A chemical reaction network prediction method and system based on a graph neural network

Abstract

The application discloses a chemical reaction network prediction method and system based on a graph neural network, and belongs to the technical field of chemical reaction networks. The method comprises the following steps: S100, a chemical reaction network image is acquired; S200, node features and edge features are selected in the chemical reaction network image; S300, the node features and the edge features are used to generate a feature dataset; S400, a neural network model is constructed, and the neural network model is trained according to the feature dataset, so that a trained neural network model is obtained; and S500, the trained neural network model is used to predict the node and the whole chemical reaction network image, and a prediction result is output. The application can more accurately and efficiently output a safety and environmental protection prediction result.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of chemical reaction network, more particularly, it relates to a chemical reaction network prediction method and system based on graph neural network. BACKGROUND

[0002] The chemical reaction network generally refers to a graphical representation method of reaction relationship between three or more chemical substances. In the chemical reaction network, not only the production correlation between various substances is included, but also the quantitative data such as the amount, the optional reaction path, and the reaction condition.

[0003] Economic prediction of a reaction network is a long-standing topic in the chemical industry. A typical example is the research and prediction of the syngas downstream product reaction network. Syngas is a mixture of carbon monoxide and hydrogen, and a large number of chemicals can be produced from them. Therefore, considering the theoretical feasibility and economy, reasonable selection of the reaction route has always been a research focus. In addition to syngas, if the reaction center is focused on methanol, there will also be a corresponding methanol reaction network that can be analogized.

[0004] In the traditional chemical industry, decisions often depend on the subjective feelings of experts, and it is difficult to comprehensively consider various raw material and product yield historical factors, safety and environmental protection requirements in the storage and transportation process, and whether there is a better reaction path.

[0005] The graph convolutional neural network is a neural network that can automatically learn from the relationship between data and adaptively change parameters to achieve a specific goal. It is developed on the basis of deep learning and has a wide range of applications in many fields. By defining different features for nodes and edges in the graph, the graph neural network can be applied in various fields such as social networks.

[0006] Therefore, how to provide a chemical reaction network prediction method and system based on graph neural network is a problem that needs to be solved by those skilled in the art. SUMMARY

[0007] Therefore, the present application provides a chemical reaction network prediction method and system based on graph neural network

[0008] In order to achieve the above purpose, the present application provides the following technical scheme:

[0009] On the one hand, the present application provides a chemical reaction network prediction method based on graph neural network, comprising the following steps:

[0010] S100: Obtain a chemical reaction network image;

[0011] S200: Select node features and edge features in the chemical reaction network image;

[0012] S300: Generate a feature dataset by combining node features and edge features;

[0013] S400: Construct a neural network model and train the neural network model based on the feature dataset to obtain a trained neural network model;

[0014] S500: Based on the trained neural network model, predict the nodes and the overall chemical reaction network image, and output the prediction results.

[0015] Preferably, the chemical reaction network image is a graphical image of the reaction relationships between three or more chemical substances, manually constructed using thermodynamic or kinetic data.

[0016] Preferably, the chemical reaction network image is a graphical image of the reaction relationships between three or more chemical substances automatically constructed using the RMG toolkit.

[0017] Preferably, after S100, the method further includes: determining the central compound of the chemical reaction network image and designating the central compound as the target compound.

[0018] Preferably, step S200 involves selecting node features in the chemical reaction network image, including:

[0019] S210: Obtain the physical property data of the target compound and the corresponding yield time series data;

[0020] S211: Encode physical attribute data and production time series data into node features to obtain the node features in the dataset.

[0021] Preferably, step S200 selects edge features in the chemical reaction network image, including:

[0022] S220: Describes the thermodynamic probability of a reaction proceeding according to the Gibbs free energy criterion;

[0023] S221: Determine the Gibbs free energy change values ​​of the chemical reactions corresponding to the reaction paths in the chemical reaction network diagram of the target compound;

[0024] S222: Obtain the edge features in the dataset based on the changing values.

[0025] Preferably, the prediction result is whether the storage and transportation process of the target compound raw materials meets safety and environmental protection requirements, and whether there is a better reaction route.

[0026] In another aspect, the present application provides a chemical reaction network prediction system based on a graph neural network, comprising:

[0027] An acquisition module: acquiring a chemical reaction network image;

[0028] A processing module connected to the acquisition module, selecting node features and edge features in the chemical reaction network image;

[0029] A generation module connected to the processing module, generating a feature data set from the node features and edge features;

[0030] A construction module connected to the generation module, constructing a neural network model, and training the neural network model according to the feature data set to obtain a trained neural network model;

[0031] A prediction module connected to the construction module, predicting the node and the overall chemical reaction network image according to the trained neural network model, and outputting a prediction result.

