A remote control method for laser processing system based on cloud computing

By correcting the weight function value according to the fluctuation changes and abnormality degree of the cutting stage in the laser processing system, a more accurate cutting parameter width correlation model is constructed, which solves the problem of inaccurate weight function value in the existing technology, and improves laser processing The accuracy of the system's remote control.

CN118543993BActive Publication Date: 2024-11-19NOBOT INTELLIGENT EQUIP (SHANDONG) CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411004604.3
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-07-25
Publication Date
2024-11-19
Estimated Expiration
2044-07-25

AI Technical Summary

Technical Problem

In the prior art, only the distance between the input variable and the training sample is considered during the training process of GRNN network, resulting in the inaccurate value of the weight function, resulting in low accuracy of the output results of the cutting parameter width correlation model, which in turn makes the laser processing system remote. The accuracy of the control is poor.

Method used

By obtaining the timing data of the cut-slit width and cutting correlation parameters in the historical data of the laser processing system, dividing the cutting stage, correcting the weight function value according to the fluctuation change and abnormality degree, and building a more accurate cutting parameter width correlation model for laser Remote control of the processing system.

Benefits of technology

The output accuracy of the cutting parameter width correlation model is improved and the accuracy of remote control of the laser processing system is improved.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118543993B_ABST
    Figure CN118543993B_ABST
Patent Text Reader

Abstract

The present invention relates to the field of laser industrial control technology, and in particular to a remote control method for a laser processing system based on cloud computing. The method first divides each cutting stage according to a time series distribution trend of a slit width in each complete cutting cycle in historical data; obtains a correlation influence degree according to the similarity of fluctuation changes between slit width time series data and associated parameter time series data of each cutting associated parameter in each cutting stage; uses the associated parameter time series data as a training set input variable parameter, and corrects a weight function value in a GRNN model training process according to a training weight correction coefficient obtained according to the corresponding abnormality degree and the associated influence degree, so that the output result of the obtained cutting parameter width associated model is more accurate, thereby making the accuracy of remote real-time control of a laser processing system based on the cutting parameter width associated model with the help of cloud computing higher.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of laser industrial control technology, and in particular to a remote control method for a laser processing system based on cloud computing. Background Art

[0002] The laser used in laser processing has the advantages of high energy, good directionality and monochromaticity. Therefore, in the manufacturing industry, the use of laser processing technology is becoming more and more widespread. In the process of laser processing, the slit width is an important evaluation index of the laser processing quality. Therefore, it is necessary to control the laser processing system to ensure the accuracy of the cut slit width. The slit width will be affected by various cutting-related parameters corresponding to the laser processing system, such as laser power, defocus, and cutting speed. Therefore, it is necessary to adjust various cutting-related parameters to varying degrees according to the progress of the laser processing process, so as to maximize the laser processing quality.

[0003] Considering that different types of cutting-related parameters have different influences on the slit width at different cutting stages during the laser processing process, in order to more accurately control the laser processing system, different cutting parameter width association models need to be set according to different influences at different cutting stages. Considering the specific data processing function of cloud computing technology, the existing technology usually combines cloud computing technology, uses the General Regression Neural Network (GRNN) network training model and the historical data of cutting processing through the laser processing control system to construct a cutting parameter width association model corresponding to each cutting stage, so as to remotely control the laser processing system according to the output results of the cutting parameter width association model.

[0004] The GRNN network determines the weight value of each cutting-related parameter through a weight function, and the weight function value in the GRNN network only takes into account the distance between the input variable and the training sample, and does not take into account the degree of influence or correlation of each input variable on the output result, that is, does not take into account the degree of influence or correlation of each cutting-related parameter on the slit width; resulting in the obtained weight function value being inaccurate, resulting in the output result of the constructed cutting parameter width association model being of low accuracy, that is, in the prior art GRNN network training process, only the weight function value obtained by considering the distance between the input variable and the training sample is inaccurate, resulting in the output result of the subsequently constructed cutting parameter width association model being of low accuracy; thereby making the accuracy of remote control of the laser processing system poor. Summary of the invention

[0005] In order to solve the technical problem that the weight function value obtained by only considering the distance between the input variable and the training sample in the prior art GRNN network training process is not accurate enough, resulting in the low accuracy of the output result of the cutting parameter width association model constructed subsequently; thus making the accuracy of the remote control of the laser processing system poor, the purpose of the present invention is to provide a remote control method for a laser processing system based on cloud computing, and the technical scheme adopted is as follows:

[0006] The present invention proposes a remote control method for a laser processing system based on cloud computing, the method comprising:

[0007] From the historical data of cutting processing performed by the laser processing control system, the slit width time series data corresponding to each complete cutting cycle and the associated parameter time series data of each cutting associated parameter are obtained;

[0008] According to the extreme value distribution of the slit width time series data of each complete cutting cycle in the time series, all complete cutting cycles are divided into at least two cutting stages; wherein the number of corresponding cutting stages in all complete cutting cycles is the same; in the historical data, according to the similarity of the fluctuation changes between all the associated parameter time series data of each cutting associated parameter in each cutting stage and the slit width time series data, the associated influence degree of each cutting associated parameter in each cutting stage is obtained;

[0009] All the associated parameter time series data corresponding to each cutting associated parameter in each cutting stage are used as the training set input variable parameters of the training process of the GRNN model in each cutting stage; in each cutting stage, according to the relative deviation of the abnormality degree of each training set input variable parameter corresponding to each cutting associated parameter and the degree of the associated influence, the training weight correction coefficient of each training set input variable parameter is obtained;

[0010] According to the training weight correction coefficient, the weight function value in the training process of the GRNN model of each cutting stage is corrected to obtain the cutting parameter width association model of each cutting stage; according to the cutting parameter width association model combined with the cutting seam width timing data and the associated parameter timing data in the real-time processing process, the laser processing system is remotely controlled.

[0011] Furthermore, the acquisition method of the cutting stage includes:

[0012] Perform curve fitting on all slit width time series data corresponding to each complete cutting cycle to obtain a slit width time series curve corresponding to each complete cutting cycle; take the mode of the number of extreme points corresponding to all slit width time series curves as a characteristic reference; take all slit width time series curves whose number of extreme points is the characteristic reference as the characteristic width time series curve;

[0013] Taking all extreme value points of each characteristic width timing curve as interval points, the complete cutting cycle corresponding to each characteristic width timing curve is divided into at least two interval time intervals; the ratio between the time length of each interval time interval in each characteristic width timing curve and the time length corresponding to each characteristic width timing curve is used as the reference length ratio of each interval time interval in each characteristic width timing curve;

[0014] Arrange the reference length proportions corresponding to the interval time intervals of each characteristic width timing curve in time order to obtain a reference length proportion sequence of each characteristic width timing curve; use the index value corresponding to each reference length proportion in the reference length proportion sequence as the stage index value corresponding to the cutting stage;

[0015] In all reference length ratio sequences, the average of all reference length ratios corresponding to each stage index value is taken as the interval ratio of each corresponding cutting stage; according to the interval ratios of all cutting stages, all complete cutting cycles are divided into at least two cutting stages in chronological order; the number of the cutting stages is the same as the number of interval time intervals.

