Environment-friendly insulating gas molecule design and performance prediction method and system

Abstract

The application discloses a novel environment-friendly insulating gas molecule design and performance prediction method and system, and aims to solve the technical problem of poor insulating gas molecule design efficiency caused by the existing novel environment-friendly insulating gas molecule design and performance prediction method. The method comprises the following steps: inputting the obtained environment-friendly insulating gas molecule bonding parameter into a preset environment-friendly insulating gas molecule structure design model, and outputting initial environment-friendly insulating gas molecule structure data to be designed; preprocessing the initial environment-friendly insulating gas molecule structure data to be designed and the environment-friendly insulating gas molecule bonding parameter, and outputting target environment-friendly insulating gas molecule structure data to be designed and target environment-friendly insulating gas molecule bonding parameter; inputting the target environment-friendly insulating gas molecule structure data to be designed and the target environment-friendly insulating gas molecule bonding parameter into a preset environment-friendly insulating gas molecule performance prediction specificity neural network model to perform performance prediction, and outputting a target environment-friendly insulating gas molecule performance value.

Description

Technical Field

[0001] This invention relates to the field of environmentally friendly insulating gas technology, and in particular to a novel method and system for molecular design and performance prediction of environmentally friendly insulating gases. Background Technology

[0002] Sulfur hexafluoride (SF6) is widely used in high-voltage electrical equipment due to its excellent insulation properties, but its extremely high global warming potential (GWP) has brought serious environmental problems. Current research and development of environmentally friendly alternative gases mainly relies on experimental methods to determine GWP and insulation performance, but this method has limitations such as high cost and long development cycles.

[0003] For critical equipment such as gas-insulated switchgear (GIS), gas-insulated transmission lines (GIL), gas-insulated transformers (GIT), and gas circuit breakers (GCB), an ideal alternative gas needs to simultaneously meet three core indicators: first, excellent electrical insulation strength, which is the foundation for safe equipment operation; second, a low GWP value to meet environmental protection requirements; and finally, a suitable liquefaction temperature to ensure the gas remains stable under different pressure and temperature environments. Finding a balance among these three will be the key breakthrough in the research and development of next-generation environmentally friendly insulating gases.

[0004] Existing methods for designing and predicting the performance of novel environmentally friendly insulating gas molecules mainly rely on trial and error. This requires manually pre-setting a large number of potential gas molecule structures, and then repeatedly synthesizing material samples and conducting insulation performance tests and GWP evaluations. This approach is too labor-intensive, resulting in poor efficiency in the design of potential insulating gas molecules. Summary of the Invention

[0005] This invention provides a novel method and system for designing and predicting the performance of environmentally friendly insulating gas molecules, which solves the technical problem of poor design efficiency of potential insulating gas molecules caused by existing methods for designing and predicting the performance of novel environmentally friendly insulating gas molecules.

[0006] The first aspect of this invention provides a novel method for the design and performance prediction of environmentally friendly insulating gas molecules, comprising:

[0007] The bonding parameters of environmentally friendly insulating gas molecules are obtained and input into a pre-set environmentally friendly insulating gas molecule structure design model. The initial environmentally friendly insulating gas molecule structure data to be designed is output. The pre-set environmentally friendly insulating gas molecule structure design model is a condition graph generation model.

[0008] The initial molecular structure data of the environmentally friendly insulating gas to be designed and the bonding parameters of the environmentally friendly insulating gas molecules are preprocessed to output the target molecular structure data of the environmentally friendly insulating gas to be designed and the target environmentally friendly insulating gas molecules.

[0009] The molecular structure data of the target environmentally friendly insulating gas to be designed and the bonding parameters of the target environmentally friendly insulating gas molecule are input into a pre-set specific neural network model for predicting the performance of the environmentally friendly insulating gas molecule, and the performance value of the target environmentally friendly insulating gas molecule is output. The pre-set specific neural network model for predicting the performance of the environmentally friendly insulating gas molecule is a multilayer perceptron structure, including an input layer, a hidden layer, and an output layer.

[0010] Optionally, the pre-set environmentally friendly insulating gas molecular structure design model includes a graph neural network encoder, a latent variable sampling module, a conditional vector embedding module, and a graph structure decoder; the step of inputting the bonding parameters of the environmentally friendly insulating gas molecules into the pre-set environmentally friendly insulating gas molecular structure design model and outputting the initial environmentally friendly insulating gas molecular structure data to be designed includes:

[0011] The graph neural network encoder is used to graph encode the bonding parameters of the environmentally friendly insulating gas molecules, and outputs a low-dimensional feature representation.

[0012] The latent variable sampling module is used to sample the low-dimensional feature representation and output latent variables;

[0013] The conditional vector embedding module is used to embed the latent variables into the target performance and output a latent vector.

[0014] The potential vector is input into the graph structure decoder for decoding, and the initial molecular structure data of the environmentally friendly insulating gas to be designed is output.

[0015] Optionally, the preprocessing of the initial environmentally friendly insulating gas molecular structure data and the bonding parameters of the environmentally friendly insulating gas molecules to output the target environmentally friendly insulating gas molecular structure data and the target environmentally friendly insulating gas molecules includes:

[0016] The initial design data of the environmentally friendly insulating gas molecule and the bonding parameters of the environmentally friendly insulating gas molecule are converted respectively, and the design data of the environmentally friendly insulating gas molecule and the bonding parameters of the environmentally friendly insulating gas molecule in array form are output.

[0017] The array-formatted design data of the environmentally friendly insulating gas molecules and the array-formatted bonding parameters of the environmentally friendly insulating gas molecules are normalized respectively, and the normalized design data of the environmentally friendly insulating gas molecules and the normalized bonding parameters of the environmentally friendly insulating gas molecules are output.

[0018] The normalized molecular structure data of the environmentally friendly insulating gas to be designed and the normalized bonding parameters of the environmentally friendly insulating gas molecules are respectively encoded to generate the target molecular structure data of the environmentally friendly insulating gas to be designed and the target bonding parameters of the environmentally friendly insulating gas molecules.

[0019] Optionally, the model training process for the pre-set environmentally friendly insulating gas molecule structure design model and the pre-set environmentally friendly insulating gas molecule performance prediction specific neural network model is as follows:

[0020] Obtain environmentally friendly insulating gas characteristic data, and input the bonding parameters of environmentally friendly insulating gas molecules used for model training from the environmentally friendly insulating gas characteristic data into the initial environmentally friendly insulating gas molecule structure design model, and output the initial environmentally friendly insulating gas molecule structure data to be designed for model training.

[0021] Based on the initial design data of the environmentally friendly insulating gas molecular structure used for model training and the environmentally friendly insulating gas characteristic data used for model training, the graph reconstruction loss is calculated.

[0022] The molecular structure data of the environmentally friendly insulating gas used for model training and the bonding parameters of the environmentally friendly insulating gas molecules used for model training are preprocessed respectively, and the preprocessed molecular structure data of the environmentally friendly insulating gas and the preprocessed bonding parameters of the environmentally friendly insulating gas molecules are output.

[0023] The preprocessed environmentally friendly insulating gas molecule structure data and the preprocessed environmentally friendly insulating gas molecule bonding parameters are input into a specific neural network model for predicting the initial environmentally friendly insulating gas molecule performance, and the gas molecule performance values ​​are output for model training.

