A distributed feature-based molecular reaction model construction method, device and equipment
By constructing a molecular reaction model based on distributed features and utilizing deep learning and neural network models, the problems of easy data tampering and transmission security in molecular refining models are solved, enabling rapid modeling and efficient prediction, and simplifying the modeling process.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- RICHFIT INFORMATION TECH
- Filing Date
- 2024-12-18
- Publication Date
- 2026-06-19
AI Technical Summary
Molecular refining model data is easily tampered with, and the security of the transmission process is difficult to guarantee. Traditional model construction is time-consuming and relies on professional knowledge, making it difficult to quickly model and predict.
A molecular reaction model construction method based on distributed features is adopted. Deep learning and neural network models are used for data feature extraction and model training to construct a distributed molecular reaction model. Data processing and prediction are performed through BP neural network and convolutional neural network.
It improves data security and accuracy, reduces noise data and human tampering interference, simplifies the modeling process, improves computational efficiency and predictive performance, and reduces reliance on professional knowledge.
Smart Images
Figure CN122245466A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of molecular refining technology, particularly to the field of computer-based molecular refining reaction model construction technology, and especially to a method, apparatus and equipment for constructing molecular reaction models based on distributed features. Background Technology
[0002] Molecular refining models typically contain key process parameters, production indicators, and product quality information, which are crucial for the production and operation of refineries. The accuracy and completeness of this data directly affect production efficiency, product quality, and production safety.
[0003] Molecular refining model data involves multiple stages and systems, and may be generated and maintained by different devices, sensors, databases, or personnel. This diversity of data sources increases the likelihood of data alteration or tampering. Furthermore, during data transmission and storage, data may be vulnerable to cyberattacks, tampering, or accidental damage. Especially in distributed environments, data transmission paths are more complex, making security difficult to guarantee.
[0004] Furthermore, traditional molecular reaction models are typically constructed based on physical and chemical principles, such as reaction rate constants and activation energies. However, these models often require extensive experimental data and specialized knowledge, and the construction and application of these models often involve multiple iterations, resulting in a lengthy process.
[0005] In recent years, with the development of distributed computing and deep learning technologies, methods for constructing molecular reaction models based on distributed features have received widespread attention. Summary of the Invention
[0006] To achieve rapid modeling and prediction of molecular reactions based on distributed computing and deep learning, effectively ensure data security and accuracy, reduce interference from factors such as noise data and human tampering during data transmission and processing, and improve the computational efficiency and prediction performance of the model; and at the same time enrich process technology routes and increase the selection space, this invention provides a method, apparatus and equipment for constructing molecular reaction models based on distributed features.
[0007] In a first aspect, embodiments of the present invention provide a method for constructing a molecular reaction model based on distributed features, which may include:
[0008] Obtain a molecular reaction dataset; wherein, the molecular reaction dataset may include: reactant data, product data, and reaction condition information data;
[0009] Feature extraction is performed on the reactant data and the product data to obtain distributed reactant feature data and product feature data;
[0010] A molecular reaction model is constructed based on the reactant characteristic data, the product characteristic data, and the reaction condition information data.
[0011] In one embodiment, constructing a molecular reaction model based on the reactant feature data, the product feature data, and the reaction condition information data may include: constructing a training sample set based on the reactant feature data, the product feature data, and the reaction condition information data to train a BP neural network molecular reaction model; using the reaction product feature data and reaction condition information data included in each sample data in the training sample set as input to the BP neural network molecular reaction model, using the product feature data included in each sample data as output to the BP neural network molecular reaction model, and optimizing the node weights and biases of the hidden layer based on gradient descent after passing through the hidden layer of the BP neural network molecular reaction model to obtain a trained BP neural network molecular reaction model.
[0012] In one embodiment, feature extraction of the reactant data and product data to obtain distributed reactant feature data and product feature data may include:
[0013] The molecular structure of the reactant data and product data is characterized using the structure-guided lumped method to obtain the molecular structure feature vectors of the reactant data and the product data, respectively.
[0014] The molecular structure feature vectors of the reactant data and the product data are respectively input into a pre-constructed convolutional neural network model, and the data output by the pooling layer in the convolutional neural network model are extracted as distributed reactant feature data and product feature data, respectively.
[0015] In one embodiment, the convolutional neural network model includes a reactant convolutional neural network model and a product convolutional neural network model;
[0016] The reactant convolutional neural network model was pre-trained using the following method:
[0017] Construct a reactant training sample set; wherein, each sample in the reactant training sample set includes a reactant molecular structure feature vector matrix composed of molecular structure feature vectors corresponding to all reactants in a set of reactant data;
[0018] The reactant convolutional neural network model is trained using samples from the reactant training sample set, wherein the reactant molecular structure feature vector matrix serves as the input and output of the reactant convolutional neural network model, respectively; after passing the reactant molecular structure feature vector matrix through a correlation layer, the trained reactant convolutional neural network model is obtained.
