Electrical performance prediction and influence factor analysis method and system based on machine learning

Abstract

The invention provides an electrical performance prediction and influence factor analysis method and system based on machine learning, and relates to the technical field of photovoltaic cells, and the method comprises the steps: obtaining technological parameters in a photovoltaic cell production process and electrical performance indexes of corresponding production batches; extracting process statistical characteristics of the process parameters; according to the production batch number and the time window identifier, matching the process statistical characteristics with the electrical performance indexes through an automatic alignment mechanism to form a mapping relation between the process statistical characteristics and the electrical performance indexes; on the basis of the mapping relation, an enhanced process feature matrix used for representing the influence of the process on the electrical performance is constructed through a multi-modal feature extension mechanism, and an electrical performance prediction result is generated through a layered weighted regression model; analyzing the marginal influence of each process parameter on an electrical performance prediction result by using an SHAP value theory, and generating an influence matrix; and according to the influence matrix, adjusting process parameters in the production process of the photovoltaic cell so as to optimize the electrical performance of the photovoltaic cell.

Description

Technical Field

[0001] This invention relates to the field of photovoltaic cell technology, and in particular to a method and system for predicting electrical performance and analyzing its influencing factors based on machine learning. Background Technology

[0002] As a crucial carrier for green energy conversion, photovoltaic (PV) cells' core performance indicators, such as photoelectric conversion efficiency, fill factor, open-circuit voltage, and short-circuit current, directly determine the energy output capacity and commercial value of PV modules. In the current context of intelligent manufacturing, PV cell production lines are gradually moving towards a highly automated, flexible, and data-driven model. Real-time collection and analysis of process data to predict the final performance of cells has become a key link in supporting stable production quality and improved yield. Especially in the promotion and application of new cell structures such as PERC, TOPCon, and HJT, complex texturing, diffusion, and deposition processes place higher demands on process control precision. Therefore, establishing a high-precision performance prediction system for multi-dimensional process flows has become a core technological task for the intelligent development of PV manufacturing.

[0003] Currently, the prediction of photovoltaic cell electrical performance mainly relies on two technical approaches: physical modeling and empirical regression methods. Physical modeling focuses on prediction through the derivation of microscopic mechanisms such as crystal structure, carrier transport, and band structure, and has strong theoretical rigor. Empirical regression methods typically employ a few features and linear relationship assumptions to establish models such as polynomial fitting, linear regression, or traditional decision tree models. Empirical regression methods are simple to implement and run quickly.

[0004] However, while existing methods possess theoretical rigor, the interactions between process parameters and boundary conditions in the complex photovoltaic cell production process are extremely intricate. Traditional physical models struggle to accurately model these complex interactions, resulting in insufficient prediction accuracy. Furthermore, empirical regression methods also struggle to capture these nonlinear relationships when faced with high-dimensional time-series inputs, dynamic gas concentration disturbances, or complex process couplings, thus failing to meet the prediction accuracy requirements of actual production. Summary of the Invention

[0005] In view of the shortcomings of the prior art, the purpose of this invention is to provide a method for predicting electrical performance and analyzing its influencing factors based on machine learning. This method addresses the problem that while existing methods possess theoretical rigor, the interactions between process parameters and boundary conditions in the complex photovoltaic cell production process are extremely complex, making it difficult for traditional physical models to accurately model these intricate interactions, resulting in insufficient prediction accuracy. Furthermore, empirical regression methods struggle to capture these nonlinear relationships when faced with high-dimensional time-series inputs, dynamic gas concentration disturbances, or complex process couplings, thus failing to meet the prediction accuracy requirements of actual production.

