A prediction method for antioxidant combinations in vegetable oils based on deep learning

Abstract

The present invention discloses a prediction method for antioxidant combinations of vegetable oils based on deep learning, which relates to the technical field of prediction of antioxidant combinations of vegetable oils. The method includes: inputting the sample data of the vegetable oil sample to be predicted obtained into a pre-trained prediction model to obtain the prediction result of the antioxidant combination corresponding to the vegetable oil sample to be predicted; the construction process of the pre-trained prediction model is: collecting data and constructing a training set and a test set; inputting the training set into the prediction model to train the model, and the prediction model is a combination model of a factorization machine with an attention mechanism and a deep neural network; verifying the prediction model through the test set, and optimizing the parameters of the prediction model according to the verification result. The present invention combines a deep learning algorithm and an electronic device to realize the construction of the prediction model, improving the accuracy and efficiency of determining the antioxidant composition.

Description

Background Art

[0002] In order to maintain the freshness of vegetable oil, adding antioxidants to vegetable oil is a common means to maintain its freshness. Usually, the effect of adding a single antioxidant is limited. Research shows that there may be a synergistic effect between different combinations of antioxidants, showing a better freshness preservation effect than adding a single antioxidant. However, which combinations of antioxidants have the greatest effect on the freshness preservation of vegetable oil still needs to be further explored, and the freshness preservation effects of different vegetable oils under different combinations of antioxidants are not the same. Moreover, screening combinations of antioxidants through experimental means is a cumbersome process. Based on this, how to quickly determine the most effective combination of antioxidants corresponding to different varieties of vegetable oil is an urgent problem to be solved at present. Summary of the Invention

[0003] The technical problem to be solved by the present invention is to provide a prediction method for antioxidant combinations of vegetable oil based on deep learning in view of the deficiencies of the prior art, specifically as follows:

[0004] 1) In the first aspect, the present invention provides a prediction method for antioxidant combinations of vegetable oil based on deep learning, and the specific technical solution is as follows:

[0005] Input the sample data of the vegetable oil sample to be predicted into the pre-trained prediction model to obtain the prediction result of the antioxidant combination corresponding to the vegetable oil sample to be predicted;

[0006] The construction process of the pre-trained prediction model is as follows:

[0007] Collect the data characteristics of experimental vegetable oil and the data characteristics of antioxidant combinations, and classify and integrate all data characteristics to construct a training set and a test set;

[0008] Input the training set into the prediction model for model training. The prediction model includes a data preprocessing module, and the main body of the model is a combined model of a factorization machine with an attention mechanism and a deep neural network;

[0009] Verify the prediction model through the test set, optimize the parameters of the prediction model according to the verification result, and use the optimized prediction model as the prediction model in the next model training until the prediction model is verified through the test set, and use the verified prediction model as the pre-trained prediction model.

[0010] The beneficial effects of the prediction method for antioxidant combinations of vegetable oil based on deep learning provided by the present invention are as follows:

[0011] This solution provides a method that can determine an antioxidant combination corresponding to a specific vegetable oil according to the characteristics of different vegetable oils. By using the method described in this solution and combining a pre-trained prediction model, the accuracy and efficiency of determining the antioxidant composition are improved.

[0012] Based on the above solution, the present invention can be further improved as follows.

[0013] Further, the model includes a data preprocessing module, which performs corresponding preprocessing on data features of different categories in the dataset samples.

[0014] Further, the process of obtaining the data features of the antioxidant combination is as follows:

[0015] After encoding the molecular string of any antioxidant in the antioxidant combination, feature extraction is performed through the embedding layer in the antioxidant molecular feature extraction module to obtain the antioxidant molecular feature, and the antioxidant molecular feature is used as the data feature of the antioxidant combination.

[0016] Further, the process of obtaining the experimental vegetable oil data features is as follows:

[0017] Input the basic parameters of the experimental vegetable oil into the vegetable oil feature extraction module to obtain the experimental vegetable oil data features;

[0018] The vegetable oil feature extraction model includes three hidden layers.

