A near-infrared spectrum feature extraction method and system based on a random frog algorithm

Abstract

This invention discloses a near-infrared spectral feature extraction method and system based on the random frog-jumping algorithm, relating to the field of online monitoring and condition assessment of power equipment. The method includes: constructing an original spectral dataset, where each spectral data point corresponds to a sample index; using the random frog-jumping algorithm to filter a set of key sample indices, and determining a key cable insulation sample set based on the correspondence between sample indices, spectral data, and cable insulation samples; using a pre-built aging prediction model to generate a predicted elongation at break retention rate corresponding to each key cable insulation sample, thereby determining the aging level set corresponding to the key cable insulation sample set; further determining whether the aging level set contains each preset aging level; if so, feature extraction is completed; otherwise, the parameters of the random frog-jumping algorithm are adjusted and the sample set is re-filtered. This invention effectively solves the problem of effective feature selection under multiple sets of high-dimensional data, providing data support for the aging and deterioration assessment of XLPE cables.

Description

Technical Field

[0001] This invention relates to the field of online monitoring and condition assessment technology for power equipment, specifically to a near-infrared spectral feature extraction method and system based on the random frog-jumping algorithm. Background Technology

[0002] In recent years, cross-linked polyethylene (XLPE) cables have been widely used in power cables and electrical equipment due to their excellent electrical insulation and heat resistance. However, with increasing service life, the cable insulation material gradually ages under the combined effects of electrical, thermal, mechanical stress, and environmental factors. This aging is mainly manifested in microstructural changes such as molecular chain breakage, oxidative degradation, and water treeing in XLPE, leading to a decline in its mechanical and electrical properties. Failure to promptly assess and replace severely aged cables can result in major safety accidents such as cable breakdown, short circuits, and even fires, causing large-scale power outages and economic losses.

[0003] Traditional cable aging testing methods, such as elongation at break testing, often require complex sample preparation or destructive operations, which are not only time-consuming and labor-intensive but also difficult to conduct large-scale screening of a vast number of in-service cables. More importantly, cable aging is often localized, and traditional sampling test results cannot represent the true health status of the entire line, easily leading to missed detections or misjudgments. Near-infrared spectroscopy has broad application potential in the performance testing of cross-linked polyethylene (XLPE) cables, but its inherent defects, such as overlapping spectral information and difficulty in directly identifying characteristics, limit its direct application in cable aging assessment.

[0004] Therefore, there is an urgent need for a method that can acquire cable aging information in a non-destructive manner and intelligently select the most representative aging characteristics to solve the technical problems of low detection efficiency, poor representativeness, and difficulty in feature extraction in existing technologies. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention provides a near-infrared spectral feature extraction method and system based on the random frog jumping algorithm.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] This invention provides a near-infrared spectral feature extraction method based on a random frog-jumping algorithm, which selects key cable insulation samples that can characterize aging levels through the following steps S1 to S4:

[0008] Step S1: For a preset number of cable insulation samples with different temperatures and aging times, use a near-infrared spectrometer to obtain the near-infrared spectral lines corresponding to each cable insulation sample, and digitize each near-infrared spectral line to generate spectral data corresponding to each cable insulation sample, thereby forming an original spectral dataset, where each spectral data corresponds to a sample index.

[0009] Step S2: Based on the original spectral dataset, the key sample index set is selected using the random frog jumping algorithm for the sample index corresponding to each spectral data. The key spectral dataset is determined according to the mapping relationship between the sample index and the spectral data. Furthermore, the key cable insulation sample set is determined according to the correspondence between the spectral data and the cable insulation samples.

[0010] Step S3: For the key cable insulation sample set, using the key spectral dataset as input, and utilizing the pre-built aging prediction model, generate the predicted elongation at break retention rate for each key cable insulation sample.

