CN117367751A

CN117367751A - Performance detection method and device for ultra-pulse thulium-doped laser

Info

Publication number: CN117367751A
Application number: CN202311357209.9A
Authority: CN
Inventors: 韦博仁; 梁瑶; 徐剑秋; 金鑫; 陈智勇; 邱武; 邓小鸿; 邹清红; 雷亚康
Original assignee: Jointo Technology Co ltd
Current assignee: Jointo Technology Co ltd
Priority date: 2023-10-19
Filing date: 2023-10-19
Publication date: 2024-01-09
Anticipated expiration: 2043-10-19
Also published as: CN117367751B

Abstract

The invention relates to the field of artificial intelligence, and discloses a performance detection method and device of an ultra-pulse thulium-doped laser, which are used for improving the performance detection accuracy of the ultra-pulse thulium-doped laser. The method comprises the following steps: performing performance test on the ultra-pulse thulium doped laser to obtain initial performance test data, and preprocessing the initial performance test data to obtain target performance test data; inputting the target performance test data into a preset target Gaussian mixture model for performance parameter grouping and data distribution analysis to obtain a plurality of performance parameter Gaussian distributions; extracting parameter distribution characteristics and converting characteristics of the Gaussian distribution of the plurality of performance parameters to obtain a target performance characteristic data set; and inputting the target performance characteristic data set into a preset extreme learning machine model to perform laser performance detection analysis, and outputting a target performance detection result.

Description

Performance detection method and device for ultra-pulse thulium-doped laser

Technical Field

The invention relates to the field of artificial intelligence, in particular to a performance detection method and device of an ultra-pulse thulium-doped laser.

Background

The super-pulse thulium-doped laser is a device widely applied in the technical field of laser and is generally used in the fields of laser radar, laser communication, medical laser, scientific research and the like. Such lasers typically have high energy, short pulse widths, and relatively narrow spectral bandwidths, making them particularly advantageous in a variety of applications. However, the performance of an ultra-pulsed thulium doped laser can be affected by a number of factors, including environmental changes, device aging, non-uniformities in the optical components, and the like. Therefore, accurate and timely detection and monitoring of its performance becomes critical. Performance detection helps to find potential problems early, perform preventive maintenance, and improve stability and reliability of the laser.

In conventional laser performance testing, raw data is typically acquired using sensors and then analyzed by statistical and mathematical modeling methods. However, due to the complexity and non-linear nature of the laser system, a single detection method cannot fully reveal the performance condition of the laser, i.e. the accuracy of the existing scheme is low.

Disclosure of Invention

The invention provides a performance detection method and device of an ultra-pulse thulium-doped laser, which are used for improving the performance detection accuracy of the ultra-pulse thulium-doped laser.

The first aspect of the invention provides a performance detection method of an ultra-pulse thulium-doped laser, which comprises the following steps:

performing performance test on the ultra-pulse thulium doped laser to obtain initial performance test data, and preprocessing the initial performance test data to obtain target performance test data;

inputting the target performance test data into a preset target Gaussian mixture model for performance parameter grouping and data distribution analysis to obtain a plurality of performance parameter Gaussian distributions;

extracting parameter distribution characteristics and converting characteristics of the Gaussian distribution of the plurality of performance parameters to obtain a target performance characteristic data set;

and inputting the target performance characteristic data set into a preset extreme learning machine model to perform laser performance detection analysis, and outputting a target performance detection result.

With reference to the first aspect, in a first implementation manner of the first aspect of the present invention, performing a performance test on the super-pulse thulium doped laser to obtain initial performance test data, and preprocessing the initial performance test data to obtain target performance test data, where the performance test data includes:

performing performance test on the ultra-pulse thulium-doped laser, and acquiring performance test data of the ultra-pulse thulium-doped laser through a preset multi-channel sensor group to obtain corresponding initial performance test data;

Performing laser performance influence factor correlation analysis on the initial performance test data to obtain a parameter correlation analysis result;

according to the parameter correlation analysis result, carrying out data standardization processing on the initial performance test data to obtain standard performance test data;

and inputting the standard performance test data into a preset LSTM neural network to perform parameter time sequence association to obtain target performance test data.

With reference to the first aspect, in a second implementation manner of the first aspect of the present invention, inputting the target performance test data into a preset target gaussian mixture model to perform performance parameter grouping and data distribution analysis, to obtain a plurality of performance parameter gaussian distributions, including:

acquiring a plurality of training performance test data and an initial Gaussian mixture model, and optimizing model parameters of the initial Gaussian mixture model according to the plurality of training performance test data to obtain a target Gaussian mixture model;

performing performance parameter grouping on a plurality of target data points in the target performance test data through the target Gaussian mixture model, and determining at least one Gaussian distribution corresponding to each target data point;

And according to at least one Gaussian distribution corresponding to each target data point, probability density distribution mapping is carried out on the target performance test data through a preset probability density distribution function, and a plurality of performance parameter Gaussian distributions are obtained.

With reference to the first aspect, in a third implementation manner of the first aspect of the present invention, the obtaining a plurality of training performance test data and an initial gaussian mixture model, and performing model parameter optimization on the initial gaussian mixture model according to the plurality of training performance test data, to obtain a target gaussian mixture model includes:

acquiring a plurality of training performance test data, calculating the quantity of mixed components of the plurality of training performance test data according to the parameter correlation analysis result, and determining the number of Gaussian distribution in an initial Gaussian mixture model according to the quantity of mixed components;

determining a plurality of corresponding Gaussian distributions according to the Gaussian distribution number, and performing posterior probability calculation on each training data point in the plurality of training performance test data to obtain posterior probability data;

performing hidden variable analysis on each training data point in the training performance test data according to the posterior probability data to obtain hidden variable data;

Parameter updating is carried out on the plurality of Gaussian distributions according to the hidden variable data, whether the initial Gaussian mixture model is converged is judged through a preset log likelihood function, and if so, the outputting of corresponding target model parameters comprises the following steps: mean, covariance matrix and mixing coefficients;

and carrying out model parameter optimization on the initial Gaussian mixture model based on the target model parameters to generate a target Gaussian mixture model.

With reference to the first aspect, in a fourth implementation manner of the first aspect of the present invention, the performing parameter distribution feature extraction and feature conversion on the plurality of performance parameter gaussian distributions to obtain a target performance feature data set includes:

the distribution extracts parameter distribution characteristics of the Gaussian distribution of the plurality of performance parameters to obtain a plurality of first parameter distribution characteristics;

calculating confidence intervals of Gaussian distribution of each performance parameter in a distribution mode, and carrying out feature verification on the first parameter distribution features based on the confidence intervals to obtain a plurality of feature verification results;

according to the feature verification results, feature selection is carried out on the first parameter distribution features to obtain second parameter distribution features;

And performing feature dimension reduction processing on the second parameter distribution features through a preset kernel principal component analysis algorithm to obtain a target performance feature data set.