[0019] The product convolutional neural network model is pre-trained using the following method:
[0020] Construct a product training sample set; wherein, each sample in the product training sample set includes a product molecular structure feature vector matrix composed of molecular structure feature vectors corresponding to all products in a set of product data;
[0021] The product convolutional neural network model is trained using samples from the product training sample set, wherein the product molecular structure feature vector matrix serves as the input and output of the product convolutional neural network model, respectively; after passing the product molecular structure feature vector matrix through a correlation layer, the trained product convolutional neural network model is obtained.
[0022] In one embodiment, the reactant molecular structure feature vector matrix, after passing through a correlation layer, yields a trained reactant convolutional neural network model, which may include:
[0023] The reactant molecular structure feature vector matrix is passed through the convolutional layer of the reactant convolutional neural network model and then subjected to a dot product operation. The reactant molecular structure feature vector matrix is moved on the convolutional kernel of the convolutional layer to output the reactant feature mapping of the reactant molecular structure feature vector matrix. The reactant feature mapping is added to the bias term of the convolutional layer and then substituted into the activation function to obtain the final reactant feature mapping.
[0024] The final reactant feature map is dimensionality reduced in the pooling layer of the reactant convolutional neural network model. The dimensionality-reduced reactant feature map is then transposed and convolved in the deconvolution layer of the reactant convolutional neural network model. The weights of the backpropagation algorithm are updated using the reactant loss function defined by the actual and output values to obtain the trained reactant convolutional neural network model.
[0025] After the product molecular structure feature vector matrix is passed through a correlation layer, a trained product convolutional neural network model is obtained, including:
[0026] The product molecular structure feature vector matrix is passed through the convolutional layer of the product convolutional neural network model and then subjected to a dot product operation. The product molecular structure feature vector matrix is moved on the convolutional kernel of the convolutional layer to output the product feature mapping of the product molecular structure feature vector matrix. The product feature mapping is added to the bias term of the convolutional layer and then substituted into the activation function to obtain the final product feature mapping.
[0027] The final product feature map is dimensionality reduced in the pooling layer of the product convolutional neural network model. The dimensionality-reduced product feature map is then transposed and convolved in the deconvolution layer of the product convolutional neural network model. The weights of the backpropagation algorithm are updated using the product loss function defined by the actual and output values to obtain the trained product convolutional neural network model.
[0028] In one embodiment, prior to training the BP neural network molecular reaction model, the following may also be included:
[0029] The reactant feature data included in each sample data is preprocessed to convert the reactant feature data into a one-dimensional vector of reactants by element-wise concatenation, and the one-dimensional vector of reactants and the reaction condition information data are used as inputs to the BP neural network molecular reaction model.
[0030] The product feature data included in each sample data is preprocessed to convert the product feature data into a one-dimensional product vector by element concatenation, and the one-dimensional product vector is used as the output of the BP neural network molecular reaction model.
[0031] Secondly, embodiments of the present invention provide a method for predicting molecular reaction products based on distributed features. This method predicts molecular reaction product data based on a molecular reaction model pre-constructed using the molecular reaction model construction method based on distributed features described in the first aspect.
[0032] In one embodiment, the prediction method in this embodiment may specifically include:
[0033] Acquire reactant data and reaction condition information from a molecular reaction dataset;
[0034] Feature extraction is performed on the reactant data to obtain distributed reactant feature data;
[0035] Based on the reactant characteristic data and the reaction condition information data, the data is input into the pre-constructed molecular reaction model to output product characteristic data.
[0036] Product data of molecular reactions can be predicted based on the product characteristic data.
[0037] In one embodiment, feature extraction of the reactant data to obtain distributed reactant feature data may include:
[0038] The molecular structure of the reactant data is characterized using the structure-guided lumped method to obtain the molecular structure feature vectors of the reactant data.
[0039] The molecular structure feature vectors of the reactant data are input into a pre-constructed convolutional neural network model, and the data output from the pooling layer of the convolutional neural network model is extracted as distributed reactant feature data.
[0040] In one embodiment, predicting product data for a molecular reaction based on the product characteristic data may include:
[0041] Based on the product feature data and the pre-built convolutional neural network model, as well as the structure-guided lumped method, the product data of the molecular reaction are predicted.
[0042] Thirdly, embodiments of the present invention provide a molecular reaction model construction apparatus based on distributed features, which may include:
[0043] The first acquisition module is used to acquire a molecular reaction dataset; wherein, the molecular reaction dataset includes: reactant data, product data, and reaction condition information data;
[0044] The first feature extraction module is used to extract features from the reactant data and the product data to obtain distributed reactant feature data and product feature data.
[0045] A construction module is used to construct a molecular reaction model based on the reactant characteristic data, the product characteristic data, and the reaction condition information data.
[0046] In one embodiment, the construction module is specifically used to: construct a training sample set based on the reactant feature data, the product feature data, and the reaction condition information data to train the BP neural network molecular reaction model; the reaction product feature data and reaction condition information data included in each sample data in the training sample set are used as the input of the BP neural network molecular reaction model, and the product feature data included in each sample data is used as the output of the BP neural network molecular reaction model; after passing through the hidden layer of the BP neural network molecular reaction model, the node weights and biases of the hidden layer are optimized based on the gradient descent method to obtain the trained BP neural network molecular reaction model.
