Method and system for monitoring over-electric stress failure of network construction new energy converter

Abstract

The invention discloses a method and a system for monitoring over-electric stress failure of a network construction new energy converter. The method comprises the following steps: collecting multi-source heterogeneous monitoring data under the operation condition of the converter; performing feature extraction on the multi-source heterogeneous monitoring data, and fusing the extracted features to obtain fused features of the multi-source heterogeneous monitoring data; based on the fusion features, establishing a machine learning model of the converter; based on a machine learning model, evaluating the health state of the converter to obtain a health evaluation result; based on the machine learning model, predicting the residual life of the converter to obtain the predicted residual life; and generating an operation and maintenance decision based on the health assessment result and the predicted residual life to guide operation and maintenance of the converter. The invention provides a new online monitoring method for over-electric stress failure of a networking new energy converter based on multi-source heterogeneous data fusion and intelligent modeling, data fusion and machine learning are introduced, and full-life-cycle and all-around health management of the converter is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of reliability management of new energy power generation equipment, and more specifically, to a monitoring method and system for over-stress failure of a grid-connected new energy converter. Background Art

[0002] Renewable energy generation technology is a key means of addressing energy shortages and environmental pollution. Large-scale integration of renewable energy into the grid has become an inevitable trend in power grid development. Renewable energy generation relies on converters to connect to the grid, enabling bidirectional energy transmission and flexible power regulation. To ensure renewable energy consumption and grid security and stability, converters must possess voltage and frequency support capabilities, rapidly adjusting their output when grid voltage and frequency fluctuate to maintain a balance between the renewable energy generation system and the grid.

[0003] However, providing continuous support for the power grid causes converters to be subjected to frequent and severe overstress, accelerating insulation aging and breakdown failure of internal components. Statistics show that overstress failure has become a major failure mode in renewable energy converters, severely impacting the reliability and grid connection efficiency of renewable energy power generation systems. Once a converter fails due to overstress, it can cause severe consequences such as a sudden drop in renewable energy output, tripping of grid connection points, and even widespread reactive power loss, threatening the safe and stable operation of the power grid.

[0004] Currently, existing technologies for overstress converter failure mainly rely on experience-based periodic maintenance strategies, passively addressing failures through manual inspections and offline testing. This traditional approach has the following shortcomings:

[0005] (1) Lack of online monitoring means, unable to perceive the health status of the converter in real time, difficult to detect potential failures in time, and posing a greater operational risk;

[0006] (2) Relying on manual experience and lacking quantitative analysis capabilities, it is impossible to accurately evaluate the insulation performance and remaining life of the converter, and maintenance decisions are blind;

[0007] (3) Passive response to faults and subsequent maintenance lead to high maintenance costs, and the loss of new energy power during the converter outage is large, resulting in poor economic efficiency.

[0008] To address these issues, academia and industry have conducted a series of research and practical applications. Some researchers have proposed methods for monitoring converter overstress based on electrical parameter measurements. Another proposed an online monitoring method for IGBT collector voltage, which uses overthreshold pulse counts to reflect overstress levels. Another proposed a frequency-domain analysis method for converter DC bus voltage, which extracts peak frequencies to characterize overvoltage failures. These methods utilize electrical parameters to diagnose converter failures, offering a certain degree of real-time performance. However, they only consider a single physical quantity and struggle to fully characterize the failure mechanism.

[0009] Other researchers have proposed data-driven converter health assessment methods. For example, a converter fault diagnosis method based on support vector machines (SVMs) uses historical operating data to train a classification model and identify fault patterns. Another paper proposes a converter RUL prediction method based on a deep belief network (DBN) that extracts degradation features and trains a mapping model to predict remaining useful life. These methods utilize machine learning to mine knowledge from data without the need for complex physical models and possess a certain level of intelligence. However, they lack the ability to integrate and analyze heterogeneous data from multiple sources, making it difficult to fully exploit the value of the data.

[0010] In summary, how to obtain the status information of converter over-stress failure online, how to extract and fuse features from multi-source heterogeneous data, and how to build evaluation and prediction models for converter health management have become key issues to be solved. Summary of the Invention

[0011] The technical solution of the present invention provides a method and system for monitoring over-electrical stress failure of a grid-connected new energy converter, so as to solve the problem of how to monitor over-electrical stress failure of a grid-connected new energy converter.

[0012] In order to solve the above problems, the present invention provides a method for monitoring over-voltage stress failure of a grid-connected new energy converter, the method comprising:

[0013] Collect multi-source heterogeneous monitoring data under converter operating conditions;

[0014] Extracting features from the multi-source heterogeneous monitoring data and fusing the extracted features to obtain fused features of the multi-source heterogeneous monitoring data;

[0015] Establishing a machine learning model of the converter based on the fused features;

[0016] Evaluate the health status of the converter based on the machine learning model and obtain a health assessment result;

[0017] Predicting the remaining life of the converter based on the machine learning model to obtain a predicted remaining life;

[0018] An operation and maintenance decision is generated based on the health assessment result and the predicted remaining life, and operation and maintenance of the converter is guided based on the operation and maintenance decision.

[0019] Preferably, the multi-source heterogeneous monitoring data includes: electrical parameters, thermal parameters and environmental parameters;

[0020] The electrical parameters are collected by voltage sensors and current sensors arranged in the main electrical circuit of the converter;

[0021] The thermal parameters are collected by an infrared thermometer arranged on the surface of the converter;

[0022] The environmental parameters are collected by meteorological sensors arranged on the surface of the converter.

[0023] Preferably, the feature extraction of the multi-source heterogeneous monitoring data includes:

[0024] Performing time-frequency domain analysis on the electrical parameters, and extracting characteristics reflecting the characteristic amplitude, duration, rising edge rate, and falling edge rate of the transient overvoltage based on the time-frequency domain analysis results;

[0025] performing image processing on the thermal parameters, and extracting features reflecting the highest temperature point, the area of the highest temperature region, and the average temperature gradient of the thermal stress distribution based on the image processing results;

[0026] Statistical analysis is performed on the environmental parameters, and based on the statistical analysis results, features of wind speed, wind direction, and air pressure that reflect the environmental stress level are extracted.

[0027] Preferably, the fusing of the extracted features includes:

[0028] The various features of the extracted multi-source heterogeneous monitoring data are sequentially spliced to generate a high-dimensional matrix, wherein the rows of the high-dimensional matrix represent sampling time points, and the columns of the high-dimensional matrix represent features of a class of multi-source heterogeneous monitoring data;

[0029] Based on the JCA algorithm, feature transformation is performed on the high-dimensional matrix to obtain a set of fusion feature vectors after dimension reduction.