With reference to the first aspect, in a fifth implementation manner of the first aspect of the present invention, the performing, by using a preset kernel principal component analysis algorithm, feature dimension reduction processing on the plurality of second parameter distribution features to obtain a target performance feature data set includes:

performing matrix conversion on the second parameter distribution characteristics based on a preset kernel principal component analysis algorithm to obtain an initial characteristic matrix, and performing matrix operation on the initial characteristic matrix through a preset kernel function to obtain a corresponding initial kernel matrix;

centering the initial kernel matrix to obtain a target kernel matrix, and decomposing the characteristic value of the target kernel matrix to obtain a corresponding characteristic value and a characteristic vector;

and selecting a corresponding main component according to the magnitude of the characteristic value, and mapping the plurality of second parameter distribution characteristics to a new characteristic space according to the main component to obtain a target performance characteristic data set.

With reference to the first aspect, in a sixth implementation manner of the first aspect of the present invention, inputting the target performance feature dataset into a preset extreme learning machine model to perform laser performance detection analysis, and outputting a target performance detection result includes:

Inputting the target performance characteristic data set into a preset extreme learning machine model, wherein the extreme learning machine model comprises: a singular spectrum analysis network and an extreme learning machine network;

carrying out data set decomposition on the target performance characteristic data set through the singular spectrum analysis network to obtain a plurality of performance characteristic subsequences;

inputting the plurality of performance characteristic subsequences into the extreme learning machine network for performance detection analysis to obtain an initial performance detection result corresponding to each performance characteristic subsequence;

and carrying out weighted fusion on the initial performance detection results corresponding to each performance characteristic subsequence to obtain the target performance detection results corresponding to the ultra-pulse thulium-doped laser.

The second aspect of the present invention provides a performance detection apparatus for an ultra-pulse thulium-doped laser, the performance detection apparatus for an ultra-pulse thulium-doped laser comprising:

the preprocessing module is used for performing performance test on the ultra-pulse thulium-doped laser to obtain initial performance test data, and preprocessing the initial performance test data to obtain target performance test data;

the analysis module is used for inputting the target performance test data into a preset target Gaussian mixture model to perform performance parameter grouping and data distribution analysis to obtain a plurality of performance parameter Gaussian distributions;

The conversion module is used for extracting parameter distribution characteristics and converting characteristics of the plurality of performance parameter Gaussian distributions to obtain a target performance characteristic data set;

and the output module is used for inputting the target performance characteristic data set into a preset extreme learning machine model to perform laser performance detection analysis and outputting a target performance detection result.

A third aspect of the present invention provides a performance detection apparatus for an ultra-pulsed thulium doped laser, comprising: a memory and at least one processor, the memory having instructions stored therein; the at least one processor invokes the instructions in the memory to cause the performance detection device of the ultra-pulse thulium-doped laser to perform the performance detection method of the ultra-pulse thulium-doped laser described above.

A fourth aspect of the present invention provides a computer readable storage medium having instructions stored therein which, when run on a computer, cause the computer to perform the above-described method of performance detection of an ultra-pulsed thulium doped laser.

In the technical scheme provided by the invention, performance test is carried out on the ultra-pulse thulium doped laser to obtain initial performance test data, and the initial performance test data is preprocessed to obtain target performance test data; inputting the target performance test data into a preset target Gaussian mixture model for performance parameter grouping and data distribution analysis to obtain a plurality of performance parameter Gaussian distributions; extracting parameter distribution characteristics and converting characteristics of the Gaussian distribution of the plurality of performance parameters to obtain a target performance characteristic data set; the invention adopts LSTM neural network to carry out data association, can better process time sequence performance data of the super-pulse thulium doped laser, is beneficial to preserving time sequence associated information in performance test data and improves sensitivity to laser performance change. The Gaussian mixture model is used for grouping and data distribution analysis of target performance test data, modeling is effectively carried out on a plurality of performance parameters, and complex characteristics of the performance of the ultra-pulse thulium-doped laser can be captured more accurately. The correlation and important characteristics among the performance parameters can be captured better by carrying out feature extraction and dimension reduction on the Gaussian distribution of the plurality of performance parameters through the analysis of the kernel principal components, and the generalization capability of the model is improved. The method has the advantages that the extreme learning machine is utilized to perform performance detection analysis, wherein the analysis and the performance detection analysis of a data set are better processed by the singular spectrum analysis network and the extreme learning machine network, and the information of different performance characteristics can be more effectively integrated by adopting a mode of weighting and fusing initial performance detection results of each performance characteristic subsequence, so that the performance detection accuracy of the ultra-pulse thulium-doped laser is improved.

Drawings

FIG. 1 is a schematic diagram of an embodiment of a performance detection method of an ultra-pulse thulium-doped laser according to an embodiment of the present invention;

FIG. 2 is a flow chart of performance parameter grouping and data distribution analysis in an embodiment of the present invention;

FIG. 3 is a flow chart of model parameter optimization in an embodiment of the invention;

FIG. 4 is a flow chart of parameter distribution feature extraction and feature transformation in an embodiment of the present invention;

FIG. 5 is a schematic diagram of an embodiment of a performance detection apparatus of an ultra-pulse thulium-doped laser according to an embodiment of the present invention;

fig. 6 is a schematic diagram of an embodiment of a performance detection apparatus of an ultra-pulse thulium-doped laser according to an embodiment of the present invention.

Detailed Description

The embodiment of the invention provides a performance detection method and device of an ultra-pulse thulium-doped laser, which are used for improving the performance detection accuracy of the ultra-pulse thulium-doped laser. The terms "first," "second," "third," "fourth" and the like in the description and in the claims and in the above drawings, if any, are used for distinguishing between similar objects and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used may be interchanged where appropriate such that the embodiments described herein may be implemented in other sequences than those illustrated or otherwise described herein. Furthermore, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, system, article, or apparatus that comprises a list of steps or elements is not necessarily limited to those steps or elements expressly listed or inherent to such process, method, article, or apparatus.

