Perovskite photovoltaic material performance influence factor analysis method

Abstract

The invention discloses a perovskite photovoltaic material performance influence factor analysis method, and particularly relates to the technical field of structure optimization. The method comprises the following steps: collecting a microstructure image of a perovskite material, extracting characteristics such as grain size, grain boundary density and defect distribution, and constructing a microstructure characteristic vector; constructing an environmental stress evolution curve as external driving input of structure evolution; establishing a stress-structure response model based on microstructure characteristics and environmental stress, and predicting a dynamic process of the microstructure changing with time; further combining the initial performance parameters to construct a performance correlation mapping model, and outputting a performance time-varying predicted value; by calculating a performance deviation function, identifying a key microscopic influence factor and carrying out spatial positioning, outputting a performance sensitive area and a distribution map thereof; according to the method, modeling and visual output of the whole process from environmental stress loading to performance degradation traceability are achieved, and the method has the advantages of being high in prediction precision, fine in analysis granularity and high in applicability.

Description

Technical Field

[0001] This invention relates to the field of structural optimization technology, specifically to a method for analyzing the factors affecting the performance of perovskite photovoltaic materials. Background Technology

[0002] Perovskite photovoltaic materials have become a research hotspot in the field of solar cells in recent years due to their high light absorption coefficient, low fabrication cost, and excellent photoelectric conversion efficiency. However, the performance of perovskite materials is easily affected by a variety of complex factors, especially in non-ideal environments such as humidity, high temperature, or complex atmospheres, where the structural stability of the perovskite layer and the device efficiency may deteriorate significantly. Existing research mostly focuses on the influence of single environmental factors (such as humidity and temperature) on perovskite performance, lacking quantitative simulation and attribution analysis of material performance changes under multi-factor coupling conditions.

[0003] Furthermore, current analytical methods largely rely on static parameter inputs or experimental post-hoc data, which cannot achieve correlation modeling between microscale structural features (such as grain boundary defect distribution and ion migration paths) and macroscopic performance indicators. This results in a lack of specificity in the material design optimization process, especially when faced with extremely small structural perturbations (such as lattice distortion or doped atom displacement), making it difficult to predict their nonlinear amplification effect on performance. Summary of the Invention

[0004] The purpose of this invention is to provide a method for analyzing the factors affecting the performance of perovskite photovoltaic materials, so as to overcome the shortcomings of the prior art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a method for analyzing the factors affecting the performance of perovskite photovoltaic materials, comprising:

[0006] Obtain the basic performance parameter set P of the target perovskite photovoltaic material under different environmental stress conditions, including photoelectric conversion efficiency, open-circuit voltage, fill factor and short-circuit current density;

[0007] Collect microstructure image data of the target material, extract the grain size distribution, grain boundary density and defect distribution matrix, and construct the microstructure feature vector Mf;

[0008] According to the set of external environmental disturbance factors Where T is temperature and H is humidity. Let represent oxygen concentration and UV represent ultraviolet irradiation intensity. Construct an environmental stress evolution curve E(t).

[0009] Based on the microstructure feature vector Mf and the environmental stress curve E(t), a stress-structure response model is constructed to predict the dynamic evolution of microstructures under different environmental stresses.

[0010] The basic performance parameter set P is input into the performance correlation mapping model, and combined with the dynamic evolution process of microstructure, the time-varying performance prediction value set P(t) is output.

[0011] Calculate the performance deviation function The performance degradation trend under the combined effects of environmental stress and microstructure disturbance was analyzed.

[0012] Based on ΔP(t), the performance sensitivity feature set Fs is inversely calculated to identify the key micro-factors that lead to performance degradation and their corresponding structural regions.

[0013] The structural regions are partitioned and sorted by sensitivity, and the key performance influencing factors of perovskite photovoltaic materials and their spatial distribution map are output.

[0014] Preferably, the obtained grain size distribution, grain boundary density, and defect distribution matrices are vectorized in a set order. The grain size distribution is arranged in statistical groups to form the first segment of the vector; the grain boundary density is added as a single value to form the second segment of the vector; the defect distribution matrix is ​​flattened into a one-dimensional sequence to form the third segment of the vector; finally, the three segments are spliced ​​together to construct a microstructure feature vector Mf of a unified dimension.

[0015] Preferably, a stress-structure response model is constructed to predict the dynamic evolution of microstructures under different environmental stresses, including:

[0016] The microstructure feature vector Mf is used as the initial structural state input, and the environmental stress evolution curve E(t) is used as the external driving variable input to construct training sample pairs with consistent input dimensions.

[0017] A long short-term memory neural network was used for modeling to construct a stress-structure response model with E(t) as the time series input and Mf(t) as the prediction output. The model contains three memory unit layers and one fully connected regression layer, and the hyperbolic tangent function was selected as the activation function.

[0018] Supervised training of the model is performed using calibrated experimental data;

[0019] The trained model is used to predict the evolution of structural features under any given E(t) condition, and the microstructure feature vector Mf(t) that changes over time is obtained.