[0019] Further, perform feature fusion and normalization processing on the data features of the antioxidant and the vegetable oil.

[0020] Further, the factorization machine is specifically:

[0021]

[0022] where y' FM and y bAFM are the output results of the factorization machine module introducing the bidirectional attention mechanism, w 0 is the weight matrix of the initial feature, x i is the feature of one antioxidant in the two antioxidants in the antioxidant combination, x j is the feature of the other antioxidant in the two antioxidants in the antioxidant combination, v i ⊙v j is the element-wise product of the feature x i and the feature x j P is the weight of the prediction layer, T is the size of the hidden layer, n is the number of antioxidant combinations, and a i,j is the weight factor calculated according to the attention mechanism;

[0023]

[0024] a' ij = h T σ(W(v i ⊙ v j )x i x j + b);

[0025] Among them, h is a model parameter, W is a weight parameter, σ is an activation function, and b is a bias parameter.

[0026] Furthermore, the deep neural network is specifically:

[0027] The high-order non-linear combination part of the input data is processed by a DNN module with an attention mechanism;

[0028] Among them, the attention mechanism is introduced in the hidden layer of the DNN module to calculate the weight of the current layer, and feature extraction is further performed by combining the output result of the previous layer with the weight of the current layer;

[0029] Before inputting the output result of the hidden layer into the output layer, the output result of the hidden layer is activated by an activation function to obtain an activation result, and the activation result is input into the output layer for output:

[0030] y DNN = σ(W L z L-1 + b L ).

[0031] Among them, y DNN is the output of the DNN, σ is the activation function, L represents the L-th layer of the DNN, and W and b respectively represent the weight parameter and bias parameter of this layer.

[0032] Furthermore, the loss function of the pre-trained prediction model is specifically:

[0033]

[0034] Among them, y i,mea is the true value of the synergy corresponding to the input parameter of the prediction model, and y i,pre is the predicted value of the synergy corresponding to the input parameter of the prediction model.

[0035] 2) In the second aspect, the present invention also provides a deep learning-based prediction system for vegetable oil antioxidant combinations, and the specific technical solution is as follows:

[0036] The prediction unit is used to: input the sample data of the vegetable oil sample to be predicted into the pre-trained prediction model to obtain the prediction result of the antioxidant combination corresponding to the vegetable oil sample to be predicted;

[0037] The construction process of the pre-trained prediction model is as follows:

[0038] Collect the data features of experimental vegetable oils and the data features of antioxidant combinations, and classify and integrate all data features to construct training set and test set samples;

[0039] Input the training set into the prediction model for model training. The prediction model includes a data preprocessing module, and the main body of the model is a combined model of a factorization machine with an attention mechanism and a deep neural network;

[0040] Verify the prediction model through the test set, optimize the parameters of the prediction model according to the verification result, and use the optimized prediction model as the prediction model in the next model training until the prediction model is verified through the test set, and use the verified prediction model as the pre-trained prediction model.

[0041] 3) Thirdly, the present invention also provides an electronic device. The electronic device includes a processor, the processor is coupled with a memory, and at least one computer program is stored in the memory. The at least one computer program is loaded and executed by the processor so that the electronic device implements any one of the above methods.

[0042] 4) Fourthly, the present invention also provides a computer-readable storage medium. At least one computer program is stored in the computer-readable storage medium. The at least one computer program is loaded and executed by a processor so that a computer implements any one of the above methods.