[0047] Fourthly, embodiments of the present invention provide a molecular reaction product prediction device based on distributed features, comprising:
[0048] The product data prediction module is used to predict the product data of molecular reactions based on pre-built molecular reaction models.
[0049] The molecular reaction model is pre-constructed according to the molecular reaction model construction method based on distributed features described in the first aspect.
[0050] In one embodiment, a molecular reaction product prediction device based on distributed characteristics includes:
[0051] The second acquisition module is used to acquire reactant data and reaction condition information data from the molecular reaction dataset.
[0052] The second feature extraction module is used to extract features from the reactant data to obtain distributed reactant feature data;
[0053] The product characteristic data determination module is used to input the reactant characteristic data and the reaction condition information data into a pre-constructed molecular reaction model to output product characteristic data.
[0054] The product data prediction module is used to predict the product data of the molecular reaction based on the product characteristic data.
[0055] The molecular reaction model is pre-constructed according to the molecular reaction model construction method based on distributed features described in the first aspect.
[0056] Fifthly, embodiments of the present invention provide a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the molecular reaction model construction method based on distributed features as described in the first aspect, or the molecular reaction product prediction method based on distributed features as described in the second aspect.
[0057] In a sixth aspect, embodiments of the present invention provide a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the molecular reaction model construction method based on distributed features as described in the first aspect, or the molecular reaction product prediction method based on distributed features as described in the second aspect.
[0058] The beneficial effects of the above-described technical solutions provided in the embodiments of the present invention include at least the following:
[0059] This invention provides a method, apparatus, and device for constructing molecular reaction models based on distributed features. The method utilizes a deep neural network model to achieve distributed feature extraction, distributing the molecular reaction model data across different nodes for processing. Data transmission and computation between nodes enable rapid modeling and prediction of molecular reactions. This effectively ensures data security and accuracy, reduces interference from noise data and human tampering during data transmission and processing, improves the model's computational efficiency and predictive performance, simplifies the modeling process, reduces reliance on specialized knowledge during model construction, and enhances modeling efficiency.
[0060] Other features and advantages of the invention will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the written description and the accompanying drawings.
[0061] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description
[0062] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:
[0063] Figure 1 This is a flowchart of a molecular reaction model construction method based on distributed features provided in an embodiment of the present invention;
[0064] Figure 2 The structure diagram for molecular structure characterization provided in the embodiment of the present invention is shown.
[0065] Figure 3 This is an example of a feature vector matrix of reactant molecular structures provided in an embodiment of the present invention;
[0066] Figure 4 This is the convolutional neural network model architecture structure provided in the embodiments of the present invention;
[0067] Figure 5 The structure of the BP neural network molecular reaction model provided in this embodiment of the invention;
[0068] Figure 6 This is a schematic diagram of the molecular reaction model construction device based on distributed features provided in an embodiment of the present invention;
[0069] Figure 7 This is a flowchart of the molecular reaction product prediction method based on distributed features provided in the embodiments of the present invention;
[0070] Figure 8 The flowchart below shows a detailed method for predicting molecular reaction products based on distributed features, as provided in this embodiment of the invention.
[0071] Figure 9 This is a schematic diagram of the molecular reaction product prediction device based on distributed features provided in an embodiment of the present invention. Detailed Implementation
[0072] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.
[0073] The inventors discovered in their practical work that distributed feature representation can improve the security of molecular reaction models by modeling and predicting molecular reactions. Utilizing neural network models can improve modeling efficiency, enabling rapid calculation and accurate prediction of molecular reactions. To avoid the increased possibility of data alteration or tampering due to the diversity of data sources, and the risk of data being attacked, tampered with, or accidentally damaged during data transmission and storage, this invention provides a method, apparatus, and device for constructing molecular reaction models based on distributed features.
[0074] Example 1
[0075] Embodiment 1 of this invention provides a method for constructing a molecular reaction model based on distributed features, referring to... Figure 1 As shown, the construction method may include the following steps:
[0076] Step S11: Obtain the molecular reaction dataset; wherein, the molecular reaction dataset includes: reactant data, product data, and reaction condition information data; specifically, the reactant data includes the molecular composition and molecular concentration data of the reactants, the product data includes the molecular composition and molecular concentration data of the products, and the reaction condition information data includes: reaction temperature, reaction pressure, reactor volume, and reactor flow rate.
[0077] The molecular reaction dataset in this embodiment of the invention is collected from a secondary processing device for a certain molecular refining process. Therefore, the constructed molecular reaction model can also be used for product prediction of this device.
[0078] Step S12: Extract features from reactant and product data to obtain distributed reactant and product feature data.
[0079] In this step, the reactant and product data are first characterized by molecular structure based on the structure-guided lumped method to obtain molecular structure feature vectors for the reactant and product data respectively. Then, the molecular structure feature vectors of the reactant and product data are input into a pre-built convolutional neural network model, and the data output from the pooling layer in the convolutional neural network model is extracted as distributed reactant feature data and product feature data respectively.