[0011] Step S4: Based on the preset elongation at break retention rate range corresponding to each preset aging level, determine the aging level corresponding to each key cable insulation sample, thereby determining the aging level set corresponding to the key cable insulation sample set. Further determine whether the aging level set contains each preset aging level. If yes, feature extraction is completed; otherwise, return to step S2 to adjust the parameters of the random frog jumping algorithm and re-filter.

[0012] Furthermore, based on the random frog jumping algorithm, step S2 generates a key sample index set according to the following steps:

[0013] Step S21: Based on the sample index corresponding to each spectral data in the original spectral dataset, randomly select a preset number of sample indexes to form an initial index subset, and use the unselected sample indexes to form the remaining index subset.

[0014] Step S22: Randomly select a sample index from the remaining index subset and use this sample index to randomly replace a sample index in the initial index subset to form a candidate index subset; Calculate the average variance of the initial index subset and the candidate index subset based on each spectral data.

[0015] Step S23: Based on the calculation result of step S22, determine whether to accept the candidate index subset. If yes, update the initial index subset using the candidate index subset and update the remaining index subset; otherwise, retain the current initial index subset with a preset probability.

[0016] Step S24: Repeat steps S22 to S23 iteratively until the preset number of iterations is reached.

[0017] Furthermore, the random frog jumping algorithm outputs a set of key sample indices determined in each iteration.

[0018] Further, step S3 constructs an aging prediction model according to the following steps:

[0019] Step S31: Based on the set of key sample indices output in each generation process, count the number of times the same sample index is selected in the entire iteration process. Combined with the preset number of iterations, determine the stability frequency corresponding to each sample index, and select sample indices with stability frequencies greater than the preset stability frequency to form a stable sample index set.

[0020] Step S32: Based on the stability sample index set, determine the stable spectral data set from the original spectral dataset according to the mapping relationship between the sample index and the spectral data, and determine the stable cable insulation sample set according to the correspondence between the spectral data and the cable insulation samples, thereby obtaining the actual elongation at break retention rate corresponding to each stable cable insulation sample.

[0021] Step S33: Using each spectral data in the stable spectral data set as the independent variable and the actual elongation at break retention rate as the target variable, construct and train a regression model to obtain an aging prediction model for predicting the elongation at break retention rate of cable insulation samples.

[0022] Furthermore, the stability frequency is the proportion of the number of times the same sample index is selected into the key sample index set in multiple executions of step S2 relative to the preset number of iterations.

[0023] Furthermore, in step S3, the predicted elongation at break retention rate corresponding to each cable insulation sample with different temperature-aging time is obtained using an aging prediction model, and the elongation at break retention rate range for each preset aging level is determined by the quartile method.

[0024] Furthermore, the preset aging levels mentioned in step S4 include Level I (good), Level II (mild aging), Level III (moderate aging), and Level IV (severe aging).

[0025] Another aspect of the present invention provides a near-infrared spectral feature extraction system based on a random frog-jumping algorithm, comprising:

[0026] The data interface module is used to obtain the original spectral dataset of cable insulation samples, where each spectral data corresponds to a sample index.

[0027] The feature filtering module is used to filter out the key sample index set using variance as the evaluation index and random frog jumping algorithm, thereby determining the key cable insulation sample set.

[0028] The aging prediction module is used to generate the predicted elongation at break retention rate for each key cable insulation sample in the cable insulation sample set using a pre-built aging prediction model.

[0029] The verification module is used to determine the aging level corresponding to each key cable insulation sample and to determine whether the aging level set corresponding to the key cable insulation sample set contains each preset aging level. If so, feature extraction is completed; otherwise, the feature filtering module is used to re-filter.

[0030] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the near-infrared spectral feature extraction method based on the random frog jumping algorithm.

[0031] The present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the near-infrared spectral feature extraction method based on the random frog-jumping algorithm.