[0020] Preferably, the output performance time-varying prediction value set P(t) includes:

[0021] The basic performance parameter set P is used as the initial performance state input, and the microstructure feature vector Mf(t) of the corresponding time series is used as the structure evolution input to construct a joint input vector set for performance prediction modeling.

[0022] A performance correlation mapping model is constructed using a dual-input regression neural network. The model contains two parallel input channels that receive P and Mf(t) respectively. The intermediate layer integrates structural information and performance information and outputs a set of performance prediction values ​​P(t) that change over time.

[0023] The performance correlation mapping model was trained using experimental data containing multiple sets of (P, Mf(t), P(t)), with the mean square error between the predicted value and the measured performance data used as the loss function during the training process.

[0024] The trained model is used in the inference phase. Given any P and Mf(t) as input, it outputs a set of time-varying performance predictions, P(t).

[0025] Preferably, the analysis of performance degradation trends under the combined effects of environmental stress and microstructure disturbance includes:

[0026] Eigenvalues ​​of ΔP(t) are decomposed to extract degradation characteristic indicators including peak deviation, cumulative deviation area and maximum rate of change, and then classified and analyzed according to performance categories.

[0027] By combining the environmental stress curve E(t) and the microstructure feature evolution sequence Mf(t), a performance deviation response model is constructed using a multivariate stepwise regression method to identify the dominant factor combination affecting the change of ΔP(t).

[0028] Based on the deviation response models for different performance parameters, a performance degradation trend diagram evolving over time is plotted, and the coupling effect of environmental stress and microstructure disturbance is evaluated based on the parameter sensitivity output by the performance deviation response model.

[0029] Preferably, identifying key microscopic factors that lead to performance degradation and their corresponding structural regions includes:

[0030] The performance deviation function ΔP(t) and the microstructure feature evolution sequence Mf(t) are time-aligned to construct a performance deviation-structural feature correlation matrix, which is used to characterize the response relationship of each microstructure feature to the performance deviation over time.

[0031] Based on the performance deviation-structural feature correlation matrix, the sensitivity index of each microstructural feature to ΔP(t) is calculated, and the performance sensitivity feature group Fs is formed according to the numerical value.

[0032] Map the performance-sensitive feature set Fs back to the corresponding microstructure feature source to identify the key micro factors that contribute the most to performance degradation.

[0033] By combining the defect distribution matrix and grain spatial location information, the specific structural regions corresponding to the key microscopic factors are determined.

[0034] Preferably, the key performance influencing factors and their spatial distribution map of the perovskite photovoltaic material are output, including:

[0035] The microstructure image is divided into multiple structural sub-regions according to a grid of equal size, and a mapping relationship is established between each sub-region and its corresponding microstructure feature vector.

[0036] The identified performance sensitivity feature groups are associated with structural sub-regions, and the clustering degree and variation of key micro factors in each sub-region are calculated as the sensitivity index of the region.

[0037] All structural sub-regions are sorted according to their sensitivity index values, and sensitivity thresholds of high, medium, and low are set.

[0038] The sensitivity levels and corresponding key influencing factors of each structural sub-region are visualized and output in the form of color heat maps, forming a spatial distribution map of the performance influencing factors of perovskite photovoltaic materials.

[0039] The technical effects and advantages provided by the present invention in the above technical solution are as follows:

[0040] 1. This invention proposes a method for analyzing the performance influencing factors of perovskite photovoltaic materials based on co-modeling of microstructure evolution and environmental stress. This method enables digital analysis of the entire process, from structural image acquisition and environmental disturbance loading to dynamic microstructure modeling and performance response prediction. Compared to traditional methods relying on static experimental data and single-factor analysis, this invention introduces multi-source time series modeling and machine learning algorithms to achieve time-varying prediction of performance deviations and quantitative identification of key factors, significantly improving the accuracy of performance evolution prediction under service conditions for photovoltaic materials.

[0041] 2. This invention, by constructing a coupled response model of microstructure features and performance deviations, and combining it with spatial distribution mapping technology, achieves for the first time the spatial localization and sensitivity visualization output of key performance influencing factors, forming a refined control basis that can be used for local structural optimization and stability improvement. The overall scheme has advantages such as high modeling accuracy, adaptability to complex environments, and strong interpretability, and is suitable for stability research and structural reliability design of perovskite and other sensitive photovoltaic materials. Attached Figure Description

[0042] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0043] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0044] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0045] For examples, please refer to Figure 1 As shown in this embodiment, the method for analyzing the factors affecting the performance of perovskite photovoltaic materials includes:

[0046] Obtain the basic performance parameter set P of the target perovskite photovoltaic material under different environmental stress conditions, including photoelectric conversion efficiency, open-circuit voltage, fill factor and short-circuit current density.