[0030] Preferably, establishing a machine learning model of the converter based on the fusion features includes:

[0031] Taking the fusion features as input, establishing a health status assessment model for the converter based on a support vector machine algorithm;

[0032] Taking the historical time series health index as input, a remaining life prediction model of the converter is established based on the long short-term memory neural network.

[0033] According to another aspect of the present invention, a monitoring system for over-voltage stress failure of a grid-connected new energy converter is provided, the system comprising:

[0034] The acquisition unit is used to collect multi-source heterogeneous monitoring data under the operating conditions of the converter;

[0035] an extraction unit, configured to extract features from the multi-source heterogeneous monitoring data and fuse the extracted features to obtain fused features of the multi-source heterogeneous monitoring data;

[0036] an establishing unit, configured to establish a machine learning model of the converter based on the fusion features;

[0037] an evaluation unit, configured to evaluate the health status of the converter based on the machine learning model and obtain a health evaluation result;

[0038] A prediction unit, configured to predict the remaining life of the converter based on the machine learning model to obtain a predicted remaining life;

[0039] A result unit is used to generate an operation and maintenance decision based on the health assessment result and the predicted remaining life, and guide the operation and maintenance of the converter based on the operation and maintenance decision.

[0040] Preferably, the multi-source heterogeneous monitoring data includes: electrical parameters, thermal parameters and environmental parameters;

[0041] The electrical parameters are collected by voltage sensors and current sensors arranged in the main electrical circuit of the converter;

[0042] The thermal parameters are collected by an infrared thermometer arranged on the surface of the converter;

[0043] The environmental parameters are collected by meteorological sensors arranged on the surface of the converter.

[0044] Preferably, the extraction unit is used to extract features from the multi-source heterogeneous monitoring data, and is also used to:

[0045] Performing time-frequency domain analysis on the electrical parameters, and extracting characteristics reflecting the characteristic amplitude, duration, rising edge rate, and falling edge rate of the transient overvoltage based on the time-frequency domain analysis results;

[0046] performing image processing on the thermal parameters, and extracting features reflecting the highest temperature point, the area of the highest temperature region, and the average temperature gradient of the thermal stress distribution based on the image processing results;

[0047] Statistical analysis is performed on the environmental parameters, and based on the statistical analysis results, features of wind speed, wind direction, and air pressure that reflect the environmental stress level are extracted.

[0048] Preferably, the extraction unit is used to fuse the extracted features and further to:

[0049] The various features of the extracted multi-source heterogeneous monitoring data are sequentially spliced to generate a high-dimensional matrix, wherein the rows of the high-dimensional matrix represent sampling time points, and the columns of the high-dimensional matrix represent features of a class of multi-source heterogeneous monitoring data;

[0050] Based on the JCA algorithm, feature transformation is performed on the high-dimensional matrix to obtain a set of fusion feature vectors after dimension reduction.

[0051] Preferably, the establishing unit is used to establish a machine learning model of the converter based on the fusion features, and is further used to:

[0052] Taking the fusion features as input, establishing a health status assessment model for the converter based on a support vector machine algorithm;

[0053] Taking the historical time series health index as input, a remaining life prediction model of the converter is established based on the long short-term memory neural network.

[0054] The present invention proposes a new online monitoring method for over-stress failure of networked new energy converters based on multi-source heterogeneous data fusion and intelligent modeling. On the basis of traditional monitoring and diagnosis technologies, new technologies such as data fusion and machine learning are introduced to realize the full life cycle and all-round health management of the converter, which is of great significance for improving the reliability level of new energy power generation systems and promoting the efficient consumption of new energy. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] A more complete understanding of exemplary embodiments of the present invention may be obtained by referring to the following drawings:

[0056] Figure 1 Flowchart of a method for monitoring over-stress failure of a grid-connected new energy converter according to a preferred embodiment of the present invention;

[0057] Figure 2 This is a block diagram of an overall technical solution for monitoring over-stress failure of a grid-connected new energy converter according to a preferred embodiment of the present invention;

[0058] Figure 3 Flowchart of converter multi-source heterogeneous data acquisition and preprocessing according to a preferred embodiment of the present invention;

[0059] Figure 4 A flowchart of converter health status assessment and remaining life prediction modeling according to a preferred embodiment of the present invention; and

[0060] Figure 5 The structure diagram of a monitoring system for over-voltage stress failure of a grid-connected new energy converter according to a preferred embodiment of the present invention is shown. DETAILED DESCRIPTION

[0061] Exemplary embodiments of the present invention will now be described with reference to the accompanying drawings. However, the present invention may be embodied in many different forms and is not limited to the embodiments described herein. These embodiments are provided to provide a thorough and complete disclosure of the present invention and to fully convey the scope of the present invention to those skilled in the art. The terminology used in the exemplary embodiments shown in the accompanying drawings is not intended to limit the present invention. In the accompanying drawings, identical elements are denoted by the same reference numerals.

[0062] Unless otherwise specified, the terms used herein (including technical terms) have the meanings commonly understood by those skilled in the art. In addition, it is understood that terms defined in commonly used dictionaries should be understood to have the same meanings as those in the context of the relevant fields, and should not be understood as idealized or overly formal meanings.

[0063] Figure 1 The present invention is a flowchart of a method for monitoring over-stress failure of a grid-connected new energy converter according to a preferred embodiment of the present invention.

[0064] The present invention proposes an online monitoring method for over-stress failure of grid-connected new energy converters. Based on multi-source heterogeneous data fusion and intelligent modeling technology, this method realizes real-time perception of the converter's operating status, quantitative assessment of its health level, and dynamic prediction of its remaining life, thereby guiding the converter's condition inspection and operation and maintenance decisions.

[0065] The present invention constructs a multi-source heterogeneous data acquisition system for converters. This system targets the key components and weak links of converters, creating a multi-parameter, all-encompassing state monitoring system. Voltage and current sensors are placed on key components of the converter's main electrical circuit, such as IGBT modules and capacitors, to collect electrical parameters in real time. Infrared thermometers are placed on the heat sinks of the converter's power devices to collect temperature parameters in real time. Meteorological sensors are placed at the converter's site to collect environmental parameters in real time. Simultaneously, through industrial Ethernet communication technology, dispersed sensor data is centralized to an edge computing gateway, enabling unified data management and preprocessing. The collaborative acquisition of multi-source heterogeneous data provides a data foundation for a comprehensive depiction of the converter's operating status.