[0006] A first aspect of this invention proposes a method for predicting electrical performance and analyzing its influencing factors based on machine learning, comprising:

[0007] S1: Obtain the process parameters of the photovoltaic cell during the production process and the electrical performance indicators of the corresponding production batch;

[0008] S2: Extract the process statistical features of the process parameters;

[0009] S3: Based on the production batch number and time window identifier, the process statistical features are matched with the electrical performance indicators through an automatic alignment mechanism to form a mapping relationship between the process statistical features and the electrical performance indicators;

[0010] S4: Based on the mapping relationship, an enhanced process feature matrix is ​​constructed to characterize the impact of the process on electrical performance through a multimodal feature extension mechanism;

[0011] S5: Based on the enhanced process feature matrix, generate electrical performance prediction results through a hierarchical weighted regression model;

[0012] S6: Analyze the marginal impact of each process parameter on the electrical performance prediction results using SHAP value theory, and generate an impact matrix;

[0013] S7: Based on the influence matrix, adjust the process parameters in the photovoltaic cell production process to optimize the electrical performance of the photovoltaic cell.

[0014] A second aspect of the present invention proposes a system for predicting electrical performance and analyzing its influencing factors based on machine learning, comprising: a processor and a memory;

[0015] The memory stores programs or instructions that can run on the processor, which, when executed by the processor, implement the steps of the machine learning-based electrical performance prediction and influencing factor analysis method as described in the first aspect.

[0016] A third aspect of the present invention provides a readable storage medium on which a program or instructions are stored, which, when executed by a processor, implement the steps of the machine learning-based electrical performance prediction and influencing factor analysis method described in the first aspect.

[0017] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following:

[0018] In this embodiment of the invention, an automatic alignment mechanism is used to match the statistical characteristics of process parameters with electrical performance indicators, ensuring accurate data alignment in high-throughput and diverse production environments and reflecting the relationship between process parameters and electrical performance. A multimodal feature extension mechanism is used to combine process data from different sources and types to generate an enhanced process feature matrix, thereby characterizing the impact of the process on electrical performance. This overcomes the limitations of physical modeling in handling complex process parameter interactions and boundary conditions, improving prediction accuracy. Based on this enhanced process feature matrix, a hierarchical weighted regression model is used to generate electrical performance prediction results. The SHAP value theory is then used to analyze the marginal impact of each process parameter on the prediction results, accurately capturing nonlinear relationships and ensuring that the prediction accuracy meets the needs of actual production. Attached Figure Description

[0019] The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Throughout the drawings, the same reference numerals denote the same parts. Obviously, the drawings described below are merely some embodiments of the present invention, and those skilled in the art can obtain other drawings based on these drawings without any creative effort.

[0020] Figure 1 This is a flowchart illustrating the method for predicting electrical performance based on machine learning and analyzing its influencing factors provided in an embodiment of the present invention.

[0021] Figure 2 This is a schematic diagram of the structure of the machine learning-based electrical performance prediction and influencing factor analysis system provided in an embodiment of the present invention. Detailed Implementation

[0022] To enable those skilled in the art to better understand the technical solutions in the embodiments of the present invention, the technical solutions 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. It should be understood that these descriptions are merely exemplary and are not intended to limit the scope of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0023] The following description, in conjunction with the accompanying drawings, details the machine learning-based electrical performance prediction and influencing factor analysis method provided by the present invention through specific embodiments and application scenarios.

[0024] Reference manual attached Figure 1 The diagram illustrates a flowchart of the method for predicting electrical performance based on machine learning and analyzing its influencing factors, provided in an embodiment of the present invention.

[0025] This invention provides a method for predicting electrical performance and analyzing its influencing factors based on machine learning, which may include the following steps:

[0026] S1: Obtain the process parameters of photovoltaic cells during the production process and the electrical performance indicators of the corresponding production batches.

[0027] In one possible implementation, the process parameters include temperature, pressure, and gas concentration collected at fixed time intervals.

[0028] Electrical performance indicators include batch-level battery conversion efficiency, fill factor, open-circuit voltage, short-circuit current, and surface grayscale value.

