Digital twin reduced-order model abnormal value repairing method and device based on point cloud recognition, and digital twin model of traction system

Abstract

The invention discloses a digital twinning reduced-order model abnormal value restoration method and device based on point cloud recognition and a digital twinning model of a traction system, and the method comprises the steps: screening point cloud sets of different working condition data, and obtaining a point cloud set of abnormal working condition data; carrying out abnormal value detection on the physical field parameters, and defining the physical field parameters exceeding the range of the fence as first abnormal values; the serial numbers of the point cloud sets of the abnormal working condition data are reordered, downsampling is carried out on the ordered point cloud sets of different voxel units, and a downsampled point cloud set composed of the first points of the different voxel units is obtained; establishing a first abnormal domain based on the boundary of the first voxel unit to which the first abnormal value belongs, and obtaining an adjacent abnormal domain; obtaining a second voxel unit based on the first voxel unit and a voxel unit in an adjacent abnormal domain; obtaining a point cloud abnormal value repairing area, and repairing the abnormal value in the second voxel unit to obtain a repaired point cloud set; and generating a corresponding order reduction model.

Description

Technical Field

[0001] The present application relates to technical fields such as point cloud processing technology and digital twins, and in particular to a method and device for repairing outliers in a digital twin reduced-order model based on point cloud recognition, and a digital twin model of a traction system. Background Art

[0002] In recent years, digital twin technology, which integrates multi-disciplinary, multi-dimensional, multi-physical quantity and multi-level simulation methods and research applications, has been widely used in many fields such as health management of production equipment and systems, product design and manufacturing, because it can use digital twin models in digital space to map the full life cycle status of physical equipment or systems in physical space.

[0003] Building digital twins and key component status prediction models based on multi-physical parameters of physical space and deep learning, realizing real-time status visualization and remaining life prediction of traction equipment in the rail transit field, and providing decision-making guidance for the operation and maintenance of subsequent products have become the key to saving product costs and improving market competitiveness.

[0004] End-to-end deep learning models utilize data-driven neural network algorithms to train on historical operating data and simulation datasets for motors and converters. By stacking multiple nonlinear intermediate layers, they model the nonlinear relationships in the data and extract useful information to effectively predict and diagnose device failures. Therefore, good training model data is a prerequisite for building accurate and reliable digital twin models.

[0005] Unreasonable values ​​in a dataset are handled using outlier detection and processing methods. Outlier detection methods include simple statistical analysis, the 3σ principle, box plots, and clustering, while outlier processing methods include deletion, interpolation, and data replacement. However, there are no fast and accurate algorithms for handling the effects of outliers on neighboring points (points that do not deviate from the normal range but differ from the surrounding points). Summary of the Invention

[0006] The present application provides a method and device for repairing outliers in a digital twin reduced-order model based on point cloud recognition, and a digital twin model of a traction system.

[0007] The technical solution of this application is achieved as follows:

[0008] In a first aspect, a method for repairing outliers in a digital twin reduced-order model based on point cloud recognition is provided, the method comprising:

[0009] Establishing a finite element simulation model of the traction system, using different working condition data as input to the finite element simulation model, and obtaining a point cloud set containing the three-dimensional spatial coordinates and physical field parameters of the traction system corresponding to the different working condition data;

[0010] Using the constructed physical field anomaly criterion model, based on the preset physical field parameter threshold, the point cloud sets of different working condition data are screened to obtain the point cloud sets of abnormal working condition data;

[0011] Using an interquartile range algorithm, performing outlier detection processing on the physical field parameters of the point cloud set of the abnormal working condition data, and defining the physical field parameters that exceed the fence range as the first outlier;

[0012] Using a point cloud voxelization grid filtering algorithm, based on the comprehensive voxel index of the point cloud set of the abnormal operating condition data, the initial sequence number of the point cloud set of the abnormal operating condition data is reordered, and then downsampling is performed on the sorted point cloud sets corresponding to different voxel units to obtain a downsampled point cloud set composed of the first points of different voxel units; wherein the downsampled point cloud set does not include the first outlier, and the comprehensive voxel index is indexed in the coordinate dimension with the smallest rate of change of the physical field, with the highest priority.

[0013] determining a first voxel unit to which the first outlier belongs;

[0014] Establishing a first abnormal domain of the largest continuous enclosing area based on the boundary of the first voxel unit, and obtaining a neighboring abnormal domain affected by the first abnormal value by using the constructed neighboring abnormal value influence domain determination function;

[0015] Based on the first voxel unit and the voxel units in the adjacent abnormal domain, obtaining a second voxel unit containing all abnormal values;

[0016] Based on the first abnormal domain, using the constructed abnormal value repair domain judgment function, a point cloud abnormal value repair area composed of third voxel units is obtained;

[0017] Repairing the outliers in the second voxel unit based on the physical field parameters of the point cloud outlier repair area to obtain a point cloud set of repaired abnormal working condition data;

[0018] Based on the repaired point cloud set and the point cloud set of normal operating data, a corresponding reduced-order model is generated to achieve full-field visual monitoring of the traction system under different line operation scenarios.

[0019] In a second aspect, a device for repairing outliers in a digital twin reduced-order model based on point cloud recognition is provided, the device comprising:

[0020] A point cloud extraction module is used to establish a finite element simulation model of the traction system, take different working condition data as input to the finite element simulation model, and obtain a point cloud set containing the three-dimensional spatial coordinates and physical field parameters of the traction system corresponding to the different working condition data;

[0021] A screening module is used to screen point cloud sets of different working condition data based on a preset physical field parameter threshold using the constructed physical field anomaly criterion model to obtain a point cloud set of abnormal working condition data;

[0022] an outlier detection and processing module, configured to perform outlier detection processing on the physical field parameters of the point cloud set of the abnormal working condition data using an interquartile range algorithm, and define the physical field parameters that exceed the fence range as first outliers;

[0023] a point cloud voxelization grid filtering module, configured to utilize a point cloud voxelization grid filtering algorithm to reorder the initial sequence number of the point cloud set of the abnormal operating condition data based on the comprehensive voxel index of the point cloud set of the abnormal operating condition data, and then perform downsampling processing on the sorted point cloud sets corresponding to different voxel units to obtain a downsampled point cloud set consisting of the first points of different voxel units; wherein the downsampled point cloud set does not include the first abnormal value, and the comprehensive voxel index is indexed in the coordinate dimension with the smallest rate of change of the physical field, with the highest priority.

[0024] A processing module, configured to determine a first voxel unit to which the first abnormal value belongs;

[0025] a neighboring outlier influence domain determination module, configured to establish a first outlier domain of the largest continuous enclosing area based on the boundary of the first voxel unit, and obtain a neighboring outlier domain affected by the first outlier using a constructed neighboring outlier influence domain determination function;

[0026] The processing module is further configured to obtain a second voxel unit containing all abnormal values ​​based on the first voxel unit and the voxel units in the adjacent abnormal domain;

[0027] an outlier repair domain determination module, configured to obtain, based on the first outlier domain and using a constructed outlier repair domain determination function, a point cloud outlier repair region composed of a third voxel unit;

[0028] an outlier repair module, configured to repair the outliers in the second voxel unit based on the physical field parameters of the point cloud outlier repair region, and obtain a point cloud set of repaired abnormal working condition data;

[0029] The reduced-order model generation module is used to generate a corresponding reduced-order model based on the repaired point cloud set and the point cloud set of normal operating data, so as to realize full-field visual monitoring of the traction system under different line operation scenarios.

[0030] In a third aspect, a digital twin model of a traction system is provided, characterized in that the digital twin model includes: a vehicle dynamics model, a lumped parameter equivalent digital circuit model, and a multi-field reduced-order model;

[0031] The vehicle dynamics model is at least used to simulate and obtain a first result; the first result includes: motor speeds corresponding to different motor torques, and the first result is transmitted to the lumped parameter equivalent digital circuit model;

[0032] The lumped parameter equivalent digital circuit model is at least used to simulate and obtain a second result based on the first result, and transmit the second result to the multi-field reduced-order model; the second result includes one or more of the following results: motor-related electrical parameters, converter-related electrical parameters, and converter electromagnetic losses; the converter electromagnetic losses include one or more of the following losses: insulated gate bipolar transistor (IGBT) losses, reactor losses, transformer losses, copper bus losses, and busbar losses;

[0033] The multi-field reduced-order model is at least used to simulate and obtain a third result based on the second result; the third result includes one or more of the following results: motor electromagnetic loss, motor temperature, converter temperature, and system structural stress; wherein, the outlier repair processing operation used in the establishment of the multi-field reduced-order model is obtained through the method of the first aspect.

[0034] In a fourth aspect, an electronic device is provided, comprising: a processor and a memory configured to store a computer program that can be run on the processor, wherein the processor is configured to execute the steps of the method of the first aspect when running the computer program.

[0035] In a fifth aspect, a computer-readable storage medium is provided, on which a computer program is stored, wherein the computer program implements the steps of the method of the first aspect when executed by a processor.

[0036] In a sixth aspect, a computer program product is provided, comprising a computer program, wherein the computer program implements the steps of the method of the first aspect when executed by a processor. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 Schematic diagram of the process of repairing outliers in the digital twin reduced-order model based on point cloud recognition in the embodiment of this application Figure 1 ;

[0038] Figure 2 This is a schematic diagram of the representation of a point cloud set after voxel gridding as exemplified in an embodiment of the present application;

[0039] Figure 3 Schematic diagram of the process of repairing outliers in the digital twin reduced-order model based on point cloud recognition in the embodiment of this application Figure 2 ;

[0040] Figure 4 Schematic diagram of the process of repairing outliers in the digital twin reduced-order model based on point cloud recognition in the embodiment of this application Figure 3 ;

[0041] Figure 5 Schematic diagram of the process of repairing outliers in the digital twin reduced-order model based on point cloud recognition in the embodiment of this application Figure 4 ;

[0042] Figure 6 This is a schematic diagram of determining a first voxel unit as exemplified in an embodiment of the present application;

[0043] Figure 7 Schematic diagram of the process of repairing outliers in the digital twin reduced-order model based on point cloud recognition in the embodiment of this application Figure 5 ;

[0044] Figure 8 Schematic diagram of the process of repairing outliers in the digital twin reduced-order model based on point cloud recognition in the embodiment of this application Figure 6 ;

[0045] Figure 9 Schematic diagram of the process of repairing outliers in the digital twin reduced-order model based on point cloud recognition in the embodiment of this application Figure 7 ;

[0046] Figure 10 Schematic diagram of the process of repairing outliers in the digital twin reduced-order model based on point cloud recognition in the embodiment of this application Figure 8 ;

[0047] Figure 11 This is a structural diagram of a certain type of motor in a traction system as exemplified in an embodiment of the present application;

[0048] Figure 12 This is a schematic diagram of a point cloud set after filtering and downsampling of a certain type of motor in a traction system as exemplified in an embodiment of the present application;

[0049] Figure 13 This is a repair process for a certain type of motor in a traction system as exemplified in an embodiment of the present application;

[0050] Figure 14 Schematic diagram of the structure of the device for repairing outliers in a digital twin reduced-order model based on point cloud recognition in an embodiment of the present application;

[0051] Figure 15This is a structural diagram of a digital twin model of a traction system in an embodiment of the present application;

[0052] Figure 16 This is a schematic diagram of the structure of the electronic device in the embodiment of the present application. DETAILED DESCRIPTION

[0053] In order to enable a more detailed understanding of the features and technical contents of the embodiments of the present application, the implementation of the embodiments of the present application is described in detail below with reference to the accompanying drawings. The attached drawings are for reference only and are not used to limit the embodiments of the present application.