[0043] It should be noted that for the beneficial effects obtained by the technical solutions and corresponding possible implementation manners of the second to fourth aspects of the present invention, reference may be made to the technical effects of the first aspect and its corresponding possible implementation manners above, which will not be elaborated here. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] By reading the detailed description of the non-restrictive embodiments with reference to the following drawings, other features, purposes and advantages of the present invention will become more obvious:

[0045] Figure 1 It is a schematic flowchart of a method for predicting antioxidant combinations of vegetable oils based on deep learning according to an embodiment of the present invention;

[0046] Figure 2Schematic diagram of the model architecture of a prediction method for a combination of vegetable oil antioxidants based on deep learning according to an embodiment of the present invention;

[0047] Figure 3 Schematic diagram of the antioxidant feature extraction module of a prediction method for a combination of vegetable oil antioxidants based on deep learning according to an embodiment of the present invention;

[0048] Figure 4 Schematic diagram of the vegetable oil feature extraction module of a prediction method for a combination of vegetable oil antioxidants based on deep learning according to an embodiment of the present invention;

[0049] Figure 5 Schematic diagram of the FM module of a prediction method for a combination of vegetable oil antioxidants based on deep learning according to an embodiment of the present invention;

[0050] Figure 6 Schematic diagram of the DNN module of a prediction method for a combination of vegetable oil antioxidants based on deep learning according to an embodiment of the present invention;

[0051] Figure 7 Schematic diagram of the attention mechanism of a prediction method for a combination of vegetable oil antioxidants based on deep learning according to an embodiment of the present invention;

[0052] Figure 8 Schematic diagram of the architecture of an electronic device according to the present invention. Detailed implementation manners

[0053] To make the objectives, technical solutions and advantages of the present invention clearer, the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings.

[0054] As Figure 1 shown, a prediction method for a combination of vegetable oil antioxidants based on deep learning according to an embodiment of the present invention includes the following steps:

[0055] S1. Input the sample data of the vegetable oil sample to be predicted into the pre-trained prediction model to obtain the prediction result of the antioxidant combination corresponding to the vegetable oil sample to be predicted;

[0056] The construction process of the pre-trained prediction model is as follows:

[0057] Collect the data of experimental vegetable oils and the data of antioxidant combinations, and classify and integrate all the data to construct a training set sample;

[0058] Input the training set into the prediction model for model training, and use the preprocessed data features as the input parameters of the prediction model. The prediction model includes a data preprocessing module, and the main body of the model is a combined model of a factorization machine with an attention mechanism and a deep neural network;

[0059] Verify the prediction model using a test set, optimize the parameters of the prediction model according to the verification results, and use the optimized prediction model as the prediction model in the next model training until the prediction model is verified by the test set, and use the verified prediction model as the pre-trained prediction model.

[0060] The beneficial effects of a method for predicting a combination of vegetable oil antioxidants based on deep learning provided by the present invention are as follows:

[0061] This solution provides a method that can determine antioxidants corresponding to different vegetable oils according to the characteristics of different vegetable oils. Through the method mentioned in this solution and combined with a pre-trained prediction model, the accuracy and efficiency of determining antioxidant compositions are improved.

[0062] Example 1, as Figure 2 shown, the construction process of the pre-trained prediction model is specifically as follows:

[0063] 1. Construct a training set and a test set for the prediction model

[0064] Obtain samples of different types of vegetable oils, samples of different combinations of antioxidants, and antioxidant vegetable oil samples obtained by adding different combinations of antioxidants to different types of vegetable oils in different addition amounts, and construct a training set and a test set.

[0065] Preprocess all the above training samples, and the specific process is as follows:

[0066] It should be noted that the training set is composed of multiple training sample groups. Each training sample group contains one or a group of antioxidants (such as antioxidant 1 and antioxidant 2), the addition amounts of this group of antioxidants (the concentration of antioxidant 1 and the concentration of antioxidant 2), and the type of vegetable oil (peanut oil, soybean oil, etc.). The above content is also the data input to the pre-trained prediction model during subsequent actual applications. Determine the label corresponding to each training sample group. The label is the synergy degree S of the antioxidant combination calculated based on the shelf life of the vegetable oil. S>1 indicates a synergistic effect. This label is a data label, and the data labels are divided into two categories, represented by 0 and 1 respectively. Among them, 1 represents a synergistic effect, and 0 represents no synergistic effect.

