Classification method and system of variable star light variable curve based on machine learning
Through a machine learning-based method, ZTF observation data and 23 light-change parameters are used to classify variable star light-change curves, which solves the problem of low accuracy of variable star classification in the existing technology, and significantly improves the accuracy and efficiency of classification.
Patent Information
- Application Number
- CN202510101368.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-01-22
- Publication Date
- 2025-05-27
AI Technical Summary
The prior art has the problem of low accuracy in the classification of variable star light curves, especially the classification effect of certain types of variable stars such as CEP, LPV, etc. is poor, and is easily disturbed by weather factors or equipment errors.
Using a machine learning-based method, by constructing ZTF observation data and performing variable star screening processing, a variable star light change curve is obtained, 23 optical change parameters are determined, and all optical change parameter data are obtained through data cross-match. Finally, a machine learning model is used for classification to improve the accuracy and recall of classification.
The classification efficiency and classification accuracy of variable star data are improved, especially in the classification of variable star types such as CEP and LPV, which significantly improves the accuracy rate and reduces inaccuracy caused by weather or equipment errors.
Smart Images
Figure CN120045997A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of astronomical data processing, and particularly to a classification method and system for variable star light curves based on machine learning. Background Art
[0002] With the rapid progress of astronomical observation means, the targets and wavebands of astronomical observations have further developed from single targets and single wavebands to multi-targets and full wavebands. Various sky survey projects with massive data outputs have emerged continuously. With the help of advanced machine learning algorithms, astronomers have more efficiently completed tasks such as automatic detection and classification of astronomical targets, parameter estimation of celestial bodies, time series data analysis, data noise reduction and generation, and have searched for rare celestial bodies and discovered new celestial body types in astronomical datasets by means of advanced data mining methods. These achievements have revealed the evolutionary processes of quasars, variable stars, and the Milky Way on the cosmic time scale, providing a solid foundation for explaining complex phenomena in the universe.
[0003] Variable sources are generally defined as stars with variable luminosities. According to the periodic characteristics of their luminosity changes, they can be divided into periodic variable stars, weak periodic variable stars, and aperiodic variable stars. The luminosity change durations and intensity levels of different types of variable stars are different. The identification and classification of variable sources have always been the basis of astronomical research work. In recent years, there have been more and more achievements in the identification and search of variable sources. With the significant increase in the number of astronomical datasets, the method of manual observation and classification has achieved little effect. Traditional observation and statistical methods can no longer meet the needs of the era of astronomical big data. Some methods for obtaining variable sources through machine learning algorithms have obtained relatively prominent results. However, machine learning has not been perfect in the aspect of variable source classification, and sometimes problems will also appear in terms of accuracy. For example, in the related technologies, the double-layer hierarchical random forest and convolutional variational autoencoder, and for the luminosity labels generated by the encoder, the classification accuracy for a part of variable stars exceeds 90%, but the accuracy for variable stars such as CEP and LPV is very low. In addition, for example, in the classification results of the method based on convolutional autoencoding for PVS, only the classification effects of three types of variable stars, namely AGN, EB, and RR, meet the requirements, and the prediction success rates of five types of variable stars, namely CEP, LPV, Mira, etc., are all lower than 65%. Therefore, these data may be interfered by reasons such as weather factors or equipment errors, and there are often problems of inaccurate data and difficulty in screening. Summary of the Invention
[0004] In order to solve the above technical problems, the purpose of the present invention is to provide a classification method and system for variable star light curves based on machine learning, which can improve the classification efficiency and classification accuracy of variable star data based on machine learning.
[0005] The first technical solution adopted by the present invention is: a classification method for variable star light curves based on machine learning, comprising the following steps:
[0006] Construct ZTF observation data and perform variable star screening processing to obtain variable star light curves;
[0007] Determine 23 kinds of light curve parameters according to the variable star light curves;
[0008] According to the 23 kinds of light curve parameters, obtain all light curve parameter data through data cross-matching and classify them through a machine learning model to obtain the classification result of the variable star light curves.
[0009] Further, the step of constructing ZTF observation data and performing variable star screening processing to obtain variable star light curves specifically includes:
[0010] Construct ZTF observation data;
[0011] Based on the variable star catalog of asassn and the standard star catalog of sdss, identify and screen the ZTF observation data to obtain the target observation data;
[0012] Extract the light curve features of the target observation data to obtain variable star light curves.
