Perovskite battery device stability classification method and system based on machine learning

Abstract

The invention relates to the technical field of perovskite batteries, in particular to a perovskite battery device stability classification method and system based on machine learning, and the method comprises the steps: obtaining a plurality of perovskite device data sets, carrying out the multi-dimensional correlation integration of the preparation process data in the perovskite device data sets, obtaining the integrated process data, carrying out the feature recognition of a target data set, and carrying out the recognition of the feature of the target data set. A verification matrix set, a test matrix set, a condition multi-dimensional matrix and a result multi-dimensional matrix are used for training and screening a pre-constructed model set, a target learning model is obtained, current process data are obtained, a current integration matrix is obtained based on the current process data and current environment data, and the target learning model is obtained. And inputting the current integrated matrix into the target learning model to obtain a stability classification matrix, and completing stability classification of the perovskite battery device based on the stability classification matrix. According to the invention, the efficiency and reliability of evaluating the stability of the perovskite cell device can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of perovskite batteries, and in particular to a method and system for classifying the stability of perovskite battery devices based on machine learning. Background Art

[0002] The stability of perovskite solar cells is a key factor in their commercial application. Although perovskite materials have high photoelectric conversion efficiency, perovskite cells are easily degraded under environmental factors such as high temperature, humidity, and ultraviolet radiation, resulting in a decline in battery performance. Therefore, improving the long-term stability of batteries is crucial for practical applications. The stability of the device is closely related to the structure of the material and the quality of the interface layer. Any slight change may lead to a decrease in charge transfer efficiency, thereby affecting the performance of the perovskite cell.

[0003] Currently, researchers are using a variety of methods to improve the stability of perovskite solar cells, such as optimizing the composition of perovskite films, adopting different packaging technologies, and improving the materials of the battery interface layer. However, most of these methods require extensive experimentation and testing to gradually screen out effective modification methods. With the rapid development of artificial intelligence technology, researchers have begun to explore methods such as machine learning to accelerate the optimization process of material design. Artificial intelligence can help researchers more quickly identify potential stability improvement paths and reduce the time and resources required by traditional experimental methods.

[0004] Although existing machine learning technology can realize the stability classification of perovskite battery devices, there are still problems such as small data set size, inaccurate feature extraction and inconsistent training data. Therefore, there is an urgent need for a perovskite battery device stability assessment method with large data scale and standardized training data to improve the efficiency and reliability of perovskite battery device stability assessment. Summary of the Invention

[0005] The present invention provides a perovskite battery device stability classification method based on machine learning and a computer-readable storage medium, the main purpose of which is to improve the efficiency and reliability of evaluating the stability of perovskite battery devices.

[0006] To achieve the above objectives, the present invention provides a perovskite battery device stability classification method based on machine learning, comprising:

[0007] Acquire multiple perovskite device data sets, wherein the perovskite device data sets include: preparation process data, environmental test data and performance evaluation data, the environmental test data includes one or more of temperature value, humidity value and light intensity value, the performance evaluation data includes T 95 Value, T 90 Value and T 80One or more of the values;

[0008] The following operation is performed on each of the plurality of perovskite device data sets:

[0009] Perform multi-dimensional correlation integration on the preparation process data in the perovskite device data set to obtain integrated process data;

[0010] Performing a gap-filling and standardization operation on the environmental test data and performance evaluation data in the perovskite device data set to obtain standard test data and standard evaluation values;

[0011] Summarize and integrate process data, standard test data and standard evaluation values ​​to obtain the target data group;

[0012] Perform feature recognition on the target data group to obtain a single-dimensional feature matrix group;

[0013] Summarizing the single-dimensional feature matrix groups to obtain multiple single-dimensional feature matrix groups, and dividing the multiple single-dimensional feature matrix groups to obtain a training matrix group set, a verification matrix group set, and a test matrix group set;

[0014] Integrate the training matrix set to obtain the conditional multidimensional matrix and the result multidimensional matrix;

[0015] The pre-built model set is trained and screened using the verification matrix set, the test matrix set, the condition multidimensional matrix, and the result multidimensional matrix to obtain a target learning model, wherein the model set includes: multiple machine learning models;

[0016] receiving a device evaluation instruction, determining a device operating environment based on the device evaluation instruction, and obtaining current environmental data of the device operating environment;

[0017] Obtain current process data, obtain the current integration matrix based on the current process data and current environmental data, input the current integration matrix into the target learning model to obtain a stability classification matrix, and complete the stability classification of the perovskite battery device based on the stability classification matrix.

[0018] Optionally, the preparation process data includes: coating process, precursor solvent, antisolvent, anti-solvent time, annealing temperature, annealing time, annealing times, ETL material, HTL material, A-site material, A-site chemical formula, B-site material, B-site chemical formula, X-site material and X-site chemical formula, wherein the coating process includes: spin coating, blade coating, spray coating or printing, the precursor solvent includes: dimethylformamide, dimethyl sulfoxide, butyrolactone or other precursors, the antisolvent includes: chlorobenzene, toluene, dichlorobenzene or other antisolvents, and the ETL material includes: other electronic materials, SnO2, PCBM and C 60One or more of the HTL materials include: other hole materials, Spiro materials, PTAA and NiO x The A-site material includes one or more of formamidinium ions, methylamine ions, cesium ions, and rubidium ions; the B-site material includes one or more of lead ions and tin ions; and the X-site material includes one or more of iodide ions, bromide ions, and chloride ions.

[0019] Optionally, performing multi-dimensional correlation integration on the preparation process data in the perovskite device data set to obtain integrated process data includes:

[0020] The A-site ion radius is calculated based on the A-site chemical formula in the preparation process data. The calculation formula is as follows:

[0021]

[0022] Among them, r A is the A-site ionic radius, r FA 、r MA 、r Cs and r Rb are respectively the preset formamidine ion radius, the preset methylamine ion radius, the preset cesium ion radius and the preset rubidium ion radius, μ FA 、μ MA 、μ Cs and μ Rb are the proportion of formamidine ion in the A-site chemical formula, the proportion of methylamine ion in the A-site chemical formula, the proportion of cesium ion in the A-site chemical formula, and the proportion of rubidium ion in the A-site chemical formula;

[0023] The B-site ion radius is calculated based on the B-site chemical formula in the preparation process data. The calculation formula is as follows:

[0024]

[0025] Among them, r B is the B-site ion radius, r Pb and r Sn are the preset lead ion radius and the preset tin ion radius, μ Pb and μ Sn are the proportion of lead ions in the B-site chemical formula and the proportion of tin ions in the B-site chemical formula respectively;

[0026] The X-position ion radius is calculated based on the X-position chemical formula in the preparation process data. The calculation formula is as follows:

[0027]

[0028] Among them, r X is the X-site ionic radius, rI 、r Br and r Cl are the preset iodine ion radius, the preset bromide ion radius and the preset chloride ion radius, μ I 、μ Br and μ Cl are the proportion of iodide ion in the chemical formula at position X, the proportion of bromide ion in the chemical formula at position X, and the proportion of chloride ion in the chemical formula at position X;

[0029] The stability factor is calculated based on the A-site ionic radius, the B-site ionic radius, and the X-site ionic radius. The calculation formula is as follows:

[0030]

[0031] in, is the stability factor;

[0032] The crystal growth rate is calculated based on the anti-dissolution time, annealing temperature, annealing time and annealing times. The calculation formula is as follows:

[0033]

[0034] Among them, ω Crystal is the crystal growth rate, t k and t H are the anti-dissolution time and annealing time, T H and T0 are annealing temperature and preset standard temperature respectively, N H is the number of annealing times, ln is the natural logarithm, and tanh is the hyperbolic tangent function;

[0035] The coating process, precursor solvent, antisolvent, ETL material, HTL material, stability factor and crystal growth degree are summarized to obtain integrated process data.

[0036] Optionally, performing a gap-filling and standardization operation on the environmental test data and performance evaluation data in the perovskite device data set to obtain standard test data and standard evaluation values ​​includes:

[0037] Determine whether there is a temperature value in the environmental test data, if so, use the temperature value as the target temperature value, otherwise use a preset default temperature value as the target temperature value;

[0038] Determine whether there is a humidity value in the environmental test data, if there is a humidity value in the environmental test data, use the humidity value as the target humidity value, otherwise use a preset default humidity value as the target humidity value;

[0039] Determine whether there is a light intensity value in the environmental test data, and if so, use the light intensity value as the target light intensity value; otherwise, use a preset default light intensity value as the target light intensity value;

[0040] Summarize the target temperature value, target humidity value and target light intensity value to obtain standard test data;

[0041] Determine whether T exists in the performance evaluation data 90 Value, if there is T in the performance evaluation data 90 value, then T 90 The value is used as the standard evaluation value, otherwise it is judged whether T exists in the performance evaluation data. 95 value;

[0042] If there is T in the performance evaluation data 95 value, then according to T 95 The calculation formula is as follows:

[0043]

[0044] Among them, τ x is the standard evaluation value, T 95 value;

[0045] If T does not exist in the performance evaluation data 95 value, then according to T 80 The calculation formula is as follows:

[0046]

[0047] in, T 80 value.

[0048] Optionally, performing feature recognition on the target data group to obtain a single-dimensional feature matrix group includes:

[0049] Determining a first condition value based on the coating process of the integrated process data in the target data set, determining a second condition value based on the precursor solvent of the integrated process data in the target data set, and determining a third condition value based on the antisolvent in the target data set;

[0050] Determining a fourth condition value, a fifth condition value, a sixth condition value, and a seventh condition value based on the ETL material of the integrated process data in the target data set;

[0051] The eighth condition value, the ninth condition value, the tenth condition value, and the eleventh condition value are determined based on the HTL material of the integrated process data in the target data set;

[0052] The ion normalization value is calculated based on the stability factor of the integrated process data in the target data set, and the calculation formula is as follows:

[0053]

[0054] Among them, δ L is the ion normalized value, e is the natural constant;

[0055] Obtaining a crystallization normalization value based on the crystallization growth degree of the integrated process data in the target data set;

[0056] Obtaining a temperature normalized value, a humidity normalized value, and a light intensity normalized value based on the target temperature value, the target humidity value, and the target light intensity value of the standard test data in the target data group, respectively;

[0057] Summarize the first condition value, the second condition value, the third condition value, the fourth condition value, the fifth condition value, the sixth condition value, the seventh condition value, the eighth condition value, the ninth condition value, the tenth condition value, the eleventh condition value, the ion normalized value, the crystallization normalized value, the temperature normalized value, the humidity normalized value, and the light intensity normalized value to obtain a condition feature group;

[0058] Calculate an evaluation normalization value based on the standard evaluation value in the target data set;

[0059] The conditional feature group and the evaluation normalized value are used to construct the conditional one-dimensional matrix and the result one-dimensional matrix respectively, and the conditional one-dimensional matrix and the result one-dimensional matrix are summarized to obtain the one-dimensional feature matrix group.

