Mixed molten salt thermophysical property prediction method based on machine learning

Abstract

The invention relates to the field of molten salt, in particular to a mixed molten salt thermophysical property prediction method based on machine learning. The method comprises the following steps: constructing a plurality of training sets, namely a fused salt melting point training set, a fused salt density training set, a fused salt specific heat training set and a fused salt upper limit temperature training set; wherein each training set comprises a plurality of groups of data, and each group of data comprises feature description parameter data and target parameter data of one kind of mixed molten salt; based on each training set, training one model to obtain a fused salt melting point prediction model, a fused salt density prediction model, a fused salt specific heat prediction model and a fused salt upper limit temperature prediction model; and utilizing the fused salt melting point prediction model, the fused salt density prediction model, the fused salt specific heat prediction model and the fused salt upper limit temperature prediction model to respectively predict the melting point, the density, the specific heat and the upper limit temperature of the mixed fused salt to be predicted. According to the method, the thermophysical property of the mixed molten salt can be predicted quickly and accurately, the trial and error cost can be remarkably reduced, and the research and development process of a molten salt material is accelerated.

Description

Technical Field

[0001] The present invention relates to the field of molten salts, and particularly to a method for predicting the thermophysical properties of mixed molten salts based on machine learning. Background Art

[0002] Molten salt thermal energy storage has the advantages of high capacity density, excellent heat transfer and storage performance, long service life, low cost, good peak shaving performance, and strong grid matching. It is considered as an optimal energy storage technology for large-scale solar thermal power generation, multi-energy complementarity of integrated energy systems, etc. The accurate measurement of its melting point and thermophysical properties has very important guiding significance for the evaluation of molten salt performance, the design and layout of solar thermal power generation systems. For example, stability directly determines the upper temperature limit of molten salt use, and affects the service life of structural materials and the system design of solar thermal power generation. Specific heat and density are important indicators to measure the heat storage capacity of molten salts, and directly determine the amount of molten salts used in the system.

[0003] At present, the trial-and-error method, as a common method for studying the thermophysical properties of multi-component molten salts, has significant defects. Although the development of mixed anion molten salts has expanded the selection range of molten salts, it has also made the design and selection of molten salts more complex. Continuing to rely on the trial-and-error method will lead to high consumption of manpower, material resources and financial resources, and it is often difficult to obtain ideal research results. In addition, the CALPHAD thermodynamic calculation method is an important tool for the design of molten salt melting points. It uses the phase equilibrium and thermodynamic data of binary molten salt systems to predict the phase behavior of multi-component systems. However, the accuracy of its phase diagram calculation is limited in the absence of experimental data. Therefore, this method still needs to rely on a large amount of experimental data to ensure the reliability and effectiveness of the results. To sum up, the existing research methods have deficiencies in both efficiency and accuracy, and it is urgent to explore new theoretical calculation methods to obtain the thermophysical properties of mixed molten salts to guide the research and development of molten salt systems. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to overcome the defects of the prior art and provide a method for predicting the thermophysical properties of mixed molten salts based on machine learning, which can quickly and accurately predict the thermophysical properties of mixed molten salts, can significantly reduce the trial-and-error cost, and accelerate the research and development process of molten salt materials.

[0005] To solve the above technical problem, the technical solution of the present invention is: a method for predicting the thermophysical properties of mixed molten salts based on machine learning, including:

[0006] Constructing multiple training sets, namely a molten salt melting point training set, a molten salt density training set, a molten salt specific heat training set, and a molten salt upper temperature training set; wherein,

[0007] Each training set includes multiple groups of data, and each group of data includes the characteristic description parameter data and target parameter data of a mixed molten salt;

[0008] Based on each training set, train a model to obtain a molten salt melting point prediction model, a molten salt density prediction model, a molten salt specific heat prediction model, and a molten salt upper limit temperature prediction model;

[0009] Use the molten salt melting point prediction model, the molten salt density prediction model, the molten salt specific heat prediction model, and the molten salt upper limit temperature prediction model to predict the melting point, density, specific heat, and upper limit temperature of the molten salt mixture to be predicted, respectively.