For easy understanding, a specific flow of an embodiment of the present invention is described below, referring to fig. 1, and an embodiment of a performance detection method of an ultra-pulse thulium doped laser in an embodiment of the present invention includes:

s101, performing performance test on an ultra-pulse thulium-doped laser to obtain initial performance test data, and preprocessing the initial performance test data to obtain target performance test data;

it is understood that the execution body of the present invention may be a performance detection device of an ultra-pulse thulium doped laser, and may also be a terminal or a server, which is not limited herein. The embodiment of the invention is described by taking a server as an execution main body as an example.

Specifically, the server uses a preset multi-channel sensor group to perform performance test on the ultra-pulse thulium-doped laser so as to obtain initial performance test data. These sensors can measure various parameters related to laser performance, such as energy output, pulse width, frequency stability, etc. These data will be used for subsequent analysis. The server performs a parameter correlation analysis. This step aims at determining which performance parameters have a correlation with each other. For example, the output energy of a laser is related to the pulse width, or to the frequency stability. The parameter correlation analysis will help the server understand these associations to better understand the performance data. And then, the server performs data standardization processing on the initial performance test data according to the result of the parameter correlation analysis. The data normalization is to put the data of different performance parameters on the same scale so that the neural network model can better understand and process the data. Normalization typically involves converting the data into a standard normal distribution with a mean of zero and a variance of one. The server inputs standard performance test data into a preset LSTM (long short time memory) neural network to carry out parameter time sequence association. The LSTM neural network is a deep learning model suitable for sequential data, and can capture the time sequence relationship between the data. And inputting the standardized data into the LSTM neural network, and predicting time sequence correlation among the performance parameters by the server so as to obtain target performance test data. For example, assume that a server is testing an ultra-pulsed thulium doped laser and that its output energy, pulse width and frequency stability are measured using sensors. Through the parameter correlation analysis, the server finds that a certain correlation exists between the output energy and the pulse width, namely the output energy is influenced by the pulse width. The server performs a normalization process on these data to ensure that they have the same dimensions. The server inputs the normalized data into an LSTM neural network that is capable of learning a timing relationship between output energy, pulse width, and frequency stability, and generating target performance test data.

S102, inputting target performance test data into a preset target Gaussian mixture model for performance parameter grouping and data distribution analysis to obtain a plurality of performance parameter Gaussian distributions;

specifically, the server prepares a plurality of training performance test data sets and an initial Gaussian mixture model. These training data sets, which contain test data for lasers of known performance, can be used to train a model of the server. Meanwhile, the initial Gaussian mixture model is a basic model, and can be used for subsequent parameter optimization. And the server optimizes the model parameters. This step is to refine the initial gaussian mixture model to make it more suitable for server applications. By adjusting and optimizing model parameters for multiple training performance test datasets, the server obtains a more accurate target Gaussian mixture model. This target model will be used for subsequent data distribution analysis. Once the target Gaussian mixture model is present, the server begins to analyze the target performance test data. The target performance test data is input into a target-si hybrid model that will divide the data points into different sets of performance parameters according to the characteristics of the data. Each group corresponds to an aspect of laser performance, such as output energy, pulse width, etc. Each target data point corresponds to at least one gaussian distribution. This is because the performance of the laser exhibits different profiles under different conditions. For example, at different temperatures, the performance of the lasers is different, so that the same performance parameter corresponds to a plurality of gaussian distributions, each distribution representing performance under one temperature condition. The server uses a preset probability density distribution function to map the probability density distribution of the target performance test data. This step will help the server to learn the data distribution of each performance parameter, including the mean, variance, and shape of the distribution. With this mapping, the server better understands the variation of laser performance under different conditions. For example, assume that there are multiple training performance test data sets, each data set corresponding to a performance test at a different temperature. The server obtains a target-specification hybrid model through model parameter optimization, and then inputs target performance test data into the model. Based on analysis of the target-gaussian mixture model, the server finds that the output energy has different gaussian distributions at different temperatures, and the pulse width is similar. The server then maps these distributions into a probability density map using a probability density distribution function, and the trend of the output energy and pulse width at different temperatures can be clearly seen.

The server obtains a plurality of training performance test data sets and an initial gaussian mixture model. These training data sets contain laser performance test data under different conditions, such as different temperatures, humidity or other environmental parameters. Meanwhile, according to the result of the parameter correlation analysis, the server calculates the number of mixed components of each training data set. This amount of mixed components reflects the different distributions exhibited by the laser performance under different conditions. And determining the number of Gaussian distributions in the initial Gaussian mixture model according to the number of the mixed components. Each gaussian represents a performance profile. For example, if the server determines that there are three blending component amounts, then the initial gaussian blending model contains three gaussian distributions. The server calculates posterior probabilities for each training data point in the plurality of training performance test data sets according to the number of gaussian distributions. The posterior probability reflects the probability that each data point belongs to each gaussian distribution. This calculation is achieved by using an Expectation-Maximization (EM) algorithm. The server then performs a hidden variable analysis. The hidden variables refer to the gaussian distribution labels corresponding to each training data point, which represent from which gaussian distribution each data point comes. These hidden variable data will be used to update parameters of the gaussian distribution, including the mean, covariance matrix, and mixing coefficients. And updating parameters by the server through the hidden variable data. This is achieved by adjusting parameters of the gaussian distribution according to the hidden variable data of each gaussian distribution. The updated parameters reflect the actual distribution of the performance data. The server uses a preset log-likelihood function to judge whether the initial Gaussian mixture model is converged. If the model converges, it means that the model of the server is already better able to fit the performance data. In this case, the server outputs the target model parameters, including the mean, covariance matrix, and mixing coefficients. For example, assume that there are multiple training performance test data sets, each data set corresponding to a laser performance test at a different temperature. The server first calculates the number of mixed components and finds three mixed component numbers. Thus, the server sets three gaussian distributions in the initial gaussian mixture model. The server then calculates the posterior probability for each training data point, and the hidden variable for each data point, using the EM algorithm. With these hidden variables, the server updates the parameters of each gaussian distribution to better fit the performance data at different temperatures. The server uses a log-likelihood function to check whether the model converges. If the model converges, the server obtains the parameters of the target-S hybrid model, and the model can more accurately describe the distribution condition of the performance of the ultra-pulse thulium-doped laser under different conditions.