[0016] Furthermore, the method for obtaining the correlation influence degree includes:

[0017] Perform curve fitting according to all associated parameter time series data of each cutting associated parameter corresponding to each complete cutting cycle, to obtain an associated parameter time series curve of each cutting associated parameter in each complete cutting cycle;

[0018] In the associated parameter time series curve, the mean of the tangent slopes of all data points on the local curve segment corresponding to each cutting stage is taken as the mean of the associated slopes of each cutting associated parameter at each cutting stage in each complete cutting cycle; the variance of the associated parameter time series data of each cutting associated parameter at each cutting stage in each complete cutting cycle is taken as the volatility of the associated data of each cutting associated parameter at each cutting stage in each complete cutting cycle;

[0019] In the slit width timing curve corresponding to each complete cutting cycle, the mean of the tangent slopes of all data points on the local curve segment corresponding to each cutting stage is taken as the mean of the width slopes at each cutting stage in each complete cutting cycle; the variance of the slit width timing data at each cutting stage in each complete cutting cycle is taken as the width data volatility at each cutting stage in each complete cutting cycle;

[0020] According to the deviation between the mean value of the correlation slope and the mean value of the width slope, and the deviation between the volatility of the correlation data and the volatility of the width data, the correlation influence degree of each cutting correlation parameter at each cutting stage is obtained.

[0021] Further, the method for obtaining the correlation influence degree of each cutting correlation parameter at each cutting stage according to the deviation between the correlation slope mean and the width slope mean, and the deviation between the correlation data volatility and the width data volatility includes:

[0022] The difference between the mean of the correlation slope and the mean of the width slope is used as the correlation slope deviation of each cutting correlation parameter at each cutting stage in each complete cutting cycle; the difference between the correlation data volatility and the width data volatility is used as the correlation volatility deviation of each cutting correlation parameter at each cutting stage in each complete cutting cycle;

[0023] The mean of all associated slope deviations corresponding to each cutting-related parameter at each cutting stage is taken as the reference slope mean; the mean of all associated volatility deviations corresponding to each cutting-related parameter at each cutting stage is taken as the reference volatility mean;

[0024] The negative correlation mapping value of the product of the reference slope mean and the reference volatility mean is used as the correlation influence degree of each cutting correlation parameter at each cutting stage.

[0025] Furthermore, the method for obtaining the training weight correction coefficient includes:

[0026] The LOF anomaly detection algorithm is used to calculate the LOF value of each training set input variable corresponding to each cutting-related parameter in each cutting stage;

[0027] According to the LOF value of each training set input variable corresponding to each cutting association parameter in each cutting stage, relative to the deviation of the overall LOF value, the variable abnormality degree of each training set input variable corresponding to each cutting association parameter in each cutting stage is obtained;

[0028] The cumulative value of the correlation influence degree corresponding to all types of cutting-related parameters in each cutting stage is used as the reference cumulative value corresponding to each cutting stage; the ratio of the correlation influence degree corresponding to each cutting-related parameter in each cutting stage to the reference cumulative value is used as the relative correlation degree of each cutting-related parameter in each cutting stage;

[0029] According to the abnormal degree of the variable and the relative correlation degree, the training weight correction coefficient of each training set input variable in each cutting correlation parameter in each cutting stage is obtained; the abnormal degree of the variable and the relative correlation degree are both positively correlated with the training weight correction coefficient.

[0030] Furthermore, the method for obtaining the abnormality degree of the variable includes:

[0031] The mean of all LOF values ​​corresponding to each cutting-related parameter in each cutting stage is used as the reference outlier value of each cutting-related parameter in each cutting stage;

[0032] The normalized value of the difference between the LOF value of each training set input variable corresponding to each cutting-related parameter in each cutting stage and the reference abnormal value is used as the variable abnormality degree of each training set input variable corresponding to each cutting-related parameter in each cutting stage.

[0033] Furthermore, the method for obtaining the training weight correction coefficient of each training set input variable in each cutting association parameter in each cutting stage according to the abnormal degree of the variable and the relative correlation degree includes:

[0034] The normalized value of the product of the negative correlation mapping value of the abnormal degree of the variable and the relative correlation degree is used as the training weight correction coefficient of each training set input variable in each cutting association parameter in each cutting stage.

[0035] Furthermore, the method for acquiring the cutting parameter width association model includes:

[0036] In the training process of the GRNN model in each cutting stage, the prior weight function value of each training set input variable corresponding to each cutting association parameter in each cutting stage is obtained; the prior weight function value is weighted by the positive correlation mapping value of the training weight correction coefficient to obtain the corrected weight function value of each training set input variable corresponding to each cutting association parameter in each cutting stage;

[0037] The GRNN model of each cutting stage is trained according to the preset smoothing factor and the modified weight function value to obtain the cutting parameter width association model of each cutting stage.

[0038] Furthermore, the method for obtaining the modified weight function value includes:

[0039] The mean of the training weight correction coefficients of all training set input variables corresponding to each cutting-related parameter in each cutting stage is used as a reference correction mean; the ratio between the training weight correction coefficient and the reference correction mean is used as a weighting coefficient of each training set input variable corresponding to each cutting-related parameter in each cutting stage;

[0040] The product of the weighting coefficient and the priori weight function value is used as the modified weight function value of each training set input variable corresponding to each cutting association parameter in each cutting stage.

[0041] Furthermore, the method for remotely controlling the laser processing system according to the cutting parameter width association model combined with the slit width time series data and the association parameter time series data in the real-time processing process includes:

[0042] According to the time length of the complete cutting cycle corresponding to each characteristic width timing curve, the cutting stage corresponding to the current moment is determined; the cutting stage corresponding to the current moment is used as the real-time cutting stage;

[0043] In all complete cutting cycles, the average of all slit width time series data corresponding to the real-time cutting stage is used as the ideal slit width in the real-time cutting stage; the slit width time series data at the current moment is used as the real-time slit width;

[0044] In the width dimension, all the slit width time series data between the ideal slit width and the real-time slit width are used as reference slit width data; the time corresponding to the reference slit width in all complete slit cycles is used as the reference time;

[0045] All the associated parameter time series data corresponding to each cutting associated parameter at all reference times are used as the screening parameter set corresponding to each cutting associated parameter;

[0046] In each screening parameter set, one associated parameter time series data is randomly selected to form a cutting associated parameter combination; and all cutting associated parameter combinations are obtained;

[0047] All the associated parameter time series data in each cutting associated parameter combination are used as the input of the cutting parameter width association model corresponding to the real-time cutting stage, and the predicted slit width corresponding to each cutting associated parameter combination is output; the predicted slit width is converted into the fitness function of the genetic algorithm, and the genetic operation screening is performed according to the elite strategy sorting according to all the predicted slit widths until the optimal slit width corresponding to the optimal solution is screened out; the associated parameter time series data in the cutting associated parameter combination corresponding to the optimal slit width is used as the optimal parameter time series data of each cutting associated parameter at the current moment;

[0048] The laser processing system is remotely controlled according to the optimal parameter timing data.