[0054] The present invention provides a method for repairing outliers in a digital twin reduced-order model based on point cloud recognition. Figure 1 Schematic diagram of the process of repairing outliers in the digital twin reduced-order model based on point cloud recognition in the embodiment of this application Figure 1 , used in electronic devices such as Figure 1 As shown in FIG, the method for repairing outliers in a digital twin reduced-order model based on point cloud recognition includes the following steps:

[0055] S101: Screening point cloud sets of data of different working conditions to obtain point cloud sets of data of abnormal working conditions.

[0056] Before executing S101, the following steps are included:

[0057] A finite element simulation model of the traction system is established, and different working condition data are used as input of the finite element simulation model to obtain a point cloud set containing the three-dimensional spatial coordinates and physical field parameters of the traction system corresponding to the different working condition data.

[0058] In an embodiment of the present application, computational fluid dynamics (CFD) simulation software is used to convert multiple sets of operating condition data collected from actual vehicle operation into boundary excitations under different input conditions. First, it is necessary to perform grid independence verification and convergence test on the finite element simulation model of the traction system to ensure that the simulation results of the finite element simulation model at the grid resolution tend to be stable and convergent. Then, the finite element simulation model is simulated and analyzed under different operating conditions to extract a point cloud set containing the three-dimensional spatial coordinates and physical field parameters of the traction system corresponding to the different operating condition data.

[0059] The physical field parameters include but are not limited to temperature field, flow field and stress field.

[0060] S101 may include the following steps: using the constructed physical field anomaly criterion model, based on a preset physical field parameter threshold, screening point cloud sets of different working condition data to obtain a point cloud set of abnormal working condition data.

[0061] It should be noted that the benchmark physical field parameter threshold under extreme operating conditions, that is, the preset physical field parameter threshold, can be determined based on the operating parameters and historical data analysis of components such as motors and converters in the traction system.

[0062] Based on this, the physical field parameters corresponding to the point cloud sets of different operating condition data are compared with the preset physical field parameter thresholds. If the physical field parameters of a point cloud set of operating condition data exceed the preset physical field parameter threshold, it is determined to be a point cloud set of abnormal operating condition data. It should be noted that when screening point cloud sets of multiple operating condition data, at least one point cloud set of abnormal operating condition data can be screened out.

[0063] S102: Perform outlier detection on the physical field parameters, and define the physical field parameters that exceed the fence range as the first outlier.

[0064] S102 may include the following steps: using an interquartile range algorithm to perform outlier detection processing on the physical field parameters of the point cloud set of abnormal working condition data, and defining the physical field parameters that exceed the fence range as the first outlier.

[0065] It's important to note that the interquartile range (IQR) algorithm is a statistical method used to measure the degree of dispersion in data. It calculates the difference between the third quartile (QU) and the first quartile (QL) of a dataset to reflect the distribution range of the middle 50% of the data. Quartiles refer to the values ​​at the three dividing points of a set of data, arranged in ascending order, and then divided into four equal parts. The first quartile (QL), also known as the lower quartile, indicates that at least 25% of the data is less than or equal to it; the second quartile is the median, indicating that at least 50% of the data is less than or equal to it; and the third quartile (QU), also known as the upper quartile, indicates that at least 75% of the data is less than or equal to it. The interquartile range is calculated as IQR = QU - QL, and it reflects the degree of dispersion in the middle 50% of the data. A larger IQR indicates a more dispersed data set; a smaller IQR indicates a more concentrated data set.

[0066] The interquartile range algorithm can be expressed as: y = f(QL, QU, IQR, k), where QL is the lower quartile, QU is the upper quartile, IQR is the interquartile range, and k is an adjustment coefficient, which is set based on the operating condition data corresponding to the traction system components. k can take values ​​of 1.5, 2, or other values, which are not limited in this embodiment of the application. Physical field parameters that are less than QL-k*IQR or greater than QU+k*IQR are defined as extreme outliers, namely, first outliers.

[0067] Based on this, the point cloud set of the abnormal operating condition data is sorted in ascending order based on its initial sequence number to obtain a sorted point cloud set. Furthermore, the upper quartile QU and lower quartile QL are determined using the interquartile range algorithm. Ultimately, physical field parameters that are less than QL - k*IQR or greater than QU + k*IQR are defined as extreme outliers, or first outliers.

[0068] For example, the data set is {3, 7, 8, 5, 12, 14, 21, 13, 18}, and after sorting, it is {3, 5, 7, 8, 12, 13, 14, 18, 21}. For a data set with n data points, the position of the lower quartile QL is (n+1)×0.25, and the position of the upper quartile QU is (n+1)×0.75. In the above example, n=9, then the position of the lower quartile QL is (9+1)×0.25=2.5, and the position of the upper quartile QU is (9+1)×0.75=7.5. When the position of the quartile is an integer, the value corresponding to the position is the corresponding quartile. When the position of the quartile is not an integer, interpolation calculation is required. For example, the position of the lower quartile QL is 2.5, indicating that the lower quartile QL is between the second and third data points, and the value of the lower quartile QL is the average of these two data points, that is, lower quartile QL = (5 + 7) / 2 = 6. Similarly, the position of the upper quartile QU is 7.5, indicating that the upper quartile QU is between the seventh and eighth data points, and upper quartile QU = (14 + 18) / 2 = 16. According to the formula IQR = upper quartile QU - lower quartile QL, we can obtain IQR = 16 - 6 = 10. Therefore, physical field parameters less than 6 - k * 10 or greater than 10 + k * 10 are defined as first outliers.

[0069] The number of the first outlier is at least one.

[0070] S103: reordering the sequence numbers of the point cloud sets of the abnormal working condition data, and downsampling the ordered point cloud sets of different voxel units to obtain downsampled point cloud sets composed of the first points of different voxel units.

[0071] S103 may include the following steps: using a point cloud voxelization grid filtering algorithm, based on the comprehensive voxel index of the point cloud set of abnormal working condition data, reordering the initial serial number of the point cloud set of abnormal working condition data, and then performing downsampling processing on the sorted point cloud sets corresponding to different voxel units to obtain a downsampled point cloud set composed of the first points of different voxel units; wherein the downsampled point cloud set does not include the first abnormal value, and the comprehensive voxel index is expanded in the order of highest priority along the coordinate dimension with the smallest rate of change of the physical field.

[0072] In the embodiment of the present application, the point cloud voxelization grid filtering optimization algorithm can be expressed as: y = f (sort H ,filter,H,R voxel ,u'). First, use the sorting algorithm sort H , based on the value of the comprehensive voxel index H, the unordered initial serial numbers corresponding to the point cloud set g(x,y,z,u)abnormal-i of the abnormal working condition data are sorted in ascending order to obtain the sorted point cloud set g(x,y,z,u)abnormal-i-sort of the abnormal working condition data. Secondly, the filtering algorithm filter is used to replace all the points in the corresponding voxel grid (unit) with the first point in the different values ​​of the above-mentioned comprehensive voxel index H to obtain the downsampled point cloud set g(x,y,z,u)abnormal-i-sort-sub. The downsampled point cloud set g(x,y,z,u)abnormal-i-sort-sub does not include the first abnormal value. The comprehensive voxel index H is indexed in the highest priority order along the coordinate dimension with the smallest rate of change of the physical field. The comprehensive voxel index H is based on the voxel resolution R voxel Calculated.

[0073] like Figure 2 The figure shows an example of how a point cloud set may be represented after being voxelized into a grid.

[0074] S104: Establish a first abnormal domain based on the boundary of the first voxel unit to which the first abnormal value belongs, and obtain an adjacent abnormal domain.

[0075] S104 may include the following steps: determining a first voxel unit to which the first outlier belongs;

[0076] A first abnormal domain of the largest continuous enclosing area is established based on the boundary of the first voxel unit, and a neighboring abnormal domain affected by the first abnormal value is obtained by using the constructed neighboring abnormal value influence domain determination function.

[0077] In the embodiment of the present application, the function of locating the abnormal area near the point cloud is middle:

[0078] Based on the index position ExtArea of ​​the first voxel unit corresponding to the first outlier in the downsampled point cloud set out The extreme value of is taken as the boundary point, and the maximum continuous enclosing area determined by it is defined as the point cloud impact-repair area out , that is, the first abnormal domain. The adjacent abnormal value influence domain judgment function can be expressed as: For Affect-RepairArea based out The first abnormal domain and u thres-inflThat is, the first threshold determines the initial adjacent abnormal domain, and then based on the KDTree adjacent search algorithm and dis thres-infl The second threshold determines whether the preliminary adjacent abnormal area is the final adjacent abnormal area AdjArea affected by the first abnormal value out .

[0079] For example, if there are 4 first outliers, the index position of the first voxel unit corresponding to it in the downsampled point cloud set is ExtArea out The values ​​of H=3, H=10, H=10, H=16 and H=28 are 15 (H=3), 62 (H=10), 102 (H=16) and 185 (H=28), respectively. The first abnormal domain of the maximum continuous enclosing area established based on the boundary of the first voxel unit is 15 (H=3) to 185 (H=28). Further, using the constructed adjacent abnormal value influence domain judgment function, the index position ExtArea of ​​the voxel unit in the adjacent abnormal domain affected by the first abnormal value in the downsampled point cloud set is obtained. out It can be 21 (H=4) and 95 (H=15).

[0080] S105: Obtain a second voxel unit based on the first voxel unit and the voxel units in the adjacent abnormal domain.

[0081] In the embodiment of the present application, a second voxel unit including all abnormal values ​​is obtained based on the first voxel unit and the voxel units in the adjacent abnormal domain.

[0082] For example, if there are 4 first outliers, the index position of the first voxel unit corresponding to it in the downsampled point cloud set is ExtArea out They are 15 (H=3), 62 (H=10), 102 (H=16) and 185 (H=28), respectively. The index position of the voxel unit in the adjacent abnormal domain in the point cloud set after downsampling is ExtArea out is 21 (H=4) and 95 (H=15), then the index position ExtArea of ​​the second voxel unit containing all abnormal values ​​in the downsampled point cloud set is obtained out They are 15 (H=3), 62 (H=10), 102 (H=16), 185 (H=28), 21 (H=4) and 95 (H=15), respectively.