[0067] The synergy degree S is calculated according to the Bliss model, and the theoretical value (E T ) of the antioxidant combination effect is calculated through the effect values of each antioxidant in the antioxidant combination, and then compared with the experimental value (E E ), as shown in the following formula:

[0068]

[0069] S = E T / E E ;

[0070] Wherein, S represents the synergy degree of the antioxidant combination, and E i represents the effect value of antioxidant component i. When S > 1, the antioxidant combination exhibits a synergistic effect. The larger the S value, the stronger the synergistic effect.

[0071] Since the description of the antioxidant combination and the type of vegetable oil are the keys for the prediction model to make predictions, and the composition of vegetable oil is complex, with dozens of fatty acid compositions, if the key features are not screened, it will cause problems such as low model learning efficiency, poor generalization ability, and low accuracy. Therefore, in this application, the SelflES method is used for the training set to convert the antioxidant molecules and vegetable oil lipid molecules in any training sample group into strings. Compared with the previous SMILES representation method, SelflES can avoid the ambiguity of rings and cross-links in molecules, thereby improving the accuracy of molecular expression.

[0072] The antioxidant combination features are obtained by the antioxidant feature extraction module. As Figure 3 shown, the SelflES string of the antioxidant is encoded and normalized. The Embedding() function is called to establish an embedding layer to reduce the feature dimension, which can improve the convergence speed and accuracy of the model. The processed string after normalization is processed through this embedding layer to obtain the antioxidant combination features.

[0073] It should be noted that SelflES is an open-source project for representing and operating chemical molecules, which can convert molecules into unambiguous strings that can be understood by machine learning algorithms. This representation method enables the molecular structure to be analyzed and predicted by a deep learning model like natural language.

[0074] The Embedding() function refers to a way of mapping discrete data into continuous vector representations, so as to capture the potential relationships and structures between data.

[0075] As Figure 4 shown is the vegetable oil feature processing module, which consists of a 5-layer neural network with 3 hidden layers and the number of neurons [350, 500, 350]. By introducing the self-attention mechanism, it autonomously learns the features of the important fatty acid composition of vegetable oil and reasonably allocates weights to avoid the interference of less relevant lipid compositions on subsequent predictions, and improves the generalization ability and prediction accuracy of the model. That is, the vegetable oil features are determined in the above way.

[0076] To ensure that the data input into the prediction module is more conducive to subsequent data recognition and analysis, it is necessary to normalize the vegetable oil characteristics and antioxidant combination characteristics, that is, map the values in all feature matrices to the range of [0, 1].

[0077]

[0078] where x * is the data after normalization processing, x is the original data, x min is the minimum value of the sample data, x max is the maximum value of the sample data.

[0079] After normalizing all the training samples, input samples can be obtained for the training of subsequent prediction models.

[0080] 2. Training of the prediction model

[0081] Based on whether the added antioxidant combination has a synergistic effect on the oxidation stability of vegetable oil, a supervised classification prediction model is established. The data labels are divided into two categories, represented by 0 and 1 respectively. Among them, 1 represents having a synergistic effect, and 0 represents not having a synergistic effect;

[0082] Based on the strength of the synergistic effect of the added antioxidant combination on the oxidation stability of vegetable oil, a supervised regression prediction model is established. The data label is the degree of synergy, which is used to measure the influence degree of the antioxidant combination on the oxidation stability of vegetable oil.

[0083] Based on the Bi-Attention mechanism, Factorization Machine (FM), and Deep Neural Network (DeepNN), a combined prediction model is built. FM is used for low-order combinations between features, and Deep NN is used for high-order combinations between features. The FM module and the Deep module share the Feature Embedding part. After all features are fused, they are input into the combined prediction module, which is composed of a neural network. There are a total of 5 layers, 3 hidden layers, and the number of neurons is [500, 500, 350].