[0013] Further, the step of constructing ZTF observation data specifically includes:
[0014] Obtain nine high-confidence variable star catalogs through the paper database. The nine high-confidence variable star catalogs include EW, SR, RRAB, EA, ROT, RRC, Mira, HADS, and CEP;
[0015] Obtain three non-periodic variable star catalogs through the Gaia database. The three non-periodic variable star catalogs include LPV, CV, and AGN;
[0016] Obtain two variable star catalogs through the paper database. The two variable star catalogs include ELL and YSO;
[0017] Merge the nine high-confidence variable star catalogs, the three non-periodic variable star catalogs, and the two variable star catalogs to construct ZTF observation data.
[0018] Further, the step of identifying and screening the ZTF observation data based on the variable star catalog of asassn and the standard star catalog of sdss to obtain the target observation data specifically includes:
[0019] Based on the variable star catalog of asassn and the standard star catalog of sdss, divide the ZTF observation data according to a preset ratio to construct an asassn training set, an asassn test set, an sdss training set, and an sdss test set;
[0020] Merge the asassn training set and the sdss training set, sort them in ascending order, and construct a copy of the training set;
[0021] Calculate the proportion of variable star labels for each segmented data on the copy of the training set, and draw a relationship graph between the variable star probability and the preset parameters by comparing the preset parameters with the proportion of variable stars within the segment;
[0022] According to the relationship graph between the variable star probability and the preset parameters, calculate the performance of all parameters on the test set, select some data with preset parameters greater than 0.95, and construct the target observation data.
[0023] Furthermore, the 23 light variation parameters specifically include skewness, kurtosis, amplitude, median magnitude, chi-square, standard deviation of 3 iterations, difference between the maximum and minimum magnitudes, median absolute deviation, small-sample kurtosis, von Neumann coefficient, Shapiro-Wilk, Stetson-K, ratio of high and low amplitudes, proportion of data points outside one error interval, gskew, MedianBRP, period, f1_power, maximum percentage difference between the maximum or minimum star and the median, ZTF observation number of the variable star itself, right ascension, declination, and number of observations.
[0024] Furthermore, the step of obtaining all light variation parameter data through data cross-matching according to the 23 light variation parameters and classifying them through a machine learning model to obtain the classification result of the variable star light variation curve specifically includes:
[0025] Based on the 23 light variation parameters, retain the coordinate information, cross-compare with the ZTF observation items, retain the part with r-band signal-to-noise ratio greater than 20 and ZTF observation times greater than 50, and cross with Gaia to obtain all light variation parameter data;
[0026] Perform data augmentation and data balancing processing on all light variation parameter data to obtain preprocessed light variation parameter data;
[0027] Classify the preprocessed light variation parameter data through a machine learning model to obtain the classification result of the variable star light variation curve.
[0028] The second technical solution adopted by the present invention is: a classification system for variable star light variation curves based on machine learning, including:
[0029] The first module is used to construct ZTF observation data and perform variable star screening processing to obtain variable star light variation curves;
[0030] The second module is used to determine 23 light variation parameters according to the variable star light variation curves;
[0031] The third module is used to obtain all the variable star light curve parameter data through data cross-matching according to 23 light variation parameters and classify them through a machine learning model to obtain the classification result of the variable star light curve.