S103, extracting parameter distribution characteristics and converting characteristics of the Gaussian distribution of the plurality of performance parameters to obtain a target performance characteristic data set;

the server performs parameter distribution feature extraction on the plurality of performance parameter gaussian distributions. Statistical information describing the characteristics of each gaussian is extracted from the gaussian. Typically, features that can be extracted include mean, variance, skewness, kurtosis, etc., which are statistical features that help describe the central trend, dispersion, and shape of the distribution. The server calculates a confidence interval for each performance parameter gaussian distribution. The confidence interval represents the degree of confidence in the server's estimate for each statistical feature. For example, a 95% confidence interval for the mean may be calculated, indicating that the mean falls within this interval with a probability of 95%. These confidence intervals are used for subsequent feature verification. Based on the confidence interval, the server performs feature verification on the plurality of first parameter distribution features. This step is used to determine whether the characteristics of each performance parameter distribution are statistically significant. For example, if the confidence interval contains zero, then the mean feature is not significant. The result of the feature verification is a plurality of feature verification results indicating which features should be preserved. And according to the multiple feature verification results, the server performs feature selection. In this step, the server decides which features are to be retained and which are to be discarded based on the result of the feature verification. Only features of statistical significance will be included in the plurality of second parameter distribution features. And performing feature dimension reduction processing on the second parameter distribution features through a preset kernel principal component analysis algorithm to obtain a target performance feature data set. Nuclear principal component analysis is a dimension reduction technique that can combine multiple features into fewer features while preserving the dominant pattern of change of the data. These new features constitute a target performance feature dataset for describing key features of laser performance. For example, assume that there is a performance test dataset of an ultra-pulsed thulium doped laser, including measurements of output energy, pulse width, and frequency stability. The server extracts statistical features, such as mean and variance, from the gaussian distribution of each performance parameter. Confidence intervals are then calculated for judging the significance of these features. The server performs feature verification to find that the mean feature of the output energy is not significant in the confidence interval, and the mean feature of the pulse width and frequency stability is significant. Thus, the server chooses to preserve the mean feature of pulse width and frequency stability. Through the analysis of the principal components of the kernel, the server combines the two features into a new feature, which constitutes the target performance feature dataset for describing the main features of the laser performance.

The server performs matrix conversion on the second parameter distribution characteristics based on a preset kernel principal component analysis algorithm to obtain an initial characteristic matrix. Nuclear principal component analysis is a technique for dimension reduction that maps high-dimensional data into low-dimensional space, preserving the dominant pattern of change of the data. This mapping is achieved by computing a kernel matrix. And the server carries out matrix operation on the initial feature matrix through a preset kernel function to obtain a corresponding initial kernel matrix. The kernel function is a function used to measure the similarity between two data points. It can map data points to a higher dimensional space to better capture the relationships between the data. The choice of kernel function depends on the nature of the data and the requirements of the problem. Then, the server performs centering processing on the initial core matrix to obtain a target core matrix. Centering refers to moving the data points to a space centered on the mean to reduce the shift of the data. The centralized core matrix helps to better capture the intrinsic structure of the data. And the server performs eigenvalue decomposition on the target kernel matrix to obtain corresponding eigenvalues and eigenvectors. The eigenvalues and eigenvectors describe the dominant pattern of change of the data in the new feature space. The feature value represents the importance of each feature vector, with larger feature values corresponding to more important features. And selecting a corresponding main component by the server according to the magnitude of the characteristic value. The principal components are eigenvectors, which correspond to the largest eigenvalues. The server then uses these principal components to map a plurality of second parameter distribution features to a new feature space, resulting in a target performance feature dataset. This new feature space typically has a lower dimension but retains the main information of the data. For example, assuming a plurality of second parameter profile characteristics, variations in the performance of an ultra-pulsed thulium doped laser under different conditions are described. The server uses a kernel principal component analysis algorithm to map these features to a new feature space. Then, the server obtains the eigenvalue and eigenvector by eigenvalue decomposition, and selects the principal component. If the server selects the first two principal components, the server maps a plurality of second parameter distribution features to a new feature space that is two-dimensional. This new feature space may better help the server understand the performance patterns of the data and may be further analyzed and visualized in a low dimension. The method for analyzing the nuclear principal component is helpful for reducing the data dimension, and simultaneously retains important information of the data, so that the performance characteristics of the ultra-pulse thulium-doped laser are better understood.

S104, inputting the target performance characteristic data set into a preset extreme learning machine model to perform laser performance detection analysis, and outputting a target performance detection result.

Specifically, the server inputs the target performance characteristic dataset into a preset extreme learning machine model. This extreme learning machine model includes two parts: a singular spectrum analysis network and an extreme learning machine network. The two parts work cooperatively to improve the accuracy and reliability of performance detection. The singular spectrum analysis network is used for carrying out data set decomposition on the target performance characteristic data set and decomposing the target performance characteristic data set into a plurality of performance characteristic subsequences. This decomposition process helps to separate the relevance between the different features to better understand and analyze the performance of each feature. The plurality of performance characteristic subsequences are fed into an extreme learning machine network for performance detection analysis. An extreme learning machine network is a machine learning model used for pattern recognition and regression tasks. It features high training speed and better generalization performance. By inputting the performance characteristic sub-sequences, the extreme learning machine network will generate initial performance detection results corresponding to each sub-sequence. These initial results represent the detection of each performance characteristic. And carrying out weighted fusion on the initial performance detection result corresponding to each performance characteristic subsequence by the server to obtain a target performance detection result of the whole super-pulse thulium-doped laser. The weighted fusion can give different weights according to the importance of each sub-sequence, and then combine them together to obtain the final performance detection result. This process helps to comprehensively consider the contributions of the various performance characteristics to more fully evaluate the performance of the laser. For example, assume that the server inputs these features into a singular spectrum analysis network, which decomposes the dataset into three subsequences: an output energy sub-sequence, a frequency stability sub-sequence, and a wavelength stability sub-sequence. Each sub-sequence is sent to an extreme learning machine network to generate initial performance test results. For example, for a sequence of output energy quanta, the server gets an initial detection result for the output energy. And the server performs weighted fusion on the three initial detection results to obtain a target performance detection result of the ultra-pulse thulium-doped laser. The result comprehensively considers the information of a plurality of performance characteristics such as output energy, frequency stability, wavelength stability and the like, and provides comprehensive evaluation for the performance of the laser.