[0049] The present invention has the following beneficial effects:

[0050] Since different cutting parameter width correlation models need to be set according to different degrees of influence at different cutting stages, it is necessary to first divide the cutting stages in the cutting process. Considering that different types of cutting-related parameters have different degrees of influence on the slit width at different cutting stages in the laser processing process, the degree of change or trend of the slit width usually differs in different cutting stages, and the extreme value point usually corresponds to the position where the change trend changes. Therefore, the extreme value distribution of the slit width time series data in the time series of the present invention is used to divide the cutting stages.

[0051] The reason why the output results of the cutting parameter width association model constructed by the prior art based on the GRNN network are less accurate is that the determination of the weight function value in the GRNN network only takes into account the distance between the input variable and the training sample, and does not take into account the influence or association degree of each cutting association parameter on the slit width. The more similar the fluctuation change of the cutting association parameter is to the fluctuation change of the slit width, the greater the influence or association of the corresponding cutting association parameter. Therefore, in each cutting stage, the present invention obtains the corresponding association influence degree of each cutting association parameter according to the similarity of the fluctuation change between all the association parameter time series data of each cutting association parameter and the slit width time series data.

[0052] Further considering that the more abnormal the training set input variable parameters are, the smaller the corresponding credibility is, therefore, the present invention obtains the training weight correction coefficient for correcting the weight function value based on the degree of association influence and the degree of abnormality of each training set input variable parameter; thereby further combining the training weight correction coefficient to correct the weight function value in the training process of the GRNN model of each cutting stage, so that the output result of the cutting parameter width association model is more accurate, thereby making the accuracy of remote real-time control of the laser processing system according to the cutting parameter width association model higher. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0054] Figure 1 A flow chart of a remote control method for a laser processing system based on cloud computing provided by an embodiment of the present invention;

[0055] Figure 2 A schematic diagram of a slit width during a laser cutting process provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0056] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the following is a detailed description of the specific implementation method, structure, features and effects of a remote control method for a laser processing system based on cloud computing proposed by the present invention in combination with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics in one or more embodiments may be combined in any suitable form.

[0057] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0058] The specific scheme of the remote control method of a laser processing system based on cloud computing provided by the present invention is described in detail below with reference to the accompanying drawings.

[0059] See also Figure 1 , which shows a flow chart of a remote control method for a laser processing system based on cloud computing provided by an embodiment of the present invention, the method comprising:

[0060] Step S1: From the historical data of cutting processing performed by the laser processing control system, obtain the slit width timing data corresponding to each complete cutting cycle and the associated parameter timing data of each cutting associated parameter.

[0061] An embodiment of the present invention aims to provide a remote control method for a laser processing system based on cloud computing, which is used to correct the determination of the weight function value in the GRNN network according to the degree of correlation influence of the associated parameter time series data of each cutting associated parameter in the historical data on the slit width time series data, so that the output result of the cutting parameter width association model obtained is more accurate, thereby making the accuracy of remote real-time control of the laser processing system based on the cutting parameter width association model higher.

[0062] Therefore, the embodiment of the present invention first obtains the slit width timing data corresponding to each complete cutting cycle and the associated parameter timing data of each cutting associated parameter in the historical data of cutting processing performed by the laser processing control system. Specifically: the types of cutting associated parameters in the embodiment of the present invention are selected from laser power, cutting speed and defocus; therefore, the embodiment of the present invention collects the slit width, laser power, cutting speed and defocus at each sampling moment through the laser processing system in the complete cutting cycle in the historical data, so that in each complete cutting cycle, all associated parameter timing data corresponding to the cutting associated parameter type corresponding to the laser power, all associated parameter timing data corresponding to the cutting associated parameter type corresponding to the cutting speed, all associated parameter timing data corresponding to the cutting associated parameter type corresponding to the defocus, and slit width timing data are obtained. In the embodiment of the present invention, the time interval of the sampling moment is set to 1 minute, and the number of complete cutting cycles is set to 100, which can be adjusted by the implementer according to the specific implementation environment. It should be noted that a complete cutting cycle corresponds to the time period corresponding to the start of cutting to the end of cutting, that is, a complete cutting process. The implementer can select other cutting associated parameters related to the slit width according to the specific implementation environment, which will not be further elaborated here. In addition, it should be noted that in order to ensure the accuracy of the cutting parameter width correlation model established subsequently, the complete cutting cycle selected in the embodiment of the present invention is a cutting cycle with qualified laser processing quality or meeting the specific cutting requirements in the implementation environment, which will not be further elaborated here.

[0063] In addition, it should be noted that in order to prevent misunderstandings, please refer to Figure 2 , which shows a schematic diagram of the slit width during a laser cutting process provided by an embodiment of the present invention; Figure 2 In the figure, the direction pointed by the arrow is the cutting direction of the laser cutting, and the black area is the slit left during the cutting process. L represents the slit width, that is, Figure 2 It should be noted that in a specific implementation environment, the corresponding slit width data usually presents a certain time series dynamic change. Figure 2 The figure below is just one of the schematic diagrams of the slit width.

[0064] Step S2: According to the extreme value distribution of the slit width time series data of each complete cutting cycle in the time series, all complete cutting cycles are divided into at least two cutting stages; wherein the number of corresponding cutting stages in all complete cutting cycles is the same; in the historical data, according to the similarity of the fluctuation changes between all associated parameter time series data of each cutting associated parameter in each cutting stage and the slit width time series data, the correlation influence degree of each cutting associated parameter in each cutting stage is obtained.

[0065] Since it is necessary to set different cutting parameter width association models according to different degrees of influence in different cutting stages, it is necessary to first divide the cutting stages in the cutting process. Taking into account that in different cutting stages during the laser processing process, different types of cutting-related parameters have different degrees of influence on the slit width, so in different cutting stages, the degree of change or trend of the slit width is usually different, and the extreme point usually corresponds to the position where the change trend changes. Therefore, the extreme value distribution of the slit width time series data in the time series of the present invention is used to divide the cutting stage. According to the extreme value distribution of the slit width time series data in the time series of each complete cutting cycle, all complete cutting cycles are divided into at least two cutting stages; wherein the number of corresponding cutting stages in all complete cutting cycles is the same. It should be noted that setting the cutting stages to be the same can facilitate the corresponding analysis of subsequent cutting stages.