[0032] The beneficial effects of adopting the above technical solution are as follows:

[0033] (1) This invention utilizes near-infrared spectroscopy to obtain aging-related information without the need for complex sample preparation, without damaging the cable body, and without requiring the cable to be taken out of service. It effectively solves the problem that traditional elongation at break tests require sampling and destruction and cannot be used for online evaluation of in-service cables, thus achieving non-destructive, in-situ, and rapid detection of operating cables.

[0034] (2) This invention utilizes a random frog-jumping algorithm with variance as the evaluation index to automatically select key feature wavelengths that are highly correlated with aging from massive spectral data. It has low computational complexity and solves the problem of effective feature selection under multiple sets of high-dimensional data, providing data support for the aging and deterioration assessment of XLPE cables.

[0035] (3) By judging whether the selected key sample set covers all preset aging levels, the present invention effectively avoids the problem of "getting trapped in a local optimal subset" that may occur in traditional unsupervised algorithms, ensuring that the final output key sample set is both mathematically optimal (high variance) and physically covers all aging states, which significantly improves the robustness and generalization ability of the algorithm. Attached Figure Description

[0036] Figure 1 This is a flowchart of the feature extraction method of the present invention;

[0037] Figure 2 This is a schematic diagram of the near-infrared spectral lines extracted in an embodiment of the present invention;

[0038] Figure 3This is a graph showing the test results of elongation at break according to an embodiment of the present invention;

[0039] Figure 4 This is a comparison chart of absorbance and elongation at break test results at 35°C in Example 1 of the present invention;

[0040] Figure 5 This is a comparison chart of absorbance and elongation at break test results at 180°C in Example 1 of the present invention. Detailed Implementation

[0041] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings.

[0042] Example 1:

[0043] refer to Figure 1 This invention provides a near-infrared spectral feature extraction method based on the random frog-jumping algorithm, which selects key cable insulation samples that can characterize the aging level according to the following steps S1 to S4:

[0044] Step S1: For 7 groups of XLPE cable insulation samples with different aging times at 4 different temperatures, near-infrared spectrometers are used to obtain the corresponding near-infrared spectral lines for each cable insulation sample. For example... Figure 2 As shown, a total of 28 near-infrared spectral lines were obtained, and each near-infrared spectral line was digitally processed to generate spectral data corresponding to each cable insulation sample, thereby forming the original spectral dataset, in which each spectral data corresponds to a sample index.

[0045] Step S2: Based on the original spectral dataset, a set of key sample indices is selected using a random frog-jumping algorithm for the sample indices corresponding to each spectral data. The key spectral dataset is determined according to the mapping relationship between the sample indices and the spectral data. Furthermore, the key cable insulation sample set is determined according to the correspondence between the spectral data and the cable insulation samples.

[0046] Specifically, based on the random frog jumping algorithm, step S2 generates a key sample index set according to the following steps:

[0047] Step S21: Based on the sample index corresponding to each spectral data in the original spectral dataset, randomly select a preset number of sample indexes to form an initial index subset, and use the unselected sample indexes to form the remaining index subset.

[0048] Step S22: Randomly select a sample index from the remaining index subset and use this sample index to randomly replace a sample index in the initial index subset to form a candidate index subset; Calculate the average variance of the initial index subset and the candidate index subset based on each spectral data.

[0049] Step S23: Based on the calculation result of step S22, determine whether to accept the candidate index subset, that is, determine whether the average variance of the new candidate index subset is greater than the average variance of the initial index subset. If yes, update the initial index subset using the candidate index subset and update the remaining index subset; otherwise, retain the current initial index subset with a preset probability of 0.05.

[0050] Step S24: Repeat steps S22 to S23 iteratively until the preset number of iterations is reached.

[0051] In this embodiment, the key sample index set selected is [21,27,8,26,24,28,12,1], and the corresponding key cable insulation sample set is [155℃-432h,180℃-30h,135℃-168h,180℃-26h,180℃-14h,180℃-32h,135℃-840h,105℃-720h], with a final average variance of 0.0174.

