Full-power high-voltage frequency converter and operation monitoring system thereof

Abstract

The invention relates to the technical field of frequency converter operation monitoring, in particular to a full-power high-voltage frequency converter and an operation monitoring system thereof, and the system specifically comprises a frequency conversion management center, a data sensing module, an edge processing module, a multi-dimensional fusion module, a multi-dimensional calibration module, a collaborative evaluation module and a rear-end alarm module. According to the method, serious abnormal scenes are rapidly identified by means of an edge health identification model, timely warning of sudden faults is achieved, meanwhile, potential fault types and fault probabilities of key components are accurately identified through a deep diagnosis model, the full-scene fault monitoring requirements are covered, a component service life coupling model is constructed, and the fault monitoring efficiency is improved. According to the method, the mutual influence among the degraded parts is fully considered, so that the problem of prediction deviation caused by neglecting a part coupling relation in traditional life prediction is solved, the accuracy of residual life prediction is improved, the overall health state of the high-voltage frequency converter is quantitatively evaluated through the comprehensive health index, and a clear decision basis is provided for operation and maintenance personnel.

Description

Technical Field

[0001] This invention relates to the field of frequency converter operation monitoring technology, and in particular to a full-power high-voltage frequency converter and its operation monitoring system. Background Technology

[0002] As a key power conversion device in industrial production, full-power high-voltage frequency converters are widely used in major industrial fields such as metallurgy, chemical industry, and power. The stability of their operation directly determines the continuity and safety of the production process. As industrial production transforms towards intelligence and efficiency, the operating conditions of high-voltage frequency converters are becoming increasingly complex, facing sudden serious faults such as overvoltage, overcurrent, and overtemperature, as well as progressive fault risks such as aging of power module insulation and degradation of cooling system performance.

[0003] Currently, traditional high-voltage frequency converter monitoring systems have the following shortcomings: The data collection dimension is singular, and the multi-source heterogeneous operational data and digital twin simulation data are not fully integrated, resulting in insufficient data support. Furthermore, the fault identification only focuses on severe sudden faults and lacks the ability to accurately diagnose potential progressive faults. The remaining useful life prediction does not take into account the life coupling effect between key components, resulting in insufficient accuracy of the prediction results. At the same time, the health status assessment lacks quantitative indicators, making it difficult to form a basis for targeted operation and maintenance decisions. This can easily lead to problems such as over-maintenance or untimely maintenance, which seriously affects the operational reliability and service life of high-voltage frequency converters.

[0004] To address the aforementioned technical shortcomings, a solution is proposed. Summary of the Invention

[0005] The purpose of this invention is to provide a full-power high-voltage frequency converter and its operation monitoring system. This system provides intelligent monitoring and operation and maintenance support for the high-voltage frequency converter throughout its entire life cycle by effectively integrating multi-source heterogeneous data, rapidly identifying severe abnormal scenarios, accurately diagnosing potential faults, predicting remaining life under coupling relationships, and quantitatively assessing health, thereby solving the aforementioned technical deficiencies.

[0006] The objective of this invention can be achieved through the following technical solution: an operation monitoring system for a full-power high-voltage frequency converter, comprising a frequency converter management center, a data sensing module, an edge processing module, a multi-dimensional fusion module, a multi-dimensional calibration module, a collaborative evaluation module, and a back-end alarm module; The data sensing module is used to collect multi-source heterogeneous operating data during the operation of the high-voltage frequency converter and preprocess it to obtain standard multi-source heterogeneous data, and then send the standard multi-source heterogeneous data to the frequency converter management center for storage. The edge processing module is used to preprocess historical operation data and fault data to build an edge health identification model. Based on the edge health identification model, it performs scene identification and analysis on standard multi-source heterogeneous data and outputs effective feature datasets or alarm signals. The multidimensional fusion module is used to process and fuse effective feature datasets and ideal output data to output multidimensional feature vectors; The multidimensional calibration module is used to generate operating condition labels from the extracted high-voltage frequency converter operating parameters and to complete the dimensional calibration analysis process based on the operating condition labels. The collaborative assessment module is used to identify degradation of multi-dimensional feature vectors and conduct collaborative health assessment feedback analysis. It then performs discrimination processing on the obtained comprehensive health index and outputs alarm operation and maintenance signals or routine monitoring signals.