[0024] Based on the gas molecule performance values ​​used for model training and the environmentally friendly insulating gas test data in the environmentally friendly insulating gas characteristic data, the regression loss is calculated;

[0025] Substitute the regression loss and the graph reconstruction loss into the preset total loss function and take the derivative to output the model gradient;

[0026] The model gradient is used to update the model parameters of the initial environmentally friendly insulating gas molecule structure design model and the specific neural network model for predicting the performance of the initial environmentally friendly insulating gas molecule, and outputs the intermediate environmentally friendly insulating gas molecule structure design model and the specific neural network model for predicting the performance of the intermediate environmentally friendly insulating gas molecule.

[0027] The number of model updates is counted in real time, and it is determined whether the number of model updates has reached the preset number of training iterations.

[0028] If this is achieved, the intermediate environmentally friendly insulating gas molecule structure design model will be used as the trained pre-set environmentally friendly insulating gas molecule structure design model, and the specific neural network model for predicting the performance of the intermediate environmentally friendly insulating gas molecule will be used as the trained specific neural network model for predicting the performance of the pre-set environmentally friendly insulating gas molecule.

[0029] Optionally, it also includes:

[0030] If the number of model updates does not reach the preset number of training iterations, then the intermediate environmentally friendly insulating gas molecule structure design model will be used as the new initial environmentally friendly insulating gas molecule structure design model, and the specific neural network model for predicting the performance of intermediate environmentally friendly insulating gas molecules will be used as the new specific neural network model for predicting the performance of initial environmentally friendly insulating gas molecules.

[0031] Jump to execute the step of inputting the bonding parameters of the environmentally friendly insulating gas molecules used for model training from the environmentally friendly insulating gas characteristic data into the initial environmentally friendly insulating gas molecule structure design model, and outputting the initial environmentally friendly insulating gas molecule structure data to be designed for model training, until the number of model updates reaches the preset number of training times;

[0032] The intermediate environmentally friendly insulating gas molecule structure design model determined when the model update count reaches the preset training count is used as the trained preset environmentally friendly insulating gas molecule structure design model, and the specific neural network model for predicting the performance of intermediate environmentally friendly insulating gas molecules is used as the trained specific neural network model for predicting the performance of preset environmentally friendly insulating gas molecules.

[0033] Optionally, the preset total loss function is specifically:

[0034] L = L e +β*L KL +λ*L property ;

[0035] Where L is the loss value corresponding to the preset total loss function; L e To compensate for the graph reconstruction loss between the generated molecular structure and the real molecular structure; L KL The KL divergence loss between the latent distribution and the standard normal distribution; Lproperty The regression loss is the difference between the predicted and target performance values ​​of the generated molecule; β and λ are adjustable weighting factors.

[0036] The second aspect of this invention provides a novel environmentally friendly insulating gas molecule design and performance prediction system, comprising:

[0037] The acquisition module is used to acquire the bonding parameters of environmentally friendly insulating gas molecules, input the bonding parameters of environmentally friendly insulating gas molecules into a preset environmentally friendly insulating gas molecule structure design model, and output the initial environmentally friendly insulating gas molecule structure data to be designed.

[0038] The preprocessing module is used to preprocess the initial environmentally friendly insulating gas molecular structure data to be designed and output the target environmentally friendly insulating gas molecular structure data to be designed.

[0039] The prediction module is used to input the molecular structure data of the target environmentally friendly insulating gas to be designed into a specific neural network model for predicting the molecular performance of the environmentally friendly insulating gas, and output the gas molecule performance value corresponding to the molecular structure data of the target environmentally friendly insulating gas to be designed.

[0040] A computer device provided in a third aspect of the present invention includes a memory and a processor. The memory stores a computer program, and when the computer program is executed by the processor, the processor performs the steps of the novel environmentally friendly insulating gas molecule design and performance prediction method as described in any of the preceding claims.

[0041] The fourth aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed, implements the steps of the novel environmentally friendly insulating gas molecule design and performance prediction method as described in any of the preceding claims.

[0042] The fifth aspect of the present invention provides a computer program product, the computer program product comprising a computer program stored on a non-transitory computer-readable storage medium, the computer program comprising program instructions, wherein, when the program instructions are executed by a computer, the computer performs the steps of the novel environmentally friendly insulating gas molecule design and performance prediction method as described in any of the preceding claims.

[0043] As can be seen from the above technical solutions, the present invention has the following advantages:

[0044] The present invention provides a novel method for the design and performance prediction of environmentally friendly insulating gas molecules. This method involves obtaining the bonding parameters of environmentally friendly insulating gas molecules and inputting these parameters into a pre-set environmentally friendly insulating gas molecule structure design model to output initial molecular structure data of the environmentally friendly insulating gas to be designed. The initial molecular structure data and bonding parameters are pre-processed to output target molecular structure data and target bonding parameters of the environmentally friendly insulating gas to be designed. These target molecular structure data and bonding parameters are then input into a pre-set specific neural network model for performance prediction of environmentally friendly insulating gas molecules to perform performance prediction, outputting the performance value of the target environmentally friendly insulating gas molecule. Based on this method, the present invention utilizes a pre-set environmentally friendly insulating gas molecule structure design model to obtain initial molecular structure data, eliminating the need for repeated synthesis of material samples and insulation performance testing and GWP evaluation. This accelerates the design process of novel insulating gases, thereby improving the efficiency of environmentally friendly insulating gas molecule design. Attached Figure Description

[0045] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0046] Figure 1 This is a flowchart illustrating the steps of a novel environmentally friendly insulating gas molecule design and performance prediction method provided in Embodiment 1 of the present invention.

[0047] Figure 2 The flowchart shows the steps of the model training process for the pre-set environmentally friendly insulating gas molecule structure design model and the pre-set environmentally friendly insulating gas molecule performance prediction specific neural network model provided in Embodiment 2 of the present invention.

[0048] Figure 3 This is a schematic diagram of the correlation coefficients of each input feature of the model provided in Embodiment 2 of the present invention;

[0049] Figure 4 This is a schematic diagram illustrating the importance of different functional groups to the properties of the output molecule, as provided in Embodiment 2 of the present invention.

[0050] Figure 5 This is a schematic diagram of the prediction results of the training set and the test set provided in Embodiment 2 of the present invention;

[0051] Figure 6 This is a flowchart illustrating a novel environmentally friendly insulating gas molecule design and performance prediction method provided in Embodiment 2 of the present invention.

[0052] Figure 7 This is a structural block diagram of a novel environmentally friendly insulating gas molecule design and performance prediction system provided in Embodiment 3 of the present invention. Detailed Implementation

[0053] This invention provides a novel method and system for designing and predicting the performance of environmentally friendly insulating gas molecules, which addresses the technical problem of poor design efficiency of potential insulating gas molecules in existing methods for designing and predicting the performance of novel environmentally friendly insulating gas molecules.

[0054] To make the objectives, features, and advantages of this invention more apparent and understandable, the technical solutions of the embodiments of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the embodiments described below are only some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0055] Please see Figure 1 , Figure 1 The flowchart illustrates the steps of a novel environmentally friendly insulating gas molecule design and performance prediction method provided in Embodiment 1 of the present invention.

[0056] This invention provides a novel method for the design and performance prediction of environmentally friendly insulating gas molecules, comprising:

[0057] Step 101: Obtain the bonding parameters of the environmentally friendly insulating gas molecules, input the bonding parameters of the environmentally friendly insulating gas molecules into the preset environmentally friendly insulating gas molecule structure design model, and output the initial environmentally friendly insulating gas molecule structure data to be designed. The preset environmentally friendly insulating gas molecule structure design model is a conditional graph generation model.

