Laser power supply life prediction method and system

Abstract

The invention discloses a laser power supply life prediction method and system, and belongs to the field of power supply life prediction, and the method comprises the steps: obtaining a plurality of historical operation parameters and a plurality of real-time operation parameters; determining a first operation influence feature set based on the plurality of historical operation parameters, screening the first operation influence feature set, and determining a second operation influence feature set; constructing a life prediction model, and synchronizing the second operation influence feature set to the life prediction model to obtain a predicted life value; carrying out life prediction evaluation on the laser power supply, carrying out feedback optimization on the life prediction model, and generating a life prediction optimization model; and performing adaptive life prediction on the laser power supply through the life prediction optimization model. The technical problems of low precision and poor adaptability of laser power supply life prediction in the prior art are solved, and the technical effect of improving the accuracy and the adaptability of laser power supply life prediction is achieved.

Description

Technical Field

[0001] The present invention relates to the field of power supply life prediction, and in particular to a laser power supply life prediction method and system. Background Art

[0002] As the core component of laser processing equipment, the life of the laser power supply directly affects the reliability and production efficiency of the equipment. Accurately predicting the life of the laser power supply is of great significance for equipment maintenance, production planning and cost control. However, the existing laser power supply life prediction only relies on the length of time the power supply has been used, which cannot fully reflect the operating status of the equipment, making it difficult to capture the performance changes of the laser power supply under different working conditions, resulting in low accuracy of the prediction results. In addition, the existing prediction models usually adopt fixed feature sets and prediction strategies, lack the ability to self-adjust according to new data and prediction results, resulting in insufficient adaptability of the prediction model when facing changes in equipment status. Therefore, the existing technology has low accuracy and poor adaptability in predicting the life of the laser power supply. Summary of the invention

[0003] The present application provides a laser power supply lifetime prediction method and system, aiming to solve the technical problems of low accuracy and poor adaptability of laser power supply lifetime prediction in the prior art.

[0004] In view of the above problems, the present application provides a method and system for predicting the life of a laser power supply.

[0005] The first aspect disclosed in the present application provides a method for predicting the life of a laser power supply, the method comprising: retrieving an operation data record log of the laser power supply to obtain a plurality of historical operation parameters; performing real-time data collection of the laser power supply through a data sensing unit to obtain a plurality of real-time operation parameters; performing an impact analysis based on the plurality of historical operation parameters to determine a first operation impact feature set, performing a correlation analysis based on the plurality of real-time operation parameters in combination with the first operation impact feature set, screening the first operation impact feature set according to the analysis results to determine a second operation impact feature set; constructing a life prediction model, synchronizing the second operation impact feature set to the life prediction model to obtain a predicted life value; performing a life prediction evaluation on the laser power supply based on the predicted life value, performing feedback optimization on the life prediction model according to the evaluation results to generate a life prediction optimization model; and performing adaptive life prediction on the laser power supply through the life prediction optimization model.

[0006] Another aspect disclosed in the present application provides a laser power supply life prediction system, which includes: a historical data acquisition module, which is used to retrieve the operation data record log of the laser power supply to obtain multiple historical operation parameters; a real-time data acquisition module, which is used to perform real-time data acquisition of the laser power supply through a data sensing unit to obtain multiple real-time operation parameters; a feature analysis and screening module, which is used to perform impact analysis based on multiple historical operation parameters to determine a first operation impact feature set, perform correlation analysis based on multiple real-time operation parameters in combination with the first operation impact feature set, screen the first operation impact feature set according to the analysis results, and determine a second operation impact feature set; a prediction model construction module, which is used to construct a life prediction model, synchronize the second operation impact feature set to the life prediction model, and obtain a predicted life value; a prediction evaluation optimization module, which is used to perform life prediction evaluation on the laser power supply based on the predicted life value, feedback optimize the life prediction model according to the evaluation result, and generate a life prediction optimization model; a life prediction execution module, which is used to perform adaptive life prediction on the laser power supply through the life prediction optimization model.

[0007] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0008] The operation data record log of the laser power supply is retrieved to obtain multiple historical operation parameters, providing historical information for subsequent analysis and model training; the laser power supply is collected in real time through the data sensing unit to obtain multiple real-time operation parameters, and the current operation status information of the laser power supply is obtained to ensure that the prediction model can be analyzed based on the latest equipment status; impact analysis is performed based on multiple historical operation parameters to determine the first operation impact feature set, and correlation analysis is performed based on multiple real-time operation parameters combined with the first operation impact feature set. The first operation impact feature set is screened according to the analysis results to determine the second operation impact feature set. Through the comprehensive analysis of historical data and real-time data, the most representative and predictive features are screened to improve the accuracy and efficiency of subsequent predictions; a life prediction model is constructed. The test model is used to synchronize the second operation impact feature set to the life prediction model to obtain the predicted life value, which provides a basis for subsequent optimization; the life prediction evaluation of the laser power supply is carried out based on the predicted life value, and the life prediction model is feedback optimized according to the evaluation results to generate a life prediction optimization model, and the model is optimized by evaluating the prediction results to improve the accuracy and reliability of the prediction model; the laser power supply is adaptively predicted by the life prediction optimization model, and the adaptive prediction of the model is realized, so that the prediction results can be continuously adjusted with the changes of the equipment status and environment, and the technical solution of improving the long-term accuracy and adaptability of the prediction is solved, which solves the technical problems of low accuracy and poor adaptability of laser power supply life prediction in the prior art, and achieves the technical effect of improving the accuracy and adaptability of laser power supply life prediction.

[0009] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Figure 1 A schematic flow chart of a method for predicting the life of a laser power supply is provided for an embodiment of the present application;

[0011] Figure 2 A structural schematic diagram of a laser power supply lifetime prediction system is provided for an embodiment of the present application.