[0080] The embodiments of the present invention are based on Figure 2 Molecular structure characterization was performed using the structure-guided lumps shown, as illustrated in Table 1. Figure 2 The physical meaning of each structural unit is explained as follows:
[0081] Table 1 Physical meaning of each structural unit
[0082]
[0083]
[0084] In this embodiment of the invention, the convolutional neural network model includes a reactant convolutional neural network model and a product convolutional neural network model.
[0085] The reactant convolutional neural network model was pre-trained using the following method:
[0086] Construct a reactant training sample set; each sample in the reactant training sample set includes a reactant molecular structure feature vector matrix composed of molecular structure feature vectors corresponding to all reactants in a set of reactant data; train the reactant convolutional neural network model using samples from the reactant training sample set, wherein the reactant molecular structure feature vector matrix serves as the input and output of the reactant convolutional neural network model, respectively; after passing the reactant molecular structure feature vector matrix through a correlation layer, the trained reactant convolutional neural network model is obtained.
[0087] Specifically, after the reactant molecular structure feature vector matrix passes through a correlation layer, a trained reactant convolutional neural network model is obtained. This process can include: the reactant molecular structure feature vector matrix undergoes a dot product operation after passing through the convolutional layer of the reactant convolutional neural network model; the reactant molecular structure feature vector matrix is then shifted across the convolutional kernel of the convolutional layer to output a reactant feature map of the reactant molecular structure feature vector matrix; the reactant feature map is added to the bias term of the convolutional layer and then substituted into the activation function to obtain the final reactant feature map; the final reactant feature map undergoes dimensionality reduction in the pooling layer of the reactant convolutional neural network model; and the dimensionality-reduced reactant feature map is then transposed and convolved through the deconvolutional layer of the reactant convolutional neural network model. The weights of the backpropagation algorithm are then updated using a reactant loss function defined by the actual and output values to obtain the trained reactant convolutional neural network model.
[0088] The resulting convolutional neural network model is pre-trained using the following method:
[0089] Construct a product training sample set; each sample in the product training sample set includes a product molecular structure feature vector matrix composed of molecular structure feature vectors corresponding to all products in a set of product data; train the product convolutional neural network model with samples in the product training sample set, wherein the product molecular structure feature vector matrix is used as the input and output of the product convolutional neural network model respectively; after the product molecular structure feature vector matrix is passed through a correlation layer, the trained product convolutional neural network model is obtained.
[0090] Specifically, after the product molecular structure feature vector matrix passes through the correlation layer, a trained product convolutional neural network model is obtained. This can be further divided into: the product molecular structure feature vector matrix undergoes a dot product operation after passing through the convolutional layer of the product convolutional neural network model; the product molecular structure feature vector matrix is then shifted across the convolutional kernel of the convolutional layer to output the product feature mapping of the product molecular structure feature vector matrix; the product feature mapping is added to the bias term of the convolutional layer and then substituted into the activation function to obtain the final product feature mapping; the final product feature mapping is dimensionality-reduced in the pooling layer of the product convolutional neural network model, and the dimensionality-reduced product feature mapping is transposed and convolved through the deconvolutional layer of the product convolutional neural network model. The weights of the backpropagation algorithm are then updated using a product loss function defined by the actual and output values to obtain the trained product convolutional neural network model.
[0091] In this embodiment of the invention, distributed feature extraction is performed using reactant data as an example, as detailed below:
[0092] 1. Variable Design: Each set of reactant data variables consists of several molecules, meaning each variable is an N*25 dimensional array (reactant molecular structure feature vector matrix), where N represents the number of constituent molecules, and 25 corresponds to the 24 structure vector segments and molecular concentrations in the structure-guided ensemble. The specific reactant molecular structure feature vector matrix is represented as follows: Figure 3 As shown.
[0093] 2. Construction of the reactant convolutional neural network model: The feature vector matrix of the reactant molecular structure is used as both input and output of the convolutional neural network. The convolutional neural network is then trained, and the output of the pooling layer is extracted as the features of the reactants (reactant feature data).
[0094] Reference Figure 4 As shown, the specific process of constructing a reaction convolutional neural network model using a convolutional neural network is as follows:
[0095] 1) Input / Output: Based on the reaction sample data of molecular composition (number of constituent molecules N*25*1), construct the input layer, that is, the number of neurons is N*25, where N is the number of constituent molecules.
[0096] 2) Constructing a convolutional neural network model for reactants
[0097] Convolutional Layer: Defines a convolution kernel and performs a dot product operation with the input data. By moving the convolution kernel across the input data, a corresponding feature map can be output. Further, the bias term is added to the output of each convolution kernel to generate the final feature map, which is then fed into the activation function for calculation to obtain the final output of the convolutional layer. To prevent data loss during calculation, zero-padding can be applied to the edges of the input data. Examples are shown in Table 2 below:
[0098] Table 2 shows an example of zero-padding operation on the edges of input data.