[0010] Furthermore, in each set of data in the molten salt melting point training set:

[0011] The characteristic description parameter data includes the molten salt mixture formula, the melting points of each single-component molten salt, each ionic radius, each ionic polarizability, each ionic electronegativity, each ionic valence, and the lattice energy of each single-component molten salt;

[0012] The target parameter data is the melting point of the molten salt mixture.

[0013] Furthermore, in each set of data in the molten salt density training set:

[0014] The characteristic description parameter data includes the molten salt mixture formula, the set temperature of the molten salt mixture, the mass numbers of each single-component molten salt, the melting points of each single-component molten salt, each ionic radius, each ionic polarizability, each ionic electronegativity, each ionic valence, and the lattice energy of each single-component molten salt;

[0015] The target parameter data is the density of the molten salt mixture at the set temperature.

[0016] Furthermore, in the molten salt density training set:

[0017] If the number of data with the same formula exceeds the first preset number of groups, retain the data of the preset number of groups with balanced set temperatures of the molten salt mixture and delete other data;

[0018] If the number of data with the same formula is less than the second preset number of groups, fit the data through an empirical formula to form the data of the second preset number of groups with balanced set temperatures of the molten salt mixture.

[0019] Furthermore, in each set of data in the molten salt specific heat training set:

[0020] The characteristic description parameter data includes the molten salt mixture formula, the set temperature of the molten salt mixture, the melting points of each single-component molten salt, each ionic radius, each ionic polarizability, each ionic electronegativity, each ionic valence, and the lattice energy of each single-component molten salt;

[0021] The target parameter data is the specific heat of the molten salt mixture at the set temperature.

[0022] Furthermore, in the molten salt specific heat training set:

[0023] If the data of the same formulation exceeds the first preset number of groups, retain the data of the preset number of groups with balanced set temperatures of the mixed molten salt, and delete other data;

[0024] If the data of the same formulation is less than the second preset number of groups, fit the data through an empirical formula to form the data of the second preset number of groups with balanced set temperatures of the mixed molten salt.

[0025] Furthermore, in each group of data in the molten salt upper limit temperature training set:

[0026] The characteristic description parameter data includes the mixed molten salt formulation, the melting points of each single-component molten salt, each ionic radius, each ionic polarizability, each ionic electronegativity, each ionic valence, and the lattice energy of each single-component molten salt;

[0027] The target parameter data is the upper limit temperature value of the mixed molten salt.

[0028] Furthermore, the method for predicting the thermal properties of the mixed molten salt based on machine learning further includes:

[0029] Train an error classification model for each prediction model respectively, and use the trained error classification model to classify and judge whether the prediction results of the corresponding model are accurate.

[0030] Furthermore, in the step of constructing the training set, it further includes:

[0031] Perform data cleaning and feature engineering on the characteristic description parameter data in each group of data; among them,

[0032] Data cleaning includes at least one of removing outliers, removing lanthanide elements, removing actinide elements, deleting abnormal values and duplicate values, and missing value processing methods;

[0033] Feature engineering includes drawing a correlation heat map to increase or decrease features, combining different features based on theory, and standardization.

[0034] Furthermore, the molten salt melting point prediction model is an artificial neural network;

[0035] The molten salt density prediction model, the molten salt specific heat prediction model, and the molten salt upper limit temperature prediction model are all support vector machines.

[0036] After adopting the above technical solution, the molten salt thermophysical property prediction system based on machine learning proposed by the present invention can effectively integrate theoretical calculations and data-driven artificial intelligence methods, providing a new solution for the material research and development of complex multi-component systems. Through the application of machine learning technology, the prediction efficiency of molten salt thermophysical properties is significantly improved, the time and cost required by traditional experimental methods are reduced, the trial-and-error cost and experimental material cost are lowered, thus effectively reducing the overall production cost. The implementation of the present invention will promote the development of molten salt technology, provide new research ideas and methods for related fields, promote overall technological progress, and have broad application prospects and popularization value. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 is the technical roadmap of the hybrid molten salt thermophysical property prediction based on machine learning proposed by the present invention;