[0012] Explanation of the accompanying drawings: historical data acquisition module 11, real-time data collection module 12, feature analysis and screening module 13, prediction model construction module 14, prediction evaluation and optimization module 15, life prediction execution module 16. DETAILED DESCRIPTION

[0013] The overall idea of ​​the technical solution provided by this application is as follows:

[0014] The embodiments of the present application provide a method and system for predicting the life of a laser power supply, which improves the accuracy and adaptability of the prediction by fusing historical data and real-time data, combining dynamic feature analysis and model optimization, and realizes adaptive prediction of the life of the laser power supply.

[0015] First, a comprehensive data foundation is established by retrieving historical operating parameters and real-time operating parameters. On this basis, multi-level feature analysis and screening are carried out, including impact analysis of historical operating parameters and correlation analysis of real-time operating parameters, so as to determine the most representative second operating impact feature set, effectively improving the pertinence and accuracy of subsequent predictions. Subsequently, a life prediction model is constructed, and a more accurate life prediction optimization model is generated through evaluation and feedback optimization of the prediction results. After that, the life prediction optimization model is used for adaptive life prediction, so that the prediction results can be continuously adjusted as the equipment status and environment change.

[0016] After introducing the basic principles of the present application, various non-limiting implementation methods of the present application will be specifically described below in conjunction with the drawings in the specification.

[0017] Embodiment 1, as Figure 1 As shown, an embodiment of the present application provides a method for predicting the life of a laser power supply, the method comprising:

[0018] S1: Retrieve the operation data record log of the laser power supply to obtain multiple historical operation parameters.

[0019] Specifically, during the operation of the laser power supply, various sensors and monitoring devices inside it will record various parameters reflecting the operating status in real time, such as voltage, current, temperature, optical power, usage time, etc., and store these parameter data in the operation data log. By accessing and reading the operation data log, historical operating parameters over a period of time are obtained to describe the past operating status and performance of the laser power supply, and comprehensively reflect the operating status of the laser power supply in different periods and under different working conditions, laying a good data foundation for accurately predicting the life span.

[0020] S2: The laser power supply is collected in real time through the data sensing unit to obtain multiple real-time operating parameters.

[0021] Specifically, the data sensing unit is a hardware module used to collect the operating data of the laser power supply in real time, which is usually connected to the control system of the laser power supply. Various types of sensors are integrated into the data sensing unit, such as voltage sensors, current sensors, temperature sensors, optical power sensors, etc., which can detect various operating indicators of the laser power supply in real time. At the same time, the data sensing unit has the functions of data processing and communication, and can perform preliminary processing on the collected data and transmit it to the host computer or server via wired or wireless means. The data sensing unit samples the operating status of the laser power supply at a certain frequency to obtain a series of real-time operating parameters reflecting its current working conditions.

[0022] The data sensing unit collects data on the laser power supply in real time to obtain multiple real-time operating parameters reflecting its current operating status, providing important data support for the prediction of the laser power supply life.

[0023] S3: Perform an impact analysis based on the multiple historical operating parameters to determine a first operating impact feature set, perform a correlation analysis based on the multiple real-time operating parameters combined with the first operating impact feature set, screen the first operating impact feature set according to the analysis results, and determine a second operating impact feature set.

[0024] Specifically, first, an impact analysis is performed based on multiple historical operating parameters. The so-called impact analysis refers to the study of the correlation and causal relationship between each historical operating parameter and the life of the laser power supply through statistical methods and machine learning algorithms, so as to determine the degree of influence of each parameter on the life. Through the impact analysis, a set of parameter subsets highly correlated with the life are screened out from the numerous historical operating parameters, which is called the first operating impact feature set, reflecting the key factors affecting the life of the laser power supply and is the focus of subsequent predictions. On this basis, the first operating impact feature set is further combined with the real-time operating parameters to perform a correlation analysis. Correlation analysis refers to studying the relationship between the real-time operating parameters and the characteristic quantities in the first operating impact feature set, and judging the similarity and consistency of the parameters in the dynamic change process. By calculating the correlation coefficient between the real-time operating parameters and each characteristic quantity, the change trend and law between them are found, and the first operating impact feature set is further screened and optimized accordingly to obtain the second operating impact feature set. The second operating impact feature set not only retains the factors that have a significant impact on the life, but also takes into account the dynamic characteristics of these factors in actual operation, and has higher representativeness and applicability.

[0025] Through impact analysis and correlation analysis, the dimension reduction and feature extraction of operating parameters are achieved, and a set of key features that can effectively reflect the life of the laser power supply is obtained, namely the second operating impact feature set. It makes full use of the complementarity of historical data and real-time data, takes into account both long-term influencing factors and short-term dynamic changes, and provides the optimal input for the subsequent life prediction model, thereby improving the prediction accuracy and reliability.

[0026] S4: constructing a life prediction model, synchronizing the second operation impact feature set to the life prediction model, and obtaining a predicted life value.

[0027] Specifically, first, according to the data type and distribution characteristics of the second operation impact feature set, select a suitable machine learning algorithm or deep learning framework, such as support vector machine, random forest, neural network, etc. Then, construct a training set and a test set based on the historical operation parameters and the corresponding life data, where the training set is used for model learning and fitting, and the test set is used for model verification and evaluation. Next, use the training set data to train and optimize the selected algorithm model, and adjust the hyperparameters and structure of the model so that it can capture the intrinsic connection between the features and the life as accurately as possible. After that, use the test set data to test and evaluate the trained model, and measure the performance and generalization ability of the model by calculating the error and correlation indicators between the predicted life and the actual life. When the life prediction model is built, it can be applied to the actual life prediction task. Specifically, the obtained second operation impact feature set is input into the trained model, and the model automatically calculates and outputs the corresponding predicted life value, which reflects the remaining use time of the laser power supply in the current state.