[0084] Ordinary deep learning models generally do not consider the interaction relationships between features, which does not conform to some actual application backgrounds. For example, the association between the type and amount of antioxidants added, the association between the fatty acid composition and content in vegetable oil, the interaction relationship between antioxidants and fatty acids, and so on are all key factors affecting the performance of antioxidant combinations on the oxidation stability of vegetable oil. The interaction relationship between features directly affects the model performance.

[0085] The introduction of the bidirectional attention mechanism enables the model to autonomously learn the interaction relationships between features, allowing the model to notice the correlations between different parts of the input data, such as the association between antioxidant types and addition amounts, and the association between the fatty acid compositions of vegetable oils, thereby improving the performance and interpretability of the model.

[0086] The final prediction result can be expressed as:

[0087]

[0088] Among them, y' FM refers to the output result obtained after being processed by the FM module, and y' DNN refers to the output result obtained after being processed by the DNN module. sigmoid(y' FM +y' DNN ) refers to fitting the output results of y' FM and y' DNN .

[0089] The introduction of the bidirectional attention mechanism is achieved by adding an attention layer (AttentionLayer) to the FM module and the DNN module. It can simultaneously learn the interaction relationships within and between input features, allowing the model to notice the correlations between different parts of the input data, such as the association between antioxidant types and addition amounts, and the association between the fatty acid compositions of vegetable oils, thereby improving the interpretability and accuracy of the model. The attention mechanism is as Figure 7 shown.

[0090] The attention effect is achieved through an attention function. The attention function:

[0091] Att = Aggregate(W a ·V)

[0092] Among them, W a is the attention weight, and V (Value) is the reagent data from which information needs to be extracted.

[0093] W a is usually obtained by calculating the similarity between Query (Q) and Key (K), and determines how much information to extract from each Value. The weight:

[0094]

[0095] Among them, Q is the request for the information to be obtained, K is the measure of the correlation with Q, and d_k is the dimension of Q and K.

[0096] Such as Figure 5As shown in the figure, the FM module realizes feature vectorization, introduces cross-term features, and decomposes the parameter matrix. The bi-attention mechanism is introduced to improve the interpretability and expressive ability of model prediction. The expression of the model is

[0097]

[0098] where w 0 is the weight matrix of the feature, P is the weight of the prediction layer, T is the size of the hidden layer, and (v i ⊙ v j ) is the element-wise product of the combined features of x i and x j . a ij is the weight factor calculated by the attention mechanism, and the calculation method is as follows

[0099] a' ij = h T σ(W(v i ⊙ v j )x i x j + b)

[0100]

[0101] where h is the model parameter, W is the weight parameter, σ is the activation function, and b is the bias parameter. The dropout method is adopted in the interaction layer to avoid overfitting.

[0102] As Figure 6 shown in the figure, the Deep NN (DNN) module: builds a deep neural network and introduces the attention mechanism in the hidden layer to extract and process the high-order non-linear combination part of the features. The Dense Embedding is input into the Hidden Layer in a fully connected manner, where the Dense Embeddings can avoid the problem of parameter explosion in the DNN.

[0103] The output of the Embedding layer is to integrate all the embedding vectors corresponding to the categorical features and input them into the DNN. The input feature matrix is

[0104] z i = [v 1 , v 2 , …, v m

[0105] where v i represents the embedding of the i-th field, and m is the number of fields. After being input into the DNN, the calculation of the hidden layer is ​

[0106] z L = σ(W L-1 z L-1 + b L-1 )

[0107] where σ represents the activation function, L represents the L-th layer of the DNN, and W and b represent the weight matrix parameters and bias parameters of this layer, respectively.

[0108] Finally, it enters the DNN part and the output is activated using the sigmoid activation function:

[0109]

[0110] The final output is

[0111] y DNN = σ(W L z L-1 + b L ).