[0066] Preferably, the acquisition method of the cutting stage includes:

[0067] According to all the slit width timing data corresponding to each complete cutting cycle, curve fitting is performed to obtain the slit width timing curve corresponding to each complete cutting cycle. By obtaining the corresponding slit width timing curve, the changing trend of the slit width can be more accurately and intuitively characterized. Further, the mode of the number of extreme points corresponding to all slit width timing curves is taken as the characteristic reference quantity; all slit width timing curves with the number of extreme points as the characteristic reference quantity are taken as the characteristic width timing curve. Since the basis for the division of subsequent cutting stages is the number of extreme points, the slit width timing curves with the same number of extreme points correspond to the same number of cutting stages. Therefore, according to the characteristic that the number of corresponding cutting stages in all complete cutting cycles is the same, by analyzing the subsequent cutting stage division of the characteristic width timing curve with the highest frequency of extreme points, the obtained cutting stage division result can be made more representative.

[0068] Since the extreme points correspond to the positions where the changing trends change, the appearance of the extreme points corresponding to the characteristic width timing curve usually represents the changes in the cutting stage. The embodiment of the present invention further uses all the extreme points of each characteristic width timing curve as interval points to divide the complete cutting cycle corresponding to each characteristic width timing curve into at least two interval time intervals, that is, each interval time interval corresponds to each cutting stage.

[0069] In order to ensure that other complete cutting cycles outside the characteristic width timing curve also correspond to the same number of cutting stages, an embodiment of the present invention uses the ratio between the time length of each interval time interval in each characteristic width timing curve and the time length corresponding to each characteristic width timing curve as the reference length ratio of each interval time interval in each characteristic width timing curve; after arranging the reference length ratios corresponding to the interval time intervals of each characteristic width timing curve in chronological order, a reference length ratio sequence of each characteristic width timing curve is obtained; the index value corresponding to each reference length ratio in the reference length ratio sequence is used as the stage index value corresponding to the cutting stage; in all reference length ratio sequences, the average of all reference length ratios corresponding to each stage index value is used as the corresponding interval ratio of each cutting stage; according to the interval ratios of all cutting stages, all complete cutting cycles are divided into at least two cutting stages in chronological order; the number of cutting stages is the same as the number of interval time intervals. The index value of the cutting stage corresponds to the index value of the interval time interval in the time sequence. Therefore, in order to characterize the cutting stages corresponding to all complete cutting cycles, the embodiment of the present invention calculates the average proportion of each cutting stage in the complete cutting cycle corresponding to each characteristic width timing curve, thereby obtaining a more representative cutting stage division method.

[0070] In the embodiment of the present invention, each cutting stage is sequentially used as the first cutting stage, then The method for obtaining the interval ratio of each cutting stage is expressed in the formula as follows:

[0071] ;

[0072] in, For the The interval ratio of each cutting stage; is the number of characteristic width timing curves; For the The characteristic width time series curve corresponds to the first The length of the interval; For the The time length corresponding to the characteristic width timing curve; For the The characteristic width time series curve corresponds to the first The reference length ratio of the interval time interval. For example, the total number of cutting stages is 3, the interval ratio of the first cutting stage is 0.3, the interval ratio of the second cutting stage is 0.5, and the interval ratio of the third cutting stage is 0.2; for any complete cutting cycle, the length of the complete cutting cycle is 100 minutes, then the time length of the corresponding first cutting stage is the product of 100 minutes and 0.3, that is, 30 minutes; the time length corresponding to the second cutting stage is the product of 100 minutes and 0.5, that is, 50 minutes; the time length of the third cutting stage is the product of 100 minutes and 0.2, that is, 20 minutes; that is, in the complete cutting cycle of 100 minutes, the first 30 minutes are the first cutting stage, the last 20 minutes are the third cutting stage, and the middle 50 minutes are the second cutting stage.

[0073] The reason why the output result of the cutting parameter width association model constructed according to the GRNN network has low accuracy is that the determination of the weight function value in the GRNN network only takes into account the distance between the input variable and the training sample, and does not take into account the influence or association degree of each cutting association parameter on the slit width. The more similar the fluctuation change of the cutting association parameter is to the fluctuation change of the slit width, the greater the influence or association of the corresponding cutting association parameter. Therefore, in the embodiment of the present invention, in the historical data, according to the similarity of the fluctuation change between all the association parameter time series data of each cutting association parameter in each cutting stage and the slit width time series data, the association influence degree of each cutting association parameter in each cutting stage is obtained.

[0074] Preferably, the method for obtaining the association influence degree includes:

[0075] According to all the associated parameter time series data of each cutting associated parameter corresponding to each complete cutting cycle, curve fitting is performed to obtain the associated parameter time series curve of each cutting associated parameter in each complete cutting cycle. Curve fitting can more accurately and intuitively characterize the change trend of the associated parameter time series data of each cutting associated parameter, so as to better perform correlation analysis with the slit width time series data.

[0076] In the time series curve of the associated parameters, the mean of the tangent slopes of all data points on the local curve segment corresponding to each cutting stage is taken as the mean of the associated slopes of each cutting associated parameter at each cutting stage in each complete cutting cycle; the variance of the associated parameter time series data of each cutting associated parameter at each cutting stage in each complete cutting cycle is taken as the volatility of the associated data of each cutting associated parameter at each cutting stage in each complete cutting cycle; in the time series curve of the slit width corresponding to each complete cutting cycle, the mean of the tangent slopes of all data points on the local curve segment corresponding to each cutting stage is taken as the mean of the width slope at each cutting stage in each complete cutting cycle; the variance of the slit width time series data at each cutting stage in each cutting stage in each complete cutting cycle is taken as the volatility of the width data at each cutting stage in each complete cutting cycle. The mean slope can characterize the overall change trend of the data at each cutting stage in each complete cutting cycle, while the variance can characterize the discreteness of the data at each cutting stage in each complete cutting cycle, and similarly reflects the corresponding stability.

[0077] Therefore, for each cutting-related parameter, the closer the corresponding correlation slope mean and width slope mean at the same cutting stage of the same complete cutting cycle are, the closer the overall change trend between the cutting-related parameter and the slit width is, the more the cutting-related parameter is correlated with the slit width, and the greater the corresponding influence degree is; similarly, the closer the corresponding correlation data volatility and width data volatility at the same cutting stage of the same complete cutting cycle are, the closer the data stability or volatility between the cutting-related parameter and the slit width is, the more the cutting-related parameter is correlated with the slit width, and the greater the corresponding influence degree is;

[0078] Therefore, the embodiment of the present invention further obtains the correlation influence degree of each cutting correlation parameter at each cutting stage according to the deviation between the correlation slope mean and the width slope mean, and the deviation between the correlation data volatility and the width data volatility.