[0007] Preferably, the analysis process of the edge processing module is as follows: Historical operating data and fault data of the high-voltage frequency converter are retrieved and preprocessed. An edge health identification model is constructed based on the preprocessed historical operating data and fault data. The edge health identification model is used to quickly identify preset severe abnormal scenarios, including overvoltage faults, overcurrent faults and overtemperature faults. The preprocessed standard multi-source heterogeneous data is input into the edge health recognition model to perform preliminary anomaly judgment on the standard multi-source heterogeneous data. If the output is a severe anomaly scenario, an alarm signal is generated. If the output is a non-severe anomaly scenario, features in the standard multi-source heterogeneous data are extracted, and an effective feature dataset is constructed based on the extracted features.

[0008] Preferably, the analysis process of the multidimensional fusion module is as follows: Construct a digital twin model of the high-voltage frequency converter and obtain the ideal output data of the digital twin model; The effective feature dataset and the ideal output data are synchronized with timestamps and standardized in terms of units. If there are non-numerical features in the standardization process, they are converted into numerical values ​​through label encoding. The processed effective feature dataset and the ideal output data are weighted and fused to obtain a multidimensional feature vector.

[0009] Preferably, the weighted fusion process is as follows: T1: Compare the matching degree between the effective feature dataset and the actual fault data of the equipment in the past 3 months (the accuracy of the association between voltage feature values ​​and actual overvoltage faults). Normalize the matching degree to obtain the weight coefficient ω1 of the effective feature dataset. Then the weight coefficient ω2 of the ideal output data is 1-ω1, and the weight must satisfy ω1+ω2=1. T2: Perform weighted fusion on each dimension of the processed effective feature dataset and the ideal output data: Let the standardized value of the i-th dimension of the effective feature dataset be X1i, and the standardized value of the i-th dimension of the digital twin model output data be X2i. After fusion, the feature value of this dimension is Yi = ω1×X1i + ω2×X2i, where i = 1, 2, 3, ..., N, and N is the total number of dimensions. T3: Retrieve the feature sorting table of the effective feature dataset, and combine the sorted N fusion feature values ​​in order to form a multidimensional feature vector of [Y1, Y2, Y3, ..., YN].

[0010] Preferably, the analysis process of the multidimensional calibration module is as follows: TT1: Extract load rate, ambient temperature, and running time from the operating parameters of the high-voltage frequency converter as operating condition dimensions to form operating condition labels; Retrieve multiple sets of historical multidimensional feature vectors that perfectly match the current operating condition labels of the high-voltage frequency converter to form a comparison dataset; TT2: Calculate the mean μH and standard deviation σH for each dimension in the comparison dataset, and then calculate the deviation between the fused value of each dimension of the current multidimensional feature vector and the mean μH. Deviation = |fused value of the current dimension - mean μH| / standard deviation σH; If the deviation of a certain dimension is less than or equal to the preset deviation, then the dimension is considered stable; if the deviation of a certain dimension is greater than the preset deviation, then the dimension is considered to be deviating. TT3: When there is a dimensional deviation, re-collect the effective feature dataset and the ideal output data and perform the weighted fusion process again until all dimensions are stable.