[0099]
[0100] The convolution kernels are shown in Table 3 below:
[0101] Table 3 Examples of Convolution Kernels
[0102] 2 3 1 3 2 5 4 1 3
[0103] The results of the convolution calculation are shown in Table 4 below:
[0104] Table 4 Convolution Calculation Results
[0105] 0 0 0 0 67 27 9 0 0 0 0 0.226 0 0 0 0 67 19 2 0 0 0 0 0.0839
[0106] Pooling layer: Max pooling captures important features and reduces the dimensionality of the convolutional layer output to decrease the number of parameters and improve computational efficiency. The pooling window size determines the range of the pooling operation. Using the above convolution results, selecting a pooling window size of 2*2 and a stride of 1, the max pooling yields the following result (Table 5), which can be selected as the mentioned feature:
[0107] Table 5. Reactant characteristic data
[0108] 0 0 0 0 67 27 9 0 0 0 0 0.226
[0109] Deconvolutional layer: Performs transposed convolution operation to upsample the feature map to the original input size.
[0110] 3) Define the loss function, initialize the weight coefficients of the network model, and continuously update the weight coefficients through the backpropagation algorithm until the loss function is minimized, thus obtaining the trained and optimized model.
[0111] 4) Based on the trained and optimized model, select the output of the pooling layer as the extracted features.
[0112] The main parameters involved in this process are:
[0113] 1) Convolution kernel: The size of the convolution kernel in a convolutional layer determines the spatial range of features that can be captured. The size of the deconvolution kernel in a deconvolutional layer is usually the same as or larger than the size of the convolution kernel.
[0114] 2) Stride: The size of the convolutional kernel's movement, affecting the size of the output features. It can be optimized based on the input data size and output features. A larger stride reduces the size of the feature map, but also reduces computational cost.
[0115] 3) Padding: Add 0 or other specific values to the edges of the input data to control the size of the output features. Here, we choose padding with 0.
[0116] 4) Activation function: ReLU is used as the activation function for nonlinear processing.
[0117] 5) Pooling layer: Max pooling is used, and the maximum value within the pooling window is selected as the pooling output.
[0118] 6) Dilation rate: Used to control the spacing between convolution kernel elements, usually set to 1 in deconvolution.
[0119] 7) Dropout: The proportion of neurons randomly dropped during training to prevent overfitting. It is typically set between 0.2 and 0.5, and adjusted during training based on specific results and the dataset. Too high a Dropout rate may lead to underfitting, while too low a Dropout rate may be insufficient to prevent overfitting.
[0120] 8) Loss function: The mean square error method is used to calculate the mean square error between the actual value and the output value, which is used as the loss function.
[0121] Similarly, feature model construction and feature extraction of product data variables are performed, and the extracted reactant and product features are used to construct molecular reaction models. This embodiment will not be elaborated further here.
[0122] Step S13: Construct a molecular reaction model based on reactant characteristic data, product characteristic data, and reaction condition information data.
[0123] In this embodiment, step S13 is implemented by first constructing a training sample set based on reactant feature data, product feature data, and reaction condition information data to train the BP neural network molecular reaction model; then, the reaction product feature data and reaction condition information data included in each sample data in the training sample set are used as the input of the BP neural network molecular reaction model, and the product feature data included in each sample data is used as the output of the BP neural network molecular reaction model. After passing through the hidden layer of the BP neural network molecular reaction model, the node weights and biases of the hidden layer are optimized based on the gradient descent method to obtain the trained BP neural network molecular reaction model.
[0124] In this step, the reactant feature data and product feature data extracted in step S12, as well as the reaction condition information data included in the molecular reaction dataset obtained in step S11, are used as training sample datasets to construct and train the molecular reaction model.
[0125] Specifically, refer to Figure 5 As shown:
[0126] 1. Input of BP neural network molecular reaction model
[0127] The reactant feature data included in each sample data is preprocessed to convert the reactant feature data into a one-dimensional vector of reactants by element-wise concatenation. The one-dimensional vector of reactants and the reaction condition information data are then used as input to the BP neural network molecular reaction model.
[0128] 2. Output of the BP neural network molecular reaction model
[0129] The product feature data included in each sample data is preprocessed to convert the product feature data into a one-dimensional product vector by element concatenation, and the one-dimensional product vector is used as the output of the BP neural network molecular reaction model.
[0130] 3. BP Neural Network Molecular Reaction Model Setup and Parameters
[0131] A backpropagation (BP) neural network with three hidden layers is constructed. The first hidden layer is designed to have a relatively large number of neurons; empirically, this is taken as (input layer node count + output layer node count) * 3 / 2. The second and third hidden layers use the same number of neurons, which is less than the number of nodes in the input layer / the first hidden layer / the output layer. The PReLU activation function is used, with the goal of minimizing the loss function. Gradient descent is employed to continuously adjust and optimize the node weights and biases, resulting in a well-trained molecular reaction network model.
[0132] PReLUde(x)=max(0,x)+α*min(0,x)
[0133] α is a trainable parameter.