[0083] S106: Acquire a point cloud outlier repair region, repair the outliers in the second voxel unit, and obtain a repaired point cloud set.

[0084] S106 may include the following steps: based on the first abnormal domain, using the constructed abnormal value repair domain judgment function, obtaining a point cloud abnormal value repair area composed of third voxel units;

[0085] Based on the physical field parameters of the point cloud outlier repair area, the outliers in the second voxel unit are repaired to obtain a repaired point cloud set.

[0086] In the embodiment of the present application, the outlier repair domain determination function can be expressed as: For Affect-RepairArea based out The first abnormal domain and u thres-rpr That is, the third threshold identifies a third voxel unit that can be used for outlier repair. This third voxel unit then forms a point cloud outlier repair region. Based on the physical field parameters of the point cloud outlier repair region, the outliers in the second voxel unit are repaired or replaced, resulting in a repaired point cloud set.

[0087] S107: Generate a corresponding reduced-order model.

[0088] S107 may include the following steps: generating a corresponding reduced-order model based on the repaired point cloud set and the point cloud set of normal operating condition data to achieve full-field visual monitoring of the traction system under different line operation scenarios.

[0089] In this embodiment, a point cloud voxelization grid filtering algorithm is used to obtain a downsampled point cloud that excludes the first outlier. This downsampled point cloud allows for rapid repair of the first outlier and adjacent outliers in the adjacent outlier domain. Furthermore, based on the repaired point cloud and the point cloud of normal operating data, a more accurate reduced-order model is generated, shortening simulation cycles and ensuring excellent engineering applicability.

[0090] In some embodiments of the present application, a point cloud voxelization grid filtering algorithm is used to reorder the initial sequence numbers of the point cloud set of the abnormal operating condition data based on the comprehensive voxel index of the point cloud set of the abnormal operating condition data, and then downsampling is performed on the sorted point cloud sets corresponding to different voxel units to obtain a downsampled point cloud set composed of the first points of different voxel units, including the following steps:

[0091] S301: Using the point cloud voxelization grid filtering algorithm, based on the comprehensive voxel index of the point cloud set of the abnormal working condition data, the initial serial number of the point cloud set of the abnormal working condition data is reordered in ascending order to obtain the sorted point cloud sets corresponding to different voxel units.

[0092] S302: performing downsampling processing on the sorted point cloud sets corresponding to different voxel units to obtain downsampled point cloud sets consisting of the first points of different voxel units.

[0093] S303: When the first outlier exists in the downsampled point cloud set, the voxel resolution is updated, the comprehensive voxel index of the point cloud set of the abnormal working condition data is updated, and the reordering and downsampling steps are repeated until the downsampled point cloud set does not include the first outlier.

[0094] In the embodiment of the present application, the point cloud voxelization grid filtering optimization algorithm can be expressed as: y = f (sort H ,filter,H,R voxel ,u'). First, use the sorting algorithm sort H , based on the value of the comprehensive voxel index H, the unordered initial serial numbers corresponding to the point cloud set g(x,y,z,u)abnormal-i of the abnormal working condition data are sorted in ascending order to obtain the sorted point cloud set g(x,y,z,u)abnormal-i-sort of the abnormal working condition data. Secondly, the filtering algorithm filter is used to replace all the points in the corresponding voxel grid (unit) with the first point in the different values ​​of the above-mentioned comprehensive voxel index H to obtain the downsampled point cloud set g(x,y,z,u)abnormal-i-sort-sub. Finally, check whether the downsampled point cloud set g(x,y,z,u)abnormal-i-sort-sub contains the above-mentioned first abnormal value u'. If so, update the voxel resolution R voxel , update the comprehensive voxel index of the point cloud set of abnormal working condition data, repeat the reordering and downsampling steps until the downsampled point cloud set g(x,y,z,u)abnormal-i-sort-sub does not have the first abnormal value u'. The comprehensive voxel index H is expanded in the order of highest priority along the coordinate dimension with the smallest rate of change of the physical field. The comprehensive voxel index H is based on the voxel resolution R voxel Calculated.

[0095] In some embodiments of the present application, the method further comprises the following steps:

[0096] S401: When the rate of change of the physical field in the first dimension is smaller than the rate of change of the physical field in the second dimension, and the rate of change of the physical field in the second dimension is smaller than the rate of change of the physical field in the third dimension, the coordinates of each dimension, the minimum coordinates in each dimension, and the voxel resolution of the point cloud set based on the abnormal working condition data are rounded down to obtain the voxel index in each dimension.

[0097] In an embodiment of the present application, the difference between the coordinates of each dimension of the point cloud set of abnormal working condition data and the minimum coordinates of each dimension direction is then rounded down with the voxel resolution to obtain the voxel index in each dimension direction.

[0098] For example, if the rate of change of the physical field u satisfies When , the voxel index in the x-dimension direction is Voxel index in the y-dimension Voxel index in the z-dimension in, is the floor symbol.

[0099] S402: Obtain the number of voxels in the first dimension based on the maximum coordinate, the minimum coordinate, and the voxel resolution in the first dimension.

[0100] In the embodiment of the present application, the difference between the maximum coordinate and the minimum coordinate in the first dimension direction and the ratio of the difference to the voxel resolution are used to obtain the number of voxels in the first dimension direction.

[0101] Based on the above example, Dx=(x max -x min ) / R voxel .

[0102] S403: Obtain the number of voxels in the second dimension based on the maximum coordinate, the minimum coordinate, and the voxel resolution in the second dimension.

[0103] In the embodiment of the present application, the difference between the maximum coordinate and the minimum coordinate in the second dimension is divided by the ratio of the voxel resolution to obtain the number of voxels in the second dimension.

[0104] Based on the above example, Dz=(z max -z min ) / R voxel .

[0105] S404: Based on the voxel index of the point cloud set of the abnormal working condition data in the first dimension direction, the voxel index in the second dimension direction and the number of voxels in the first dimension direction, as well as the voxel index in the third dimension direction, the number of voxels in the first dimension direction and the number of voxels in the second dimension direction, the comprehensive voxel index of the point cloud set of the abnormal working condition data is obtained.

[0106] In the embodiment of the present application, when the rate of change of the physical field in the first dimension is less than that in the second dimension, and the rate of change of the physical field in the second dimension is less than that in the third dimension, the expression for the comprehensive voxel index is: H = Hi + Hj*Di + Hk*Di*Dj, where Hi, Hj, and Hk are the three-dimensional voxel indices of the point cloud in the first, second, and third dimensions, respectively, and Di, Dj, and Dk are the number of voxels in the point cloud in the first, second, and third dimensions, respectively. H is the index expanded in the order of highest priority along the coordinate dimension with the smallest rate of change of the physical field.

[0107] For example, if the rate of change of the physical field u satisfies When, H=Hx+Hy*Dx+Hz*Dx*Dz.

[0108] Or when the rate of change of the physical field u satisfies When, H=Hz+Hx*Dz+Hy*Dz*Dx.

[0109] In some embodiments of the present application, determining the first voxel unit to which the first outlier belongs includes the following steps:

[0110] S501: Using an indexing algorithm, mark the downsampled point cloud set at the first index position of the sorted point cloud set, and mark the point corresponding to the first outlier at the second index position of the sorted point cloud set.

[0111] S502: Using a proximity algorithm, search for a target first index position that is closest to the second index position.

[0112] S503: When the second index position is smaller than the target first index position, determine the index position of the first voxel unit to which the first abnormal value belongs as a first index position before the target first index position in the downsampled point cloud set.

[0113] S504: When the second index position is greater than the target first index position, determine the index position of the first voxel unit to which the first abnormal value belongs as the target first index position in the downsampled point cloud set.

[0114] In the embodiment of the present application, the extreme outlier location allocation function of the point cloud middle:

[0115] The function is a point cloud positioning function, which uses the index algorithm to mark the downsampled point cloud set g(x, y, z, u) abnormal-i-sort-sub at the first index position Sub of the sorted point cloud set id , and mark the first outlier u' at the second index position Out of the sorted point cloud set id .

[0116] Function is the point cloud extreme outlier distribution function First, through KDTree

[0117] The proximity algorithm searches for any Out id (i) Recent Sub id (i). Then by Function Comparison Out id (i) and Sub id (i) The size of the final return is less than Out id(i) Sub id (i) or Sub id (i-1), that is, assigning the first outlier to the voxel unit to which it belongs, and obtaining the index position ExtArea of ​​the voxel unit in the point cloud set after downsampling out (i.e. the first voxel unit), specifically as follows: ExtArea out =Sub id (i-1); Out id (i)>Sub id (i),ExtArea out =Sub id (i)).

[0118] For example, Figure 6 As shown in (a), Out id (i)<Sub id (i), then ExtArea out =Sub id (i-1). like Figure 6 As shown in (b), Out id (i)>Sub id (i), then ExtArea out =Sub id (i)).

[0119] In an embodiment of the present application, the method of establishing a first abnormal domain of the maximum continuous enclosing area based on the boundary of the first voxel unit and obtaining an adjacent abnormal domain affected by the first abnormal value by using the constructed adjacent abnormal value influence domain determination function includes the following steps:

[0120] S701: Determine the average value and median based on the physical field parameters of the downsampled points corresponding to each voxel unit and the N voxel units before and after it in the first abnormal domain; N is an integer greater than or equal to 1.

[0121] S702: When the absolute value of the difference between the mean value and the median value is greater than or equal to a first threshold, the corresponding voxel unit is identified as a preliminary adjacent abnormal domain.

[0122] S703: Using a proximity algorithm, search for a target third index position that is closest to the fourth index position; wherein the third index position is the index position of the downsampled point corresponding to the voxel unit in the preliminary neighboring abnormal domain in the sorted point cloud set, and the fourth index position is the index position of the downsampled point of the first voxel unit in the sorted point cloud set.

[0123] S704: When the difference between the fourth index position and the target third index position is less than or equal to the second threshold, the preliminary adjacent abnormal domain is confirmed as an adjacent abnormal domain.

[0124] S705: When the difference between the fourth index position and the target third index position is greater than a second threshold, the preliminary adjacent abnormal region is identified as a noise point.