[0112] The output of the FM module and the output of the DNN module are integrated together through the concatenate() function and input into the combined prediction module.

[0113] Loss function: Mean Square Error (MSE) is used as the error between the true value and the predicted value of the antioxidant synergy score in this model.

[0114]

[0115] The output result of the prediction model is the synergy degree (regression model) and whether there is synergy (classification model).

[0116] Optimize the key parameters of the prediction model (i.e., the hyperparameters in Table 1) to improve the prediction accuracy of the model.

[0117] That is, use the Adam (Adaptive Movement Estimation) optimization algorithm, which is an optimization algorithm improved on the basis of the gradient descent algorithm and can update the weights during the model training process.

[0118] In addition, it should be particularly noted that the setting results of the hyperparameters of the prediction model involved in this solution are shown in Table 1 below.

[0119] Table 1

[0120]

[0121]

[0122] 2) Second aspect, the present invention also provides a prediction system for antioxidant combinations in vegetable oils based on deep learning. The specific technical solution is as follows:

[0123] The prediction unit is used to: input the sample data of the vegetable oil sample to be predicted obtained into a pre-trained prediction model to obtain the prediction result of the antioxidant combination corresponding to the vegetable oil sample to be predicted;

[0124] The construction process of the pre-trained prediction model is as follows:

[0125] Collect the data features of experimental vegetable oils and the data features of antioxidant combinations, and classify and integrate all the data features to construct a training set sample;

[0126] Perform corresponding preprocessing on the data features of different categories in the training set sample, and use the preprocessed data features as the input parameters of the prediction model. The prediction model is a combined model of a factorization machine and a deep neural network;

[0127] Verify the prediction model through a test set, optimize the parameters of the prediction model according to the verification result, and use the optimized prediction model as the prediction model in the next model training until the prediction model is verified through the test set, and use the verified prediction model as the pre-trained prediction model.

[0128] In the above embodiments, although the steps are numbered S1, S2, etc., these are only specific embodiments given by the present invention. Those skilled in the art can adjust the execution order of S1, S2, etc. according to the actual situation, and this is also within the protection scope of the present invention. It can be understood that in some embodiments, it may include some or all of the above embodiments.

[0129] It should be noted that the beneficial effects of the prediction system for antioxidant combinations in vegetable oils based on deep learning provided in the above embodiments are the same as those of the prediction method for antioxidant combinations in vegetable oils based on deep learning, and will not be elaborated here. In addition, when the system provided in the above embodiments realizes its functions, only the above-mentioned functional module division is used for illustration. In actual applications, the above functions can be allocated to different functional modules according to needs, that is, the system is divided into different functional modules according to the actual situation to complete all or part of the functions described above. In addition, the system provided in the above embodiments and the method embodiments belong to the same concept, and the specific implementation process is detailed in the method embodiments and will not be elaborated here.

[0130] Such as Figure 8As shown in the figure, an electronic device 300 according to an embodiment of the present invention includes a processor 320, the processor 320 is coupled to a memory 310, and at least one computer program 330 is stored in the memory 310. The at least one computer program 330 is loaded and executed by the processor 320 so that the electronic device 300 implements any one of the above methods. Specifically:

[0131] The electronic device 300 may vary greatly due to configuration or performance differences, and may include one or more processors 320 (Central Processing Units, CPUs) and one or more memories 310. Among them, at least one computer program 330 is stored in the one or more memories 310, and the at least one computer program 330 is loaded and executed by the one or more processors 320 so that the electronic device 300 implements a method for predicting a combination of vegetable oil antioxidants based on deep learning provided in the above embodiment. Of course, the electronic device 300 may also have components such as a wired or wireless network interface, a keyboard, and an input / output interface for input and output. The electronic device 300 may also include other components for implementing device functions, which will not be elaborated here.

