Processing method and device for predicting photoelectric characteristics of solvent molecules

Abstract

The embodiment of the invention relates to a processing method and device for predicting photoelectric characteristics of solvent molecules. The method comprises the following steps: constructing a deep learning model for predicting various photoelectric characteristics of the solvent molecules, and recording the deep learning model as a corresponding multi-task prediction model; constructing a model training data set and recording the model training data set as a corresponding first data set; training the multi-task prediction model based on the first data set; after model training is finished, solvent optimization processing is conducted on batch optimization tasks input by the user through the multi-task prediction model, and a processing result is fed back to the current user. The method can reduce the analysis complexity, shorten the analysis time, improve the analysis efficiency and improve the prediction accuracy.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, in particular to a processing method and device for predicting photoelectric properties of solvent molecules. BACKGROUND

[0002] The photoelectric material solution is a complex system composed of solvent molecules and solute molecules, and the photoelectric properties of the solution are jointly determined by the intermolecular interaction between the two. Among them, the solute molecules are the direct executors of photoelectric conversion, and the solvent molecules affect the photoelectric properties of the solute molecules through dissolution, conformation regulation and spectral regulation. Under the condition that the solute molecules are fixed, it is necessary to analyze the photoelectric properties of different solvent molecules for the solvent optimization engineering of photoelectric materials. The commonly used photoelectric properties of solvent molecules mentioned here are: the peak point of the emission spectrum of the solvent molecule (referred to as the emission peak), the residence time of the solvent molecule in the excited state (referred to as the molecular lifetime), the photoelectric conversion efficiency of the solvent molecule, the peak point of the absorption spectrum of the solvent molecule (referred to as the absorption peak), and the half-peak width of the absorption spectrum of the solvent molecule (referred to as the absorption half-peak width).

[0003] At present, the technical means for analyzing the above-mentioned several types of photoelectric properties of solvent molecules are mostly based on quantum chemical calculation methods (such as density functional theory calculation method, etc.). The biggest problem of this conventional analysis method is high computational complexity, long calculation time and low calculation efficiency, which is difficult to meet the high-throughput calculation requirements. With the application of artificial intelligence models in the field of photoelectric materials, by constructing a target property prediction model, the purpose of reducing analysis complexity, shortening analysis time and improving analysis efficiency can be achieved. However, the conventional model only predicts the self-structure of the solvent molecule as a reference, and does not consider the influence of solute structure, solute properties and solute-solvent interaction properties on solvent properties, which makes the prediction accuracy of the conventional model need to be further improved. SUMMARY

[0004] The present application aims at the defects of the prior art, and provides a processing method and device for predicting photoelectric properties of solvent molecules, electronic equipment and computer readable storage medium. The present application takes five types of solute / solvent properties which can be quickly obtained by simple calculation as prior properties (total number of solute aromatic rings, solute interatomic interaction energy, solvent dipole moment, solvent polarity index, solute-solvent orbital overlap integral calculated by the GFN2-xTB method), takes five types of solvent molecular photoelectric properties (emission peak, molecular lifetime, photoelectric conversion efficiency, absorption peak, absorption half-width) which need to be obtained by long-time calculation as prediction targets, constructs a multi-task prediction model which can predict the five types of prediction targets according to the solute molecular structure, solvent molecular structure and five types of prior properties, and provides two types of optional structures for the multi-task prediction model, constructs a model training data set to train the model, and assists the user in processing the high-throughput solvent optimization task by using the multi-task prediction model after the model training is completed. The present application can not only reduce the analysis complexity, shorten the analysis time and improve the analysis efficiency, but also improve the prediction accuracy, and improve the flexibility and generalization of the prediction model.

[0005] To achieve the above object, the first aspect of the embodiment of the present application provides a processing method for predicting photoelectric properties of solvent molecules, which comprises:

[0006] A deep learning model for feature fusion of solute molecular structure, solvent molecular structure and solute-solvent system prior properties and prediction of multiple types of photoelectric properties of solvent molecules is constructed, which is called a corresponding multi-task prediction model; wherein the multi-task prediction model is used for multi-task prediction of photoelectric properties of solvent molecules according to model input solute molecular structure M 1 , solvent molecular structure M 2 and prior property vector X and outputs corresponding solvent property prediction vector Y; the solute molecular structure and the solvent molecular structure are both three-dimensional molecular structures; the solute-solvent system prior properties include total number of aromatic rings of solute molecules, interatomic interaction energy of solute molecules, dipole moment of solvent molecules, polarity index of solvent molecules, and orbital overlap integral of solute molecules and solvent molecules; the multiple types of photoelectric properties include emission peak, molecular lifetime, photoelectric conversion efficiency, absorption peak and absorption half-width; the solute molecular structure M 1 is composed of multiple atomic features , 1≤index i≤N 1 , N 1 is the total number of atoms of the solute molecule; the atomic feature includes atomic element type and atomic three-dimensional coordinates; the solvent molecular structure M 2 is composed of multiple atomic features , 1≤index j≤N 2, N 2 is the total number of atoms of the solvent molecule; the atomic features include the atomic element type, the atomic three-dimensional coordinates; the prior characteristic vector X includes the total number of solute aromatic rings x1, the solute interatomic interaction energy x2, the solvent dipole moment x3, the solvent polarity index x4, the solute-solvent orbital overlap integral x5; the solvent characteristic prediction vector Y includes the emission peak y1, the molecular lifetime y2, the photoelectric conversion efficiency y3, the absorption peak y4, the absorption half-peak width y5;

[0007] The model training data set is denoted as a corresponding first data set; and the multi-task prediction model is trained based on the first data set;

[0008] After the model training is completed, the multi-task prediction model is used to perform solvent optimization processing on a batch of preferred tasks input by a user and feedback the processing results to the current user; the batch of preferred tasks includes a first solute molecule sequence, a first solvent molecule sequence set, and a five-category index threshold range; the first solvent molecule sequence set includes a plurality of first solvent molecule sequences; the first solvent molecule sequence and each of the first solvent molecule sequences are each an SMILES sequence; and the five-category index threshold range includes an emission peak threshold range, a molecular lifetime threshold range, a photoelectric conversion efficiency threshold range, an absorption peak threshold range, and an absorption half-peak width threshold range.

[0009] Preferably, the first data set includes a plurality of first data records; the first data record includes a first solute molecule structure, a first solvent molecule structure, a first prior characteristic vector, and a first solvent characteristic label vector; the data structures of the first solute molecule structure, the first solvent molecule structure, the first prior characteristic vector, and the first solvent characteristic label vector are consistent with the corresponding solute molecule structure M 1 , the solvent molecule structure M 2 , the prior characteristic vector X, and the solvent characteristic prediction vector Y.

[0010] Preferably, the first model input end of the multi-task prediction model is used to receive the solute molecule structure M 1 , the second model input end is used to receive the solvent molecule structure M 2 , the third model input end is used to receive the prior characteristic vector X, and the model output end is used to output the corresponding solvent characteristic prediction vector Y.

[0011] The model structure of the multi-task prediction model has two optional structures, namely a first optional structure and a second optional structure; the model components of the first optional structure include a first structure feature encoder, a solute feature extraction module, a second structure feature encoder, a solvent feature extraction module, a molecular feature fusion module, a fused feature encoder, a prior feature extraction module, a prior feature fusion module, a first regression prediction model, a second regression prediction model, a third regression prediction model, a fourth regression prediction model, a fifth regression prediction model, and a prediction output layer; the model components of the second optional structure include the first structure feature encoder, the solute feature extraction module, the second structure feature encoder, the solvent feature extraction module, the molecular feature fusion module, the fused feature encoder, the prior feature extraction module, the prior feature fusion module, and a fused feature decoder.

[0012] The shared components of the first optional structure and the second optional structure include the first structure feature encoder, the solute feature extraction module, the second structure feature encoder, the solvent feature extraction module, the molecular feature fusion module, the fused feature encoder, the prior feature extraction module, and the prior feature fusion module.

[0013] The personalized components of the first optional structure include the first regression prediction model, the second regression prediction model, the third regression prediction model, the fourth regression prediction model, the fifth regression prediction model, and the prediction output layer.

[0014] The personalized components of the second optional structure include the fused feature decoder.

[0015] Further, the component connection relationship of the first optional structure is: the input end of the first structure feature encoder is connected with the first model input end, and the output end is connected with the input end of the solute feature extraction module; the output end of the solute feature extraction module is connected with the first input end of the molecular feature fusion module; the input end of the second structure feature encoder is connected with the second model input end, and the output end is connected with the input end of the solvent feature extraction module; the output end of the solvent feature extraction module is connected with the second input end of the molecular feature fusion module; the output end of the molecular feature fusion module is connected with the input end of the fusion feature encoder; the output end of the fusion feature encoder is connected with the first input end of the prior feature fusion module; the input end of the prior feature extraction module is connected with the third model input end, and the output end is connected with the second input end of the prior feature fusion module; the output end of the prior feature fusion module is connected with the input end of the first, second, third, fourth and fifth regression prediction models respectively; the input end of the prediction output layer is connected with the output end of the first, second, third, fourth and fifth regression prediction models respectively, and the output end is connected with the model output end.

[0016] The component connection relationship of the second optional structure is: the input end of the first structure feature encoder is connected with the first model input end, and the output end is connected with the input end of the solute feature extraction module; the output end of the solute feature extraction module is connected with the first input end of the molecular feature fusion module; the input end of the second structure feature encoder is connected with the second model input end, and the output end is connected with the input end of the solvent feature extraction module; the output end of the solvent feature extraction module is connected with the second input end of the molecular feature fusion module; the output end of the molecular feature fusion module is connected with the input end of the fusion feature encoder; the output end of the fusion feature encoder is connected with the first input end of the prior feature fusion module; the input end of the prior feature extraction module is connected with the third model input end, and the output end is connected with the second input end of the prior feature fusion module; the output end of the prior feature fusion module is connected with the input end of the fusion feature decoder; the output end of the fusion feature decoder is connected with the model output end.

[0017] Further, the component function of the shared component of the first optional structure and the second optional structure is:

[0018] The first structure feature encoder is realized based on a Uni-Mol model; the first structure feature encoder is used for performing high-dimensional feature coding processing on the solute molecular structure M 1 at an atomic level to obtain a corresponding feature tensor E1, and transmitting the feature tensor E1 to the solute feature extraction module; the shape of the feature tensor E1 is N 1 ×CA , C A is a preset atomic feature dimension;

[0019] The solute feature extraction module is configured to perform pooling processing on each feature channel of the feature tensor E1 according to a preset solute feature pooling rule to obtain a pooled feature vector P1 with a vector length of C A , and perform molecular feature space mapping on the pooled feature vector P1 based on an embedded MLP model to obtain a molecular feature vector H1 with a vector length of C B , and send the molecular feature vector H1 to the molecular feature fusion module; the solute feature pooling rule includes maximum pooling, average pooling, and attention pooling; C B is a preset molecular feature dimension;

[0020] The second structure feature encoder is implemented based on a Uni-Mol model; the second structure feature encoder is configured to perform atomic-level high-dimensional feature encoding processing on the solvent molecule structure M 2 to obtain a corresponding feature tensor E2, and send the feature tensor E2 to the solvent feature extraction module; the shape of the feature tensor E2 is N 2 × C A ;

[0021] The solvent feature extraction module is configured to perform pooling processing on each feature channel of the feature tensor E2 according to a preset solvent feature pooling rule to obtain a pooled feature vector P2 with a vector length of C A , and perform molecular feature space mapping on the pooled feature vector P2 based on an embedded MLP model to obtain a molecular feature vector H2 with a vector length of C B , and send the molecular feature vector H2 to the molecular feature fusion module; the solvent feature pooling rule includes maximum pooling, average pooling, and attention pooling;

[0022] The molecular feature fusion module is configured to sequentially sort the molecular feature vector H1 and the molecular feature vector H2 to form a corresponding feature vector sequence H3, and send the feature vector sequence H3 to the fusion feature encoder;

[0023] The fusion feature encoder is implemented based on an encoder model of a Transformer architecture; the fusion feature encoder is configured to perform feature encoding processing on the feature vector sequence H3 through a multi-head self-attention encoding mechanism to obtain a corresponding encoding feature sequence H4, and send the encoding feature sequence H4 to the prior feature fusion module; the encoding feature sequence H4 is sequentially sorted by two feature encoding vectors B with a vector length of C ; the feature encoding vector corresponds to the molecular feature vector H1 and H2 one by one;

[0024] The prior feature extraction module is configured to use an embedded MLP model to perform prior feature space mapping on the prior characteristic vector X to obtain a prior feature vector H5 with a length of C C ; C C is a preset prior feature dimension;

[0025] The prior feature fusion module is configured to extract a corresponding feature encoding vector H4 from the encoding feature sequence H4 , and perform vector splicing on the feature encoding vector H4 and the prior feature vector H5 to obtain a splicing feature vector H6 with a length of C D ; in the first type of optional structure, the splicing feature vector H6 is sent to the first regression prediction model, the second regression prediction model, the third regression prediction model, the fourth regression prediction model, and the fifth regression prediction model; and in the second type of optional structure, the splicing feature vector H6 is sent to the fusion feature decoder; C D is a preset splicing feature dimension, C D =C B +C C .