[0029] It should be noted that the main task in this step is to obtain the key process parameters (such as temperature, pressure, and gas concentration) involved in the photovoltaic cell production process and the corresponding electrical performance indicators (such as conversion efficiency, fill factor, open-circuit voltage, and short-circuit current) of the production batches. This provides a data foundation for subsequent prediction and analysis.

[0030] S2: Extract process statistical features of process parameters.

[0031] In one possible implementation, the process statistics include maximum, minimum, average, median, variance, standard deviation, and skewness.

[0032] Optionally, the combined process features can be combined to form the following process statistical features:

[0033]

[0034] Where X represents the process statistical characteristics. x represents the mean of the process parameters. max x represents the maximum value of the process parameter. min σ represents the minimum value of the process parameter, median represents the median of the process parameter, and σ represents the variance.

[0035] Specifically, the goal of this step is to extract statistical features (such as maximum, minimum, mean, variance, etc.) from process parameters. These features help simplify the data and provide more representative characteristics, which in turn help the model make better predictions.

[0036] S3: Based on the production batch number and time window identifier, the process statistical characteristics are matched with electrical performance indicators through an automatic alignment mechanism to form a mapping relationship between process statistical characteristics and electrical performance indicators.

[0037] Automatic alignment mechanisms are commonly used in data processing, especially in the fields of time series data and experimental data, to ensure that data from different data sources or at different points in time can be accurately aligned according to specific rules. It ensures that process parameters and electrical performance indicators from different sources can be compared or analyzed on the same basis, thereby ensuring data consistency and integrity.

[0038] In one possible implementation, S3 specifically includes:

[0039] S301: Calculate the process statistical characteristics of the process parameters collected in each time period.

[0040] S302: Obtain the electrical performance indicators of the corresponding batch of batteries in each time period.

[0041] Optionally, the electrical performance index is Y = [y1, y2, y3, y4, y5], where Y represents the electrical performance index, y1 represents the efficiency, y2 represents the fill factor, y3 represents the open-circuit voltage, y4 represents the short-circuit current, and y5 represents the surface gray value.

[0042] S303: Using production batch numbers and time window identifiers, a dataset is formed by matching process statistical characteristics with electrical performance indicators through an automatic alignment mechanism.

[0043] In this embodiment of the invention, an automatic alignment mechanism is used to match process statistical characteristics with corresponding electrical performance indicators, ensuring data accuracy and consistency. This avoids data inconsistencies caused by manual processing or time misalignment between different data sources, providing a reliable data foundation for subsequent modeling.

[0044] S304: By using a combined mechanism of statistical constraints and dynamic thresholds to remove missing and outlier data from the dataset, a structured dataset containing process parameter input vectors and corresponding electrical performance parameter output vectors is obtained.

[0045] Optionally, the statistical constraint is specifically a three-standard-deviation test, and the dynamic threshold joint mechanism is specifically a Mahalanobis distance threshold model.

[0046] In this embodiment of the invention, a joint mechanism of statistical constraints and dynamic thresholds is introduced. Through a three-standard-deviation test and a Mahalanobis distance threshold model, missing and outlier values ​​in the data are effectively removed. This ensures the quality of the dataset and improves the accuracy of model training. By removing outlier data, the model can avoid noise interference, thereby improving prediction accuracy.

[0047] S305: Based on structured datasets, construct the mapping relationship between process statistical characteristics and electrical performance indicators.

[0048] Optionally, the mapping relationship is as follows:

[0049] Y b =f(X) b,t )

[0050] Among them, Y b Let X represent the electrical performance index, f represent the mapping function, and X represent the electrical performance index. b,t This indicates the statistical characteristics of the process.

[0051] Specifically, process parameters such as temperature, pressure, and gas concentration are collected from the MES database at fixed time intervals. Within each time period, statistical characteristics such as maximum, minimum, average, median, and variance are calculated. Simultaneously, electrical performance indicators such as conversion efficiency and fill factor of the corresponding batch of batteries are obtained. Using an automatic alignment mechanism between production batch number and time window identifier, process statistical characteristics are matched with electrical performance indicators. Then, missing and abnormal data are eliminated through a joint mechanism of statistical constraints (three-standard deviation test) and dynamic threshold (Mahavir distance threshold model). Finally, a structured dataset containing process parameter input vectors and corresponding electrical performance parameter output vectors is formed, thereby constructing the mapping relationship between the two.