[0134] The molecular reaction model construction method based on distributed features provided in this embodiment of the invention utilizes a deep neural network model to achieve distributed feature extraction, distributing the molecular reaction model data across different nodes for processing, and performing data transmission and computation between nodes. This enables rapid modeling and prediction of molecular reactions, effectively ensuring data security and accuracy, reducing interference from factors such as noise data and human tampering during data transmission and processing, improving the model's computational efficiency and prediction performance, simplifying the modeling process, reducing reliance on professional knowledge during model construction, and improving modeling efficiency.
[0135] Based on the same inventive concept, this embodiment of the invention provides a molecular reaction model construction device based on distributed features, referring to... Figure 6 As shown, the construction device may include a first acquisition module 61, a first feature extraction module 62, and a construction module 63, and its working principle is as follows:
[0136] The first acquisition module 61 is used to acquire a molecular reaction dataset; wherein, the molecular reaction dataset includes: reactant data, product data and reaction condition information data; specifically, the reactant data includes the molecular composition and molecular concentration data of the reactants, the product data includes the molecular composition and molecular concentration data of the products, and the reaction condition information data includes: reaction temperature, reaction pressure, reactor volume and reactor flow rate.
[0137] The first feature extraction module 62 is used to extract features from reactant data and product data to obtain distributed reactant feature data and product feature data;
[0138] Module 63 is used to construct molecular reaction models based on reactant characteristic data, product characteristic data, and reaction condition information data.
[0139] In one embodiment, the construction module 63 is specifically used to: construct a training sample set based on reactant feature data, product feature data, and reaction condition information data to train the BP neural network molecular reaction model; the reaction product feature data and reaction condition information data included in each sample data in the training sample set are used as inputs to the BP neural network molecular reaction model, and the product feature data included in each sample data is used as outputs to the BP neural network molecular reaction model; after passing through the hidden layer of the BP neural network molecular reaction model, the node weights and biases of the hidden layer are optimized based on the gradient descent method to obtain the trained BP neural network molecular reaction model.
[0140] In one embodiment, the first feature extraction module 62 is specifically used for:
[0141] The molecular structure of the reactant data and product data is characterized using the structure-guided lumped method to obtain the molecular structure feature vectors of the reactant data and the product data, respectively.
[0142] The molecular structure feature vectors of the reactant data and the product data are respectively input into a pre-constructed convolutional neural network model, and the data output by the pooling layer in the convolutional neural network model are extracted as distributed reactant feature data and product feature data, respectively.
[0143] In another embodiment, the construction module 63 is further configured to:
[0144] The reactant feature data included in each sample data is preprocessed to convert the reactant feature data into a one-dimensional vector of reactants by element-wise concatenation, and the one-dimensional vector of reactants and the reaction condition information data are used as inputs to the BP neural network molecular reaction model.
[0145] The product feature data included in each sample data is preprocessed to convert the product feature data into a one-dimensional product vector by element concatenation, and the one-dimensional product vector is used as the output of the BP neural network molecular reaction model.
[0146] Based on the same inventive concept, this embodiment of the invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method for constructing a molecular reaction model based on distributed features.
[0147] Based on the same inventive concept, this embodiment of the invention provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the above-described method for constructing a molecular reaction model based on distributed features.
[0148] The principle by which the above-mentioned apparatus, medium, and related equipment in this embodiment of the invention solve the problem is similar to the aforementioned method for constructing molecular reaction models based on distributed features. Therefore, its implementation can refer to the implementation of the aforementioned method for constructing molecular reaction models based on distributed features, and the repeated parts will not be described again.
[0149] Example 2
[0150] Embodiment 2 of the present invention provides a method for predicting molecular reaction products based on distributed features. This prediction method predicts the product data of a molecular reaction based on a molecular reaction model pre-constructed using the method for constructing a molecular reaction model based on distributed features in Embodiment 1. (Refer to...) Figure 7 As shown, the prediction method may include the following steps:
[0151] Step S71: Obtain reactant data and reaction condition information from the molecular reaction dataset.
[0152] The specific implementation of this step can refer to step S11 in the above embodiment one, and the embodiment of the present invention will not be repeated here.
[0153] Step S72: Extract features from the reactant data to obtain distributed reactant feature data.
[0154] In this step, the reactant data are first characterized using the structure-guided lumped method to obtain molecular structure feature vectors for each reactant. Then, these feature vectors are input into a pre-constructed convolutional neural network model, and the data output from the pooling layers is extracted as distributed reactant feature data. The specific implementation of this step can refer to step S12 in Example 1 above; further details are omitted here.
[0155] Step S73: Based on reactant characteristic data and reaction condition information data, input them into a pre-constructed molecular reaction model to output product characteristic data.
[0156] Step S74: Predict product data of molecular reactions based on product characteristic data.
[0157] Specifically, in this second embodiment, step S74 is based on product feature data, a pre-built convolutional neural network model, and product data of molecular reactions predicted by the structure-guided lumped method.
[0158] Combination Figure 8As shown, in this embodiment of the invention, for the reactants to be calculated, the extracted reactant feature data and reaction condition information data are input into the molecular reaction network model to calculate the product features. Furthermore, the product feature model (such as the pre-trained product convolutional neural network model in Example 1) is combined to achieve product data prediction.