[0066] The present invention provides a data fusion method for overvoltage failure characteristics. The present invention proposes a data fusion method for converter overvoltage failure characteristics, aiming to extract and fuse key information reflecting failure characteristics from multi-source heterogeneous data. First, the collected electrical parameters are analyzed in the time and frequency domains to extract indicators such as peak value, duration, and rate of rise that reflect transient overvoltage characteristics; image processing is performed on the temperature parameters to extract indicators such as temperature gradient and maximum temperature that reflect thermal stress distribution; and statistical analysis is performed on the environmental parameters to extract indicators such as mean and variance that reflect external environmental stress. On this basis, the Joint Component Analysis (JCA) method is used to map characteristic indicators from different sources and scales into a unified low-dimensional space to obtain fused indicators that comprehensively reflect overvoltage failure characteristics. This fusion method fully exploits the inherent correlation of multi-source data, effectively eliminates data redundancy and measurement noise, and improves the accuracy and robustness of feature representation.

[0067] The present invention provides a converter health status assessment and remaining life prediction model. The present invention constructs a data-driven converter health status assessment and remaining life prediction model to achieve intelligent management of the converter from health to failure. First, the Support Vector Machine (SVM) algorithm is used to train a binary classification model of converter over-stress failure with fusion indicators as input to achieve real-time diagnosis of the converter health status. On this basis, the Support Vector Regression (SVR) algorithm is introduced to quantify the current health level of the converter and output a health index (HI) between 0 and 1. The closer HI is to 1, the healthier the converter; the closer HI is to 0, the higher the risk of converter failure.

[0068] Based on the real-time health index of the converter, the present invention further constructs a remaining useful life prediction model. Utilizing the HI degradation sequence in the converter's historical operating data, with the time series HI value as input and the actual remaining useful life (RUL) as output, the long short-term memory (LSTM) neural network is trained to learn the intrinsic mapping relationship between the health degradation law and the remaining useful life. By inputting the current and historical HI sequences of the converter in real time, the probability distribution of the remaining useful life of the converter at the current health level is dynamically predicted. This prediction model fully considers the nonlinearity and uncertainty of the converter degradation process, and improves the long-term accuracy of life prediction.

[0069] The present invention provides an adaptive converter condition-based maintenance decision-making method. Based on health status assessment and life prediction results, the method dynamically optimizes maintenance strategies to maximize the converter's lifecycle value. When the diagnostic results indicate the converter is healthy, no maintenance action is taken and status changes continue to be monitored. When the diagnostic results indicate the converter is subhealthy, the method combines the health index and remaining life prediction results to determine whether to trigger an early warning and plan the optimal maintenance time. When the diagnostic results indicate the converter is in a failed state, a maintenance task is immediately triggered, and historical operating and environmental data are analyzed to locate the root cause of the failure and develop a targeted maintenance plan. During the maintenance process, maintenance effectiveness is continuously evaluated, and maintenance strategies are dynamically adjusted to ensure rapid recovery of the converter. This decision-making method, while balancing equipment utilization and power supply reliability, reduces unnecessary preventive maintenance and improves the economic efficiency of maintenance resource allocation.

[0070] This invention provides an integrated software platform for converter health management. This platform integrates and visualizes functions such as condition monitoring, health assessment, lifespan prediction, and maintenance decision-making. Through a human-computer interaction interface, the platform enables real-time presentation of converter operating status, quantitative assessment of health levels, trend prediction of remaining lifespan, and dynamic generation of maintenance strategies. This platform enables managers to gain comprehensive control of their converters and make coordinated decisions and allocate resources based on assessment results. Ultimately, this platform achieves closed-loop management of the converter's entire process, from condition perception, health diagnosis, trend prediction, to maintenance decision-making, thereby enhancing the intelligent level of converter operation and maintenance.

[0071] This invention focuses on the reliability problem of over-stress failure of new energy converters, and starts from the source of state monitoring data to build a complete and intelligent health management technology solution. This solution is based on the industry demand for intelligent operation and maintenance of energy and power equipment, fully taps the value of industrial big data, and uses the fusion analysis of multi-source heterogeneous data to refine and enhance the characteristic representation of equipment failure; introduces machine learning algorithms to break through the limitations of traditional physical models and achieve accurate assessment of equipment health status and remaining life; optimizes operation and maintenance strategies, maximizes the benefits of equipment life cycle, and promotes the formation of an intelligent closed loop of equipment management. The technical route of this invention is complete, the method is advanced, and the practicality is strong. It is of great significance to improve the reliability of new energy power generation systems and promote the high-quality development of new energy. At the same time, the exploration of the application of multi-source data fusion and machine learning in industrial scenarios by this invention has also set a benchmark for intelligent operation and maintenance of energy equipment, and will provide new ideas and examples for the digital transformation of the energy and power industry.

[0072] The technical field of this invention intersects multiple disciplines, including renewable energy power generation, power electronics, sensor technology, signal processing, data fusion, and machine learning, embodying a distinct multidisciplinary cross-integration feature. In the field of energy and power, this invention addresses the reliability pain points of renewable energy converters, providing new ideas and methods for improving renewable energy power generation efficiency and reducing grid connection risks. In the field of artificial intelligence, this invention introduces a data-driven modeling paradigm into the health management of industrial equipment, enabling intelligent operation of the entire process from condition monitoring, fault diagnosis, and life prediction, and promoting the deep integration of industrial big data and artificial intelligence technologies.

[0073] like Figure 1 As shown, the present invention provides a method for monitoring over-stress failure of a grid-connected new energy converter, the method comprising:

[0074] Step 101: Collect multi-source heterogeneous monitoring data under wind power converter operating conditions;

[0075] Preferably, the multi-source heterogeneous monitoring data includes: electrical parameters, thermal parameters and environmental parameters;

[0076] Electrical parameters are collected through voltage sensors and current sensors arranged in the main electrical circuit of the wind power converter;

[0077] Thermal parameters are collected by infrared thermometers placed on the surface of the wind turbine converter;

[0078] Environmental parameters are collected through meteorological sensors arranged on the surface of the wind turbine converter.

[0079] The present invention collects multi-source heterogeneous monitoring data such as electrical parameters, thermal parameters, and environmental parameters under the operating conditions of the converter, including:

[0080] (1) Collect electrical parameters through voltage and current sensors arranged in the main electrical circuit of the converter;

[0081] (2) Collect thermal parameters through infrared thermometers placed on the surface of key components of the converter;

[0082] (3) Collect environmental parameters through meteorological sensors arranged at the site where the converter is located.