[0125] For example, let's take a set of data:

[0126] When the comprehensive voxel index H=1, the serial number of the first point of the voxel unit, that is, the point after downsampling, is 1, and the physical field parameter of the point is u=336; when the comprehensive voxel index H=2, the serial number of the first point of the voxel unit, that is, the point after downsampling, is 8, and the physical field parameter of the point is u=336.3; when the comprehensive voxel index H=3, the serial number of the first point of the voxel unit, that is, the point after downsampling, is 15, and the physical field parameter of the point is u=337.5; when the comprehensive voxel index H=4, the serial number of the first point of the voxel unit, that is, the point after downsampling, is 21, and the physical field parameter of the point is u=340.1; when the comprehensive voxel index H=5, the serial number of the first point of the voxel unit, that is, the point after downsampling, is 29, and the physical field parameter of the point is u=337.3; when the comprehensive voxel index H=6, the serial number of the first point of the voxel unit, that is, the point after downsampling, is 35, and the physical field parameter of the point is u=337.7; when the comprehensive voxel index H=7, the serial number of the first point of the voxel unit, that is, the point after downsampling, is 42, and the physical field parameter of the point is u=337.4; when the comprehensive voxel index H=8, the serial number of the first point of the voxel unit, that is, the point after downsampling, is 50, and the physical field parameter of the point is u=337.1; when the comprehensive voxel index H=9, the serial number of the first point of the voxel unit, that is, the point after downsampling, is 56, and the physical field parameter of the point is u=336.4; when the comprehensive voxel index H=10, the serial number of the first point of the voxel unit, that is, the point after downsampling, is is 62, and the physical field parameter u of this point is 336.1; when the comprehensive voxel index H is 11, the serial number of the first point of the voxel unit, that is, the point after downsampling, is 70, and the physical field parameter u of this point is 338.2; when the comprehensive voxel index H is 12, the serial number of the first point of the voxel unit, that is, the point after downsampling, is 76, and the physical field parameter u of this point is 338.6; when the comprehensive voxel index H is 13, the serial number of the first point of the voxel unit, that is, the point after downsampling, is 82, and the physical field parameter u of this point is 339; when the comprehensive voxel index H is 14, the serial number of the first point of the voxel unit, that is, the point after downsampling, is 88, and the physical field parameter u of this point is 339.5; when the comprehensive voxel index H is 15, the serial number of the first point of the voxel unit, that is, the point after downsampling, is 88, and the physical field parameter u of this point is 339.5 The first point, or downsampled point, has a serial number of 95, and its physical field parameter is u = 342.5. When the integrated voxel index H = 16, the first point, or downsampled point, of this voxel unit has a serial number of 102, and its physical field parameter is u = 340.5. When the integrated voxel index H = 17, the first point, or downsampled point, of this voxel unit has a serial number of 109, and its physical field parameter is u = 340.1. When the integrated voxel index H = 18, the first point, or downsampled point, of this voxel unit has a serial number of 116, and its physical field parameter is u = 339.6. When the integrated voxel index H = 19, the first point, or downsampled point, of this voxel unit has a serial number of 122, and its physical field parameter is u = 339.2; when the comprehensive voxel index H=20, the serial number of the first point of the voxel unit, that is, the point after downsampling, is 128, and the physical field parameter u of the point is 338.8; when the comprehensive voxel index H=21, the serial number of the first point of the voxel unit, that is, the point after downsampling, is 134, and the physical field parameter u of the point is 345; when the comprehensive voxel index H=22, the serial number of the first point of the voxel unit, that is, the point after downsampling, is 140, and the physical field parameter u of the point is 345.5; when the comprehensive voxel index H=23, the serial number of the first point of the voxel unit, that is, the point after downsampling, is 147, and the physical field parameter u of the point is 346; when the comprehensive voxel index H=24, the serial number of the first point of the voxel unit, that is, the point after downsampling, is 155, and the physical field parameter u of the point is 346.6; when the comprehensive voxel index H=25, the serial number of the first point of the voxel unit, that is, the point after downsampling, is 157, and the physical field parameter u of the point is 350. The serial number of the point is 163, and the physical field parameter of this point is u = 347.1; when the integrated voxel index H = 26, the serial number of the first point of this voxel unit, that is, the point after downsampling, is 169, and the physical field parameter of this point is u = 347.6; when the integrated voxel index H = 27, the serial number of the first point of this voxel unit, that is, the point after downsampling, is 177, and the physical field parameter of this point is u = 347; when the integrated voxel index H = 28, the serial number of the first point of this voxel unit, that is, the point after downsampling, is 185, and the physical field parameter of this point is u = 346.5; when the integrated voxel index H = 29, the serial number of the first point of this voxel unit, that is, the point after downsampling, is 191, and the physical field parameter of this point is u = 345.8; when the integrated voxel index H = 30, the serial number of the first point of this voxel unit, that is, the point after downsampling, is 197, and the physical field parameter of this point is u = 345.5.

[0127] For example, the first abnormal domain is 15 (H=3) to 185 (H=28).

[0128] Based on this, if N=2, the example calculates the physical field parameters u of the downsampled points corresponding to the voxel unit 21 (H=4) and the two voxel units before and after in the first abnormal domain, determines the average value μ and the median η, and then calculates |μ-η|, and then compares |μ-η| with the first threshold. When |μ-η| is greater than or equal to the first threshold u thres-infl When the voxel unit 21 (H=4) is identified as the preliminary adjacent abnormal area AdjArea out1 In addition, voxel unit 95 (H=15), voxel unit 128 (H=20) and voxel unit 134 (H=21) are confirmed as preliminary adjacent abnormal area AdjArea out1 .

[0129] Among them, the example voxel unit 15 (H=3), the voxel unit 62 (H=10), the voxel unit 102 (H=16) and the voxel unit 185 (H=28) are first voxel units.

[0130] Secondly, use the KDTree proximity algorithm to search for any ExtArea out (k) (i.e., the fourth index position) the nearest AdjArea out1 (k) (i.e. the third index position of the target), get ExtArea out (k) and AdjArea out1 (k) is the distance dis(k), and the preset distance threshold dis thres-infl For comparison, when dis(k) is less than or equal to the second threshold dis thres-infl When the initial adjacent abnormal area AdjArea out1 (k) Marked as the adjacent abnormal area AdjArea out , where dis(k) is greater than dis thres-infl It is an outlier (noise point), corresponding to voxel units 128 (H=20) and 134 (H=21).

[0131] In some embodiments of the present application, the step of obtaining a point cloud outlier repair region composed of a third voxel unit based on the first outlier domain and using a constructed outlier repair domain determination function comprises the following steps:

[0132] S801: Determine the average value and median based on the physical field parameters of the downsampled points corresponding to each voxel unit and the N voxel units before and after it in the first abnormal domain; N is an integer greater than or equal to 1.

[0133] S802: When the absolute value of the difference between the mean and the median is less than or equal to a third threshold, the corresponding voxel unit is used as a third voxel unit, and the area formed by the third voxel unit is used as a point cloud outlier repair area.

[0134] Based on a set of data in the above example, the first abnormal domain is 15 (H=3) to 185 (H=28). After executing S801 and S802, the third voxel units include: 29 (H=5), 35 (H=6), 42 (H=7), 50 (H=8), 56 (H=9), 70 (H=11), 76 (H=13), 88 (H=14), 109 (H=17), 116 (H=18), 122 (H=19), 128 (H=20), 134 (H=21), 140 (H=22), 147 (H=23), 155 (H=24), 163 (H=25), 169 (H=26), and 177 (H=27).

[0135] In some embodiments of the present application, repairing the outliers in the second voxel unit based on the physical field parameters of the point cloud outlier repair area to obtain a repaired point cloud set includes the following steps:

[0136] S901: Using a neighbor algorithm, search for a third voxel unit that is closest to the second voxel unit.

[0137] In an embodiment of the present application, a proximity algorithm is used to search for a third voxel unit that is closest to the second voxel unit based on the index position of the downsampled point of the second voxel unit in the sorted point cloud set and the index position of the downsampled point of the third voxel unit in the sorted point cloud set.

[0138] Based on a set of data in the above example, the second voxel units include 15 (H=3), 62 (H=10), 102 (H=16), 185 (H=28), 21 (H=4) and 95 (H=15), and the third voxel units include: 29 (H=5), 35 (H=6), 42 (H=7), 50 (H=8), 56 (H=9), 70 (H=11), 76 (H=13), 88 (H=14), 109 (H=17), 116 (H=18), 122 (H=19), 128 (H=20), and 132 (H=21). ), 134 (H=21), 140 (H=22), 147 (H=23), 155 (H=24), 163 (H=25), 169 (H=26), and 177 (H=27), then the neighboring algorithm is used to search for the third voxel unit closest to the second voxel unit 15 (H=3), which is 29 (H=5); the third voxel unit closest to the second voxel unit 21 (H=4) is 29 (H=5); the third voxel unit closest to the second voxel unit 62 (H=10) is 56 (H=9), and so on.

[0139] S902: When the difference between the physical field parameters of the downsampled points in the second voxel unit and the physical field parameters of the downsampled points in the third voxel unit is less than or equal to the fourth threshold, the physical field parameters of the downsampled points in the third voxel unit are updated with the outliers in the second voxel unit using a replacement function to obtain a repaired point cloud set.

[0140] In the embodiment of the present application, when a third voxel unit is searched, the difference between the physical field parameters of the downsampled point of the second voxel unit and the physical field parameters of the downsampled point of the third voxel unit is usually less than or equal to the fourth threshold value, and the physical field parameters of the downsampled point of the third voxel unit are updated to all abnormal values ​​in the second voxel unit (the number is at least one), and a repaired point cloud set is obtained. When two third voxel units are searched, the difference between the physical field parameters of the downsampled point of the second voxel unit and the physical field parameters of the downsampled points of the two third voxel units is further determined to see which difference is less than or equal to the fourth threshold value, and it is used as the target third voxel unit, and then the physical field parameters of the downsampled point of the target third voxel unit are updated to all abnormal values ​​in the second voxel unit (the number is at least one), and a repaired point cloud set is obtained.

[0141] In some embodiments of the present application, generating a corresponding reduced-order model based on the repaired point cloud set of abnormal operating condition data and the point cloud set of normal operating condition data to achieve full-field visual monitoring of the traction system under different line operation scenarios includes:

[0142] Taking different operating condition data as input items, and the point cloud set of the repaired abnormal operating condition data and the point cloud set of normal operating condition data as output items, a reduced-order model corresponding to the traction system is constructed using a neural network and a deep learning algorithm to achieve full-field visual monitoring of the traction system under different line operation scenarios.

[0143] Based on the above embodiment, this application specifically illustrates a method for repairing outliers in a digital twin reduced-order model based on point cloud recognition, including the following steps:

[0144] S1001: Establish a finite element simulation model of the traction system (motor, converter) in the fields of rail transit, urban rail transit, etc., and extract the point cloud set g(x, y, z, u) of the three-dimensional physical field corresponding to different working condition data of the traction system; the physical field u of the point cloud set includes but is not limited to temperature field, flow field, and stress field.

[0145] like Figure 11 FIG. 1 is a schematic diagram of the structure of a certain type of motor 110 of an exemplary traction system.

[0146] S1002: Establish a physical field anomaly criterion model, based on the analysis of operating parameters, filter and output the point cloud set g(x, y, z, u) corresponding to the above different operating condition data, where there is an abnormal operating condition data exceeding the physical field parameter threshold. abnormal .