[0159] The molecular reaction product prediction method based on distributed features provided in this embodiment of the invention distributes the molecular reaction model data across different nodes for processing through a distributed feature extraction model, enabling data transmission and computation between nodes. This achieves rapid modeling and prediction of molecular reactions, effectively ensuring data security and accuracy. It also reduces interference from factors such as noise data and human tampering during data transmission and processing on the molecular reaction model data, thereby improving the model's computational efficiency and prediction performance.
[0160] Based on the same inventive concept, this embodiment of the invention provides a molecular reaction product prediction device based on distributed features, referring to... Figure 9 As shown, the device may include: a second acquisition module 91, a second feature extraction module 92, a product feature data determination module 93, and a product data prediction module 94, and its working principle is as follows:
[0161] The second acquisition module 91 is used to acquire reactant data and reaction condition information data from the molecular reaction dataset;
[0162] The second feature extraction module 92 is used to extract features from the reactant data to obtain distributed reactant feature data;
[0163] The product characteristic data determination module 93 is used to input reactant characteristic data and reaction condition information data into a pre-constructed molecular reaction model to output product characteristic data; wherein, the molecular reaction model is pre-constructed according to the molecular reaction model construction method based on distributed features in Example 1;
[0164] The product data prediction module 94 is used to predict product data of molecular reactions based on product characteristic data.
[0165] In one embodiment, the second feature extraction module 92 is specifically used to: firstly, characterize the molecular structure of the reactant data based on the structure-guided lumped method to obtain the molecular structure feature vectors of the reactant data respectively; then, input the molecular structure feature vectors of the reactant data into a pre-built convolutional neural network model, and extract the data output by the pooling layer in the convolutional neural network model as distributed reactant feature data.
[0166] In another embodiment, the product data prediction module 94 is specifically used to predict the product data of molecular reactions based on product feature data and a pre-built convolutional neural network model, as well as the structure-guided lumped method.
[0167] Based on the same inventive concept, this embodiment of the invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method for predicting molecular reaction products based on distributed features.
[0168] Based on the same inventive concept, this embodiment of the invention provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the above-mentioned method for predicting molecular reaction products based on distributed features.
[0169] The principles by which the above-mentioned devices, media, and related equipment in the embodiments of the present invention solve the problem are similar to those of the aforementioned methods. Therefore, their implementation can refer to the implementation of the aforementioned methods, and repeated details will not be repeated.
[0170] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage and optical storage) containing computer-usable program code.
[0171] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.
[0172] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1The function specified in one or more boxes.
[0173] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.
[0174] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.
Claims
1. A method for constructing a molecular reaction model based on distributed features, characterized in that, include: Obtain a molecular reaction dataset; wherein the molecular reaction dataset includes: reactant data, product data, and reaction condition information data; Feature extraction is performed on the reactant data and the product data to obtain distributed reactant feature data and product feature data; A molecular reaction model is constructed based on the reactant characteristic data, the product characteristic data, and the reaction condition information data.
2. The method of claim 1, wherein, The construction of a molecular reaction model based on the reactant characteristic data, the product characteristic data, and the reaction condition information data includes: A training sample set is constructed based on the reactant feature data, the product feature data, and the reaction condition information data to train the BP neural network molecular reaction model. The reaction product feature data and reaction condition information data included in each sample data in the training sample set are used as the input of the BP neural network molecular reaction model, and the product feature data included in each sample data is used as the output of the BP neural network molecular reaction model. After passing through the hidden layer of the BP neural network molecular reaction model, the node weights and biases of the hidden layer are optimized based on the gradient descent method to obtain the trained BP neural network molecular reaction model.
3. The method of claim 1, wherein, The step of extracting features from the reactant data and the product data to obtain distributed reactant feature data and product feature data includes: The molecular structure of the reactant data and product data is characterized using the structure-guided lumped method to obtain the molecular structure feature vectors of the reactant data and the product data, respectively. The molecular structure feature vectors of the reactant data and the product data are respectively input into a pre-constructed convolutional neural network model, and the data output by the pooling layer in the convolutional neural network model are extracted as distributed reactant feature data and product feature data, respectively.
4. The method according to claim 3, characterized in that, The convolutional neural network model includes a reactant convolutional neural network model and a product convolutional neural network model; The reactant convolutional neural network model was pre-trained using the following method: Construct a reactant training sample set; wherein, each sample in the reactant training sample set includes a reactant molecular structure feature vector matrix composed of molecular structure feature vectors corresponding to all reactants in a set of reactant data; The reactant convolutional neural network model is trained using samples from the reactant training sample set, wherein the reactant molecular structure feature vector matrix serves as the input and output of the reactant convolutional neural network model, respectively; after passing the reactant molecular structure feature vector matrix through a correlation layer, the trained reactant convolutional neural network model is obtained. The product convolutional neural network model is pre-trained using the following method: Construct a product training sample set; wherein, each sample in the product training sample set includes a product molecular structure feature vector matrix composed of molecular structure feature vectors corresponding to all products in a set of product data; The product convolutional neural network model is trained using samples from the product training sample set, wherein the product molecular structure feature vector matrix serves as the input and output of the product convolutional neural network model, respectively; after passing the product molecular structure feature vector matrix through a correlation layer, the trained product convolutional neural network model is obtained.