[0028] S5: Performing lifetime prediction evaluation on the laser power source based on the predicted lifetime value, performing feedback optimization on the lifetime prediction model according to the evaluation result, and generating a lifetime prediction optimization model.

[0029] Specifically, first, based on the life cycle experiment of the laser power supply, its real failure life data is obtained as a reference standard for evaluating the life prediction model. The so-called life cycle experiment refers to an accelerated aging test of the laser power supply sample in an environment similar to the actual working conditions, recording the whole process data from the initial state to the final failure, and taking the cumulative working time at the time of failure as its real life. Then, the predicted life value of the life prediction model on the same sample is compared with the real life value, and the error between the two, that is, the prediction error value, is calculated. By analyzing the distribution characteristics and change trends of the prediction error value, the prediction deviation and stability of the current model are judged. If the prediction error value exceeds the preset acceptable range, it means that the performance of the current model is not ideal and needs to be optimized and improved in a targeted manner. For example, adjust the structure and parameters of the model, such as increasing or decreasing the number of hidden layers, the number of neurons, changing the activation function and the loss function, etc.; select a better feature subset and eliminate redundant or irrelevant features; introduce new data sources to expand the number and diversity of training samples; improve the training method of the model, such as using cross-validation, ensemble learning and other techniques. Through continuous iterative optimization, a life prediction model with better performance and stronger generalization ability is obtained, namely, the life prediction optimization model.

[0030] Through life cycle experimental data and prediction error analysis, the constructed life prediction model is comprehensively evaluated and optimized, which not only improves the prediction accuracy and robustness of the model, but also enhances the model's ability to adapt to complex working conditions.

[0031] S6: Adaptively predict the lifetime of the laser power supply using the lifetime prediction optimization model.

[0032] Specifically, various parameter data of the laser power supply during operation are continuously collected in an online monitoring manner and transmitted to the life prediction optimization model. The life prediction optimization model first preprocesses and extracts features of the received real-time data to obtain a feature vector that matches the model input. Then, based on the learned feature-life mapping relationship, the life prediction optimization model calculates the predicted life value corresponding to the current feature vector as the evaluation result of the current health status of the laser power supply, thereby realizing adaptive life prediction of the laser power supply.

[0033] By optimizing the life prediction optimization model, the laser power supply can be predicted in real time and online to achieve continuous evaluation and tracking of the remaining life of the laser power supply, thereby improving the accuracy and adaptability of the laser power supply life prediction.

[0034] Furthermore, the embodiment of the present application also includes:

[0035] Performing data cleaning on the multiple historical operation parameters to generate multiple historical operation standard parameters;

[0036] Performing statistical analysis based on the plurality of historical operating standard parameters to determine parameter distribution characteristics;

[0037] Performing univariate analysis on the plurality of historical operating standard parameters according to the parameter distribution characteristics to generate a first analysis result;

[0038] Performing a multivariate analysis on the plurality of historical operating standard parameters according to the parameter distribution characteristics to generate a second analysis result;

[0039] Influencing features are selected according to the first analysis result and the second analysis result to determine the first operation influencing feature set.

[0040] In a feasible implementation, first, multiple historical operation parameters are cleaned to generate multiple historical operation standard parameters. Among them, data cleaning includes checking and processing outliers, missing values ​​and other problems in historical operation parameters to ensure the accuracy and consistency of the data. After completing data cleaning, the data is standardized by min-max standardization, z-score standardization and other methods to eliminate the differences in dimensions and numerical ranges between different parameters. The standardized parameters are historical operation standard parameters, which lay the foundation for subsequent analysis. Secondly, statistical analysis is performed on the obtained multiple historical operation standard parameters. Statistical analysis includes calculating various statistics of each parameter, such as mean, median, standard deviation, etc., to describe the central tendency and degree of dispersion of the parameters; drawing frequency distribution diagrams and probability density curves of parameters, etc., to analyze the distribution type of parameters; calculating the correlation coefficient matrix between parameters to reveal the correlation between parameters. Through statistical analysis, the inherent statistical characteristics of the historical operation standard parameters are accurately portrayed, providing a basis for subsequent modeling analysis.

[0041] Next, the parameter distribution characteristics are fully utilized to perform univariate analysis on each parameter. Univariate analysis mainly includes correlation analysis, regression analysis, variance analysis and information theory analysis. By calculating the correlation coefficient between the parameter and the life span, the strength and direction of the correlation between each parameter and the life span can be determined. For example, a regression model of the parameter and the life span is established, the functional relationship between each parameter and the life span is fitted, the goodness of fit and significance of the model are evaluated, the mean difference of the parameters under different life span levels is compared, and whether the effect of the parameter on the life span is significant is determined. Through univariate analysis, a subset of parameters with high correlation with the life span can be preliminarily screened out, revealing the influence of each parameter on the life span, and providing a basis for further multivariate analysis, thereby obtaining the first analysis result. Then, on the basis of the univariate analysis, the interaction and combination effects between the parameters are further considered, and the influence of multiple parameters on the life span of the laser power supply is comprehensively evaluated. Using the parameter distribution characteristics, a variety of commonly used multivariate analysis methods, such as multivariate regression, principal component analysis, partial least squares regression, etc., are used to establish a multivariate relationship model between parameters and life span. Compared with univariate analysis, multivariate analysis takes into account the correlation and interaction between parameters, overcomes the limitations of univariate analysis, reveals the internal structure and combination pattern between parameters, and finds the most critical combination of influencing factors. The relationship model between the parameter combination and life span obtained by multivariate analysis reflects the law of the comprehensive influence of multiple parameters on life span, which is the second analysis result. Subsequently, the first and second analysis results are combined to give priority to parameters that are highly correlated with life span and have a significant impact on life span, avoid selecting redundant parameters, and thus determine the first operating influence feature set.