[0047] In a preferred embodiment of the present invention, perovskite photovoltaic devices are sequentially placed in a pre-defined temperature and humidity controlled cavity. The cavity has the ability to independently regulate temperature and humidity, with a temperature control range of -10°C to 85°C and a humidity control range of 10% to 95% relative humidity. By setting multiple different combined stress conditions, such as temperatures of 25°C, 45°C, and 65°C, and humidity levels of 30%, 60%, and 90%, environmental stress is applied to the devices group by group to ensure coverage of actual usage scenarios under typical service conditions.

[0048] After each set of temperature and humidity stress loading has stabilized for more than 10 minutes, the perovskite photovoltaic device is measured in real time using a solar simulator and photovoltaic performance testing platform. The photoelectric conversion efficiency, open-circuit voltage, fill factor, and short-circuit current density under the current conditions are recorded. The solar simulator should meet the AM1.5G standard with an illuminance of 1000 watts per square meter to ensure the accuracy and comparability of the test data. The test results under each set of stress conditions are recorded and numbered to form an initial data matrix.

[0049] The collected performance parameter data were normalized according to the time series using a linear transformation, converting all performance parameter values ​​to the [0,1] interval. Specifically, for each performance index, a linear proportional relationship was established using its maximum and minimum values, projecting the actual measured values ​​into the normalized interval. The normalized data was then reorganized according to the environmental stress combination number, forming a four-dimensional performance parameter dataset with temperature and humidity as input dimensions and photoelectric conversion efficiency, open-circuit voltage, fill factor, and short-circuit current density as output dimensions.

[0050] Based on the constructed multidimensional performance parameter dataset, a regression algorithm is used to build a quantitative mapping model between environmental stress and basic performance parameters. The regression algorithm employs a multivariate nonlinear least squares fitting method, using temperature and humidity as independent variables and performance parameters as dependent variables to establish a function model. The function form adopts a quadratic polynomial fitting model, establishing a response function of the following form for each performance index: Performance parameter Y = a × temperature squared + b × humidity squared + c × temperature × humidity + d × temperature + e × humidity + f, where a to f are fitting parameters. The parameter values ​​are determined by minimizing the sum of squared errors between the actual observed values ​​and the model predicted values, and then... The accuracy of the fit of the indicator should be evaluated, and it should be no less than 0.95.

[0051] Finally, a quantitative mapping function between the basic performance parameter set P and the external environmental stress is obtained, which is used as input data for subsequent microstructure dynamic evolution modeling and performance evolution trend analysis, realizing causal tracing and prediction of performance evolution path.

[0052] Microstructure image data of the target material are acquired, and the grain size distribution, grain boundary density and defect distribution matrix are extracted to construct the microstructure feature vector Mf.

[0053] Scanning electron microscopy was used to perform multi-magnification imaging of perovskite photovoltaic material samples. The imaging magnification was set to 1000x, 5000x, and 10000x sequentially to ensure the capture of multi-scale information on the sample surface, from macroscopic morphology to microscopic grain structure.

[0054] During the imaging process, a metal coating is applied to the sample surface to enhance conductivity and reduce electron beam scattering interference. The acquired image data is then subjected to grayscale histogram equalization in its original format to improve contrast, and Gaussian filtering is used to suppress noise. The filter kernel size is set to 5×5 pixels to eliminate random noise during image acquisition while maintaining the clarity of grain boundary edges.

[0055] Based on the image preprocessing, the image is binarized, and the adaptive threshold is dynamically adjusted according to the local gray average value and standard deviation to adapt to the gray difference between grains.

[0056] Adjacent grains are separated using a connected component labeling method. Each grain is statistically analyzed using the diameter of its smallest circumscribed circle as its equivalent diameter, constructing a grain size distribution dataset. The size distributions are grouped at 5-nanometer intervals to form a continuous probability density function, used to characterize the granular features of the microstructure.

[0057] Based on the boundary image of the segmented grains, a pixel-level path length statistical method is used to calculate the total length of the grain boundaries within a unit area (e.g., per 100 square micrometers). This length value is then divided by the unit area to obtain the grain boundary density, in micrometers per square micrometer.

[0058] For defect identification, the location and morphological features of gray-level abrupt change regions in the image are analyzed, and point-like and line-like defects are identified using the edge gradient method and Laplacian transform. The identified defects are then located using coordinates to construct a two-dimensional spatial matrix, where the matrix element values ​​indicate whether a defect exists at that location (1 for presence, 0 for absence). This matrix is ​​the defect distribution matrix, used to describe the density, spatial clustering, and distribution patterns of defects.

[0059] The obtained grain size distribution, grain boundary density, and defect distribution matrices are vectorized according to a set order. The grain size distribution is arranged in statistical groups to form the first vector segment; the grain boundary density is added as a single value to form the second vector segment; the defect distribution matrix is ​​flattened into a one-dimensional sequence to form the third vector segment.