[0038] Figure 2 is the prediction scatter plot of the melting point mixing prediction model in the present invention;

[0039] Figure 3 is the prediction scatter plot of the density mixing prediction model in the present invention;

[0040] Figure 4 is the prediction scatter plot of the specific heat mixing prediction model in the present invention;

[0041] Figure 5 is the prediction scatter plot of the upper limit temperature mixing prediction model in the present invention;

[0042] Figure 6 is the DSC melting point test chart of the predicted hybrid molten salt in Examples 1, 2, and 3;

[0043] Figure 7 is the density test chart of the predicted hybrid molten salt in Example 1;

[0044] Figure 8 is the DSC specific heat test chart of the predicted hybrid molten salt corresponding to Example 1;

[0045] Figure 9 is the TG test chart of the predicted hybrid molten salt corresponding to Examples 1, 2, and 3;

[0046] Figure 10 is the density test chart of the predicted hybrid molten salt corresponding to Example 2;

[0047] Figure 11 is the DSC specific heat test chart of the predicted hybrid molten salt corresponding to Example 2;

[0048] Figure 12 is the density test chart of the predicted hybrid molten salt corresponding to Example 3;

[0049] Figure 13 It is the DSC specific heat test chart corresponding to the predicted hybrid molten salt in Example 3. Detailed implementation mode

[0050] To make the content of the present invention easier to be clearly understood, the following further detailed description of the present invention is provided according to specific embodiments in conjunction with the accompanying drawings.

[0051] As Figure 1 shown, a method for predicting the thermal properties of hybrid molten salts based on machine learning includes:

[0052] Constructing multiple training sets, namely a molten salt melting point training set, a molten salt density training set, a molten salt specific heat training set, and a molten salt upper limit temperature training set; wherein,

[0053] Each training set includes multiple groups of data, and each group of data includes the characteristic description parameter data and target parameter data of a hybrid molten salt.

[0054] Based on each training set, training a model to obtain a molten salt melting point prediction model, a molten salt density prediction model, a molten salt specific heat prediction model, and a molten salt upper limit temperature prediction model.

[0055] Using the molten salt melting point prediction model, the molten salt density prediction model, the molten salt specific heat prediction model, and the molten salt upper limit temperature prediction model to predict the melting point, density, specific heat, and upper limit temperature of the hybrid molten salt to be predicted respectively.

[0056] Specifically, this embodiment can quickly and accurately predict the thermal properties of hybrid molten salts, and through the prediction of the thermal properties of hybrid molten salts, the best low-melting-point, wide-temperature-range, and large-capacity high-temperature molten salts can be screened out.

[0057] It should be noted that considering the huge molten salt system, it is necessary to screen and narrow the range. The hybrid molten salts involved in the training set of this embodiment and the hybrid molten salts to be predicted are all made of selected high-quality single-component molten salts, and the high-quality single-component molten salts are mainly screened with cost, safety, stability, and specific heat as the screening indicators.

[0058] In this embodiment, in each group of data of the molten salt melting point training set:

[0059] The characteristic description parameter data includes the hybrid molten salt formula, the melting points of each single-component molten salt, each ionic radius, each ionic polarizability, each ionic electronegativity, each ionic valence, and the lattice energy of each single-component molten salt.

[0060] The target parameter data is the melting point of the hybrid molten salt.

[0061] In this embodiment, in each group of data of the molten salt density training set:

[0062] The characteristic description parameter data includes the mixed molten salt formula, the set temperature of the mixed molten salt, the mass numbers of each single-component molten salt, the melting points of each single-component molten salt, each ionic radius, each ionic polarizability, each ionic electronegativity, each ionic valence, and the lattice energy of each single-component molten salt;

[0063] The target parameter data is the density of the mixed molten salt at the set temperature.

[0064] In this embodiment, in the molten salt density training set:

[0065] If the data of the same formula exceeds the first preset number of groups, retain the data of the preset number of groups with balanced set temperatures of the mixed molten salt, and delete other data;

[0066] If the data of the same formula is less than the second preset number of groups, fit the data through an empirical formula to form the data of the second preset number of groups with balanced set temperatures of the mixed molten salt.