[0058] The bonding parameters of environmentally friendly insulating gas molecules include molecular coordinates, bond lengths, and bond angles.

[0059] The initial molecular structure data of the environmentally friendly insulating gas to be designed includes the quantum chemical parameters of the novel environmentally friendly insulating gas molecules output by the model, including the positive electrostatic potential surface area (Å). 2 ), molecular volume (Å) 3 ), highest occupied molecular orbital energy level (eV), lowest occupied molecular orbital energy level (eV), polarizability (10 -24 cm 3 ), dipole moment (D), vertical and horizontal ionization energies (eV), bond order, ionization potential, electron affinity, molecular mass, characteristic functional groups, bond energy, etc., wherein characteristic functional groups include, but are not limited to, sulfur-fluorine groups (-SF). x(x=1~5)), -CN (cyano), carbon-fluorine group (-CF) x (x=1~3)), carbonyl (-CO), thiocarbonyl (-CS), nitrogen-containing fluorine group (-NF) x (x=1~2)), carbon-carbon double bond (-C=C), carbon-carbon triple bond, carbon-hydrogen group (-CH) x (x=1~3) etc.

[0060] The pre-built environmentally friendly insulating gas molecular structure design model includes a graph neural network encoder, a latent variable sampling module, a conditional vector embedding module, and a graph structure decoder.

[0061] Specifically, step 101 may include the following sub-steps S11-S14:

[0062] Step S11: The bonding parameters of the environmentally friendly insulating gas molecules are graph encoded using a graph neural network encoder, and a low-dimensional feature representation is output.

[0063] Step S12: Use the latent variable sampling module to sample the low-dimensional feature representation and output the latent variables;

[0064] Step S13: Use the conditional vector embedding module to embed the latent variables into the target performance and output the latent vector;

[0065] Step S14: Input the potential vector into the graph structure decoder for decoding, and output the initial molecular structure data of the environmentally friendly insulating gas to be designed.

[0066] It should be noted that the pre-set environmentally friendly insulating gas molecular structure design model is a conditional graph variational autoencoder (c-GraphVAE). This invention is based on SF6 and other mainstream environmentally friendly insulating gases. The gas skeleton is extracted by configuration stripping method to construct molecular graph structure. Combined with the target performance conditions, the structure of the molecule to be designed is generated by the conditional graph variational autoencoder (c-GraphVAE).

[0067] Furthermore, a graph neural network encoder is used to graph encode the backbone (i.e., the bonding parameters of the environmentally friendly insulating gas molecules), outputting a low-dimensional feature representation. The low-dimensional feature representation is sampled by a latent variable sampling module, and the latent variables are output and embedded with the target performance to jointly generate a latent vector z. Finally, a graph structure decoder is used to decode the latent vector, outputting the initial environmentally friendly insulating gas molecule structure data to be designed. The processing of the graph structure decoder is as follows: the graph structure decoder adds new atoms / functional groups through the addNode module, then establishes chemical bonds through the addEdge module, and finally selects the lowest energy stereoconfiguration, i.e., the initial environmentally friendly insulating gas molecule structure data to be designed, by the selectIsomer module.

[0068] Step 102: Preprocess the initial molecular structure data and bonding parameters of the environmentally friendly insulating gas to be designed, and output the target molecular structure data and bonding parameters of the target environmentally friendly insulating gas.

[0069] It should be noted that this invention performs preprocessing operations to standardize, format, screen features, and reduce dimensionality of the initial environmentally friendly insulating gas molecular structure data and bonding parameters, thereby eliminating noise, unifying the format, and enhancing the validity of the data, providing high-quality input for subsequent performance prediction.

[0070] Specifically, step 102 may include the following sub-steps S21-S23:

[0071] Step S21: Convert the initial design data of the environmentally friendly insulating gas molecule structure and the bonding parameters of the environmentally friendly insulating gas molecule respectively, and output the design data of the environmentally friendly insulating gas molecule structure and the bonding parameters of the environmentally friendly insulating gas molecule in array form.

[0072] Step S22: Normalize the array-formatted environmentally friendly insulating gas molecular structure data and the array-formatted environmentally friendly insulating gas molecular bonding parameters respectively, and output the normalized environmentally friendly insulating gas molecular structure data and the normalized environmentally friendly insulating gas molecular bonding parameters.

[0073] Step S23: Encode the normalized molecular structure data of the environmentally friendly insulating gas to be designed and the normalized bonding parameters of the environmentally friendly insulating gas molecules respectively to generate the target molecular structure data of the environmentally friendly insulating gas to be designed and the target bonding parameters of the environmentally friendly insulating gas molecules.

[0074] It should be noted that before inputting the feature data into the specific neural network model for predicting the performance of environmentally friendly insulating gas molecules, the original data is converted into array form, the data is normalized with SF6 as the reference and encoded to obtain the input of the specific neural network model for predicting the performance of environmentally friendly insulating gas molecules, namely the target environmentally friendly insulating gas molecule structure data and the target environmentally friendly insulating gas molecule bonding parameters.

[0075] Step 103: Input the molecular structure data of the target environmentally friendly insulating gas to be designed and the bonding parameters of the target environmentally friendly insulating gas molecule into the specific neural network model for performance prediction of the pre-set environmentally friendly insulating gas molecule to perform performance prediction, and output the performance value of the target environmentally friendly insulating gas molecule. The specific neural network model for performance prediction of the pre-set environmentally friendly insulating gas molecule is a multilayer perceptron structure, which includes an input layer, a hidden layer and an output layer.

[0076] The target environmentally friendly insulating gas molecule performance values ​​include insulation parameters such as DC / AC critical breakdown field strength, partial discharge initiation voltage, and impulse breakdown voltage; decomposition characteristics such as discharge decomposition products and thermal decomposition products; toxicity parameters such as gas inhalation toxicity, target organ toxicity, and cell action pathway; and performance data such as metal compatibility and non-metal compatibility.

[0077] It should be noted that the specific neural network model for predicting the molecular performance of pre-designed environmentally friendly insulating gases is a multilayer perceptron structure, comprising one input layer, two hidden layers, and one output layer. The input layer receives all molecular feature variables obtained from theory and experiments; the two hidden layers have 64–128 and 32–64 neurons respectively, both using nonlinear activation functions, and a dropout operation (dropout rate of 0.1–0.15) is introduced between hidden layers to reduce the risk of overfitting; the output layer outputs predicted values ​​for the insulation (critical partial discharge / breakdown field strength, etc.), stability (electro / thermal decomposition conditions, rate, product composition, etc.), environmental friendliness (GWP, atmospheric lifetime, infrared radiation intensity, etc.), and biosafety (acute inhalation toxicity, target organ toxicity, toxic pathways, etc.) of the environmentally friendly insulating gas, i.e., the performance values ​​of the environmentally friendly insulating gas molecules to be designed.