[0060] Optionally, calculating the evaluation normalized value according to the standard evaluation value in the target data group includes:

[0061] Compare the standard evaluation value with the preset average T 90 value;

[0062] If the standard evaluation value is greater than the average T 90 value, the preset unit value is used as the evaluation normalization value, otherwise the preset zero value is used as the evaluation normalization value.

[0063] Optionally, the integration of the training matrix set to obtain a conditional multidimensional matrix and a result multidimensional matrix includes:

[0064] The conditional multidimensional matrix is ​​obtained by using the training matrix set, where the conditional multidimensional matrix is ​​as follows:

[0065]

[0066] Among them, Z X1 is a conditional multidimensional matrix, [Z a1 ] is the conditional single-dimensional matrix of the first training matrix group in the training matrix group set, [Z a2] is the conditional single-dimensional matrix of the second training matrix group in the training matrix group set, [Z ai ] is the conditional single-dimensional matrix of the i-th training matrix group in the training matrix group set, [Z an ] is the conditional single-dimensional matrix of the nth training matrix group in the training matrix group set, where n is the number of training matrix groups in the training matrix group set;

[0067] The result multidimensional matrix is ​​obtained by using the training matrix set, wherein the result multidimensional matrix is ​​as follows:

[0068]

[0069] Among them, Z Y is the resulting multidimensional matrix, [Z b1 ] is the result single-dimensional matrix of the first training matrix group in the training matrix group set, [Z b2 ] is the single-dimensional matrix of the second training matrix group in the training matrix group set, [Z bi ] is the result single-dimensional matrix of the i-th training matrix group in the training matrix group set, [Z bn ] is the resulting single-dimensional matrix of the nth training matrix group in the training matrix group set.

[0070] Optionally, the method of using the verification matrix set, the test matrix set, the conditional multidimensional matrix, and the result multidimensional matrix to train and screen the pre-constructed model set to obtain the target learning model includes:

[0071] For each machine learning model in the model set, perform the following operations:

[0072] The machine learning model is trained using the conditional multidimensional matrix and the result multidimensional matrix to obtain a preliminary training model;

[0073] For each verification matrix group in the verification matrix group set, perform the following operations:

[0074] Inputting the conditional single-dimensional matrix in the verification matrix group into the preliminary training model to obtain a result output matrix, and identifying a result output value from the result output matrix;

[0075] Determine whether the evaluation normalized value corresponding to the result single-dimensional matrix in the verification matrix group is equal to the unit value;

[0076] If the evaluation normalized value is equal to the unit value, determining whether the result output value is equal to the unit value;

[0077] If the result output value is equal to the unit value, the result output value is regarded as a true positive value; otherwise, the result output value is regarded as a false negative value;

[0078] If the evaluation normalized value is not equal to the unit value, determining whether the result output value is equal to the unit value;

[0079] If the result output value is equal to the unit value, the result output value is regarded as a false positive value; otherwise, the result output value is regarded as a true negative value;

[0080] Summarize the true positive values, false negative values, false positive values, and true negative values ​​respectively to obtain the true positive value set, false negative value set, false positive value set, and true negative value set;

[0081] Determining a correct positive number based on the correct positive value set, wherein the correct positive number is the number of correct positive values ​​in the correct positive value set;

[0082] Determine the number of false negatives, the number of false positives, and the number of true negatives based on the false negative value set, the false positive value set, and the true negative value set, respectively;

[0083] The success degree of the Ith training is calculated based on the number of correct positives, the number of false negatives, the number of false positives, and the number of correct negatives, where the initial value of I is 2;

[0084] The Ith item in the pre-constructed training record sequence is replaced with the Ith training success degree to obtain an updated record sequence, wherein the first item in the training record sequence is zero. The training record sequence is as follows:

[0085] {P1,P2…P m}

[0086] Among them, P1 is the first item in the training record sequence, P2 is the second item in the training record sequence, and P m is the mth item of the training record sequence;

[0087] Using the 1st training success degree to confirm the 1-1th training success degree in the update record sequence, wherein the 1-1th training success degree is the item before the 1st training success degree in the update record sequence;

[0088] Subtract the I-1 training success degree from the I-th training success degree to obtain the success degree difference;

[0089] Comparing the success degree difference with a preset difference threshold and comparing the I-th training success degree with a preset success threshold;

[0090] If the success degree difference is less than or equal to the difference threshold and the I-th training success degree is greater than or equal to the success threshold, the preliminary training model is used as the final training model;

[0091] If the success degree difference is greater than the difference threshold or the I-th training success degree is less than the success threshold, the preliminary training model is used as the machine learning model, the updated record sequence is used as the training record sequence, J=I+1, J is used as I, and the step of training the machine learning model using the conditional multidimensional matrix and the result multidimensional matrix is ​​returned to, until the success degree difference is less than or equal to the difference threshold and the I-th training success degree is greater than or equal to the success threshold or I=m, the preliminary training model is used as the final training model;

[0092] Aggregate the final training models to obtain multiple final training models;

[0093] The following operations are performed on each of the multiple final training models:

[0094] Get the final success rate based on the test matrix set and the final training model;

[0095] Summarizing the final success degrees to obtain multiple final success degrees, and determining a maximum success degree based on the multiple final success degrees, wherein the maximum success degree is the maximum final success degree among the multiple final success degrees;

[0096] The final training model corresponding to the maximum success degree is used as the target learning model.

[0097] Optionally, calculating the first training success degree according to the number of correct positives, the number of false negatives, the number of false positives, and the number of correct negatives includes:

[0098] The precision is calculated based on the number of true positives and false positives. The calculation formula is as follows:

[0099]

[0100] Among them, H Pre is the precision rate, R1 and F1 are the number of correct positives and the number of false positives, respectively;

[0101] The recall rate is calculated based on the number of true negatives and false negatives. The calculation formula is as follows:

[0102]

[0103] Among them, H Re is the recall rate, R2 and F2 are the number of true negatives and false negatives, respectively;

[0104] The success rate of the first training is calculated based on the precision, recall, number of true positives, number of false negatives, number of false positives, and number of true negatives. The calculation formula is as follows:

[0105]

[0106] in, is the success degree of the first training.

[0107] To achieve the above objectives, the present invention also provides a perovskite battery device stability classification system based on machine learning, comprising:

[0108] The device data acquisition module is used to acquire multiple perovskite device data groups, wherein the perovskite device data groups include: preparation process data, environmental test data and performance evaluation data, the environmental test data includes one or more of temperature value, humidity value and light intensity value, the performance evaluation data includes T 95 Value, T 90 Value and T 80 One or more of the values;

[0109] a data feature extraction module for performing the following operations on each of the multiple perovskite device data groups: performing multi-dimensional correlation integration on the preparation process data in the perovskite device data group to obtain integrated process data; performing a gap filling and standardization operation on the environmental test data and performance evaluation data in the perovskite device data group to obtain standard test data and standard evaluation values; summarizing and integrating the process data, standard test data, and standard evaluation values ​​to obtain a target data group; and performing feature recognition on the target data group to obtain a single-dimensional feature matrix group;

[0110] A classification model training module is used to summarize a single-dimensional feature matrix group to obtain multiple single-dimensional feature matrix groups, divide the multiple single-dimensional feature matrix groups to obtain a training matrix group set, a verification matrix group set, and a test matrix group set, integrate the training matrix group set to obtain a conditional multidimensional matrix and a result multidimensional matrix, and use the verification matrix group set, the test matrix group set, the conditional multidimensional matrix, and the result multidimensional matrix to train and screen a pre-constructed model set to obtain a target learning model, wherein the model set includes: multiple machine learning models;

[0111] The device stability prediction module is used to receive device evaluation instructions, confirm the device working environment based on the device evaluation instructions, obtain the current environmental data of the device working environment, obtain the current process data, obtain the current integration matrix based on the current process data and the current environmental data, input the current integration matrix into the target learning model, obtain the stability classification matrix, and complete the stability classification of the perovskite battery device based on the stability classification matrix.

[0112] In order to solve the above problem, the present invention further provides an electronic device, comprising:

[0113] a memory storing at least one instruction;

[0114] A processor executes instructions stored in the memory to implement the above-mentioned perovskite battery device stability classification method based on machine learning.

[0115] In order to solve the above problems, the present invention also provides a computer-readable storage medium, which stores at least one instruction, and the at least one instruction is executed by a processor in an electronic device to implement the above-mentioned machine learning-based perovskite battery device stability classification method.

[0116] The present invention aims to solve the problem described in the background technology. The present invention obtains multiple perovskite device data sets, wherein the perovskite device data sets include: preparation process data, environmental test data and performance evaluation data, the environmental test data includes one or more of temperature value, humidity value and light intensity value, the performance evaluation data includes T 95 Value, T 90 Value and T 80One or more of the values, it can be seen that the embodiment of the present invention obtains multiple perovskite device data groups from the literature, thereby providing a huge data set for subsequent model training, thereby improving the accuracy of the prediction results after training, and then performing the following operations on each of the multiple perovskite device data groups: multi-dimensionally correlating and integrating the preparation process data in the perovskite device data group to obtain integrated process data. It can be seen that the embodiment of the present invention summarizes and integrates the multi-dimensional data in the preparation process data into more streamlined values ​​by multi-dimensionally correlating and integrating the preparation process data in the perovskite device data group. At the same time, the stability factor and the crystal growth degree can also more accurately reflect the performance of the perovskite crystal recorded in the literature, thereby reducing the subsequent training model. The total amount of training data at the time of implementation improves the efficiency of subsequent model training and the accuracy of model prediction, performs a gap filling and standardization operation on the environmental test data and performance evaluation data in the perovskite device data group, and obtains standard test data and standard evaluation values. It can be seen that the embodiment of the present invention performs a gap filling and standardization operation on the environmental test data and performance evaluation data in the perovskite device data group, thereby supplementing and standardizing the missing data, making the input data for subsequent model training more unified, summarizing and integrating process data, standard test data and standard evaluation values ​​to obtain a target data group, performing feature recognition on the target data group to obtain a single-dimensional feature matrix group, it can be seen that the embodiment of the present invention performs feature recognition on the target data group, reduces the dimensionality of high-dimensional data, and obtains a single-dimensional feature matrix group. dimensional feature matrix group, so that the matrix can be directly input into the model for training in the future, and the single-dimensional feature matrix group is summarized to obtain multiple single-dimensional feature matrix groups. The multiple single-dimensional feature matrix groups are divided to obtain training matrix group sets, verification matrix group sets and test matrix group sets. The training matrix group sets are integrated to obtain conditional multidimensional matrices and result multidimensional matrices. It can be seen that the embodiment of the present invention facilitates the subsequent machine learning model to learn the mapping relationship between the conditional multidimensional matrix and the result multidimensional matrix by classifying and integrating the training matrix group sets into conditional multidimensional matrices and result multidimensional matrices. The pre-constructed model set is trained and screened using the verification matrix group set, the test matrix group set, the conditional multidimensional matrix and the result multidimensional matrix to obtain the target learning model, wherein the model set includes: multiple machine learning models. It can be seen that the embodiment of the present invention improves the efficiency and reliability of the target learning model in evaluating the stability of the perovskite battery device by simultaneously training and screening multiple machine learning models, and taking the machine learning model with the best prediction effect after training as the target learning model, receives a device evaluation instruction, confirms the device working environment based on the device evaluation instruction, obtains the current environmental data of the device working environment, obtains the current process data, obtains the current integration matrix based on the current process data and the current environmental data, inputs the current integration matrix into the target learning model, obtains the stability classification matrix, and completes the stability classification of the perovskite battery device based on the stability classification matrix. It can be seen that the embodiment of the present invention obtains the current environmental data of the device working environment,The stability of the perovskite battery device is predicted according to the environment in which the perovskite battery device is located, thereby improving the reliability of the stability evaluation of the perovskite battery device. Therefore, the present invention can improve the efficiency and reliability of the stability evaluation of the perovskite battery device. BRIEF DESCRIPTION OF THE DRAWINGS