[0052] In this embodiment of the invention, a mapping relationship between process statistical characteristics and electrical performance indicators is constructed based on a structured dataset, which helps to build a more accurate model. This step accurately represents the relationship between the statistical characteristics of process parameters and the electrical performance indicators of the battery, which helps to understand how process parameters affect battery performance and provides quantitative basis for subsequent optimization and prediction.

[0053] S4: Based on the mapping relationship, an enhanced process feature matrix is ​​constructed through a multimodal feature extension mechanism to characterize the impact of the process on electrical performance.

[0054] Among them, the multimodal feature extension mechanism is a data processing technique designed to extract information from multiple different types of data sources (or "modals") and merge or transform them into a more expressive feature representation. This mechanism is particularly suitable for processing heterogeneous data from multiple sources, such as multimodal data in photovoltaic cell production, including process parameters (such as temperature, pressure, and gas concentration) and electrical performance indicators (such as conversion efficiency, fill factor, and open-circuit voltage). Through this mechanism, an enhanced feature matrix that comprehensively reflects the impact of the process on electrical performance can be constructed.

[0055] In one possible implementation, S4 specifically includes:

[0056] S401: Define multiple process windows based on different process segments.

[0057] S402: Extract process statistical features of each process parameter from each process window:

[0058]

[0059] Among them, F j This represents the process statistical characteristics of the j-th process parameter. σ represents the mean of the j-th process parameter. j Skew represents the variance of the j-th process parameter. j x represents the skewness of the j-th process parameter. j,max x represents the maximum value of the j-th process parameter. j,min This represents the minimum value of the j-th process parameter.

[0060] Optionally, the formula for calculating skewness is:

[0061]

[0062] Where n represents the number of measurement points or samples for the process parameter, x j,i This represents the i-th data point or sampled value of the j-th process parameter.

[0063] S403: It integrates the cross terms and high-order nonlinear interaction terms between different process parameters, and uses the feature construction engine to upgrade each process parameter from its original dimension to a multi-dimensional tensor representation, forming an enhanced process feature matrix.

[0064] Optionally, cross terms and higher-order nonlinear interaction terms between process parameters can be constructed, such as the product term x of pairwise parameters. j ·x k Squared terms cubic terms etc., forming interactive feature vectors Among them, I j,k Represents the interaction feature vector, x j Let x represent the measured value of the j-th process parameter. k Let Fk represent the measured value of the k-th process parameter, denoted as Fk, .... Finally, concatenate all statistical features with the interaction features to obtain the enhanced process feature matrix: M = [F1, F2, ..., Fk]. m ,I 1,2 ,I 1,3 ,...,I m-1,m ], where M represents the enhanced process feature matrix, F P I represents the statistical characteristic vector of the p-th process parameter. p Let p = 1, 2, ..., m be the interaction feature vector, where p = m represents the number of process parameters.

[0065] It should be noted that this matrix uses a feature construction engine to elevate the original dimensions into a multidimensional tensor representation, thereby enhancing the features of complex process response patterns.

[0066] In this embodiment of the invention, complex interactions between process parameters can be effectively captured through cross-terms and higher-order nonlinear interaction terms (such as product terms, square terms, etc.). These interaction effects are often key factors affecting battery electrical performance, and traditional single process parameters (such as temperature, pressure, etc.) may not fully reflect these effects. By introducing interaction features, the model can more accurately simulate the complexity of the process. Simultaneously, by introducing cross-terms and higher-order nonlinear interaction features, the relationship between process parameters and electrical performance indicators can be more comprehensively reflected, thereby improving the accuracy of the prediction model. Since the influence of many parameters on electrical performance during the process is nonlinear, expanding the feature space can help the model better capture these complex influences, ultimately achieving more accurate electrical performance predictions.