In the embodiment of the invention, performance test is carried out on the ultra-pulse thulium doped laser to obtain initial performance test data, and the initial performance test data is preprocessed to obtain target performance test data; inputting target performance test data into a preset target Gaussian mixture model for performance parameter grouping and data distribution analysis to obtain a plurality of performance parameter Gaussian distributions; extracting parameter distribution characteristics and converting characteristics of the Gaussian distribution of the plurality of performance parameters to obtain a target performance characteristic data set; the invention adopts LSTM neural network to carry out data association, can better process time sequence performance data of the super-pulse thulium doped laser, is beneficial to preserving time sequence association information in performance test data and improves sensitivity to laser performance change. The Gaussian mixture model is used for grouping and data distribution analysis of target performance test data, modeling is effectively carried out on a plurality of performance parameters, and complex characteristics of the performance of the ultra-pulse thulium-doped laser can be captured more accurately. The correlation and important characteristics among the performance parameters can be captured better by carrying out feature extraction and dimension reduction on the Gaussian distribution of the plurality of performance parameters through the analysis of the kernel principal components, and the generalization capability of the model is improved. The method has the advantages that the extreme learning machine is utilized to perform performance detection analysis, wherein the analysis and the performance detection analysis of a data set are better processed by the singular spectrum analysis network and the extreme learning machine network, and the information of different performance characteristics can be more effectively integrated by adopting a mode of weighting and fusing initial performance detection results of each performance characteristic subsequence, so that the performance detection accuracy of the ultra-pulse thulium-doped laser is improved.

In a specific embodiment, the process of executing step S101 may specifically include the following steps:

(1) Performing performance test on the ultra-pulse thulium-doped laser, and acquiring performance test data of the ultra-pulse thulium-doped laser through a preset multi-channel sensor group to obtain corresponding initial performance test data;

(2) Performing laser performance influence factor correlation analysis on the initial performance test data to obtain a parameter correlation analysis result;

(3) According to the parameter correlation analysis result, carrying out data standardization processing on the initial performance test data to obtain standard performance test data;

(4) And inputting the standard performance test data into a preset LSTM neural network to perform parameter time sequence association, so as to obtain the target performance test data.

Specifically, the performance test of the server ultra-pulse thulium-doped laser needs to obtain performance test data through a preset multi-channel sensor group. These sensors may include photodetectors, frequency meters, power meters, etc. for measuring various performance parameters of the laser output. For example, the server measures the output energy of the laser using a power meter, the frequency stability of the laser using a frequency meter, and the like. Original performance test data are obtained, and correlation analysis of laser performance influence factors is needed. The purpose of this step is to know the correlation between different performance parameters, i.e. whether a change in one parameter will have an effect on the other. Correlation analysis is typically implemented using statistical methods, such as pearson correlation coefficients or spearman rank correlation coefficients. By analyzing the correlation, the server identifies which parameters have the greatest impact on laser performance. And then, according to the result of the parameter correlation analysis, the server performs data standardization processing on the initial performance test data. Data normalization is the conversion of measured values of different performance parameters into values of the same scale for comparison and analysis. Common normalization methods include mean removal and variance scaling to ensure that the data is consistent across the same scale. And inputting the standard performance test data into a preset LSTM (Long Short-Term Memory) neural network to perform parameter time sequence association. LSTM is a cyclic neural network suitable for time series data, has memory capacity, and can capture time series dependency of data. Through the LSTM neural network, the server correlates the standard performance test data at different time points, thereby obtaining target performance test data. For example, it is assumed that the server acquires time-series data of performance parameters such as output energy, frequency stability, and wavelength stability using sensors such as a power meter, a frequency meter, and a wavelength meter. The server performs parameter correlation analysis to find that there is a certain correlation between the output energy and the frequency stability, i.e. the change of the output energy affects the frequency stability. The server then data normalizes the time series data of these performance parameters, ensuring that they have consistent dimensions. The server inputs standard performance test data into the LSTM neural network, which will learn and capture the time-series dependencies between the performance parameters. Finally, the server obtains target performance test data through the LSTM neural network, and the data comprise time sequence association information among different performance parameters, so that the performance of the laser can be more comprehensively known.

In a specific embodiment, as shown in fig. 2, the process of executing step S102 may specifically include the following steps:

s201, acquiring a plurality of training performance test data and an initial Gaussian mixture model, and optimizing model parameters of the initial Gaussian mixture model according to the plurality of training performance test data to obtain a target Gaussian mixture model;

s202, grouping performance parameters of a plurality of target data points in target performance test data through a target Gaussian mixture model, and determining at least one Gaussian distribution corresponding to each target data point;

s203, probability density distribution mapping is carried out on the target performance test data through a preset probability density distribution function according to at least one Gaussian distribution corresponding to each target data point, and a plurality of performance parameter Gaussian distributions are obtained.

Specifically, a server obtains a plurality of training performance test data and an initial Gaussian mixture model. These training performance test data are data obtained from previous performance tests on ultra-pulsed thulium doped lasers, which were used to train a gaussian mixture model. The initial gaussian mixture model is a model for the start and may typically be a set of randomly initialized gaussian distributions. The server performs model parameter optimization on the initial gaussian mixture model by using a plurality of training performance test data. This process typically uses maximum likelihood estimation or a expectation maximization algorithm, etc., to optimize the parameters of the gaussian mixture model to better fit the training data. After optimization, the server obtains a target-si hybrid model, which can better describe the distribution of the performance test data. The server then groups the performance parameters for a plurality of target data points in the target performance test data using the target gaussian mixture model. The target performance test data is data that the server wishes to perform performance test, and is new performance test data. Through the target gaussian mixture model, the server determines at least one gaussian distribution corresponding to each target data point. The server correlates the target data points with gaussian distributions to understand their performance under the gaussian mixture model. And according to at least one Gaussian distribution corresponding to each target data point, the server performs probability density distribution mapping on the target performance test data through a preset probability density distribution function. This mapping process maps the target data points to a performance parameter gaussian distribution to determine their probability density in the gaussian distribution. This helps the server understand the location and nature of the target performance data in the overall distribution. For example, assume that there is a set of training performance test data, including measurements of output energy. The server uses these data to train a gaussian mixture model and gets a target gaussian mixture model that contains two gaussian distributions: one corresponding to a low energy state and the other corresponding to a high energy state. The server has a set of target performance test data including some measure of output energy. The server maps the target data points to one of two gaussian distributions through a target-gaussian mixture model to determine the probability that each target data point corresponds to a low energy state or a high energy state. This helps the server to know the performance status of the target data point and can be used for performance detection and classification.

In a specific embodiment, as shown in fig. 3, the process of executing step S201 may specifically include the following steps:

s301, acquiring a plurality of training performance test data, calculating the quantity of mixed components of the plurality of training performance test data according to a parameter correlation analysis result, and determining the number of Gaussian distribution in an initial Gaussian mixture model according to the quantity of mixed components;

s302, determining a plurality of corresponding Gaussian distributions according to the Gaussian distribution number, and performing posterior probability calculation on each training data point in the plurality of training performance test data to obtain posterior probability data;

s303, performing hidden variable analysis on each training data point in the plurality of training performance test data according to the posterior probability data to obtain hidden variable data;

s304, updating parameters of a plurality of Gaussian distributions according to hidden variable data, judging whether an initial Gaussian mixture model is converged through a preset log-likelihood function, and outputting corresponding target model parameters if the initial Gaussian mixture model is converged, wherein the steps of: mean, covariance matrix and mixing coefficients;

and S305, performing model parameter optimization on the initial Gaussian mixture model based on the target model parameters to generate the target Gaussian mixture model.