[0117] Figure 1 A schematic flow chart of a method for classifying the stability of perovskite battery devices based on machine learning according to an embodiment of the present invention;

[0118] Figure 2 A functional module diagram of a perovskite battery device stability classification system based on machine learning provided by one embodiment of the present invention;

[0119] Description of reference numerals:

[0120] 1. Electronic device; 10. Processor; 11. Memory; 12. Bus.

[0121] The purpose, features and advantages of the present invention will be further described with reference to the accompanying drawings and in conjunction with the embodiments. DETAILED DESCRIPTION

[0122] It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0123] The embodiments of the present application provide a method for classifying the stability of a perovskite battery device based on machine learning. The execution subject of the method includes, but is not limited to, at least one of electronic devices such as a server and a terminal that can be configured to execute the method provided in the embodiments of the present application. In other words, the method for classifying the stability of a perovskite battery device based on machine learning can be executed by software or hardware installed on a terminal device or a server device, and the software can be a blockchain platform. The server includes, but is not limited to, a single server, a server cluster, a cloud server, or a cloud server cluster.

[0124] Reference Figure 1 FIG2 is a flow chart of a method for classifying the stability of a perovskite battery device based on machine learning according to an embodiment of the present invention. In this embodiment, the method for classifying the stability of a perovskite battery device based on machine learning includes:

[0125] S1. Acquire multiple perovskite device data sets, wherein the perovskite device data sets include: preparation process data, environmental test data, and performance evaluation data, wherein the environmental test data includes one or more of temperature, humidity, and light intensity values, and the performance evaluation data includes T 95 Value, T 90 Value and T80 One or more of the values.

[0126] In the embodiment of the present invention, the perovskite cell device refers to a perovskite solar cell.

[0127] In detail, the preparation process data includes: coating process, precursor solvent, antisolvent, anti-solvent time, annealing temperature, annealing time, annealing times, ETL material, HTL material, A-site material, A-site chemical formula, B-site material, B-site chemical formula, X-site material and X-site chemical formula, wherein the coating process includes: spin coating, blade coating, spray coating or printing, the precursor solvent includes: dimethylformamide, dimethyl sulfoxide, butyrolactone or other precursors, the antisolvent includes: chlorobenzene, toluene, dichlorobenzene or other antisolvents, and the ETL material includes: other electronic materials, SnO2, PCBM and C 60 One or more of the HTL materials include: other hole materials, Spiro materials, PTAA and NiO x The A-site material includes one or more of formamidinium ions, methylamine ions, cesium ions, and rubidium ions; the B-site material includes one or more of lead ions and tin ions; and the X-site material includes one or more of iodide ions, bromide ions, and chloride ions.

[0128] Exemplarily, multiple documents related to perovskite battery devices are retrieved and collected, and multiple data related to perovskite battery devices are extracted from the multiple documents to obtain the multiple perovskite device data groups, and one document corresponds to one perovskite device data group.

[0129] It should be explained that the perovskite device data group includes: preparation process data, environmental test data and performance evaluation data. The preparation process data refers to the data related to the preparation process of perovskite battery devices recorded in the literature, including: coating process, precursor solvent, anti-solvent, anti-dissolution time, annealing temperature, annealing time, annealing times, ETL material, HTL material, A-site material, A-site chemical formula, B-site material, B-site chemical formula, X-site material and X-site chemical formula. The coating process refers to the deposition method of the perovskite precursor solution on the substrate of the device recorded in the literature, and the coating process corresponding to a document is one of spin coating, blade coating, spray coating and printing, and the specific operating processes of spin coating, blade coating, spray coating and printing are all existing technologies in the field of perovskite battery device preparation, which will not be repeated here. A precursor solution refers to a solution containing a variety of precursor compounds and a precursor solvent, and a precursor solvent is a solvent for dissolving a variety of precursor compounds. In the preparation process of perovskite battery devices recorded in the literature, the precursor solution is usually pre-configured by the experimenter, and the precursor solution is used to be coated on the substrate of the device to form a perovskite crystal structure. For example, a document records the process of preparing a perovskite battery device with the chemical formula FAPbI3 (FA is formamidine, Pb is lead, and I is iodine). In this preparation process, the experimenter dissolves formamidine iodide (FAI) and lead iodide (PbI2) in N, N-dimethylformamide solution in a certain proportion in advance to configure the required precursor solution, wherein formamidine iodide and lead iodide are the precursor compounds, and N, N-dimethylformamide solution is the precursor solvent. An antisolvent is a solution added during the coating process in the preparation of a perovskite battery device recorded in the literature, which is used to induce the precursor solution to crystallize rapidly on the substrate. The anti-solvent time refers to the time difference between the addition of the antisolvent and the start of the coating process. For example, if the antisolvent is added 40 seconds after the start of the coating process, the anti-solvent time is 40 seconds. Precursor solvents include dimethylformamide, dimethyl sulfoxide, butyrolactone, or other precursors. Dimethylformamide refers to N,N-dimethylformamide, and butyrolactone refers to 1,4-butyrolactone. When the precursor solvent used in the literature is not dimethylformamide, dimethyl sulfoxide, or butyrolactone, it is denoted as "other precursors." Antisolvents include chlorobenzene, toluene, dichlorobenzene, or other antisolvents. Dichlorobenzene refers to 1,2-dichlorobenzene. When the antisolvent used in the literature is not chlorobenzene, toluene, or dichlorobenzene, it is denoted as "other antisolvents."

[0130] It should be understood that after the precursor solution is coated on the substrate through a coating process, the coated film is heated at a certain temperature for a period of time to remove the solvent, promote crystal growth and structural rearrangement, and the annealing temperature is the temperature during the above-mentioned heating treatment, and the annealing time is the duration of the above-mentioned heating treatment, and the annealing temperature and annealing time are both obtained from the specific operation records in the literature.

[0131] For example, if the annealing process recorded in the literature is a single annealing, the number of annealing times is 1; if the annealing process recorded in the literature is a double annealing, the number of annealing times is 2.

[0132] It should be explained that ETL materials refer to the materials used in the electron transport layer of perovskite battery devices. ETL materials include other electronic materials, SnO2, PCBM and C 60 For example, if the electron transport layer in the perovskite battery device described in a certain document is a mixed layer composed of SnO2 and PCBM, then the ETL materials in the perovskite device data set corresponding to the document are SnO2 and PCBM, PCBM is a fullerene derivative, C 60 Refers to carbon 60, SnO2 refers to tin dioxide, when the ETL material used in the literature is not SnO2, PCBM and C 60 When the ETL material is used, it is recorded as other electronic materials. HTL materials refer to the materials used for the hole transport layer in perovskite battery devices. HTL materials include other hole materials, Spiro materials, PTAA and NiO x One or more of the following, Spiro material refers to Spiro-OMeTAD, PTAA refers to: poly [bis (4-phenyl) (2,4,6-trimethylphenyl) amine], NiO x Refers to nickel oxide. When the HTL material used in the literature is not Spiro material, PTAA and NiO x When the HTL material is used, it is recorded as other hole materials.

[0133] It is understandable that the general formula of the chemical formula of the perovskite crystal in the perovskite battery device is ABX3, wherein A represents an organic or inorganic monovalent cation, and the part of the chemical formula of the perovskite crystal recorded in the literature where the organic or inorganic monovalent cation is located is the A-site chemical formula, and the substance included in the A-site chemical formula is the A-site material, B represents a divalent metal cation, and the part of the chemical formula of the perovskite crystal recorded in the literature where the divalent metal cation is located is the B-site chemical formula, and the substance included in the B-site chemical formula is the B-site material, X represents a monovalent halogen anion, and the part of the chemical formula of the perovskite crystal recorded in the literature where the monovalent halogen anion is located is the X-site chemical formula, and the substance included in the X-site chemical formula is the X-site material. For example, the chemical formula of the perovskite crystal in a perovskite battery device recorded in a certain document is FA 0.6 MA 0.4 PbI3, then the chemical formula of A is FA 0.6 MA 0.4 The A-site materials are FA and MA, where FA represents formamidinium ion and MA represents methylamine ion. The B-site chemical formula is Pb, and the B-site material is lead ion. The X-site chemical formula is I3, and the X-site material is iodide ion.

[0134] It should be explained that environmental test data refers to data related to the light aging test conditions of perovskite cell devices recorded in the literature, including one or more of the temperature value, humidity value and light intensity value. For example: the experimenter corresponding to a certain document conducted a light aging test on the perovskite cell device it prepared. The test conditions are as follows: temperature is 35 degrees Celsius, relative humidity is 65%, and light intensity is 1sun, where sun is a commonly used standard unit of light intensity, representing the intensity of standard sunlight on the earth's surface (air quality AM1.5). At this time, the temperature value in the environmental test data is 35 degrees Celsius, the humidity value is 65%, and the light intensity value is 1sun. Performance evaluation data refers to data related to the changes in the photoelectric conversion efficiency of perovskite cell devices recorded in the literature, including: T 95 Value, T 90 Value and T 80 One or more values, T 95 Value, T 90 Value and T 80 The values ​​refer to the time required for the photoelectric conversion efficiency of the perovskite cell device to drop to 95%, 90% and 80% of the initial value during light aging test.