[0032] The beneficial effects of the method and system of the present invention are as follows: By constructing ZTF observation data and performing variable star screening and processing, the variable star light curve is obtained to screen the variable stars in the LAMOST observation targets to be classified, thereby reducing the scope of variable star classification and improving the classification accuracy. Further, according to the variable star light curve, 23 light variation parameters are determined. Finally, all the light variation parameter data is obtained through data cross-matching and classified through a machine learning model. By balancing the number of different types of variable stars in the observation data, a quantity distribution beneficial to the machine learning classification algorithm is obtained. Then, according to the coordinates of each variable star, cross-matching with the ZTF observation data, and a certain number of other parameters are obtained according to Gaia DR3 Astrophysical Params. Using the above data for LightGBM classification can improve the accuracy and recall rate of the classification result. Description of the Drawings
[0033] Figure 1 is the flowchart of the steps of a method for classifying variable star light curves based on machine learning according to the present invention;
[0034] Figure 2 is the structural block diagram of a system for classifying variable star light curves based on machine learning according to the present invention;
[0035] Figure 3 is the schematic diagram of the steps of data information integration provided by a specific embodiment of the present invention;
[0036] Figure 4 is the schematic flowchart of data parameter calculation provided by a specific embodiment of the present invention;
[0037] Figure 5 is the schematic diagram of the comparison between the final result and the high-confidence star catalog provided by a specific embodiment of the present invention. Detailed Embodiment
[0038] The present invention will be further described in detail below with reference to the drawings and specific embodiments. For the step numbers in the following embodiments, they are only set for the convenience of description and explanation, and no limitation is imposed on the order between the steps. The execution order of each step in the embodiments can be adaptively adjusted according to the understanding of those skilled in the art.
[0039] First of all, it should be noted that currently, the overall classification methods for variable stars in major astronomical observation projects tend to be consistent. Taking the GCVS variable star classification table as an example, variable stars are generally divided into eight categories: eruptive variables, pulsating variables, rotating variables, cataclysmic variables, eclipsing binaries, X-ray variables, other variables, and new variables. Under different types of variable stars, several to dozens of different subtypes can be further divided. The light variation characteristics of different subtypes have relatively small differences overall, which makes it extremely difficult to classify each specific type of variable star.
[0040] Taking machine learning algorithms as an example, a large number of classification types require the trained model to have extremely high robustness. At the same time, in terms of classification parameters, the model needs to have more key parameters for different subtypes of variable stars, which makes it difficult to select variable stars and training set parameters. Taking the embodiments of the present invention as an example, the training set includes nine types of periodic variable stars, which cover three major categories of pulsating variables, rotating variables, and eclipsing binaries in the variable star catalog. On this basis, to obtain non-periodic variable stars such as other variables and new variables, which account for a relatively small proportion of variable stars, it is necessary to make the machine learning training set cover all categories of the variable star general catalog, and in terms of classification parameters, cover as comprehensively as possible to ensure that the accuracy rate and recall rate on the test set are high enough. This can ensure that in the final non-periodic star catalog of specific subcategories, the proportion of target variable stars in the total amount is high enough.
[0041] Based on the above situation, the embodiments of the present invention choose to add AGN, LPV, ELL, CV, and YSO variable stars to the classification training set of periodic variable stars. The added new type of training set covers the other seven categories except other variables. Since other variables and new variables account for a relatively low proportion of each subcategory of variable stars in the variable star general catalog, therefore, the embodiments of the present invention select two categories with a relatively large proportion of AGN and YSO in the new variables, and screen and give the final star catalog by comparing the Hα emission line and Li absorption line in the spectra of YSO variable stars.
[0042] Refer to Figure 1 , the present invention provides a method for classifying the light curves of variable stars based on machine learning, and the method includes the following steps:
[0043] S100. Construct ZTF observation data and perform variable star screening processing to obtain the light curves of variable stars;
[0044] S110. Construct ZTF observation data;
[0045] Specifically, nine high-confidence variable star catalogs are obtained through the paper database. The nine high-confidence variable star catalogs include EW, SR, RRAB, EA, ROT, RRC, Mira, HADS, and CEP. Three non-periodic variable star catalogs are obtained through the Gaia database. The three non-periodic variable star catalogs include LPV, CV, and AGN. Two variable star catalogs are obtained through the paper database. The two variable star catalogs include ELL and YSO. The nine high-confidence variable star catalogs, the three non-periodic variable star catalogs, and the two variable star catalogs are merged to construct ZTF observation data.
[0046] In this embodiment, as Figure 3 shown, the present invention uses multi-observation project data. This method uses the data training set published in the paper by Qiao et al. A total of 9 types of periodic variable stars are used as the main part of the data set to exclude periodic variable stars, and then three non-periodic variable stars from Gaia are added. The characteristics of these three variable stars are obvious and cover the remaining main categories of variable stars in the GCVS variable star catalog. Finally, two variable stars with a relatively large distribution quantity in the remaining variable star catalog are added as category distinctions to improve the recognition accuracy of YSO variable stars.