[0052] Step S3: For the key cable insulation sample set, using the key spectral dataset as input, and utilizing the pre-built aging prediction model, generate the predicted elongation at break retention rate for each key cable insulation sample.

[0053] Step S4: Based on the preset elongation at break retention rate range corresponding to each preset aging level, determine the aging level corresponding to each key cable insulation sample, thereby determining the aging level set corresponding to the key cable insulation sample set. Further determine whether the aging level set contains each preset aging level. If yes, feature extraction is completed; otherwise, return to step S2 to adjust the parameters of the random frog jumping algorithm and re-filter.

[0054] Furthermore, based on the fact that the random frog jumping algorithm outputs a set of key sample indices determined in each iteration, step S3 constructs the aging prediction model according to the following steps:

[0055] Step S31: Execute the random frog-jump algorithm 100 times. Based on the key sample index set output in each iteration, count the number of times the same sample index is selected throughout the entire iteration process. Determine the stability frequency corresponding to each sample index by calculating the proportion of times the same sample index is selected into the key sample index set relative to the preset number of iterations. Select sample indices with a stability frequency greater than 50% to form a stable sample index set. Furthermore, a higher stability frequency indicates that the selected key cable insulation samples are more truly related to the material aging mechanism, rather than being selected by chance.

[0056] Step S32: Based on the stability sample index set, determine the stable spectral data set from the original spectral dataset according to the mapping relationship between the sample index and the spectral data, and determine the stable cable insulation sample set according to the correspondence between the spectral data and the cable insulation samples, thereby obtaining the actual elongation at break retention rate corresponding to each stable cable insulation sample.

[0057] Step S33: Using each spectral data in the stable spectral data set as the independent variable and the actual measured elongation at break retention rate as the target variable, construct and train a regression model to obtain an aging prediction model for predicting the elongation at break retention rate of cable insulation samples.

[0058] Furthermore, in step S3, the predicted elongation at break retention rate corresponding to each cable insulation sample with different temperature-aging time is obtained using the aging prediction model, and the elongation at break retention rate range for each preset aging level is obtained by using the quartile method, as shown in Table 1.

[0059] Table 1

[0060]

[0061] To ensure comparability of test results, this embodiment simultaneously conducted elongation at break tests on 28 groups of samples. Elongation at break is a commonly used aging test method for polymers, and is a destructive test. The test results are as follows: Figure 3 As shown.

[0062] The technical effects of the present invention will be further explained in detail below using key cable insulation samples at 135°C and 180°C as examples.

[0063] refer to Figure 4 Firstly, taking key cable insulation samples at 135°C as examples, the constructed aging prediction model shows that the elongation at break of the cable insulation samples from 135°C to 168h is 582.19%, and the elongation at break of the cable insulation samples from 135°C to 840h is 547.05%. Furthermore, in the healthy state (initial state), the elongation at break of the XLPE samples is 650%. Therefore, the elongation at break retention rate of the cable insulation samples from 135°C to 168h is 89.57%, and the elongation at break retention rate of the cable insulation samples from 135°C to 840h is 84.16%. According to Table 1, the insulation condition of all cable insulation samples at 135°C is Grade I (good), but the insulation condition of the samples from 135°C to 840h is worse than that of the samples from 135°C to 168h.

[0064] Meanwhile, the absorbance of the cable insulation samples from 135℃ to 168h was 0.8538, and the absorbance of the cable insulation samples from 135℃ to 168h was 0.6188. Referring to Table 2, it can be seen that the insulation condition of the cables at 135℃ shows a consistent trend, with all samples exhibiting a Class I (good) insulation condition.