[0079] Preferably, according to the deviation between the mean of the correlation slope and the mean of the width slope, and the deviation between the volatility of the correlation data and the volatility of the width data, the method for obtaining the correlation influence degree of each cutting correlation parameter at each cutting stage includes:

[0080] The difference between the mean of the associated slope and the mean of the width slope is taken as the associated slope deviation of each cutting associated parameter at each cutting stage in each complete cutting cycle; the difference between the associated data volatility and the width data volatility is taken as the associated volatility deviation of each cutting associated parameter at each cutting stage in each complete cutting cycle; the mean of all corresponding associated slope deviations of each cutting associated parameter at each cutting stage is taken as the reference slope mean; the mean of all corresponding associated volatility deviations of each cutting associated parameter at each cutting stage is taken as the reference volatility mean.

[0081] Since for each cutting correlation parameter, the closer the mean of the correlation slope is to the mean of the width slope, the closer the correlation data volatility is to the width data volatility, the more the corresponding cutting correlation parameter is correlated with the slit width, and the greater the corresponding influence; the correlation slope deviation represents the difference between the mean of the correlation slope and the mean of the width slope, and the correlation volatility deviation represents the difference between the correlation data volatility and the width data volatility;

[0082] Therefore, for all complete cutting cycles, the smaller the overall corresponding associated slope deviation of each cutting-related parameter in each cutting stage, the smaller the overall associated volatility deviation, the more associated the corresponding cutting-related parameter is with the slit width; that is, the smaller the reference slope mean, the smaller the reference volatility mean, the greater the degree of associated influence of the corresponding cutting-related parameter in the corresponding cutting stage. Therefore, the embodiment of the present invention further uses the negative correlation mapping value of the product between the reference slope mean and the reference volatility mean as the degree of associated influence of each cutting-related parameter in each cutting stage. It should be noted that the implementer can also obtain the degree of associated influence based on the reference slope mean and the reference volatility mean by other methods, such as using the negative correlation mapping value of the sum of the reference slope mean and the reference volatility mean as the degree of associated influence.

[0083] In the embodiment of the present invention, each cutting-related parameter is sequentially used as the first The cutting association parameters are The cutting-related parameters are The method for obtaining the degree of correlation influence at each cutting stage is expressed in the formula as follows:

[0084] ;

[0085] in, For the The cutting-related parameters are The degree of correlation influence at each cutting stage; is the number of complete cutting cycles; For the In a complete cutting cycle The mean width slope at each cutting stage; For the The cutting-related parameters are In a complete cutting cycle The mean of the correlation slopes at the cutting stages; For the In a complete cutting cycle The fluctuation of width data at each cutting stage; For the The cutting-related parameters are In a complete cutting cycle The volatility of the correlation data at each cutting stage; For the The cutting-related parameters are In a complete cutting cycle The correlation slope deviation at each cutting stage; For the The cutting-related parameters are In a complete cutting cycle Correlation volatility deviations at each cutting stage; For the The cutting-related parameters are The mean of the reference slopes over the cutting stages; For the The cutting-related parameters are The reference volatility mean over the cutting stages; It is an exponential function with a natural constant as the base. The implementer may also perform negative correlation mapping by other negative correlation mapping methods, such as the inverse.

[0086] Step S3: Use all the associated parameter time series data corresponding to each cutting associated parameter in each cutting stage as the training set input variable parameters of the training process of the GRNN model in each cutting stage; in each cutting stage, obtain the training weight correction coefficient of each training set input variable parameter according to the relative deviation of the abnormality degree and the degree of associated influence of each training set input variable parameter corresponding to each cutting associated parameter.

[0087] The weight function value in the training process of the GRNN model is further modified in combination with the obtained correlation influence degree; therefore, the corresponding training set is first obtained. Considering that the constructed model is a cutting parameter width correlation model, all the correlation parameter time series data corresponding to each cutting correlation parameter in each cutting stage are used as the training set input variable parameters of the training process of the GRNN model in each cutting stage; that is, all the correlation parameter time series data of each cutting correlation parameter in each cutting stage in all complete cutting cycles are used as the training set input variable parameters. It should be noted that the implementer can also use part of the complete cutting cycles as the training set according to the specific implementation environment, for example, select 60 complete cutting cycles, which will not be further elaborated here.

[0088] The subsequent weight function value is further corrected in combination with the degree of associated influence; however, considering that for each training set variable parameter, the more abnormal the corresponding numerical value is, the smaller the corresponding credibility is, then the weight function value corresponding to the training set variable parameter should be smaller; and when the corresponding degree of associated influence is greater, it means that the type of cutting associated parameter corresponding to the training set variable parameter has a greater influence on the slit width, and the corresponding weight function value should be greater. Therefore, the embodiment of the present invention combines the degree of abnormality and the degree of associated influence. In each cutting stage, the training weight correction coefficient of each training set input variable parameter is obtained according to the relative deviation of the degree of abnormality and the degree of associated influence of each training set input variable parameter corresponding to each cutting associated parameter. The correction weight of the weight function value is represented by the training weight correction coefficient. The larger the training weight correction coefficient, the larger the corresponding weight function value, and vice versa.

[0089] Preferably, the method for obtaining the training weight correction coefficient includes:

[0090] The LOF value of each training set input variable corresponding to each cutting association parameter in each cutting stage is calculated by the LOF anomaly detection algorithm; the larger the LOF value of the training set input variable relative to the overall LOF value corresponding to each cutting association parameter in each cutting stage, the more abnormal the input variable of the training set is, and the greater the corresponding variable abnormality. According to the LOF value of each training set input variable corresponding to each cutting association parameter in each cutting stage, relative to the deviation of the overall LOF value, the variable abnormality degree of each training set input variable corresponding to each cutting association parameter in each cutting stage is obtained. In the embodiment of the present invention, the LOF value calculated by the LOF anomaly detection algorithm measures the abnormality degree of each training set input variable. The implementer may also use other methods to characterize the variable abnormality degree of each training set input variable, such as the isolation forest algorithm, which will not be further elaborated here.

[0091] Preferably, the method for obtaining the abnormality degree of a variable includes:

[0092] The mean of all LOF values ​​corresponding to each cutting-related parameter in each cutting stage is used as the reference abnormal value of each cutting-related parameter in each cutting stage; the normalized value of the difference between the LOF value of each training set input variable corresponding to each cutting-related parameter in each cutting stage and the reference abnormal value is used as the variable abnormality degree of each training set input variable corresponding to each cutting-related parameter in each cutting stage. Since the larger the LOF value of the training set input variable relative to the overall LOF value corresponding to each cutting-related parameter in each cutting stage, the greater the corresponding abnormality degree of the variable, the embodiment of the present invention characterizes the overall LOF value by the mean of all LOF values ​​corresponding to each cutting-related parameter in each cutting stage, that is, the reference abnormal value. The larger the normalized value of the difference between the LOF value of each training set input variable and the reference abnormal value, the greater the corresponding abnormality degree of the variable.