[0047] S120. Identify and screen the ZTF observation data based on the variable star catalog of asassn and the standard star catalog of sdss to obtain the target observation data;
[0048] Specifically, based on the variable star catalog of asassn and the standard star catalog of sdss, the ZTF observation data is divided according to a preset ratio to construct an asassn training set, an asassn test set, an sdss training set, and an sdss test set. The asassn training set and the sdss training set are merged and sorted in ascending order to construct a copy of the training set. Calculate the ratio of variable star labels on each segment of data in the copy of the training set. By comparing the preset parameters with the ratio of variable stars within the segment, draw a relationship diagram between the variable star probability and the preset parameters. According to the relationship diagram between the variable star probability and the preset parameters, calculate the performance of all parameters on the test set, and select the part of the data with a preset parameter greater than 0.95 to construct the target observation data.
[0049] In this embodiment, after obtaining the observation data, first, an identification standard is established using the variable stars and standard stars provided by sdss to screen the variable stars in the LAMOST observation targets to be classified, thereby reducing the scope of variable star classification and improving the classification accuracy. Among them, the parameter selection for identification and the process of determining variable stars are as follows: First, obtain the variable star catalog from asassn and the standard star catalog from sdss, and divide the two into a training set asassn_train, sdss_train and a test set asassn_test, sdss_test in a ratio of seven to three, where the variable star label is 1 and the non-variable star label is 0. Then, merge the training sets, sort and save copies of the training sets separately in ascending order for each parameter, and calculate the proportion of variable star labels on each segment of data on each copy. By comparing the parameter with the proportion of variable stars within the segment, a relationship diagram of variable star probability and parameter is drawn, and it can be seen the accuracy of different parameters in determining whether a star is a variable star. Further, calculate the performance of all parameters on the test set, comprehensively judge and select a parameter as the final evaluation standard. Finally, calculate the part of the LAMOST observation targets where the parameter is greater than 0.95, extract the light curves of this part, and calculate the light curve parameters.
[0050] S130. Extract the light variation characteristics of the target observation data to obtain the light curve of the variable star.
[0051] S200. Determine 23 light variation parameters according to the light curve of the variable star;
[0052] Specifically, simple statistical features are obtained from the mathematical characteristics of the light curve, including the mean magnitude, median magnitude, standard deviation of magnitude, number of observations, difference between the maximum and minimum magnitudes, etc.
[0053] Furthermore, 19 parameters calculated from the simple statistical features, such as skewness, kurtosis, amplitude, median magnitude, chi-square, three-iteration standard deviation, difference between the maximum and minimum magnitudes, median absolute deviation, small-sample kurtosis, von Neumann coefficient, ShapiroWilk, StetsonK, ratio of high and low amplitudes, proportion of data points outside one error interval, gskew, MedianBRP, period, f1_power, maximum percentage difference between the maximum or minimum magnitude and the median, together with the ZTF observation number, right ascension, declination, and number of observations of the variable star itself, form a basic parameter set, totaling 23 light variation parameters.
[0054] Taking some of the parameters as an example, the calculation methods of each parameter are as follows:
[0055] Skewness, kurtosis: Calculated by the skew and kurtosis functions in the scipy statistical function library.
[0056] Amplitude calculation: By sorting the magnitude values of the light curves in ascending order, subtracting the median of the 5% smallest magnitude values from the median of the 5% largest magnitude values, and dividing the result by 2, the amplitude is obtained.
[0057] Median magnitude: Obtained by the std function of numpy.
[0058] Chi-square: Let N_data be the length of the magnitude observation data (len(mag_list)), w be the square root of the magnitude error (magerr_list**-2), and Wmean be the mean of mag_list. Then the calculation formula for chi-square (chisquare) is
[0059]
[0060] The calculation methods of the remaining parameters are relatively complex and will not be elaborated here.
[0061] S300. According to 23 light curve parameters, all light curve parameter data are obtained through data cross-matching and classified by a machine learning model to obtain the classification results of variable star light curves.
[0062] Specifically, based on the 23 light curve parameters, the coordinate information is retained, cross-compared with the ZTF observation project, the part with an r-band signal-to-noise ratio greater than 20 and the ZTF observation times greater than 50 is retained, and cross-compared with Gaia to obtain all light curve parameter data; data augmentation and data balancing processing are performed on all light curve parameter data to obtain preprocessed light curve parameter data; the preprocessed light curve parameter data is classified by a machine learning model to obtain the classification results of variable star light curves.