[0042] Furthermore, the embodiment of the present application also includes:

[0043] Time-aligning the multiple real-time operating parameters according to the multiple historical operating parameters to generate multiple real-time aligned operating parameters;

[0044] Sorting the multiple real-time alignment operation parameters in descending order according to the first operation impact feature set to construct an operation parameter impact list;

[0045] Performing weighted calculation on the first operation impact feature set based on the operation parameter impact list to generate a plurality of weight correlation coefficients, wherein the plurality of weight correlation coefficients include a plurality of correlation directions and a plurality of correlation levels;

[0046] The plurality of correlation directions and the plurality of correlation levels are added to the analysis result.

[0047] In a feasible implementation, first, multiple real-time operation parameters are time-aligned according to multiple historical operation parameters to generate multiple real-time aligned operation parameters. Specifically, time alignment means that if there is a deviation or inconsistency between the real-time operation parameters and the historical operation parameters in the timestamp, appropriate processing is required to ensure that the real-time data at each time point can be accurately matched and corresponded with the corresponding historical data when performing correlation analysis. Through time alignment, the differences between the real-time data and the historical data in terms of sampling frequency, time accuracy, etc. are eliminated to achieve synchronization and consistency, and provide a unified time reference for subsequent correlation calculations. Then, the multiple real-time aligned operation parameters are sorted in descending order according to the first operation impact feature set to construct an operation parameter impact list. The operation parameter impact list is used to quantitatively describe the impact strength of each impact feature on each parameter under the current operating condition, and provide a basis for subsequent weighted analysis. Specifically, the real-time aligned operation parameters are sorted in descending order according to the importance score or weight of each feature in the first operation impact feature set. The feature with the greatest impact strength is ranked first, and the feature with the smallest impact strength is ranked last.

[0048] Next, a weighted calculation is performed on the first operation impact feature set based on the operation parameter impact list to generate multiple weighted correlation coefficients. During the calculation process, features with greater impact strength are given higher weights, and features with smaller impact strength are given lower weights. For example, according to the ranking position of each feature in the impact list, weights are assigned by methods such as exponential decay and power law decay, so that features with higher rankings obtain higher weights. Then, a weighted correlation calculation is performed on each real-time alignment operation parameter and each impact feature to obtain multiple weighted correlation coefficients. Among them, each weighted correlation coefficient contains two important attributes: correlation direction and correlation level. The correlation direction indicates the positive or negative correlation between the impact feature and the operation parameter; the correlation level indicates the strength of the correlation between the impact feature and the operation parameter, for example, it is divided into strong correlation, moderate correlation, weak correlation, etc. Afterwards, the calculated multiple correlation directions and multiple correlation levels are added to the analysis results to obtain the analysis results.

[0049] Furthermore, the embodiment of the present application also includes:

[0050] Calculating a weight coefficient of the first operation impact feature set based on the operation parameter impact list combined with an expert scoring method;

[0051] By calculating the correlation coefficient of the first operation influencing feature set using a statistical analysis method, a correlation coefficient matrix is ​​obtained;

[0052] Multiplying the weight coefficient based on the first operation influencing feature set by the correlation coefficient matrix to generate the plurality of weight correlation coefficients;

[0053] Dividing the relevant directions according to the positive and negative values ​​of the plurality of weight correlation coefficients to generate a plurality of relevant directions;

[0054] Dividing the correlation levels according to the absolute values ​​of the plurality of weight correlation coefficients to generate a plurality of correlation levels;

[0055] The plurality of correlation directions and the plurality of correlation levels are added to the plurality of weight correlation coefficients.

[0056] In a preferred embodiment, first, according to the sorting positions of the features in the operating parameter impact list, experts in the relevant field are invited to score the importance of each feature in the first operating impact feature set. Then, using expert scoring methods such as the Delphi method or the analytic hierarchy process, the scoring results of multiple experts are integrated to obtain the weight coefficient of each feature. This not only considers the objective sorting driven by data but also incorporates the subjective judgment of expert experience, and can more comprehensively reflect the actual importance of each feature. Then, statistical methods such as the Pearson correlation coefficient and the Spearman rank correlation coefficient are used to calculate the correlation coefficients between the features in the first operating impact feature set. All the calculated correlation coefficients are arranged according to the feature correspondence relationship to form a correlation coefficient matrix, which comprehensively describes the correlation structure within the feature set. Next, matrix multiplication operation is performed on the obtained weight coefficient and the obtained correlation coefficient matrix. Considering both the importance of the features and the correlation between the features, the obtained weight correlation coefficient can more accurately reflect the comprehensive impact degree of each feature on the laser power supply life.

[0057] After that, for each weight correlation coefficient, if its value is greater than zero, it is determined to be positively correlated; if its value is less than zero, it is determined to be negatively correlated; if its value is equal to zero, it is determined to be uncorrelated, so as to clearly distinguish the impact direction between each feature and the laser power supply life, providing an important qualitative basis for subsequent life prediction. Then, the correlation levels are divided. For example, the threshold method is used to divide the correlation levels. Two thresholds a and b are preset (where 0 < a < b < 1). When the absolute value of the weight correlation coefficient is less than a, it is determined to be weakly correlated; when the absolute value is greater than or equal to a and less than b, it is determined to be moderately correlated; when the absolute value is greater than or equal to b, it is determined to be strongly correlated. Subsequently, the obtained multiple correlation directions and the obtained multiple correlation levels are associated with the corresponding weight correlation coefficients, providing a comprehensive and rich information basis for subsequent feature screening and model construction.