[0065] Table 2

[0066]

[0067] refer to Figure 5 Taking key cable insulation samples at 180°C as examples, the aging prediction model shows that the elongation at break of the cable insulation sample at 180°C-14h is 611.81%, the elongation at break of the cable insulation sample at 180°C-26h is 500.20%, the elongation at break of the cable insulation sample at 180°C-30h is 119.89%, and the elongation at break of the cable insulation sample at 180°C-32h is 86.12%. Therefore, based on the 650% elongation at break of the XLPE samples in the healthy state (initial state), and referring to Table 1, the retention rate of the cable insulation samples after 180℃-14h is 94.12%, and the insulation condition is Grade I (good); the retention rate of the cable insulation samples after 180℃-26h is 76.95%, and the insulation condition is Grade II (mild aging); the retention rate of the cable insulation samples after 180℃-30h is 18.44%, and the insulation condition is Grade IV (severe aging); and the retention rate of the cable insulation samples after 180℃-32h is 13.25%, and the insulation condition is Grade IV (severe aging).

[0068] Meanwhile, the absorbance of the cable insulation sample from 180℃-14h was 0.6173, the absorbance from the sample from 180℃-26h was 0.5845, the absorbance from the sample from 180℃-30h was 0.5487, and the absorbance from the sample from 180℃-32h was 0.5098. Referring to Table 2, the insulation condition of the cable insulation sample from 180℃-14h is Class I (good); the insulation condition of the sample from 180℃-26h is Class II (slightly aged); the insulation condition of the sample from 180℃-30h is Class IV (severely aged); and the insulation condition of the sample from 180℃-32h is Class IV (severely aged).

[0069] The comparative results show that the absorbance value of the near-infrared spectrum is positively correlated with the elongation at break retention rate of the cable insulation, and the trends of both are completely consistent with the degree of aging. Therefore, the key samples screened by near-infrared spectroscopy and their absorbance characteristics can replace traditional destructive mechanical tests for accurately assessing the aging state and aging level of cable insulation.

[0070] Example 2:

[0071] This invention provides a near-infrared spectral feature extraction system based on a random frog-jumping algorithm, comprising:

[0072] The data interface module is used to obtain the original spectral dataset of cable insulation samples, where each spectral data corresponds to a sample index.

[0073] The feature filtering module is used to filter out the key sample index set using variance as the evaluation index and random frog jumping algorithm, thereby determining the key cable insulation sample set.

[0074] The aging prediction module is used to generate the predicted elongation at break retention rate for each key cable insulation sample in the cable insulation sample set using a pre-built aging prediction model.

[0075] The verification module is used to determine the aging level corresponding to each key cable insulation sample and to determine whether the aging level set corresponding to the key cable insulation sample set contains each preset aging level. If so, feature extraction is completed; otherwise, the feature filtering module is used to re-filter.

[0076] Furthermore, embodiments of the present invention also provide an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the near-infrared spectral feature extraction method based on the random frog jumping algorithm.

[0077] Meanwhile, this embodiment of the invention also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the near-infrared spectral feature extraction method based on the random frog jumping algorithm.

[0078] The above description is merely a preferred embodiment of the present invention and does not constitute any limitation on the present invention. Any simple modifications, alterations, or equivalent structural changes made to the above embodiments based on the technical essence of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A near-infrared spectral feature extraction method based on a random frog algorithm, characterized in that, Select key cable insulation samples that can characterize the aging level by following steps S1 to S4: Step S1: For a preset number of cable insulation samples with different temperatures and aging times, use a near-infrared spectrometer to obtain the near-infrared spectral lines corresponding to each cable insulation sample, and digitize each near-infrared spectral line to generate spectral data corresponding to each cable insulation sample, thereby forming an original spectral dataset, where each spectral data corresponds to a sample index. Step S2: Based on the original spectral dataset, the key sample index set is selected using the random frog jumping algorithm for the sample index corresponding to each spectral data. The key spectral dataset is determined according to the mapping relationship between the sample index and the spectral data. Furthermore, the key cable insulation sample set is determined according to the correspondence between the spectral data and the cable insulation samples. Step S3: For the key cable insulation sample set, using the key spectral dataset as input, and utilizing the pre-built aging prediction model, generate the predicted elongation at break retention rate for each key cable insulation sample. Step S4: Based on the preset elongation at break retention rate range corresponding to each preset aging level, determine the aging level corresponding to each key cable insulation sample, thereby determining the aging level set corresponding to the key cable insulation sample set. Further determine whether the aging level set contains each preset aging level. If yes, feature extraction is completed; otherwise, return to step S2 to adjust the parameters of the random frog jumping algorithm and re-filter.