[0083] This invention monitors the electrical parameters of the IGBT switching devices in the grid-side converter unit of a wind turbine converter. It uses a high-voltage differential probe and Hall current sensor to sample the IGBT's collector voltage and emitter current in real time at a 10kHz data acquisition frequency. The data is then uploaded to an intelligent gateway via an RJ45 Ethernet port.

[0084] This invention monitors the thermal parameters of the converter power unit's heat sink. An infrared temperature measurement array is placed on the IGBT module heat sink surface with a spatial resolution of 1 cm and a temporal resolution of 1 s. This array collects IGBT junction temperature distribution data in real time and uploads it to an intelligent gateway via the CAN bus.

[0085] This system monitors environmental parameters within the wind turbine tower and nacelle. Temperature and humidity sensors are placed at different heights on the tower to monitor ambient temperature and humidity, while vibration sensors are placed within the nacelle to monitor mechanical vibration levels. Environmental data is uploaded to an intelligent gateway via wireless LoRa communication.

[0086] The wind farm of the present invention is equipped with a weather station to monitor meteorological parameters such as wind speed, wind direction, and air pressure with a time granularity of 1 minute. The meteorological data is uploaded to the wind farm main control system via GPRS communication.

[0087] The present invention preprocesses the collected data:

[0088] After the intelligent gateway receives the multi-source heterogeneous data collected by the wind turbine, it first performs data preprocessing:

[0089] Outlier processing: Tukey algorithm (box plot) was used to detect and eliminate outliers.

[0090] Time synchronization: Data from different data sources are timestamped to achieve millisecond-level synchronization.

[0091] Data cleaning: remove incomplete data, out-of-range data, etc.

[0092] Data normalization: The Z-score method was used to normalize the data.

[0093] Step 102: extracting features from the multi-source heterogeneous monitoring data and fusing the extracted features to obtain fused features of the multi-source heterogeneous monitoring data;

[0094] Preferably, feature extraction is performed on multi-source heterogeneous monitoring data, including:

[0095] Perform time-frequency domain analysis on electrical parameters and extract the characteristic amplitude, duration, rising edge rate, and falling edge rate of transient overvoltage based on the time-frequency domain analysis results;

[0096] Perform image processing on thermal parameters and extract the features of the highest temperature point, the area of the highest temperature region, and the average temperature gradient reflecting the thermal stress distribution based on the image processing results;

[0097] Statistical analysis is performed on environmental parameters, and based on the statistical analysis results, the characteristics of wind speed, wind direction, and air pressure that reflect the environmental stress level are extracted.

[0098] Preferably, the extracted features are fused, including:

[0099] The various features of the extracted multi-source heterogeneous monitoring data are sequentially spliced to generate a high-dimensional matrix. The rows of the high-dimensional matrix represent the sampling time points, and the columns of the high-dimensional matrix represent the features of a class of multi-source heterogeneous monitoring data.

[0100] Based on the JCA algorithm, the high-dimensional matrix is transformed to obtain a set of fusion feature vectors after dimensionality reduction.

[0101] The present invention extracts features from multi-source heterogeneous monitoring data and uses a joint component analysis method to achieve consistent representation and fusion of multi-source features;

[0102] Feature extraction includes:

[0103] (1) Perform time-frequency domain analysis on electrical parameters to extract indicators such as amplitude and duration that reflect transient overvoltage characteristics;

[0104] (2) Perform image processing on thermal parameters to extract indicators such as temperature gradient and maximum temperature that reflect thermal stress distribution;

[0105] (3) Conduct statistical analysis on environmental parameters and extract indicators such as mean and variance that reflect the level of environmental stress.

[0106] The present invention performs feature extraction on the preprocessed data:

[0107] 1. Electrical parameters: Short-time Fourier transform is used to perform time-frequency analysis on the IGBT collector voltage and emitter current to extract characteristic indicators reflecting overvoltage, including overvoltage amplitude, duration, rising edge rate, falling edge rate, etc. A characteristic vector is formed every 10 cycles.

[0108] 2. Thermal parameters: Perform image segmentation and regional analysis on the infrared thermal imaging image of the IGBT heat sink to extract indicators such as the highest temperature point, the area of the highest temperature area, and the average temperature gradient, and form a feature vector every minute.

[0109] 3. Environmental parameters: The sliding window method is used to perform statistical analysis on temperature, humidity, and vibration data. The mean, variance, and extreme values of temperature / humidity / vibration are extracted, and a feature vector is formed every 10 minutes.

[0110] 4. Meteorological parameters: Wind speed, wind direction, and air pressure are selected as feature indicators, with one feature vector generated every 10 minutes. All feature vectors are aligned according to a unified timestamp to form a time series multidimensional feature set.

[0111] The present invention performs data feature fusion and adopts Joint Component Analysis (JCA) to fuse multi-source heterogeneous features:

[0112] 1. Concatenate various types of features into a high-dimensional matrix in sequence. Each row represents a sampling time point, each column represents a type of feature index, with a total of N sampling points and M-dimensional features.

[0113] 2. Use the JCA algorithm to perform feature transformation on the high-dimensional matrix to obtain a set of K-dimensional (K << M) fused feature vectors. JCA learns the projection matrix by maximizing the correlation of features from different sources, enabling dimensionality reduction of the fused features with the minimum reconstruction error.

[0114] In the present invention, high-dimensional heterogeneous features such as electrical, thermal, environmental, and meteorological features are projected into a low-dimensional space through JCA to obtain low-dimensional overvoltage failure fused features. While the feature dimension is greatly compressed, the feature expression ability is enhanced.

[0115] Applying the online monitoring method provided by the present invention to the target offshore wind farm for continuous operation for 1 year has achieved good results:

[0116] 1. The monitoring system operates stably, and the multi-source heterogeneous data collected is complete and reliable, with a data missing rate lower than 1%, providing high-quality data support for condition assessment.

[0117] 2. The accuracy of the converter fault diagnosis model reaches 95%. Overvoltage failure warnings are issued for 12 wind turbine converters, and fault symptoms are identified 2 - 7 days in advance, enabling timely arrangement of maintenance to avoid the expansion of accidents.

[0118] 3. The health status assessment model continuously monitors that the health index of 1 wind turbine converter continues to decline, with the HI reaching as low as 0.65. Through fault analysis, the aging IGBT module and DC bus capacitor are replaced in a timely manner, restoring the health level of the equipment.