[0011] Preferably, the analysis process of the collaborative evaluation module is as follows: The obtained multidimensional feature vectors are input into a pre-set deep diagnostic model, and the failure probability vectors of each key component of the high-voltage frequency converter are output. The failure probability of each key component is judged one by one to obtain high failure risk and low failure risk. The key component corresponding to high failure risk is set as a degraded component. The preset degradation model corresponding to each degraded component is obtained. The degradation data of the preprocessed degraded component is input into the model, and the remaining time of the degraded component from the current state to the failure threshold is output and set as the independent remaining useful life prediction value SYg, where g represents the degraded component number.

[0012] Preferably, a component life coupling model is constructed: a matrix M is defined, where the row index is the target degraded component g, the column index is the influencing component j, and the matrix element Mij is the coupling influence coefficient agj of the influencing component j on the target degraded component g; The comprehensive influence coefficient Zg of the target degraded component g is obtained based on the component life coupling model; The collaborative remaining useful life is calculated by multiplying the predicted independent remaining useful life of the degraded component (SYg) by the comprehensive influence coefficient (Zg). The remaining useful life (XTg) of all degraded components is summarized, and the preset weight coefficient of each degraded component is obtained. The sum of the remaining useful life (XTg) multiplied by the corresponding preset weight coefficient is set as the comprehensive health index. The comprehensive health index is then processed to obtain alarm maintenance signals or regular monitoring signals.

[0013] The present invention also proposes a full-power high-voltage frequency converter, including a frequency converter body, a protective door hinged inside the front surface of the frequency converter body, and a protective window fixedly connected inside the upper end of the protective door.

[0014] The beneficial effects of this invention are as follows: This invention utilizes an edge health recognition model to quickly identify severe abnormal scenarios such as overvoltage, overcurrent, and overtemperature, enabling timely alarms for sudden faults. Simultaneously, it employs a deep diagnostic model to accurately identify the potential fault types and probabilities of each key component, covering fault monitoring needs across all scenarios and improving the comprehensiveness and accuracy of fault identification.

[0015] This invention also constructs a component life coupling model, which fully considers the mutual influence between various degraded components. Based on the independent remaining useful life and the comprehensive influence coefficient, it calculates the collaborative remaining useful life to solve the prediction deviation problem caused by the neglect of component coupling relationship in traditional life prediction, thereby improving the accuracy of remaining life prediction. Finally, it uses a comprehensive health index to quantitatively assess the overall health status of the high-voltage frequency converter, generating targeted alarm operation and maintenance signals or routine monitoring signals to provide clear decision-making basis for operation and maintenance personnel, realize intelligent and precise operation and maintenance work, effectively reduce operation and maintenance costs, and extend the service life of the high-voltage frequency converter. Attached Figure Description

[0016] The invention will now be further described with reference to the accompanying drawings; Figure 1 This is a three-dimensional view of the structure of the present invention; Figure 2 This is a flowchart of the system of the present invention; Figure 3 This is a reference diagram for multidimensional eigenvector analysis; Figure 4 This is a reference diagram for the analysis of the multidimensional calibration module of this invention.