[0135] It should be understood that since some documents do not fully disclose all the conditions in the light aging test, the environmental test data includes one or more of the temperature, humidity and light intensity values. For example, a certain document only records the temperature and light intensity when the perovskite cell device is subjected to the light aging test, so the environmental test data corresponding to the document only includes the temperature and light intensity values. At the same time, since some documents only record the changes in the photoelectric conversion efficiency during the light aging test, when recording the changes in the photoelectric conversion efficiency, one or more of the time required for the photoelectric conversion efficiency to drop to 95%, 90% and 80% of the initial value is recorded, the performance evaluation data includes T 95 Value, T 90 Value and T 80 One or more of the values.

[0136] S2. Perform the following operation on each of the multiple perovskite device data groups: perform multi-dimensional correlation integration on the preparation process data in the perovskite device data group to obtain integrated process data.

[0137] In detail, the multi-dimensional correlation integration of the preparation process data in the perovskite device data set to obtain the integrated process data includes:

[0138] The A-site ion radius is calculated based on the A-site chemical formula in the preparation process data. The calculation formula is as follows:

[0139]

[0140] Among them, r A is the A-site ionic radius, r FA 、r MA 、r Cs and r Rb are respectively the preset formamidine ion radius, the preset methylamine ion radius, the preset cesium ion radius and the preset rubidium ion radius, μ FA 、μ MA 、μ Cs and μ Rb are the proportion of formamidine ion in the A-site chemical formula, the proportion of methylamine ion in the A-site chemical formula, the proportion of cesium ion in the A-site chemical formula, and the proportion of rubidium ion in the A-site chemical formula;

[0141] The B-site ion radius is calculated based on the B-site chemical formula in the preparation process data. The calculation formula is as follows:

[0142]

[0143] Among them, r B is the B-site ion radius, r Pb and r Snare the preset lead ion radius and the preset tin ion radius, μ Pb and μ Sn are the proportion of lead ions in the B-site chemical formula and the proportion of tin ions in the B-site chemical formula respectively;

[0144] The X-position ion radius is calculated based on the X-position chemical formula in the preparation process data. The calculation formula is as follows:

[0145]

[0146] Among them, r X is the X-site ionic radius, r I 、r Br and r Cl are the preset iodine ion radius, the preset bromide ion radius and the preset chloride ion radius, μ I 、μ Br and μ Cl are the proportion of iodide ion in the chemical formula at position X, the proportion of bromide ion in the chemical formula at position X, and the proportion of chloride ion in the chemical formula at position X;

[0147] The stability factor is calculated based on the A-site ionic radius, the B-site ionic radius, and the X-site ionic radius. The calculation formula is as follows:

[0148]

[0149] in, is the stability factor;

[0150] The crystal growth rate is calculated based on the anti-dissolution time, annealing temperature, annealing time and annealing times. The calculation formula is as follows:

[0151]

[0152] Among them, ω Crystal is the crystal growth rate, t k and t H are the anti-dissolution time and annealing time, T H and T0 are annealing temperature and preset standard temperature respectively, N H is the number of annealing times, ln is the natural logarithm, and tanh is the hyperbolic tangent function;

[0153] The coating process, precursor solvent, antisolvent, ETL material, HTL material, stability factor and crystal growth degree are summarized to obtain integrated process data.

[0154] It should be understood that since the A-site material includes one or more of formamidinium ions, methylamine ions, cesium ions, and rubidium ions, the A-site chemical formula in the preparation process data does not necessarily include every ion of formamidinium ions, methylamine ions, cesium ions, and rubidium ions. If a certain ion of formamidinium ions, methylamine ions, cesium ions, and rubidium ions does not exist in the A-site chemical formula, the proportion of the ion in the A-site chemical formula is 0. For example: the A-site chemical formula is FA 0.6 MA 0.4 , then μ FA 、μ MA 、μ Cs and μ Rb are 0.6, 0.4, 0 and 0 respectively, and the chemical formula of B is Pb, then μ Pb and μ Sn are 1 and 0 respectively, and the chemical formula of X is I3, then μ I 、μ Br and μ Cl They are 1, 0 and 0 respectively.

[0155] It should be explained that the radius of formamidine ion, the radius of methylamine ion, the radius of cesium ion, the radius of rubidium ion, the radius of lead ion, the radius of tin ion, the radius of iodide ion, the radius of bromide ion and the radius of chloride ion are respectively the radius of formamidine ion, the radius of methylamine ion, the radius of cesium ion, the radius of rubidium ion, the radius of lead ion, the radius of tin ion, the radius of iodide ion, the radius of bromide ion and the radius of chloride ion, among which the radius of cesium ion, the radius of rubidium ion, the radius of lead ion, the radius of tin ion, the radius of iodide ion, the radius of bromide ion and the radius of chloride ion are all common knowledge, while the radius of formamidine ion and methylamine ion does not have a strictly fixed radius, and the actual value of their radius may vary depending on the crystal symmetry. The radius of the formamidine ion is slightly different due to the effects of the structure, configuration and hydrogen bond. Therefore, the embodiment of the present invention selects the average value of the geometric fitting values ​​of the formamidine ion radius in the perovskite crystal structure recorded in all documents as the formamidine ion radius, and selects the average value of the geometric fitting values ​​of the methylamine ion radius in the perovskite crystal structure recorded in all documents as the methylamine ion radius. Optionally, the formamidine ion radius is 253, the methylamine ion radius is 217, the cesium ion radius is 167, the rubidium ion radius is 148, the lead ion radius is 119, the tin ion radius is 112, the iodine ion radius is 220, the bromide ion radius is 196, and the chloride ion radius is 181, and the units of the above ion radii are all picometers.

[0156] Optionally, the average value of the annealing temperatures in all the literatures used in the embodiments of the present invention is used as the standard temperature.

[0157] It is understood that the A-site ionic radius, the B-site ionic radius, and the X-site ionic radius refer to the weighted average radius of all ions in the A-site chemical formula, the weighted average radius of all ions in the B-site chemical formula, and the weighted average radius of the X-site chemical formula, respectively. The stability factor reflects the stability of the crystal structure corresponding to the chemical formula of the perovskite battery device. The larger the stability factor, the more stable the crystal structure corresponding to the chemical formula of the perovskite battery device.

[0158] It should be understood that, generally speaking, the longer the anti-solvent time is, the later the anti-solvent is added during the coating process, and the longer the crystallization process of the perovskite crystals in the perovskite cell device is. The longer the annealing time, the higher the annealing temperature, and the more annealing times, the larger the grains in the perovskite crystals and the more complete the crystallization. Therefore, the crystal growth degree reflects the degree of crystallization of the perovskite crystals in the perovskite cell device. The greater the crystal growth degree, the higher the degree of crystallization of the perovskite crystals in the perovskite cell device.

[0159] It can be understood that the embodiment of the present invention calculates the stability factor and crystal growth degree by multi-dimensionally correlating and integrating the preparation process data, thereby summarizing and integrating the multi-dimensional data in the preparation process data into more streamlined values. At the same time, the stability factor and crystal growth degree can more accurately reflect the performance of the perovskite crystals recorded in the literature, thereby reducing the total amount of training data in the subsequent model training, and improving the efficiency of subsequent model training and the accuracy of model prediction.

[0160] S3. Perform a gap-filling and standardization operation on the environmental test data and performance evaluation data in the perovskite device data group to obtain standard test data and standard evaluation values.

[0161] In detail, performing a gap-filling and standardization operation on the environmental test data and performance evaluation data in the perovskite device data set to obtain standard test data and standard evaluation values ​​includes:

[0162] Determine whether there is a temperature value in the environmental test data, if so, use the temperature value as the target temperature value, otherwise use a preset default temperature value as the target temperature value;

[0163] Determine whether there is a humidity value in the environmental test data, if there is a humidity value in the environmental test data, use the humidity value as the target humidity value, otherwise use a preset default humidity value as the target humidity value;

[0164] Determine whether there is a light intensity value in the environmental test data, and if so, use the light intensity value as the target light intensity value; otherwise, use a preset default light intensity value as the target light intensity value;

[0165] Summarize the target temperature value, target humidity value and target light intensity value to obtain standard test data;

[0166] Determine whether T exists in the performance evaluation data 90 Value, if there is T in the performance evaluation data 90 value, then T 90 The value is used as the standard evaluation value, otherwise it is judged whether T exists in the performance evaluation data. 95 value;

[0167] If there is T in the performance evaluation data 95 value, then according to T 95 The calculation formula is as follows:

[0168]

[0169] Among them, τ x is the standard evaluation value, T 95 value;

[0170] If T does not exist in the performance evaluation data 95 value, then according to T 80 The calculation formula is as follows:

[0171]

[0172] in, T 80 value.

[0173] It should be understood that since some documents do not fully disclose all the conditions in the light aging test, that is, the environmental test data may include one or more of the temperature value, humidity value and light intensity value, the embodiment of the present invention fills in the gaps in the environmental test data to facilitate more unified and complete input data for subsequent model training. At the same time, since some documents only record one or more of the time required for the photoelectric conversion efficiency to drop to 95%, 90% and 80% of the initial value when recording the change in photoelectric conversion efficiency during the light aging test, in order to facilitate subsequent model training, the embodiment of the present invention uses T 95 Value and T 80 Values ​​are uniformly converted to T 90 value, so that the input data will be more unified when the model is trained later.

[0174] It should be explained that the default temperature value, default humidity value and default light intensity value are all values ​​set manually by the trainer of the machine learning model. Optionally, the default temperature value is 25°C, the default humidity value is 30%, and the default light intensity value is 0sun.

[0175] S4. Summarize and integrate the process data, standard test data, and standard evaluation values ​​to obtain a target data group, perform feature recognition on the target data group, and obtain a single-dimensional feature matrix group.