[0093] In the embodiment of the present invention, the In the cutting stage Each training set input variable corresponding to the cutting association parameter is used as the first In the cutting stage The first cutting parameter corresponding to training set input variables; then In the cutting stage The first cutting parameter corresponding to The method for obtaining the abnormal degree of the input variables of the training set is expressed in the formula as follows:

[0094] ;

[0095] in, For the In the cutting stage The first cutting parameter corresponding to The degree of variable abnormality of the input variables of the training set; For the In the cutting stage The first cutting parameter corresponding to LOF value of the input variable of the training set; For the In the cutting stage The mean LOF value of all training set input variables corresponding to the cutting association parameter, that is, In the cutting stage The reference outlier values ​​corresponding to the cutting-related parameters; is a normalization function. In the embodiment of the present invention, the normalization method adopts linear normalization. The implementer may also adopt other normalization methods according to the specific implementation environment, but it is necessary to ensure that the adopted normalization method can normalize negative numbers, which will not be further elaborated here.

[0096] The cumulative value of the correlation influence degree corresponding to all types of cutting correlation parameters in each cutting stage is used as the reference cumulative value corresponding to each cutting stage; the ratio of the correlation influence degree corresponding to each cutting correlation parameter in each cutting stage to the reference cumulative value is used as the relative correlation degree of each cutting correlation parameter in each cutting stage. That is, the correlation influence degree is normalized by the reference cumulative value, so that the relative correlation degree of each cutting correlation parameter in each cutting stage is taken into account in terms of relative size, and the corresponding value is more reasonable. The implementer can also directly calculate the correlation influence degree as the relative correlation degree instead of the correlation influence degree.

[0097] The degree of variable abnormality characterizes the degree of abnormality of each training set input variable; and the relative degree of association characterizes the degree of association influence, so the smaller the degree of variable abnormality and the larger the relative degree of association, the larger the corresponding training weight correction coefficient characterizing the weight correction weight of the weight function value. Therefore, the embodiment of the present invention obtains the training weight correction coefficient of each training set input variable in each cutting association parameter in each cutting stage according to the degree of variable abnormality and the relative degree of association; both the degree of variable abnormality and the relative degree of association are positively correlated with the training weight correction coefficient.

[0098] Preferably, according to the abnormal degree and relative correlation degree of the variables, the method for obtaining the training weight correction coefficient of each training set input variable in each cutting correlation parameter in each cutting stage includes:

[0099] The normalized value of the product of the negative correlation mapping value of the variable abnormality degree and the relative correlation degree is used as the training weight correction coefficient of each training set input variable in each cutting correlation parameter in each cutting stage. It should be noted that the implementer can also obtain the training weight correction coefficient according to the variable abnormality degree and the relative correlation degree by other methods, such as using the normalized value of the ratio between the relative correlation degree and the variable abnormality degree as the training weight correction coefficient, which will not be further elaborated here.

[0100] Step S4: Correct the weight function value of the GRNN model in the training process of each cutting stage according to the training weight correction coefficient to obtain the cutting parameter width association model of each cutting stage; remotely control the laser processing system according to the cutting parameter width association model combined with the cutting seam width timing data and the associated parameter timing data in the real-time processing process.

[0101] According to the purpose of the embodiment of the present invention, after obtaining the training weight correction coefficient, the weight function value in the training process of the GRNN model of each cutting stage is further corrected according to the training weight correction coefficient to obtain the cutting parameter width association model of each cutting stage. Preferably, the method for obtaining the cutting parameter width association model includes:

[0102] During the training process of the GRNN model in each cutting stage, the a priori weight function value of each training set input variable corresponding to each cutting-related parameter in each cutting stage is obtained; that is, the a priori weight function value obtained only according to the distance between the input variable and the training sample in the prior art. Since the larger the training weight correction function, the larger the corresponding weight function value should be, the embodiment of the present invention weights the a priori weight function value by the positive correlation mapping value of the training weight correction coefficient to obtain the corrected weight function value of each training set input variable corresponding to each cutting-related parameter in each cutting stage. It should be noted that GRNN is a regression model based on a neural network well known to those skilled in the art, and its training process and the method for obtaining the a priori weight function will not be further described here. Preferably, the method for obtaining the corrected weight function value includes:

[0103] The mean of the training weight correction coefficients of all training set input variables corresponding to each cutting association parameter in each cutting stage is used as the reference correction mean; the ratio between the training weight correction coefficient and the reference correction mean is used as the weighting coefficient of each training set input variable corresponding to each cutting association parameter in each cutting stage. The obtained weighting coefficient is more reasonable by the relative size between the training weight correction coefficient of each training set input variable and the mean of all training weight correction coefficients; for each training set input variable, when its corresponding training weight correction coefficient is greater than the reference correction mean, it means that the training weight correction coefficient has a greater degree of association or influence than the overall training weight correction coefficient, so the obtained weighting coefficient is greater than 1, and after weighting the priori weight function value, the obtained modified weight function value is larger; otherwise, the corresponding modified weight function value is smaller. Therefore, the embodiment of the present invention further uses the product of the weighting coefficient and the priori weight function value as the modified weight function value of each training set input variable corresponding to each cutting association parameter in each cutting stage.

[0104] After obtaining the corrected weight function value, the GRNN model of each cutting stage is further trained according to the preset smoothing factor and the corrected weight function value to obtain the cutting parameter width association model of each cutting stage. In the embodiment of the present invention, the preset smoothing factor is set to 0.8, and the implementer can adjust it according to the specific implementation environment. It should be noted that since GRNN is a regression model based on a neural network well known to those skilled in the art, after determining the smoothing factor and the weight function value in the training process, the subsequent process of obtaining the trained cutting parameter width association model will not be further described.

[0105] Since the laser processing system usually requires real-time control, it is necessary to analyze the slit width timing data and the associated parameter timing data in the real-time processing process, so as to obtain the optimal parameter timing data required by the embodiment of the present invention at the current moment in combination with the cutting parameter width association model, so as to remotely control the laser processing system more accurately according to the optimal parameter timing data. Therefore, the embodiment of the present invention remotely controls the laser processing system according to the cutting parameter width association model combined with the slit width timing data and the associated parameter timing data in the real-time processing process.