[0017] Illustration: 1. Inverter body; 2. Protective door; 3. Protective window; Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the invention. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments; Example 1: Please refer to Figures 1 to 4 As shown, the present invention is a full-power high-voltage frequency converter, including a frequency converter body 1, a protective door 2 hinged inside the front surface of the frequency converter body 1, and a protective window 3 fixedly connected inside the upper end of the protective door 2. A monitoring system for the operation of a full-power high-voltage frequency converter includes a frequency converter management center, a data sensing module, an edge processing module, a multi-dimensional fusion module, a multi-dimensional calibration module, a collaborative evaluation module, and a back-end alarm module. The frequency converter management center has a one-way communication connection with the data sensing module, the data sensing module has a one-way communication connection with the edge processing module, the edge processing module has a one-way communication connection with the frequency converter management center, the edge processing module has a one-way communication connection with both the multi-dimensional fusion module and the collaborative evaluation module, the multi-dimensional fusion module has a one-way communication connection with the multi-dimensional calibration module, the multi-dimensional calibration module has a one-way communication connection with the frequency converter management center, and the collaborative evaluation module has a one-way communication connection with the back-end alarm module. The data sensing module is used to collect multi-source heterogeneous operating data during the operation of the high-voltage frequency converter and preprocess it to obtain standard multi-source heterogeneous data, and then send the standard multi-source heterogeneous data to the frequency converter management center for storage. This involves collecting multi-source heterogeneous operating data during the operation of the high-voltage frequency converter, including operating voltage, operating temperature, etc. The collected multi-source heterogeneous operational data is preprocessed, including cleaning and noise reduction, to obtain standard multi-source heterogeneous data. The edge processing module is used to preprocess historical operational and fault data to build an edge health identification model. Based on the edge health identification model, it performs scene identification and analysis on standard multi-source heterogeneous data, and outputs effective feature datasets or alarm signals, specifically including: Historical operating data and fault data of the high-voltage frequency converter are retrieved and preprocessed. An edge health identification model is constructed based on the preprocessed historical operating data and fault data. The edge health identification model is used to quickly identify preset severe abnormal scenarios, including overvoltage faults, overcurrent faults and overtemperature faults. The preprocessed standard multi-source heterogeneous data is input into the edge health recognition model to perform preliminary anomaly judgment. If the output indicates a severe anomaly, an alarm signal is generated. The backend alarm module responds to the alarm signal and immediately performs the preset early warning operation corresponding to the alarm signal, so as to timely manage the anomalies of the high-voltage frequency converter. If the output is a non-severe abnormal scenario, features are extracted from the standard multi-source heterogeneous data. These features include current fluctuation variance, temperature trend slope, etc. An effective feature dataset is constructed based on the extracted features and sent to the frequency converter management center for storage.