[0026] Further, the component function of the individualization component of the first type of optional structure is as follows:

[0027] The first regression prediction model is implemented based on a feedforward neural network; the first regression prediction model is configured to predict the peak point wavelength of the solvent molecule emission spectrum according to the splicing feature vector H6 to obtain a corresponding first wavelength, which is sent to the prediction output layer;

[0028] The second regression prediction model is implemented based on a feedforward neural network; the second regression prediction model is configured to predict the residence time of the solvent molecule in the excited state according to the splicing feature vector H6 to obtain a corresponding first time, which is sent to the prediction output layer;

[0029] The third regression prediction model is implemented based on a feedforward neural network; the third regression prediction model is configured to predict the photoelectric conversion efficiency of the solvent molecule according to the splicing feature vector H6 to obtain a corresponding first conversion efficiency, which is sent to the prediction output layer;

[0030] The fourth regression prediction model is implemented based on a feedforward neural network; the fourth regression prediction model is configured to predict the peak point wavelength of the solvent molecule absorption spectrum according to the splicing feature vector H6 to obtain a corresponding second wavelength, which is sent to the prediction output layer;

[0031] The fourth regression prediction model is based on a feedforward neural network and is used to predict the half-peak width of the solvent molecule absorption spectrum according to the splicing feature vector H6 to obtain a corresponding first half-peak width, which is sent to the prediction output layer;

[0032] The prediction output layer takes the obtained first wavelength, first time length, first conversion efficiency, second wavelength and first half-peak width as corresponding emission peak y1, molecular lifetime y2, photoelectric conversion efficiency y3, absorption peak y4 and absorption half-peak width y5 to form a corresponding solvent property prediction vector Y and output.

[0033] Further, the component function of the personalized component of the second type of optional structure is:

[0034] The fusion feature decoder is based on a decoder model of a Transformer architecture and is used to convert the splicing feature vector H6 into an encoded feature sequence composed of the feature encoding vector and the prior feature vector H5, initialize a corresponding decoding output sequence based on the data format of the solvent property prediction vector Y, and perform output sequence decoding processing according to the sequence decoding logic of the Transformer architecture decoder based on the encoded feature sequence and the decoding output sequence to obtain and output the corresponding solvent property prediction vector Y.

[0035] Preferably, the model training data set is denoted as a corresponding first data set, and specifically includes:

[0036] A large amount of data of solute-solvent molecule pairs in the field of optoelectronic materials is collected through a plurality of preset data channels to obtain a corresponding molecule pair data set; the plurality of data channels include public electrolyte information library, public optoelectronic material information library and public technical literature; the molecule pair data set includes a plurality of solute-solvent molecule pairs; the solute-solvent molecule pair includes a solute molecule sequence and a solvent molecule sequence; the solute molecule sequence and the solvent molecule sequence are each a one-dimensional SMILES sequence;

[0037] Each of the solute-solvent molecule pairs is taken as a corresponding current molecule pair;

[0038] Based on a preset chemoinformatics tool, a three-dimensional molecular conformation is created according to the solute molecule sequence and the solvent molecule sequence of the current molecule pair to obtain a corresponding current solute conformation and a current solvent conformation; the chemoinformatics tool includes Open Babel software and RDKit software;

[0039] and based on a preset molecular dynamics simulation tool, a stable conformation of the current solute conformation and the current solvent conformation is optimized to obtain a corresponding current solute optimized conformation and a current solvent optimized conformation; and based on the molecular dynamics simulation tool, a three-dimensional solute-solvent system conformation composed of the current solute optimized conformation and the current solvent optimized conformation is simulated to obtain a corresponding current system conformation; the molecular dynamics simulation tool includes Gaussian software, GROMACS software;

[0040] and the element type and three-dimensional coordinates of each atom of the solute molecule in the current system conformation are extracted as the corresponding atomic element type and atomic three-dimensional coordinates to form a corresponding atomic feature and all the atomic features obtained form a corresponding solute molecule structure M 1 ; and the element type and three-dimensional coordinates of each atom of the solvent molecule in the current system conformation are extracted as the corresponding atomic element type and atomic three-dimensional coordinates to form a corresponding atomic feature and all the atomic features obtained form a corresponding solvent molecule structure M 2 ; and the current solute molecule structure M 1 and the solvent molecule structure M 2 are taken as the first solute molecule structure and the first solvent molecule structure of the corresponding current molecule pair;

[0041] and the total number of aromatic rings of the solute molecule in the current system conformation is identified based on the cheminformatics tool, and the identification result is taken as the corresponding solute aromatic ring total number x1; and the interatomic interaction energy of the solute molecule in the current system conformation is calculated based on a preset quantum chemistry calculation tool, and the calculation result is taken as the corresponding solute interatomic interaction energy x2; and the molecular dipole moment of the solvent molecule in the current system conformation is calculated based on the quantum chemistry calculation tool, and the calculation result is taken as the corresponding solvent dipole moment x3; and the polarity index of the solvent molecule in the current system conformation is calculated based on the quantum chemistry calculation tool, and the calculation result is taken as the corresponding solvent polarity index x4; and the orbital overlap integral of the solute molecule and the solvent molecule in the current system conformation is calculated based on the quantum chemistry calculation tool, and the calculation result is taken as the corresponding solute-solvent orbital overlap integral x5; and the obtained solute aromatic ring total number x1, solute interatomic interaction energy x2, solvent dipole moment x3, solvent polarity index x4, and solute-solvent orbital overlap integral x5 form the corresponding prior characteristic vector X; and the obtained prior characteristic vector X is taken as the first prior characteristic vector of the corresponding current molecule pair; the quantum chemistry calculation tool includes Multiwfn software, Gaussian software, ORCA software, GAMESS software, NWChem software, SHARC software, GPAW software, and Newton-X software;

[0042] and the peak wavelength of the emission spectrum of the solute molecule in the current system conformation is calculated based on the quantum chemistry calculation tool, and the calculation result is taken as the corresponding emission peak y1; and the residence time of the solvent molecule in the excited state in the current system conformation is calculated based on the quantum chemistry calculation tool, and the calculation result is taken as the corresponding molecular lifetime y2; and the photoelectric conversion efficiency of the solvent molecule in the current system conformation is calculated based on the quantum chemistry calculation tool, and the calculation result is taken as the corresponding photoelectric conversion efficiency y3; and the peak wavelength of the absorption spectrum of the solvent molecule in the current system conformation is calculated based on the quantum chemistry calculation tool, and the calculation result is taken as the corresponding absorption peak y4; and the half-peak width of the absorption spectrum of the solvent molecule in the current system conformation is calculated based on the quantum chemistry calculation tool, and the calculation result is taken as the corresponding absorption half-peak width y5; and the obtained emission peak y1, molecular lifetime y2, photoelectric conversion efficiency y3, absorption peak y4, and absorption half-peak width y5 form the corresponding solvent characteristic prediction vector Y; and the obtained solvent characteristic prediction vector Y is taken as the first solvent characteristic label vector of the corresponding current molecule pair;

[0043] and the first solute molecular structure, the first solvent molecular structure, the first prior characteristic vector and the first solvent characteristic label vector of the current molecular pair form a corresponding first data record;

[0044] and all the first data records obtained form a corresponding first data set.

[0045] Preferably, the multi-task prediction model is trained based on the first data set, specifically including:

[0046] Step 91, the first data set is randomly divided into two sub-data sets based on a preset first segmentation ratio, denoted as a corresponding first training set and a first evaluation set; and the total number of records of the first training set is counted to obtain a corresponding total number N TR ; and the total number of records of the first training set is counted to obtain a corresponding total number N EV ;

[0047] Wherein, the first training set and the first evaluation set are both composed of a plurality of first data records; the total number ratio of records of the first training set and the first evaluation set satisfies the first segmentation ratio, N TR :N EV ≈ the first segmentation ratio;

[0048] Step 92, each first data record of the first training set is taken as a corresponding current training record; and the first solute molecular structure, the first solvent molecular structure and the first prior characteristic vector of the current training record are respectively subjected to noise adding processing to obtain corresponding noise-added solute molecular structure, noise-added solvent molecular structure and noise-added prior characteristic vector; the noise-added solute molecular structure, the noise-added solvent molecular structure and the noise-added prior characteristic vector are input into the multi-task prediction model for prediction, and the solvent characteristic prediction vector Y obtained by this time prediction is taken as a corresponding first prediction vector; and the first prediction vector and the first solvent characteristic label vector of the current training record form a corresponding first prediction-label pair;

[0049] Wherein, the noise adding mode of the noise-added solute molecular structure is to add noise to part or all of the atomic three-dimensional coordinates of the first solute molecular structure, and the overall noise added should satisfy the Gaussian distribution rule that the average noise is zero; the noise adding mode of the noise-added solvent molecular structure is to add noise to all or part of the atomic three-dimensional coordinates of the first solvent molecular structure, and the overall noise added should satisfy the Gaussian distribution rule that the average noise is zero; and the noise adding mode of the noise-added prior characteristic vector is to add noise to all vector data of the first prior characteristic vector, and the overall noise added should satisfy the Gaussian distribution rule that the average noise is zero.

[0050] each of the first prediction vectors is denoted as a corresponding prediction vector Y g each of the first solvent property label vectors is denoted as a corresponding label vector 1≤index g≤N TR ; each of the prediction vectors Y g the emission peak y1, the molecular lifetime y2, the photoelectric conversion efficiency y3, the absorption peak y4, and the absorption half-peak width y5 in each of the prediction vectors Y g,1 , y g,2 , y g,3 , y g,4 , y g,5 ; each of the label vectors the emission peak y1, the molecular lifetime y2, the photoelectric conversion efficiency y3, the absorption peak y4, and the absorption half-peak width y5 in each of the label vectors

[0051] Step 93, the obtained N TR first prediction-label pairs are brought into a preset overall loss function L total to obtain a corresponding five classification loss values and an overall loss value;

[0052] The overall loss function L total is specifically:

[0053] L total = w1·L1 + w2·L2 + w3·L3 + w4·L4 + w5·L5,

[0054]

[0055] w1, w2, w3, w4, and w5 are preset 5 weight coefficients;

[0056] The five classification loss values include a first loss value, a second loss value, a third loss value, a fourth loss value, and a fifth loss value; the first loss value, the second loss value, the third loss value, the fourth loss value, and the fifth loss value are respectively loss values output by a corresponding first loss function L1, a second loss function L2, a third loss function L3, a fourth loss function L4, and a fifth loss function L5;

[0057] The overall loss value is a loss value output by the overall loss function L total ;

[0058] Step 94, identifying whether the overall loss value meets a preset overall loss value range; if yes, going to step 95; if no, based on a preset first model optimizer, performing a round of parameter modulation on the multi-task prediction model towards a direction of making the overall loss function L total reach a minimum value, and returning to step 92 at the end of the current round of parameter modulation;

[0059] wherein the first model optimizer comprises an Adam optimizer, an SGD optimizer;

[0060] Step 95, identifying whether the first loss value, the second loss value, the third loss value, the fourth loss value and the fifth loss value all meet the respective first loss value range, the second loss value range, the third loss value range, the fourth loss value range and the fifth loss value range; if yes, going to step 96; if no, taking the first loss value, the second loss value, the third loss value, the fourth loss value or the fifth loss value that does not meet the corresponding loss value range as a corresponding to-be-optimized loss value, taking the first loss function L1, the second loss function L2, the third loss function L3, the fourth loss function L4 or the fifth loss function L5 corresponding to the to-be-optimized loss value as a corresponding target loss function, taking the first regression prediction model, the second regression prediction model, the third regression prediction model, the fourth regression prediction model or the fifth regression prediction model corresponding to the to-be-optimized loss value as a corresponding target regression model, taking the model parameters of the target regression model as corresponding target model parameters, assigning a corresponding second model optimizer to the target regression model, and based on the second model optimizer, performing a round of parameter modulation on the target model parameters towards a direction of making the target loss function reach a minimum value, and returning to step 92 after the parameter modulation of all the target model parameters in this time is completed;

[0061] wherein each second model optimizer comprises an Adam optimizer, an SGD optimizer;

[0062] Step 96, performing a round of traversal on all the first data records in the first evaluation set; and in the current round of traversal, taking the first data record currently traversed as a corresponding current evaluation record; and taking the first solute molecule structure, the first solvent molecule structure and the first prior characteristic vector of the current evaluation record as the current solute molecule structure M 1 , the current solvent molecule structure M 2and inputting the prior characteristic vector X into the multi-task prediction model for prediction, and taking the solvent characteristic prediction vector Y obtained through the prediction as a corresponding second prediction vector; and taking the second prediction vector and the first solvent characteristic label vector of the current evaluation record as a corresponding second prediction-label pair; and at the end of the current iteration, taking the N EV second prediction-label pairs obtained into a preset first model evaluation function to obtain a corresponding first evaluation value;

[0063] The first model evaluation function is implemented based on an RMSE function.

[0064] Step 97, identifying whether the first evaluation value meets a preset first evaluation value range; if not, returning to step 91 for continuous training; if yes, confirming that the model training is completed.