[0106] Preferably, the method for remotely controlling the laser processing system according to the cutting parameter width association model combined with the slit width time series data and the association parameter time series data in the real-time processing process includes:

[0107] According to the time length of the complete cutting cycle corresponding to each characteristic width timing curve, the cutting stage corresponding to the current moment is determined; the cutting stage corresponding to the current moment is used as the real-time cutting stage;

[0108] In all complete cutting cycles, the mean of all slit width time series data corresponding to the real-time cutting stage is used as the ideal slit width in the real-time cutting stage; the slit width time series data at the current moment is used as the real-time slit width; in the width dimension, all slit width time series data between the ideal slit width and the real-time slit width are used as reference slit width data; the moment corresponding to the reference slit width in all complete slit cycles is used as the reference moment. When the real-time slit width is greater than the ideal slit width, it means that the slit width needs to be reduced; when the real-time slit width is less than the ideal slit width, the slit width needs to be increased. That is, when the real-time slit width is greater than the ideal slit width, the slit width time series data that is greater than or equal to the ideal slit width and less than or equal to the real-time slit width is used as the reference slit width data; when the real-time slit width is less than the ideal slit width, the slit width time series data that is greater than or equal to the real-time slit width and less than or equal to the ideal slit width is used as the reference slit width data. In addition, when the real-time slit width is equal to the ideal slit width, the embodiment of the present invention does not adjust the various cutting-related parameters in the laser processing system; in another embodiment of the present invention, when the real-time slit width is equal to the ideal slit width, the slit width timing data equal to the ideal slit width is used as reference slit width data.

[0109] All the corresponding associated parameter time series data of each cutting-related parameter at all reference moments are used as the screening parameter set corresponding to each cutting-related parameter. In each screening parameter set, one associated parameter time series data is randomly selected to form a cutting-related parameter combination; all cutting-related parameter combinations are obtained. That is, in an embodiment of the present invention, the laser power data at all reference moments are used as the screening parameter set corresponding to the laser power, the cutting speed data at all reference moments are used as the screening parameter set corresponding to the cutting speed, and the defocus data at all reference moments are used as the screening parameter set corresponding to the defocus. In the screening parameter set corresponding to the laser power, the screening parameter set corresponding to the cutting speed, and the screening parameter set corresponding to the defocus, one associated parameter time series data is randomly selected respectively to form a combination corresponding to the laser power data, the cutting speed data, and the defocus data, that is, the cutting-related parameter combination; continue to randomly select to obtain all kinds of cutting-related parameter combinations.

[0110] All the associated parameter time series data in each cutting associated parameter combination are used as the input of the cutting parameter width association model corresponding to the real-time cutting stage, and the predicted slit width corresponding to each cutting associated parameter combination is output; the predicted slit width is converted into the fitness function of the genetic algorithm, and the genetic operation screening is performed according to the elite strategy sorting according to all the predicted slit widths until the optimal slit width corresponding to the optimal solution is screened out; the associated parameter time series data in the cutting associated parameter combination corresponding to the optimal slit width is used as the optimal parameter time series data of each cutting associated parameter at the current moment. That is, according to all the predicted slit widths corresponding to all cutting associated parameter combinations, genetic operation screening is performed in combination with the elite strategy sorting to screen out the optimal solution, and the laser power data, cutting speed data and defocus data in the cutting associated parameter combination corresponding to the optimal solution are used as the optimal parameter time series data that needs to be adjusted by the laser processing system at the current moment in the embodiment of the present invention. Finally, the laser processing system is remotely controlled according to the optimal parameter time series data, that is, the associated parameter time series data corresponding to each cutting associated parameter in the laser processing system is adjusted to the size of the corresponding optimal parameter time series data. It should be noted that elite strategy ranking is an existing technology well known to those skilled in the art and will not be further defined or elaborated herein.

[0111] To summarize, the present invention first divides each cutting stage according to the time series distribution trend of the slit width in each complete cutting cycle in the historical data; obtains the degree of correlation influence according to the similarity of the fluctuation change of the slit width time series data and the associated parameter time series data of each cutting associated parameter in each cutting stage; uses the associated parameter time series data as the input variable parameter of the training set, and corrects the weight function value in the GRNN model training process according to the training weight correction coefficient obtained according to the corresponding abnormality degree and the associated influence degree, so that the output result of the cutting parameter width association model obtained is more accurate, thereby making the remote real-time control of the laser processing system according to the cutting parameter width association model more accurate.

[0112] It should be noted that the sequence of the above embodiments of the present invention is only for description and does not represent the advantages and disadvantages of the embodiments. The processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0113] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.

Claims

1. A remote control method for a laser processing system based on cloud computing, characterized in that: The method comprises: From the historical data of cutting processing performed by the laser processing control system, the slit width time series data corresponding to each complete cutting cycle and the associated parameter time series data of each cutting associated parameter are obtained; According to the extreme value distribution of the slit width time series data of each complete cutting cycle in the time series, all complete cutting cycles are divided into at least two cutting stages; wherein the number of corresponding cutting stages in all complete cutting cycles is the same; in the historical data, according to the similarity of the fluctuation changes between all the associated parameter time series data of each cutting associated parameter in each cutting stage and the slit width time series data, the associated influence degree of each cutting associated parameter in each cutting stage is obtained; All the associated parameter time series data corresponding to each cutting associated parameter in each cutting stage are used as the training set input variable parameters of the training process of the GRNN model in each cutting stage; in each cutting stage, according to the relative deviation of the abnormality degree of each training set input variable parameter corresponding to each cutting associated parameter and the degree of the associated influence, the training weight correction coefficient of each training set input variable parameter is obtained; The weight function value in the training process of the GRNN model of each cutting stage is corrected according to the training weight correction coefficient to obtain a cutting parameter width association model for each cutting stage; the laser processing system is remotely controlled according to the cutting parameter width association model combined with the slit width time series data and the association parameter time series data in the real-time processing process; The acquisition method of the cutting stage includes: Perform curve fitting on all slit width time series data corresponding to each complete cutting cycle to obtain a slit width time series curve corresponding to each complete cutting cycle; take the mode of the number of extreme points corresponding to all slit width time series curves as a characteristic reference; take all slit width time series curves whose number of extreme points is the characteristic reference as the characteristic width time series curve; Taking all extreme value points of each characteristic width timing curve as interval points, the complete cutting cycle corresponding to each characteristic width timing curve is divided into at least two interval time intervals; the ratio between the time length of each interval time interval in each characteristic width timing curve and the time length corresponding to each characteristic width timing curve is used as the reference length ratio of each interval time interval in each characteristic width timing curve; Arrange the reference length proportions corresponding to the interval time intervals of each characteristic width timing curve in time order to obtain a reference length proportion sequence of each characteristic width timing curve; use the index value corresponding to each reference length proportion in the reference length proportion sequence as the stage index value corresponding to the cutting stage; In all reference length ratio sequences, the average of all reference length ratios corresponding to each stage index value is used as the interval ratio of each corresponding cutting stage; according to the interval ratios of all cutting stages, all complete cutting cycles are divided into at least two cutting stages in chronological order; the number of the cutting stages is the same as the number of interval time intervals; The method for obtaining the association influence degree includes: Perform curve fitting according to all associated parameter time series data of each cutting associated parameter corresponding to each complete cutting cycle, to obtain an associated parameter time series curve of each cutting associated parameter in each complete cutting cycle; In the associated parameter time series curve, the mean of the tangent slopes of all data points on the local curve segment corresponding to each cutting stage is taken as the mean of the associated slopes of each cutting associated parameter at each cutting stage in each complete cutting cycle; the variance of the associated parameter time series data of each cutting associated parameter at each cutting stage in each complete cutting cycle is taken as the volatility of the associated data of each cutting associated parameter at each cutting stage in each complete cutting cycle; In the slit width timing curve corresponding to each complete cutting cycle, the mean of the tangent slopes of all data points on the local curve segment corresponding to each cutting stage is taken as the mean of the width slopes at each cutting stage in each complete cutting cycle; the variance of the slit width timing data at each cutting stage in each complete cutting cycle is taken as the width data volatility at each cutting stage in each complete cutting cycle; According to the deviation between the mean of the correlation slope and the mean of the width slope, and the deviation between the volatility of the correlation data and the volatility of the width data, the correlation influence degree of each cutting correlation parameter at each cutting stage is obtained; The method for obtaining the training weight correction coefficient includes: The LOF anomaly detection algorithm is used to calculate the LOF value of each training set input variable corresponding to each cutting-related parameter in each cutting stage; According to the LOF value of each training set input variable corresponding to each cutting association parameter in each cutting stage, relative to the deviation of the overall LOF value, the variable abnormality degree of each training set input variable corresponding to each cutting association parameter in each cutting stage is obtained; The cumulative value of the correlation influence degree corresponding to all types of cutting-related parameters in each cutting stage is used as the reference cumulative value corresponding to each cutting stage; the ratio of the correlation influence degree corresponding to each cutting-related parameter in each cutting stage to the reference cumulative value is used as the relative correlation degree of each cutting-related parameter in each cutting stage; According to the abnormal degree of the variable and the relative correlation degree, a training weight correction coefficient of each training set input variable in each cutting correlation parameter in each cutting stage is obtained; the abnormal degree of the variable and the relative correlation degree are both positively correlated with the training weight correction coefficient; The method for obtaining the cutting parameter width association model includes: In the training process of the GRNN model in each cutting stage, the prior weight function value of each training set input variable corresponding to each cutting association parameter in each cutting stage is obtained; the prior weight function value is weighted by the positive correlation mapping value of the training weight correction coefficient to obtain the corrected weight function value of each training set input variable corresponding to each cutting association parameter in each cutting stage; The GRNN model of each cutting stage is trained according to the preset smoothing factor and the modified weight function value to obtain the cutting parameter width association model of each cutting stage.