[0020] Example 2: The multidimensional fusion module is used to process and fuse the effective feature dataset and the ideal output data, outputting a multidimensional feature vector, specifically including: By constructing a digital twin model of a high-voltage frequency converter using existing digital twin technology, the ideal output data of the digital twin model can be obtained. The ideal output data includes simulated voltage deviation, simulated current loss, etc. The effective feature dataset and the ideal output data are processed by timestamp synchronization and unit unification standardization. If there are non-numerical features in the unit unification standardization process (such as "temperature risk level: low / medium / high" output by the twin model), they are converted into numerical values ​​through label encoding, such as low=1, medium=2, high=3. The processed effective feature dataset and the ideal output data are then weighted and fused. The specific steps are as follows: T1: Compare the matching degree between the effective feature dataset and the actual fault data of the equipment in the past 3 months (such as the accuracy of the correlation between voltage feature values ​​and actual overvoltage faults), and normalize the matching degree to obtain the weight coefficient ω1 of the effective feature dataset. Then the weight coefficient ω2 of the ideal output data is 1-ω1, and the weight must satisfy ω1+ω2=1. T2: Perform weighted fusion on each dimension of the processed effective feature dataset and the ideal output data: Let the standardized value of the i-th dimension of the effective feature dataset be X1i, and the standardized value of the i-th dimension of the digital twin model output data be X2i. After fusion, the feature value of this dimension is Yi = ω1×X1i + ω2×X2i, where i = 1, 2, 3, ..., N, and N is the total number of dimensions. T3: Retrieve the feature sorting table of the effective feature dataset, and combine the sorted N fusion feature values ​​in order to form a multidimensional feature vector of [Y1, Y2, Y3, ..., YN]. The multidimensional calibration module is used to generate operating condition labels from the extracted high-voltage frequency converter operating parameters, and to complete the dimensional calibration analysis process based on these labels. Specifically, this includes: TT1: Extract load rate, ambient temperature, and running time from the operating parameters of the high-voltage frequency converter as operating condition dimensions to form operating condition labels; For example: load rate 80%-90%, ambient temperature 25-30℃, operating time 1000-1200h; Retrieve multiple sets of historical multidimensional feature vectors that perfectly match the current operating condition labels of the high-voltage frequency converter to form a comparison dataset; TT2: Calculate the mean μH and standard deviation σH for each dimension in the comparison dataset, and then calculate the deviation between the fused value of each dimension of the current multidimensional feature vector and the mean μH. Deviation = |fused value of the current dimension - mean μH| / standard deviation σH; If the deviation of a certain dimension is less than or equal to the preset deviation, then the dimension is considered stable; if the deviation of a certain dimension is greater than the preset deviation, then the dimension is considered to be deviating. TT3: When there is a dimensional deviation, the effective feature dataset and ideal output data are collected again and the weighted fusion process is executed again until all dimensions are stable. The multidimensional feature vector is then sent to the frequency converter management center for storage. The collaborative assessment module is used to identify degradation in multi-dimensional feature vectors and perform collaborative health assessment feedback analysis. It then processes the resulting comprehensive health index and outputs alarm and maintenance signals or routine monitoring signals, specifically including: The obtained multidimensional feature vector is input into the pre-set deep diagnostic model, and the fault probability vector of each key component of the high voltage frequency converter (such as power module and cooling system) is output. The fault probability vector represents the key component name-fault type-fault probability. For example: power module-insulation aging fault-probability is 0.85. The failure probability of each key component is judged and processed one by one. If the failure probability exceeds the preset failure probability, it is judged as high failure risk; if the failure probability does not exceed the preset failure probability, it is judged as low failure risk. The critical components corresponding to high failure risk are set as degraded components. The preset degradation model corresponding to each degraded component is obtained. The degradation data of the preprocessed degraded components (such as the insulation resistance of the degraded component: insulation component) is input into the model. The remaining time of the degraded component from the current state to the failure threshold is output and set as the independent remaining useful life prediction value SYg, where g represents the degraded component number. Construct a component lifespan coupling model: Define a matrix M, where the row index is "target degraded component g" (such as power unit P, heat dissipation system S, control module C, filter component F), the column index is "affected component j", and the matrix element Mij is the coupling influence coefficient agj of affected component j on target degraded component g; The comprehensive influence coefficient Zg of the target degraded component g is obtained based on the component life coupling model; The collaborative remaining useful life is calculated by multiplying the predicted independent remaining useful life of the degraded component (SYg) by the comprehensive influence coefficient (Zg). The remaining useful life XTg of all degraded components is summarized, and the preset weight coefficient of each degraded component is obtained. The sum of the remaining useful life XTg multiplied by the corresponding preset weight coefficient is set as the comprehensive health index. The comprehensive health index is judged. If the comprehensive health index is less than the preset comprehensive health index threshold, an alarm operation and maintenance signal is generated. If the comprehensive health index is greater than or equal to the preset comprehensive health index threshold, a regular monitoring signal is generated. The back-end alarm module is used to respond to alarm operation and maintenance messages or regular monitoring signals, and immediately generate the preset warning text corresponding to the alarm operation and maintenance messages or regular monitoring signals, so as to provide timely feedback information for reasonable and targeted management of high-voltage frequency converters; By considering the life-cycle coupling effect between components, the limitations of traditional independent life-cycle prediction are overcome, making the remaining useful life prediction more consistent with the actual operating status of the frequency converter. Furthermore, through real-time monitoring and precise maintenance throughout the entire life cycle, the probability of sudden failure of the frequency converter is effectively reduced, and the service life of the equipment is extended. In summary, the ideal output data from the high-voltage frequency converter is achieved by integrating multi-source heterogeneous operating data, historical data, and digital twin model data through the data acquisition module. Combined with preprocessing methods such as cleaning, noise reduction, and tag encoding, the data is standardized and complete, providing solid data support for subsequent monitoring and analysis. Furthermore, the edge health recognition model can quickly identify severe abnormal scenarios such as overvoltage, overcurrent, and overtemperature, enabling timely alarms for sudden faults. At the same time, the deep diagnostic model can accurately identify the potential fault types and probabilities of each key component, covering the fault monitoring needs of the entire scenario and improving the comprehensiveness and accuracy of fault identification. Furthermore, a component life coupling model is constructed to fully consider the mutual influence between various degraded components. Based on the independent remaining useful life and the comprehensive influence coefficient, the collaborative remaining useful life is calculated to solve the prediction bias problem caused by the neglect of component coupling relationship in traditional life prediction, thereby improving the accuracy of remaining life prediction. Finally, the overall health status of the high-voltage frequency converter is quantitatively evaluated through the comprehensive health index, generating targeted alarm operation and maintenance signals or routine monitoring signals to provide clear decision-making basis for operation and maintenance personnel, realize intelligent and precise operation and maintenance work, effectively reduce operation and maintenance costs, and extend the service life of the high-voltage frequency converter.