[0132] A computer-readable storage medium according to an embodiment of the present invention stores at least one computer program, and the at least one computer program is loaded and executed by a processor so that a computer implements any one of the above methods.

[0133] Optionally, the computer-readable storage medium may be a read-only memory (ROM), a random access memory (RAM), a compact disc read-only memory (CD-ROM), a magnetic tape, a floppy disk, and an optical data storage device, etc.

[0134] In an exemplary embodiment, a computer program product or a computer program is further provided. The computer program product or the computer program includes computer instructions, and the computer instructions are stored in a computer-readable storage medium. The processor of the electronic device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions so that the electronic device executes any one of the above methods.

[0135] It should be noted that the terms "first", "second", etc. in the description and claims of the present application are used to distinguish similar objects, and do not represent a specific order or sequence. In appropriate cases, the order of use of similar objects may be interchanged so that the embodiments of the present application described here can be implemented in an order other than the illustrated or described order.

[0136] Those skilled in the art of the present technology know that the present invention can be implemented as a system, method or computer program product. Therefore, the present disclosure can be specifically implemented in the following forms, namely: it can be completely hardware, can also be completely software (including firmware, resident software, microcode, etc.), and can also be in the form of a combination of hardware and software, which is generally referred to as "circuit", "module" or "system" in this article. In addition, in some embodiments, the present invention can also be implemented in the form of a computer program product in one or more computer-readable media, which contain computer-readable program codes.

[0137] Any combination of one or more computer-readable media can be adopted. The computer-readable media can be computer-readable signal media or computer-readable storage media. The computer-readable storage media can be, for example - but not limited to - electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices or components, or any combination of the above. More specific examples (non-exhaustive list) of the computer-readable storage media include: electrical connections with one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the above. In this document, the computer-readable storage media can be any tangible medium that contains or stores a program, and this program can be used by or in combination with an instruction execution system, device or component.

[0138] Although the embodiments of the present invention have been shown and described above, it can be understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those of ordinary skill in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A method for predicting a combination of vegetable oil antioxidants based on deep learning, characterized in that: include: Inputting the acquired sample data of the vegetable oil sample to be predicted into the pre-trained prediction model to obtain the prediction result of the antioxidant combination corresponding to the vegetable oil sample to be predicted; The construction process of the pre-trained prediction model is as follows: Collect the experimental vegetable oil data features and antioxidant combination data features, classify and integrate all data features to construct training sets and test sets; Inputting the training set into a prediction model for model training, wherein the prediction model includes a data preprocessing module, and the model body is a factor decomposition machine with an attention mechanism and a deep neural network combination model; The prediction model is verified through the test set, and the parameters of the prediction model are optimized according to the verification result, and the optimized prediction model is used as the prediction model in the next model training, until the prediction model is verified through the test set, and the verified prediction model is used as the pre-trained prediction model; The factorization machine is specifically: Among them, y′ FM and bAFM is the output result of the factorization machine module that introduces the bidirectional attention mechanism, w0 is the weight matrix of the initial feature, x i is a characteristic of one of the two antioxidants in the antioxidant combination, x j is a characteristic of the other antioxidant in the antioxidant combination, v i ⊙v j is feature x i With feature x j The element-wise product of , P is the weight of the prediction layer, T is the size of the hidden layer, n is the number of antioxidant combinations, a i,j is the weight factor calculated according to the attention mechanism, w i is the weight matrix of the i-th initial feature; a′ ij =h T σ(W(v i ⊙v j )x i x j +b); Among them, h is the model parameter, W is the weight parameter, σ is the activation function, and b is the bias parameter; Among them, a combined prediction model is built based on the bidirectional attention mechanism Bi-Attention, factorization machine FM and deep neural network Deep NN. FM is used for low-order combination between features, and Deep NN is used for high-order combination between features. The FM module and the Deep module share the Feature Embedding part. All features are fused and input into the combined prediction module, which is composed of a neural network. Among them, the introduction of the bidirectional attention mechanism is achieved by adding an attention layer AttentionLayer in the FM module and the DNN module.