[0078] In this embodiment of the invention, a novel method for designing and predicting the performance of environmentally friendly insulating gas molecules is provided. The method involves obtaining bonding parameters of environmentally friendly insulating gas molecules and inputting these parameters into a pre-set environmentally friendly insulating gas molecule structure design model to output initial molecular structure data of the environmentally friendly insulating gas to be designed. The initial molecular structure data and bonding parameters are pre-processed to output target molecular structure data and target bonding parameters of the environmentally friendly insulating gas to be designed. The target molecular structure data and bonding parameters are then input into a pre-set specific neural network model for performance prediction of environmentally friendly insulating gas molecules to perform performance prediction, outputting the performance value of the target environmentally friendly insulating gas molecule. Based on the above scheme, this invention utilizes a pre-set environmentally friendly insulating gas molecule structure design model to obtain initial molecular structure data, eliminating the need for repeated synthesis of material samples and insulation performance testing and GWP evaluation. This accelerates the design process of novel insulating gases, thereby improving the efficiency of environmentally friendly insulating gas molecule design.

[0079] For better illustration, refer to Figure 2 The diagram illustrates the model training process of the pre-set environmentally friendly insulating gas molecule structure design model and the specific neural network model for predicting the performance of pre-set environmentally friendly insulating gas molecules provided in Embodiment 2 of the present invention. This process may include the following steps:

[0080] Step 201: Obtain the property data of the environmentally friendly insulating gas, and input the bonding parameters of the environmentally friendly insulating gas molecules used for model training into the initial environmentally friendly insulating gas molecule structure design model, and output the initial environmentally friendly insulating gas molecule structure data to be designed for model training.

[0081] The initial environmentally friendly insulating gas molecular structure design model is an untrained pre-set environmentally friendly insulating gas molecular structure design model.

[0082] It should be noted that the analysis of existing characteristic data of environmentally friendly insulating gases and the establishment of a molecular characteristic dataset for insulating gases constitute the collection of characteristic data for all existing known environmentally friendly insulating gases. This characteristic data includes molecular bonding parameters, molecular structure data, and experimental data for environmentally friendly insulating gases used in model training. The molecular bonding parameters used in model training include molecular coordinates, bond lengths, and bond angles. The molecular structure data used in model training includes quantum chemical parameters, such as the surface area of ​​the positive electrostatic potential (Å). 2 ), molecular volume (Å) 3), highest occupied molecular orbital energy level (eV), lowest occupied molecular orbital energy level (eV), polarizability (10 -24 cm 3 ), dipole moment (D), vertical ionization energy and horizontal ionization energy (eV), bond order, ionization potential, electron affinity, as well as molecular mass, characteristic functional groups, bond energy, etc.; environmental insulating gas test data include insulation parameters such as DC / AC critical breakdown field strength, partial discharge initiation voltage, and impulse breakdown voltage, decomposition characteristics such as discharge decomposition products and thermal decomposition products, toxicity parameters such as gas inhalation toxicity, target organ toxicity, and cell action pathways, and experimental data such as metal compatibility and non-metal compatibility.

[0083] Furthermore, the molecular structure data and bonding parameters of environmentally friendly insulating gases used for model training can be calculated based on first-principles density functional theory and reactive molecular dynamics methods. Experimental data for environmentally friendly insulating gases can be obtained through insulating gas discharge characteristic testing, gas discharge basic parameter measurement and diagnostic experiments, gas decomposition characteristic testing, gas-solid material compatibility testing, and gas biosafety testing. Specifically, insulating gas discharge characteristic testing involves applying different voltages to the insulating gas to determine the breakdown field strength and voltage; discharge basic parameter measurement and diagnostic experiments measure partial discharge, ionization coefficient, electron adhesion coefficient, and critical breakdown field strength; gas decomposition characteristic testing applies voltage or high temperature to the gas to promote decomposition and measures the type and content of decomposition gas products; gas-solid material compatibility testing uses accelerated thermal aging tests to determine the morphology, structure, and mechanical / electrothermal properties of gas components and solid materials; and gas biosafety testing uses animal inhalation exposure tests to determine acute inhalation toxicity.

[0084] Furthermore, an initial environmentally friendly insulating gas molecular structure design model is established. After model training, a molecular backbone expansion strategy is used to obtain the novel environmentally friendly insulating gas molecule to be designed. That is, the molecular structure is designed using the trained pre-set environmentally friendly insulating gas molecular structure design model to obtain the initial environmentally friendly insulating gas molecule structure data to be designed. The novel environmentally friendly insulating gas molecule obtained by the model design (the initial environmentally friendly insulating gas molecule structure data to be designed) is used as input to a trained pre-set specific neural network model for predicting the performance of environmentally friendly insulating gas molecules. The core performance is then predicted and obtained using the trained pre-set specific neural network model for predicting the performance of environmentally friendly insulating gas molecules.

[0085] It is worth mentioning that this invention uses the idea of ​​molecular skeleton expansion to obtain preliminary insulating gas molecules to be designed. The properties of these gas molecules are then used as input to a molecular prediction model to predict insulating gas molecules with excellent insulating properties. Specifically, based on SF6 and other mainstream environmentally friendly insulating gases, the gas skeleton is extracted using a configuration stripping method to construct a molecular graph structure. Combined with target performance conditions, a Conditional Graph Variational Autoencoder (c-GraphVAE) is used to generate the structure of the molecules to be designed. During the training process, constraints on the rationality of the molecular structure are incorporated to ensure the integrity and chemical validity of the generated molecules, ultimately obtaining a generative model with performance condition control capabilities.

[0086] In this embodiment, the input characteristic molecular properties are divided into quantum chemical parameters, including the surface area of ​​the positive electrostatic potential (Å). 2 ), molecular volume (Å) 3 The molecular structure is calculated using Material Studio software. The parameters include the highest occupied molecular orbital energy level (eV), the lowest occupied molecular orbital energy level (eV), polarizability (10⁻²⁴ cm³), dipole moment (D), vertical and horizontal ionization energies (eV), and structural parameters such as molecular mass and total number of electrons. The output molecular environmental insulation properties include relative SF6 breakdown strength, liquefaction temperature, and GWP value. The results are obtained through simulation using quantum chemistry and molecular dynamics methods, while the structural parameters are directly derived from the molecular formula. Specifically, before calculation, molecular structure optimization is performed using the COMPASS II force field, with an energy convergence threshold set to 1 × 10⁻⁴. -5 With a maximum atomic displacement limited to 0.001 Å and kcal / mol, the molecular structure was optimized using Gaussian 09 software with the B3LYP / 6-311G(d,p) method. Subsequently, the Dmol3 module was used to perform relevant molecular structure model calculations to obtain the relevant parameters of the input molecular properties.

[0087] Step 202: Calculate the graph reconstruction loss based on the molecular structure data of the environmentally friendly insulating gas to be designed and the molecular structure data of the environmentally friendly insulating gas used for model training.

[0088] It should be noted that the computational graph reconstruction loss is based on the molecular structure generated by the model and the real molecular structure. That is, it is based on the molecular structure data of the environmentally friendly insulating gas to be designed in the initial data of the environmentally friendly insulating gas to be trained and the molecular structure data of the environmentally friendly insulating gas in the data of the environmentally friendly insulating gas properties used for model training. The computational graph reconstruction loss can refer to the process of existing computational graph reconstruction loss, and will not be elaborated on further in this invention.

[0089] Step 203: Preprocess the environmentally friendly insulating gas molecule structure data and the environmentally friendly insulating gas molecule bonding parameters used for model training, and output the preprocessed environmentally friendly insulating gas molecule structure data and the preprocessed environmentally friendly insulating gas molecule bonding parameters.

[0090] It should be noted that the preprocessing process for the environmentally friendly insulating gas molecular structure data and the bonding parameters of the environmentally friendly insulating gas molecules used for model training can refer to the preprocessing process for the initial environmentally friendly insulating gas molecular structure data and the bonding parameters of the environmentally friendly insulating gas molecules in the above embodiment. This invention will not elaborate further.