2. The remote control method of a laser processing system based on cloud computing according to claim 1, characterized in that: The method for obtaining the correlation influence degree of each cutting correlation parameter at each cutting stage according to the deviation between the correlation slope mean and the width slope mean, and the deviation between the correlation data volatility and the width data volatility comprises: The difference between the mean of the correlation slope and the mean of the width slope is used as the correlation slope deviation of each cutting correlation parameter at each cutting stage in each complete cutting cycle; the difference between the correlation data volatility and the width data volatility is used as the correlation volatility deviation of each cutting correlation parameter at each cutting stage in each complete cutting cycle; The mean of all associated slope deviations corresponding to each cutting-related parameter at each cutting stage is taken as the reference slope mean; the mean of all associated volatility deviations corresponding to each cutting-related parameter at each cutting stage is taken as the reference volatility mean; The negative correlation mapping value of the product of the reference slope mean and the reference volatility mean is used as the correlation influence degree of each cutting correlation parameter at each cutting stage.

3. The remote control method of a laser processing system based on cloud computing according to claim 1, characterized in that: The method for obtaining the abnormal degree of the variable includes: The mean of all LOF values ​​corresponding to each cutting-related parameter in each cutting stage is used as the reference outlier value of each cutting-related parameter in each cutting stage; The normalized value of the difference between the LOF value of each training set input variable corresponding to each cutting-related parameter in each cutting stage and the reference abnormal value is used as the variable abnormality degree of each training set input variable corresponding to each cutting-related parameter in each cutting stage.

4. The remote control method of a laser processing system based on cloud computing according to claim 3 is characterized in that: The method for obtaining the training weight correction coefficient of each training set input variable in each cutting association parameter in each cutting stage according to the abnormal degree of the variable and the relative correlation degree includes: The normalized value of the product of the negative correlation mapping value of the abnormal degree of the variable and the relative correlation degree is used as the training weight correction coefficient of each training set input variable in each cutting association parameter in each cutting stage.

5. The remote control method of a laser processing system based on cloud computing according to claim 1, characterized in that: The method for obtaining the modified weight function value comprises: The mean of the training weight correction coefficients of all training set input variables corresponding to each cutting-related parameter in each cutting stage is used as a reference correction mean; the ratio between the training weight correction coefficient and the reference correction mean is used as a weighting coefficient of each training set input variable corresponding to each cutting-related parameter in each cutting stage; The product of the weighting coefficient and the priori weight function value is used as the modified weight function value of each training set input variable corresponding to each cutting association parameter in each cutting stage.

6. The remote control method of a laser processing system based on cloud computing according to claim 1, characterized in that: The method for remotely controlling the laser processing system according to the cutting parameter width association model combined with the slit width time series data and the association parameter time series data in the real-time processing process includes: According to the time length of the complete cutting cycle corresponding to each characteristic width timing curve, the cutting stage corresponding to the current moment is determined; the cutting stage corresponding to the current moment is used as the real-time cutting stage; In all complete cutting cycles, the average of all slit width time series data corresponding to the real-time cutting stage is used as the ideal slit width in the real-time cutting stage; the slit width time series data at the current moment is used as the real-time slit width; In the width dimension, all the slit width time series data between the ideal slit width and the real-time slit width are used as reference slit width data; the time corresponding to the reference slit width in all complete slit cycles is used as the reference time; All the associated parameter time series data corresponding to each cutting associated parameter at all reference times are used as the screening parameter set corresponding to each cutting associated parameter; In each screening parameter set, one associated parameter time series data is randomly selected to form a cutting associated parameter combination; and all cutting associated parameter combinations are obtained; All the associated parameter time series data in each cutting associated parameter combination are used as the input of the cutting parameter width association model corresponding to the real-time cutting stage, and the predicted slit width corresponding to each cutting associated parameter combination is output; the predicted slit width is converted into the fitness function of the genetic algorithm, and the genetic operation screening is performed according to the elite strategy sorting according to all the predicted slit widths until the optimal slit width corresponding to the optimal solution is screened out; the associated parameter time series data in the cutting associated parameter combination corresponding to the optimal slit width is used as the optimal parameter time series data of each cutting associated parameter at the current moment; The laser processing system is remotely controlled according to the optimal parameter timing data.