Search

CN-121997743-A - Service life prediction method based on CIGS photovoltaic panel material degradation of each layer

CN121997743ACN 121997743 ACN121997743 ACN 121997743ACN-121997743-A

Abstract

The invention belongs to the technical field of material life prediction, and particularly discloses a life prediction method based on material degradation of each layer of a CIGS photovoltaic panel, which comprises the following steps of S1, establishing baseline data and a CIGS degradation mode library; S2, obtaining a time sequence data set, S3, training a random forest model, S4, encoding a degradation type label into numerical characteristics, splicing the numerical characteristics with physicochemical data characteristics, calculating a minimum accumulated distance, judging a confidence level and a degradation level to obtain an input characteristic set of the LSTM model, S5, constructing the LSTM model, and S6, predicting the service life. According to the life prediction method based on the degradation of each layer of material of the CIGS photovoltaic panel, the parameters with the largest influence on the life can be rapidly positioned through multi-source data fusion and machine learning model analysis, the prediction error is reduced, and meanwhile, as outdoor measured data are accumulated, the model can be continuously and iteratively optimized, so that the prediction precision is further improved.

Inventors

  • HU ZHENFEI
  • CHEN GUI

Assignees

  • 成都威尔纳思科技有限公司

Dates

Publication Date
20260508
Application Date
20260127

Claims (10)

  1. 1. A life prediction method based on the material degradation of each layer of CIGS photovoltaic panel is characterized by comprising the following steps: S1, establishing baseline data and a CIGS degradation mode library, wherein physicochemical data comprise CIGS crystallinity, znO light transmittance, interface adhesion, temperature coefficient, CIGS element proportion, cdS layer S content, packaging layer water vapor permeability and environmental humidity; s2, data acquisition, outlier processing, missing value filling, normalization processing and time sequence alignment are carried out, and a time sequence data set is obtained; S3, training a random forest model, outputting importance scores of physicochemical data, and screening physicochemical data characteristics input as a model; S4, encoding the degradation type label into a numerical value characteristic, then splicing the numerical value characteristic with the physicochemical data characteristic as a static characteristic, unifying and standardizing the length, calculating the minimum accumulated distance through a dynamic time warping algorithm, and judging the confidence level and the degradation level to obtain an input characteristic set of the LSTM model; S5, constructing an LSTM model, wherein the LSTM model comprises an input layer, three hidden layers, a full-connection layer and an output layer; s6, life prediction, namely setting an invalidation threshold value, and predicting the invalidation time point to be the time point which is lower than the invalidation threshold value for the first time to obtain the residual life.
  2. 2. The method for predicting the service life of each layer of material degradation of a CIGS photovoltaic panel according to claim 1, wherein in S1, the CIGS degradation mode library comprises degradation type labels, characteristic thresholds and time sequence change rules, and the degradation type labels comprise CIGS element segregation, znO window layer oxidation, encapsulation layer aging and interface stripping.
  3. 3. The method for predicting the service life of each layer of material degradation based on the CIGS photovoltaic panel according to claim 1, wherein in S1, non-core power generation areas at the edges and corners of the assembly are taken, a sealing glue is used for sealing a notch after sampling, a glass substrate is peeled off, a surface packaging layer is removed, the CIGS crystallinity is calculated through X-ray diffraction measurement, and the CIGS element proportion is measured through an inductively coupled plasma-mass spectrometry.
  4. 4. The life prediction method based on the degradation of the material of each layer of the CIGS photovoltaic panel according to claim 1, wherein S2 is specifically: And acquiring 1 time of nondestructive index in each quarter, acquiring 1 time of the nondestructive index in each 6 months after a new component, acquiring 1 time of the destructive index in each year from the 2 nd year, replacing an abnormal value by a moving average value of the previous 3 times of data of the index, if the data of a certain quarter is missing, supplementing the data by a linear interpolation method, mapping the indexes of different dimensions to a [0,1] interval, and integrating all indexes by using time stamps to form a time sequence data set of the component, time and multidimensional characteristics.
  5. 5. The life prediction method based on the degradation of each layer of material of the CIGS photovoltaic panel according to claim 1, wherein in S3, physicochemical data characteristics of at least 5 points before being classified are screened as model inputs.
  6. 6. The life prediction method based on the degradation of the material of each layer of the CIGS photovoltaic panel according to claim 1, wherein in S4, the normalized calculation formula is as follows: ; Where x is the input data, μ is the mean, and σ is the standard deviation.
  7. 7. The method for predicting lifetime of material degradation of layers of CIGS photovoltaic panels according to claim 1, wherein in S4, the confidence level and degradation level are determined as follows: similarity=1- (minimum cumulative distance/sequence length), the closer the similarity is to 1, the higher the matching degree is; Confidence level Wherein sim is the current similarity, conf ε [0,1]; high confidence match, determining a degradation type label, and judging that the degradation is moderate; Conf < 0.6-0.8: the message is matched, the slight degradation is judged, and the suspected degradation type is marked; conf <0.6, low confidence match, no significant degradation is determined.
  8. 8. The method for predicting the lifetime of a material degradation of layers of a CIGS photovoltaic panel according to claim 1, wherein in S5, the number of neurons in the fully connected layer is 16 and the activation function is ReLU.
  9. 9. The method for predicting the service life of each layer of material degradation of a CIGS photovoltaic panel according to claim 1, wherein in S5, the number of neurons in the three hidden layers is 128, 64 and 32, respectively, the activation function is tanh, and the drop out is 0.2.
  10. 10. The method for predicting lifetime based on degradation of materials of layers of CIGS photovoltaic panels according to claim 1, wherein in S6, a failure threshold value=0.7 is set, and the predicted failure time point is a time point that is lower than the failure threshold value for the first time, and the remaining lifetime=the current operated lifetime+ (predicted failure time point/4) - (operated season degree/4).