[0060] Finally, the three feature segments are concatenated to construct a microstructure feature vector Mf with a unified dimension. This vector contains all the information features of the material structure under a given observation area, serving as the input parameter for the subsequent microstructure evolution prediction model under environmental stress. The length of the Mf vector is determined by the dimension and distribution accuracy of the defect matrix, and is maintained within a stable dimensional range to meet the modeling consistency requirements.

[0061] According to the set of external environmental disturbance factors Where T is temperature and H is humidity. Let represent oxygen concentration and UV represent ultraviolet irradiation intensity. Construct an environmental stress evolution curve E(t).

[0062] First, four perturbation factors were set: the range and rate of change of temperature, humidity, oxygen concentration, and ultraviolet radiation intensity. The temperature range was set from 20°C to 80°C, the humidity range was set from 20% to 90% relative humidity, the oxygen concentration was set between 0% and 21% volume fraction, and the ultraviolet radiation intensity covered 0 to 1 solar constant (i.e., 0 to 1000 watts per square meter).

[0063] For each perturbation factor, based on the expected experimental duration Tmax, a template for its variation function with respect to time t is constructed, and preliminary parameterization is performed using a linear function, a piecewise function, or a sine function. For example, for the temperature factor T(t), it can be set as a linear heating process, i.e. Where T0 is the initial temperature and α is the heating rate. All perturbation functions have a unified time step, and a corresponding time control table is generated in minutes to specify the perturbation parameter values ​​to be applied every minute.

[0064] A programmable environmental stress loading platform was used to control disturbances in perovskite photovoltaic devices. The platform is equipped with a high-precision temperature and humidity control unit, an oxygen regulation module, and an ultraviolet light irradiation component, each corresponding to one of the four disturbance factors.

[0065] According to the time control schedule, the platform updates the set values ​​of each factor every minute and outputs them stably, ensuring the consistency and timeliness of the applied perturbations. Simultaneously, the equipped sensor array collects the actual effect values ​​of each perturbation factor in real time and records them in the database, forming a raw data sequence containing timestamps and measured perturbation values. The sampling frequency is uniformly set to once per minute to ensure that the data resolution meets the requirements of subsequent model processing.

[0066] Numerical smoothing was performed on the four sets of disturbance data collected to eliminate abnormal fluctuations caused by factors such as equipment response delay and sensor error. A weighted moving average algorithm was used to update each disturbance data point. The calculation method is as follows: the weight of the current point is equal to the weighted average of the data in the previous 3 minutes, where the weight of the most recent minute is 0.5, the weights of the previous two minutes are 0.3 and 0.2 respectively, and the sum of the weights is 1.

[0067] This algorithm can preserve the trend of disturbance changes while suppressing local oscillations, and the processed data curve is smooth and continuous, which meets the differentiability requirement and provides a stable foundation for subsequent mathematical modeling.

[0068] The four smooth curves of temperature, humidity, oxygen concentration, and ultraviolet irradiation intensity are uniformly mapped to the time dimension t, forming a four-dimensional vector function. Each component is a time function of the measured perturbation factor. This four-dimensional function is defined over the time interval from 0 to Tmax, with a time step of 1 minute.

[0069] The E(t) is used to simulate the external multi-factor composite disturbance environment of the target material. It can be used as the driving input parameter for microstructure evolution prediction model and performance degradation modeling, so as to realize the controllable analysis of the structural behavior and performance change trend of photovoltaic materials under real service conditions.

[0070] Based on the microstructure feature vector Mf and the environmental stress curve E(t), a stress-structure response model is constructed to predict the dynamic evolution of microstructures under different environmental stresses.

[0071] First, the constructed microstructure feature vector Mf is used as the initial structural state input. The vector dimension is a fixed dimension d, for example, d=128, representing the joint feature expression of grain size distribution, grain boundary density, and defect distribution matrix. Simultaneously, the obtained environmental stress evolution curve E(t) is used as the external driving input. E(t) is a time series function with four variables: the joint evolution function of temperature, humidity, oxygen concentration, and ultraviolet irradiation intensity.

[0072] By organizing the real observation data of multiple experimental samples under different perturbation conditions, a training sample pair in the form of (E(t), Mf(t)) is constructed, where Mf(t) is the sequence of microstructure feature vectors corresponding to each time step, ensuring that the input and output dimensions are consistent, which facilitates the training of the neural network model.

[0073] A long short-term memory neural network is used for modeling. The neural network structure includes three consecutively stacked long short-term memory unit layers. The hidden state dimension of each unit layer is set to 64 to capture the short-term fluctuations and long-term trends of environmental stress on the evolution of microstructure. A fully connected regression layer is connected at the end to map the hidden state output to an output vector Mf(t) with the same dimension as the microstructure feature vector.

[0074] The network employs a hyperbolic tangent function as its activation function, enabling it to maintain a nonlinear response capability when handling positive and negative perturbations. The model input is an environmental stress sequence E(t) with a time length of T, and the output is a predicted microstructure feature vector Mf(t) in the same time dimension, used to simulate the continuous evolution of the structure in a perturbation field.