[0119] 4. The remaining useful life prediction model successfully warns of the end of life of multiple wind turbines, with the RUL prediction error lower than 10%, providing a reliable basis for maintenance decision-making. By optimizing the maintenance time window, two typhoon weather conditions are avoided, reducing power generation losses.

[0120] 5. The online monitoring system realizes the full-process management of wind power converters from state perception, health assessment to life prediction and maintenance decision-making, improving the intelligent operation and maintenance level of the wind farm. Compared with the same period of the previous year, the number of converter fault shutdowns is reduced by 30%, the average maintenance time is shortened by 12%, the power generation of the wind farm is increased by 2.5%, and the operation and maintenance cost is reduced by 15%.

[0121] The present invention is feasible and effective for online monitoring of wind power converters. It overcomes the difficulties in data collection and communication caused by the harsh environment of offshore wind farms, and realizes real-time perception and panoramic characterization of converter status. It constructs a complete multi-source data fusion and intelligent modeling system for the typical overvoltage failure mode of wind power converters, and forms an end-to-end closed-loop solution covering status monitoring, health assessment, life prediction, and operation and maintenance decision-making, which significantly improves the reliability of wind power converters, provides new ideas and methods for the intelligent operation of wind farms, and has good prospects for promotion and application.

[0122] Step 103: Establishing a machine learning model for the wind power converter based on the fusion features;

[0123] The present invention constructs a machine learning model to achieve real-time evaluation of the converter's health status and dynamic prediction of its remaining life. The constructed model includes:

[0124] (1) Taking multi-source fusion features as input, a support vector machine algorithm is used to build a converter health status assessment model;

[0125] (2) Taking the historical time series health index as input, a long short-term memory neural network is used to construct a converter remaining life prediction model.

[0126] Step 104: Evaluate the health status of the wind turbine converter based on the machine learning model and obtain a health evaluation result;

[0127] Step 105: Predicting the remaining life of the wind turbine converter based on the machine learning model to obtain a predicted remaining life;

[0128] Step 106: Generate an operation and maintenance decision based on the health assessment result and the predicted remaining lifespan, and guide the operation and maintenance of the wind turbine converter based on the operation and maintenance decision.

[0129] Preferably, a machine learning model for wind power converter is established based on the fusion features, including:

[0130] Taking the fusion features as input, a health status assessment model for wind power converters is established based on the support vector machine algorithm.

[0131] Taking the historical time series health index as input, a remaining life prediction model for wind turbine converters is established based on long short-term memory neural network.

[0132] The maintenance decision of the present invention includes:

[0133] (1) Determine whether the converter needs maintenance based on the health status assessment results;

[0134] (2) Determine the optimal maintenance time based on the remaining life prediction results;

[0135] (3) If maintenance is triggered, diagnose the cause of the failure and formulate a maintenance plan;

[0136] (4) Evaluate maintenance effects and dynamically optimize maintenance strategies.

[0137] The present invention integrates health assessment and life prediction results to form an adaptive condition-based maintenance decision-making plan to guide converter operation and maintenance.

[0138] The present invention performs health status assessment. Based on the fusion features, the present invention constructs a health status assessment model for wind power converters:

[0139] 1. Manually label historical operating data, selecting 500 samples from normal operating conditions and 500 samples from overvoltage failure conditions to form the training set. Each sample uses the JCA fusion feature as input, with the normal state labeled as 0 and the failure state labeled as 1.

[0140] 2. The SVM algorithm is used to train the binary classification model, and the RBF kernel function and grid search are used to optimize the model hyperparameters. Finally, the converter overvoltage failure diagnosis model is obtained, and the real-time health status of the newly collected monitoring data is judged.

[0141] 3. Use an SVM+regression model to quantitatively assess the converter's health. Historical samples are manually scored for health, with a health index (HI) ranging from 0 to 1 representing healthy to failed. Using the fused features as input and the HI as output, an SVM regression model is trained to calculate the HI value in real time for new monitoring data.

[0142] The present invention predicts the remaining life. The present invention selects the accelerated life test data in the historical operating conditions of the converter to construct a remaining life prediction model:

[0143] 1. Data acquisition: Apply overvoltage stress higher than the rated operating condition to the experimental converter and run it continuously until failure, while collecting multi-source monitoring data.

[0144] 2. Data annotation: Calculate the health index HI for the collected data, and manually record the actual remaining life RUL at each sampling moment to form a HI-RUL data pair.

[0145] 3. Data preparation: Use a sliding time window of length m to extract HI sequence samples. Each sample corresponds to the RUL value at the end of the window to form a feature-label dataset.

[0146] 4. Model training: Build an LSTM network with the HI sequence as input and RUL as output. The network structure is m->32->16->1, the loss function is mean square error, and the Adam optimizer is used. Train for 1000 epochs.

[0147] 5. Model evaluation: The leave-one-out method was used to evaluate the generalization performance of the model, with an average prediction error (RMSE) of 0.15 and an average relative error (MAPE) of 5%.

[0148] 6. Model Application: Use the trained LSTM network to dynamically predict the RUL of the real-time HI sequence. Output the prediction results as a probability density distribution, and obtain the RUL interval estimate at a confidence level of 95%.

[0149] The present invention optimizes operation and maintenance decisions by integrating wind power converter health status assessment and remaining life prediction results to form intelligent operation and maintenance decisions:

[0150] 1. Condition monitoring: When the converter is running, the monitoring system collects multi-source data in real time, calculates fusion features, and inputs them into the SVM diagnostic model to determine the health status. When an overvoltage failure state is detected, an alarm is immediately issued, and the converter's reactive output is suppressed to prevent the failure from expanding.

[0151] 2. Health Assessment: Newly collected monitoring data is fed into the SVM regression model in real time, and the health index (HI) value is calculated online and a health trend curve is plotted. If the HI value falls below the warning threshold of 0.7 for three consecutive days, a health report is generated, and a fault tree analysis is performed on the converter to infer the cause of degradation and develop a maintenance plan.

[0152] 3. Lifespan Prediction: A 30-day HI sequence is fed into the LSTM prediction model to obtain the current RUL distribution. If the lower bound of the 95% confidence interval for the RUL is less than 90 days, the unit is designated as a high-risk O&M target and included in the annual maintenance plan. If the lower bound of the 95% confidence interval for the RUL is less than 30 days, prompt downtime for maintenance and replacement of faulty components is recommended.

[0153] 4. Decision optimization: Compare and analyze the RUL prediction results of different wind turbines, combine factors such as wind resource forecast, electricity price forecast, spare parts reserves, and comprehensively arrange wind farm maintenance plans, formulate optimal maintenance time windows and maintenance projects, and minimize maintenance costs while ensuring power generation efficiency.