[0067] S5: Based on the enhanced process feature matrix, electrical performance prediction results are generated through a hierarchical weighted regression model.

[0068] Among them, the hierarchical weighted regression model is an extension of the traditional regression model, aiming to capture more complex feature relationships and optimize prediction accuracy through a hierarchical structure and weighting mechanism. It is particularly suitable for handling multi-objective prediction problems or complex process / production data problems, and is especially useful in the prediction of photovoltaic cell electrical performance.

[0069] In one possible implementation, S5 specifically includes:

[0070] S501: Using the enhanced process feature matrix as input, in the first layer of the hierarchical weighted regression model, a sub-model based on gradient boosting tree is used to independently predict each electrical performance index, capturing the local patterns between each process parameter and each single electrical performance index, and obtaining the prediction results of each sub-model.

[0071] Optionally, the first-layer GBDT sub-model adopts a weight-aware objective function:

[0072]

[0073] Among them, L k Let w represent the weight-aware objective function. i,k y represents the sample weighting factor. i,k f represents the actual electrical performance value. k (x i ) represents the predicted electrical performance value of the i-th sample calculated by the k-th sub-model.

[0074] The objective function constructs independent sub-models by iteratively fitting the residuals.

[0075] In this embodiment of the invention, in the first layer, the model independently predicts each electrical performance index using a Gradient Boosting Tree (GBDT) sub-model. GBDT is able to capture local patterns between each electrical performance index and process parameters, which ensures accurate modeling of each electrical performance index.

[0076] S502: The prediction results of each sub-model are used as intermediate features and input into the overall predictor of the second layer of the hierarchical weighted regression model. By introducing an error-sensitive weight mechanism, dynamic weighting compensation is performed on the high error interval to optimize the prediction accuracy of the hierarchical weighted regression model in the high error region.

[0077] Optionally, the second-layer master predictor introduces an error-sensitive weighting mechanism, using the outputs of each sub-model as intermediate features, and dynamically compensates for them through a weighted residual loss function.

[0078]

[0079] Where L represents the weighted residual loss function, α k w′ represents the weight of the k-th electrical performance index in the loss function of the overall control predictor. i,k y represents the sample weighting factor. i,k F(x) represents the actual electrical performance value. k This represents the predicted value of the k-th electrical performance index.

[0080] In this embodiment of the invention, by introducing an error-sensitive weighting mechanism, the second-layer overall control predictor can identify which samples have large prediction errors and dynamically compensate for these high-error intervals. The model will pay more attention to these high-error samples, thereby improving the prediction accuracy in these regions.

[0081] S503: Based on the results of dynamic weighted compensation, the prediction results of each sub-model are comprehensively analyzed through the second-layer master control predictor to generate electrical performance prediction results.

[0082] Specifically, the enhanced process feature matrix is ​​first input into a hierarchical weighted regression model. The first layer employs independent sub-models based on Gradient Boosting Tree (GBDT) to predict five electrical performance indicators: conversion efficiency, fill factor, open-circuit voltage, short-circuit current, and surface grayscale value, capturing the local patterns of process features and single target variables. The second layer embeds the prediction results of all sub-models as intermediate features into the overall predictor, introducing an error-sensitive weighting mechanism to dynamically compensate for high-error intervals, thus obtaining the electrical performance prediction results.

[0083] In this embodiment of the invention, by accurately integrating the results of different sub-models and optimizing them through a master predictor, the final electrical performance prediction results are more reliable. In particular, the weighting and compensation of the prediction results enable the model to adapt to the complex and variable processes in actual production.

[0084] S6: Analyze the marginal impact of each process parameter on the electrical performance prediction results using SHAP value theory, and generate an impact matrix.