[0176] Specifically, the feature recognition is performed on the target data group to obtain a single-dimensional feature matrix group, including:

[0177] Determining a first condition value based on the coating process of the integrated process data in the target data set, determining a second condition value based on the precursor solvent of the integrated process data in the target data set, and determining a third condition value based on the antisolvent in the target data set;

[0178] Determining a fourth condition value, a fifth condition value, a sixth condition value, and a seventh condition value based on the ETL material of the integrated process data in the target data set;

[0179] The eighth condition value, the ninth condition value, the tenth condition value, and the eleventh condition value are determined based on the HTL material of the integrated process data in the target data set;

[0180] The ion normalization value is calculated based on the stability factor of the integrated process data in the target data set, and the calculation formula is as follows:

[0181]

[0182] Among them, δ L is the ion normalized value, e is the natural constant;

[0183] Obtaining a crystallization normalization value based on the crystallization growth degree of the integrated process data in the target data set;

[0184] Obtaining a temperature normalized value, a humidity normalized value, and a light intensity normalized value based on the target temperature value, the target humidity value, and the target light intensity value of the standard test data in the target data group, respectively;

[0185] Summarize the first condition value, the second condition value, the third condition value, the fourth condition value, the fifth condition value, the sixth condition value, the seventh condition value, the eighth condition value, the ninth condition value, the tenth condition value, the eleventh condition value, the ion normalized value, the crystallization normalized value, the temperature normalized value, the humidity normalized value, and the light intensity normalized value to obtain a condition feature group;

[0186] Calculate an evaluation normalization value based on a standard evaluation value in a target data set;

[0187] The conditional feature group and the evaluation normalized value are used to construct the conditional one-dimensional matrix and the result one-dimensional matrix respectively, and the conditional one-dimensional matrix and the result one-dimensional matrix are summarized to obtain the one-dimensional feature matrix group.

[0188] It should be understood that the method for obtaining a crystallization normalized value based on the crystal growth degree of the integrated process data in the target data set is the same as the method for obtaining an ion normalized value using the stability factor of the integrated process data in the target data set, and will not be described in detail here. The method for obtaining a temperature normalized value based on the target temperature value of the standard test data in the target data set, the method for obtaining a humidity normalized value based on the target humidity value of the standard test data in the target data set, and the method for obtaining a light intensity normalized value based on the target light intensity value of the standard test data in the target data set are all the same as the method for obtaining an ion normalized value using the stability factor of the integrated process data in the target data set, and will not be described in detail here.

[0189] In detail, determining the first condition value based on the coating process in the integrated process data includes:

[0190] If the coating process is a spin coating method, the preset first single value is used as the first condition value;

[0191] If the coating process is a blade coating method, the preset second single value is used as the first condition value;

[0192] If the coating process is a spray coating method, the preset third single value is used as the first condition value;

[0193] If the coating process is a printing method, the preset fourth single value is used as the first condition value.

[0194] In detail, the determining of the second condition value based on the precursor solvent of the integrated process data in the target data set includes:

[0195] If the precursor solvent is dimethylformamide, the first single value is used as the second condition value;

[0196] If the precursor solvent is dimethyl sulfoxide, the second single value is used as the second condition value;

[0197] If the precursor solvent is butyrolactone, the third single value is used as the second condition value;

[0198] If the precursor solvent is other precursors, the fourth single value is used as the second condition value.

[0199] In detail, determining the third condition value based on the antisolvent in the target data set includes:

[0200] If the antisolvent is chlorobenzene, the first single value is used as the third condition value;

[0201] If the antisolvent is toluene, the second single value is used as the third condition value;

[0202] If the antisolvent is dichlorobenzene, the third single value is used as the third condition value;

[0203] If the antisolvent is another antisolvent, the fourth single value is used as the third condition value.

[0204] It should be explained that the first single value is 0.25, the second single value is 0.5, the third single value is 0.75, and the fourth single value is 1.

[0205] Specifically, the fourth condition value, the fifth condition value, the sixth condition value, and the seventh condition value are determined based on the ETL material of the integrated process data in the target data set, including:

[0206] If SnO2 exists in the ETL material, the fourth condition value is taken as a unit value, otherwise the fourth condition value is taken as a zero value;

[0207] If PCBM exists in the ETL material, the unit value is used as the fifth condition value, otherwise the zero value is used as the fifth condition value;

[0208] If there is C in the ETL material 60 , then the unit value is used as the sixth condition value, otherwise the zero value is used as the sixth condition value;

[0209] If other electronic materials exist in the ETL material, a unit value is used as the seventh condition value; otherwise, a zero value is used as the seventh condition value.

[0210] Specifically, the eighth condition value, the ninth condition value, the tenth condition value, and the eleventh condition value are determined based on the HTL material of the integrated process data in the target data set, including:

[0211] If Spiro material exists in the HTL material, the unit value is used as the eighth condition value, otherwise the zero value is used as the eighth condition value;

[0212] If PTAA exists in the HTL material, the unit value is used as the ninth condition value, otherwise the zero value is used as the ninth condition value;

[0213] If NiO exists in HTL materials x , then the unit value is used as the tenth condition value, otherwise the zero value is used as the tenth condition value;

[0214] If other hole materials exist in the HTL material, the unit value is used as the eleventh condition value; otherwise, the zero value is used as the eleventh condition value.

[0215] In detail, the conditional one-dimensional matrix is ​​as follows:

[0216] Z a =[a1,a2,…a i ,…a 11 ,a L ,a J ,a T ,a S ,aG ]

[0217] Among them, Z a is a conditional single-dimensional matrix, a1 is the first condition value, a2 is the second condition value, a 11 is the eleven conditional values, a L and a J are ion normalized value and crystal normalized value, respectively, a T 、a S and a G are respectively the normalized value of temperature, humidity and light intensity, a i It refers to the i-th conditional value among the third conditional value, the fourth conditional value, the fifth conditional value, the sixth conditional value, the seventh conditional value, the eighth conditional value, the ninth conditional value and the tenth conditional value.

[0218] In detail, the resulting one-dimensional matrix is ​​as follows:

[0219] Z b =[b x ]

[0220] Among them, Z b is the resulting one-dimensional matrix, b x is the evaluation normalization value.

[0221] In detail, the calculation of the evaluation normalization value according to the standard evaluation value in the target data group includes:

[0222] Compare the standard evaluation value with the preset average T 90 value;

[0223] If the standard evaluation value is greater than the average T 90 value, the preset unit value is used as the evaluation normalization value, otherwise the preset zero value is used as the evaluation normalization value.

[0224] It should be explained that the unit value is 1 and the zero value is 0. Optionally, all T 90 The average value of the T 90 value.

[0225] It is understandable that the embodiment of the present invention is based on the average T 90 The value is used as the stability classification standard. The perovskite battery device with a standard evaluation value above the average value is considered to be stable, so it is assigned a value of 1 in the result single-dimensional matrix. Otherwise, the perovskite battery device corresponding to the standard evaluation value is considered to be unstable, so it is assigned a value of 0 in the result single-dimensional matrix.

[0226] S5. Summarize the single-dimensional feature matrix groups to obtain multiple single-dimensional feature matrix groups, and divide the multiple single-dimensional feature matrix groups to obtain a training matrix group set, a verification matrix group set, and a test matrix group set.

[0227] In detail, the division of the plurality of single-dimensional feature matrix groups to obtain a training matrix group set, a verification matrix group set, and a test matrix group set includes:

[0228] Confirm the number of single-dimensional feature matrix groups in the multiple single-dimensional feature matrix groups to obtain the total number of matrices;

[0229] The target number of training samples is calculated based on the total number of matrices. The calculation formula is as follows:

[0230] E A1 =80%×E A

[0231] Among them, E A1 is the target training quantity, E A is the total number of matrices;

[0232] Based on the target training quantity, a plurality of single-dimensional feature matrix groups are randomly extracted to obtain a training matrix group set, wherein the number of training matrix groups in the training matrix group set is the target training quantity;

[0233] The multiple single-dimensional feature matrix groups after random extraction are recorded as updated single-dimensional feature matrix group sets;

[0234] The updated single-dimensional feature matrix set is evenly divided to obtain a verification matrix set and a test matrix set.

[0235] Exemplarily, if the multiple single-dimensional feature matrix groups consist of 100 single-dimensional feature matrix groups, 80 single-dimensional feature matrix groups are randomly extracted from the 100 single-dimensional feature matrix groups and summarized to obtain a training matrix group set, and the remaining 20 unit feature matrix groups are used as the updated single-dimensional feature matrix group set, and then 10 unit feature matrix groups are randomly extracted from the updated single-dimensional feature matrix group set and summarized to obtain a verification matrix group set, and the remaining 10 unit feature matrix groups in the updated single-dimensional feature matrix group set after the 10 unit feature matrix groups are extracted are summarized to obtain a test matrix group set.

[0236] S6. Integrate the training matrix set to obtain a conditional multidimensional matrix and a result multidimensional matrix.

[0237] In detail, the training matrix set is integrated to obtain a conditional multidimensional matrix and a result multidimensional matrix, including:

[0238] The conditional multidimensional matrix is ​​obtained by using the training matrix set, where the conditional multidimensional matrix is ​​as follows:

[0239]

[0240] Among them, Z X1is a conditional multidimensional matrix, [Z a1 ] is the conditional single-dimensional matrix of the first training matrix group in the training matrix group set, [Z a2 ] is the conditional single-dimensional matrix of the second training matrix group in the training matrix group set, [Z ai ] is the conditional single-dimensional matrix of the i-th training matrix group in the training matrix group set, [Z an ] is the conditional single-dimensional matrix of the nth training matrix group in the training matrix group set, where n is the number of training matrix groups in the training matrix group set;

[0241] The result multidimensional matrix is ​​obtained by using the training matrix set, wherein the result multidimensional matrix is ​​as follows:

[0242]

[0243] Among them, Z Y is the resulting multidimensional matrix, [Z b1 ] is the result single-dimensional matrix of the first training matrix group in the training matrix group set, [Z b2 ] is the single-dimensional matrix of the second training matrix group in the training matrix group set, [Z bi ] is the result single-dimensional matrix of the i-th training matrix group in the training matrix group set, [Z bn ] is the resulting single-dimensional matrix of the nth training matrix group in the training matrix group set.

[0244] S7. Use the verification matrix set, the test matrix set, the conditional multidimensional matrix and the result multidimensional matrix to train and screen the pre-built model set to obtain the target learning model, wherein the model set includes: multiple machine learning models.

[0245] It should be understood that the embodiment of the present invention simultaneously trains multiple different types of machine learning models, and after the training is completed, the training effects of different machine learning models are tested by a test matrix set, thereby screening out the machine learning model with the best training effect. The machine learning model refers to DT, SVC, KNN, MLP, LGBM, XGB or CAT, DT is Decision Tree, SVC is Support Vector Classification, KNN is K-nearest Neighbor Classification, MLP is Multi-layer Perceptron, LGBM is Light Gradient Boosting Machine, XGB is Extreme Gradient Boosting, and CAT is Categorical Boosting.