5. The method according to claim 4, characterized in that, After passing through a correlation layer, the feature vector matrix of the reactant molecular structure yields a trained reactant convolutional neural network model, including: The reactant molecular structure feature vector matrix is passed through the convolutional layer of the reactant convolutional neural network model and then subjected to a dot product operation. The reactant molecular structure feature vector matrix is moved on the convolutional kernel of the convolutional layer to output the reactant feature mapping of the reactant molecular structure feature vector matrix. The reactant feature mapping is added to the bias term of the convolutional layer and then substituted into the activation function to obtain the final reactant feature mapping. The final reactant feature map is dimensionality reduced in the pooling layer of the reactant convolutional neural network model. The dimensionality-reduced reactant feature map is then transposed and convolved in the deconvolution layer of the reactant convolutional neural network model. The weights of the backpropagation algorithm are updated using the reactant loss function defined by the actual and output values to obtain the trained reactant convolutional neural network model. After the product molecular structure feature vector matrix is passed through a correlation layer, a trained product convolutional neural network model is obtained, including: The product molecular structure feature vector matrix is passed through the convolutional layer of the product convolutional neural network model and then subjected to a dot product operation. The product molecular structure feature vector matrix is moved on the convolutional kernel of the convolutional layer to output the product feature mapping of the product molecular structure feature vector matrix. The product feature mapping is added to the bias term of the convolutional layer and then substituted into the activation function to obtain the final product feature mapping. The final product feature map is dimensionality reduced in the pooling layer of the product convolutional neural network model. The dimensionality-reduced product feature map is then transposed and convolved in the deconvolution layer of the product convolutional neural network model. The weights of the backpropagation algorithm are updated using the product loss function defined by the actual and output values to obtain the trained product convolutional neural network model.
6. The method according to claim 2, characterized in that, Before training the BP neural network molecular reaction model, the following steps are also required: The reactant feature data included in each sample data is preprocessed to convert the reactant feature data into a one-dimensional vector of reactants by element-wise concatenation, and the one-dimensional vector of reactants and the reaction condition information data are used as inputs to the BP neural network molecular reaction model. The product feature data included in each sample data is preprocessed to convert the product feature data into a one-dimensional product vector by element concatenation, and the one-dimensional product vector is used as the output of the BP neural network molecular reaction model.
7. A method for predicting molecular reaction products based on distributed features, characterized in that, The molecular reaction model based on distributed features, as described in any one of claims 1 to 6, pre-constructs a molecular reaction model to predict product data of a molecular reaction.
8. The method according to claim 7, characterized in that, include: Acquire reactant data and reaction condition information from a molecular reaction dataset; Feature extraction is performed on the reactant data to obtain distributed reactant feature data; Based on the reactant characteristic data and the reaction condition information data, the data is input into the pre-constructed molecular reaction model to output product characteristic data. Product data of molecular reactions can be predicted based on the product characteristic data.
9. The method according to claim 8, characterized in that, Feature extraction is performed on the reactant data to obtain distributed reactant feature data, including: The molecular structure of the reactant data is characterized using the structure-guided lumped method to obtain the molecular structure feature vectors of the reactant data. The molecular structure feature vectors of the reactant data are input into a pre-constructed convolutional neural network model, and the data output from the pooling layer of the convolutional neural network model is extracted as distributed reactant feature data.
10. The method according to claim 8, characterized in that, Predicting product data for molecular reactions based on the aforementioned product characteristic data includes: Based on the product feature data and the pre-built convolutional neural network model, as well as the structure-guided lumped method, the product data of the molecular reaction are predicted.
11. A molecular reaction model construction device based on distributed features, characterized in that, include: The first acquisition module is used to acquire a molecular reaction dataset; wherein, the molecular reaction dataset includes: reactant data, product data, and reaction condition information data; The first feature extraction module is used to extract features from the reactant data and the product data to obtain distributed reactant feature data and product feature data. A construction module is used to construct a molecular reaction model based on the reactant characteristic data, the product characteristic data, and the reaction condition information data.
12. A molecular reaction product prediction device based on distributed characteristics, characterized in that, include: The product data prediction module is used to predict the product data of molecular reactions based on pre-built molecular reaction models. The molecular reaction model is pre-constructed using the molecular reaction model construction method based on distributed features according to any one of claims 1 to 6.
13. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the program implements the molecular reaction model construction method based on distributed features as described in any one of claims 1 to 6, or the molecular reaction product prediction method based on distributed features as described in any one of claims 7 to 10.
14. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the molecular reaction model construction method based on distributed features as described in any one of claims 1 to 6, or the molecular reaction product prediction method based on distributed features as described in any one of claims 7 to 10.