[0147] Among them, the physical field anomaly judgment model includes: determining the physical field parameter thresholds under extreme working conditions based on the operating parameters and historical data analysis of devices such as motors and converters in the traction system, and comparing the physical field parameters corresponding to each operating condition point cloud set with the preset physical field parameter thresholds. If the physical field parameters of a certain operating condition point cloud set exceed the physical field parameter threshold, it is judged to be a point cloud set of abnormal operating condition data.

[0148] S1003: The point cloud set g(x,y,z,u) of the abnormal working condition data is abnormal For any working condition data in , the interquartile range algorithm y=f(QL,QU,IQR,k) is used to detect and identify the abnormal value of the physical field parameter u. The u value data less than QL-k*IQR or greater than QU+k*IQR is defined as an extreme abnormal value, marked as u', that is, u'={(u<QL-k*IQR)|(u> QU+k*IQR)}.

[0149] Among them, the extreme outlier is the first outlier.

[0150] Among them, in the interquartile range algorithm y=f(QL, QU, IQR, k), QL is the lower quartile, QU is the upper quartile, IQR is the interquartile range, and k is the adjustment coefficient, which is set according to the working condition data corresponding to the traction system components; k can take a value of 1.5 or 2 or other, and the embodiment of the present application does not limit this.

[0151] S1004: Constructing a point cloud voxelized grid filtering optimization algorithm y=f(sort H ,filter,H,R voxel ,u'), the unordered serial numbers corresponding to the point cloud set g(x,y,z,u)abnormal-i of any abnormal working condition data are reordered in a form consistent with the arrangement of adjacent voxels and the slowly changing direction of the physical field, and at the same time, voxel filtering is performed on g(x,y,z,u)abnormal-i based on the extreme outliers to obtain the downsampled point cloud set g(x,y,z,u)abnormal-i-sort-sub; g(x,y,z,u)abnormal-i-sort-sub does not contain extreme outliers u'.

[0152] Specifically, it may include: in the point cloud voxelization grid filtering optimization algorithm y=f(sort H ,filter,H,R voxel ,u') in:

[0153] First, use the sort algorithm of comprehensive voxel index H, based on the value of the comprehensive voxel index H, sort the unordered serial numbers corresponding to the point cloud set g(x,y,z,u)abnormal-i of any abnormal working condition data in ascending order to obtain the sorted point cloud set g(x,y,z,u)abnormal-i-sort. The expression of the comprehensive voxel index H (one-dimensional voxel index) is H=Hi+Hj*Di+Hk*Di*Dj, where Hi / Hj / Hk is the three-dimensional voxel index of the point cloud in each dimensional direction, and Di / Dj / Dk is the number of voxels in the point cloud in each dimensional direction. H is the index expanded in the highest priority order along the coordinate dimension with the smallest rate of change of the physical field, that is, when the rate of change of the physical field u satisfies H=Hx+Hy*Dx+Hz*Dx*Dz, or when the rate of change of the physical field u satisfies H=Hz+Hx*Dz+Hy*Dz*Dx.

[0154] Secondly, the filtering algorithm filter is used to replace all points in the corresponding voxel grid (unit) with the first point in different values ​​of the above-mentioned comprehensive voxel index H to obtain the downsampled point cloud set g(x, y, z, u) abnormal-i-sort-sub.

[0155] Finally, check whether the downsampled point cloud set g(x,y,z,u)abnormal-i-sort-sub contains the above-mentioned extreme outlier u'. If so, update the voxel resolution R voxel Repeat the above steps until g(x,y,z,u)abnormal-i-sort-sub does not have an extreme outlier u'; where R voxel Used to calculate the values ​​of Hi / Hj / Hk and Di / Dj / Dk, that is For floor operation, D i =(i max -i min ) / R voxeli .

[0156] like Figure 2 The figure shows an example of how a point cloud set may be represented after being voxelized into a grid.

[0157] like Figure 12 As shown in FIG, a schematic diagram of a point cloud set of a certain type of motor of a traction system after filtering and downsampling is given. Figure 12 There are extreme outliers in (a). Figure 12 There are no extreme outliers in (b).

[0158] S1005: Constructing point cloud extreme outlier location allocation function Determine the index position ExtArea of ​​the voxel corresponding to the extreme outlier in the sorted point cloud set out .

[0159] Specifically, it may include: positioning allocation function of extreme outliers in point cloud middle:

[0160] The function is a point cloud positioning function, which uses the index algorithm to mark the downsampled point cloud set g(x,y,z,u)abnormal-i-sort-sub and the extreme outlier u'. The index positions of the sorted point cloud set g(x,y,z,u)abnormal-i-sort are Sub id and Out id .

[0161] Function is the point cloud extreme outlier distribution function First, use the KDTree proximity algorithm to search for any Out id (i) Recent Sub id (i). Then by Function Comparison Out id (i) and Sub id (i) The size of the final return is less than Out id (i) Sub id (i) or Sub id (i-1), that is, assigning the extreme outlier to the voxel unit to which it belongs, and obtaining the index position ExtArea of ​​the voxel unit in the sorted point cloud set out (first voxel unit), specifically as follows:

[0162]

[0163] For example, Figure 6 As shown in (a), Out id (i)<Sub id (i), then ExtArea out =Sub id (i-1). like Figure 6 As shown in (b), Out id (i)>Sub id (i), then ExtArea out =Sub id (i)).

[0164] S1006: Constructing point cloud adjacent abnormal domain positioning function Mark the adjacent abnormal area AdjArea affected by extreme outliers out .

[0165] Specifically, it may include: positioning function in the abnormal area near the point cloud middle:

[0166] Based on the index position ExtArea of ​​the voxel corresponding to the extreme outlier value in the sorted abnormal working condition point cloud set out The extreme value of is taken as the boundary point, and the maximum continuous enclosing area determined by it is defined as the point cloud impact-repair area out (i.e. the first abnormal domain), using the constructed adjacent abnormal value influence domain determination function Determine the adjacent abnormal area AdjArea affected by extreme outliers out Specifically, first calculate the point cloud impact-repair area Affect-RepairArea out The absolute deviation |μ-η| of the mean μ and median η of the physical field u in each voxel unit i and its spatial neighborhood (i.e., including i and the N voxel units before and after i) and the preset influence threshold u thres-infl (i.e. the first threshold) is compared, when |μ-η| is greater than or equal to u thres-infl When the voxel unit i is confirmed as the preliminary adjacent abnormal area AdjArea out1 . Secondly, use the KDTree proximity algorithm to search for any ExtArea out (k) Nearest AdjArea out1 (k), get ExtArea out (k) and AdjArea out1 (k) is the distance dis(k), and the preset distance threshold dis thres-infl (i.e. the second threshold) for comparison, when dis(k) is less than or equal to dis thres-infl When the initial adjacent abnormal area AdjArea out1 (k) Marked as the adjacent abnormal area AdjArea out , where dis(k) is greater than dis thres-infl are outliers (noise points), corresponding to 128 (H=20) and 134 (H=21) in the figure.

[0167] The details are as follows:

[0168]

[0169] dis=KDTress(ExtArea out ,AdjArea out1 );

[0170] dis(k)≤dis thres-infll,AdjArea out =AdjArea out1 (k)).

[0171] For example, let's take a set of data:

[0172] When the comprehensive voxel index H=1, the serial number of the first point of the voxel unit, that is, the point after downsampling, is 1, and the physical field parameter of the point is u=336; when the comprehensive voxel index H=2, the serial number of the first point of the voxel unit, that is, the point after downsampling, is 8, and the physical field parameter of the point is u=336.3; when the comprehensive voxel index H=3, the serial number of the first point of the voxel unit, that is, the point after downsampling, is 15, and the physical field parameter of the point is u=337.5; when the comprehensive voxel index H=4, the serial number of the first point of the voxel unit, that is, the point after downsampling, is 21, and the physical field parameter of the point is u=340.1; when the comprehensive voxel index H=5, the serial number of the first point of the voxel unit, that is, the point after downsampling, is 29, and the physical field parameter of the point is u=337.3; when the comprehensive voxel index H=6, the serial number of the first point of the voxel unit, that is, the point after downsampling, is 35, and the physical field parameter of the point is u=337.7; when the comprehensive voxel index H=7, the serial number of the first point of the voxel unit, that is, the point after downsampling, is 42, and the physical field parameter of the point is u=337.4; when the comprehensive voxel index H=8, the serial number of the first point of the voxel unit, that is, the point after downsampling, is 50, and the physical field parameter of the point is u=337.1; when the comprehensive voxel index H=9, the serial number of the first point of the voxel unit, that is, the point after downsampling, is 56, and the physical field parameter of the point is u=336.4; when the comprehensive voxel index H=10, the serial number of the first point of the voxel unit, that is, the point after downsampling, is is 62, and the physical field parameter u of this point is 336.1; when the comprehensive voxel index H is 11, the serial number of the first point of the voxel unit, that is, the point after downsampling, is 70, and the physical field parameter u of this point is 338.2; when the comprehensive voxel index H is 12, the serial number of the first point of the voxel unit, that is, the point after downsampling, is 76, and the physical field parameter u of this point is 338.6; when the comprehensive voxel index H is 13, the serial number of the first point of the voxel unit, that is, the point after downsampling, is 82, and the physical field parameter u of this point is 339; when the comprehensive voxel index H is 14, the serial number of the first point of the voxel unit, that is, the point after downsampling, is 88, and the physical field parameter u of this point is 339.5; when the comprehensive voxel index H is 15, the serial number of the first point of the voxel unit, that is, the point after downsampling, is 88, and the physical field parameter u of this point is 339.5 The first point, or downsampled point, has a serial number of 95, and its physical field parameter is u = 342.5. When the integrated voxel index H = 16, the first point, or downsampled point, of this voxel unit has a serial number of 102, and its physical field parameter is u = 340.5. When the integrated voxel index H = 17, the first point, or downsampled point, of this voxel unit has a serial number of 109, and its physical field parameter is u = 340.1. When the integrated voxel index H = 18, the first point, or downsampled point, of this voxel unit has a serial number of 116, and its physical field parameter is u = 339.6. When the integrated voxel index H = 19, the first point, or downsampled point, of this voxel unit has a serial number of 122, and its physical field parameter is u = 339.2; when the comprehensive voxel index H=20, the serial number of the first point of the voxel unit, that is, the point after downsampling, is 128, and the physical field parameter u of the point is 338.8; when the comprehensive voxel index H=21, the serial number of the first point of the voxel unit, that is, the point after downsampling, is 134, and the physical field parameter u of the point is 345; when the comprehensive voxel index H=22, the serial number of the first point of the voxel unit, that is, the point after downsampling, is 140, and the physical field parameter u of the point is 345.5; when the comprehensive voxel index H=23, the serial number of the first point of the voxel unit, that is, the point after downsampling, is 147, and the physical field parameter u of the point is 346; when the comprehensive voxel index H=24, the serial number of the first point of the voxel unit, that is, the point after downsampling, is 155, and the physical field parameter u of the point is 346.6; when the comprehensive voxel index H=25, the serial number of the first point of the voxel unit, that is, the point after downsampling, is 157, and the physical field parameter u of the point is 350. The serial number of the point is 163, and the physical field parameter of this point is u = 347.1; when the integrated voxel index H = 26, the serial number of the first point of this voxel unit, that is, the point after downsampling, is 169, and the physical field parameter of this point is u = 347.6; when the integrated voxel index H = 27, the serial number of the first point of this voxel unit, that is, the point after downsampling, is 177, and the physical field parameter of this point is u = 347; when the integrated voxel index H = 28, the serial number of the first point of this voxel unit, that is, the point after downsampling, is 185, and the physical field parameter of this point is u = 346.5; when the integrated voxel index H = 29, the serial number of the first point of this voxel unit, that is, the point after downsampling, is 191, and the physical field parameter of this point is u = 345.8; when the integrated voxel index H = 30, the serial number of the first point of this voxel unit, that is, the point after downsampling, is 197, and the physical field parameter of this point is u = 345.5.