[0075] The model was trained using supervised learning with microstructure evolution data collected under known perturbation conditions. The training data included multiple time series pairs of different E(t) and their corresponding Mf(t) to ensure the model's generalization ability. During training, the loss function was defined as the mean squared error between the predicted and observed values, and the backpropagation algorithm was used to iteratively update the weights and bias parameters in the network.

[0076] To ensure training stability, the Adam optimizer was used with an initial learning rate of 0.001 and 500 training epochs. Each training epoch contained a batch of samples with a batch size of 32. The training termination condition was that the validation set loss did not decrease for 10 consecutive epochs.

[0077] The trained stress-structure response model is deployed in the prediction pipeline. The environmental stress evolution curve E(t) at any given time dimension is input, and the model automatically generates a sequence of microstructure feature vectors Mf(t) that evolves over time. The output includes changes in grain size distribution, grain boundary density evolution trends, and defect distribution migration states, serving as key inputs for subsequent material property evolution prediction models.

[0078] This step transforms the process of microstructural change from qualitative observation to quantitative prediction, enabling digital modeling of the structural response mechanism.

[0079] The basic performance parameter set P is input into the performance correlation mapping model, and combined with the dynamic evolution process of microstructure, the time-varying performance prediction value set P(t) is output.

[0080] The obtained basic performance parameter set P is used as the initial performance state input. The parameter set P includes photoelectric conversion efficiency, open-circuit voltage, fill factor and short-circuit current density, forming a fixed-dimensional vector, denoted as P0.

[0081] Simultaneously, the time-varying sequence Mf(t) of the acquired microstructure feature vectors is used as the structural state input, where each time step t corresponds to a vector Mf(t). P0 is copied and extended to the same time length as Mf(t) and concatenated with Mf(t) of each frame to form a joint input vector group [P0, Mf(t)], which serves as the input sample for the performance prediction model, preserving the coupling relationship between performance and structure.

[0082] A dual-input regression neural network is used for modeling. The network structure consists of two parallel input channels: one channel receives the initial performance parameter P0, and the other channel receives the time series structural features Mf(t). The two channels are respectively processed through two fully connected layers for feature transformation and abstraction extraction. The intermediate output features are then fused in the third layer by concatenation before being fed into a nonlinear transformation layer. The activation function is a modified linear unit function.

[0083] The model output is a performance parameter prediction sequence P(t) of the same length as the input time step, with the same dimension as the initial P0, representing the performance evolution path of the photovoltaic device within the prediction time window.

[0084] The performance correlation mapping model is trained under supervision using multiple sample data. The training samples consist of three parts: the initial performance parameter P0, the corresponding structural evolution vector sequence Mf(t), and the corresponding sequence of real performance observations P(t).

[0085] The loss function is defined as the mean squared error between the predicted performance sequence and the actual performance sequence, denoted as . Where N is the number of time steps, P(t) represents the predicted value, and P(t) represents the measured value. During training, the Adam optimization algorithm is used for gradient descent optimization, with a learning rate of 0.0005, 300 training epochs, and a batch size of 64 for training iterations until the validation set error converges.

[0086] The trained performance correlation mapping model is used in the inference phase. Given the initial performance parameter set P0 and the structural dynamic sequence Mf(t) as input, the model automatically outputs the time-varying performance prediction set P(t), which includes the photoelectric conversion efficiency, open-circuit voltage, fill factor, and short-circuit current density at different time points.

[0087] The P(t) is used to quantify the specific impact of structural evolution on performance degradation, enabling lifetime prediction and reliability assessment of devices under complex service conditions.

[0088] The performance deviation function ΔP(t) = P(t) – P is calculated to analyze the performance degradation trend under the combined effects of environmental stress and microstructure disturbance.

[0089] The predicted performance time series P(t) is compared with the basic performance parameter set P, which corresponds to four performance dimensions: photoelectric conversion efficiency, open-circuit voltage, fill factor, and short-circuit current density.

[0090] At each time step t, the performance deviation value ΔP(t) is calculated, which is defined as follows: This is the difference between the current performance value and the initial performance value, reflecting the degree of performance degradation relative to the initial state. The result of this function is a time series, with units consistent with the original performance parameters.

[0091] Eigenvalue decomposition is performed on the performance deviation function ΔP(t) to extract the following three key degradation indicators:

[0092] Peak deviation: The minimum value of ΔP(t) over the entire time period, representing the maximum degree of performance degradation;

[0093] Cumulative deviation area: The integral of ΔP(t) over time, reflecting the total performance loss;

[0094] Maximum rate of change: Calculate the first-order difference of ΔP(t) between each adjacent time step, and take the maximum absolute value to represent the most severe rate of performance degradation.

[0095] The above indicators are extracted and categorized according to performance type, and a degradation index matrix for each performance dimension (such as open-circuit voltage, photoelectric conversion efficiency, etc.) is established for the next step of regression modeling.