In particular, the server obtains a plurality of training performance test data, which may be the result of historical performance tests, for training the gaussian mixture model. These training data typically contain measurements of a number of performance parameters, such as output energy, frequency stability, etc. The server performs a parameter correlation analysis to learn the correlation between different performance parameters. Correlation analysis may reveal which parameters there is a correlation between to assist the server in selecting the appropriate hybrid model. Based on the results of the parameter correlation analysis, the server calculates the number of mixed components of the plurality of training performance test data. The number of blending components represents how many gaussian distributions are needed in the gaussian blending model to describe the distribution of the data. If certain performance parameters are highly correlated, less mixed ingredients are required, and if the correlation between them is low, more mixed ingredients are required. This number is typically determined by some statistical method or information criteria. Once the number of blend components is determined, the server creates an initial Gaussian mixture model containing a corresponding number of Gaussian distributions. Each gaussian distribution represents a mixture of components that describe the distribution of training performance test data. The initial gaussian mixture model includes the mean, covariance matrix, and mixture coefficients for each gaussian distribution. The server performs a posterior probability calculation for each training data point in the plurality of training performance test data. The posterior probability represents the probability that each data point belongs to each mixed component. This may be calculated by using an Expectation-Maximization (EM) algorithm or the like. The calculation of posterior probabilities will help the server to understand the relationship between each data point and the different blend components. Then, the server performs a hidden variable analysis based on the posterior probability data. The hidden variable refers to the hidden information of which mixed component each data point belongs to. By analyzing hidden variables, the server better understands the location and distribution of each data point in the Gaussian mixture model. And according to the hidden variable data, the server updates parameters of a plurality of Gaussian distributions. This is accomplished by using an Expectation Maximization (EM) algorithm that iteratively updates the mean, covariance matrix, and mixing coefficients of each gaussian distribution to better fit the training data. The server uses a preset log-likelihood function to determine whether the initial gaussian mixture model has converged. If the model has converged, the server outputs corresponding target model parameters including the mean, covariance matrix, and mixing coefficients for each Gaussian distribution. These parameters will constitute a target-si hybrid model for describing the distribution of performance test data. For example, assume that a parametric correlation analysis indicates that there is a correlation between output energy and frequency stability. Based on the calculation of the number of blend components, the server determines that two blend components are needed to describe the data distribution. The server creates an initial gaussian mixture model containing two gaussian distributions and then iteratively updates the model parameters using the EM algorithm. As the iteration proceeds, the server continuously optimizes the model until the convergence condition is satisfied. Finally, the server obtains a target gaussian mixture model, which contains parameters of two gaussian distributions. This model can be used to describe the distribution of the output energy, thereby helping the server to perform performance detection and analysis.

In a specific embodiment, as shown in fig. 4, the process of performing step S103 may specifically include the following steps:

s401, carrying out parameter distribution feature extraction on a plurality of performance parameter Gaussian distributions by distribution to obtain a plurality of first parameter distribution features;

s402, distributing and calculating confidence intervals of Gaussian distribution of each performance parameter, and carrying out feature verification on a plurality of first parameter distribution features based on the confidence intervals to obtain a plurality of feature verification results;

s403, selecting the characteristics of the first parameter distribution characteristics according to the characteristic verification results to obtain second parameter distribution characteristics;

s404, performing feature dimension reduction processing on the second parameter distribution features through a preset kernel principal component analysis algorithm to obtain a target performance feature data set.

Specifically, the server performs parameter distribution feature extraction on the gaussian distribution of the plurality of performance parameters. And analyzing the Gaussian distribution of each performance parameter, and extracting statistical characteristics related to the distribution shape and characteristics. These statistical features may include mean, variance, skewness, kurtosis, etc. to describe the nature of each gaussian distribution. The server calculates a confidence interval for each performance parameter gaussian distribution. Confidence intervals are ranges used to measure the uncertainty of one parameter estimate. By calculating the confidence interval, the server knows the confidence level of the estimated value. Then, based on the confidence intervals, feature verification is performed on the plurality of first parameter distribution features. Feature verification is a process used to verify the stability and trustworthiness of distributed features. The feature verification results will tell the server which features have highly reliable estimates and which need further inspection or adjustment. For example, if the mean estimate of a certain performance parameter has a high confidence, then the feature is reliable. And according to the characteristic verification results, the server performs characteristic selection on the first parameter distribution characteristics. The server selects those features that perform well in feature verification and that have a high degree of confidence to construct a second parameter distribution feature. These second parameter distribution features will describe the distribution characteristics of the performance parameters more accurately. And performing feature dimension reduction processing on the second parameter distribution features through a preset kernel principal component analysis algorithm. Nuclear principal component analysis is a method for reducing data dimensions by mapping high-dimensional data into low-dimensional space to extract the most important features. This will help reduce the complexity of the data while retaining important information to arrive at the target performance characteristic dataset. For example, assume that the server extracts gaussian distribution characteristics, such as mean and variance, for each parameter. The confidence interval is then calculated and a feature verification is performed, for example, to verify the mean value of the output energy. If the mean value estimate of the output energy has a high confidence level, the server selects the feature and constructs a second parameter distribution feature. And performing dimension reduction processing on the second parameter distribution characteristics by the server through kernel principal component analysis to obtain a target performance characteristic data set. This dataset will contain reduced-dimension features that better characterize the performance of the ultra-pulsed thulium-doped laser, facilitating further performance analysis and detection.

In a specific embodiment, the process of executing step S404 may specifically include the following steps:

(1) Performing matrix conversion on a plurality of second parameter distribution characteristics based on a preset kernel principal component analysis algorithm to obtain an initial characteristic matrix, and performing matrix operation on the initial characteristic matrix through a preset kernel function to obtain a corresponding initial kernel matrix;

(2) Centering the initial core matrix to obtain a target core matrix, and decomposing the characteristic value of the target core matrix to obtain a corresponding characteristic value and a characteristic vector;

(3) And selecting a corresponding main component according to the magnitude of the characteristic value, and mapping a plurality of second parameter distribution characteristics to a new characteristic space according to the main component to obtain a target performance characteristic data set.