[0032] According to the above technical solution, compared with the prior art, the present application provides a chemical reaction network prediction method and prediction system based on a graph neural network. The graph neural network is used to complete the above tasks, and various feature engineering factors are encoded as the features of the nodes and edges in the reaction network. The neural network and chemical knowledge are used comprehensively to assist relevant staff in completing decision-making work. The present application is different from the application scenarios of the neural network in the prior art. The present method can use the trained neural network model to quickly give a conclusion on the safety and environmental protection of a certain route or the overall reaction network, providing a reference for relevant decision-makers. Compared with the traditional monotonous linear analysis result, the present method comprehensively considers the historical data of various reactants and products, the amount ratio, the reaction conditions, and many other factors, and can more accurately and efficiently give a result. Compared with the subjective prediction relying on experts, the result of the present application is more objective and accurate. BRIEF DESCRIPTION OF DRAWINGS

[0033] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or the prior art description. Obviously, the drawings in the following description are only embodiments of the present application, and for those skilled in the art, other drawings can be obtained without creative labor on the basis of the provided drawings.

[0034] Figure 1 The method flowchart of the present application;

[0035] Figure 2 The chemical reaction network graph with methane as the center;

[0036] Figure 3The system structure schematic diagram of the present application. DETAILED DESCRIPTION

[0037] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative work are within the protection scope of the present application.

[0038] Referring to the drawings Figure 1 As shown in the drawings, the embodiments of the present application disclose a chemical reaction network prediction method based on a graph neural network, comprising the following steps:

[0039] S100: Obtain a chemical reaction network image;

[0040] S200: Select node features and edge features in the chemical reaction network image;

[0041] S300: Generate a feature dataset from the node features and the edge features;

[0042] S400: Construct a neural network model, and train the neural network model according to the feature dataset to obtain a trained neural network model;

[0043] S500: Predict the node and the overall chemical reaction network image according to the trained neural network model, and output a prediction result.

[0044] In one specific embodiment, the chemical reaction network image is a graphical image of reaction relationships between three or more chemical substances, which is manually constructed through thermodynamic data or kinetic data.

[0045] In one specific embodiment, the chemical reaction network image is a graphical image of reaction relationships between three or more chemical substances, which is automatically constructed through an RMG toolkit.

[0046] Specifically, RMG-Reaction Mechanism Generator is only one method, and other methods can also be used for construction, such as the website https: / / rmg.mit.edu / . The first two are reaction networks constructed using thermodynamic data.

[0047] In one specific embodiment, after S100, it further comprises: determining a central compound of the chemical reaction network image, and taking the central compound as a target compound.

[0048] Specifically, the acquisition of the chemical reaction network image: can be manually constructed from thermodynamic or kinetic data, or automatically constructed using toolkits such as RMG. The embodiment selects a reaction network centered on methane (CH4), see FIG. 1, which is a reaction network image that has been constructed. Figure 2

[0049] In one specific embodiment, S200 selects node features in the chemical reaction network image, including:

[0050] S210: Acquire physical property data and corresponding yield time series data of the target compound;

[0051] S211: Encode the physical property data and yield time series data into node features to obtain part of the node features in the data set.

[0052] Specifically, taking methane as an example, its physical properties are: boiling point -161.5℃, flash point -188℃, melting point -182.5℃, density (standard condition) 0.717g / L, safety description S9; S16; S33, hazard symbol R12, etc. These data are encoded into node features to generate a node feature data set. At the same time, LSTM and other time series neural networks can be used to encode the historical price of methane, which will be used as node features.

[0053] In one specific embodiment, S200 selects edge features in the chemical reaction network image to generate an edge feature data set, including:

[0054] S220: According to the Gibbs free energy criterion to describe the thermodynamic probability of the reaction;

[0055] S221: Determine the Gibbs free energy change value of the chemical reaction corresponding to the reaction path in the chemical reaction network image of the target compound;

[0056] S222: Generate an edge feature data set according to the change value.

[0057] Specifically, the Gibbs free energy criterion can be used to describe the thermodynamic probability of the reaction, and FIG. 2 shows the Gibbs free energy change value of the chemical reaction corresponding to the reaction path. Figure 2 In FIG. 2, the five nodes correspond to the Gibbs free energy change value of the chemical reaction corresponding to the reaction path. The values corresponding to the right upper (1 o'clock direction) to the left upper (11 o'clock direction) in the figure are: -0.023eV, 0.018eV, 0.0057eV, -0.109eV, -0.0093eV. In the figure, the node features are selected: taking methane as an example, its physical properties are: boiling point -161.5℃, flash point -188℃, melting point -182.5℃, density, which is encoded as edge features in this example.

[0058] ​Specifically, the edge feature is different from the node feature. The node is a substance, so the node feature is an index of the substance. The edge is a reaction, so the edge feature corresponds to a reaction index. Figure 2 The small origin in the formula represents the edge (placed on the middle intersection of the edge), so Gibbs is the attribute of the edge. Figure 2 All molecular formulas (bond-line formulas) in the formula are regarded as nodes, so the attributes of the substance are regarded as node features.