[0085] SHAP (SHapley Additive exPlanations) is a game theory-based interpretation method used to quantify the contribution of each feature in a machine learning model to the model's prediction results. The core idea of ​​SHAP value theory is to interpret the model's output by calculating the "marginal contribution" of each feature to the final prediction result.

[0086] In one possible implementation, S6 specifically includes:

[0087] S601: By applying small perturbations to each process parameter and using the finite perturbation difference formula, the marginal influence of each process parameter is calculated:

[0088]

[0089] Among them, S j,k This represents the marginal impact of the j-th process parameter on the predicted result of the k-th electrical performance index, where ε represents the disturbance magnitude, and e j f represents the unit vector of the j-th process parameter. k (x+εe j f represents the perturbed predicted value of the model output. k (x) represents the original predicted value of the model.

[0090] S602: Using the SHAP value theory, calculate the contribution of each process parameter to the electrical performance prediction results.

[0091] S603: Combine marginal impact and contribution to generate an impact matrix, where the elements in the impact matrix represent the impact and contribution of each process parameter on different electrical performance indicators.

[0092] Specifically, a method combining SHAP value theory with a custom perturbation response function is introduced to calculate the marginal impact of any process parameter on various electrical performance indicators under small perturbation conditions. By combining the finite perturbation difference formula with the embedded structure of the model, the marginal effect of the process parameter on the electrical performance indicators is expressed in matrix form, and finally an influence matrix is ​​formed.

[0093] In this embodiment of the invention, the marginal impact of each process parameter on the predicted electrical performance can be accurately calculated using the difference formula for small and finite perturbations. This method helps identify the sensitivity of process parameters to battery electrical performance, i.e., the magnitude of change in electrical performance indicators when a small change in a certain process parameter occurs. This is crucial for optimizing the production process and adjusting process parameters. Furthermore, by combining SHAP value theory and perturbation response functions, not only can the contribution of individual process parameters be calculated, but also the nonlinear relationships and interactions between process parameters can be identified. For example, temperature and gas concentration may have a joint effect on battery conversion efficiency, and the interaction between these two factors is often ignored in traditional linear modeling. This method can capture such complex interaction effects.

[0094] S7: Based on the influence matrix, adjust the process parameters in the photovoltaic cell production process to optimize the electrical performance of the photovoltaic cells.

[0095] In one possible implementation, S7 specifically includes:

[0096] S701: Select several key process parameters that affect the electrical performance of photovoltaic cells from the influence matrix.

[0097] It should be noted that the final output impact matrix not only serves as a performance response indicator under the current process environment, but also provides a clear direction for parameter tuning for the upstream control system: for any performance target, the positive or negative parameter with the largest impact value in the corresponding column is selected as the key process control quantity for improving or suppressing the target.

[0098] In this embodiment of the invention, an influence matrix can be used to systematically identify which process parameters have the greatest impact on battery electrical performance. These key process parameters can provide a clear direction for subsequent optimization.

[0099] S702: Analyze the marginal impact of each key process parameter on the electrical performance of photovoltaic cells.

[0100] It should be noted that the marginal impact refers to how the battery's electrical performance changes when these process parameters change, and whether the intensity of this change increases or decreases.

[0101] In this embodiment of the invention, the marginal impact measures the specific effect of changes in each process parameter on the battery's electrical performance. By quantifying this impact, the contribution of each process parameter to the electrical performance can be precisely understood.

[0102] S703: Based on the analysis results of each marginal impact degree, determine the adjustment range of each key process parameter.

[0103] In this embodiment of the invention, the value of each key process parameter can be precisely adjusted by defining a specific adjustment range. Unlike traditional experience-based adjustment methods, this method is based on data-driven approaches and marginal impact analysis, ensuring that the optimization direction and range of each process parameter have a scientific basis.

[0104] S704: Adjust the values ​​of the corresponding process parameters according to the various adjustment ranges to optimize the electrical performance of photovoltaic cells.