[0067] In this embodiment, in each group of data in the molten salt specific heat training set:

[0068] The characteristic description parameter data includes the mixed molten salt formula, the set temperature of the mixed molten salt, the melting points of each single-component molten salt, each ionic radius, each ionic polarizability, each ionic electronegativity, each ionic valence, and the lattice energy of each single-component molten salt;

[0069] The target parameter data is the specific heat of the mixed molten salt at the set temperature.

[0070] In this embodiment, in the molten salt specific heat training set:

[0071] If the data of the same formula exceeds the first preset number of groups, retain the data of the preset number of groups with balanced set temperatures of the mixed molten salt, and delete other data;

[0072] If the data of the same formula is less than the second preset number of groups, fit the data through an empirical formula to form the data of the second preset number of groups with balanced set temperatures of the mixed molten salt.

[0073] Among them, the empirical formula includes one or more of a linear fitting formula, an Arrhenius formula, a polynomial fitting formula, and an Eyring equation.

[0074] Each formula ensures sufficient and reasonable data, and can also provide the accuracy and rationality of the prediction model obtained by training.

[0075] In this embodiment, in each group of data in the molten salt upper limit temperature training set:

[0076] The characteristic description parameter data includes the mixed molten salt formula, the melting points of each single-component molten salt, each ionic radius, each ionic polarizability, each ionic electronegativity, each ionic valence, and the lattice energy of each single-component molten salt;

[0077] The target parameter data is the upper limit temperature value of the mixed molten salt.

[0078] Specifically, based on the melting point characteristics, upper limit temperature characteristics, density characteristics, and specific heat characteristics of the molten salt, this embodiment analyzes the melting point characteristics, upper limit temperature characteristics, density characteristics, and specific heat characteristics of the molten salt, and selects the influencing factors related to the melting point of the molten salt, the influencing factors related to the upper limit temperature of the molten salt, the influencing factors related to the density of the molten salt, and the influencing factors related to the specific heat of the molten salt. This embodiment reasonably selects the characteristic description parameter data of each training set, laying a good foundation for accurate prediction.

[0079] In this embodiment, in the step of constructing the training set, it further includes:

[0080] Performing data cleaning and feature engineering on the characteristic description parameter data in each group of data; where

[0081] Data cleaning includes at least one of removing outliers, removing lanthanide elements, removing actinide elements, deleting abnormal values and duplicate values, and missing value processing methods;

[0082] Feature engineering includes increasing or decreasing features by drawing a correlation heat map, combining different features based on theory, and standardization.

[0083] In this embodiment, the method for predicting the thermal physical properties of the mixed molten salt based on machine learning further includes:

[0084] Training an error classification model for each prediction model respectively, and using the trained error classification model to classify and judge whether the prediction results of the corresponding model are accurate. The error classification model can be a decision tree, logistic regression, etc.

[0085] Among them, the molten salt melting point prediction model and its corresponding error classification model jointly form the melting point mixed prediction model; the molten salt density prediction model and its corresponding error classification model jointly form the density mixed prediction model; the molten salt specific heat prediction model and its corresponding error classification model jointly form the specific heat mixed prediction model; the molten salt upper limit temperature prediction model and its corresponding error classification model jointly form the upper limit temperature mixed prediction model.

[0086] In this embodiment, a support vector machine is selected as the prediction model for the upper limit temperature, density, and specific heat of the molten salt, and an artificial neural network is selected as the prediction model for the melting point of the molten salt. The data processing algorithm and parameter tuning algorithm combined during this period are leave-one-out cross-validation and grid search to prevent model overfitting, and the accuracy rate and R2 score are used as evaluation indicators for judging the model accuracy.

[0087] The calculation formula for the prediction accuracy rate is:

[0088]

[0089] Wherein, Y i is the true value of the target variable of the i-th sample, is the predicted value of the target variable of the i-th sample, er is the error threshold, N is the total number of samples, is the indicator function, defined as follows:

[0090]

[0091] Among them, the calculation formula of the R2 score is:

[0092]

[0093] Wherein, Y i is the true value of the target variable of the i-th sample, is the predicted value of the target variable of the i-th sample, is the average value of the target variable of the i-th sample, and N is the total number of samples.