[0246] In detail, the method of using the verification matrix set, the test matrix set, the conditional multidimensional matrix and the result multidimensional matrix to train and screen the pre-built model set to obtain the target learning model includes:

[0247] For each machine learning model in the model set, perform the following operations:

[0248] The machine learning model is trained using the conditional multidimensional matrix and the result multidimensional matrix to obtain a preliminary training model;

[0249] For each verification matrix group in the verification matrix group set, perform the following operations:

[0250] Inputting the conditional single-dimensional matrix in the verification matrix group into the preliminary training model to obtain a result output matrix, and identifying a result output value from the result output matrix;

[0251] Determine whether the evaluation normalized value corresponding to the result single-dimensional matrix in the verification matrix group is equal to the unit value;

[0252] If the evaluation normalized value is equal to the unit value, determining whether the result output value is equal to the unit value;

[0253] If the result output value is equal to the unit value, the result output value is regarded as a true positive value; otherwise, the result output value is regarded as a false negative value;

[0254] If the evaluation normalized value is not equal to the unit value, determining whether the result output value is equal to the unit value;

[0255] If the result output value is equal to the unit value, the result output value is regarded as a false positive value; otherwise, the result output value is regarded as a true negative value;

[0256] Summarize the true positive values, false negative values, false positive values, and true negative values ​​respectively to obtain the true positive value set, false negative value set, false positive value set, and true negative value set;

[0257] Determining a correct positive number based on the correct positive value set, wherein the correct positive number is the number of correct positive values ​​in the correct positive value set;

[0258] Determine the number of false negatives, the number of false positives, and the number of true negatives based on the false negative value set, the false positive value set, and the true negative value set, respectively;

[0259] The success degree of the Ith training is calculated based on the number of correct positives, the number of false negatives, the number of false positives, and the number of correct negatives, where the initial value of I is 2;

[0260] The Ith item in the pre-constructed training record sequence is replaced with the Ith training success degree to obtain an updated record sequence, wherein the first item in the training record sequence is zero. The training record sequence is as follows:

[0261] {P1,P2…P m}

[0262] Among them, P1 is the first item in the training record sequence, P2 is the second item in the training record sequence, and P m is the mth item of the training record sequence;

[0263] Using the 1st training success degree to confirm the 1-1th training success degree in the update record sequence, wherein the 1-1th training success degree is the item before the 1st training success degree in the update record sequence;

[0264] Subtract the I-1 training success degree from the I-th training success degree to obtain the success degree difference;

[0265] Comparing the success degree difference with a preset difference threshold and comparing the I-th training success degree with a preset success threshold;

[0266] If the success degree difference is less than or equal to the difference threshold and the I-th training success degree is greater than or equal to the success threshold, the preliminary training model is used as the final training model;

[0267] If the success degree difference is greater than the difference threshold or the I-th training success degree is less than the success threshold, the preliminary training model is used as the machine learning model, the updated record sequence is used as the training record sequence, J=I+1, J is used as I, and the step of training the machine learning model using the conditional multidimensional matrix and the result multidimensional matrix is ​​returned to, until the success degree difference is less than or equal to the difference threshold and the I-th training success degree is greater than or equal to the success threshold or I=m, the preliminary training model is used as the final training model;

[0268] Summarize the final training models to obtain multiple final training models;

[0269] The following operations are performed on each of the multiple final training models:

[0270] Get the final success rate based on the test matrix set and the final training model;

[0271] Summarizing the final success degrees to obtain multiple final success degrees, and determining a maximum success degree based on the multiple final success degrees, wherein the maximum success degree is the maximum final success degree among the multiple final success degrees;

[0272] The final training model corresponding to the maximum success degree is used as the target learning model.

[0273] It is understandable that the use of the conditional multidimensional matrix and the result multidimensional matrix to train the machine learning model to obtain a preliminary training model means: inputting the conditional multidimensional matrix into the machine learning model, the machine learning model will generate an output matrix based on the input conditional multidimensional matrix, and then calculate the difference between the output matrix and the result multidimensional matrix through the logarithmic loss function, and correct the machine learning model according to the difference and gradient boosting algorithm, and then input the conditional multidimensional matrix into the corrected machine learning model again. Through the continuous iteration of the above process, the machine learning model gradually learns the mapping relationship between the conditional multidimensional matrix and the result multidimensional matrix. The technology of calculating the difference between the output matrix and the result multidimensional matrix through the logarithmic loss function and the technology of correcting the machine learning model according to the difference and gradient boosting algorithm are both existing technologies and will not be repeated here.

[0274] Exemplarily, the conditional one-dimensional matrix in the verification matrix group is input into the preliminary training model, and the matrix output by the preliminary training model according to the conditional one-dimensional matrix is ​​used as the result output matrix.

[0275] It should be explained that identifying the result output value from the result output matrix means: using the numerical value in the result output matrix as the result output value, for example: if the result output matrix is ​​[1], then the result output value is 1.

[0276] Exemplarily, since the initial value of I is 2, the training success degree calculated for the first time based on the number of correct positives, the number of false negatives, the number of false positives, and the number of correct negatives is the second training success degree, and then the first training success degree is confirmed in the update record sequence based on the second training success degree. Since the second training success degree is located in the second item in the update record sequence, the first item in the update record sequence is the first training success degree. Then, the success degree difference is calculated based on the second training success degree and the first training success degree. If the success degree difference is greater than the difference threshold or the second training success degree is less than the success threshold, the preliminary training model is used as the machine learning model, the update record sequence is used as the training record sequence, and I is updated. The new value is 3, and the step of training the machine learning model using the conditional multidimensional matrix and the result multidimensional matrix is ​​returned. When the training success degree is calculated according to the number of correct positives, the number of false negatives, the number of false positives, and the number of correct negatives next time, since I=3, the training success degree is the third training success degree, and then the second training success degree is confirmed in the update record series according to the third training success degree, and then the success degree difference is calculated again, and so on. If the 50th training success degree is greater than or equal to the success threshold and the success degree difference corresponding to the 50th training success degree is greater than or equal to the success threshold is less than or equal to the difference threshold, the preliminary training model corresponding to the 50th training success degree is used as the final training model.

[0277] It should be understood that when the I-th training success degree is greater than or equal to the success threshold, it means that the preliminary training model after verification by the verification matrix set has been basically trained successfully, and when the success degree difference is less than or equal to the difference threshold, it means that the preliminary training model trained this time has no obvious improvement compared with the model trained last time, and the training can be terminated in advance to prevent overfitting, or when I=m, the I-th training success degree is still not greater than or equal to the success threshold. Since the training time is too long, the training process will also be stopped. Among them, the success threshold, difference threshold and m are all values ​​set manually by the trainer of the machine learning model. Optionally, the success threshold is 80%, the difference threshold is 1%, and m is 200.

[0278] In detail, the calculating of the I-th training success degree according to the number of correct positives, the number of false negatives, the number of false positives and the number of correct negatives includes:

[0279] The precision is calculated based on the number of true positives and false positives. The calculation formula is as follows:

[0280]

[0281] Among them, H Pre is the precision rate, R1 and F1 are the number of correct positives and the number of false positives, respectively;

[0282] The recall rate is calculated based on the number of true negatives and false negatives. The calculation formula is as follows:

[0283]

[0284] Among them, H Re is the recall rate, R2 and F2 are the number of true negatives and false negatives, respectively;

[0285] The success rate of the first training is calculated based on the precision, recall, number of true positives, number of false negatives, number of false positives, and number of true negatives. The calculation formula is as follows:

[0286]

[0287] in, is the success degree of the first training.

[0288] It should be explained that the Ith training success degree reflects the prediction accuracy of the preliminary training model corresponding to the Ith training success degree for the stability of the perovskite battery device. The greater the Ith training success degree, the higher the prediction accuracy of the preliminary training model corresponding to the Ith training success degree for the stability of the perovskite battery device.

[0289] S8. Receive a device evaluation instruction, confirm the device working environment based on the device evaluation instruction, and obtain current environmental data of the device working environment.

[0290] It should be explained that the device evaluation instruction is initiated by the experimenter of the perovskite battery device. For example, Xiao Zhang is an experimenter studying perovskite battery devices. Xiao Zhang now needs to predict the stability of the performance of the perovskite battery device in a specific environment, so he initiates the device evaluation instruction. If Xiao Zhang wants to predict the stability of a new type of perovskite battery device when it is applied at a certain location in reality, the environment of the location is the device working environment, and the current environmental data includes: current temperature value, current humidity value and current light intensity value. By collecting multiple temperatures, multiple humidity and multiple light intensities of the device working environment over a period of time, the average value of the multiple temperatures, the average value of the multiple humidity and the average value of the multiple light intensities during the period are used as the current temperature value, the current humidity value and the current light intensity value, respectively, thereby obtaining the current environmental data of the device working environment.

[0291] S9. Obtain current process data, obtain a current integration matrix based on the current process data and current environmental data, input the current integration matrix into the target learning model to obtain a stability classification matrix, and complete the stability classification of the perovskite battery device based on the stability classification matrix.

[0292] It should be explained that the current process data include: current coating process, current precursor solvent, current antisolvent, current anti-solvent time, current annealing temperature, current annealing time, current annealing times, current ETL material, current HTL material, current A-site material, current A-site chemical formula, current B-site material, current B-site chemical formula, current X-site material and current X-site chemical formula, and the current process data are prepared and recorded in advance by experimenters studying perovskite battery devices when designing perovskite battery devices.

[0293] It should be understood that the method for obtaining the current integration matrix based on the current process data and the current environmental data is the same as the method for obtaining the conditional unit matrix using the preparation process data and the environmental test data, and will not be repeated here.

[0294] Exemplarily, the current integration matrix is ​​input into the target learning model, and the matrix output by the target learning model based on the current integration matrix is ​​used as the stability classification matrix. If the value in the stability classification matrix is ​​1, it means that the perovskite battery device corresponding to the stability classification matrix has stable performance when working in the device working environment. If the value in the stability classification matrix is ​​not 1, it means that the perovskite battery device corresponding to the stability classification matrix has unstable performance when working in the device working environment, thereby completing the stability classification of the perovskite battery device.