[0059] In one specific embodiment, the prediction result is whether the raw material storage and transportation process of the target compound meets the safety and environmental protection requirements, and whether there is a more optimal reaction path.

[0060] Specifically, node and whole graph training and prediction: using the above method to complete data accumulation, that is, starting to train the time series neural network model, after obtaining the trained time series neural network model, the node and the whole graph can be predicted. The node prediction index includes: whether the sub-path is the optimal reaction path. The whole graph prediction index includes: whether the whole chemical reaction network meets the safety and environmental protection requirements.

[0061] On the other hand, referring to FIG. 8, the embodiment of the present application also discloses a chemical reaction network prediction system based on a graph neural network, which comprises: Figure 3 An acquisition module: acquiring a chemical reaction network image;

[0062] A processing module connected with the acquisition module, selecting node features and edge features in the chemical reaction network image;

[0063] A generation module connected with the processing module, generating feature data sets from the node features and the edge features;

[0064] A construction module connected with the generation module, constructing a neural network model, and training the neural network model according to the feature data sets to obtain a trained neural network model;

[0065] A prediction module connected with the construction module, predicting the node and the whole chemical reaction network image according to the trained neural network model, and outputting a prediction result.

[0066] Specifically, the node of the present application is the material attribute (including toxicity, storage condition, etc.) of a certain chemical substance, the yield (using LSTM to extract features from time series data), etc. The edge mainly describes the dependence of a certain substance on raw materials and the production cost of the path.

[0067] The existing graph neural network applications in the economic and chemical fields related to the present application are significantly different from the present application. Briefly described as follows:

[0068]

[0069] ​Different from the graph neural network in the field of economics, the node in the present application is a chemical substance, rather than a stock of a company. Meanwhile, the final goal of the present application is whether a chemical reaction path is of construction value, which needs to comprehensively consider various storage and transportation, environmental protection and safety and other factors.

[0070] Different from the graph neural network in the field of chemical materials, the graph in the present application is not to represent a chemical microstructure, but the relationship between substances.

[0071] Different from the reaction network computing tool, the present application focuses on the prediction of network security and environmental protection, and is mainly completed by using a machine learning model, rather than focusing on the construction of a reaction network. On the contrary, RMG can be one of the methods for generating a chemical reaction network in the present application.

[0072] The various embodiments in the present specification are described in a progressive manner, and each embodiment focuses on the difference from other embodiments. The same or similar parts between the various embodiments can be referred to each other. For the device disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple, and the relevant part can be referred to the method part.

[0073] The above description of the disclosed embodiments enables a person skilled in the art to implement or use the present application. Various modifications to the embodiments will be apparent to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for predicting chemical reaction networks based on graph neural networks, characterized in that, Includes the following steps: S100: Acquire chemical reaction network images; S200: Select node features and edge features in the chemical reaction network image; S300: Generate a feature dataset by combining node features and edge features; S400: Construct a neural network model and train the neural network model based on the feature dataset to obtain a trained neural network model; S500: Based on the trained neural network model, predict the overall image of the nodes and chemical reaction network, and output the prediction result; The chemical reaction network image is a graphical representation of the reaction relationships between three or more chemical substances, manually constructed using thermodynamic or kinetic data. S200 selects node features in the chemical reaction network image, including: S210: Obtain the physical property data of the target compound and the corresponding yield time series data; S211: Encode physical attribute data and production time series data into node features to obtain a portion of the node features in the dataset; S200 selects edge features in the chemical reaction network image, including: S220: Describes the thermodynamic probability of a reaction proceeding according to the Gibbs free energy criterion; S221: Determine the Gibbs free energy change values ​​of the chemical reactions corresponding to the reaction paths in the chemical reaction network diagram of the target compound; S222: Obtain the edge features in the dataset based on the changing values; The prediction results indicate whether the storage and transportation process of the target compound raw materials meets safety and environmental protection requirements, and whether there is a better reaction route.

2. The chemical reaction network prediction method based on graph neural networks according to claim 1, characterized in that, The chemical reaction network image is a graphical representation of the reaction relationships between three or more chemical substances, automatically constructed using the RMG toolkit.

3. The chemical reaction network prediction method based on graph neural networks according to claim 1, characterized in that, Following S100, the method further includes: identifying the central compound of the chemical reaction network image and designating the central compound as the target compound.

4. A graph neural network-based chemical reaction network prediction system utilizing the graph neural network-based chemical reaction network prediction method according to any one of claims 1-3, characterized in that, include: Acquisition module: Acquires chemical reaction network images; The processing module, connected to the acquisition module, selects node features and edge features in the chemical reaction network image; The generation module, connected to the processing module, generates a feature dataset from node features and edge features; A construction module is connected to the generation module to construct a neural network model and train the neural network model based on the feature dataset to obtain a trained neural network model. The prediction module, connected to the construction module, predicts the nodes and the overall chemical reaction network image based on the trained neural network model and outputs the prediction results.

Images