[0058] Furthermore, the embodiment of the present application further includes:

[0059] Performing positive correlation screening on the first operating impact feature set based on the multiple correlation directions to generate a first screened feature set;

[0060] Performing strong correlation screening on the first operating impact feature set based on the multiple correlation levels to generate a second screened feature set;

[0061] Performing data intersection extraction according to the first screened feature set and the second screened feature set to obtain a screened feature intersection;

[0062] Performing screening on the first operating impact feature set based on the screened feature intersection to determine the second operating impact feature set.

[0063] In a feasible implementation, first, analyze the correlation direction corresponding to the weight correlation coefficient. Select features with positive correlation, that is, features in which the life of the laser power supply increases accordingly as the eigenvalue increases, to form a first screening feature set, which includes all features that are positively correlated with the life of the laser power supply. At the same time, select features with strong correlation, that is, features that are highly correlated with the life of the laser power supply, to form a second screening feature set, which includes all features that are strongly correlated with the life of the laser power supply, regardless of their correlation direction. Subsequently, perform an intersection operation on the first screening feature set obtained and the second screening feature set obtained to obtain a feature subset that satisfies both positive correlation and strong correlation conditions, and obtain the intersection of screening features. Through intersection extraction, the feature range is further narrowed, the most representative and predictive features are retained, the feature dimension is effectively reduced, redundant information is reduced, and the efficiency and accuracy of subsequent prediction models are improved. Afterwards, the obtained intersection of screening features is used as the screening basis to conduct a final screening on the original first operation impact feature set, retaining the features that are both in the first operation impact feature set and in the intersection of screening features to form the second operation impact feature set, ensuring that the final selected features not only have a strong positive correlation, but also have an important position in the original feature set, providing high-quality input features for the subsequent life prediction model, thereby improving the accuracy and reliability of laser power supply life prediction.

[0064] Furthermore, the embodiment of the present application also includes:

[0065] Conduct life cycle experiments based on laser power supplies to determine the actual life value of the laser power supply;

[0066] The actual life value is calculated to be different from the predicted life value to obtain a prediction error value;

[0067] Determining whether the prediction error value is within a confidence interval;

[0068] If the prediction error value is not within the confidence interval, performing a deviation assessment on the predicted life value to generate the assessment result;

[0069] Based on the evaluation result, synchronizing the predicted life value to a feedback mechanism to obtain a feedback adjustment parameter set;

[0070] Performing dynamic feedback optimization on the life prediction model according to the feedback adjustment parameter set to generate multiple feedback optimization results;

[0071] The plurality of feedback optimization results are reversely verified, and when the verification is passed, the life prediction model is updated according to the plurality of feedback optimization results to generate the life prediction optimization model.

[0072] In a feasible implementation, first, an accelerated life test is performed on the laser power supply. Specifically, on the basis of simulating actual working conditions, the environmental stress (such as temperature, humidity, current density, etc.) is appropriately increased to accelerate the aging process of the laser power supply. By recording the entire process data of the laser power supply from the initial state to the final failure, its actual life value is determined. Then, the actual life value obtained is compared with the predicted life value output by the life prediction model, and the difference between the two, that is, the prediction error value, is calculated, which intuitively reflects the accuracy level of the current life prediction model and provides a quantitative basis for subsequent model evaluation and optimization. Subsequently, a confidence interval is pre-set, such as an acceptable error range determined based on historical data and expert experience. The obtained prediction error value is compared with the confidence interval to determine whether the error is within an acceptable range, which provides a decision-making basis for whether model optimization is needed in the future.

[0073] When the prediction error value exceeds the confidence interval, it means that the prediction result of the current model has a significant deviation. At this time, an in-depth deviation evaluation is conducted on the predicted life value to analyze the possible causes of the error, such as improper feature selection, unreasonable model structure, and poor parameter setting. Through the evaluation, detailed evaluation results are generated to provide targeted guidance for subsequent model optimization. Then, the evaluation results are input into the pre-designed feedback mechanism. Based on the evaluation results, the feedback mechanism generates parameters for adjusting the model in combination with the difference between the predicted life value and the actual life value, forming a feedback adjustment parameter set, including the adjustment coefficient of the feature weight, the modification suggestion of the model structure, the update value of the learning rate, etc. After that, the original life prediction model is systematically optimized using the obtained feedback adjustment parameter set, including adjusting the feature weight, modifying the model structure, updating the learning parameters, etc. Through multiple rounds of optimization, multiple feedback optimization results are generated, each of which represents the improvement of the model in a certain aspect or to a certain extent. Thereafter, the generated multiple feedback optimization results are rigorously reversed. Specifically, an independent validation data set is used to perform performance evaluation on each feedback optimization result to verify whether it has indeed improved the prediction accuracy and generalization ability of the model. If the reverse validation is passed, that is, the feedback optimization results have indeed improved the model performance, then the optimization measures are adopted, the original life prediction model is updated, and a new life prediction optimization model is generated. If the reverse validation fails, it means that there are problems in the optimization process, and the optimization strategy is readjusted.

[0074] Through continuous evaluation and dynamic optimization of the laser power supply life prediction model, the prediction accuracy and adaptability of the model are continuously improved, so that the model always maintains a high prediction performance and adapts to the life prediction needs under different working conditions and environmental conditions, thereby providing a more accurate and reliable basis for the reliability evaluation and maintenance decision-making of the laser power supply.

[0075] Furthermore, the embodiment of the present application also includes:

[0076] Based on the evaluation result and the predicted life value, deviation backtracking is performed to determine multiple error data sources;

[0077] Constructing the feedback mechanism, wherein the feedback mechanism receives the predicted life value according to a data collection cycle;

[0078] Marking the predicted life value according to the multiple error data sources, and synchronizing the marking result to the feedback mechanism;

[0079] Perform deep learning by combining the feedback mechanism with the identification result to calculate and obtain multiple feedback adjustment parameters;

[0080] Constructing simulation operation environment information, and performing cyclic feedback on the plurality of feedback adjustment parameters based on the simulation operation environment information to obtain the feedback adjustment parameter set.