[0096] The obtained environmental stress evolution curve E(t) and the obtained microstructure feature vector sequence Mf(t) are jointly expanded, and correlation analysis is performed with the corresponding ΔP(t).

[0097] A multivariate performance deviation response model is constructed using the stepwise regression method. The specific process is as follows:

[0098] The initial regression model includes all perturbation factors and structural feature dimensions as independent variables, and ΔP(t) as the dependent variable;

[0099] Variables were screened using the Akaike information criterion, gradually eliminating variables that did not contribute significantly to the model's explanatory power, and retaining factors with a significance level of less than 0.05.

[0100] The variables retained in the final output regression equation are the combinations of dominant factors that significantly affect performance deviation, corresponding to specific environmental stress parameters (such as temperature and ultraviolet irradiation intensity) or microstructure change characteristics (such as defect density and grain size change rate).

[0101] The completed performance deviation response model was used to fit the trend of ΔP(t) over time, and the degradation curve of each performance parameter was plotted. The horizontal axis of the graph is time t, and the vertical axis is the value of ΔP(t).

[0102] Further, the standardized regression coefficients of the regression model are used to quantify the sensitivity of each input variable; the larger the absolute value of the coefficient, the more significant the impact of that factor on performance changes. Specifically, this includes:

[0103] In stepwise regression models, to make input variables with different dimensions comparable, all independent variables are first standardized. This involves subtracting the mean from the original value of each variable and dividing by the standard deviation, thus converting it into a dimensionless variable with a mean of zero and a standard deviation of one.

[0104] For the standardized multivariate regression equation, each regression coefficient represents the unit standard deviation of the influence of the standardized variable on the performance deviation function ΔP(t). The absolute value of the standardized regression coefficient of each variable is used as its sensitivity index, denoted as Si. The larger Si is, the higher the degree to which the variable dominates the performance change.

[0105] By sorting all variables in descending order of Si, the sensitivity distribution of the performance degradation influencing factors can be obtained.

[0106] By comparing the sensitivity of structural factors and environmental disturbance factors, the coupling effect strength between the two is assessed, and three influence level labels—"dominant single factor effect," "weak coupling," and "strong coupling"—are output to provide a decision-making basis for subsequent material stability optimization and structural design. Specifically, this includes:

[0107] Input variables are divided into two main categories:

[0108] Structural variables: derived from the microstructure feature vector Mf(t), including grain size distribution, grain boundary density, defect distribution matrix, etc.

[0109] Environmental variables: derived from the environmental stress evolution curve E(t), including temperature, humidity, oxygen concentration, ultraviolet radiation intensity, etc.

[0110] According to the above classification results, sum the sensitivity indicators Si of the structural variables and environmental variables in the regression model respectively to obtain the total structural sensitivity Ss and the total environmental sensitivity Se. Define the coupling effect strength index R as the ratio of the two sensitivities, specifically as follows: R = Ss / Se; according to the R value, divide the dominant mode of performance change: when R≥2 or R≤0.5: it is considered that the performance deviation is mainly driven by a single factor, and the output label is "dominant single factor effect"; when 0.5 < R < 1.5 and there are at least two highly sensitive variables (Si > 0.3) in any category of factors: it is considered that there is a joint influence between the structural and environmental factors, and the output label is "strong coupling"; when 0.5 < R < 1.5 and there is a dominant variable in only one category of factors: the output label is "weak coupling".

[0111] Finally, output the sensitivity ranking results, the coupling strength index R, and the classification label in the form of structured data. This result can be used to guide material formulation selection, device packaging structure design, and stability enhancement path decision-making. For example, in the "strong coupling" mode, the structural stability and environmental shielding ability should be optimized simultaneously; in the "dominant single factor effect" mode, the corresponding link of the dominant factor can be preferentially intervened to improve the service reliability of the material.

[0112] Based on ΔP(t), inversely deduce the performance sensitivity feature group Fs, and identify the key microscopic factors leading to performance degradation and their corresponding structural regions.

[0113] Perform time alignment processing on the obtained performance deviation function ΔP(t) and the obtained microstructure feature vector sequence Mf(t).

[0114] Set a unified time step Δt, for example, 1 minute, and construct a sample pair of ΔP(t) and Mf(t) at the corresponding moment t. If the total time length is T, then T groups of samples are formed, constituting an association matrix R, with the matrix dimension of T×(n + 1), where n is the microstructure feature dimension, and +1 is the corresponding performance deviation value ΔP(t). This matrix is used to characterize the co-action relationship of each structural feature on the performance deviation at different time points.

[0115] Based on the constructed association matrix R, adopt the method of variable-by-variable linear regression. Respectively take each microstructure feature column as the independent variable and ΔP(t) as the dependent variable, and calculate its linear regression coefficient βi and the determination coefficient .

[0116] To simultaneously consider the regression strength and time correlation, define the comprehensive sensitivity index , and calculate its Si value for each microstructure feature, representing its influence ability on the performance deviation function.

