evaluation results of the transient fault mode and the degradation fault mode.

4. The method according to claim 1, characterized in that: Calculate the power transfer fluctuation value between devices and analyze the energy transfer efficiency of the device group. When the energy transfer efficiency is lower than the preset transfer threshold, locate the fault source device in combination with the device fault mode library, identify the fault propagation path based on the fault source device, and generate the fault impact link diagram including: Based on the device output power and the adjacent device input power within the sliding time window, the power difference squared at each sampling moment is multiplied by the corresponding time weight coefficient and then accumulated to obtain the power transfer fluctuation value between devices; The physical connection relationship between the collected devices is used to construct a topological connection matrix, a device transfer fluctuation matrix is ​​constructed based on the power transfer fluctuation value and the topological connection matrix, and the energy transfer efficiency of the device group is obtained according to the ratio of the device transfer fluctuation matrix to a preset reference fluctuation matrix; When the energy transfer efficiency of the device group is lower than the preset transfer threshold, the fault feature vector of each device is calculated based on the device transfer fluctuation matrix, and the fault feature vector is matched with the feature vector in the dual-frequency domain fault feature library for similarity to determine the fault source device; Taking the fault source device as the starting point, a fault propagation influence matrix is ​​constructed based on the device transfer fluctuation matrix and the topological connection matrix, wherein the propagation influence value of a directly connected device pair is its power transfer fluctuation value, and the propagation influence value of an indirectly connected device pair is the multiplication of the power transfer fluctuation value on the path divided by the power of the shortest path length; The priority path of fault propagation is determined based on the size of each element in the fault propagation influence matrix, the fault propagation probability of the priority path is calculated in combination with historical fault data, and a fault influence link diagram including the fault propagation path and propagation probability is generated.

5. The method according to claim 4, characterized in that Determining a priority path for fault propagation based on the size of each element in the fault propagation influence matrix, and calculating the fault propagation probability of the priority path in combination with historical fault data includes: Constructing a propagation path candidate set based on the fault propagation impact matrix, wherein each propagation path includes a sequence of propagation impact values ​​of adjacent device pairs; Performing wavelet decomposition on the propagation impact value sequence to obtain impact value components of different frequency bands, taking the components whose impact value components are less than a preset component threshold as the propagation baseline sequence, and taking the components whose impact value components are greater than the preset component threshold as the fluctuation sequence; The path stability is obtained by dividing the mean of the propagation baseline sequence by the standard deviation of the fluctuation sequence, and the overlap degree between the peak-to-valley distribution characteristics of the fluctuation sequence and the historical fault time is used as the path credibility. The propagation path candidate set is screened according to the path stability; For the selected propagation paths, the sample entropy of the propagation impact value sequence is calculated, the size of the sample entropy is used as the path complexity evaluation index, and the propagation path with the smallest path complexity is selected as the fault propagation priority path; Based on historical fault data, the fault occurrence time sequence correlation of adjacent devices on the propagation priority path is calculated, the fault occurrence time sequence correlation is constructed as a priori probability distribution in the form of a Gaussian distribution, the current propagation impact value sequence is subjected to wavelet transform, the wavelet coefficients at different scales are extracted, the amplitude and phase characteristics of the wavelet coefficients are combined into an observation vector, and the prediction probability at the current moment is calculated based on the posterior probability at the previous moment; A likelihood function is calculated according to the observation vector, the predicted probability is multiplied by the likelihood function and normalized to obtain the posterior probability at the current moment, iteration is repeated until the posterior probability converges, and the converged posterior probability is used as the fault propagation probability of the fault propagation priority path.

6. The method according to claim 1, characterized in that Establish fault diagnosis evaluation indicators based on the fault impact link diagram and dual-frequency domain fault feature library, perform graded diagnosis on the equipment based on the fault diagnosis evaluation indicators, and evaluate the fault development trend in combination with the equipment's historical fault data, output fault warning information and equipment health status evaluation reports including: Extracting common nodes of intersecting links from the fault-affected link graph, calculating the state coordination degree of the common nodes to construct a link coupling matrix, screening features in a dual-frequency domain fault feature library according to the link coupling matrix, calculating the matching degree between the features and the link coupling matrix, and selecting features with high matching degree to construct a candidate feature set; Extract the change sequence of the candidate feature set in the historical fault data, analyze the causal relationship of the change sequence to construct a feature propagation network, identify the strongest causal path from the feature propagation network as a feature diagnosis chain, calculate the propagation strength of each node in the feature diagnosis chain, take the product of the propagation strength and the feature matching degree as the feature weight, and perform weighted combination of the candidate feature set according to the feature weight to obtain the fault diagnosis evaluation index; The probability distribution of fault diagnosis evaluation indicators in historical fault data is statistically analyzed, and the health level division threshold is determined according to the probability distribution. The interval to which the current fault diagnosis evaluation indicator belongs is determined as the equipment health level. The fault development trend is predicted according to the state transition law of the characteristic diagnosis chain, and the fault warning information and health status assessment report are output in combination with the equipment health level.

7. The method according to claim 6, characterized in that Statistically analyzing the probability distribution of fault diagnosis evaluation indicators in historical fault data, and determining the health level classification threshold according to the probability distribution includes: Calculate the time series fluctuation characteristics of the fault diagnosis evaluation index in the historical fault data to obtain a local stability index, dynamically adjust the resampling disturbance intensity according to the local stability index, and use the disturbance intensity to resample the fault diagnosis evaluation index to generate multiple groups of sample data; The inter-class distribution distance, intra-class distribution distance and time series continuity constraints of multiple groups of sample data are combined to construct a multi-objective optimization function, and multiple groups of initial thresholds are obtained through iterative optimization. The kernel density estimation method was used to analyze the distribution characteristics of multiple groups of initial thresholds, and the dynamic intervals of health grade division were constructed based on quantiles at different confidence levels. The distance distribution from the current fault diagnosis evaluation index to the dynamic interval is calculated, the distance distribution is converted into a probability distribution, the information entropy is calculated based on the probability distribution and converted into a credibility index, and the dynamic interval is optimized in combination with the credibility index to obtain the final health level division threshold.

8. A multi-source data fusion diagnosis system for new energy station equipment, used to implement the method described in any one of claims 1 to 7, characterized in that: include: The first unit is used to collect and pre-process the operating data of new energy station equipment to obtain standardized operating data; The second unit is used to extract transient features and drift features from the standardized operation data, identify the transient fault mode of the equipment based on the transient features, identify the degradation fault mode of the equipment based on the drift features, and establish a dual-frequency domain fault feature library based on the transient fault mode and the degradation fault mode; The third unit is used to calculate the power transfer fluctuation value between devices, analyze the energy transfer efficiency of the device group, and when the energy transfer efficiency is lower than the preset transfer threshold, locate the fault source device in combination with the device fault mode library, identify the fault propagation path based on the fault source device, and generate a fault impact link diagram; The fourth unit is used to establish fault diagnosis evaluation indicators based on the fault impact link diagram and the dual-frequency domain fault feature library, perform hierarchical diagnosis on the equipment based on the fault diagnosis evaluation indicators, and evaluate the fault development trend in combination with the equipment's historical fault data, and output fault warning information and equipment health status assessment report; The fifth unit is used to generate a maintenance decision plan based on the fault warning information and the equipment health status assessment report, execute the maintenance decision plan, and record the maintenance effect data.

9. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the method described in any one of claims 1 to 7.

10. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images