[0081] In a preferred embodiment, first, the evaluation results are analyzed in depth, and combined with the predicted life value, the potential causes of the prediction deviation are traced. Through deviation backtracking, multiple possible error data sources are determined, including but not limited to: data quality problems (such as noise, outliers, missing values, etc.), insufficient model complexity (such as underfitting), improper feature selection (such as missing key features or too many redundant features), unreasonable parameter settings, etc. Through comprehensive error source analysis, a clear direction and basis are provided for subsequent model adjustments. Then, a flexible and efficient feedback mechanism is designed, which can regularly receive the latest predicted life values ​​according to the preset data collection cycle based on the actual operation of the laser power supply and the data acquisition strategy. Periodic data reception ensures that the feedback mechanism can capture changes in the performance of the prediction model in a timely manner, providing a real-time basis for subsequent dynamic adjustments.

[0082] Then, based on the determined multiple error data sources, each predicted life value is identified. Specifically, multidimensional labels are added to each predicted life value, which correspond to different error data sources, indicating which factors may affect the predicted value. For example, the identification results include "poor data quality", "insufficient feature representativeness", "model complexity mismatch", etc. Multidimensional identification not only helps to understand the reliability of each predicted value, but also provides precise guidance for subsequent model adjustments. After the identification is completed, the identification results are synchronized to the constructed feedback mechanism to prepare for subsequent adjustments. Subsequently, the accumulated identification results in the feedback mechanism are systematically analyzed and learned. By mining the potential patterns and laws in the identification results, the key factors affecting the prediction accuracy are identified, and multiple feedback adjustment parameters are generated, including the adjustment coefficient of the feature weight, the optimization suggestions of the model structure, the dynamic update value of the learning rate, etc. After that, a simulation environment that is highly similar to the actual laser power supply operating environment is constructed, that is, the simulated operating environment information, which includes various factors that may affect the life of the laser power supply, such as workload, ambient temperature, humidity, etc. Then, the obtained multiple feedback adjustment parameters are applied to the simulation environment to observe their impact on the prediction results. Through multiple rounds of feedback and adjustment, these parameters are continuously optimized until the best prediction effect is obtained in the simulation environment. After that, the parameters optimized through feedback are integrated into a feedback adjustment parameter set, which includes various adjustment parameters that have been fully verified and optimized.

[0083] By automating the entire process from error analysis to parameter optimization, the source of the prediction error is accurately located, and the optimal adjustment strategy is quickly found through intelligent algorithms and virtual verification. This greatly improves the efficiency and accuracy of model optimization, enhances the predictive model's ability to adapt to various complex working conditions, and provides strong support for laser power supply life prediction.

[0084] In summary, the laser power supply life prediction method provided in the embodiment of the present application has the following technical effects:

[0085] Retrieving the operation data log of the laser power supply to obtain multiple historical operation parameters provides long-term operation reference information for subsequent analysis, which helps to identify long-term factors affecting life. The laser power supply is collected in real time through the data sensing unit to obtain multiple real-time operation parameters, capture the immediate performance of the equipment, and help discover short-term factors and sudden changes that may affect the life. Based on multiple historical operation parameters, impact analysis is performed to determine the first operation impact feature set, and correlation analysis is performed based on multiple real-time operation parameters combined with the first operation impact feature set. The first operation impact feature set is screened according to the analysis results to determine the second operation impact feature set. Through comprehensive analysis of historical and real-time data, the most representative and predictive features are screened to improve the accuracy and efficiency of subsequent predictions and ensure that the model focuses on the most relevant influencing factors. Construct a life prediction model, synchronize the second operation impact feature set to the life prediction model, obtain the predicted life value, and provide a basis for subsequent model evaluation and optimization. Based on the predicted life value, the life prediction evaluation of the laser power supply is performed, and the life prediction model is feedback optimized according to the evaluation results to generate a life prediction optimization model. The model is feedback optimized to improve the accuracy and reliability of the prediction model, so that the model can continuously learn and improve. The life prediction optimization model is used to adaptively predict the life of the laser power supply, realizing the adaptive prediction of the model, so that the prediction results can be dynamically adjusted as the equipment status and environment change, thereby improving the long-term accuracy and adaptability of the prediction.

[0086] Embodiment 2, based on the same inventive concept as a laser power supply life prediction method in the above embodiment, Figure 2 As shown, an embodiment of the present application provides a laser power supply life prediction system, the system comprising:

[0087] A historical data acquisition module 11 is used to retrieve the operation data record log of the laser power supply to obtain multiple historical operation parameters;

[0088] A real-time data acquisition module 12 is used to collect data of the laser power supply in real time through a data sensing unit to obtain a plurality of real-time operating parameters;

[0089] A feature analysis and screening module 13 is used to perform an impact analysis based on the multiple historical operating parameters to determine a first operating impact feature set, perform a correlation analysis based on the multiple real-time operating parameters in combination with the first operating impact feature set, screen the first operating impact feature set according to the analysis result, and determine a second operating impact feature set;

[0090] A prediction model building module 14 is used to build a life prediction model, synchronize the second operation impact feature set to the life prediction model, and obtain a predicted life value;

[0091] A prediction evaluation optimization module 15 is used to perform a life prediction evaluation on the laser power supply based on the predicted life value, and to perform feedback optimization on the life prediction model according to the evaluation result to generate a life prediction optimization model;

[0092] The life prediction execution module 16 is used to perform adaptive life prediction on the laser power supply through the life prediction optimization model.