[0063] In this embodiment, the data flow processing is as Figure 4 shown. First, the obtained training set star catalog is cross-compared with the ZTF observation project, and data with an r-band and observation times greater than 50 are screened. Then, the cross-obtained data is cross-compared with Gaia to obtain the remaining parameters, and these characteristic parameters constitute all parts of the training set and the test set.
[0064] Retain the coordinate information of the data with the obtained parameters, and cross-compare with observation projects such as Gaia to obtain the remaining parameters. Among them, through the Topcat software, star catalogs of observation projects such as simbad, MASS, and Gaia DR3 Astrophysical Params can be obtained. By setting the cross-comparison radius to 1 through the built-in comparison program of the software, the cross-comparison result of the training set and Gaia can be obtained.
[0065] Further evaluate the classification effect, including precision, recall, F1-score, ROC curve, etc. Compare the results of different machine learning methods. Calculate the precision through the following formula, and its expression is:
[0066]
[0067] Among them, TP represents True Positive: a positive sample predicted as positive by the model. True Negative (TN): a negative sample predicted as negative by the model. TP and TN represent the part where a model predicts correctly. FP is False Positive: a sample predicted as positive by the model but actually negative; False Negative (FN): a sample predicted as negative by the model but actually positive; FP and FN represent the part where a model predicts wrongly, and are respectively called Type I Error and Type II Error in statistics.
[0068] Through the above classification effect evaluation, it can be seen that on the training set and the test set, the precision and recall rates of different variable stars meet the requirements for variable star determination in astronomy.
[0069] The embodiment of the present invention further adds a data enhancement and data balancing part to try to weaken the influence brought by the difference in the number of different variable stars in the training set. The enhancement methods used include SMOTE and ADASYN. By introducing the fit_resample function in the SMOTE and ADASYN modules under imblearn.over_sampling, data enhancement is performed on the variable star samples with a smaller number. Although the data enhancement does not have a relatively obvious improvement effect in this study, this method of solving the unbalanced data distribution provides a reference for subsequent similar studies.
[0070] Finally, integrate the LAMOST low-resolution and medium-resolution data, retain the coordinate information, cross-compare with the ZTF observation project, retain the part with an r-band signal-to-noise ratio greater than 20 and the number of ZTF observations greater than 50, and cross with Gaia to obtain the remaining parameters. Use different machine learning classification methods to classify the LAMOST observation targets, compare the YSO variable star table given in each classification result with the published total YSO variable star table, and select and screen a certain number of YSO variable stars for the final manual verification.
[0071] The classification results are as Figure 5 shown, and the machine learning classification results conform to the coordinate distribution law of YSO variable stars
[0072] In summary, in the embodiments of the present invention, nine types of periodic variable stars from published papers, three types of aperiodic variable stars from Gaia publications, and two types of aperiodic variable stars disclosed in papers are integrated; the above data is cross-compared with ZTF to obtain light curves; the light curves are calculated to obtain 23 light curve parameters; the training set is cross-compared with observation items such as Gaia DR3 AstrophysicalParams to obtain other parameters; the obtained training set is classified using machine learning methods such as LightGBM, XGBoost, and RF, and the effects and efficiencies of different classifications are compared; the trained model is used for the classification of LAMOST observation targets, and the integration and calculation of parameters are the same as those for obtaining the training set; YSO variable stars are selected as the final identified target variable stars; the spectral data of the YSO candidate list is observed, and the certification results are given.
[0073] Referring to Figure 2 , a classification system for variable star light curves based on machine learning, comprising:
[0074] The first module 201 is used to construct ZTF observation data and perform variable star screening processing to obtain variable star light curves;
[0075] The second module 202 is used to determine 23 light curve parameters according to the variable star light curves;
[0076] The third module 203 is used to obtain all light curve parameter data through data cross-matching according to the 23 light curve parameters and classify them through a machine learning model to obtain the classification results of variable star light curves.
[0077] The content in the above method embodiments is applicable to the system embodiments of the present invention. The functions specifically implemented by the system embodiments of the present invention are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those of the above method embodiments.