[0091] Furthermore, all data were divided into a 60%–90% training set and a 10%–40% test set for a large machine learning model, with the Bayesian regularized backpropagation algorithm used for training. Both models were trained using the Adam optimizer (Adaptive Moment Estimation). The loss function of the pre-set specific neural network model for predicting the molecular performance of environmentally friendly insulating gases was Mean Squared Error (MSE). The pre-set molecular structure design model for environmentally friendly insulating gases used reconstruction loss, KL divergence loss, and performance control loss as optimization objectives. During training, molecular structure rationality constraints were added to ensure the integrity and chemical validity of the generated molecules, ultimately obtaining a generative model with performance condition control capabilities. The number of training rounds was set to 200–400.

[0092] It is worth mentioning that the model training employs k-fold cross-validation (k=5~10), which involves dividing the data into 5~10 groups, alternately using 1~2 groups as the validation set, and the remaining 4~8 groups as the training set for both training and evaluation; finally, the average of the 5~10 validation results is taken as the evaluation metric for the overall performance of the network model. The coefficient of determination R can be used as the evaluation metric for the network model. 2 The Coefficient of Determination (R0) and the standard deviation σ are used to evaluate the accuracy of the prediction model. 2 σ is used to measure the proportion of the target variable's variability explained by the model and to evaluate the model's overall predictive ability; σ is used to reflect the degree of data dispersion and to further evaluate the model's error stability.

[0093] In this embodiment, the GWP value in the training data is obtained from the IPCC AR5 (Intergovernmental Panel on Climate Change Fifth Assessment Report) public database or literature values, the insulation strength is extracted through experiments or literature, normalized with SF6 as the benchmark, and the liquefaction temperature is referenced from the NIST (National Institute of Standards and Technology) database or literature.

[0094] Step 204: Input the preprocessed environmentally friendly insulating gas molecule structure data and the preprocessed environmentally friendly insulating gas molecule bonding parameters into the specific neural network model for the initial environmentally friendly insulating gas molecule performance prediction, and output the gas molecule performance values ​​for model training.

[0095] The gas molecule performance values ​​used for model training include insulation parameters such as DC / AC critical breakdown field strength, partial discharge initiation voltage, and impulse breakdown voltage of the gas output by the model; decomposition characteristics such as discharge decomposition products and thermal decomposition products; toxicity parameters such as gas inhalation toxicity, target organ toxicity, and cellular action pathways; and data on metal compatibility and non-metal compatibility.

[0096] It should be noted that a specific neural network model for predicting the molecular properties of novel environmentally friendly insulating gases (the initial specific neural network model for predicting the molecular properties of environmentally friendly insulating gases) is established. The input features of the environmentally friendly insulating gas molecules are used as the input vector of the feedforward neural network, and the core gas properties (relative breakdown strength, liquefaction temperature, global warming potential (GWP), electrothermal stability, compatibility with metallic / non-metallic materials, biosafety, etc.) are used as the output of the neural network. The neural network is iteratively trained using a large machine learning model.

[0097] In this embodiment, the input molecular features are used as the input vector of a specific neural network model for predicting the molecular performance of initial environmentally friendly insulating gases, and the output molecular characteristics are used as the output of the same model. The specific neural network model is iteratively trained to train a neural network model specifically designed for molecular prediction. Specifically, before the feature data is input, the original data is converted into an array format, the data is normalized, and data that deviates excessively from the average value of all gases is corrected. Finally, the data is encoded, and all data is split into training and test sets at a ratio of 90% to 10%.

[0098] Step 205: Calculate the regression loss based on the gas molecule performance values ​​and environmental insulating gas characteristic data used for model training, including the environmental insulating gas test data.

[0099] It should be noted that the regression loss is calculated using the predicted performance values ​​of the molecules generated by the model and the target performance values ​​(i.e., the test data of the environmentally friendly insulating gas obtained through experiments). The process of calculating the regression loss can refer to the existing process of calculating regression loss, and will not be elaborated on further in this invention.

[0100] Step 206: Substitute the regression loss and graph reconstruction loss into the preset total loss function and calculate the derivative to output the model gradient.

[0101] It should be noted that the preset total loss function is: L = L e +β*L KL +λ*L property Where L is the loss value corresponding to the preset total loss function; L e To generate the graph reconstruction loss between the molecular structure and the real molecular structure, L KL The KL (Kullback-Leibler) divergence loss is the difference between the latent distribution and the standard normal distribution, while L... property The regression loss is calculated between the predicted and target performance values ​​of the generated molecules, where β and λ are adjustable weighting factors.

[0102] Step 207: Use model gradient to update the model parameters of the initial environmentally friendly insulating gas molecular structure design model and the specific neural network model for predicting the performance of the initial environmentally friendly insulating gas molecules, and output the intermediate environmentally friendly insulating gas molecular structure design model and the specific neural network model for predicting the performance of the intermediate environmentally friendly insulating gas molecules.

[0103] Step 208: Count the number of model updates in real time and determine whether the number of model updates has reached the preset number of training iterations.

[0104] Step 209: If achieved, use the intermediate environmentally friendly insulating gas molecule structure design model as the trained pre-set environmentally friendly insulating gas molecule structure design model, and use the specific neural network model for predicting the performance of intermediate environmentally friendly insulating gas molecules as the trained pre-set specific neural network model for predicting the performance of environmentally friendly insulating gas molecules.

[0105] Optionally, if the number of model updates does not reach the preset number of training iterations, the intermediate environmentally friendly insulating gas molecule structure design model is used as the new initial environmentally friendly insulating gas molecule structure design model, and the specific neural network model for predicting the performance of intermediate environmentally friendly insulating gas molecules is used as the new specific neural network model for predicting the performance of initial environmentally friendly insulating gas molecules.

[0106] Jump to execute the steps of inputting the bonding parameters of the environmentally friendly insulating gas molecules used for model training from the environmentally friendly insulating gas property data into the initial environmentally friendly insulating gas molecule structure design model, and outputting the initial environmentally friendly insulating gas molecule structure data to be designed for model training, until the number of model updates reaches the preset number of training times;

[0107] The intermediate environmentally friendly insulating gas molecule structure design model determined when the model update count reaches the preset training count is used as the pre-trained environmentally friendly insulating gas molecule structure design model, and the specific neural network model for predicting the performance of intermediate environmentally friendly insulating gas molecules is used as the pre-trained specific neural network model for predicting the performance of environmentally friendly insulating gas molecules.

[0108] It should be noted that if the number of model updates does not reach the preset number of training iterations, the intermediate environmentally friendly insulating gas molecule structure design model will be used as the new initial environmentally friendly insulating gas molecule structure design model, and the specific neural network model for predicting the performance of intermediate environmentally friendly insulating gas molecules will be used as the new initial specific neural network model for predicting the performance of environmentally friendly insulating gas molecules. Then, step 201 will be executed until the number of model updates reaches the preset number of training iterations.

[0109] For example, neural networks were used to predict the main environmental insulation properties of molecules after functional group substitution. Relevant data and literature were consulted and quantum chemical calculations were performed to establish a database of environmentally friendly insulating molecules, as shown in Table 1 (partial list):

[0110] Table 1 Environmental Insulation Molecular Database

[0111]

[0112] Furthermore, the dataset was split into training and testing sets in a 9:1 ratio, and constructed using the Tensorflow library provided by Python. A neural network was then built based on the aforementioned structure, and the training results are as follows: Figure 1 As shown. After training, R 2 The values ​​of σ are 0.821 and 0.138, respectively, indicating that the model has good predictive performance.