2. A method for predicting a combination of vegetable oil antioxidants based on deep learning according to claim 1, characterized in that: The data preprocessing module is specifically used to preprocess the training set input into the prediction model, and the specific process is: The category features and numerical features in the training set are respectively converted into strings and the conversion results are encoded to obtain category feature encoding results and numerical feature encoding results corresponding to the category features, and the category feature encoding results and the numerical feature encoding results are standardized and normalized to obtain preprocessed data features.

3. The method for predicting a combination of vegetable oil antioxidants based on deep learning according to claim 1, characterized in that: The process of acquiring the data features of the antioxidant combination is as follows: After encoding the molecular character string of any antioxidant in the antioxidant combination, feature extraction is performed through an antioxidant molecular feature extraction module to obtain antioxidant molecular features, and the antioxidant molecular features are used as data features of the antioxidant combination.

4. The method for predicting a combination of vegetable oil antioxidants based on deep learning according to claim 1, characterized in that: The process of acquiring the experimental vegetable oil data features is as follows: Input the basic parameters of the experimental vegetable oil into the vegetable oil feature extraction module to obtain the experimental vegetable oil data features; The vegetable oil feature extraction module is an attention neural network model, which includes three hidden layers, and an attention mechanism is introduced in the first and third hidden layers respectively.

5. The method for predicting a combination of vegetable oil antioxidants based on deep learning according to claim 1, characterized in that: The deep neural network is specifically: The DNN module with attention mechanism processes the high-order nonlinear combination of input data. Among them, the attention mechanism is introduced into the hidden layer of the DNN module to calculate the weight of the current layer, and the output result of the previous layer is combined with the weight of the current layer to further extract features; Before the output result of the hidden layer is input to the output layer, the output result of the hidden layer is activated by the activation function to obtain the activation result, and the activation result is input to the output layer for output: and DNN =σ(W L With L-1 +b L ); Among them, y DNN is the output of DNN, σ is the activation function, L represents the Lth layer of DNN, W and b represent the weight parameter and bias parameter of this layer respectively.

6. The method for predicting a combination of vegetable oil antioxidants based on deep learning according to claim 1, characterized in that: The loss function of the pre-trained prediction model is specifically: Among them, y i,mea is the true value of the synergy corresponding to the input parameters of the prediction model, y i,pre is the synergistic prediction value corresponding to the input parameters of the prediction model, and MSE is the loss value.

7. A vegetable oil antioxidant combination prediction system based on deep learning, using a vegetable oil antioxidant combination prediction method based on deep learning as claimed in claim 1, characterized in that: The system includes: The prediction unit is used to: input the acquired sample data of the vegetable oil sample to be predicted into the pre-trained prediction model to obtain the prediction result of the antioxidant combination corresponding to the vegetable oil sample to be predicted; The construction process of the pre-trained prediction model is as follows: Collect the experimental vegetable oil data features and antioxidant combination data features, classify and integrate all data features to construct training sets and test sets; Inputting the training set into a prediction model for model training, wherein the prediction model includes a data preprocessing module, and the model body is a factor decomposition machine with an attention mechanism and a deep neural network combination model; The prediction model is verified through the test set, and the parameters of the prediction model are optimized according to the verification result, and the optimized prediction model is used as the prediction model in the next model training, until the prediction model is verified through the test set, and the verified prediction model is used as the pre-trained prediction model.

8. An electronic device, characterized in that: The electronic device comprises a processor, the processor is coupled to a memory, at least one computer program is stored in the memory, and the at least one computer program is loaded and executed by the processor so that the electronic device implements the method according to any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores at least one computer program, and the at least one computer program is loaded and executed by a processor so that a computer implements the method according to any one of claims 1 to 6.

Images