[0094] As Figures 2 - 5 shown, the R 2 score between the predicted melting point of the molten salt and the true value is as high as 0.7937, and the accuracy rate reaches 90.11%; the R 2 score between the upper limit temperature of the molten salt and the true value is as high as 0.9106, and the accuracy rate reaches 88.09%; the R 2 score between the density of the molten salt and the true value is as high as 0.9918, and the accuracy rate reaches 99.17%; the R 2 score between the specific heat of the molten salt and the true value is as high as 0.9455, and the accuracy rate reaches 97.18%.

[0095] Next, in combination with specific embodiments, the solutions involved in the above embodiments will be introduced in detail.

[0096] Embodiment 1

[0097] In this embodiment, the above machine learning model and corresponding data processing and feature engineering algorithms are used to predict the melting point, upper limit temperature, density, and specific heat of the ternary molten salt NaNO3-KNO3-KCl.

[0098] The network parameters of the melting point hybrid prediction model constructed in this embodiment are shown in Table 1.

[0099] Table 1

[0100] Serial number Network parameter Set value 1 hidden_layer_sizes (30,15) 2 activation logistic 3 solver sgd 4 alpha 0.01 5 max_iter 1000 6 batch_size 32 7 C 1 8 penalty 11 9 solver liblinear

[0101] Collect data such as the mass number (M), melting point (T), ionic radius (r), polarizability (p), electronegativity (e), valence (z), lattice energy (E), etc. of each component of the untrained NaNO₃-KNO₃-KCl ternary molten salt. The specific data are listed in Table 2:

[0102] Table 2

[0103]

[0104]

[0105] Input the NaNO₃-KNO₃-KCl parameter dataset in Table 2 into the melting point mixing prediction model established in this embodiment, and output the predicted melting point value of NaNO₃-KNO₃-KCl: 217.67 °C. The experimentally measured melting point is 211.1 °C. As Figure 6 shown, the error is only 3%.

[0106] The network parameters of the density mixing prediction model constructed in this embodiment are shown in Table 3.

[0107] Table 3

[0108] Serial number Network parameter Set value 1 kernel rbf 2 c 1 3 gamma [0.01,0.1] 4 epsilon 0.1 5 max_depth 5 6 min_samples_leaf 5 7 min_samples_split 10 8 criterion entropy

[0109] Collect data such as the ratio (c), set temperature (Ts), mass number (M), melting point (T), ionic radius (r), polarizability (p), electronegativity (e), valence (z), lattice energy (E), etc. of each component of the untrained NaNO₃-KNO₃-KCl ternary molten salt. The specific data are listed in Table 4:

[0110] Table 4

[0111]

[0112]

[0113] Input the NaNO₃-KNO₃-KCl parameter dataset in Table 4 into the density mixing prediction model established in this embodiment, and output the predicted density value of NaNO₃-KNO₃-KCl: 1.797 g / cm 3 , and the experimentally measured density is 1.805 g / cm 3 , as Figure 7 shown, the error is less than 1%.

[0114] The network parameters of the specific heat mixing prediction model constructed in this embodiment are shown in Table 5.

[0115] Table 5

[0116] Serial number Network parameter Set value 1 kernel rbf 2 c 1 3 gamma [0.01,0.1] 4 epsilon 0.1 5 kernel rbf 6 c 1 7 class_weight balanced

[0117] Collect data such as the ratio (c) of each component of the untrained NaNO3-KNO3-KCl ternary molten salt, the set temperature (Ts), the melting point (T), the ionic radius (r), the polarizability (p), the electronegativity (e), the valence (z), and the lattice energy (E). The specific data are listed in Table 6:

[0118] Table 6

[0119]

[0120] Input the NaNO3-KNO3-KCl parameter data set in Table 6 into the specific heat mixing prediction model established in this embodiment, and output the predicted specific heat value of NaNO3-KNO3-KCl: 1.642 J / g*K. The experimentally measured specific heat is 1.649 J / g*K, as Figure 8 shown, and the error is less than 1%.