[0173] For example, the first abnormal domain is 15 (H=3) to 185 (H=28).

[0174] Based on this, if N=2, the example calculates the physical field parameters u of the downsampled points corresponding to the voxel unit 21 (H=4) and the two voxel units before and after in the first abnormal domain, determines the average value μ and the median η, and then calculates |μ-η|, and then compares |μ-η| with the first threshold. When |μ-η| is greater than or equal to the first threshold u thres-infl When the voxel unit 21 (H=4) is identified as the preliminary adjacent abnormal area AdjArea out1 In addition, voxel unit 95 (H=15), voxel unit 128 (H=20) and voxel unit 134 (H=21) are confirmed as preliminary adjacent abnormal area AdjArea out1 .

[0175] Among them, the example voxel unit 15 (H=3), the voxel unit 62 (H=10), the voxel unit 102 (H=16) and the voxel unit 185 (H=28) are first voxel units.

[0176] Secondly, use the KDTree proximity algorithm to search for any ExtArea out (k) (i.e., the fourth index position) the nearest AdjArea out1 (k) (i.e. the third index position of the target), get ExtArea out (k) and AdjArea out1 (k) is the distance dis(k), and the preset distance threshold dis thres-infl For comparison, when dis(k) is less than or equal to the second threshold dis thres-infl When the initial adjacent abnormal area AdjArea out1 (k) Marked as the adjacent abnormal area AdjArea out , where dis(k) is greater than dis thres-infl It is an outlier (noise point), corresponding to voxel units 128 (H=20) and 134 (H=21).

[0177] S1007: Perform point cloud fusion on the voxel unit corresponding to the extreme outlier value and the adjacent abnormal area affected by the extreme outlier value to obtain the voxel unit Area containing all abnormal data. out .

[0178] In the embodiment of the present application, based on the voxel unit (i.e., the first voxel unit) corresponding to the extreme outlier (i.e., the first outlier) and the voxel units in the adjacent outlier domain, the voxel unit Area containing all outliers is obtained. out (ie the second voxel unit).

[0179] For example, if there are 4 first outliers, the index position of the first voxel unit corresponding to it in the downsampled point cloud set is ExtArea out They are 15 (H=3), 62 (H=10), 102 (H=16) and 185 (H=28), respectively. The index position of the voxel unit in the adjacent abnormal domain in the point cloud set after downsampling is ExtArea out is 21 (H=4) and 95 (H=15), then the index position ExtArea of ​​the second voxel unit containing all abnormal values ​​in the downsampled point cloud set is obtained out They are 15 (H=3), 62 (H=10), 102 (H=16), 185 (H=28), 21 (H=4) and 95 (H=15), respectively.

[0180] S1008: Constructing a point cloud outlier repair domain positioning function Mark the area where the abnormal point cloud is repaired repair .

[0181] Specifically, it may include: repairing the localization function in the point cloud outlier domain middle:

[0182] Based on the above point cloud impact-repair area out , using the constructed outlier repair domain decision function Calculate the impact of point cloud outliers-Repair Area out The absolute deviation |μ-η| of the mean μ and median η of the physical field u in each voxel unit i and its spatial neighborhood (i.e., including i and the n voxel points before and after it) is greater than the preset repair threshold u thres-rpr (ie the third threshold value) for comparison, when |μ-η| is less than or equal to the third threshold value u thres-rpr When , the voxel point i is marked as an outlier repair area Area repair .

[0183] The details are as follows:

[0184]

[0185] S1009: Constructing a point cloud outlier repair function Use the above outliers to repair the domain Area repair The corresponding physical field replaces and updates the above voxel unit Area containing all abnormal data out The physical fields corresponding to all points in the system are used to repair the point cloud set of abnormal working condition data.

[0186] Specifically, it may include: using the proximity algorithm to search for any area out (k) Nearest Area repair (k), when Area out (k) and Area repair When the physical field difference Δu of (k) is within the preset threshold (i.e., the fourth threshold), the replacement function is used The Area out (k) The physical field values ​​corresponding to all points in the voxel unit are repaired using the area repair (k) The physical field u corresponding to the voxel unit rpr (k) Perform replacement and update to complete the repair of the point cloud set of abnormal working condition data.

[0187] like Figure 13 As shown, an example of a repair process for a certain type of motor in a traction system is shown. Figure 13 (a) is the initial model, Figure 13 (b) After the extreme outliers are repaired, Figure 13 (c) After repairing the neighboring outliers.

[0188] S1010: For the remaining point cloud sets of abnormal working condition data, repeat the above steps S1003-S1009 to complete the outlier repair.

[0189] S1011: Based on the repaired abnormal operating condition point cloud set and the normal operating condition point cloud set, a corresponding reduced-order model is generated to achieve full-field visual monitoring of the traction system under different line operation scenarios.

[0190] The method includes taking different working condition data as input items, and taking the repaired abnormal working condition point cloud set and the normal working condition point cloud set physical field data corresponding to the different working condition data as output items, using neural networks and deep learning algorithms to construct a reduced-order model corresponding to the traction system, and realizing full-field visual monitoring of the traction system under different line operation scenarios.

[0191] Based on the above embodiment, the following beneficial effects are achieved:

[0192] 1) For large-scale point cloud training models with local abnormal data, it can achieve rapid and automatic positioning and repair of abnormal areas, improving processing efficiency;

[0193] 2) For devices with complex geometric structures and large scale, the mesh quality requirements of the finite element simulation model required in the early stage of constructing its twin model are reduced, thus shortening the simulation cycle;

[0194] 3) While retaining the scale of the point cloud, it is repaired and calibrated based on the normal data attached to the outliers to ensure that the model data is more consistent with the actual change laws of the physical field and to build a more accurate digital twin model.

[0195] The key technical points of this application are as follows:

[0196] 1) A digital twin reduced-order model outlier repair method based on point cloud recognition is characterized by constructing four key functions: a point cloud voxelized grid filtering optimization algorithm function, a point cloud neighboring anomaly domain positioning function, a point cloud outlier repair domain positioning function, and a point cloud outlier repair function. The point cloud voxelized grid filtering optimization algorithm function is used to obtain a filtered point cloud free of extreme outliers as a reference for the physical field repair value in the point cloud outlier repair domain positioning function. The outlier domain formed by the point cloud extreme outlier positioning allocation function and the point cloud neighboring anomaly domain positioning function is quickly repaired based on the point cloud outlier repair function.

[0197] 2) The point cloud voxelization grid filtering optimization algorithm function is characterized by combining the voxel resolution determined by the extreme outliers identified by the point cloud outlier detection algorithm with a voxel filtering method determined by the rate of change of the physical field to screen out a reference point cloud set for repairing the outliers;

[0198] 3) The point cloud neighboring outlier domain positioning function is characterized by determining the point cloud outlier impact-repair area based on the voxel unit boundary of the extreme outlier, and screening out the neighboring outlier areas affected by the extreme outlier by constructing the outlier impact domain determination function;

[0199] 4) The point cloud outlier repair domain positioning function is characterized in that the point cloud outlier impact-repair area is determined based on the voxel unit boundary of the extreme outlier, and the area for repairing the extreme outlier and the adjacent outlier area affected by the extreme outlier is screened out by constructing an outlier repair domain judgment function.

[0200] In order to implement the method of the embodiment of the present application, based on the same inventive concept, the embodiment of the present application also provides a digital twin reduced-order model outlier repair device based on point cloud recognition, Figure 14 This is a structural diagram of the device for repairing outliers in a digital twin reduced-order model based on point cloud recognition in an embodiment of the present application. Figure 14 As shown, the digital twin reduced-order model outlier repair device 140 based on point cloud recognition includes:

[0201] The point cloud extraction module 1401 is used to establish a finite element simulation model of the traction system, use different working condition data as input to the finite element simulation model, and obtain a point cloud set containing the three-dimensional spatial coordinates and physical field parameters of the traction system corresponding to the different working condition data;

[0202] A screening module 1402 is configured to screen point cloud sets of different operating condition data using the constructed physical field anomaly criterion model based on a preset physical field parameter threshold to obtain a point cloud set of abnormal operating condition data;

[0203] An outlier detection processing module 1403 is configured to perform outlier detection processing on the physical field parameters of the point cloud set of the abnormal operating condition data using an interquartile range algorithm, and define the physical field parameters that exceed the fence range as first outliers;

[0204] The point cloud voxelization grid filtering module 1404 is configured to utilize a point cloud voxelization grid filtering algorithm to reorder the initial sequence numbers of the point cloud set of the abnormal operating condition data based on the comprehensive voxel index of the point cloud set, and then perform downsampling processing on the sorted point cloud sets corresponding to different voxel units to obtain downsampled point cloud sets consisting of the first points of different voxel units; wherein the downsampled point cloud sets do not include the first abnormal value, and the comprehensive voxel index is indexed in the coordinate dimension with the smallest rate of change of the physical field, with the highest priority.

[0205] The processing module 1405 is configured to determine a first voxel unit to which the first abnormal value belongs;

[0206] A neighboring outlier influence domain determination module 1406 is configured to establish a first outlier domain of the largest continuous enclosing area based on the boundary of the first voxel unit, and obtain a neighboring outlier domain affected by the first outlier using the constructed neighboring outlier influence domain determination function;

[0207] The processing module 1405 is further configured to obtain a second voxel unit containing all abnormal values ​​based on the first voxel unit and the voxel units in the adjacent abnormal domain;

[0208] An outlier repair domain determination module 1407 is configured to obtain a point cloud outlier repair region composed of a third voxel unit based on the first outlier domain and using the constructed outlier repair domain determination function;

[0209] An outlier repair module 1408 is configured to repair the outliers in the second voxel unit based on the physical field parameters of the point cloud outlier repair region to obtain a point cloud set of repaired abnormal working condition data;

[0210] The reduced-order model generation module 1409 is used to generate a corresponding reduced-order model based on the repaired point cloud set and the point cloud set of normal operating data, so as to realize full-field visual monitoring of the traction system under different line operation scenarios.