[0117] Sort all Si values ​​from largest to smallest, and select the top few features (e.g., the top 5) to form the performance sensitivity feature group Fs, which represents the set of microstructure features that have the most dominant effect on performance degradation under the current environment and structural evolution conditions.

[0118] Map each feature in Fs back to its source dimension in Mf(t) to identify its physical meaning, such as the rate of change of grain size, the magnitude of increase in defect point density, and the magnitude of grain boundary migration.

[0119] Based on the ranking of the sensitivity index Si, the key micro-factors that contribute the most to the performance degradation are identified, and their evolution curves over time are established to further track their action paths and key time periods.

[0120] By combining the constructed defect distribution matrix with the grain spatial coordinate information, the identified key microscopic factors are spatially reflected in the image data.

[0121] If the key factor is a defect density-related feature, its concentrated region is extracted by the position coordinates in the defect matrix; if it is a grain size or grain boundary feature, the grain region with the most significant feature changes is located by segmenting the label map in the image.

[0122] The final output includes the structural region number, location coordinates, and corresponding degradation index values, enabling spatial mapping of performance-sensitive factors and providing a basis for subsequent regional repair, local material modification, or structural optimization.

[0123] The structural regions are partitioned and sorted by sensitivity, and the key performance influencing factors of perovskite photovoltaic materials and their spatial distribution map are output.

[0124] First, the acquired raw microstructure image data is spatially divided according to an equal-sized grid. The division method adopts a fixed-size sliding window method, with the window size set to 50 pixels × 50 pixels and the overlap step size to 0 pixels, forming several non-overlapping structural sub-regions, each sub-region corresponding to an image segment.

[0125] In the image segmentation and feature extraction stage, the microstructure feature vector Mf(t) has been extracted from the entire image. The spatial coordinates of each sub-region in the image are matched with the position index in the global feature vector to establish a one-to-one correspondence between "structural sub-region - local feature vector" for subsequent local localization of sensitive factors.

[0126] Based on the performance-sensitive feature group Fs identified in the previous steps, the feature values ​​corresponding to each structural sub-region are extracted, and the aggregation degree and variation magnitude of key micro-factors in the region are calculated.

[0127] Here, the clustering degree is defined as the proportion of pixels in the sub-region whose feature value is higher than the global average, and the change magnitude is defined as the ratio of the maximum change of the feature value in the sub-region throughout the entire time series to the initial value.

[0128] The two are weighted and averaged, with weight ratios set to 0.6 and 0.4, respectively, to reflect anomaly density and dynamic evolution capability. The resulting comprehensive index is defined as the sensitivity score Sr of the sub-region, with a value range between 0 and 1, representing the degree of contribution of the region to the overall performance degradation.

[0129] Sort the sensitivity scores Sr of all sub-regions from highest to lowest value, and set the grading threshold according to the distribution of the score values.

[0130] The specific settings are as follows: Sr≥0.7 is marked as a highly sensitive region, 0.4≤Sr<0.7 is marked as a medium sensitive region, and Sr<0.4 is marked as a low sensitive region.

[0131] After the classification is completed, a sensitivity level label is assigned to each sub-region, and a corresponding spatial index table is established to facilitate subsequent visualization annotation and region identification.

[0132] The structural sub-regions labeled with sensitivity levels are remapped to the original image space and visualized using a color heatmap.

[0133] The color coding scheme is as follows: red represents high-sensitivity areas, orange represents medium-sensitivity areas, and blue represents low-sensitivity areas. In the heatmap overlay, the names of the dominant key micro-factors in each high-sensitivity area are labeled (e.g., "increased defect density" or "enhanced grain boundary migration").

[0134] The final output is a spatial distribution map of the performance influencing factors of perovskite photovoltaic materials, which can be used to guide the implementation of optimization strategies such as local structural regulation, defect repair, and local material reconstruction, and provide a visual basis for improving the long-term stability of devices.

[0135] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A method for analyzing the factors affecting the performance of perovskite photovoltaic materials, characterized in that: include: Obtain the basic performance parameter set P of the target perovskite photovoltaic material under different environmental stress conditions, including photoelectric conversion efficiency, open-circuit voltage, fill factor and short-circuit current density; Collect microstructure image data of the target material, extract the grain size distribution, grain boundary density and defect distribution matrix, and construct the microstructure feature vector Mf; According to the set of external environmental disturbance factors Where T is temperature and H is humidity. Let represent oxygen concentration and UV represent ultraviolet irradiation intensity. Construct an environmental stress evolution curve E(t). Based on the microstructure feature vector Mf and the environmental stress curve E(t), a stress-structure response model is constructed to predict the dynamic evolution of microstructures under different environmental stresses. The basic performance parameter set P is input into the performance correlation mapping model, and combined with the dynamic evolution process of microstructure, the time-varying performance prediction value set P(t) is output. Calculate the performance deviation function The performance degradation trend under the combined effects of environmental stress and microstructure disturbance was analyzed. Based on ΔP(t), the performance sensitivity feature set Fs is inversely calculated to identify the key micro-factors that lead to performance degradation and their corresponding structural regions. The structural regions are partitioned and sorted by sensitivity, and the key performance influencing factors of perovskite photovoltaic materials and their spatial distribution map are output.