[0093] Furthermore, the feature analysis and screening module 13 includes the following execution steps:

[0094] Performing data cleaning on the multiple historical operation parameters to generate multiple historical operation standard parameters;

[0095] Performing statistical analysis based on the plurality of historical operating standard parameters to determine parameter distribution characteristics;

[0096] Performing univariate analysis on the plurality of historical operating standard parameters according to the parameter distribution characteristics to generate a first analysis result;

[0097] Performing a multivariate analysis on the plurality of historical operating standard parameters according to the parameter distribution characteristics to generate a second analysis result;

[0098] Influencing features are selected according to the first analysis result and the second analysis result to determine the first operation influencing feature set.

[0099] Furthermore, the feature analysis and screening module 13 also includes the following execution steps:

[0100] Time-aligning the multiple real-time operating parameters according to the multiple historical operating parameters to generate multiple real-time aligned operating parameters;

[0101] Sorting the multiple real-time alignment operation parameters in descending order according to the first operation impact feature set to construct an operation parameter impact list;

[0102] Performing weighted calculation on the first operation impact feature set based on the operation parameter impact list to generate a plurality of weight correlation coefficients, wherein the plurality of weight correlation coefficients include a plurality of correlation directions and a plurality of correlation levels;

[0103] The plurality of correlation directions and the plurality of correlation levels are added to the analysis result.

[0104] Furthermore, the feature analysis and screening module 13 also includes the following execution steps:

[0105] Calculating a weight coefficient of the first operation impact feature set based on the operation parameter impact list combined with an expert scoring method;

[0106] By calculating the correlation coefficient of the first operation influencing feature set using a statistical analysis method, a correlation coefficient matrix is ​​obtained;

[0107] Multiplying the weight coefficient based on the first operation influencing feature set by the correlation coefficient matrix to generate the plurality of weight correlation coefficients;

[0108] Dividing the relevant directions according to the positive and negative values ​​of the plurality of weight correlation coefficients to generate a plurality of relevant directions;

[0109] Dividing the correlation levels according to the absolute values ​​of the plurality of weight correlation coefficients to generate a plurality of correlation levels;

[0110] The plurality of correlation directions and the plurality of correlation levels are added to the plurality of weight correlation coefficients.

[0111] Furthermore, the feature analysis and screening module 13 also includes the following execution steps:

[0112] Performing positive correlation screening on the first operation impact feature set based on the multiple correlation directions to generate a first screening feature set;

[0113] Performing strong correlation screening on the first operation impact feature set based on the multiple correlation levels to generate a second screening feature set;

[0114] Perform data intersection extraction according to the first screening feature set and the second screening feature set to obtain a screening feature intersection;

[0115] The first operation influencing feature set is screened based on the screening feature intersection to determine the second operation influencing feature set.

[0116] Furthermore, the prediction evaluation optimization module 15 includes the following execution steps:

[0117] Conduct life cycle experiments based on laser power supplies to determine the actual life value of the laser power supply;

[0118] The actual life value is calculated to be different from the predicted life value to obtain a prediction error value;

[0119] Determining whether the prediction error value is within a confidence interval;

[0120] If the prediction error value is not within the confidence interval, performing a deviation assessment on the predicted life value to generate the assessment result;

[0121] Based on the evaluation result, synchronizing the predicted life value to a feedback mechanism to obtain a feedback adjustment parameter set;

[0122] Performing dynamic feedback optimization on the life prediction model according to the feedback adjustment parameter set to generate multiple feedback optimization results;

[0123] The plurality of feedback optimization results are reversely verified, and when the verification is passed, the life prediction model is updated according to the plurality of feedback optimization results to generate the life prediction optimization model.

[0124] Furthermore, the prediction evaluation optimization module 15 also includes the following execution steps:

[0125] Based on the evaluation result and the predicted life value, deviation backtracking is performed to determine multiple error data sources;

[0126] Constructing the feedback mechanism, wherein the feedback mechanism receives the predicted life value according to a data collection cycle;

[0127] Marking the predicted life value according to the multiple error data sources, and synchronizing the marking result to the feedback mechanism;

[0128] Perform deep learning by combining the feedback mechanism with the identification result to calculate and obtain multiple feedback adjustment parameters;

[0129] Constructing simulation operation environment information, and performing cyclic feedback on the plurality of feedback adjustment parameters based on the simulation operation environment information to obtain the feedback adjustment parameter set.

[0130] Any step of the method described above can be stored as a computer instruction or program in an unlimited computer memory, and can be called and recognized by an unlimited computer processor to implement any method in the embodiments of the present application, without any unnecessary restrictions.

[0131] Furthermore, the first or second mentioned above may not only represent an order relationship, but may also represent a specific concept, and / or refer to the selection of multiple elements individually or in whole. Obviously, those skilled in the art may make various changes and modifications to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the present application and its equivalents, the present application intends to include these changes and variations.

Claims

1. A method for predicting the life of a laser power supply, characterized in that: The method comprises: Retrieve the operation data record log of the laser power supply to obtain multiple historical operation parameters; The laser power supply is collected in real time through the data sensing unit to obtain multiple real-time operating parameters; Performing an impact analysis based on the multiple historical operating parameters to determine a first operating impact feature set, performing a correlation analysis based on the multiple real-time operating parameters combined with the first operating impact feature set, screening the first operating impact feature set according to the analysis result, and determining a second operating impact feature set; constructing a life prediction model, synchronizing the second operation impact feature set to the life prediction model, and obtaining a predicted life value; Performing a life prediction evaluation on the laser power supply based on the predicted life value, performing feedback optimization on the life prediction model according to the evaluation result, and generating a life prediction optimization model; The lifetime prediction optimization model is used to perform adaptive lifetime prediction on the laser power supply.