[0078] The above is a specific description of the preferred embodiments of the present invention, but the present invention is not limited to the described embodiments. Those skilled in the art can make various equivalent deformations or substitutions without departing from the spirit of the present invention, and these equivalent deformations or substitutions are all included within the scope defined by the claims of this application.
Claims
1. A classification method for variable star light curves based on machine learning, characterized in that: The following steps are involved: Construct ZTF observation data and perform variable star screening to obtain variable star light curves; According to the light curve of variable stars, 23 light variation parameters are determined; Based on the 23 light variation parameters, all light variation parameter data are obtained through data cross-matching and classified through machine learning models to obtain the classification results of variable star light curves.
2. The method for classifying variable star light curves based on machine learning according to claim 1, characterized in that: The step of constructing ZTF observation data and performing variable star screening processing to obtain variable star light curves specifically includes: Construct ZTF observation data; Based on the variable star catalog of asassn and the standard star catalog of sdss, the ZTF observation data is identified and screened to obtain the target observation data; The light variation characteristics of the target observation data are extracted to obtain the light variation curve of the variable star.
3. The method for classifying variable star light curves based on machine learning according to claim 2, characterized in that: The step of constructing ZTF observation data specifically includes: Nine high-confidence variable star catalogs were obtained through the paper database, including EW, SR, RRAB, EA, ROT, RRC, Mira, HADS and CEP; Obtain three non-periodic variable star catalogs through the Gaia database, including LPV, CV and AGN; Obtain two variable star catalogs through the paper database, including ELL and YSO; Nine high-confidence variable star catalogs, three non-periodic variable star catalogs and two variable star catalogs are combined to construct the ZTF observation data.
4. The method for classifying variable star light curves based on machine learning according to claim 3, characterized in that: The step of identifying and screening ZTF observation data based on the variable star catalog of ASASSN and the standard star catalog of SDSSS to obtain target observation data specifically includes: Based on the variable star catalog of asassn and the standard star catalog of sdss, the ZTF observation data is divided according to the preset ratio to construct the asassn training set, asassn test set, sdss training set and sdss test set; Merge the asassn training set with the sdss training set, sort them in ascending order, and construct a training set copy; Calculate the ratio of variable star labels in each segment of the training set copy, and draw a relationship between the variable star probability and the preset parameters by comparing the preset parameters with the ratio of variable stars in the segment; According to the relationship diagram between the probability of variable stars and the preset parameters, the performance of all parameters on the test set is calculated, and some data with preset parameters greater than 0.95 are selected to construct the target observation data.
5. The method for classifying variable star light curves based on machine learning according to claim 4, characterized in that: The 23 light variable parameters specifically include skewness, kurtosis, amplitude, median magnitude, chi-square, three-iteration standard deviation, difference between maximum and minimum magnitude, median absolute deviation, small sample kurtosis, von Neumann coefficient, ShapiroWilk, StetsonK, ratio of high and low amplitudes, proportion of data points outside one error interval, gskew, MedianBRP, period, f1_power, maximum percentage difference between the largest or smallest star and the median, ZTF observation number of the variable star itself, right ascension, declination and number of observations.
6. The method for classifying variable star light curves based on machine learning according to claim 5, characterized in that: The step of obtaining all light variable parameter data by data cross-matching according to the 23 light variable parameters and classifying them by a machine learning model to obtain the classification result of the variable star light curve specifically includes: Based on the coordinate information of 23 light-variable parameters, cross-check with the ZTF observation project, retain the parts with r-band signal-to-noise ratio greater than 20 and ZTF observation times greater than 50, cross-check with Gaia, and obtain all light-variable parameter data; Perform data enhancement and data balancing on all light-variable parameter data to obtain pre-processed light-variable parameter data; The preprocessed light variation parameter data is classified through a machine learning model to obtain the classification results of the variable star light curve.
7. A classification system for variable star light curves based on machine learning, characterized in that: Includes the following modules: The first module is used to construct ZTF observation data and perform variable star screening processing to obtain variable star light curves; The second module is used to determine 23 light variation parameters based on the light variation curve of variable stars; The third module is used to obtain all light variation parameter data through data cross-matching based on 23 light variation parameters and classify them through machine learning models to obtain the classification results of variable star light variation curves.