Description

Service life prediction method based on CIGS photovoltaic panel material degradation of each layer Technical Field The invention relates to the technical field of material life prediction, in particular to a life prediction method based on material degradation of each layer of a CIGS photovoltaic panel. Background Copper indium gallium diselenide (CIGS) photovoltaic panels have entered commercial production and are being implemented in a number of application areas for large-scale applications. The CIGS technology has irreplaceable advantages in the fields of building integration, mobile energy and the like by virtue of excellent dim light performance, light weight and flexible characteristics, and has wide future market prospect. However, the commercial time is short (the large-scale mass production is less than 10 years), the long-term outdoor reliability data is lost, the definite life expectancy is lacked, the traditional life assessment depends on an accelerated aging test and an empirical formula, the coupling effect of a complex outdoor environment cannot be simulated, the prediction precision is low, the test period is long, and the market penetration is directly restricted. In the prior art, CIGS photovoltaic panels typically have a lifetime (typically 10-25 years) that is lower than crystalline silicon photovoltaic panels (about 25-30 years). Due to the flexibility and the non-silicon structural characteristics, the aging mechanism of the crystalline silicon photovoltaic panel is obviously different from that of the crystalline silicon photovoltaic panel, and the crystalline silicon photovoltaic panel is influenced by factors such as material characteristics, manufacturing process, use environment, packaging technology and the like, and has different standard core indexes such as service life, annual attenuation rate and the like. In addition, the core internal factor of the shortened service life of the CIGS photovoltaic panel is deterioration of physicochemical properties of a multilayer film (a CIGS absorption layer, a CdS buffer layer and a ZnO window layer) and a packaging structure, so that a service life prediction method based on the deterioration of materials of each layer of the CIGS photovoltaic panel is urgently needed. Disclosure of Invention The invention aims to provide a life prediction method based on the degradation of each layer of material of a CIGS photovoltaic panel, which can quickly locate parameters with the largest influence on the life by analyzing each layer of material through multi-source data fusion and a machine learning model, and can reduce prediction errors, and simultaneously, the model can be continuously and iteratively optimized along with the accumulation of outdoor measured data, so that the prediction precision is further improved. In order to achieve the above purpose, the invention provides a life prediction method based on the degradation of materials of each layer of a CIGS photovoltaic panel, which comprises the following steps: S1, establishing baseline data and a CIGS degradation mode library, wherein physicochemical data comprise CIGS crystallinity, znO light transmittance, interface adhesion, temperature coefficient, CIGS element proportion, cdS layer S content, packaging layer water vapor permeability and environmental humidity; s2, data acquisition, outlier processing, missing value filling, normalization processing and time sequence alignment are carried out, and a time sequence data set is obtained; S3, training a random forest model, outputting importance scores of physicochemical data, and screening physicochemical data characteristics input as a model; S4, encoding the degradation type label into a numerical value characteristic, then splicing the numerical value characteristic with the physicochemical data characteristic as a static characteristic, unifying and standardizing the length, calculating the minimum accumulated distance through a dynamic time warping algorithm, and judging the confidence level and the degradation level to obtain an input characteristic set of the LSTM model; S5, constructing an LSTM model, wherein the LSTM model comprises an input layer, three hidden layers, a full-connection layer and an output layer; s6, life prediction, namely setting an invalidation threshold value, and predicting the invalidation time point to be the time point which is lower than the invalidation threshold value for the first time to obtain the residual life. Preferably, in S1, the CIGS degradation pattern library includes degradation type tags, feature thresholds and time sequence variation rules, wherein the degradation type tags include CIGS element segregation, znO window layer oxidation, encapsulation layer aging and interface lift-off. Preferably, in S1, the non-core power generation regions at the edges and corners of the module are sampled, the cut is sealed with a sealant, the glass substrate is peeled off, the surface encapsulati