2. The method for analyzing the influencing factors of perovskite photovoltaic material performance according to claim 1, characterized in that: The obtained grain size distribution, grain boundary density, and defect distribution matrices are vectorized according to a set order. The grain size distribution is arranged in statistical groups to form the first segment of the vector; the grain boundary density is added as a single value to form the second segment of the vector; the defect distribution matrix is ​​flattened into a one-dimensional sequence to form the third segment of the vector; finally, the three segments are spliced ​​together to construct a microstructure feature vector Mf of a unified dimension.

3. The method for analyzing the influencing factors of perovskite photovoltaic material performance according to claim 1, characterized in that: A stress-structure response model is constructed to predict the dynamic evolution of microstructures under different environmental stresses, including: The microstructure feature vector Mf is used as the initial structural state input, and the environmental stress evolution curve E(t) is used as the external driving variable input to construct training sample pairs with consistent input dimensions. A long short-term memory neural network was used for modeling to construct a stress-structure response model with E(t) as the time series input and Mf(t) as the prediction output. The model contains three memory unit layers and one fully connected regression layer, and the hyperbolic tangent function was selected as the activation function. Supervised training of the model is performed using calibrated experimental data; The trained model is used to predict the evolution of structural features under any given E(t) condition, and the microstructure feature vector Mf(t) that changes over time is obtained.

4. The method for analyzing the influencing factors of perovskite photovoltaic material performance according to claim 1, characterized in that: The output performance time-varying prediction value set P(t) includes: The basic performance parameter set P is used as the initial performance state input, and the microstructure feature vector Mf(t) of the corresponding time series is used as the structural evolution input to construct a joint input vector set for performance prediction modeling. A performance correlation mapping model is constructed using a dual-input regression neural network. The model contains two parallel input channels that receive P and Mf(t) respectively. The intermediate layer integrates structural information and performance information and outputs a set of performance prediction values ​​P(t) that change over time. The performance correlation mapping model was trained using experimental data containing multiple sets of (P, Mf(t), P(t)), with the mean square error between the predicted value and the measured performance data used as the loss function during the training process. The trained model is used in the inference phase. Given any P and Mf(t) as input, it outputs a set of time-varying performance predictions, P(t).

5. The method for analyzing the influencing factors of perovskite photovoltaic material performance according to claim 1, characterized in that: The analysis of performance degradation trends under the combined effects of environmental stress and microstructure disturbances includes: Eigenvalues ​​of ΔP(t) are decomposed to extract degradation characteristic indicators, including peak deviation, cumulative deviation area and maximum rate of change, and then classified and analyzed according to performance categories. By combining the environmental stress curve E(t) and the microstructure feature evolution sequence Mf(t), a performance deviation response model is constructed using a multivariate stepwise regression method to identify the dominant factor combination affecting the change of ΔP(t). Based on the deviation response models for different performance parameters, a performance degradation trend diagram evolving over time is plotted, and the coupling effect of environmental stress and microstructure disturbance is evaluated based on the parameter sensitivity output by the performance deviation response model.

6. The method for analyzing the influencing factors of perovskite photovoltaic material performance according to claim 1, characterized in that: Identify the key microscopic factors that lead to performance degradation and their corresponding structural regions, including: The performance deviation function ΔP(t) and the microstructure feature evolution sequence Mf(t) are time-aligned to construct a performance deviation-structural feature correlation matrix, which is used to characterize the response relationship of each microstructure feature to the performance deviation over time. Based on the performance deviation-structural feature correlation matrix, the sensitivity index of each microstructural feature to ΔP(t) is calculated, and the performance sensitivity feature group Fs is formed according to the numerical value. Map the performance-sensitive feature set Fs back to the corresponding microstructure feature source to identify the key micro factors that contribute the most to performance degradation. By combining the defect distribution matrix and grain spatial location information, the specific structural regions corresponding to the key microscopic factors are determined.

7. The method for analyzing the influencing factors of perovskite photovoltaic material performance according to claim 1, characterized in that: The key performance influencing factors of perovskite photovoltaic materials and their spatial distribution diagrams are output, including: The microstructure image is divided into multiple structural sub-regions according to a grid of equal size, and a mapping relationship is established between each sub-region and its corresponding microstructure feature vector. The identified performance sensitivity feature groups are associated with structural sub-regions, and the clustering degree and variation of key micro factors in each sub-region are calculated as the sensitivity index of the region. All structural sub-regions are sorted according to their sensitivity index values, and sensitivity thresholds of high, medium, and low are set. The sensitivity levels and corresponding key influencing factors of each structural sub-region are visualized and output in the form of color heat maps, forming a spatial distribution map of the performance influencing factors of perovskite photovoltaic materials.

Images