[0295] The present invention aims to solve the problem described in the background technology. The present invention obtains multiple perovskite device data sets, wherein the perovskite device data sets include: preparation process data, environmental test data and performance evaluation data, the environmental test data includes one or more of temperature value, humidity value and light intensity value, the performance evaluation data includes T 95 Value, T 90 Value and T 80One or more of the values, it can be seen that the embodiment of the present invention obtains multiple perovskite device data groups from the literature, thereby providing a huge data set for subsequent model training, thereby improving the accuracy of the prediction results after training, and then performing the following operations on each of the multiple perovskite device data groups: multi-dimensionally correlating and integrating the preparation process data in the perovskite device data group to obtain integrated process data. It can be seen that the embodiment of the present invention summarizes and integrates the multi-dimensional data in the preparation process data into more streamlined values ​​by multi-dimensionally correlating and integrating the preparation process data in the perovskite device data group. At the same time, the stability factor and the crystal growth degree can also more accurately reflect the performance of the perovskite crystal recorded in the literature, thereby reducing the subsequent training model. The total amount of training data at the time of implementation improves the efficiency of subsequent model training and the accuracy of model prediction, performs a gap filling and standardization operation on the environmental test data and performance evaluation data in the perovskite device data group, and obtains standard test data and standard evaluation values. It can be seen that the embodiment of the present invention performs a gap filling and standardization operation on the environmental test data and performance evaluation data in the perovskite device data group, thereby supplementing and standardizing the missing data, making the input data for subsequent model training more unified, summarizing and integrating process data, standard test data and standard evaluation values ​​to obtain a target data group, performing feature recognition on the target data group to obtain a single-dimensional feature matrix group, it can be seen that the embodiment of the present invention performs feature recognition on the target data group, reduces the dimensionality of high-dimensional data, and obtains a single-dimensional feature matrix group. dimensional feature matrix group, so that the matrix can be directly input into the model for training in the future, and the single-dimensional feature matrix group is summarized to obtain multiple single-dimensional feature matrix groups. The multiple single-dimensional feature matrix groups are divided to obtain training matrix group sets, verification matrix group sets and test matrix group sets. The training matrix group sets are integrated to obtain conditional multidimensional matrices and result multidimensional matrices. It can be seen that the embodiment of the present invention facilitates the subsequent machine learning model to learn the mapping relationship between the conditional multidimensional matrix and the result multidimensional matrix by classifying and integrating the training matrix group sets into conditional multidimensional matrices and result multidimensional matrices. The pre-constructed model set is trained and screened using the verification matrix group set, the test matrix group set, the conditional multidimensional matrix and the result multidimensional matrix to obtain the target learning model, wherein the model set includes: multiple machine learning models. It can be seen that the embodiment of the present invention improves the efficiency and reliability of the target learning model in evaluating the stability of the perovskite battery device by simultaneously training and screening multiple machine learning models, and taking the machine learning model with the best prediction effect after training as the target learning model, receives a device evaluation instruction, confirms the device working environment based on the device evaluation instruction, obtains the current environmental data of the device working environment, obtains the current process data, obtains the current integration matrix based on the current process data and the current environmental data, inputs the current integration matrix into the target learning model, obtains the stability classification matrix, and completes the stability classification of the perovskite battery device based on the stability classification matrix. It can be seen that the embodiment of the present invention obtains the current environmental data of the device working environment,The stability of the perovskite battery device is predicted according to the environment in which the perovskite battery device is located, thereby improving the reliability of the stability evaluation of the perovskite battery device. Therefore, the present invention can improve the efficiency and reliability of the stability evaluation of the perovskite battery device.

[0296] like Figure 2 , which is a functional module diagram of a perovskite battery device stability classification system based on machine learning provided by one embodiment of the present invention.

[0297] The machine learning-based perovskite battery device stability classification system 100 of the present invention can be installed in an electronic device 1. Depending on the functions implemented, the machine learning-based perovskite battery device stability classification system 100 can include a device data acquisition module 101, a data feature extraction module 102, a classification model training module 103, and a device stability prediction module 104. The module of the present invention can also be referred to as a unit, which refers to a series of computer program segments that can be executed by an electronic device processor and can perform fixed functions, and is stored in the memory of the electronic device.

[0298] The device data acquisition module 101 is used to acquire multiple perovskite device data groups, wherein the perovskite device data groups include: preparation process data, environmental test data and performance evaluation data, the environmental test data includes one or more of temperature value, humidity value and light intensity value, the performance evaluation data includes T 95 Value, T 90 Value and T 80 One or more of the values;

[0299] The data feature extraction module 102 is configured to perform the following operations on each of the multiple perovskite device data groups: performing multi-dimensional correlation integration on the preparation process data in the perovskite device data group to obtain integrated process data; performing a gap filling and standardization operation on the environmental test data and performance evaluation data in the perovskite device data group to obtain standard test data and standard evaluation values; summarizing and integrating the process data, standard test data, and standard evaluation values ​​to obtain a target data group; and performing feature recognition on the target data group to obtain a single-dimensional feature matrix group.

[0300] The classification model training module 103 is used to summarize the single-dimensional feature matrix group to obtain multiple single-dimensional feature matrix groups, divide the multiple single-dimensional feature matrix groups to obtain training matrix group sets, verification matrix group sets and test matrix group sets, integrate the training matrix group sets to obtain conditional multidimensional matrices and result multidimensional matrices, and use the verification matrix group sets, test matrix group sets, conditional multidimensional matrices and result multidimensional matrices to train and screen a pre-constructed model set to obtain a target learning model, wherein the model set includes: multiple machine learning models;

[0301] The device stability prediction module 104 is used to receive device evaluation instructions, confirm the device working environment based on the device evaluation instructions, obtain current environmental data of the device working environment, obtain current process data, obtain the current integration matrix based on the current process data and the current environmental data, input the current integration matrix into the target learning model, obtain the stability classification matrix, and complete the stability classification of the perovskite battery device based on the stability classification matrix.

[0302] In detail, each module in the perovskite battery device stability classification system 100 based on machine learning in the embodiment of the present invention adopts the same method as above when in use. Figure 1 The same technical means are used in the stability classification method of perovskite battery devices based on machine learning, and can produce the same technical effects, so they will not be repeated here.

Claims

1. A perovskite battery device stability classification method based on machine learning, characterized in that: The method comprises: Acquire multiple perovskite device data sets, wherein the perovskite device data sets include: preparation process data, environmental test data and performance evaluation data, the environmental test data includes one or more of temperature value, humidity value and light intensity value, the performance evaluation data includes T 95 Value, T 90 Value and T 80 One or more of the values; The following operation is performed on each of the plurality of perovskite device data sets: Perform multi-dimensional correlation integration on the preparation process data in the perovskite device data set to obtain integrated process data; Performing a gap-filling and standardization operation on the environmental test data and performance evaluation data in the perovskite device data set to obtain standard test data and standard evaluation values; Summarize and integrate process data, standard test data and standard evaluation values ​​to obtain the target data set; Perform feature recognition on the target data group to obtain a single-dimensional feature matrix group; Summarizing the single-dimensional feature matrix groups to obtain multiple single-dimensional feature matrix groups, and dividing the multiple single-dimensional feature matrix groups to obtain a training matrix group set, a verification matrix group set, and a test matrix group set; Integrate the training matrix set to obtain the conditional multidimensional matrix and the result multidimensional matrix; The pre-built model set is trained and screened using the verification matrix set, the test matrix set, the condition multidimensional matrix, and the result multidimensional matrix to obtain a target learning model, wherein the model set includes: multiple machine learning models; receiving a device evaluation instruction, determining a device operating environment based on the device evaluation instruction, and obtaining current environmental data of the device operating environment; Obtain current process data, obtain the current integration matrix based on the current process data and current environmental data, input the current integration matrix into the target learning model to obtain a stability classification matrix, and complete the stability classification of the perovskite battery device based on the stability classification matrix.

2. The method for classifying the stability of a perovskite battery device based on machine learning according to claim 1, wherein: The preparation process data includes: coating process, precursor solvent, antisolvent, anti-solvent time, annealing temperature, annealing time, annealing times, ETL material, HTL material, A-site material, A-site chemical formula, B-site material, B-site chemical formula, X-site material and X-site chemical formula, wherein the coating process includes: spin coating, blade coating, spray coating or printing, the precursor solvent includes: dimethylformamide, dimethyl sulfoxide, butyrolactone or other precursors, the antisolvent includes: chlorobenzene, toluene, dichlorobenzene or other antisolvents, and the ETL material includes: other electronic materials, SnO2, PCBM and C 60 One or more of the HTL materials include: other hole materials, Spiro materials, PTAA and NiO x The A-site material includes one or more of formamidinium ions, methylamine ions, cesium ions, and rubidium ions; the B-site material includes one or more of lead ions and tin ions; and the X-site material includes one or more of iodide ions, bromide ions, and chloride ions.

3. The method for classifying the stability of a perovskite battery device based on machine learning according to claim 2, wherein: The multi-dimensional correlation integration of the preparation process data in the perovskite device data set to obtain integrated process data includes: The A-site ion radius is calculated based on the A-site chemical formula in the preparation process data. The calculation formula is as follows: Among them, r A is the A-site ionic radius, r FA 、r MA 、r Cs and r Rb are respectively the preset formamidine ion radius, the preset methylamine ion radius, the preset cesium ion radius and the preset rubidium ion radius, μ FA 、μ MA 、μ Cs and μ Rb are the proportion of formamidine ion in the A-site chemical formula, the proportion of methylamine ion in the A-site chemical formula, the proportion of cesium ion in the A-site chemical formula, and the proportion of rubidium ion in the A-site chemical formula; The B-site ion radius is calculated based on the B-site chemical formula in the preparation process data. The calculation formula is as follows: Among them, r B is the B-site ion radius, r Pb and r Sn are the preset lead ion radius and the preset tin ion radius, μ Pb and μ Sn are the proportion of lead ions in the B-site chemical formula and the proportion of tin ions in the B-site chemical formula respectively; The X-position ion radius is calculated based on the X-position chemical formula in the preparation process data. The calculation formula is as follows: Among them, r X is the X-site ionic radius, r I 、r Br and r Cl are the preset iodine ion radius, the preset bromide ion radius and the preset chloride ion radius, μ I 、μ Br and μ Cl are the proportion of iodide ion in the chemical formula at position X, the proportion of bromide ion in the chemical formula at position X, and the proportion of chloride ion in the chemical formula at position X; The stability factor is calculated based on the A-site ionic radius, the B-site ionic radius, and the X-site ionic radius. The calculation formula is as follows: in, is the stability factor; The crystal growth rate is calculated based on the anti-dissolution time, annealing temperature, annealing time and annealing times. The calculation formula is as follows: Among them, ω Crystal is the crystal growth rate, t k and t H are the anti-dissolution time and annealing time, T H and T0 are annealing temperature and preset standard temperature respectively, N H is the number of annealing times, ln is the natural logarithm, and tanh is the hyperbolic tangent function; The coating process, precursor solvent, antisolvent, ETL material, HTL material, stability factor and crystal growth degree are summarized to obtain integrated process data.