[0154] The present invention provides an online monitoring method for over-stress failure of grid-connected new energy converters. Based on the converter's multi-source heterogeneous data fusion and intelligent modeling technology, it achieves real-time perception of the converter's health status, early warning of failure risks, and optimization of maintenance decisions.

[0155] The present invention constructs a real-time acquisition system for the converter's electrical, thermal, and environmental parameters, and obtains status monitoring data under the converter's operating conditions through Internet of Things sensing technology, breaking through the bottleneck of traditional offline monitoring data acquisition.

[0156] The present invention proposes a multi-source heterogeneous data fusion analysis method for over-voltage stress failure characteristics. Through feature extraction and joint component analysis, the consistent representation and feature enhancement of different modal monitoring data are achieved, the intrinsic correlation of the data is mined, and the accuracy and robustness of the feature representation are improved.

[0157] This paper uses machine learning algorithms to construct a converter health status assessment and remaining life prediction model, quantify the current health level of the equipment, predict its degradation trend, and play a core data-driven role in the full life cycle management of the equipment.

[0158] The present invention proposes an adaptive converter condition maintenance decision-making method that integrates health assessment and life prediction results, automatically determines maintenance needs and timing, optimizes maintenance plans, and achieves continuous improvement of the maintenance process through closed-loop feedback, promoting the formation of an intelligent operation and maintenance model that prioritizes prevention, prioritizes diagnosis, performs maintenance on demand, and continuously optimizes.

[0159] The online monitoring method for over-stress failure of grid-connected new energy converters proposed in the present invention realizes real-time perception of converter health status, early warning of failure risks, and optimization of maintenance decisions through fusion analysis and intelligent modeling of multi-source heterogeneous data. It effectively solves the difficulties in converter over-stress failure monitoring and operation and maintenance, and has achieved remarkable results in improving converter reliability and ensuring stable operation of new energy power generation systems.

[0160] The present invention realizes online real-time monitoring of converter over-electrical stress failure, thereby improving the sensitivity and timeliness of monitoring.

[0161] The present invention deploys multi-parameter sensors for electrical, thermal, and environmental parameters at key locations on the converter to collect real-time status data under operating conditions. It then fuses and analyzes multi-source heterogeneous data to extract key indicators reflecting overstress failure characteristics, dynamically characterizing the converter's health status and ensuring the converter's operating status is always under monitoring. Compared to traditional offline detection and regular inspections, the present invention enables continuous monitoring of the converter's status. Once signs of failure are detected, an immediate warning is issued, enabling maintenance personnel to take timely action, thereby minimizing the occurrence or expansion of failures. This improves the sensitivity and timeliness of converter monitoring and reduces the risk of accidents. Preliminary estimates suggest that the online monitoring method of the present invention can increase the detection rate of converter fault hazards by over 30%.

[0162] The present invention constructs a fusion analysis framework for multi-source heterogeneous data of converters, fully tapping the value of massive monitoring data.

[0163] The present invention collects massive amounts of status monitoring data during the operation of the converter. These data contain important information about the health status of the converter, but data from different sources and types are often scattered and heterogeneous, and a single data source is difficult to fully reflect the status of the equipment. The present invention innovatively proposes a data fusion analysis framework for over-stress failure characteristics. By extracting features from multi-source parameters such as electrical, thermal, and environmental parameters, and using a joint component analysis method to map multi-source features to a unified feature space, data from different modalities are consistently represented, the inherent correlation behind the data is mined, and measurement errors and data redundancy are suppressed. It is calculated that the accuracy of feature representation can be improved by more than 20% using the data fusion method of the present invention. The present invention realizes the value conversion of massive monitoring data and provides high-quality data support for subsequent equipment health assessment and life prediction.

[0164] The present invention establishes a data-driven equipment health assessment and life prediction model, realizing quantitative analysis and trend prediction of equipment degradation status.

[0165] Current inverter condition assessment and life prediction methods rely primarily on expert experience and theoretical models, which are subject to significant subjectivity and low prediction accuracy. This invention utilizes machine learning algorithms to automatically learn equipment degradation patterns from a large amount of historical operating and monitoring data. This approach establishes an inverter health index assessment model and a remaining life prediction model. This mathematical model characterizes the intrinsic connection between the equipment's health status and the data manifold, overcoming the limitations of empirical models. The health assessment model automatically determines the equipment's health level and quantifies its degree of degradation based on current monitoring data. The life prediction model estimates the probability distribution of the equipment's remaining useful life and characterizes its degradation trend. Statistics show that the health assessment model of this invention can achieve an accuracy rate of over 95% for condition identification, while the life prediction model of this invention can keep the life estimation error within 10%. Through quantitative analysis of equipment status and trend prediction, operations and maintenance personnel can intuitively understand the health of the equipment, enabling them to conduct predictive maintenance and implement repairs before equipment failure, effectively reducing the risk of unplanned downtime.

[0166] The present invention forms an intelligent closed-loop operation and maintenance mode for the converter from state perception, health diagnosis to maintenance decision-making, which improves the pertinence and economy of maintenance.

[0167] This invention addresses the challenges of traditional converter maintenance, which primarily rely on post-maintenance or periodic maintenance strategies. These strategies often result in untimely or excessive maintenance, leading to low equipment utilization and high maintenance costs. Based on real-time converter status monitoring, this invention integrates the results of health assessment and lifespan prediction to form an adaptive condition-based maintenance decision-making method. Based on the converter's health status, this method automatically determines whether maintenance is necessary and the optimal timing. After triggering a maintenance task, it also diagnoses the cause of the failure, optimizes the maintenance plan, and provides feedback on the maintenance process, ultimately forming a closed-loop, dynamically optimized maintenance decision-making chain. By replacing periodic maintenance with condition-based maintenance and post-maintenance with on-demand maintenance, unnecessary downtime for maintenance is reduced, improving equipment utilization. According to estimates, the intelligent maintenance decision-making method of this invention can extend the converter's mean time between failures (MTBF) by over 20% and reduce annual maintenance costs by over 10%. The present invention realizes end-to-end intelligent management of the converter from state perception, health diagnosis, trend prediction to maintenance decision-making, and promotes the formation of a new intelligent operation and maintenance model with prevention as the main focus, diagnosis first, on-demand maintenance and continuous optimization.

[0168] The present invention improves the overall reliability level of the new energy power generation converter and ensures the efficient and friendly grid connection of new energy.