[0065] Preferably, the multi-task prediction model is used for solvent selection processing on a batch of user inputted selection tasks, and the processing result is fed back to the current user, specifically including:

[0066] extracting the corresponding first solute molecular sequence, the first solvent molecular sequence set and the five types of index threshold range from the batch of selection tasks;

[0067] and based on a preset cheminformatics tool, creating a first solute conformation according to the first solute molecular sequence to obtain a corresponding first solute conformation; and extracting the element type and three-dimensional coordinates of each atom of the first solute conformation as a corresponding atom element type and a corresponding atom three-dimensional coordinate to form a corresponding atom feature and all the atom features obtained form a corresponding solute molecular structure M 1 ; and based on the cheminformatics tool, identifying the total number of aromatic rings of the first solute conformation and taking the identification result as a corresponding solute aromatic ring total number x1; and based on a preset quantum chemistry calculation tool, calculating the interatomic interaction energy of the first solute conformation and taking the calculation result as a corresponding solute interatomic interaction energy x2;

[0068] and based on the cheminformatics tool, creating a first solvent conformation according to each first solvent molecular sequence of the first solvent molecular sequence set to obtain a corresponding first solvent conformation; and taking each first solvent conformation as a corresponding current solvent conformation; and extracting the element type and three-dimensional coordinates of each atom of the current solvent conformation as a corresponding atom element type and a corresponding atom three-dimensional coordinate to form a corresponding atom feature and all the atom features obtained a corresponding said solvent molecule structure M 2 ; and based on the quantum chemistry calculation tool, calculating a molecular dipole moment corresponding to the current solvent conformation and taking the calculation result as a corresponding said solvent dipole moment x3; and based on the quantum chemistry calculation tool, calculating a polarity index corresponding to the current solvent molecule structure and taking the calculation result as a corresponding said solvent polarity index x4; and based on the quantum chemistry calculation tool, calculating a solute-solvent orbital overlap integral corresponding to the first solute conformation and the current solvent conformation by the GFN2-xTB method and taking the calculation result as a corresponding said solute-solvent orbital overlap integral x5;

[0069] and a set of said solute aromatic ring total number x1, said solute interatomic interaction energy x2, said solvent dipole moment x3, said solvent polarity index x4, and said solute-solvent orbital overlap integral x5 corresponding to the first solute molecule sequence and each said first solvent molecule sequence form a corresponding said prior characteristic vector X; and a set of said solute molecule structure M 1 , solvent molecule structure M 2 , and prior characteristic vector X corresponding to each said first solvent molecule sequence are input into the multi-task prediction model for prediction to obtain a corresponding said solvent characteristic prediction vector Y; and the first solvent molecule sequence corresponding to each said solvent characteristic prediction vector Y satisfying the threshold range of the five types of indexes is recorded as a corresponding preferred molecule sequence;

[0070] and each said preferred molecule sequence and its corresponding said solvent characteristic prediction vector Y form a corresponding preferred record; and all said preferred records obtained form a corresponding solvent molecule preferred report which is fed back to the current user.

[0071] The second aspect of the embodiment of the application provides a device for implementing the processing method for predicting the photoelectric characteristics of a solvent molecule according to the first aspect.

[0072] The model construction module is used to construct a deep learning model for feature fusion of solute molecule structure, solvent molecule structure, and solute-solvent system prior characteristics and prediction of multi-class photoelectric characteristics of a solvent molecule, which is recorded as a corresponding multi-task prediction model; wherein the multi-task prediction model is used to predict the photoelectric characteristics of a solvent molecule based on the input solute molecule structure M 1 , solvent molecule structure M 2The photoelectric properties of the solvent molecule are multi-tasked predicted by the prior characteristic vector X and the corresponding solvent characteristic prediction vector Y is output; the solute molecular structure and the solvent molecular structure are both three-dimensional molecular structures; the prior characteristics of the solute-solvent system include the total number of aromatic rings of the solute molecule, the interatomic interaction energy of the solute molecule, the dipole moment of the solvent molecule, the polarity index of the solvent molecule, and the orbital overlap integral of the solute molecule and the solvent molecule; the multiple types of photoelectric properties include emission peak, molecular lifetime, photoelectric conversion efficiency, absorption peak, and absorption half-peak width; the solute molecular structure M 1 Multiple atomic features Composition, 1≤index i≤N 1 , N 1 is the total number of atoms in the solute molecule; the atomic characteristics Including atomic element type, atomic three-dimensional coordinates; the solvent molecular structure M 2 Multiple atomic features Composition, 1≤indexj≤N 2 , N 2 is the total number of atoms in the solvent molecule; the atomic characteristics Including the atomic element type and the atomic three-dimensional coordinates; the prior characteristic vector X includes the total number of solute aromatic rings x1, the solute interatomic interaction energy x2, the solvent dipole moment x3, the solvent polarity index x4, and the solute-solvent orbital overlap integral x5; the solvent characteristic prediction vector Y includes the emission peak y1, the molecular lifetime y2, the photoelectric conversion efficiency y3, the absorption peak y4, and the absorption half-peak width y5;

[0073] The model training module is used to construct a model training data set recorded as a corresponding first data set; and train the multi-task prediction model based on the first data set;

[0074] The model application module is used to use the multi-task prediction model to perform solvent optimization processing on the batch optimization tasks input by the user after the model training is completed, and to feedback the processing results to the current user; the batch optimization tasks include a first solute molecule sequence, a first solvent molecule sequence set and five types of indicator threshold ranges; the first solvent molecule sequence set includes multiple first solvent molecule sequences; the first solvent molecule sequence and each of the first solvent molecule sequences is a SMILES sequence; the five types of indicator threshold ranges include an emission peak threshold range, a molecular lifetime threshold range, a photoelectric conversion efficiency threshold range, an absorption peak threshold range, and an absorption half-width threshold range.

[0075] A third aspect of an embodiment of the present invention provides an electronic device, including: a memory, a processor, and a transceiver;

[0076] The processor is coupled with the memory, reads and executes instructions in the memory to realize the method steps of the first aspect described above.

[0077] The transceiver is coupled with the processor, and the transceiver is controlled by the processor to perform message transmission and reception.

[0078] The fourth aspect of the embodiment of the present application provides a computer readable storage medium, which stores computer instructions, when the computer instructions are executed by a computer, the computer instructions make the computer execute the instructions of the method of the first aspect described above.

[0079] The embodiment of the present application provides a processing method and device for predicting photoelectric properties of solvent molecules, electronic equipment and computer readable storage medium. From the above content, it can be known that the embodiment of the present application takes five types of solute / solvent properties which can be quickly obtained by simple calculation as prior properties (total number of solute aromatic rings, solute atomic interaction energy, solvent dipole moment, solvent polarity index, solute-solvent orbital overlap integral calculated by the GFN2-xTB method), takes five types of solvent molecular photoelectric properties (emission peak, molecular lifetime, photoelectric conversion efficiency, absorption peak and absorption half-width) which need to be obtained by long-time calculation as prediction targets, constructs a multi-task prediction model which can predict the five types of prediction targets according to the solute molecular structure, the solvent molecular structure and the five types of prior properties, and provides two types of optional structures for the multi-task prediction model; and constructs a model training data set to train the model; and after the model training is completed, the multi-task prediction model is used to assist a user in processing a high-throughput solvent optimization task. The embodiment of the present application not only reduces the analysis complexity, shortens the analysis time and improves the analysis efficiency, but also improves the prediction accuracy, and improves the flexibility and generalization of the prediction model. BRIEF DESCRIPTION OF DRAWINGS

[0080] Figure 1 A processing method for predicting photoelectric properties of solvent molecules is provided for the first embodiment of the present application;

[0081] Figure 2 A module structure diagram of one type of optional structure of the multi-task prediction model provided for the first embodiment of the present application;

[0082] Figure 3 A module structure diagram of two types of optional structures of the multi-task prediction model provided for the first embodiment of the present application;

[0083] Figure 4 A module structure diagram of a processing device for predicting photoelectric properties of solvent molecules provided for the second embodiment of the present application;

[0084] Figure 5A structural schematic diagram of an electronic device provided for the third embodiment of the present application. DETAILED DESCRIPTION

[0085] In order to make the objects, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, but not all the embodiments. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of the present application.

[0086] The first embodiment of the present application provides a processing method for predicting photoelectric characteristics of solvent molecules. Figure 1 As shown in a structural schematic diagram of an electronic device provided for the first embodiment of the present application, the method mainly comprises the following steps:

[0087] Step 1, constructing a deep learning model for fusing characteristics of solute molecular structure, solvent molecular structure and solute-solvent system prior characteristics and predicting multi-class photoelectric characteristics of solvent molecules based on the fused characteristics, which is corresponding to a multi-task prediction model.

[0088] Here, the solute molecular structure and the solvent molecular structure of the embodiments of the present application are both three-dimensional molecular structures. The solute-solvent system prior characteristics of the embodiments of the present application include total number of aromatic rings of the solute molecule, interatomic interaction energy of the solute molecule, dipole moment of the solvent molecule, polarity index of the solvent molecule and orbital overlap integral of the solute molecule and the solvent molecule. The multi-class photoelectric characteristics of the embodiments of the present application include emission peak, molecular lifetime, photoelectric conversion efficiency, absorption peak and absorption half-width.

[0089] It should be noted that the five types of solute-solvent system prior characteristics can be quickly obtained by a calculation tool: 1) total number of aromatic rings of the solute molecule: input the solute molecular structure into a preset chemical informatics tool to count the total number of aromatic rings, and the characteristic parameter can be obtained, which is a second-level process in a conventional case; 2) interatomic interaction energy of the solute molecule: input the solute molecular structure into a preset quantum chemistry calculation tool to calculate, and the characteristic parameter can be obtained, which is a minute-level process or even faster in a conventional case; 3) dipole moment and polarity index of the solvent molecule: input the solvent molecular structure into a quantum chemistry calculation tool to calculate, and the two characteristic parameters can be obtained simultaneously, which is a minute-level process or even faster in a conventional case; 4) orbital overlap integral of the solute molecule and the solvent molecule: input the molecular structures of the solute and the solvent into a quantum chemistry calculation tool and calculate the orbital overlap integral between the solute molecule and the solvent molecule based on the GFN2-xTB method, and the characteristic parameter can be obtained, which is a second-level process in a conventional case.

[0090] It also needs to be explained that the multiple photoelectric characteristics (emission peak, molecular lifetime, photoelectric conversion efficiency, absorption peak, absorption half-peak width) of the embodiments of the present application are the five types of prediction targets of the embodiments of the present application. The time to obtain the five types of prediction targets according to the quantum chemical calculation method is relatively long: 1) The calculation process of the emission peak, the absorption peak and the absorption half-peak width is usually a process of hours, and even for some complex molecular structures, the calculation process can reach a single day; 2) The calculation process of the molecular lifetime and the photoelectric conversion efficiency is usually in units of days or weeks. If high-throughput screening of solvent molecules is to be carried out based on the five types of prediction targets according to the quantum chemical calculation method, it is difficult to complete in a short period of time without sufficient large computing resources.

[0091] The multi-task prediction model of the embodiments of the present application is used to perform multi-task prediction of the photoelectric characteristics of the solvent molecules according to the solute molecular structure M 1 , the solvent molecular structure M 2 and the prior characteristic vector X input by the model, and output the corresponding solvent characteristic prediction vector Y.

[0092] 1) The solute molecular structure M 1 is composed of multiple atomic features , 1≤index i≤N 1 , N 1 is the total number of atoms of the solute molecule; the atomic feature includes the atomic element type and the atomic three-dimensional coordinates. 2) The solvent molecular structure M 2 is composed of multiple atomic features , 1≤index j≤N 2 , N 2 is the total number of atoms of the solvent molecule; the atomic feature includes the atomic element type and the atomic three-dimensional coordinates. 3) The prior characteristic vector X includes the total number of solute aromatic rings x1, the solute interatomic interaction energy x2, the solvent dipole moment x3, the solvent polarity index x4, and the solute-solvent orbital overlap integral x5. 4) The solvent characteristic prediction vector Y includes the emission peak y1, the molecular lifetime y2, the photoelectric conversion efficiency y3, the absorption peak y4, and the absorption half-peak width y5.

[0093] The first model input end of the multi-task prediction model of the embodiments of the present application is used to receive the solute molecular structure M 1 , the second model input end is used to receive the solvent molecular structure M 2 , the third model input end is used to receive the prior characteristic vector X, and the model output end is used to output the corresponding solvent characteristic prediction vector Y.

[0094] It needs to be explained that the embodiments of the present application provide two types of optional model structures for the multi-task prediction model, which are a first type of optional structure and a second type of optional structure; the first type of optional structure is as followsFigure 2 This is a module structure diagram of a first type of optional structure of the multi-task prediction model provided in the first embodiment of the present invention. The second type of optional structure is as follows Figure 3 The module structure diagram of the two optional structures of the multi-task prediction model provided in the first embodiment of the present invention is shown.

[0095] like Figure 2 As shown, the model components of an optional structure of an embodiment of the present invention include: a first structural feature encoder, a solute feature extraction module, a second structural feature encoder, a solvent feature extraction module, a molecular feature fusion module, a fusion feature encoder, a priori feature extraction module, a priori feature fusion module, a first regression prediction model, a second regression prediction model, a third regression prediction model, a fourth regression prediction model, a fifth regression prediction model, and a prediction output layer.