[0113] The Pearson correlation coefficients of each parameter are calculated based on the model training results, as follows: Figure 3 According to the correlation analysis results, molecular volume, polarizability, positive electrostatic potential surface area and molecular mass have a high correlation coefficient, which should be the main focus in the initial design of environmentally friendly insulating molecules.

[0114] For the structural design of molecular backbone extension, SF6 was used as the basic backbone input graph neural network encoder, and insulation properties were set as the output target, thus realizing the preliminary design of a novel environmentally friendly insulating gas molecule. Seven output molecules with relatively obvious characteristics are listed, and their relevant molecular properties are shown in Table 2:

[0115] Table 2 Relevant Molecular Properties

[0116]

[0117] Based on the model results, the importance parameters of different functional groups on the properties of each output molecule can be obtained. For the four functional groups mentioned above, their importance parameters for the model output molecule properties such as relative breakdown voltage, liquefaction temperature, and GWP are as follows: Figure 4 As shown, the comparison between the predicted values ​​and the true values ​​on the training and test sets after model training is as follows: Figure 5 As shown.

[0118] In this embodiment, a feedforward neural network (a specific neural network model for predicting the initial properties of environmentally friendly insulating gas molecules) is trained using known environmentally friendly insulating gas molecule data as the training set. This yields a predictive model for unknown molecules, predicting their relative SF6 breakdown strength, liquefaction temperature, and GWP value based on the input molecular properties. This effectively improves the gas design efficiency. For any configuration of unknown gas molecules, this neural network model can effectively predict and design the relevant properties of environmentally friendly insulating gases simply by inputting the molecular physicochemical properties. It achieves simultaneous prediction from input features to multiple target properties, improving the flexibility and parallelism of the molecular design process, reducing the complexity of model deployment, and providing experimental directions for the optimization and design of related media.

[0119] For comparison of technical effects, existing technologies can be referenced. Traditional insulating gases such as sulfur hexafluoride (SF6) have excellent insulating properties and have long been widely used in high-voltage electrical equipment. Currently, research on environmentally friendly alternative gases mainly involves experimentally determining the gas power consumption (GWP) and insulating properties of candidate gases. However, experimental methods are time-consuming, labor-intensive, and costly. Therefore, there is an urgent need for an efficient and economical alternative technology to rapidly design potential environmentally friendly insulating gases.

[0120] For equipment such as electrically insulated switchgear (GIS), gas-insulated lines (GIL), gas-insulated transformers (GIT), and high-voltage circuit breakers (GCB), the electrical strength of the insulating gas is the most critical parameter. Furthermore, in addition to insulation performance, the gas pressure performance (GWP) and liquefaction temperature (LST) must be considered to meet environmental insulation requirements and operational capabilities under high-pressure environments.

[0121] In existing technologies, the design methods for insulating gases mainly include trial and error and computer-aided design. However, the trial and error method is too time- and resource-intensive. As for computer-aided design, some studies have used single artificial neural networks (ANNs) to predict the gas's gas power-to-insulation (GWP) or insulation performance separately. However, these methods fail to effectively achieve joint evaluation of multiple indicators, resulting in low model training efficiency and insufficient data utilization. Therefore, it is necessary to propose a new method to effectively generate novel environmentally friendly insulating gas molecules that meet target insulation characteristics, while simultaneously predicting their GWP and insulation performance, significantly improving the design efficiency of environmentally friendly gases.

[0122] To address the above problems, this invention proposes a novel method for the design and performance prediction of environmentally friendly insulating gas molecules. Please refer to [link / reference]. Figure 6 This study analyzes existing characteristic data of environmentally friendly insulating gases and establishes a dataset of insulating gas molecular characteristics. A specific neural network model is developed for predicting the performance of novel environmentally friendly insulating gas molecules. The input molecular characteristics of the environmentally friendly insulating gas are used as the input vector of a feedforward neural network, and the output molecular characteristics are used as the output of the neural network. A large-scale machine learning model is used to iteratively train the neural network. A design model for the molecular structure of novel environmentally friendly insulating gases is also established. Known molecular structures of environmentally friendly insulating gases are input into the model in a graph structure form for training. A molecular skeleton expansion strategy is used to obtain the novel environmentally friendly insulating gas molecules to be designed. Quantum chemical simulations are performed on the novel environmentally friendly insulating gas molecules obtained from the model design. The obtained gas molecule properties are used as input to the molecular prediction model. The aforementioned specific neural network model is used to predict and obtain their core properties, such as relative breakdown strength, liquefaction temperature, global warming potential (GWP), electrothermal stability, metal / non-metal compatibility, and biocompatibility.

[0123] In this embodiment of the invention, characteristic parameters of gas molecules are obtained through quantum chemical calculations, constructing a molecular property dataset containing multi-dimensional features such as electronic structure, polarizability, dipole moment, and ionization energy. Relative breakdown strength, liquefaction temperature, global warming potential (GWP), and biocompatibility are used as target outputs to form standard training samples. Subsequently, a feedforward neural network model is used to model the mapping relationship between molecular features and target properties. Bayesian regularization and k-fold cross-validation are employed to optimize network parameters, improving the model's prediction accuracy and generalization ability. Furthermore, an environmentally friendly insulating gas molecule design model based on molecular backbone extension is constructed to create novel environmentally friendly insulating gas molecules. These molecules are then input into a trained environmentally friendly insulating gas molecule performance prediction model for performance prediction and design. This invention achieves rapid evaluation of a large number of candidate molecules while maintaining accuracy, significantly improving the design efficiency of environmentally friendly insulating gases, reducing R&D costs, and accelerating the design process of novel insulating gases. It can be used for rapid and efficient initial screening of gas molecules, thereby effectively supporting the development of alternatives to SF6 in high-voltage insulation applications.

[0124] Please see Figure 7 , Figure 7 This is a structural block diagram of a novel environmentally friendly insulating gas molecule design and performance prediction system provided in Embodiment 3 of the present invention.

[0125] This invention provides a novel environmentally friendly insulating gas molecule design and performance prediction system, comprising:

[0126] The acquisition module 701 is used to acquire the bonding parameters of environmentally friendly insulating gas molecules, input the bonding parameters of environmentally friendly insulating gas molecules into the preset environmentally friendly insulating gas molecule structure design model, and output the initial environmentally friendly insulating gas molecule structure data to be designed. The preset environmentally friendly insulating gas molecule structure design model is a condition graph generation model.

[0127] Preprocessing module 702 is used to preprocess the initial molecular structure data of the environmentally friendly insulating gas to be designed and output the target molecular structure data of the environmentally friendly insulating gas to be designed.

[0128] The prediction module 703 is used to input the molecular structure data of the target environmentally friendly insulating gas to be designed and the bonding parameters of the target environmentally friendly insulating gas molecule into a pre-set specific neural network model for predicting the performance of the environmentally friendly insulating gas molecule, and output the performance value of the target environmentally friendly insulating gas molecule. The pre-set specific neural network model for predicting the performance of the environmentally friendly insulating gas molecule is a multilayer perceptron structure, which includes an input layer, a hidden layer and an output layer.