[0169] As a key component in the grid integration of renewable energy power generation, the reliable operation of converters directly impacts renewable energy generation efficiency and power quality. Converter overstress failure can lead to sudden drops in output power and excessive harmonic content. In severe cases, this can trigger cascading trips, leading to widespread curtailment of renewable energy resources, impacting grid security and stability. This invention minimizes unplanned converter downtime through online monitoring, integrated diagnosis, lifespan prediction, and maintenance optimization of converter failures, ensuring optimal converter health. This not only improves converter reliability but also enhances the overall availability of renewable energy power generation systems. In the context of high-proportion renewable energy grid integration, improving converter reliability is crucial for increasing renewable energy absorption capacity and achieving friendly grid integration. Application of this invention is expected to increase converter reliability by over 10% and extend mean time to failure (MTTF) by over 15%. This could lead to a 5% increase in annual renewable energy power generation, representing significant economic benefits.

[0170] This invention sets a new benchmark for intelligent operation and maintenance of energy equipment and provides a demonstration for the intelligent development of the energy and power industry.

[0171] This invention is a successful example of digital and intelligent upgrading and transformation in the energy equipment sector, and represents an innovative application of emerging technologies such as big data and artificial intelligence in energy equipment management. On the one hand, based on real-time status perception of energy equipment, this invention utilizes technologies such as the Internet of Things and edge computing to build a data collection system covering the entire equipment lifecycle, breaking through the limitations of traditional offline data silos. On the other hand, through data fusion and intelligent modeling, this invention uses machine learning to achieve online assessment of equipment status, risk warnings, and optimized maintenance decisions, breaking through the bottlenecks of traditional physical models and shifting energy equipment status management from experience-driven to data-driven, and from passive to active management. The comprehensive technical architecture and management model formed by this invention provides new ideas and methods for the intelligent development of the energy and power industry, innovatively constructing an end-to-end closed-loop intelligent system from the perception layer to the data layer to the application layer, and has exemplary significance for promoting the digitization, networking, and intelligence of energy and power systems.

[0172] In summary, the online monitoring method for over-stress failure of grid-connected new energy converters proposed in the present invention is based on the urgent need for equipment reliability management. In response to the shortcomings of traditional monitoring and operation and maintenance models, it innovatively combines emerging technologies such as multi-source heterogeneous data fusion and machine learning with energy equipment management, and has achieved significant technical breakthroughs and application results in converter online monitoring, health assessment, life prediction, and maintenance decision-making. Through the present invention, converter failure accidents can be avoided and reduced to the greatest extent, converter reliability can be improved, and its service life can be extended. While reducing maintenance costs, the efficiency of new energy power generation systems can be improved, the safety and stability of the power grid can be guaranteed, and significant economic and social benefits can be generated. The present invention has made fruitful explorations into the application of multi-source data fusion and machine learning in the field of energy equipment, established a new technical benchmark for the intelligent development of the energy and power industry, and has important demonstration significance for promoting the digital transformation of energy and power systems, and assisting energy revolution and industrial upgrading.

[0173] The present invention uses an offshore wind farm with a 500MW installed capacity as its application target. The farm comprises 100 5MW wind turbines, each equipped with a full-power converter and connected to the grid via a 35kV cable. The wind farm is located in a harsh marine environment, with frequent typhoons, lightning, and other extreme weather conditions. The wind turbine converters are frequently subjected to grid-side overvoltage shocks, resulting in a high risk of insulation failure and an urgent need for condition monitoring and intelligent operations and maintenance.

[0174] Figure 5 The structure diagram of a monitoring system for over-voltage stress failure of a grid-connected new energy converter according to a preferred embodiment of the present invention is shown.

[0175] like Figure 5As shown, the present invention provides a monitoring system for over-voltage stress failure of a grid-connected new energy converter, the system comprising:

[0176] The collection unit 501 is used to collect multi-source heterogeneous monitoring data under the operating conditions of the wind power converter;

[0177] Preferably, the multi-source heterogeneous monitoring data includes: electrical parameters, thermal parameters and environmental parameters;

[0178] Electrical parameters are collected through voltage sensors and current sensors arranged in the main electrical circuit of the wind power converter;

[0179] Thermal parameters are collected by infrared thermometers placed on the surface of the wind turbine converter;

[0180] Environmental parameters are collected through meteorological sensors arranged on the surface of the wind turbine converter.

[0181] The extraction unit 502 is used to extract features from the multi-source heterogeneous monitoring data and fuse the extracted features to obtain fused features of the multi-source heterogeneous monitoring data;

[0182] Preferably, the extraction unit 502 is used to extract features from multi-source heterogeneous monitoring data, and is also used to:

[0183] Perform time-frequency domain analysis on electrical parameters and extract the characteristic amplitude, duration, rising edge rate, and falling edge rate of transient overvoltage based on the time-frequency domain analysis results;

[0184] Perform image processing on thermal parameters and extract the features of the highest temperature point, the area of the highest temperature region, and the average temperature gradient reflecting the thermal stress distribution based on the image processing results;

[0185] Statistical analysis is performed on environmental parameters, and based on the statistical analysis results, the characteristics of wind speed, wind direction, and air pressure that reflect the environmental stress level are extracted.

[0186] Preferably, the extraction unit 502 is used to fuse the extracted features and further to:

[0187] The various features of the extracted multi-source heterogeneous monitoring data are sequentially spliced to generate a high-dimensional matrix. The rows of the high-dimensional matrix represent the sampling time points, and the columns of the high-dimensional matrix represent the features of a class of multi-source heterogeneous monitoring data.

[0188] Based on the JCA algorithm, the high-dimensional matrix is transformed to obtain a set of fusion feature vectors after dimensionality reduction.

[0189] An establishing unit 503 is used to establish a machine learning model of the wind power converter based on the fusion features;

[0190] An evaluation unit 504 is configured to evaluate the health status of the wind power converter based on a machine learning model and obtain a health evaluation result;

[0191] A prediction unit 505 is configured to predict the remaining life of the wind power converter based on a machine learning model to obtain a predicted remaining life;

[0192] The result unit 506 is used to generate an operation and maintenance decision based on the health assessment result and the predicted remaining life, and guide the operation and maintenance of the wind power converter based on the operation and maintenance decision.

[0193] Preferably, the establishing unit 503 is configured to establish a machine learning model for the wind power converter based on the fusion features, and is further configured to:

[0194] Taking the fusion features as input, a health status assessment model for wind power converters is established based on the support vector machine algorithm.