[0096] The component connection relationship of this type of optional structure is: the input end of the first structural feature encoder is connected to the input end of the first model, and the output end is connected to the input end of the solute feature extraction module; the output end of the solute feature extraction module is connected to the first input end of the molecular feature fusion module; the input end of the second structural feature encoder is connected to the input end of the second model, and the output end is connected to the input end of the solvent feature extraction module; the output end of the solvent feature extraction module is connected to the second input end of the molecular feature fusion module; the output end of the molecular feature fusion module is connected to the input end of the fusion feature encoder; the output end of the fusion feature encoder is connected to the first input end of the prior feature fusion module; the input end of the prior feature extraction module is connected to the input end of the third model, and the output end is connected to the second input end of the prior feature fusion module; the output end of the prior feature fusion module is respectively connected to the input end of the first, second, third, fourth and fifth regression prediction models; the input end of the prediction output layer is respectively connected to the output end of the first, second, third, fourth and fifth regression prediction models, and the output end is connected to the model output end.

[0097] like Figure 3 As shown, the model components of the second type of optional structure of the embodiment of the present invention include: a first structural feature encoder, a solute feature extraction module, a second structural feature encoder, a solvent feature extraction module, a molecular feature fusion module, a fusion feature encoder, a priori feature extraction module, a priori feature fusion module, and a fusion feature decoder.

[0098] The component connection relationship of the two optional structures is as follows: the input end of the first structure feature encoder is connected with the first model input end, and the output end is connected with the input end of the solute feature extraction module; the output end of the solute feature extraction module is connected with the first input end of the molecular feature fusion module; the input end of the second structure feature encoder is connected with the second model input end, and the output end is connected with the input end of the solvent feature extraction module; the output end of the solvent feature extraction module is connected with the second input end of the molecular feature fusion module; the output end of the molecular feature fusion module is connected with the input end of the fusion feature encoder; the output end of the fusion feature encoder is connected with the first input end of the prior feature fusion module; the input end of the prior feature extraction module is connected with the third model input end, and the output end is connected with the second input end of the prior feature fusion module; the output end of the prior feature fusion module is connected with the input end of the fusion feature decoder; and the output end of the fusion feature decoder is connected with the model output end.

[0099] As shown in Figure 2 , Figure 3 , the one type and the two types of optional structures provided by the embodiment of the application have eight shared components, which are: a first structure feature encoder, a solute feature extraction module, a second structure feature encoder, a solvent feature extraction module, a molecular feature fusion module, a fusion feature encoder, a prior feature extraction module, and a prior feature fusion module. The functions of the eight shared components in the two types of optional structures are the same, and are as follows.

[0100] 1) The first structure feature encoder:

[0101] The first structure feature encoder is implemented based on a Uni-Mol model. The first structure feature encoder is used for performing high-dimensional feature coding processing on the solute molecular structure M 1 at the atomic level to obtain a corresponding feature tensor E1, which is sent to the solute feature extraction module. The shape of the feature tensor E1 is N 1 xC A , and C A is a preset atomic feature dimension.

[0102] Here, the Uni-Mol model of the first structure feature encoder has been pre-trained. The detailed model structure and the pre-training method of the Uni-Mol model can be understood through the public technical document “Uni-Mol: A Universal 3D Molecular Representation Learning Framework”, and will not be described further here. The reason why the Uni-Mol model is used as the feature encoder of the solute molecular structure M 1 in the embodiment of the application is that the model can refine the precision of the coded features to the atomic level, which helps to further improve the prediction accuracy of the final predicted properties.

[0103] 2) Solute feature extraction module:

[0104] The solute feature extraction module is configured to pool each feature channel of the feature tensor E1 according to a preset solute feature pooling rule to obtain a pooled feature vector P1 with a length of C A B The solute feature extraction module is configured to perform molecular feature space mapping on the pooled feature vector P1 based on an embedded MLP model to obtain a molecular feature vector H1 with a length of C

[0105] Here, the solute feature pooling rule of the embodiment of the present application includes Max Pooling, Average Pooling, and Attentive Pooling. The Max Pooling rule takes the maximum value in the N 1 feature data in each feature channel as the pooled value of the current feature channel. The Average Pooling rule takes the average value in the N 1 feature data in each feature channel as the pooled value of the current feature channel. The Attentive Pooling rule first calculates the attention weights of the N 1 feature data in each feature channel, and then performs weighted summation calculation on the N 1 feature data based on the obtained N 1 attention weights and takes the calculation result as the pooled value of the current feature channel. In the specific implementation process, one of the three rules can be selected as the adaptive rule based on the specific implementation requirements. C B is the preset molecular feature dimension.

[0106] 3) Second structure feature encoder:

[0107] The second structure feature encoder is implemented based on a Uni-Mol model. The second structure feature encoder is configured to perform high-dimensional feature encoding processing on the solvent molecular structure M 2 at the atomic level to obtain a corresponding feature tensor E2, which is sent to the solvent feature extraction module. The shape of the feature tensor E2 is N 2 × C A .

[0108] Here, the Uni-Mol model of the second structure feature encoder has also been pre-trained. The reason why the embodiment of the present application uses the Uni-Mol model as the feature encoder of the solvent molecular structure M 2 is that this model can refine the precision of the encoded features to the atomic level, thereby helping to further improve the prediction accuracy of the final predicted properties.

[0109] 4) Solvent feature extraction module: ​

[0110] The solvent feature extraction module is configured to perform pooling processing on each feature channel of the feature tensor E2 according to a preset solvent feature pooling rule to obtain a pooled feature vector P2 with a length of C A The solvent feature extraction module is configured to perform pooling processing on each feature channel of the feature tensor E2 according to a preset solvent feature pooling rule to obtain a pooled feature vector P2 with a length of C B The solvent feature extraction module is configured to perform pooling processing on each feature channel of the feature tensor E2 according to a preset solvent feature pooling rule to obtain a pooled feature vector P2 with a length of C

[0111] Here, the solvent feature pooling rule of the embodiment of the present application includes maximum pooling, average pooling, and attention pooling, and in the specific implementation process, one of the three types of rules can be selected as the adaptive rule based on the specific implementation requirements.

[0112] 5) The molecular feature fusion module:

[0113] The molecular feature fusion module is configured to sequentially sort the molecular feature vector H1 and the molecular feature vector H2 to form a corresponding feature vector sequence H3 and send the feature vector sequence H3 to the fusion feature encoder.

[0114] 6) The fusion feature encoder:

[0115] The fusion feature encoder is implemented based on an encoder model of a Transformer architecture. The fusion feature encoder is configured to perform feature encoding processing on the feature vector sequence H3 through a multi-head self-attention encoding mechanism to obtain a corresponding encoded feature sequence H4 and send the encoded feature sequence H4 to the prior feature fusion module. The encoded feature sequence H4 is sequentially sorted by two feature encoding vectors B The feature encoding vector corresponds to the molecular feature vector H1 and the molecular feature vector H2 one by one.

[0116] It should be noted that according to the public information of the Transformer architecture, the encoder (Encoder) model under the architecture actually performs feature association between each token feature vector and all token feature vectors in the input feature through a series of multi-head attention operations. In the embodiment of the present application, the molecular feature vector H1 and the molecular feature vector H2 in the feature vector sequence H3 can be regarded as two token feature vectors. After a series of multi-head attention operations of the fusion feature encoder, the molecular feature vector H2 can be associated with the molecular feature vector H1 to obtain a corresponding feature encoding vector The molecular feature vector H1 can be associated with the molecular feature vector H2 to obtain a corresponding feature encoding vector The feature encoding vector is actually a solvent molecular feature fused with solute feature information. Property prediction based on this molecular feature with solute information helps to improve the prediction accuracy.​

[0117] 7) Prior feature extraction module:

[0118] The prior feature extraction module is used to map the prior characteristic vector X to a prior feature space using an embedded MLP model to obtain a prior feature vector H5 with a length of C C and send it to the prior feature fusion module. Wherein, C C is the preset prior feature dimension.

[0119] 8) Prior feature fusion module:

[0120] The prior feature fusion module is used to extract the corresponding feature encoding vector from the encoding feature sequence H4, and perform vector splicing on the feature encoding vector and the prior feature vector H5 to obtain a splicing feature vector H6 with a length of C D ; and in a class of optional structures, the splicing feature vector H6 is sent to the first regression prediction model, the second regression prediction model, the third regression prediction model, the fourth regression prediction model and the fifth regression prediction model; and in a class of optional structures, the splicing feature vector H6 is sent to the fusion feature decoder; C D is the preset splicing feature dimension, C D =C B +C C .

[0121] Here, as known from the foregoing, the feature encoding vector is a solvent molecule feature that fuses solute feature information, and the splicing feature vector H6 obtained by splicing the feature encoding vector and the prior feature vector H5 can be regarded as a solvent molecule feature that fuses solute and prior information.

[0122] As shown in Figure 2 , the personalized components of the first class of optional structures of the embodiment of the application are: the first regression prediction model, the second regression prediction model, the third regression prediction model, the fourth regression prediction model, the fifth regression prediction model, and the prediction output layer. The component functions of the six personalized components of the first class of optional structures are as follows.

[0123] 1) First regression prediction model:

[0124] The first regression prediction model is realized based on a feedforward neural network; the first regression prediction model is used to predict the peak point wavelength of the solvent molecule emission spectrum according to the splicing feature vector H6 to obtain the corresponding first wavelength and send it to the prediction output layer.

[0125] 2) Second regression prediction model:

[0126] The second regression prediction model is implemented based on a feedforward neural network; the second regression prediction model is used to predict the residence time of the solvent molecules in the excited state according to the spliced ​​feature vector H6 and obtain the corresponding first duration and send it to the prediction output layer.

[0127] 3) The third regression prediction model:

[0128] The third regression prediction model is implemented based on a feedforward neural network; the third regression prediction model is used to predict the photoelectric conversion efficiency of the solvent molecules according to the spliced ​​feature vector H6 to obtain the corresponding first conversion efficiency and send it to the prediction output layer.

[0129] 4) The fourth regression prediction model:

[0130] The fourth regression prediction model is implemented based on a feedforward neural network; the fourth regression prediction model is used to predict the peak point wavelength of the solvent molecule absorption spectrum according to the spliced ​​feature vector H6 to obtain the corresponding second wavelength and send it to the prediction output layer.

[0131] 5) Fifth regression prediction model:

[0132] The fifth regression prediction model is implemented based on a feedforward neural network; the fifth regression prediction model is used to predict the half-peak width of the solvent molecule absorption spectrum according to the spliced ​​eigenvector H6 to obtain the corresponding first half-peak width and send it to the prediction output layer.

[0133] 6) Prediction output layer:

[0134] The prediction output layer uses the obtained first wavelength, first duration, first conversion efficiency, second wavelength and first half-peak width as the corresponding emission peak y1, molecular lifetime y2, photoelectric conversion efficiency y3, absorption peak y4 and absorption half-peak width y5 to form the corresponding solvent characteristic prediction vector Y and output it.

[0135] like Figure 3 As shown, the second optional structure of the personalized component of the embodiment of the present invention is: a fusion feature decoder. The component functions of the decoder are as follows.

[0136] The fusion feature decoder of the embodiment of the present invention is implemented based on the decoder model of the Transformer architecture. The fusion feature decoder is used to convert the spliced ​​feature vector H6 into a feature encoding vector and the prior feature vector H5; and initialize a corresponding decoding output sequence based on the data format of the solvent property prediction vector Y; and according to the sequence decoding logic of the Transformer architecture decoder, perform output sequence decoding processing according to the encoding feature sequence and the decoding output sequence to obtain the corresponding solvent property prediction vector Y and output it.

[0137] It should be noted that the fusion feature decoder is sequentially connected by a plurality of decoding blocks and a linear prediction network according to the disclosed decoder model structure of the Transformer architecture, and each decoding block is composed of a multi-head self-attention unit and a cross-attention unit. The input of the fusion feature decoder is the encoded feature sequence (denoted as the encoded feature sequence A) and the initialized decoding output sequence, and the output is the solvent property prediction vector Y. According to the disclosed decoding logic of the Transformer architecture, the decoding logic of the fusion feature decoder is simply that: first, the initialized decoding output sequence is embedded and encoded to obtain an initialized encoded feature sequence Z 0 , and the encoded feature sequence Z 0 is sent to the first decoding block for operation; the multi-head self-attention unit of the first decoding block is used for multi-head self-attention operation on the encoded feature sequence Z 0 , and the operation result is sent to the cross-attention unit of the module as the corresponding self-attention encoded feature sequence S 1 ; the cross-attention unit of the first decoding block is used for cross-attention operation on the encoded feature sequence A and the self-attention encoded feature sequence S 1 , and the operation result is sent to the second decoding block as the new encoded feature sequence Z 1 ; the multi-head self-attention unit of the second decoding block is used for multi-head self-attention operation on the encoded feature sequence Z 1 , and the operation result is sent to the cross-attention unit of the module as the corresponding self-attention encoded feature sequence S 2 ; the cross-attention unit of the second decoding block is used for cross-attention operation on the encoded feature sequence A and the self-attention encoded feature sequence S 2 , and the operation result is sent to the third decoding block as the new encoded feature sequence Z 2 ; and so on, until the cross-attention unit of the last decoding block sends the cross-attention operation result obtained by it as the final encoded feature sequence Z end to the linear prediction network; and the linear prediction network is used for five types of solvent property prediction according to the encoded feature sequence Z end , and outputs the corresponding solvent property prediction vector Y.

[0138] It should be further noted that although the embodiments of the present application provide two types of optional structures for the multi-task prediction model, the user needs to select one as the base model from the two types of optional structures based on the user's habits and / or specific task implementation requirements before processing the prediction task based on the multi-task prediction model. After confirming the base model, first, the subsequent step 2 is trained, and then the subsequent step 3 is applied after the training is completed.