[0121] The network parameters of the upper limit temperature mixing prediction model constructed in this embodiment are shown in Table 7.

[0122] Table 7

[0123] Serial number Network parameter Set value 1 kernel rbf 2 c 1 3 gamma [0.01,0.1] 4 epsilon 0.1 5 max_depth 3 6 min_samples_leaf 5 7 min_samples_split 10 8 criterion entropy

[0124] Collect data such as the ratio (c) of each component of the untrained NaNO3-KNO3-KCl ternary molten salt, the melting point (T), the ionic radius (r), the polarizability (p), the electronegativity (e), the valence (z), and the lattice energy (E). The specific data are listed in Table 8:

[0125] Table 8

[0126] <![CDATA[NaNO3]]> <![CDATA[KNO3]]> KCl c 32.6 53.2 14.2 T(K) 581.15 607.15 1049.15 r+ 116 133 133 r- 260 260 181 p+ 0.49 0.73 0.73 p- 1.42 1.42 0.4 e+ 0.93 0.82 0.82 e- 2.96 2.96 3.16 z+ 1 1 1 z- -1 -1 -1 E 520 670 700

[0127] Input the NaNO3-KNO3-KCl parameter data set in Table 8 into the upper limit temperature mixing prediction model established in this embodiment, and output the predicted upper limit temperature value of NaNO3-KNO3-KCl: 632 °C. The experimentally measured upper limit temperature is 619 °C, as Figure 9 shown, and the error is only 2%.

[0128] Example 2

[0129] Different from Example 1, the formula for predicting the melting point in this example is NaNO3-KCl. Data such as the mass number (M), melting point (T), ionic radius (r), polarizability (p), electronegativity (e), valence (z), and lattice energy (E) of NaNO3-KCl are listed in Table 9:

[0130] Table 9

[0131]

[0132]

[0133] The NaNO3-KCl parameter dataset in Table 9 was input into the melting point mixing prediction model established in this embodiment, and the predicted melting point value of NaNO3-KCl was output: 201.69 °C. The experimentally measured melting point was 206.2 °C. As Figure 6 shown, the error was only 2%.

[0134] Different from Example 1, the formula for predicting density in this embodiment is NaNO3-KCl. Data such as the ratio (c) of NaNO3-KCl, the set temperature (Ts), the mass number (M), the melting point (T), the ionic radius (r), the polarizability (p), the electronegativity (e), the valence (z), and the lattice energy (E) are listed in Table 10:

[0135] Table 10

[0136]

[0137]

[0138] The NaNO3-KCl parameter dataset in Table 10 was input into the density mixing prediction model established in this embodiment, and the predicted density value of NaNO3-KCl was output: 1.841 g / cm3. The experimentally measured density was 1.847 g / cm3. As Figure 10 shown, the error was less than 1%.

[0139] Different from Example 1, the formula for predicting specific heat in this embodiment is NaNO3-KCl. Data such as the ratio (c) of NaNO3-KCl, the set temperature (Ts), the mass number (M), the melting point (T), the ionic radius (r), the polarizability (p), the electronegativity (e), the valence (z), and the lattice energy (E) are listed in Table 11:

[0140] Table 11

[0141]

[0142] The NaNO3-KCl parameter dataset in Table 11 was input into the specific heat mixing prediction model established in this embodiment, and the predicted specific heat value of NaNO3-KCl was output: 1.47 J / g*K. The experimentally measured specific heat was 1.49 J / g*K. As Figure 11 shown, the error was approximately 1%.