Specifically, the server uses a preset Kernel principal component analysis (Kernel Principal Component Analysis, kernel PCA) algorithm to perform matrix conversion on the plurality of second parameter distribution features to obtain an initial feature matrix. Nuclear PCA is a nonlinear dimension reduction technique that can map high-dimensional data into low-dimensional space while preserving the nonlinear structure of the data. When the kernel PCA is carried out, the server uses a preset kernel function to carry out matrix operation on the initial feature matrix, and a corresponding initial kernel matrix is obtained. Kernel functions are typically used to measure similarity or distance between data and map the data into a high-dimensional feature space. Different kernel functions may be used to capture different types of data structures. And the server performs centering processing on the initial core matrix to obtain a target core matrix. Centering is to adjust the value of the matrix by subtracting the mean of each element to ensure that the data has zero mean in the new feature space. This helps to better describe the distribution characteristics of the data. Then, the server performs eigenvalue decomposition on the target kernel matrix to obtain corresponding eigenvalues and eigenvectors. Eigenvalue decomposition is a common linear algebraic method for decomposing the matrix into the form of eigenvalues and eigenvectors. The eigenvalues represent the importance of the data in the new feature space, while the eigenvectors describe the direction of the data in the new space. And selecting a corresponding main component by the server according to the magnitude of the characteristic value. The principal components are eigenvectors of the eigenvalue decomposition, which represent the principal directions of the data in the new feature space. Typically, the server selects the first few principal components with the greatest eigenvalues to preserve the most important information. The server uses these principal components to map a plurality of second parameter distribution features to a new feature space, resulting in a target performance feature dataset. This dataset contains reduced-dimension features that better describe the nature of the raw data, facilitating further performance analysis and detection. For example, assume that there is a set of second parameter profile characteristics of the ultra-pulsed thulium-doped laser, including frequency stability, wavelength stability, pulse width, and the like. The server uses the kernel PCA algorithm to perform nonlinear dimensionality reduction on these features, mapping them into a new feature space. The server then performs principal component analysis on the data in the new feature space, selecting the most important principal component. Finally, the server obtains a target performance characteristic data set, which contains the characteristics after dimension reduction, and the target performance characteristic data set better reflects the performance characteristics of the ultra-pulse thulium-doped laser. This data set can be used for further performance analysis and detection, helping to better understand the performance characteristics and performance issues of the laser.

In a specific embodiment, the process of executing step S104 may specifically include the following steps:

(1) Inputting the target performance characteristic data set into a preset extreme learning machine model, wherein the extreme learning machine model comprises: a singular spectrum analysis network and an extreme learning machine network;

(2) Carrying out data set decomposition on the target performance characteristic data set through a singular spectrum analysis network to obtain a plurality of performance characteristic subsequences;

(3) Inputting the multiple performance characteristic subsequences into an extreme learning machine network for performance detection analysis to obtain an initial performance detection result corresponding to each performance characteristic subsequence;

(4) And carrying out weighted fusion on the initial performance detection results corresponding to each performance characteristic subsequence to obtain the target performance detection results corresponding to the ultra-pulse thulium-doped laser.

Specifically, the server inputs the target performance characteristic dataset into a preset extreme learning machine model. The extreme learning machine model consists of two parts, namely a singular spectrum analysis network and an extreme learning machine network. The two parts will work cooperatively to achieve performance detection and analysis. The singular spectrum analysis network is responsible for performing dataset decomposition on the target performance characteristic dataset. It divides the data set into a plurality of sub-sequences of performance characteristics, each sub-sequence containing data relating to a different performance characteristic. This decomposition helps to better understand the contribution and impact of each feature. These multiple performance feature subsequences are input into an extreme learning machine network for performance detection analysis. An extreme learning machine network is a deep learning model that learns and predicts patterns and trends of data. According to the input performance characteristic subsequences, initial performance detection results corresponding to each subsequence can be obtained. And then, carrying out weighted fusion on the initial performance detection result corresponding to each performance characteristic subsequence. This weighted fusion process may determine weights based on the importance or confidence level of each feature sub-sequence. Finally, the target performance detection result corresponding to the ultra-pulse thulium-doped laser is obtained through weighted fusion. For example, assume that the server first inputs the target performance feature dataset into an extreme learning machine model, including a singular spectrum analysis network and an extreme learning machine network. The singular spectrum analysis network decomposes the data into three performance characteristic subsequences, which respectively correspond to output energy, frequency stability and wavelength stability. And then, the extreme learning machine network performs performance detection analysis on each subsequence to obtain an initial performance detection result. For example, it may determine the trend of the output energy under certain conditions, or evaluate the degree of fluctuation of the frequency stability. The server performs weighted fusion on the three initial performance detection results. If the server considers the frequency stability to be more important for performance evaluation, the server assigns a higher weight to the frequency stability result. Finally, the server obtains the target performance detection result of the super-pulse thulium-doped laser through weighted fusion, and the result comprehensively considers the influence of different performance characteristics.

The performance detection method of the ultra-pulse thulium-doped laser according to the embodiment of the present invention is described above, and the performance detection device of the ultra-pulse thulium-doped laser according to the embodiment of the present invention is described below, referring to fig. 5, one embodiment of the performance detection device of the ultra-pulse thulium-doped laser according to the embodiment of the present invention includes:

the preprocessing module 501 is configured to perform performance test on the ultra-pulse thulium doped laser to obtain initial performance test data, and perform preprocessing on the initial performance test data to obtain target performance test data;

the analysis module 502 is configured to input the target performance test data into a preset target gaussian mixture model to perform performance parameter grouping and data distribution analysis, so as to obtain a plurality of performance parameter gaussian distributions;

a conversion module 503, configured to perform parameter distribution feature extraction and feature conversion on the plurality of performance parameter gaussian distributions, so as to obtain a target performance feature dataset;

and the output module 504 is used for inputting the target performance characteristic data set into a preset extreme learning machine model to perform laser performance detection analysis and outputting a target performance detection result.