[0129] Furthermore, the pre-built environmentally friendly insulating gas molecular structure design model includes a graph neural network encoder, a latent variable sampling module, a conditional vector embedding module, and a graph structure decoder; the acquisition module 701 is specifically used for:

[0130] The bonding parameters of environmentally friendly insulating gas molecules are graphically encoded using a graph neural network encoder, and a low-dimensional feature representation is output.

[0131] A latent variable sampling module is used to sample the low-dimensional feature representation and output the latent variables;

[0132] The latent variables are embedded using a conditional vector embedding module to perform target performance embedding and output latent vectors.

[0133] The potential vectors are input into the graph structure decoder for decoding, and the initial molecular structure data of the environmentally friendly insulating gas to be designed is output.

[0134] Furthermore, the preprocessing module 702 is specifically used for:

[0135] The initial molecular structure data and bonding parameters of the environmentally friendly insulating gas to be designed are converted respectively, and the data are output in array form.

[0136] Normalize the array-formatted molecular structure data of the environmentally friendly insulating gas to be designed and the array-formatted bonding parameters of the environmentally friendly insulating gas molecules respectively, and output the normalized molecular structure data of the environmentally friendly insulating gas to be designed and the normalized bonding parameters of the environmentally friendly insulating gas molecules.

[0137] The normalized molecular structure data and bonding parameters of the environmentally friendly insulating gas to be designed are encoded to generate the target molecular structure data and bonding parameters of the target environmentally friendly insulating gas.

[0138] In one optional system embodiment, it further includes:

[0139] The first module is used to acquire the property data of environmentally friendly insulating gas, input the bonding parameters of environmentally friendly insulating gas molecules used for model training from the property data into the initial environmentally friendly insulating gas molecule structure design model, and output the initial environmentally friendly insulating gas molecule structure data to be designed for model training.

[0140] The second module is used to calculate the graph reconstruction loss based on the molecular structure data of the environmentally friendly insulating gas to be designed and the property data of the environmentally friendly insulating gas used for model training.

[0141] The third module is used to preprocess the environmentally friendly insulating gas molecule structure data and the environmentally friendly insulating gas molecule bonding parameters used for model training, and output the preprocessed environmentally friendly insulating gas molecule structure data and the preprocessed environmentally friendly insulating gas molecule bonding parameters.

[0142] The fourth module is used to input the preprocessed environmentally friendly insulating gas molecule structure data and the preprocessed environmentally friendly insulating gas molecule bonding parameters into the specific neural network model for the initial environmentally friendly insulating gas molecule performance prediction, and output the gas molecule performance values ​​for model training.

[0143] The fifth module is used to calculate the regression loss based on the gas molecule performance values ​​and environmental insulating gas characteristic data used for model training, and the environmental insulating gas test data.

[0144] The sixth module is used to substitute the regression loss and graph reconstruction loss into the preset total loss function and perform differentiation to output the model gradient;

[0145] The seventh module is used to update the model parameters of the initial environmentally friendly insulating gas molecular structure design model and the specific neural network model for predicting the initial environmentally friendly insulating gas molecular performance using model gradients, and output the intermediate environmentally friendly insulating gas molecular structure design model and the intermediate environmentally friendly insulating gas molecular performance prediction specific neural network model.

[0146] The eighth module is used to count the number of model updates in real time and determine whether the number of model updates has reached the preset number of training iterations.

[0147] The ninth module is used to, if necessary, use the intermediate environmentally friendly insulating gas molecule structure design model as a pre-trained environmentally friendly insulating gas molecule structure design model, and use the specific neural network model for predicting the performance of intermediate environmentally friendly insulating gas molecules as a pre-trained specific neural network model for predicting the performance of environmentally friendly insulating gas molecules.

[0148] Optionally, it also includes:

[0149] The tenth module is used to take the intermediate environmentally friendly insulating gas molecule structure design model as the new initial environmentally friendly insulating gas molecule structure design model and the specific neural network model for predicting the performance of intermediate environmentally friendly insulating gas molecules as the new initial specific neural network model for predicting the performance of environmentally friendly insulating gas molecules if the number of model updates has not reached the preset number of training times.

[0150] The eleventh module is used to jump to the execution of the steps of inputting the bonding parameters of the environmentally friendly insulating gas molecules used for model training from the environmentally friendly insulating gas property data into the initial environmentally friendly insulating gas molecule structure design model, and outputting the initial environmentally friendly insulating gas molecule structure data to be designed for model training, until the number of model updates reaches the preset number of training times.

[0151] The twelfth module is used to take the intermediate environmentally friendly insulating gas molecule structure design model determined when the model update number reaches the preset training number as the pre-trained environmentally friendly insulating gas molecule structure design model, and to take the specific neural network model for predicting the performance of intermediate environmentally friendly insulating gas molecules as the pre-trained specific neural network model for predicting the performance of environmentally friendly insulating gas molecules.

[0152] Optionally, a preset total loss function is defined as follows:

[0153] L = L e +β*L KL +λ*L property ;

[0154] Where L is the loss value corresponding to the preset total loss function; L e To compensate for the graph reconstruction loss between the generated molecular structure and the real molecular structure; L KLThe KL divergence loss between the latent distribution and the standard normal distribution; L property The regression loss is the difference between the predicted and target performance values ​​of the generated molecule; β and λ are adjustable weighting factors.

[0155] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working process of the system and modules described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0156] This invention also provides a computer device, including a memory and a processor, wherein the memory stores a computer program; when the computer program is executed by the processor, the processor performs the steps of the novel environmentally friendly insulating gas molecule design and performance prediction method as described in any of the above embodiments.

[0157] This invention also provides a computer-readable storage medium storing a computer program / instructions thereon, which, when executed by a processor, implements the steps of the novel environmentally friendly insulating gas molecule design and performance prediction method as described in any of the above embodiments.

[0158] This invention also provides a computer program product, including a computer program / instructions, which, when executed by a processor, implement the steps of the novel environmentally friendly insulating gas molecule design and performance prediction method as described in any of the above embodiments.