[0139] Step 2, a model training data set is constructed, denoted as a corresponding first data set; and the multi-task prediction model is trained based on the first data set;

[0140] Specifically, step 21, a model training data set is constructed, denoted as a corresponding first data set;

[0141] The first data set includes a plurality of first data records; the first data record includes a first solute molecular structure, a first solvent molecular structure, a first prior characteristic vector, and a first solvent characteristic label vector; the data structures of the first solute molecular structure, the first solvent molecular structure, the first prior characteristic vector, and the first solvent characteristic label vector are consistent with the corresponding solute molecular structure M 1 , the solvent molecular structure M 2 , the prior characteristic vector X, and the solvent characteristic prediction vector Y;

[0142] Specifically, step 211, a plurality of data channels are used to collect big data of solute-solvent molecular pairs in the field of optoelectronic materials to obtain a corresponding molecular pair data set;

[0143] Here, the plurality of data channels include a public electrolyte information library, a public optoelectronic material information library, and public technical literature; the molecular pair data set includes a plurality of solute-solvent molecular pairs; the solute-solvent molecular pair includes a solute molecular sequence and a solvent molecular sequence; the solute molecular sequence and the solvent molecular sequence are each a one-dimensional SMILES sequence;

[0144] Step 212, and each group of solute-solvent molecular pairs is taken as a corresponding current molecular pair;

[0145] Step 213, and based on a preset cheminformatics tool, a three-dimensional molecular conformation is created according to the solute molecular sequence and the solvent molecular sequence of the current molecular pair to obtain a corresponding current solute conformation and a current solvent conformation;

[0146] Here, the cheminformatics tool of the embodiment of the application includes Open Babel software and RDKit software;

[0147] Step 214, and based on a preset molecular dynamics simulation tool, a stable conformation of the current solute conformation and the current solvent conformation is optimized to obtain a corresponding current solute optimized conformation and a current solvent optimized conformation; and a three-dimensional solute-solvent system conformation composed of the current solute optimized conformation and the current solvent optimized conformation is simulated based on the molecular dynamics simulation tool to obtain a corresponding current system conformation;

[0148] Here, the molecular dynamics simulation tool of the embodiment of the application includes Gaussian software and GROMACS software;

[0149] Step 215, and extract the element type and three-dimensional coordinates of each atom of the solute molecule in the current conformation as the corresponding atomic element type and atomic three-dimensional coordinates to form the corresponding atomic features And all the atomic features obtained Form the corresponding solute molecule structure M 1 ; and extract the element type and three-dimensional coordinates of each atom of the solvent molecule in the current conformation as the corresponding atomic element type and atomic three-dimensional coordinates to form the corresponding atomic features And all the atomic features obtained Form the corresponding solvent molecule structure M 2 ; and the current solute molecule structure M 1 And the solvent molecule structure M 2 As the corresponding first solute molecule structure and the first solvent molecule structure of the current molecular pair;

[0150] Step 216, and identify the total number of aromatic rings of the solute molecule in the current conformation based on the chemoinformatics tool and take the identification result as the corresponding total number of solute aromatic rings x1; and calculate the interatomic interaction energy of the solute molecule in the current conformation based on the preset quantum chemistry calculation tool and take the calculation result as the corresponding solute interatomic interaction energy x2; and calculate the molecular dipole moment of the solvent molecule in the current conformation based on the quantum chemistry calculation tool and take the calculation result as the corresponding solvent dipole moment x3; and calculate the polarity index of the solvent molecule in the current conformation based on the quantum chemistry calculation tool and take the calculation result as the corresponding solvent polarity index x4; and calculate the orbital overlap integral of the solute molecule and the solvent molecule in the current conformation based on the quantum chemistry calculation tool and take the calculation result as the corresponding solute-solvent orbital overlap integral x5; and form the corresponding prior characteristic vector X from the total number of solute aromatic rings x1, solute interatomic interaction energy x2, solvent dipole moment x3, solvent polarity index x4, and solute-solvent orbital overlap integral x5; and take the prior characteristic vector X obtained this time as the first prior characteristic vector corresponding to the current molecular pair;

[0151] Here, the quantum chemistry calculation tool of the embodiment of the application includes Multiwfn software, Gaussian software, ORCA software, GAMESS software, NWChem software, SHARC software, GPAW software, Newton-X software;

[0152] Step 217, and based on the quantum chemistry calculation tool, the peak wavelength of the emission spectrum of the solute molecule in the current system conformation is calculated, and the calculation result is taken as the corresponding emission peak y1; and based on the quantum chemistry calculation tool, the residence time of the solvent molecule in the excited state in the current system conformation is calculated, and the calculation result is taken as the corresponding molecular lifetime y2; and based on the quantum chemistry calculation tool, the photoelectric conversion efficiency of the solvent molecule in the current system conformation is calculated, and the calculation result is taken as the corresponding photoelectric conversion efficiency y3; and based on the quantum chemistry calculation tool, the peak wavelength of the absorption spectrum of the solvent molecule in the current system conformation is calculated, and the calculation result is taken as the corresponding absorption peak y4; and based on the quantum chemistry calculation tool, the half-peak width of the absorption spectrum of the solvent molecule in the current system conformation is calculated, and the calculation result is taken as the corresponding absorption half-peak width y5; and the obtained emission peak y1, molecular lifetime y2, photoelectric conversion efficiency y3, absorption peak y4, and absorption half-peak width y5 form a corresponding solvent characteristic prediction vector Y; and the obtained solvent characteristic prediction vector Y is taken as the first solvent characteristic label vector corresponding to the current molecule pair;

[0153] Step 218, and a corresponding first data record is formed by the first solute molecule structure, the first solvent molecule structure, the first prior characteristic vector, and the first solvent characteristic label vector corresponding to the current molecule pair;

[0154] Step 219, and all the obtained first data records form a corresponding first data set;

[0155] Step 22, and the multitask prediction model is trained based on the first data set;

[0156] Specifically includes: step 221, based on the first segmentation ratio, the first data set is randomly divided into two sub-data sets, which are respectively taken as the corresponding first training set and the first evaluation set; and the total number of records of the first training set is counted to obtain the corresponding total number N TR ; and the total number of records of the first training set is counted to obtain the corresponding total number N EV ;

[0157] Wherein, the first segmentation ratio is a pre-set ratio parameter, for example, 8:2; the first training set and the first evaluation set are both composed of a plurality of first data records; the total number ratio of the records of the first training set and the first evaluation set satisfies the first segmentation ratio, N TR :N EV ≈ the first segmentation ratio;

[0158] Step 222, each first data record of the first training set is taken as a corresponding current training record; and the first solute molecular structure, the first solvent molecular structure and the first priori characteristic vector of the current training record are respectively subjected to noise adding processing to obtain corresponding noise-added solute molecular structure, noise-added solvent molecular structure and noise-added priori characteristic vector; and the noise-added solute molecular structure, the noise-added solvent molecular structure and the noise-added priori characteristic vector are input into the multi-task prediction model for prediction, and a solvent characteristic prediction vector Y obtained by this time prediction is taken as a corresponding first prediction vector; and a corresponding first prediction-label pair is formed by the first prediction vector and the first solvent characteristic label vector of the current training record;

[0159] Here, the noise adding mode of the noise-added solute molecular structure is to add noise to part or all of the three-dimensional coordinates of the atoms of the first solute molecular structure, and the overall noise added should satisfy the Gaussian distribution rule that the average noise is zero; the noise adding mode of the noise-added solvent molecular structure is to add noise to all or part of the three-dimensional coordinates of the atoms of the first solvent molecular structure, and the overall noise added should satisfy the Gaussian distribution rule that the average noise is zero; the noise adding mode of the noise-added priori characteristic vector is to add noise to all vector data of the first priori characteristic vector, and the overall noise added should satisfy the Gaussian distribution rule that the average noise is zero; it should be noted that the reason for adding noise to the first solute molecular structure, the first solvent molecular structure and the first priori characteristic vector is to improve the generalization ability of the model through model training;

[0160] In addition, each first prediction vector of the embodiment of the present application is denoted as a corresponding prediction vector Y g , and each first solvent characteristic label vector can be denoted as a corresponding label vector 1≤index g≤N TR ; the emission peak y1, the molecular lifetime y2, the photoelectric conversion efficiency y3, the absorption peak y4 and the absorption half-peak width y5 in each prediction vector Y g are denoted as corresponding prediction values y g,1 , y g,2 , y g,3 , y g,4 , y g,5 ; the emission peak y1, the molecular lifetime y2, the photoelectric conversion efficiency y3, the absorption peak y4 and the absorption half-peak width y5 in each label vector are denoted as corresponding prediction values

[0161] Step 223, the N TR first prediction-label pairs obtained are brought into a preset overall loss function L total to obtain five classification loss values and an overall loss value corresponding thereto;

[0162] Here, the overall loss function L of the embodiment of the application total Specifically,

[0163] L total = w1·L1+ w2·L2+ w3·L3+ w4·L4+ w5·L5,

[0164]

[0165] wherein w1, w2, w3, w4, w5 are five preset weight coefficients; L1, L2, L3, L4, L5 are the corresponding first loss function, second loss function, third loss function, fourth loss function, and fifth loss function; the five classification loss values include a first loss value, a second loss value, a third loss value, a fourth loss value, and a fifth loss value; the first loss value, the second loss value, the third loss value, the fourth loss value, and the fifth loss value are loss values output by the corresponding first loss function L1, second loss function L2, third loss function L3, fourth loss function L4, and fifth loss function L5, respectively; and the overall loss value is the loss value output by the overall loss function L total .

[0166] Step 224, whether the overall loss value meets a preset overall loss value range is identified; if yes, step 225 is turned to; if not, a round of parameter modulation is performed on the multi-task prediction model based on a preset first model optimizer in a direction of making the overall loss function L total reach a minimum value, and step 222 is returned at the end of the round of parameter modulation;

[0167] wherein the overall loss value range is a preset numerical range; and the first model optimizer includes an Adam optimizer and an SGD optimizer.

[0168] Step 225, identify whether the first loss value, the second loss value, the third loss value, the fourth loss value and the fifth loss value all satisfy the respective first loss value range, the second loss value range, the third loss value range, the fourth loss value range and the fifth loss value range; if all satisfy, go to step 226; if at least one loss value does not satisfy, take the first loss value, the second loss value, the third loss value, the fourth loss value or the fifth loss value which does not satisfy the corresponding loss value range as a corresponding to-be-optimized loss value, and take the first loss function L1, the second loss function L2, the third loss function L3, the fourth loss function L4 or the fifth loss function L5 corresponding to each to-be-optimized loss value as a corresponding target loss function, and take the first regression prediction model, the second regression prediction model, the third regression prediction model, the fourth regression prediction model or the fifth regression prediction model corresponding to each to-be-optimized loss value as a corresponding target regression model, and take the model parameters of each target regression model as a corresponding target model parameter, and assign a corresponding second model optimizer to each target regression model, and based on each second model optimizer, perform a round of parameter modulation on the corresponding target model parameter in the direction of making the corresponding target loss function reach the minimum value, and return to step 222 after the parameter modulation of all target model parameters in this time is completed;

[0169] Here, the first loss value range, the second loss value range, the third loss value range, the fourth loss value range and the fifth loss value range are five pre-set numerical ranges; each second model optimizer includes an Adam optimizer, an SGD optimizer;

[0170] Step 226, perform a round of traversal on all first data records of the first evaluation set; and in this round of traversal, take the first data record currently traversed as a corresponding current evaluation record; and take the first solute molecular structure, the first solvent molecular structure and the first prior characteristic vector of the current evaluation record as the current solute molecular structure M 1 , the current solvent molecular structure M 2 and the current prior characteristic vector X input the multi-task prediction model for prediction and take the solvent characteristic prediction vector Y obtained by this time prediction as a corresponding second prediction vector; and form a corresponding second prediction-label pair by the second prediction vector and the first solvent characteristic label vector of the current evaluation record; and when this round of traversal ends, take the N EV second prediction-label pairs obtained to calculate the corresponding first evaluation value by using a pre-set first model evaluation function;

[0171] Here, the first model evaluation function is implemented based on the RMSE function;

[0172] Step 227, identify whether the first evaluation value meets the preset first evaluation value range; if not, return to step 221 for continuous training; if yes, confirm that the model training is completed.

[0173] Here, the first evaluation value range is a pre-set numerical range.

[0174] Step 3, after the model training is completed, the multi-task prediction model is used to process the solvent optimization of the batch optimization task input by the user and feed back the processing result to the current user;

[0175] The batch optimization task includes a first solute molecular sequence, a first solvent molecular sequence set, and a five-category index threshold range; the first solvent molecular sequence set includes a plurality of first solvent molecular sequences; the first solvent molecular sequence and each first solvent molecular sequence are each an SMILES sequence; the five-category index threshold range includes an emission peak threshold range, a molecular lifetime threshold range, a photoelectric conversion efficiency threshold range, an absorption peak threshold range, and an absorption half-peak width threshold range.