2. A laser power supply life prediction method as claimed in claim 1, characterized in that: Performing an impact analysis based on the multiple historical operating parameters to determine a first operating impact feature set, the method comprising: Performing data cleaning on the multiple historical operation parameters to generate multiple historical operation standard parameters; Performing statistical analysis based on the plurality of historical operating standard parameters to determine parameter distribution characteristics; Performing univariate analysis on the plurality of historical operating standard parameters according to the parameter distribution characteristics to generate a first analysis result; Performing a multivariate analysis on the plurality of historical operating standard parameters according to the parameter distribution characteristics to generate a second analysis result; Influencing features are selected according to the first analysis result and the second analysis result to determine the first operation influencing feature set.

3. A laser power supply life prediction method as claimed in claim 1, characterized in that: Based on the multiple real-time operating parameters combined with the first operating impact feature set, a correlation analysis is performed, the method comprising: Time-aligning the multiple real-time operating parameters according to the multiple historical operating parameters to generate multiple real-time aligned operating parameters; Sorting the multiple real-time alignment operation parameters in descending order according to the first operation impact feature set to construct an operation parameter impact list; Performing weighted calculation on the first operation impact feature set based on the operation parameter impact list to generate a plurality of weight correlation coefficients, wherein the plurality of weight correlation coefficients include a plurality of correlation directions and a plurality of correlation levels; The plurality of correlation directions and the plurality of correlation levels are added to the analysis result.

4. A laser power supply life prediction method as claimed in claim 3, characterized in that: Performing weighted calculation on the first operation impact feature set based on the operation parameter impact list to generate a plurality of weight correlation coefficients, the method comprising: Calculating a weight coefficient of the first operation impact feature set based on the operation parameter impact list combined with an expert scoring method; By calculating the correlation coefficient of the first operation influencing feature set using a statistical analysis method, a correlation coefficient matrix is ​​obtained; Multiplying the weight coefficient based on the first operation influencing feature set by the correlation coefficient matrix to generate the plurality of weight correlation coefficients; Dividing the relevant directions according to the positive and negative values ​​of the plurality of weight correlation coefficients to generate a plurality of relevant directions; Dividing the correlation levels according to the absolute values ​​of the plurality of weight correlation coefficients to generate a plurality of correlation levels; The plurality of correlation directions and the plurality of correlation levels are added to the plurality of weight correlation coefficients.

5. A method for predicting the life of a laser power supply according to claim 4, characterized in that: The first operation impact feature set is screened according to the analysis result to determine a second operation impact feature set, the method comprising: Performing positive correlation screening on the first operation impact feature set based on the multiple correlation directions to generate a first screening feature set; Performing strong correlation screening on the first operation impact feature set based on the multiple correlation levels to generate a second screening feature set; Perform data intersection extraction according to the first screening feature set and the second screening feature set to obtain a screening feature intersection; The first operation influencing feature set is screened based on the screening feature intersection to determine the second operation influencing feature set.

6. A laser power supply life prediction method as claimed in claim 1, characterized in that: Based on the predicted life value, a life prediction evaluation is performed on the laser power supply, and according to the evaluation result, the life prediction model is feedback optimized to generate a life prediction optimization model, the method comprising: Conduct life cycle experiments based on laser power supplies to determine the actual life value of the laser power supply; The actual life value is calculated to be different from the predicted life value to obtain a prediction error value; Determining whether the prediction error value is within a confidence interval; If the prediction error value is not within the confidence interval, performing a deviation assessment on the predicted life value to generate the assessment result; Based on the evaluation result, synchronizing the predicted life value to a feedback mechanism to obtain a feedback adjustment parameter set; Performing dynamic feedback optimization on the life prediction model according to the feedback adjustment parameter set to generate multiple feedback optimization results; The plurality of feedback optimization results are reversely verified, and when the verification is passed, the life prediction model is updated according to the plurality of feedback optimization results to generate the life prediction optimization model.

7. A method for predicting the life of a laser power supply according to claim 6, characterized in that: Based on the evaluation result, the predicted life value is synchronized to a feedback mechanism to obtain a feedback adjustment parameter set, the method comprising: Based on the evaluation result and the predicted life value, deviation backtracking is performed to determine multiple error data sources; Constructing the feedback mechanism, wherein the feedback mechanism receives the predicted life value according to a data collection cycle; Marking the predicted life value according to the multiple error data sources, and synchronizing the marking result to the feedback mechanism; Perform deep learning by combining the feedback mechanism with the identification result to calculate and obtain multiple feedback adjustment parameters; Constructing simulation operation environment information, and performing cyclic feedback on the plurality of feedback adjustment parameters based on the simulation operation environment information to obtain the feedback adjustment parameter set.

8. A laser power supply life prediction system, characterized in that: A system for implementing a laser power supply life prediction method according to any one of claims 1 to 7, comprising: A historical data acquisition module, wherein the historical data acquisition module is used to retrieve the operation data record log of the laser power supply to obtain multiple historical operation parameters; A real-time data acquisition module, which is used to collect data from the laser power supply in real time through a data sensing unit to obtain multiple real-time operating parameters; a feature analysis and screening module, the feature analysis and screening module being used to perform an impact analysis based on the multiple historical operating parameters to determine a first operating impact feature set, perform a correlation analysis based on the multiple real-time operating parameters in combination with the first operating impact feature set, screen the first operating impact feature set according to the analysis result, and determine a second operating impact feature set; A prediction model building module, wherein the prediction model building module is used to build a life prediction model, synchronize the second operation impact feature set to the life prediction model, and obtain a predicted life value; A prediction evaluation optimization module, wherein the prediction evaluation optimization module is used to perform a life prediction evaluation on the laser power supply based on the predicted life value, and to perform feedback optimization on the life prediction model according to the evaluation result to generate a life prediction optimization model; A life prediction execution module is used to perform adaptive life prediction on the laser power supply through the life prediction optimization model.

Images