[0159] In the several embodiments provided in this application, it should be understood that the disclosed systems and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0160] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0161] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for designing and predicting the performance of environmentally friendly insulating gas molecules, characterized in that, include: The bonding parameters of environmentally friendly insulating gas molecules are obtained and input into a pre-set environmentally friendly insulating gas molecule structure design model. The initial environmentally friendly insulating gas molecule structure data to be designed is output. The pre-set environmentally friendly insulating gas molecule structure design model is a condition graph generation model. The initial molecular structure data of the environmentally friendly insulating gas to be designed and the bonding parameters of the environmentally friendly insulating gas molecules are preprocessed to output the target molecular structure data of the environmentally friendly insulating gas to be designed and the target environmentally friendly insulating gas molecules. The molecular structure data of the target environmentally friendly insulating gas to be designed and the bonding parameters of the target environmentally friendly insulating gas molecule are input into a specific neural network model for predicting the performance of the pre-set environmentally friendly insulating gas molecule to perform performance prediction, and the performance value of the target environmentally friendly insulating gas molecule to be designed is output. The specific neural network model for predicting the performance of the pre-set environmentally friendly insulating gas molecule is a multilayer perceptron structure, which includes an input layer, a hidden layer and an output layer. The model training process for the pre-designed environmentally friendly insulating gas molecule structure design model and the specific neural network model for predicting the performance of the pre-designed environmentally friendly insulating gas molecules is as follows: Obtain environmentally friendly insulating gas characteristic data, and input the bonding parameters of environmentally friendly insulating gas molecules used for model training from the environmentally friendly insulating gas characteristic data into the initial environmentally friendly insulating gas molecule structure design model, and output the initial environmentally friendly insulating gas molecule structure data to be designed for model training. Based on the initial design data of the environmentally friendly insulating gas molecular structure used for model training and the environmentally friendly insulating gas characteristic data used for model training, the graph reconstruction loss is calculated. The molecular structure data of the environmentally friendly insulating gas used for model training and the bonding parameters of the environmentally friendly insulating gas molecules used for model training are preprocessed respectively, and the preprocessed molecular structure data of the environmentally friendly insulating gas and the preprocessed bonding parameters of the environmentally friendly insulating gas molecules are output. The preprocessed environmentally friendly insulating gas molecule structure data and the preprocessed environmentally friendly insulating gas molecule bonding parameters are input into a specific neural network model for predicting the initial environmentally friendly insulating gas molecule performance, and the gas molecule performance values ​​are output for model training. Based on the gas molecule performance values ​​used for model training and the environmentally friendly insulating gas test data in the environmentally friendly insulating gas characteristic data, the regression loss is calculated; Substitute the regression loss and the graph reconstruction loss into the preset total loss function and take the derivative to output the model gradient; The model gradient is used to update the model parameters of the initial environmentally friendly insulating gas molecule structure design model and the specific neural network model for predicting the performance of the initial environmentally friendly insulating gas molecule, and outputs the intermediate environmentally friendly insulating gas molecule structure design model and the specific neural network model for predicting the performance of the intermediate environmentally friendly insulating gas molecule. The number of model updates is counted in real time, and it is determined whether the number of model updates has reached the preset number of training iterations. If this is achieved, the intermediate environmentally friendly insulating gas molecule structure design model will be used as the trained pre-set environmentally friendly insulating gas molecule structure design model, and the specific neural network model for predicting the performance of the intermediate environmentally friendly insulating gas molecule will be used as the trained specific neural network model for predicting the performance of the pre-set environmentally friendly insulating gas molecule.

2. The method for designing and predicting the performance of environmentally friendly insulating gas molecules according to claim 1, characterized in that, The pre-set environmentally friendly insulating gas molecular structure design model includes a graph neural network encoder, a latent variable sampling module, a conditional vector embedding module, and a graph structure decoder; the step of inputting the bonding parameters of the environmentally friendly insulating gas molecules into the pre-set environmentally friendly insulating gas molecular structure design model and outputting the initial environmentally friendly insulating gas molecular structure data to be designed includes: The graph neural network encoder is used to graph encode the bonding parameters of the environmentally friendly insulating gas molecules, and outputs a low-dimensional feature representation. The latent variable sampling module is used to sample the low-dimensional feature representation and output latent variables; The conditional vector embedding module is used to embed the latent variables into the target performance and output a latent vector. The potential vector is input into the graph structure decoder for decoding, and the initial molecular structure data of the environmentally friendly insulating gas to be designed is output.

3. The method for designing and predicting the performance of environmentally friendly insulating gas molecules according to claim 1, characterized in that, The preprocessing of the initial environmentally friendly insulating gas molecular structure data and the bonding parameters of the environmentally friendly insulating gas molecules to output the target environmentally friendly insulating gas molecular structure data and the target environmentally friendly insulating gas molecules includes: The initial design data of the environmentally friendly insulating gas molecule and the bonding parameters of the environmentally friendly insulating gas molecule are converted respectively, and the design data of the environmentally friendly insulating gas molecule and the bonding parameters of the environmentally friendly insulating gas molecule in array form are output. The array-formatted design data of the environmentally friendly insulating gas molecules and the array-formatted bonding parameters of the environmentally friendly insulating gas molecules are normalized respectively, and the normalized design data of the environmentally friendly insulating gas molecules and the normalized bonding parameters of the environmentally friendly insulating gas molecules are output. The normalized molecular structure data of the environmentally friendly insulating gas to be designed and the normalized bonding parameters of the environmentally friendly insulating gas molecules are respectively encoded to generate the target molecular structure data of the environmentally friendly insulating gas to be designed and the target bonding parameters of the environmentally friendly insulating gas molecules.

4. The method for designing and predicting the performance of environmentally friendly insulating gas molecules according to claim 1, characterized in that, Also includes: If the number of model updates does not reach the preset number of training iterations, then the intermediate environmentally friendly insulating gas molecule structure design model will be used as the new initial environmentally friendly insulating gas molecule structure design model, and the specific neural network model for predicting the performance of intermediate environmentally friendly insulating gas molecules will be used as the new specific neural network model for predicting the performance of initial environmentally friendly insulating gas molecules. Jump to execute the step of inputting the bonding parameters of the environmentally friendly insulating gas molecules used for model training from the environmentally friendly insulating gas characteristic data into the initial environmentally friendly insulating gas molecule structure design model, and outputting the initial environmentally friendly insulating gas molecule structure data to be designed for model training, until the number of model updates reaches the preset number of training times; The intermediate environmentally friendly insulating gas molecule structure design model determined when the model update count reaches the preset training count is used as the trained preset environmentally friendly insulating gas molecule structure design model, and the specific neural network model for predicting the performance of intermediate environmentally friendly insulating gas molecules is used as the trained specific neural network model for predicting the performance of preset environmentally friendly insulating gas molecules.

5. The method for designing and predicting the performance of environmentally friendly insulating gas molecules according to claim 1, characterized in that, The preset total loss function is specifically as follows: L = L e +β*L KL +λ*L property ; Where L is the loss value corresponding to the preset total loss function; L e To compensate for the graph reconstruction loss between the generated molecular structure and the real molecular structure; L KL The KL divergence loss between the latent distribution and the standard normal distribution; L property The regression loss is the difference between the predicted and target performance values ​​of the generated molecule; β and λ are adjustable weighting factors.

6. An environmentally friendly insulating gas molecule design and performance prediction system, applied to the environmentally friendly insulating gas molecule design and performance prediction method described in claim 1, characterized in that, include: The acquisition module is used to acquire the bonding parameters of environmentally friendly insulating gas molecules, input the bonding parameters of environmentally friendly insulating gas molecules into a preset environmentally friendly insulating gas molecule structure design model, and output the initial environmentally friendly insulating gas molecule structure data to be designed. The preset environmentally friendly insulating gas molecule structure design model is a conditional graph generation model. The preprocessing module is used to preprocess the initial environmentally friendly insulating gas molecular structure data to be designed and output the target environmentally friendly insulating gas molecular structure data to be designed. The prediction module is used to input the molecular structure data of the target environmentally friendly insulating gas to be designed into a specific neural network model for predicting the molecular performance of the environmentally friendly insulating gas, and output the gas molecule performance value corresponding to the molecular structure data of the target environmentally friendly insulating gas. The specific neural network model for predicting the molecular performance of the environmentally friendly insulating gas is a multilayer perceptron structure, which includes an input layer, a hidden layer and an output layer.

7. A computer device, characterized in that, The method includes a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor performs the steps of the environmentally friendly insulating gas molecule design and performance prediction method as described in any one of claims 1-5.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed, it implements the environmentally friendly insulating gas molecule design and performance prediction method as described in any one of claims 1-5.

9. A computer program product, characterized in that, The computer program product includes a computer program stored on a non-transitory computer-readable storage medium, the computer program including program instructions, wherein when the program instructions are executed by a computer, the computer performs the environmentally friendly insulating gas molecule design and performance prediction method as described in any one of claims 1-5.

Images