[0176] Specifically, it includes: step 31, extracting the corresponding first solute molecular sequence, first solvent molecular sequence set, and five-category index threshold range from the batch optimization task;

[0177] Step 32, based on the chemical informatics tool, creating a corresponding first solute conformation according to the first solute molecular sequence; and extracting the element type and three-dimensional coordinates of each atom of the first solute conformation as a corresponding atomic element type and atomic three-dimensional coordinates to form a corresponding atomic feature and all atomic features obtained form a corresponding solute molecular structure M 1 ; and based on the chemical informatics tool, identifying the total number of aromatic rings of the first solute conformation and taking the identification result as a corresponding solute aromatic ring total number x1; and based on the quantum chemistry calculation tool, calculating the interatomic interaction energy corresponding to the first solute conformation and taking the calculation result as a corresponding solute interatomic interaction energy x2;

[0178] Step 33, based on the chemical informatics tool, creating a corresponding first solvent conformation according to each first solvent molecular sequence of the first solvent molecular sequence set; and taking each first solvent conformation as a corresponding current solvent conformation; and extracting the element type and three-dimensional coordinates of each atom of the current solvent conformation as a corresponding atomic element type and atomic three-dimensional coordinates to form a corresponding atomic feature and all atomic features obtained form a corresponding solvent molecular structure M 2; Based on the quantum chemical calculation tool, the molecular dipole moment corresponding to the current solvent conformation is calculated and the calculation result is used as a corresponding solvent dipole moment x3; Based on the quantum chemical calculation tool, the polarity index corresponding to the current solvent molecular structure is calculated and the calculation result is used as a corresponding solvent polarity index x4; Based on the quantum chemical calculation tool, the solute-solvent orbital overlap integral corresponding to the first solute conformation and the current solvent conformation is calculated according to the GFN2-xTB method and the calculation result is used as a corresponding solute-solvent orbital overlap integral x5;

[0179] Step 34, a corresponding prior characteristic vector X is formed by a set of solute aromatic ring totals x1, solute interatomic interaction energy x2, solvent dipole moment x3, solvent polarity index x4, and solute-solvent orbital overlap integral x5 corresponding to the first solute molecular sequence and each first solvent molecular sequence; and a set of solute molecular structures M corresponding to each first solvent molecular sequence is formed. 1 , solvent molecular structure M 2 The prior characteristic vector X is input into the multi-task prediction model to obtain a corresponding solvent characteristic prediction vector Y; and the first solvent molecule sequence corresponding to each solvent characteristic prediction vector Y that meets the threshold range of the five indicators is recorded as the corresponding preferred molecule sequence;

[0180] In step 35, each preferred molecular sequence and its corresponding solvent property prediction vector Y form a corresponding preferred record; and all obtained preferred records form a corresponding solvent molecule preferred report which is fed back to the current user.

[0181] Figure 4 This is a module structure diagram of a processing device for predicting the photoelectric properties of solvent molecules provided in the second embodiment of the present invention. The device is a terminal device or server that implements the aforementioned method embodiment, and can also be a device that enables the aforementioned terminal device or server to implement the aforementioned method embodiment. For example, the device can be a device or chip system of the aforementioned terminal device or server. Figure 4 As shown, the device includes: a model construction module 201, a model training module 202 and a model application module 203.

[0182] The model building module 201 is used to build a deep learning model that fuses the solute molecular structure, solvent molecular structure and solute-solvent system prior characteristics and predicts multiple types of photoelectric properties of solvent molecules based on the fusion characteristics, which is recorded as the corresponding multi-task prediction model; wherein the multi-task prediction model is used to predict the solute molecular structure M according to the model input 1 , solvent molecular structure M 2The photoelectric properties of the solvent molecule are predicted by multi-task and the prior characteristic vector X, and the corresponding solvent characteristic prediction vector Y is output; the solute molecular structure and the solvent molecular structure are both three-dimensional molecular structures; the prior characteristics of the solute-solvent system include the total number of aromatic rings of the solute molecule, the interatomic interaction energy of the solute molecule, the dipole moment of the solvent molecule, the polarity index of the solvent molecule, and the orbital overlap integral of the solute molecule and the solvent molecule; multiple types of photoelectric properties include emission peak, molecular lifetime, photoelectric conversion efficiency, absorption peak, and absorption half-peak width; the solute molecular structure M 1 Multiple atomic features Composition, 1≤index i≤N 1 , N 1 is the total number of atoms in the solute molecule; atomic characteristics Including atomic element type, atomic three-dimensional coordinates; solvent molecular structure M 2 Multiple atomic features Composition, 1≤indexj≤N 2 , N 2 is the total number of atoms in the solvent molecule; atomic characteristics It includes the atomic element type and atomic three-dimensional coordinates; the prior characteristic vector X includes the total number of solute aromatic rings x1, the solute interatomic interaction energy x2, the solvent dipole moment x3, the solvent polarity index x4, and the solute-solvent orbital overlap integral x5; the solvent characteristic prediction vector Y includes the emission peak y1, molecular lifetime y2, photoelectric conversion efficiency y3, absorption peak y4, and absorption half-peak width y5.

[0183] The model training module 202 is used to construct a model training data set recorded as the corresponding first data set; and train the multi-task prediction model based on the first data set.

[0184] The model application module 203 is used to perform solvent optimization processing on the batch optimization tasks input by the user using the multi-task prediction model after the model training is completed, and to feed back the processing results to the current user; the batch optimization tasks include a first solute molecule sequence, a first solvent molecule sequence set and five types of indicator threshold ranges; the first solvent molecule sequence set includes multiple first solvent molecule sequences; the first solvent molecule sequence and each first solvent molecule sequence are each a SMILES sequence; the five types of indicator threshold ranges include an emission peak threshold range, a molecular lifetime threshold range, a photoelectric conversion efficiency threshold range, an absorption peak threshold range, and an absorption half-width threshold range.

[0185] An embodiment of the present invention provides a processing device for predicting the photoelectric properties of solvent molecules, which can execute the method steps in the above method embodiment. Its implementation principles and technical effects are similar and will not be repeated here.

[0186] It should be noted that the division of the various modules of the above apparatus is only a logical functional division, and in actual implementation, all or part of them can be integrated into one physical entity, or can be physically separated. These modules can all be implemented in the form of software invoked by a processing element; all can be implemented in the form of hardware; or some modules can be implemented in the form of software invoked by a processing element, and some modules can be implemented in the form of hardware. For example, the model construction module can be a separately established processing element, or can be integrated in a certain chip of the above apparatus, in addition, it can also be stored in the form of program code in the memory of the above apparatus, and the function of the above determination module is invoked and executed by a certain processing element of the above apparatus. The implementation of other modules is similar. In addition, all or part of these modules can be integrated together, or can be independently implemented. The processing element described herein can be an integrated circuit with signal processing capability. In the implementation process, each step of the above method or each module can be completed by the integrated logic circuit of the hardware in the processor element or the instructions in the form of software.

[0187] For example, the above modules can be one or more integrated circuits configured to implement the above method, such as one or more application specific integrated circuits (ASICs), or one or more digital signal processors (DSPs), or one or more field programmable gate arrays (FPGAs), etc. For another example, when a certain module above is implemented in the form of program code invoked by a processing element, the processing element can be a general purpose processor, such as a central processing unit (CPU) or other processor that can invoke program code. For another example, these modules can be integrated together to implement in the form of system on a chip (SOC).

[0188] In the above embodiments, all or part of the embodiments may be implemented using software, hardware, firmware, or any combination thereof. When implemented using software, all or part of the embodiments may be implemented in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the processes or functions described in the above method embodiments are generated in whole or in part. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The above-mentioned computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the above-mentioned computer instructions can be transmitted from one website, computer, server or data center to another website, computer, server or data center via wired (such as coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (such as infrared, wireless, Bluetooth, microwave, etc.) means. The above-mentioned computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or data center that includes one or more available media. The above-mentioned available medium can be a magnetic medium (such as a floppy disk, hard disk, tape), an optical medium (such as a DVD), or a semiconductor medium (such as a solid state disk (SSD)), etc.

[0189] Figure 5 This is a schematic diagram of the structure of an electronic device provided in the third embodiment of the present invention. The electronic device can be a terminal device or server that implements the method of the aforementioned embodiment, or it can be a terminal device or server that implements the method of the aforementioned embodiment connected to the aforementioned terminal device or server. Figure 5 As shown, the electronic device may include: a processor 301 (such as a CPU), a memory 302, and a transceiver 303; the transceiver 303 is coupled to the processor 301, and the processor 301 controls the transceiver 303's transceiver actions. Various instructions may be stored in the memory 302 for completing various processing functions and implementing the processing steps described in the aforementioned embodiment method. Preferably, the electronic device involved in the embodiment of the present invention further includes: a power supply 304, a system bus 305, and a communication port 306. The system bus 305 is used to realize communication connections between components. The above-mentioned communication port 306 is used for connecting and communicating between the electronic device and other peripherals.

[0190] exist Figure 5The system bus 305 mentioned above can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. The system bus can be divided into an address bus, a data bus, a control bus, etc. For the convenience of representation, only one thick line is used in the figure, but it does not mean that there is only one bus or only one type of bus. The communication interface is used to realize the communication between the database access device and other devices (such as a client, a read-write library and a read-only library). The memory can include a Random Access Memory (RAM), and can also include a Non-Volatile Memory, such as at least one disk memory.

[0191] The processor mentioned above can be a general-purpose processor, including a Central Processing Unit (CPU), a Network Processor (NP), a Graphics Processing Unit (GPU), etc.; can also be a Digital Signal Processor (DSP), an Application-Specific Integrated Circuit (ASIC), a Field-Programmable Gate Array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0192] It should be noted that the embodiments of the present application also provide a computer readable storage medium, which stores instructions, and when the instructions run on a computer, the computer executes the method and process provided in the above embodiments.

[0193] The embodiment of the present application provides a processing method and device for predicting photoelectric characteristics of solvent molecules, electronic equipment and computer readable storage medium. According to the above content, the embodiment of the present application takes five types of solute / solvent characteristics that can be quickly obtained by simple calculation as prior characteristics (total number of solute aromatic rings, solute atomic interaction energy, solvent dipole moment, solvent polarity index, solute-solvent orbital overlap integral calculated according to the GFN2-xTB method), takes five types of solvent molecular photoelectric characteristics (emission peak, molecular lifetime, photoelectric conversion efficiency, absorption peak and absorption half-width) that need to be calculated for a long time as prediction targets, constructs a multi-task prediction model capable of predicting the five types of prediction targets according to the solute molecular structure, the solvent molecular structure and the five types of prior characteristics, and provides two types of optional structures for the multi-task prediction model; and constructs a model training data set to train the model; and after the model training is completed, the multi-task prediction model is used to assist the user in processing the high-throughput solvent optimization task. The embodiment of the present application not only reduces the analysis complexity, shortens the analysis time and improves the analysis efficiency, but also improves the prediction accuracy, and improves the flexibility and generalization of the prediction model.

[0194] The steps of a method or algorithm described in connection with the embodiments disclosed herein can be embodied directly in hardware, in a software module executed by a processor, or in a combination of the two. A software module can reside in random access memory (RAM), flash memory, read-only memory (ROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), registers, hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.

[0195] The above specific embodiments further illustrate the purpose, technical solutions and beneficial effects of the present application. It should be understood that the above description is only a specific embodiment of the present application and is not used to limit the protection scope of the present application. Any modification, equivalent replacement, improvement, etc. within the spirit and principle of the present application should be included in the protection scope of the present application.

Claims

1. A method for predicting the photoelectric properties of solvent molecules, characterized in that: The method comprises: Construct a deep learning model that fuses the solute molecular structure, solvent molecular structure and solute-solvent system prior characteristics and predicts multiple types of photoelectric properties of solvent molecules based on the fusion characteristics, which is recorded as the corresponding multi-task prediction model; wherein, the multi-task prediction model is used to predict the solute molecular structure M according to the model input 1 、Solvent molecular structure M 2 The photoelectric properties of the solvent molecule are multi-tasked predicted by the prior characteristic vector X and the corresponding solvent characteristic prediction vector Y is output; the solute molecular structure and the solvent molecular structure are both three-dimensional molecular structures; the prior characteristics of the solute-solvent system include the total number of aromatic rings of the solute molecule, the interatomic interaction energy of the solute molecule, the dipole moment of the solvent molecule, the polarity index of the solvent molecule, and the orbital overlap integral of the solute molecule and the solvent molecule; the multiple types of photoelectric properties include emission peak, molecular lifetime, photoelectric conversion efficiency, absorption peak, and absorption half-peak width; the solute molecular structure M 1 Multiple atomic features Composition, 1≤index i≤N 1 , N 1 is the total number of atoms in the solute molecule; the atomic characteristics Including atomic element type, atomic three-dimensional coordinates; the solvent molecular structure M 2 Multiple atomic features Composition, 1≤indexj≤N 2 , N 2 is the total number of atoms in the solvent molecule; the atomic characteristics Including the atomic element type and the atomic three-dimensional coordinates; the prior characteristic vector X includes the total number of solute aromatic rings x1, the solute interatomic interaction energy x2, the solvent dipole moment x3, the solvent polarity index x4, and the solute-solvent orbital overlap integral x5; the solvent characteristic prediction vector Y includes the emission peak y1, the molecular lifetime y2, the photoelectric conversion efficiency y3, the absorption peak y4, and the absorption half-peak width y5; Constructing a model training data set as a corresponding first data set; and training the multi-task prediction model based on the first data set; After the model training is completed, the multi-task prediction model is used to perform solvent optimization processing on the batch optimization tasks input by the user and the processing results are fed back to the current user; the batch optimization tasks include a first solute molecule sequence, a first solvent molecule sequence set and five types of indicator threshold ranges; the first solvent molecule sequence set includes multiple first solvent molecule sequences; the first solvent molecule sequence and each of the first solvent molecule sequences is a SMILES sequence; the five types of indicator threshold ranges include an emission peak threshold range, a molecular lifetime threshold range, a photoelectric conversion efficiency threshold range, an absorption peak threshold range, and an absorption half-width threshold range.