[0021] The threshold is set for comparative analysis of results to determine whether they are good or bad. The value of the threshold is determined by a combination of large-scale model analysis of sample data and human experience. It can also be adjusted appropriately based on seasonal or common-sense influencing factors.

[0022] The above are merely preferred embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A monitoring system for the operation of a full-power high-voltage frequency converter, characterized in that, It includes a frequency converter management center, a data sensing module, an edge processing module, a multi-dimensional fusion module, a multi-dimensional calibration module, a collaborative evaluation module, and a back-end alarm module; The data sensing module is used to collect multi-source heterogeneous operating data during the operation of the high-voltage frequency converter and preprocess it to obtain standard multi-source heterogeneous data, and then send the standard multi-source heterogeneous data to the frequency converter management center for storage. The edge processing module is used to preprocess historical operation data and fault data to build an edge health identification model. Based on the edge health identification model, it performs scene identification and analysis on standard multi-source heterogeneous data and outputs effective feature datasets or alarm signals. The multidimensional fusion module is used to process and fuse effective feature datasets and ideal output data to output multidimensional feature vectors; The multidimensional calibration module is used to generate operating condition labels from the extracted high-voltage frequency converter operating parameters and to complete the dimensional calibration analysis process based on the operating condition labels. The collaborative assessment module is used to identify degradation of multi-dimensional feature vectors and conduct collaborative health assessment feedback analysis. It then performs discrimination processing on the obtained comprehensive health index and outputs alarm operation and maintenance signals or routine monitoring signals.

2. The operation monitoring system for a full-power high-voltage frequency converter according to claim 1, characterized in that, The analysis process of the edge processing module is as follows: Historical operating data and fault data of the high-voltage frequency converter are retrieved and preprocessed. An edge health identification model is constructed based on the preprocessed historical operating data and fault data. The edge health identification model is used to quickly identify preset severe abnormal scenarios, including overvoltage faults, overcurrent faults and overtemperature faults. The preprocessed standard multi-source heterogeneous data is input into the edge health recognition model to perform preliminary anomaly judgment on the standard multi-source heterogeneous data. If the output is a severe anomaly scenario, an alarm signal is generated. If the output is a non-severe anomaly scenario, features in the standard multi-source heterogeneous data are extracted, and an effective feature dataset is constructed based on the extracted features.

3. The operation monitoring system for a full-power high-voltage frequency converter according to claim 1, characterized in that, The analysis process of the multidimensional fusion module is as follows: Construct a digital twin model of the high-voltage frequency converter and obtain the ideal output data of the digital twin model; The effective feature dataset and the ideal output data are synchronized with timestamps and standardized in terms of units. If there are non-numerical features in the standardization process, they are converted into numerical values ​​through label encoding. The processed effective feature dataset and the ideal output data are weighted and fused to obtain a multidimensional feature vector.