[0211] In this embodiment, a point cloud voxelization grid filtering algorithm is used to obtain a downsampled point cloud that excludes the first outlier. This downsampled point cloud allows for rapid repair of the first outlier and adjacent outliers in the adjacent outlier domain. Furthermore, based on the repaired point cloud and the point cloud of normal operating data, a more accurate reduced-order model is generated, shortening simulation cycles and ensuring excellent engineering applicability.

[0212] In some embodiments of the present application, the point cloud voxelization grid filtering module 1404 is specifically used to use the point cloud voxelization grid filtering algorithm to reorder the initial serial number of the point cloud set of the abnormal working condition data in ascending order based on the comprehensive voxel index of the point cloud set of the abnormal working condition data to obtain sorted point cloud sets corresponding to different voxel units; downsampling is performed on the sorted point cloud sets corresponding to different voxel units to obtain downsampled point cloud sets composed of the first points of different voxel units; when the first outlier exists in the downsampled point cloud set, the comprehensive voxel index of the point cloud set of the abnormal working condition data is updated by updating the voxel resolution, and the reordering and downsampling steps are repeated until the downsampled point cloud set does not include the first outlier.

[0213] In some embodiments of the present application, the point cloud voxelization grid filtering module 1404 is specifically used to round down the coordinates of each dimension, the minimum coordinates of each dimension, and the voxel resolution of the point cloud set of the abnormal working condition data to obtain the voxel index in each dimension when the rate of change of the physical field in the first dimension is less than the rate of change of the physical field in the second dimension, and the rate of change of the physical field in the second dimension is less than the rate of change of the physical field in the third dimension; obtain the number of voxels in the first dimension based on the maximum coordinate, minimum coordinate, and the voxel resolution in the first dimension; obtain the number of voxels in the second dimension based on the maximum coordinate, minimum coordinate, and the voxel resolution in the second dimension; obtain the comprehensive voxel index of the point cloud set of the abnormal working condition data based on the voxel index of the point cloud set of the abnormal working condition data in the first dimension, the voxel index in the second dimension, and the number of voxels in the first dimension, as well as the voxel index in the third dimension, the number of voxels in the first dimension, and the number of voxels in the second dimension.

[0214] In some embodiments of the present application, the processing module 1405 is specifically used to use an index algorithm to mark the downsampled point cloud set at the first index position of the sorted point cloud set, and to mark the point corresponding to the first outlier at the second index position of the sorted point cloud set; use a proximity algorithm to search for the target first index position closest to the second index position; when the second index position is smaller than the target first index position, determine that the index position of the first voxel unit to which the first outlier belongs is the previous first index position of the target first index position in the downsampled point cloud set; when the second index position is greater than the target first index position, determine that the index position of the first voxel unit to which the first outlier belongs is the target first index position in the downsampled point cloud set.

[0215] In some embodiments of the present application, the neighboring outlier influence domain determination module 1406 is specifically used to determine the average value and the median based on the physical field parameters of the downsampled points corresponding to each voxel unit in the first abnormal domain and the N voxel units before and after; N is an integer greater than or equal to 1; when the absolute value of the difference between the average value and the median is greater than or equal to a first threshold, the corresponding voxel unit is confirmed as a preliminary neighboring abnormal domain; using a neighboring algorithm, search for the target third index position closest to the fourth index position; wherein the third index position is the index position of the downsampled point corresponding to the voxel unit in the preliminary neighboring abnormal domain in the sorted point cloud set, and the fourth index position is the index position of the downsampled point of the first voxel unit in the sorted point cloud set; when the difference between the fourth index position and the target third index position is less than or equal to a second threshold, the preliminary neighboring abnormal domain is confirmed as a neighboring abnormal domain.

[0216] In some embodiments of the present application, the neighboring outlier impact domain determination module 1406 is further configured to identify the preliminary neighboring outlier domain as a noise point when the difference between the fourth index position and the target third index position is greater than the second threshold.

[0217] In some embodiments of the present application, the outlier repair domain determination module 1407 is specifically used to determine the average value and the median based on the physical field parameters of the downsampled points corresponding to each voxel unit and the previous and next N voxel units in the first outlier domain; N is an integer greater than or equal to 1; when the absolute value of the difference between the average value and the median is less than or equal to a third threshold, the corresponding voxel unit is used as the third voxel unit, and the area composed of the third voxel unit is used as the point cloud outlier repair area.

[0218] In some embodiments of the present application, the outlier repair module 1408 is specifically used to use a proximity algorithm to search for a third voxel unit that is closest to the second voxel unit; when the difference between the physical field parameters of the downsampled point of the second voxel unit and the physical field parameters of the downsampled point of the third voxel unit is less than or equal to a fourth threshold, the physical field parameters of the downsampled point of the third voxel unit are updated with the outlier value in the second voxel unit using a replacement function to obtain the repaired point cloud set.

[0219] In some embodiments of the present application, the reduced-order model generation module 1409 is specifically used to take different operating condition data as input items, and take the point cloud set of the repaired abnormal operating condition data and the point cloud set of the normal operating condition data as output items, and use neural networks and deep learning algorithms to construct a reduced-order model corresponding to the traction system, so as to realize full-field visual monitoring of the traction system under different line operation scenarios.

[0220] refer to Figure 15 , Figure 15 Schematic diagram of the structure of a digital twin model of a traction system in an embodiment of the present application, wherein the digital twin model 150 includes: a vehicle dynamics model 1501, a lumped parameter equivalent digital circuit model 1502, and a multi-field reduced-order model 1503;

[0221] The vehicle dynamics model is at least used to simulate and obtain a first result; the first result includes: motor speeds corresponding to different motor torques, and the first result is transmitted to the lumped parameter equivalent digital circuit model;

[0222] The lumped parameter equivalent digital circuit model is at least used to simulate and obtain a second result based on the first result, and transmit the second result to the multi-field reduced-order model; the second result includes one or more of the following results: motor-related electrical parameters, converter-related electrical parameters, and converter electromagnetic losses; the converter electromagnetic losses include one or more of the following losses: insulated gate bipolar transistor (IGBT) losses, reactor losses, transformer losses, copper bus losses, and busbar losses;

[0223] The multi-field reduced-order model is at least used to simulate and obtain a third result based on the second result; the third result includes one or more of the following results: motor electromagnetic loss, motor temperature, converter temperature, and system structural stress; wherein, the outlier repair processing operation used in the establishment of the multi-field reduced-order model is obtained by the method provided in any embodiment of the present application.

[0224] The digital twin model provided in the embodiment of the present application can allow multiple models to participate in the simulation together. Multiple system parameters can be obtained in one simulation, which can comprehensively reflect the system status. In addition, the simulation results of each model can influence each other, which can improve the simulation accuracy of each model.

[0225] In an embodiment of the present application, the multi-field reduced-order model further includes: a motor electromagnetic loss reduced-order model, a motor temperature field reduced-order model, and a stress field reduced-order model.

[0226] In the embodiment of the present application, the output frequency of the lumped parameter equivalent digital circuit model is 10 5 -10 7Hz, the input frequency of the multi-field reduced-order model is 0.1-1 Hz, and the lumped parameter equivalent digital circuit model and the multi-field reduced-order model achieve temporal matching between the lumped parameter equivalent digital circuit model output data and the reduced-order model input data through a multi-time-scale data matching interface.

[0227] The present application also provides another electronic device. Figure 16 This is a schematic diagram of the structure of the electronic device in the embodiment of the present application. Figure 16 As shown, the electronic device 160 includes: a processor 1601 and a memory 1602 configured to store a computer program that can be run on the processor;

[0228] The processor 1601 is configured to execute the method steps in the aforementioned embodiment when running a computer program.

[0229] Of course, in actual application, Figure 16 As shown, the various components in the electronic device 160 are coupled together via a bus system 1603. It is understood that the bus system 1603 is used to achieve connection and communication between these components. In addition to the data bus, the bus system 1603 also includes a power bus, a control bus, and a status signal bus. However, for the sake of clarity, Figure 16 Various buses are labeled as bus system 1603.

[0230] In practical applications, the processor may be at least one of an application-specific integrated circuit (ASIC), a digital signal processing device (DSPD), a programmable logic device (PLD), a field-programmable gate array (FPGA), a controller, a microcontroller, and a microprocessor. It is understood that for different devices, the electronic device used to implement the functions of the processor may also be other, and the embodiments of the present application do not specifically limit this.

[0231] The above-mentioned memory can be a volatile memory (volatile memory), such as a random-access memory (RAM); or a non-volatile memory (non-volatile memory), such as a read-only memory (ROM), a flash memory, a hard disk (HDD) or a solid-state drive (SSD); or a combination of the above types of memory, and provides instructions and data to the processor.

[0232] In an exemplary embodiment, the present application further provides a computer-readable storage medium for storing a computer program.

[0233] Optionally, the computer-readable storage medium can be applied to any one of the methods in the embodiments of the present application, and the computer program enables the computer to execute the corresponding processes implemented by the processor in each method in the embodiments of the present application. For the sake of brevity, they are not repeated here.

[0234] Illustratively, an embodiment of the present application further provides a computer program product, including a computer program, which can be executed by a processor of an electronic device to complete the steps of any of the aforementioned methods.

[0235] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely schematic. For example, the division of the units is merely a logical function division. In actual implementation, there may be other division methods, such as: multiple units or components can be combined, or can be integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the components shown or discussed can be through some interfaces, and the indirect coupling or communication connection of the devices or units can be electrical, mechanical or other forms.

[0236] The units described above as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units; some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0237] In addition, the functional units in the embodiments of the present invention can all be integrated into one processing module, or each unit can be a separate unit, or two or more units can be integrated into one unit; the above-mentioned integrated unit can be implemented in the form of hardware or in the form of hardware plus software functional units. It can be understood by those skilled in the art that all or part of the steps of the above-mentioned method embodiments can be completed by hardware related to program instructions, and the above-mentioned program can be stored in a computer-readable storage medium. When the program is executed, it executes the steps of the above-mentioned method embodiments; and the above-mentioned storage medium includes various media that can store program codes, such as mobile storage devices, read-only memories (ROMs), random access memories (RAMs), magnetic disks or optical disks.

[0238] The methods disclosed in the several method embodiments provided in this application can be arbitrarily combined without conflict to obtain new method embodiments.

[0239] The features disclosed in the several product embodiments provided in this application can be arbitrarily combined without conflict to obtain new product embodiments.

[0240] The features disclosed in the several method or device embodiments provided in this application can be arbitrarily combined without conflict to obtain new method embodiments or device embodiments.

[0241] The above description is only a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with this technical field can easily think of changes or replacements within the technical scope disclosed by the present invention, which should be covered by the scope of protection of the present invention.