2. The method for predicting the photoelectric properties of solvent molecules according to claim 1, characterized in that: The first data set includes a plurality of first data records; the first data record includes a first solute molecular structure, a first solvent molecular structure, a first priori characteristic vector and a first solvent characteristic label vector; the data structure of the first solute molecular structure, the first solvent molecular structure, the first priori characteristic vector and the first solvent characteristic label vector is consistent with the corresponding solute molecular structure M 1 , the solvent molecular structure M 2 , the prior characteristic vector X and the solvent characteristic prediction vector Y remain consistent.

3. The method for predicting the photoelectric properties of solvent molecules according to claim 1, characterized in that: The first model input end of the multi-task prediction model is used to receive the solute molecular structure M 1 The second model input terminal is used to receive the solvent molecular structure M 2 The third model input terminal is used to receive the priori characteristic vector X, and the model output terminal is used to output the corresponding solvent characteristic prediction vector Y; The multi-task prediction model has two optional structures, namely, a first-class optional structure and a second-class optional structure; The model components of the first type of optional structure include: a first structural feature encoder, a solute feature extraction module, a second structural feature encoder, a solvent feature extraction module, a molecular feature fusion module, a fusion feature encoder, a priori feature extraction module, a priori feature fusion module, a first regression prediction model, a second regression prediction model, a third regression prediction model, a fourth regression prediction model, a fifth regression prediction model, and a prediction output layer; the model components of the second type of optional structure include: the first structural feature encoder, the solute feature extraction module, the second structural feature encoder, the solvent feature extraction module, the molecular feature fusion module, the fusion feature encoder, the priori feature extraction module, the priori feature fusion module, and a fusion feature decoder; The shared components of the first type of optional structure and the second type of optional structure are: the first structural feature encoder, the solute feature extraction module, the second structural feature encoder, the solvent feature extraction module, the molecular feature fusion module, the fusion feature encoder, the prior feature extraction module, and the prior feature fusion module; The personalized components of the optional structure are: the first regression prediction model, the second regression prediction model, the third regression prediction model, the fourth regression prediction model, the fifth regression prediction model, and the prediction output layer; The personalized component of the second type of optional structure is: the fusion feature decoder.

4. The method for predicting the photoelectric properties of solvent molecules according to claim 3, characterized in that: The component connection relationship of the said optional structure is as follows: the input end of the first structural feature encoder is connected to the input end of the first model, and the output end is connected to the input end of the solute feature extraction module; the output end of the solute feature extraction module is connected to the first input end of the molecular feature fusion module; the input end of the second structural feature encoder is connected to the input end of the second model, and the output end is connected to the input end of the solvent feature extraction module; the output end of the solvent feature extraction module is connected to the second input end of the molecular feature fusion module; the output end of the molecular feature fusion module is connected to the input end of the fusion feature encoder; the output end of the fusion feature encoder is connected to the first input end of the prior feature fusion module; the input end of the prior feature extraction module is connected to the input end of the third model, and the output end is connected to the second input end of the prior feature fusion module; the output end of the prior feature fusion module is respectively connected to the input end of the first, second, third, fourth and fifth regression prediction models; the input end of the prediction output layer is respectively connected to the output end of the first, second, third, fourth and fifth regression prediction models, and the output end is connected to the output end of the model; The component connection relationship of the two types of optional structures is: the input end of the first structural feature encoder is connected to the input end of the first model, and the output end is connected to the input end of the solute feature extraction module; the output end of the solute feature extraction module is connected to the first input end of the molecular feature fusion module; the input end of the second structural feature encoder is connected to the input end of the second model, and the output end is connected to the input end of the solvent feature extraction module; the output end of the solvent feature extraction module is connected to the second input end of the molecular feature fusion module; the output end of the molecular feature fusion module is connected to the input end of the fusion feature encoder; the output end of the fusion feature encoder is connected to the first input end of the prior feature fusion module; the input end of the prior feature extraction module is connected to the input end of the third model, and the output end is connected to the second input end of the prior feature fusion module; the output end of the prior feature fusion module is connected to the input end of the fusion feature decoder; and the output end of the fusion feature decoder is connected to the output end of the model.

5. The method for predicting the photoelectric properties of solvent molecules according to claim 3, characterized in that: The functions of the shared components of the first type of optional structure and the second type of optional structure are: The first structural feature encoder is implemented based on the Uni-Mol model; the first structural feature encoder is used to encode the solute molecular structure M 1 Perform atomic-level high-dimensional feature encoding processing to obtain the corresponding feature tensor E1 and send it to the solute feature extraction module; the shape of the feature tensor E1 is N 1 ×C A , C A is the preset atomic feature dimension; The solute feature extraction module is used to perform pooling processing on each feature channel of the feature tensor E1 according to the preset solute feature pooling rule to obtain a vector length of C A The pooled feature vector P1 is mapped to the molecular feature space based on the built-in MLP model to obtain a vector with a length of C B The molecular feature vector H1 is sent to the molecular feature fusion module; the solute feature pooling rules include maximum pooling, average pooling, and attention pooling; C B is the preset molecular feature dimension; The second structural feature encoder is implemented based on the Uni-Mol model; the second structural feature encoder is used to 2 Perform atomic-level high-dimensional feature encoding processing to obtain the corresponding feature tensor E2 and send it to the solvent feature extraction module; the shape of the feature tensor E2 is N 2 ×C A ; The solvent feature extraction module is used to perform pooling processing on each feature channel of the feature tensor E2 according to the preset solvent feature pooling rule to obtain a vector length of C A The pooled feature vector P2 is mapped to the molecular feature space based on the built-in MLP model to obtain a vector with a length of C B The molecular feature vector H2 is sent to the molecular feature fusion module; the solvent feature pooling rules include maximum pooling, average pooling, and attention pooling; The molecular feature fusion module is used to sequentially sort the molecular feature vector H1 and the molecular feature vector H2 to form a corresponding feature vector sequence H3 and send it to the fusion feature encoder; The fusion feature encoder is implemented based on the encoder model of the Transformer architecture; the fusion feature encoder is used to perform feature encoding processing on the feature vector sequence H3 through a multi-head self-attention encoding mechanism to obtain the corresponding encoding feature sequence H4 and send it to the prior feature fusion module; the encoding feature sequence H4 consists of two vectors with a length of C B The feature encoding vector The feature coding vector is sorted in sequence; One-to-one correspondence with the molecular characteristic vectors H1 and H2; The prior feature extraction module is used to use the built-in MLP model to perform prior feature space mapping on the prior feature vector X to obtain a vector with a length of C C The prior feature vector H5 is sent to the prior feature fusion module; C C is the preset prior feature dimension; The prior feature fusion module is used to extract the corresponding feature coding vector from the coding feature sequence H4 And the feature encoding vector The vector length C is obtained by vector concatenation with the prior feature vector H5. D and in the first optional structure, sending the spliced ​​feature vector H6 to the first regression prediction model, the second regression prediction model, the third regression prediction model, the fourth regression prediction model, and the fifth regression prediction model; and in the second optional structure, sending the spliced ​​feature vector H6 to the fusion feature decoder; C D is the preset splicing feature dimension, C D =C B +C C .

6. The method for predicting the photoelectric properties of solvent molecules according to claim 3, characterized in that: The component functions of the personalized components of the optional structure are: The first regression prediction model is implemented based on a feedforward neural network; the first regression prediction model is used to predict the peak point wavelength of the solvent molecule emission spectrum according to the spliced ​​feature vector H6 to obtain the corresponding first wavelength and send it to the prediction output layer; The second regression prediction model is implemented based on a feedforward neural network; the second regression prediction model is used to predict the residence time of the solvent molecules in the excited state according to the spliced ​​feature vector H6, obtain the corresponding first duration and send it to the prediction output layer; The third regression prediction model is implemented based on a feedforward neural network; the third regression prediction model is used to predict the photoelectric conversion efficiency of the solvent molecules according to the spliced ​​feature vector H6 to obtain a corresponding first conversion efficiency and send it to the prediction output layer; The fourth regression prediction model is implemented based on a feedforward neural network; the fourth regression prediction model is used to predict the peak point wavelength of the solvent molecule absorption spectrum according to the spliced ​​feature vector H6 to obtain the corresponding second wavelength and send it to the prediction output layer; The fifth regression prediction model is implemented based on a feedforward neural network; the fifth regression prediction model is used to predict the half-peak width of the solvent molecule absorption spectrum according to the spliced ​​feature vector H6 to obtain the corresponding first half-peak width and send it to the prediction output layer; The prediction output layer uses the obtained first wavelength, the first duration, the first conversion efficiency, the second wavelength and the first half-width as the corresponding emission peak y1, the molecular lifetime y2, the photoelectric conversion efficiency y3, the absorption peak y4 and the absorption half-width y5 to form the corresponding solvent characteristic prediction vector Y and output it.

7. The method for predicting the photoelectric properties of solvent molecules according to claim 3, characterized in that: The functions of the personalized components of the second type of optional structures are as follows: The fusion feature decoder is implemented based on the decoder model of the Transformer architecture; the fusion feature decoder is used to convert the spliced ​​feature vector H6 into a feature encoding vector and the encoding feature sequence composed of the prior feature vector H5; and initialize a corresponding decoding output sequence based on the data format of the solvent property prediction vector Y; and perform output sequence decoding processing according to the encoding feature sequence and the decoding output sequence according to the sequence decoding logic of the Transformer architecture decoder to obtain the corresponding solvent property prediction vector Y and output it.

8. The method for predicting the photoelectric properties of solvent molecules according to claim 2, characterized in that: The model building training dataset is recorded as the corresponding first dataset, specifically including: A corresponding molecular pair dataset is obtained by collecting big data on solute-solvent molecular pairs in the field of optoelectronic materials through a plurality of preset data channels; the plurality of data channels include a public electrolyte information library, a public optoelectronic material information library, and public technical literature; the molecular pair dataset includes a plurality of solute-solvent molecular pairs; the solute-solvent molecular pairs include a solute molecular sequence and a solvent molecular sequence; the solute molecular sequence and the solvent molecular sequence are each a one-dimensional SMILES sequence; and taking each group of solute-solvent molecule pairs as the corresponding current molecule pairs; and creating a three-dimensional molecular conformation based on the solute molecule sequence and the solvent molecule sequence of the current molecular pair based on a preset chemical informatics tool, wherein the chemical informatics tool includes Open Babel software and RDKit software; Based on a preset molecular dynamics simulation tool, the stable conformations of the current solute conformation and the current solvent conformation are optimized to obtain the corresponding current solute optimized conformation and current solvent optimized conformation; and based on the molecular dynamics simulation tool, a three-dimensional solute-solvent system conformation composed of the current solute optimized conformation and the current solvent optimized conformation is simulated to obtain the corresponding current system conformation; the molecular dynamics simulation tool includes Gaussian software and GROMACS software; The element type and three-dimensional coordinates of each atom of the solute molecule in the current system conformation are extracted as the corresponding atomic element type and the atomic three-dimensional coordinates to form the corresponding atomic features. And all the atomic features m obtained 1 i The solute molecular structure M corresponding to the composition 1 ; and extracting the element type and three-dimensional coordinates of each atom of the solvent molecule in the current system conformation as the corresponding atomic element type and the atomic three-dimensional coordinate composition of the corresponding atomic features And all the atomic features obtained The solvent molecular structure M corresponding to the composition 2 ; and the current solute molecular structure M 1 and the solvent molecular structure M 2 the first solute molecular structure and the first solvent molecular structure corresponding to the current molecular pair; And based on the chemical informatics tool, the total number of aromatic rings of the solute molecules in the current system conformation is identified and the identification result is used as the corresponding total number of solute aromatic rings x1; and based on the preset quantum chemical calculation tool, the interatomic interaction energy of the solute molecules in the current system conformation is calculated and the calculation result is used as the corresponding solute interatomic interaction energy x2; and based on the quantum chemical calculation tool, the molecular dipole moment of the solvent molecules in the current system conformation is calculated and the calculation result is used as the corresponding solvent dipole moment x3; and based on the quantum chemical calculation tool, the polarity index of the solvent molecules in the current system conformation is calculated and the calculation result is used as the corresponding solvent polarity index x4; and based on the quantum chemical calculation tool, the current The orbital overlap integrals of the solute molecules and the solvent molecules in the system conformation are calculated and the calculation result is used as the corresponding solute-solvent orbital overlap integral x5; and the corresponding prior characteristic vector X is composed of the obtained total number of solute aromatic rings x1, the solute interatomic interaction energy x2, the solvent dipole moment x3, the solvent polarity index x4, and the solute-solvent orbital overlap integral x5; and the prior characteristic vector X obtained this time is used as the first prior characteristic vector corresponding to the current molecular pair; the quantum chemical calculation tools include Multiwfn software, Gaussian software, ORCA software, GAMESS software, NWChem software, SHARC software, GPAW software, and Newton-X software; Based on the quantum chemical calculation tool, the peak wavelength of the emission spectrum of the solute molecules in the current system conformation is calculated and the calculation result is used as the corresponding emission peak y1; based on the quantum chemical calculation tool, the residence time of the solvent molecules in the excited state in the current system conformation is calculated and the calculation result is used as the corresponding molecular lifetime y2; based on the quantum chemical calculation tool, the photoelectric conversion efficiency of the solvent molecules in the current system conformation is calculated and the calculation result is used as the corresponding photoelectric conversion efficiency y3; based on the quantum chemical calculation tool, the peak wavelength of the absorption spectrum of the solvent molecules in the current system conformation is calculated and the calculation result is used as the corresponding absorption peak y4; based on the quantum chemical calculation tool, the half-peak width of the absorption spectrum of the solvent molecules in the current system conformation is calculated and the calculation result is used as the corresponding absorption half-peak width y5; and the obtained emission peak y1, the molecular lifetime y2, the photoelectric conversion efficiency y3, the absorption peak y4, and the absorption half-peak width y5 form the corresponding solvent characteristic prediction vector Y; and the solvent characteristic prediction vector Y obtained this time is used as the first solvent characteristic label vector corresponding to the current molecule pair; and forming a corresponding first data record from the first solute molecular structure, the first solvent molecular structure, the first priori characteristic vector, and the first solvent characteristic label vector corresponding to the current molecular pair; All the obtained first data records form the corresponding first data set.