[0143] Different from Example 1, the formula for predicting the upper limit temperature in this example is NaNO3-KCl. The data such as the ratio (c) of NaNO3-KCl, set temperature (Ts), mass number (M), melting point (T), ionic radius (r), polarizability (p), electronegativity (e), valence (z), lattice energy (E), etc. are listed in Table 12:

[0144] Table 12

[0145] <![CDATA[NaNO3]]> KCl c 26.7 73.3 T(K) 581.15 1049.15 r+ 116 133 r- 260 181 p+ 0.49 0.73 p- 1.42 0.4 e+ 0.93 0.82 e- 2.96 3.16 z+ 1 1 z- -1 -1 E 520 700

[0146] Input the NaNO3-KCl parameter data set in Table 12 into the upper limit temperature mixing prediction model established in this example, and the predicted value of the upper limit temperature of NaNO3-KCl is output: 613 °C. The experimentally measured upper limit temperature is 626 °C. As Figure 9 shown, the error is only 2%.

[0147] Example 3

[0148] Different from Example 1, the formula predicted in this example is NaNO3-Ca(NO3)2. The data such as the mass number (M), melting point (T), ionic radius (r), polarizability (p), electronegativity (e), valence (z), lattice energy (E), etc. of NaNO3-Ca(NO3)2 are listed in Table 13:

[0149] Table 13

[0150]

[0151]

[0152] Input the NaNO3-Ca(NO3)2 parameter data set in Table 13 into the melting point mixing prediction model established in this example, and the predicted value of the melting point of NaNO3-Ca(NO3)2 is output: 213.23 °C. The experimentally measured melting point is 223.1 °C. As Figure 6 shown, the error is only 4%.

[0153] Different from Example 1, the formula for predicting density in this example is NaNO3-Ca(NO3)2. The data such as the ratio (c) of NaNO3-Ca(NO3)2, set temperature (Ts), mass number (M), melting point (T), ionic radius (r), polarizability (p), electronegativity (e), valence (z), lattice energy (E), etc. are listed in Table 14:

[0154] Table 14

[0155]

[0156]

[0157] The parameter dataset of NaNO3-Ca(NO3)2 in Table 14 was input into the density mixing prediction model established in this example, and the predicted density value of NaNO3-Ca(NO3)2 was output: 2.142 g / cm 3 , and the experimentally measured density was 2.134 g / cm 3 , as Figure 12 shown, the error was only 3%.

[0158] Different from Example 1, the formula for predicting specific heat in this example is NaNO3-Ca(NO3)2. The data such as the ratio (c), set temperature (Ts), mass number (M), melting point (T), ionic radius (r), polarizability (p), electronegativity (e), valence (z), lattice energy (E), etc. of NaNO3-Ca(NO3)2 are listed in Table 15:

[0159] Table 15

[0160]

[0161] The parameter dataset of NaNO3-Ca(NO3)2 in Table 15 was input into the specific heat mixing prediction model established in this example, and the predicted specific heat value of NaNO3-Ca(NO3)2 was output: 1.452 J / g*K. The experimentally measured specific heat was 1.461 J / g*K, as Figure 13 shown, the error was about 6%.

[0162] Different from Example 1, the formula for predicting the upper limit temperature in this example is NaNO3-Ca(NO3)2. The data such as the ratio (c), set temperature (Ts), mass number (M), melting point (T), ionic radius (r), polarizability (p), electronegativity (e), valence (z), lattice energy (E), etc. of NaNO3-Ca(NO3)2 are listed in Table 16:

[0163] Table 16

[0164] <![CDATA[NaNO3]]> <![CDATA[Ca(NO3)2]]> c 26.7 42.4 T(K) 581.15 834.15 r+ 116 100 r- 260 260 p+ 0.49 1.09 p- 1.42 1.42 e+ 0.93 1 e- 2.96 2.96 z+ 1 2 z- -1 -1 E 520 370

[0165] The parameter dataset of NaNO3-Ca(NO3)2 in Table 16 was input into the upper limit temperature mixing prediction model established in this example, and the predicted upper limit temperature value of NaNO3-Ca(NO3)2 was output: 544 °C. The experimentally measured upper limit temperature was 537 °C, as Figure 9 shown, the error was only 1%.

[0166] Enlightened by the above-described ideal embodiments of the present invention, through the above description, relevant staff can fully make various changes and modifications without departing from the technical idea of this invention. The technical scope of this invention is not limited to the content in the specification, and its technical scope must be determined according to the scope of the claims.