Claims

1. A method for repairing outliers in a digital twin reduced-order model based on point cloud recognition, characterized in that: The method comprises: Establishing a finite element simulation model of the traction system, using different working condition data as input to the finite element simulation model, and obtaining a point cloud set containing the three-dimensional spatial coordinates and physical field parameters of the traction system corresponding to the different working condition data; Using the constructed physical field anomaly criterion model, based on the preset physical field parameter threshold, the point cloud sets of different working condition data are screened to obtain the point cloud sets of abnormal working condition data; Using an interquartile range algorithm, performing outlier detection processing on the physical field parameters of the point cloud set of the abnormal working condition data, and defining the physical field parameters that exceed the fence range as the first outlier; Using a point cloud voxelization grid filtering algorithm, based on the comprehensive voxel index of the point cloud set of the abnormal operating condition data, the initial sequence number of the point cloud set of the abnormal operating condition data is reordered, and then downsampling is performed on the sorted point cloud sets corresponding to different voxel units to obtain a downsampled point cloud set composed of the first points of different voxel units; wherein the downsampled point cloud set does not include the first outlier, and the comprehensive voxel index is indexed in the coordinate dimension with the smallest rate of change of the physical field, with the highest priority. determining a first voxel unit to which the first outlier belongs; Establishing a first abnormal domain of the largest continuous enclosing area based on the boundary of the first voxel unit, and obtaining a neighboring abnormal domain affected by the first abnormal value by using the constructed neighboring abnormal value influence domain determination function; Based on the first voxel unit and the voxel units in the adjacent abnormal domain, obtaining a second voxel unit containing all abnormal values; Based on the first abnormal domain, using the constructed abnormal value repair domain judgment function, a point cloud abnormal value repair area composed of third voxel units is obtained; Repairing the outliers in the second voxel unit based on the physical field parameters of the point cloud outlier repair area to obtain a repaired point cloud set; Based on the repaired point cloud set and the point cloud set of normal operating data, a corresponding reduced-order model is generated to achieve full-field visual monitoring of the traction system under different line operation scenarios.

2. The method according to claim 1, characterized in that The point cloud voxelization grid filtering algorithm is used to reorder the initial sequence numbers of the point cloud set of the abnormal working condition data based on the comprehensive voxel index of the point cloud set of the abnormal working condition data, and then downsample the sorted point cloud sets corresponding to different voxel units to obtain a downsampled point cloud set composed of the first points of different voxel units, including: Using the point cloud voxelization grid filtering algorithm, based on the comprehensive voxel index of the point cloud set of the abnormal working condition data, the initial sequence numbers of the point cloud set of the abnormal working condition data are reordered in ascending order to obtain sorted point cloud sets corresponding to different voxel units; Downsampling is performed on the sorted point cloud sets corresponding to different voxel units to obtain downsampled point cloud sets consisting of the first points of different voxel units; When the first outlier exists in the downsampled point cloud set, the voxel resolution is updated, the comprehensive voxel index of the point cloud set of the abnormal working condition data is updated, and the reordering and downsampling steps are repeated until the downsampled point cloud set does not include the first outlier.

3. The method according to claim 1 or 2, characterized in that The method further comprises: When the rate of change of the physical field in the first dimension is less than the rate of change of the physical field in the second dimension, and the rate of change of the physical field in the second dimension is less than the rate of change of the physical field in the third dimension, the coordinates of each dimension, the minimum coordinates in each dimension, and the voxel resolution of the point cloud set of the abnormal operating condition data are rounded down to obtain a voxel index in each dimension; Obtaining the number of voxels in the first dimension based on the maximum coordinate, the minimum coordinate, and the voxel resolution in the first dimension; Obtaining the number of voxels in the second dimension based on the maximum coordinate, the minimum coordinate, and the voxel resolution in the second dimension; Based on the voxel index of the point cloud set of the abnormal working condition data in the first dimension direction, the voxel index in the second dimension direction and the number of voxels in the first dimension direction, as well as the voxel index in the third dimension direction, the number of voxels in the first dimension direction and the number of voxels in the second dimension direction, the comprehensive voxel index of the point cloud set of the abnormal working condition data is obtained.

4. The method according to claim 1 or 2, characterized in that The determining the first voxel unit to which the first abnormal value belongs includes: Using an indexing algorithm, marking the downsampled point cloud set at a first index position of the sorted point cloud set, and marking the point corresponding to the first outlier at a second index position of the sorted point cloud set; Using a proximity algorithm, searching for the target first index position closest to the second index position; When the second index position is smaller than the target first index position, determining the index position of the first voxel unit to which the first outlier belongs to be a first index position preceding the target first index position in the downsampled point cloud set; In a case where the second index position is greater than the target first index position, the index position of the first voxel unit to which the first abnormal value belongs is determined to be the target first index position in the downsampled point cloud set.

5. The method according to claim 1 or 2, characterized in that The step of establishing a first abnormal domain of the largest continuous enclosing area based on the boundary of the first voxel unit, and obtaining an adjacent abnormal domain affected by the first abnormal value by using the constructed adjacent abnormal value influence domain determination function, comprises: Determine the average and median based on the physical field parameters of the downsampled points corresponding to each voxel unit and the N preceding and following voxel units in the first abnormal domain, where N is an integer greater than or equal to 1; When the absolute value of the difference between the mean value and the median value is greater than or equal to a first threshold, the corresponding voxel unit is identified as a preliminary adjacent abnormal domain; Using a proximity algorithm, search for a target third index position that is closest to the fourth index position; wherein the third index position is the index position of the downsampled point corresponding to the voxel unit in the preliminary neighboring abnormal domain in the sorted point cloud set, and the fourth index position is the index position of the downsampled point of the first voxel unit in the sorted point cloud set; In a case where the difference between the fourth index position and the target third index position is less than or equal to a second threshold, the preliminary adjacent abnormal domain is confirmed as an adjacent abnormal domain.

6. The method according to claim 5, characterized in that The method further comprises: When the difference between the fourth index position and the target third index position is greater than the second threshold, the preliminary adjacent abnormal region is identified as a noise point.

7. The method according to claim 1 or 2, characterized in that The method of obtaining a point cloud outlier repair region composed of a third voxel unit based on the first outlier domain and utilizing the constructed outlier repair domain determination function comprises: Determine the average and median based on the physical field parameters of the downsampled points corresponding to each voxel unit and the N preceding and following voxel units in the first abnormal domain, where N is an integer greater than or equal to 1; When the absolute value of the difference between the mean value and the median value is less than or equal to the third threshold value, the corresponding voxel unit is used as the third voxel unit, and the area formed by the third voxel unit is used as the point cloud outlier repair area.

8. The method according to claim 1 or 2, characterized in that The step of repairing the outliers in the second voxel unit based on the physical field parameters of the point cloud outlier repair region to obtain a repaired point cloud set includes: Using a neighbor algorithm, searching for a third voxel unit that is closest to the second voxel unit; When the difference between the physical field parameters of the downsampled points of the second voxel unit and the physical field parameters of the downsampled points of the third voxel unit is less than or equal to a fourth threshold, the physical field parameters of the downsampled points of the third voxel unit are updated with the outliers in the second voxel unit using a replacement function to obtain the repaired point cloud set.

9. The method according to claim 1 or 2, characterized in that The method generates a corresponding reduced-order model based on the repaired point cloud set and the point cloud set of normal operating condition data to achieve full-field visual monitoring of the traction system under different line operation scenarios, including: Taking different operating condition data as input items, and the point cloud set of the repaired abnormal operating condition data and the point cloud set of normal operating condition data as output items, a reduced-order model corresponding to the traction system is constructed using a neural network and a deep learning algorithm to achieve full-field visual monitoring of the traction system under different line operation scenarios.

10. A device for repairing outliers in a digital twin reduced-order model based on point cloud recognition, characterized in that: The device comprises: A point cloud extraction module is used to establish a finite element simulation model of the traction system, take different working condition data as input to the finite element simulation model, and obtain a point cloud set containing the three-dimensional spatial coordinates and physical field parameters of the traction system corresponding to the different working condition data; A screening module is used to screen point cloud sets of different working condition data based on a preset physical field parameter threshold using the constructed physical field anomaly criterion model to obtain a point cloud set of abnormal working condition data; an outlier detection and processing module, configured to perform outlier detection processing on the physical field parameters of the point cloud set of the abnormal working condition data using an interquartile range algorithm, and define the physical field parameters that exceed the fence range as first outliers; a point cloud voxelization grid filtering module, configured to utilize a point cloud voxelization grid filtering algorithm to reorder the initial sequence number of the point cloud set of the abnormal operating condition data based on the comprehensive voxel index of the point cloud set of the abnormal operating condition data, and then perform downsampling processing on the sorted point cloud sets corresponding to different voxel units to obtain a downsampled point cloud set consisting of the first points of different voxel units; wherein the downsampled point cloud set does not include the first abnormal value, and the comprehensive voxel index is indexed in the coordinate dimension with the smallest rate of change of the physical field, with the highest priority. A processing module, configured to determine a first voxel unit to which the first abnormal value belongs; a neighboring outlier influence domain determination module, configured to establish a first outlier domain of the largest continuous enclosing area based on the boundary of the first voxel unit, and obtain a neighboring outlier domain affected by the first outlier using a constructed neighboring outlier influence domain determination function; The processing module is further configured to obtain a second voxel unit containing all abnormal values ​​based on the first voxel unit and the voxel units in the adjacent abnormal domain; an outlier repair domain determination module, configured to obtain, based on the first outlier domain and using a constructed outlier repair domain determination function, a point cloud outlier repair region composed of a third voxel unit; an outlier repair module, configured to repair the outliers in the second voxel unit based on the physical field parameters of the point cloud outlier repair region, and obtain a point cloud set of repaired abnormal working condition data; The reduced-order model generation module is used to generate a corresponding reduced-order model based on the repaired point cloud set and the point cloud set of normal operating data, so as to realize full-field visual monitoring of the traction system under different line operation scenarios.

11. A digital twin model of a traction system, characterized in that: The digital twin model includes: a vehicle dynamics model, a lumped parameter equivalent digital circuit model, and a multi-field reduced-order model; The vehicle dynamics model is at least used to simulate and obtain a first result; the first result includes: motor speeds corresponding to different motor torques, and the first result is transmitted to the lumped parameter equivalent digital circuit model; The lumped parameter equivalent digital circuit model is at least used to simulate and obtain a second result based on the first result, and transmit the second result to the multi-field reduced-order model; the second result includes one or more of the following results: motor-related electrical parameters, converter-related electrical parameters, and converter electromagnetic losses; the converter electromagnetic losses include one or more of the following losses: insulated gate bipolar transistor (IGBT) losses, reactor losses, transformer losses, copper bus losses, and busbar losses; The multi-field reduced-order model is at least used to simulate and obtain a third result based on the second result; the third result includes one or more of the following results: motor electromagnetic loss, motor temperature, converter temperature, and system structural stress; wherein the outlier repair processing operation used in the establishment of the multi-field reduced-order model is obtained by the method described in any one of claims 1 to 9.

Images