2. The random frog algorithm-based near-infrared spectral feature extraction method according to claim 1, characterized in that, Based on the random frog jumping algorithm, step S2 generates a key sample index set according to the following steps: Step S21: Based on the sample index corresponding to each spectral data in the original spectral dataset, randomly select a preset number of sample indexes to form an initial index subset, and use the unselected sample indexes to form the remaining index subset. Step S22: Randomly select a sample index from the remaining index subset and use this sample index to randomly replace a sample index in the initial index subset, thereby forming a candidate index subset; Based on the spectral data, calculate the average variances of the initial index subset and the candidate index subset, respectively. Step S23: Based on the calculation result of step S22, determine whether to accept the candidate index subset. If yes, update the initial index subset using the candidate index subset and update the remaining index subset. Otherwise, the current initial index subset will be retained with a preset probability; Step S24: Repeat steps S22 to S23 iteratively until the preset number of iterations is reached.

3. The random frog algorithm-based near infrared spectral feature extraction method according to claim 2, characterized in that, The random frog jumping algorithm outputs a set of key sample indices determined in each iteration.

4. The method according to claim 3, wherein, Step S3 involves constructing an aging prediction model as follows: Step S31: Based on the set of key sample indices output in each generation process, count the number of times the same sample index is selected in the entire iteration process. Combined with the preset number of iterations, determine the stability frequency corresponding to each sample index, and select sample indices with stability frequencies greater than the preset stability frequency to form a stable sample index set. Step S32: Based on the stability sample index set, determine the stable spectral data set from the original spectral dataset according to the mapping relationship between the sample index and the spectral data, and determine the stable cable insulation sample set according to the correspondence between the spectral data and the cable insulation samples, thereby obtaining the actual elongation at break retention rate corresponding to each stable cable insulation sample. Step S33: Using each spectral data in the stable spectral data set as the independent variable and the actual elongation at break retention rate as the target variable, construct and train a regression model to obtain an aging prediction model for predicting the elongation at break retention rate of cable insulation samples.

5. The random frog algorithm-based near infrared spectral feature extraction method according to claim 4, characterized in that, The stability frequency is the proportion of the number of times the same sample index is selected into the key sample index set in multiple executions of step S2 relative to the preset number of iterations.

6. The random frog algorithm-based near infrared spectral feature extraction method according to claim 4, characterized in that, Step S3 uses an aging prediction model to obtain the predicted elongation at break retention rate for cable insulation samples with different temperatures and aging times, and uses the quartile method to determine the elongation at break retention rate range for each preset aging level.

7. The random frog algorithm-based near infrared spectral feature extraction method according to claim 1, characterized in that, The preset aging levels mentioned in step S4 include Level I (good), Level II (mild aging), Level III (moderate aging), and Level IV (severe aging).

8. A near infrared spectrum feature extraction system based on random frog algorithm, characterized in that, include: The data interface module is used to obtain the original spectral dataset of cable insulation samples, where each spectral data corresponds to a sample index. The feature filtering module is used to filter out the key sample index set using variance as the evaluation index and random frog jumping algorithm, thereby determining the key cable insulation sample set. The aging prediction module is used to generate the predicted elongation at break retention rate for each key cable insulation sample in the cable insulation sample set using a pre-built aging prediction model. The verification module is used to determine the aging level corresponding to each key cable insulation sample and to determine whether the aging level set corresponding to the key cable insulation sample set contains each preset aging level. If so, feature extraction is completed; otherwise, the feature filtering module is used to re-filter.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method as described in any one of claims 1 to 7.

Images