4. The method for classifying the stability of a perovskite battery device based on machine learning according to claim 3, wherein: The performing of a gap filling and standardization operation on the environmental test data and performance evaluation data in the perovskite device data set to obtain standard test data and standard evaluation values ​​includes: Determine whether there is a temperature value in the environmental test data, if so, use the temperature value as the target temperature value, otherwise use a preset default temperature value as the target temperature value; Determine whether there is a humidity value in the environmental test data, if there is a humidity value in the environmental test data, use the humidity value as the target humidity value, otherwise use a preset default humidity value as the target humidity value; Determine whether there is a light intensity value in the environmental test data, and if so, use the light intensity value as the target light intensity value; otherwise, use a preset default light intensity value as the target light intensity value; Summarize the target temperature value, target humidity value and target light intensity value to obtain standard test data; Determine whether T exists in the performance evaluation data 90 If T exists in the performance evaluation data 90 value, then T 90 The value is used as the standard evaluation value, otherwise it is judged whether T exists in the performance evaluation data. 95 value; If there is T in the performance evaluation data 95 value, then according to T 95 The calculation formula is as follows: Among them, τ x is the standard evaluation value, T 95 value; If T does not exist in the performance evaluation data 95 value, then according to T 80 The calculation formula is as follows: in, T 80 value.

5. The method for classifying the stability of a perovskite battery device based on machine learning according to claim 4, wherein: The feature recognition is performed on the target data group to obtain a single-dimensional feature matrix group, including: Determining a first condition value based on the coating process of the integrated process data in the target data set, determining a second condition value based on the precursor solvent of the integrated process data in the target data set, and determining a third condition value based on the antisolvent in the target data set; Determining a fourth condition value, a fifth condition value, a sixth condition value, and a seventh condition value based on the ETL material of the integrated process data in the target data set; The eighth condition value, the ninth condition value, the tenth condition value, and the eleventh condition value are determined based on the HTL material of the integrated process data in the target data set; The ion normalization value is calculated based on the stability factor of the integrated process data in the target data set, and the calculation formula is as follows: Among them, δ L is the ion normalized value, e is the natural constant; Obtaining a crystallization normalization value based on the crystallization growth degree of the integrated process data in the target data set; Obtaining a temperature normalized value, a humidity normalized value, and a light intensity normalized value based on the target temperature value, the target humidity value, and the target light intensity value of the standard test data in the target data group, respectively; Summarize the first condition value, the second condition value, the third condition value, the fourth condition value, the fifth condition value, the sixth condition value, the seventh condition value, the eighth condition value, the ninth condition value, the tenth condition value, the eleventh condition value, the ion normalized value, the crystallization normalized value, the temperature normalized value, the humidity normalized value, and the light intensity normalized value to obtain a condition feature group; Calculate an evaluation normalization value based on the standard evaluation value in the target data set; The conditional feature group and the evaluation normalized value are used to construct the conditional one-dimensional matrix and the result one-dimensional matrix respectively, and the conditional one-dimensional matrix and the result one-dimensional matrix are summarized to obtain the one-dimensional feature matrix group.

6. The method for classifying the stability of a perovskite battery device based on machine learning according to claim 5, wherein: The step of calculating the evaluation normalized value based on the standard evaluation value in the target data set includes: Compare the standard evaluation value with the preset average T 90 value; If the standard evaluation value is greater than the average T 90 value, the preset unit value is used as the evaluation normalization value, otherwise the preset zero value is used as the evaluation normalization value.

7. The method for classifying the stability of a perovskite battery device based on machine learning according to claim 6, wherein: The training matrix set is integrated to obtain a conditional multidimensional matrix and a result multidimensional matrix, including: The conditional multidimensional matrix is ​​obtained by using the training matrix set, where the conditional multidimensional matrix is ​​as follows: Among them, Z X1 is a conditional multidimensional matrix, [Z a1 ] is the conditional single-dimensional matrix of the first training matrix group in the training matrix group set, [Z a2 ] is the conditional single-dimensional matrix of the second training matrix group in the training matrix group set, [Z ai ] is the conditional single-dimensional matrix of the i-th training matrix group in the training matrix group set, [Z an ] is the conditional single-dimensional matrix of the nth training matrix group in the training matrix group set, where n is the number of training matrix groups in the training matrix group set; The result multidimensional matrix is ​​obtained by using the training matrix set, wherein the result multidimensional matrix is ​​as follows: Among them, Z Y is the resulting multidimensional matrix, [Z b1 ] is the result single-dimensional matrix of the first training matrix group in the training matrix group set, [Z b2 ] is the single-dimensional matrix of the second training matrix group in the training matrix group set, [Z bi ] is the result single-dimensional matrix of the i-th training matrix group in the training matrix group set, [Z bn ] is the resulting single-dimensional matrix of the nth training matrix group in the training matrix group set.

8. The method for classifying the stability of a perovskite battery device based on machine learning according to claim 7, wherein: The method of using the verification matrix set, the test matrix set, the conditional multidimensional matrix and the result multidimensional matrix to train and screen the pre-built model set to obtain the target learning model includes: For each machine learning model in the model set, perform the following operations: The machine learning model is trained using the conditional multidimensional matrix and the result multidimensional matrix to obtain a preliminary training model; For each verification matrix group in the verification matrix group set, perform the following operations: Inputting the conditional single-dimensional matrix in the verification matrix group into the preliminary training model to obtain a result output matrix, and identifying a result output value from the result output matrix; Determine whether the evaluation normalized value corresponding to the result single-dimensional matrix in the verification matrix group is equal to the unit value; If the evaluation normalized value is equal to the unit value, determining whether the result output value is equal to the unit value; If the result output value is equal to the unit value, the result output value is regarded as a true positive value; otherwise, the result output value is regarded as a false negative value; If the evaluation normalized value is not equal to the unit value, determining whether the result output value is equal to the unit value; If the result output value is equal to the unit value, the result output value is regarded as a false positive value; otherwise, the result output value is regarded as a true negative value; Summarize the true positive values, false negative values, false positive values, and true negative values ​​respectively to obtain the true positive value set, false negative value set, false positive value set, and true negative value set; Determining a correct positive number based on the correct positive value set, wherein the correct positive number is the number of correct positive values ​​in the correct positive value set; Determine the number of false negatives, the number of false positives, and the number of true negatives based on the false negative value set, the false positive value set, and the true negative value set, respectively; The success degree of the training step I is calculated based on the number of correct positives, the number of false negatives, the number of false positives, and the number of correct negatives, where the initial value of I is 2; The Ith item in the pre-constructed training record sequence is replaced with the Ith training success degree to obtain an updated record sequence, where the first item in the training record sequence is zero. The training record sequence is as follows: {P1,P2…P m } Among them, P1 is the first item in the training record sequence, P2 is the second item in the training record sequence, and P m is the mth item of the training record sequence; Using the 1st training success degree to confirm the 1-1th training success degree in the update record sequence, wherein the 1-1th training success degree is the item before the 1st training success degree in the update record sequence; Subtract the I-1 training success degree from the I-th training success degree to obtain the success degree difference; Comparing the success degree difference with a preset difference threshold and comparing the I-th training success degree with a preset success threshold; If the success degree difference is less than or equal to the difference threshold and the I-th training success degree is greater than or equal to the success threshold, the preliminary training model is used as the final training model; If the success degree difference is greater than the difference threshold or the I-th training success degree is less than the success threshold, the preliminary training model is used as the machine learning model, the updated record sequence is used as the training record sequence, J=I+1, J is used as I, and the step of training the machine learning model using the conditional multidimensional matrix and the result multidimensional matrix is ​​returned to, until the success degree difference is less than or equal to the difference threshold and the I-th training success degree is greater than or equal to the success threshold or I=m, the preliminary training model is used as the final training model; Aggregate the final training models to obtain multiple final training models; The following operations are performed on each of the multiple final training models: Get the final success rate based on the test matrix set and the final training model; Summarizing the final success degrees to obtain multiple final success degrees, and determining a maximum success degree based on the multiple final success degrees, wherein the maximum success degree is the maximum final success degree among the multiple final success degrees; The final training model corresponding to the maximum success degree is used as the target learning model.

9. The method for classifying the stability of a perovskite battery device based on machine learning according to claim 8, wherein: The calculating of the I-th training success degree according to the number of correct positives, the number of false negatives, the number of false positives and the number of correct negatives comprises: The precision is calculated based on the number of true positives and false positives. The calculation formula is as follows: Among them, H Pre is the precision rate, R1 and F1 are the number of correct positives and the number of false positives, respectively; The recall rate is calculated based on the number of true negatives and false negatives. The calculation formula is as follows: Among them, H Re is the recall rate, R2 and F2 are the number of true negatives and false negatives, respectively; The success rate of the first training is calculated based on the precision, recall, number of true positives, number of false negatives, number of false positives, and number of true negatives. The calculation formula is as follows: in, is the success degree of the first training.

10. A perovskite battery device stability classification system based on machine learning, characterized in that: The system comprises: The device data acquisition module is used to acquire multiple perovskite device data groups, wherein the perovskite device data groups include: preparation process data, environmental test data and performance evaluation data, the environmental test data includes one or more of temperature value, humidity value and light intensity value, the performance evaluation data includes T 95 Value, T 90 Value and T 80 One or more of the values; a data feature extraction module for performing the following operations on each of the multiple perovskite device data groups: performing multi-dimensional correlation integration on the preparation process data in the perovskite device data group to obtain integrated process data; performing a gap filling and standardization operation on the environmental test data and performance evaluation data in the perovskite device data group to obtain standard test data and standard evaluation values; summarizing and integrating the process data, standard test data, and standard evaluation values ​​to obtain a target data group; and performing feature recognition on the target data group to obtain a single-dimensional feature matrix group; A classification model training module is used to summarize a single-dimensional feature matrix group to obtain multiple single-dimensional feature matrix groups, divide the multiple single-dimensional feature matrix groups to obtain a training matrix group set, a verification matrix group set, and a test matrix group set, integrate the training matrix group set to obtain a conditional multidimensional matrix and a result multidimensional matrix, and use the verification matrix group set, the test matrix group set, the conditional multidimensional matrix, and the result multidimensional matrix to train and screen a pre-constructed model set to obtain a target learning model, wherein the model set includes: multiple machine learning models; The device stability prediction module is used to receive device evaluation instructions, confirm the device working environment based on the device evaluation instructions, obtain the current environmental data of the device working environment, obtain the current process data, obtain the current integration matrix based on the current process data and the current environmental data, input the current integration matrix into the target learning model, obtain the stability classification matrix, and complete the stability classification of the perovskite battery device based on the stability classification matrix.

Images