[0105] It's important to note that once the process parameters are adjusted, their effects need to be tracked through real-time monitoring. If the battery performance meets the expected targets, the optimization is considered successful. If the expected results are not achieved, further adjustments are needed, and new results should be obtained.

[0106] In this embodiment of the invention, through real-time monitoring, the adjustment of process parameters is not limited to a one-time change, but can be continuously tracked and repeatedly optimized. This means that the production line can more flexibly respond to changes that may occur during the production process, and gradually improve battery performance.

[0107] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following:

[0108] In this embodiment of the invention, an automatic alignment mechanism is used to match the statistical characteristics of process parameters with electrical performance indicators, ensuring accurate data alignment in high-throughput and diverse production environments and reflecting the relationship between process parameters and electrical performance. A multimodal feature extension mechanism is used to combine process data from different sources and types to generate an enhanced process feature matrix, thereby characterizing the impact of the process on electrical performance. This overcomes the limitations of physical modeling in handling complex process parameter interactions and boundary conditions, improving prediction accuracy. Based on this enhanced process feature matrix, a hierarchical weighted regression model is used to generate electrical performance prediction results. The SHAP value theory is then used to analyze the marginal impact of each process parameter on the prediction results, accurately capturing nonlinear relationships and ensuring that the prediction accuracy meets the needs of actual production.

[0109] Reference manual attached Figure 2 The diagram shows a schematic representation of the structure of the machine learning-based electrical performance prediction and influencing factor analysis system provided in an embodiment of the present invention.

[0110] This invention provides a machine learning-based electrical performance prediction and influencing factor analysis system 20, comprising: a processor 201 and a memory 202;

[0111] The memory 202 stores programs or instructions that can run on the processor 201. When the program or instructions are executed by the processor 201, they implement the steps of the above-mentioned method for predicting electrical performance based on machine learning and analyzing its influencing factors, and can achieve the same technical effect. To avoid repetition, the present invention will not elaborate further.

[0112] It should be understood that the processor 201 in this embodiment of the invention may be a central processing unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor.

[0113] It should also be understood that the memory 202 in the embodiments of the present invention can be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. The non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. The volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of random access memory (RAM) are available, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate synchronous DRAM (DDRSDRAM), enhanced synchronous DRAM (ESDRAM), synchronous linked DRAM (SLDRAM), and direct rambus RAM (DR RAM).

[0114] The above embodiments can be implemented, in whole or in part, by software, hardware (such as circuits), firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of the present invention are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. A semiconductor medium can be a solid-state drive.

[0115] It should be understood that, in various embodiments of the present invention, the sequence number of each process does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0116] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0117] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the devices, apparatuses, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0118] In the several embodiments provided by this invention, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another device, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0119] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0120] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0121] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, essentially, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0122] This invention provides a readable storage medium comprising: storing a program or instructions on the readable storage medium, wherein when the program or instructions are executed by a processor, the program or instructions implement the steps of the above-described method for predicting electrical performance based on machine learning and analyzing its influencing factors, and can achieve the same technical effect. To avoid repetition, this invention will not elaborate further.

[0123] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the embodiments of the present invention, and are not intended to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the protection scope of the present invention.