9. The method for predicting the photoelectric properties of solvent molecules according to claim 2, wherein: The training of the multi-task prediction model based on the first data set specifically includes: Step 91: randomly split the first data set into two sub-data sets based on a preset first split ratio, recorded as a corresponding first training set and a first evaluation set; and count the total number of records in the first training set to obtain the corresponding total number N TR ; and count the total number of records in the first training set to obtain the corresponding total number N EV ; Wherein, the first training set and the first evaluation set are both composed of a plurality of the first data records; the ratio of the total number of records of the first training set to the first evaluation set satisfies the first segmentation ratio, N TR :N EV ≈ first division ratio; Step 92: Use each of the first data records in the first training set as a corresponding current training record; perform noise addition processing on the first solute molecular structure, the first solvent molecular structure, and the first prior characteristic vector of the current training record to obtain corresponding noisy solute molecular structure, noisy solvent molecular structure, and noisy prior characteristic vector; input the noisy solute molecular structure, the noisy solvent molecular structure, and the noisy prior characteristic vector into the multi-task prediction model for prediction, and use the solvent characteristic prediction vector Y obtained in this prediction as the corresponding first prediction vector; and form a corresponding first prediction-label pair with the first prediction vector and the first solvent characteristic label vector of the current training record; The noise adding method of the noisy solute molecular structure is to add noise to part or all of the three-dimensional coordinates of the atoms of the first solute molecular structure, and the overall noise added should satisfy the Gaussian distribution rule with an average noise of zero; the noise adding method of the noisy solvent molecular structure is to add noise to all or part of the three-dimensional coordinates of the atoms of the first solvent molecular structure, and the overall noise added should satisfy the Gaussian distribution rule with an average noise of zero; the noise adding method of the noisy prior characteristic vector is to add noise to all vector data of the first prior characteristic vector, and the overall noise added should satisfy the Gaussian distribution rule with an average noise of zero; Each of the first prediction vectors is recorded as the corresponding prediction vector Y g , each of the first solvent characteristic label vectors is recorded as the corresponding label vector 1≤indexg≤N TR ; Each of the prediction vectors Y g The emission peak y1, the molecular lifetime y2, the photoelectric conversion efficiency y3, the absorption peak y4, and the absorption half-maximum width y5 in are recorded as the corresponding predicted values ​​y g,1 、y g,2 、y g,3 、y g,4 、y g,5 ; Each of the label vectors The emission peak y1, the molecular lifetime y2, the photoelectric conversion efficiency y3, the absorption peak y4, and the absorption half-maximum width y5 in are recorded as the corresponding predicted values Step 93: Get N TR The first prediction-label pair is brought into the preset overall loss function L total Calculate and obtain the corresponding five classification loss values ​​and one overall loss value; Among them, the overall loss function L total Specifically: L total =w1·L1+w2·L2+w3·L3+w4·L4+w5·L5, w1, w2, w3, w4, and w5 are the five preset weight coefficients; The five classification loss values ​​include a first loss value, a second loss value, a third loss value, a fourth loss value, and a fifth loss value; the first loss value, the second loss value, the third loss value, the fourth loss value, and the fifth loss value are the loss values ​​output by the corresponding first loss function L1, second loss function L2, third loss function L3, fourth loss function L4, and fifth loss function L5, respectively; The overall loss value is the overall loss function L total Output loss value; Step 94, identifying whether the overall loss value meets the preset overall loss value range; if so, go to step 95; if not, based on the preset first model optimizer, move towards making the overall loss function L total Perform a round of parameter modulation on the multi-task prediction model in the direction of reaching the minimum value, and return to step 92 when this round of parameter modulation is completed; Wherein, the first model optimizer includes Adam optimizer and SGD optimizer; Step 95, identify whether the first loss value, the second loss value, the third loss value, the fourth loss value and the fifth loss value all satisfy their respective corresponding first loss value range, second loss value range, third loss value range, fourth loss value range and fifth loss value range; if they all satisfy, go to step 96; if at least one loss value does not satisfy, then take each of the first loss value, the second loss value, the third loss value, the fourth loss value or the fifth loss value that does not satisfy the corresponding loss value range as a corresponding loss value to be optimized, and use the first loss function L1, the second loss function L2, the third loss function L3, the fourth loss function L4 corresponding to each of the loss values ​​to be optimized 4 or the fifth loss function L5 as the corresponding target loss function, and use the first regression prediction model, the second regression prediction model, the third regression prediction model, the fourth regression prediction model or the fifth regression prediction model corresponding to each of the loss values ​​to be optimized as the corresponding target regression model, and use the model parameters of each of the target regression models as the corresponding target model parameters, and assign a corresponding second model optimizer to each of the target regression models, and based on each of the second model optimizers, perform a round of parameter modulation on the corresponding target model parameters in the direction of minimizing the corresponding target loss function, and return to step 92 after the parameter modulation of all the target model parameters is completed; Wherein, each of the second model optimizers includes an Adam optimizer and an SGD optimizer; Step 96: perform a round of traversal on all the first data records of the first evaluation set; and in this round of traversal, use the first data record currently traversed as the corresponding current evaluation record; and use the first solute molecular structure, the first solvent molecular structure and the first prior characteristic vector of the current evaluation record as the current solute molecular structure M 1 , the solvent molecular structure M 2 The prior characteristic vector X is input into the multi-task prediction model for prediction and the solvent characteristic prediction vector Y obtained in this prediction is used as the corresponding second prediction vector; and the second prediction vector and the first solvent characteristic label vector of the current evaluation record form a corresponding second prediction-label pair; and at the end of this round of traversal, the obtained N EV The second prediction-label pair is brought into a preset first model evaluation function to calculate a corresponding first evaluation value; Wherein, the first model evaluation function is implemented based on the RMSE function; Step 97 , identifying whether the first evaluation value satisfies a preset first evaluation value range; if not, returning to step 91 to continue training; if so, confirming that the model training is complete.

10. The method for predicting the photoelectric properties of solvent molecules according to claim 1, characterized in that: The method of using the multi-task prediction model to perform solvent optimization processing on the batch optimization task input by the user and feeding back the processing result to the current user specifically includes: Extracting the corresponding first solute molecule sequence, the first solvent molecule sequence set, and the five-category indicator threshold ranges from the batch optimization task; Based on the preset chemical informatics tool, a three-dimensional molecular conformation is created according to the first solute molecular sequence to obtain the corresponding first solute conformation; and the element type and three-dimensional coordinates of each atom in the first solute conformation are extracted as the corresponding atomic element type and the atomic three-dimensional coordinates to form a corresponding atomic feature And all the atomic features obtained Form a corresponding solute molecular structure M 1 ; and based on the chemical informatics tool, identifying the total number of aromatic rings of the first solute conformation and using the identification result as a corresponding total number of aromatic rings of the solute x1; and based on a preset quantum chemical calculation tool, calculating the interatomic interaction energy corresponding to the first solute conformation and using the calculation result as a corresponding interatomic interaction energy of the solute x2; Based on the chemical informatics tool, a three-dimensional molecular conformation is created according to each of the first solvent molecular sequences in the first solvent molecular sequence set to obtain a corresponding first solvent conformation; each of the first solvent conformations is used as the corresponding current solvent conformation; and the element type and three-dimensional coordinates of each atom in the current solvent conformation are extracted as the corresponding atomic element type and the atomic three-dimensional coordinates to form a corresponding atomic feature. And all the atomic features obtained Form a corresponding solvent molecular structure M 2 ; and based on the quantum chemical calculation tool, calculating the molecular dipole moment corresponding to the current solvent conformation and using the calculation result as the corresponding solvent dipole moment x3; and based on the quantum chemical calculation tool, calculating the polarity index corresponding to the current solvent molecular structure and using the calculation result as the corresponding solvent polarity index x4; and based on the quantum chemical calculation tool, calculating the solute-solvent orbital overlap integral corresponding to the first solute conformation and the current solvent conformation according to the GFN2-xTB method and using the calculation result as the corresponding solute-solvent orbital overlap integral x5; The total number of solute aromatic rings x1, the solute interatomic interaction energy x2, the solvent dipole moment x3, the solvent polarity index x4, and the solute-solvent orbital overlap integral x5 corresponding to the first solute molecule sequence and each of the first solvent molecule sequences form a corresponding priori characteristic vector X; and the solute molecular structure M corresponding to each of the first solvent molecule sequences is 1 、Solvent molecular structure M 2 and the prior characteristic vector X are input into the multi-task prediction model to predict a corresponding solvent characteristic prediction vector Y; and the first solvent molecule sequence corresponding to each solvent characteristic prediction vector Y that satisfies the threshold range of the five categories of indicators is recorded as the corresponding preferred molecule sequence; Each of the preferred molecular sequences and its corresponding solvent property prediction vector Y forms a corresponding preferred record; and all the obtained preferred records form a corresponding solvent molecule preferred report to be fed back to the current user.

11. A device for executing the processing method for predicting the photoelectric properties of solvent molecules according to any one of claims 1 to 10, characterized in that: The device includes: a model construction module, a model training module and a model application module; The model building module is used to build a deep learning model that fuses the solute molecular structure, solvent molecular structure and solute-solvent system prior characteristics and predicts multiple types of photoelectric properties of solvent molecules based on the fusion characteristics, which is recorded as the corresponding multi-task prediction model; wherein, the multi-task prediction model is used to predict the solute molecular structure M according to the model input 1 、Solvent molecular structure M 2 The photoelectric properties of the solvent molecule are multi-tasked predicted by the prior characteristic vector X and the corresponding solvent characteristic prediction vector Y is output; the solute molecular structure and the solvent molecular structure are both three-dimensional molecular structures; the prior characteristics of the solute-solvent system include the total number of aromatic rings of the solute molecule, the interatomic interaction energy of the solute molecule, the dipole moment of the solvent molecule, the polarity index of the solvent molecule, and the orbital overlap integral of the solute molecule and the solvent molecule; the multiple types of photoelectric properties include emission peak, molecular lifetime, photoelectric conversion efficiency, absorption peak, and absorption half-peak width; the solute molecular structure M 1 Multiple atomic features Composition, 1≤index i≤N 1 , N 1 is the total number of atoms in the solute molecule; the atomic characteristics Including atomic element type, atomic three-dimensional coordinates; the solvent molecular structure M 2 Multiple atomic features Composition, 1≤indexj≤N 2 , N 2 is the total number of atoms in the solvent molecule; the atomic characteristics Including the atomic element type and the atomic three-dimensional coordinates; the prior characteristic vector X includes the total number of solute aromatic rings x1, the solute interatomic interaction energy x2, the solvent dipole moment x3, the solvent polarity index x4, and the solute-solvent orbital overlap integral x5; the solvent characteristic prediction vector Y includes the emission peak y1, the molecular lifetime y2, the photoelectric conversion efficiency y3, the absorption peak y4, and the absorption half-peak width y5; The model training module is used to construct a model training data set recorded as a corresponding first data set; and train the multi-task prediction model based on the first data set; The model application module is used to use the multi-task prediction model to perform solvent optimization processing on the batch optimization tasks input by the user after the model training is completed, and to feedback the processing results to the current user; the batch optimization tasks include a first solute molecule sequence, a first solvent molecule sequence set and five types of indicator threshold ranges; the first solvent molecule sequence set includes multiple first solvent molecule sequences; the first solvent molecule sequence and each of the first solvent molecule sequences is a SMILES sequence; the five types of indicator threshold ranges include an emission peak threshold range, a molecular lifetime threshold range, a photoelectric conversion efficiency threshold range, an absorption peak threshold range, and an absorption half-width threshold range.

12. An electronic device, characterized in that: include: memory, processors, and transceivers; The processor is configured to be coupled to the memory, read and execute instructions in the memory, so as to implement the method according to any one of claims 1 to 10; The transceiver is coupled to the processor, and the processor controls the transceiver to send and receive messages.

13. A computer-readable storage medium, characterized in that The computer-readable storage medium stores computer instructions, and when the computer instructions are executed by a computer, the computer is caused to execute the method according to any one of claims 1 to 10.

Images