4. The operation monitoring system for a full-power high-voltage frequency converter according to claim 3, characterized in that, The weighted fusion process is as follows: T1: Compare the matching degree between the effective feature dataset and the actual fault data of the equipment in the past 3 months (the accuracy of the association between voltage feature values ​​and actual overvoltage faults). Normalize the matching degree to obtain the weight coefficient ω1 of the effective feature dataset. Then the weight coefficient ω2 of the ideal output data is 1-ω1, and the weight must satisfy ω1+ω2=1. T2: Perform weighted fusion on each dimension of the processed effective feature dataset and the ideal output data: Let the standardized value of the i-th dimension of the effective feature dataset be X1i, and the standardized value of the i-th dimension of the digital twin model output data be X2i. After fusion, the feature value of this dimension is Yi = ω1×X1i + ω2×X2i, where i = 1, 2, 3, ..., N, and N is the total number of dimensions. T3: Retrieve the feature sorting table of the effective feature dataset, and combine the sorted N fusion feature values ​​in order to form a multidimensional feature vector of [Y1, Y2, Y3, ..., YN].

5. The operation monitoring system for a full-power high-voltage frequency converter according to claim 1, characterized in that, The analysis process of the multidimensional calibration module is as follows: TT1: Extract load rate, ambient temperature, and running time from the operating parameters of the high-voltage frequency converter as operating condition dimensions to form operating condition labels; Retrieve multiple sets of historical multidimensional feature vectors that perfectly match the current operating condition labels of the high-voltage frequency converter to form a comparison dataset; TT2: Calculate the mean μH and standard deviation σH for each dimension in the comparison dataset, and then calculate the deviation between the fused value of each dimension of the current multidimensional feature vector and the mean μH. Deviation = |fused value of the current dimension - mean μH| / standard deviation σH; If the deviation of a certain dimension is less than or equal to the preset deviation, then the dimension is considered stable; if the deviation of a certain dimension is greater than the preset deviation, then the dimension is considered to be deviating. TT3: When there is a dimensional deviation, re-collect the effective feature dataset and the ideal output data and perform the weighted fusion process again until all dimensions are stable.

6. The operation monitoring system for a full-power high-voltage frequency converter according to claim 1, characterized in that, The analysis process of the collaborative evaluation module is as follows: The obtained multidimensional feature vectors are input into a pre-set deep diagnostic model, and the failure probability vectors of each key component of the high-voltage frequency converter are output. The failure probability of each key component is judged one by one to obtain high failure risk and low failure risk. The key component corresponding to high failure risk is set as a degraded component. The preset degradation model corresponding to each degraded component is obtained. The degradation data of the preprocessed degraded component is input into the model, and the remaining time of the degraded component from the current state to the failure threshold is output and set as the independent remaining useful life prediction value SYg, where g represents the degraded component number.

7. The operation monitoring system for a full-power high-voltage frequency converter according to claim 6, characterized in that, Construct a component life coupling model: Define a matrix M, where the row index is the target degraded component g, the column index is the influencing component j, and the matrix element Mij is the coupling influence coefficient agj of the influencing component j on the target degraded component g; The comprehensive influence coefficient Zg of the target degraded component g is obtained based on the component life coupling model; The collaborative remaining useful life is calculated by multiplying the predicted independent remaining useful life of the degraded component (SYg) by the comprehensive influence coefficient (Zg). The remaining useful life (XTg) of all degraded components is summarized, and the preset weight coefficient of each degraded component is obtained. The sum of the remaining useful life (XTg) multiplied by the corresponding preset weight coefficient is set as the comprehensive health index. The comprehensive health index is then processed to obtain alarm maintenance signals or regular monitoring signals.

8. A full-power high-voltage frequency converter, applied to an operation monitoring system for a full-power high-voltage frequency converter as described in any one of claims 1-7, characterized in that, Includes a frequency converter body (1), with a protective door (2) hinged inside the front surface of the frequency converter body (1), and a protective window (3) fixedly connected inside the upper end of the protective door (2).

Images