Performing performance test on the ultra-pulse thulium-doped laser through the cooperative cooperation of the components to obtain initial performance test data, and preprocessing the initial performance test data to obtain target performance test data; inputting the target performance test data into a preset target Gaussian mixture model for performance parameter grouping and data distribution analysis to obtain a plurality of performance parameter Gaussian distributions; extracting parameter distribution characteristics and converting characteristics of the Gaussian distribution of the plurality of performance parameters to obtain a target performance characteristic data set; the invention adopts LSTM neural network to carry out data association, can better process time sequence performance data of the super-pulse thulium doped laser, is beneficial to preserving time sequence associated information in performance test data and improves sensitivity to laser performance change. The Gaussian mixture model is used for grouping and data distribution analysis of target performance test data, modeling is effectively carried out on a plurality of performance parameters, and complex characteristics of the performance of the ultra-pulse thulium-doped laser can be captured more accurately. The correlation and important characteristics among the performance parameters can be captured better by carrying out feature extraction and dimension reduction on the Gaussian distribution of the plurality of performance parameters through the analysis of the kernel principal components, and the generalization capability of the model is improved. The method has the advantages that the extreme learning machine is utilized to perform performance detection analysis, wherein the analysis and the performance detection analysis of a data set are better processed by the singular spectrum analysis network and the extreme learning machine network, and the information of different performance characteristics can be more effectively integrated by adopting a mode of weighting and fusing initial performance detection results of each performance characteristic subsequence, so that the performance detection accuracy of the ultra-pulse thulium-doped laser is improved.

The above fig. 5 describes the performance detection device of the ultra-pulse thulium-doped laser in the embodiment of the present invention in detail from the point of view of modularized functional entities, and the following describes the performance detection device of the ultra-pulse thulium-doped laser in the embodiment of the present invention in detail from the point of view of hardware processing.

Fig. 6 is a schematic structural diagram of a performance detecting apparatus for an ultra-pulse thulium doped laser according to an embodiment of the present invention, where the performance detecting apparatus 600 for an ultra-pulse thulium doped laser may have relatively large differences due to different configurations or performances, and may include one or more processors (central processing units, CPU) 610 (e.g., one or more processors) and a memory 620, and one or more storage media 630 (e.g., one or more mass storage devices) storing application programs 633 or data 632. Wherein the memory 620 and the storage medium 630 may be transitory or persistent storage. The program stored in the storage medium 630 may include one or more modules (not shown), each of which may include a series of instruction operations in the performance detection apparatus 600 for the ultra-pulsed thulium-doped laser. Still further, the processor 610 may be configured to communicate with the storage medium 630 to execute a series of instruction operations in the storage medium 630 on the performance detection apparatus 600 of the ultra-pulsed thulium doped laser.

The performance detection apparatus 600 of the ultra-pulse thulium-doped laser may also include one or more power supplies 640, one or more wired or wireless network interfaces 650, one or more input/output interfaces 660, and/or one or more operating systems 631, such as Windows Server, mac OS X, unix, linux, freeBSD, and the like. It will be appreciated by those skilled in the art that the configuration of the performance detection device of the ultra-pulse thulium-doped laser illustrated in fig. 6 is not limiting of the performance detection device of the ultra-pulse thulium-doped laser, and may include more or fewer components than illustrated, or may combine certain components, or a different arrangement of components.

The invention also provides a performance detection device of the ultra-pulse thulium-doped laser, which comprises a memory and a processor, wherein the memory stores computer readable instructions, and the computer readable instructions, when executed by the processor, cause the processor to execute the steps of the performance detection method of the ultra-pulse thulium-doped laser in the above embodiments.

The present invention also provides a computer readable storage medium, which may be a non-volatile computer readable storage medium, and may also be a volatile computer readable storage medium, where instructions are stored in the computer readable storage medium, where the instructions, when executed on a computer, cause the computer to perform the steps of the performance detection method of the super-pulse thulium doped laser.

It will be clear to those skilled in the art that, for convenience and brevity of description, specific working procedures of the above-described systems, apparatuses and units may refer to corresponding procedures in the foregoing method embodiments, which are not repeated herein.

The integrated units, if implemented in the form of software functional units and sold or used as stand-alone products, may be stored in a computer readable storage medium. Based on such understanding, the technical solution of the present invention may be embodied essentially or in part or all of the technical solution or in part in the form of a software product stored in a storage medium, including instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method according to the embodiments of the present invention. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a read-only memory (ROM), a random access memory (random access memory, RAM), a magnetic disk, or an optical disk, or other various media capable of storing program codes.

The above embodiments are only for illustrating the technical solution of the present invention, and not for limiting the same; although the invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical scheme described in the foregoing embodiments can be modified or some technical features thereof can be replaced by equivalents; such modifications and substitutions do not depart from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. The performance detection method of the ultra-pulse thulium-doped laser is characterized by comprising the following steps of:

2. The method for detecting performance of an ultra-pulse thulium doped laser according to claim 1, wherein the performing performance test on the ultra-pulse thulium doped laser to obtain initial performance test data, and preprocessing the initial performance test data to obtain target performance test data, includes:

3. The method for detecting performance of an ultra-pulse thulium doped laser according to claim 2, wherein the inputting the target performance test data into a preset target gaussian mixture model for performance parameter grouping and data distribution analysis to obtain a plurality of performance parameter gaussian distributions includes:

4. The method for detecting performance of an ultra-pulse thulium doped laser according to claim 3, wherein the steps of obtaining a plurality of training performance test data and an initial gaussian mixture model, and optimizing model parameters of the initial gaussian mixture model according to the plurality of training performance test data, and obtaining a gaussian mixture model include:

5. The method for detecting performance of an ultra-pulse thulium doped laser according to claim 1, wherein the performing parameter distribution feature extraction and feature conversion on the plurality of performance parameter gaussian distributions to obtain a target performance feature data set includes:

6. The method for detecting performance of an ultra-pulse thulium doped laser according to claim 5, wherein the performing feature dimension reduction processing on the plurality of second parameter distribution features through a preset kernel principal component analysis algorithm to obtain a target performance feature data set includes:

7. The method for detecting performance of an ultra-pulse thulium doped laser according to claim 1, wherein inputting the target performance characteristic dataset into a preset extreme learning machine model for laser performance detection analysis, outputting a target performance detection result, comprises:

8. The utility model provides a performance detection device of super pulse thulium doped laser, its characterized in that, super pulse thulium doped laser's performance detection device includes:

9. The utility model provides a performance detection equipment of super pulse thulium doped laser, its characterized in that, super pulse thulium doped laser's performance detection equipment includes: a memory and at least one processor, the memory having instructions stored therein;

the at least one processor invokes the instructions in the memory to cause the performance detection apparatus of the ultra-pulsed thulium doped laser to perform the performance detection method of the ultra-pulsed thulium doped laser according to any one of claims 1-7.

10. A computer readable storage medium having instructions stored thereon, which when executed by a processor, implement the performance detection method of the ultra-pulsed thulium doped laser according to any one of claims 1-7.