Claims

1. A method for predicting electrical performance based on machine learning and analyzing influencing factors thereof, characterized by, The method comprises the following steps: S1: obtaining process parameters of the photovoltaic cell in the production process and corresponding electrical performance indicators of the production batch; S2: extracting process statistical features of the process parameters; S3: according to the production batch number and the time window identifier, the process statistical features are matched with the electrical performance indicators through an automatic alignment mechanism to form a mapping relationship between the process statistical features and the electrical performance indicators; S4: based on the mapping relationship, an enhanced process feature matrix for representing the influence of the process on the electrical performance is constructed through a multi-modal feature expansion mechanism; S5: based on the enhanced process feature matrix, an electrical performance prediction result is generated through a hierarchical weighted regression model; S6: the marginal influence of each process parameter on the electrical performance prediction result is analyzed using SHAP value theory to generate an influence matrix; S7: according to the influence matrix, the process parameters in the photovoltaic cell production process are adjusted to optimize the electrical performance of the photovoltaic cell. 2.The method of claim 1, wherein, The process parameters include temperature, pressure and gas concentration collected at fixed time intervals; The electrical performance indicators include conversion efficiency, fill factor, open circuit voltage, short circuit current and surface gray value of the batch-level cell. 3.The method of claim 1, wherein, The process statistical features include maximum value, minimum value, average value, median, variance, standard deviation and skewness. 4.The method of claim 1, wherein, The S3 specifically comprises: S301: calculating the process statistical features of the collected process parameters in each time period; S302: obtaining the electrical performance indicators of the corresponding batch cells in each time period; S303: using the production batch number and the time window identifier, the process statistical features are matched with the electrical performance indicators through an automatic alignment mechanism to form a data set; S304: through a statistical constraint and dynamic threshold joint mechanism, the missing data and abnormal data in the data set are removed to obtain a structured data set containing process parameter input vectors and corresponding electrical performance parameter output vectors; S305: based on the structured data set, the mapping relationship between the process statistical features and the electrical performance indicators is constructed. 5.The method of claim 1, wherein, The S4 specifically comprises: S401: defining a plurality of process windows based on different process sections; S402: extracting process statistical features of each process parameter from each process window; S403: fusing cross terms and high-order nonlinear interaction terms between different process parameters, using a feature construction engine to promote each process parameter from the original dimension to a multi-dimensional tensor representation, forming the enhanced process feature matrix. 6.The method of claim 1, wherein, The S5 specifically comprises: S501: taking the enhanced process feature matrix as input, in the first layer of the hierarchical weighted regression model, using a gradient boosting tree-based sub-model to independently predict each electrical performance indicator, capturing the local patterns between each process parameter and each single electrical performance indicator, and obtaining the prediction results of each sub-model; S502: input the prediction results of each sub-model as intermediate features into a total control predictor of the second layer of the hierarchical weighted regression model, dynamically weight and compensate high error intervals by introducing an error sensitive weight mechanism, and optimize the prediction accuracy of the hierarchical weighted regression model in the high error area; S503: based on the results of dynamic weighted compensation, the prediction results of each sub-model are analyzed by the total control predictor of the second layer to generate the electrical performance prediction results. 7.The method of claim 5, wherein the machine learning-based electrical performance prediction and its influencing factor analysis is based on a machine learning algorithm. The S6 specifically includes: S601: calculate the marginal influence degree of each process parameter by slightly perturbing each process parameter and using a finite perturbation difference formula; S602: calculate the contribution degree of each process parameter to the electrical performance prediction results using the SHAP value theory; S603: combine the marginal influence degree and the contribution degree to generate the influence matrix, wherein the elements in the influence matrix are the influence degree and the contribution degree of each process parameter on different electrical performance indicators. 8.The method of claim 1, wherein, The S7 specifically includes: S701: select a plurality of key process parameters that affect the electrical performance of the photovoltaic cell from the influence matrix; S702: analyze the marginal influence degree of each key process parameter on the electrical performance of the photovoltaic cell; S703: determine the adjustment range of each key process parameter according to the analysis results of each marginal influence degree; S704: adjust the value of the corresponding process parameter according to each adjustment range to optimize the electrical performance of the photovoltaic cell. 9.A system for machine learning-based electrical performance prediction and analysis of influencing factors, characterized in that, It includes: a processor and a memory; The memory stores programs or instructions that can be run on the processor, and the programs or instructions are executed by the processor to realize the steps of the machine learning-based electrical performance prediction and its influencing factor analysis method according to any one of claims 1 to 8.

10. A readable storage medium, characterized by, The program or instruction is stored on the readable storage medium, and the program or instruction is executed by the processor to realize the steps of the machine learning-based electrical performance prediction and its influencing factor analysis method according to any one of claims 1 to 8.

Images