[0195] Taking the historical time series health index as input, a remaining life prediction model for wind turbine converters is established based on long short-term memory neural network.

[0196] A monitoring system for over-electrical stress failure of a grid-connected new energy converter according to a preferred embodiment of the present invention corresponds to a monitoring method for over-electrical stress failure of a grid-connected new energy converter according to another preferred embodiment of the present invention, and will not be described in detail here.

[0197] It will be understood by those skilled in the art that the embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. The solutions in the embodiments of the present invention may be implemented in various computer languages, for example, the object-oriented programming language Java and the interpreted scripting language JavaScript.

[0198] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0199] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0200] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0201] Although the preferred embodiments of the present invention have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0202] Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if such changes and modifications fall within the scope of the claims and their equivalents, the present invention is intended to include such changes and modifications.

[0203] The invention has been described above with reference to a few embodiments. However, it is readily apparent to a person skilled in the art that other embodiments than the ones disclosed above are equally within the scope of the invention, as defined by the appended patent claims.

[0204] Generally, all terms used in the claims are to be interpreted according to their ordinary meaning in the technical field, unless explicitly defined otherwise herein. All references to "a / the [means, component, etc.]" are to be interpreted openly as referring to at least one instance of the means, component, etc., unless explicitly stated otherwise. The steps of any method disclosed herein do not necessarily need to be performed in the exact order disclosed, unless explicitly stated otherwise.

Claims

1. A method for monitoring over-stress failure of a grid-connected new energy converter, the method comprising: Collect multi-source heterogeneous monitoring data under converter operating conditions; Extracting features from the multi-source heterogeneous monitoring data and fusing the extracted features to obtain fused features of the multi-source heterogeneous monitoring data; Establishing a machine learning model of the converter based on the fused features; Evaluate the health status of the converter based on the machine learning model and obtain a health assessment result; Predicting the remaining life of the converter based on the machine learning model to obtain a predicted remaining life; An operation and maintenance decision is generated based on the health assessment result and the predicted remaining life, and operation and maintenance of the converter is guided based on the operation and maintenance decision.

2. The method according to claim 1, wherein the multi-source heterogeneous monitoring data comprises: Electrical parameters, thermal parameters and environmental parameters; The electrical parameters are collected by voltage sensors and current sensors arranged in the main electrical circuit of the converter; The thermal parameters are collected by an infrared thermometer arranged on the surface of the converter; The environmental parameters are collected by meteorological sensors arranged on the surface of the converter.

3. The method according to claim 2, wherein extracting features from the multi-source heterogeneous monitoring data comprises: Performing time-frequency domain analysis on the electrical parameters, and extracting characteristics reflecting the characteristic amplitude, duration, rising edge rate, and falling edge rate of the transient overvoltage based on the time-frequency domain analysis results; performing image processing on the thermal parameters, and extracting features reflecting the highest temperature point, the area of the highest temperature region, and the average temperature gradient of the thermal stress distribution based on the image processing results; Statistical analysis is performed on the environmental parameters, and based on the statistical analysis results, features of wind speed, wind direction, and air pressure that reflect the environmental stress level are extracted.

4. The method according to claim 1, wherein fusing the extracted features comprises: The various features of the extracted multi-source heterogeneous monitoring data are sequentially spliced to generate a high-dimensional matrix, wherein the rows of the high-dimensional matrix represent sampling time points, and the columns of the high-dimensional matrix represent features of a class of multi-source heterogeneous monitoring data; Based on the JCA algorithm, feature transformation is performed on the high-dimensional matrix to obtain a set of fusion feature vectors after dimension reduction.

5. The method according to claim 1, wherein establishing a machine learning model of the converter based on the fusion features comprises: Taking the fusion features as input, establishing a health status assessment model for the converter based on a support vector machine algorithm; Taking the historical time series health index as input, a remaining life prediction model of the converter is established based on the long short-term memory neural network.

6. A monitoring system for over-voltage stress failure of a grid-connected new energy converter, the system comprising: The acquisition unit is used to collect multi-source heterogeneous monitoring data under the operating conditions of the converter; an extraction unit, configured to extract features from the multi-source heterogeneous monitoring data and fuse the extracted features to obtain fused features of the multi-source heterogeneous monitoring data; an establishing unit, configured to establish a machine learning model of the converter based on the fusion features; an evaluation unit, configured to evaluate the health status of the converter based on the machine learning model and obtain a health evaluation result; A prediction unit, configured to predict the remaining life of the converter based on the machine learning model to obtain a predicted remaining life; A result unit is used to generate an operation and maintenance decision based on the health assessment result and the predicted remaining life, and guide the operation and maintenance of the converter based on the operation and maintenance decision.

7. The system according to claim 6, wherein the multi-source heterogeneous monitoring data comprises: Electrical parameters, thermal parameters and environmental parameters; The electrical parameters are collected by voltage sensors and current sensors arranged in the main electrical circuit of the converter; The thermal parameters are collected by an infrared thermometer arranged on the surface of the converter; The environmental parameters are collected by meteorological sensors arranged on the surface of the converter.

8. The system according to claim 7, wherein the extraction unit is configured to extract features from the multi-source heterogeneous monitoring data, and further configured to: Performing time-frequency domain analysis on the electrical parameters, and extracting characteristics reflecting the characteristic amplitude, duration, rising edge rate, and falling edge rate of the transient overvoltage based on the time-frequency domain analysis results; performing image processing on the thermal parameters, and extracting features reflecting the highest temperature point, the area of the highest temperature region, and the average temperature gradient of the thermal stress distribution based on the image processing results; Statistical analysis is performed on the environmental parameters, and based on the statistical analysis results, features of wind speed, wind direction, and air pressure that reflect the environmental stress level are extracted.

9. The system according to claim 6, wherein the extraction unit is configured to fuse the extracted features and further configured to: The various features of the extracted multi-source heterogeneous monitoring data are sequentially spliced to generate a high-dimensional matrix, wherein the rows of the high-dimensional matrix represent sampling time points, and the columns of the high-dimensional matrix represent features of a class of multi-source heterogeneous monitoring data; Based on the JCA algorithm, feature transformation is performed on the high-dimensional matrix to obtain a set of fusion feature vectors after dimension reduction.

10. The system according to claim 6, wherein the establishing unit is configured to establish a machine learning model of the converter based on the fused features, and is further configured to: Taking the fusion features as input, establishing a health status assessment model for the converter based on a support vector machine algorithm; Taking the historical time series health index as input, a remaining life prediction model of the converter is established based on the long short-term memory neural network.

Images