Claims

1. A method for predicting the thermophysical properties of a mixed molten salt based on machine learning, characterized in that: It includes: Construct multiple training sets, namely a molten salt melting point training set, a molten salt density training set, a molten salt specific heat training set, and a molten salt upper limit temperature training set; where Each training set includes multiple groups of data, and each group of data includes the characteristic description parameter data and target parameter data of a mixed molten salt; Based on each training set, train a model to obtain a molten salt melting point prediction model, a molten salt density prediction model, a molten salt specific heat prediction model, and a molten salt upper limit temperature prediction model; Use the molten salt melting point prediction model, the molten salt density prediction model, the molten salt specific heat prediction model, and the molten salt upper limit temperature prediction model to predict the melting point, density, specific heat, and upper limit temperature of the mixed molten salt to be predicted, respectively.

2. The method for predicting the thermophysical properties of a mixed molten salt based on machine learning according to claim 1, characterized in that: In each group of data of the molten salt melting point training set: The characteristic description parameter data includes the mixed molten salt formula, the melting points of each single-component molten salt, each ionic radius, each ionic polarizability, each ionic electronegativity, each ionic valence, and the lattice energy of each single-component molten salt; The target parameter data is the melting point of the mixed molten salt.

3. The method for predicting the thermophysical properties of a mixed molten salt based on machine learning according to claim 1, characterized in that: In each group of data of the molten salt density training set: The characteristic description parameter data includes the mixed molten salt formula, the set temperature of the mixed molten salt, the mass numbers of each single-component molten salt, the melting points of each single-component molten salt, each ionic radius, each ionic polarizability, each ionic electronegativity, each ionic valence, and the lattice energy of each single-component molten salt; The target parameter data is the density of the mixed molten salt at the set temperature.

4. The method for predicting the thermophysical properties of a mixed molten salt based on machine learning according to claim 3, characterized in that: In the molten salt density training set: If the number of data with the same formula exceeds the first preset number of groups, retain the preset number of groups of data with balanced set temperatures of the mixed molten salt, and delete other data; If the number of data with the same formula is less than the second preset number of groups, fit the data through an empirical formula to form the second preset number of groups of data with balanced set temperatures of the mixed molten salt.

5. The method for predicting the thermophysical properties of a mixed molten salt based on machine learning according to claim 1, characterized in that: In each group of data of the molten salt specific heat training set: The characteristic description parameter data includes the mixed molten salt formula, the set temperature of the mixed molten salt, the melting points of each single-component molten salt, each ionic radius, each ionic polarizability, each ionic electronegativity, each ionic valence, and the lattice energy of each single-component molten salt; The target parameter data is the specific heat of the mixed molten salt at the set temperature.

6. The method for predicting the thermophysical properties of a mixed molten salt based on machine learning according to claim 5, characterized in that: In the molten salt specific heat training set: If the number of data with the same formula exceeds the first preset number of groups, retain the preset number of groups of data with balanced set temperatures of the mixed molten salt, and delete other data; If the number of data with the same formula is less than the second preset number of groups, fit the data through an empirical formula to form the second preset number of groups of data with balanced set temperatures of the mixed molten salt.

7. The method for predicting the thermophysical properties of a hybrid molten salt based on machine learning according to claim 1, wherein: In each set of data in the training set of the upper limit temperature of the molten salt: The feature description parameter data includes the hybrid molten salt formula, the melting points of each single-component molten salt, each ionic radius, each ionic polarizability, each ionic electronegativity, each ionic valence, and the lattice energy of each single-component molten salt; The target parameter data is the upper limit temperature value of the hybrid molten salt.

8. The method for predicting the thermophysical properties of a hybrid molten salt based on machine learning according to claim 1, wherein: It further includes: Training an error classification model for each prediction model respectively, and using the trained error classification model to classify and judge whether the prediction results of the corresponding model are accurate.

9. The method for predicting the thermophysical properties of a hybrid molten salt based on machine learning according to claim 1, wherein: In the step of constructing the training set, it further includes: Performing data cleaning and feature engineering on the feature description parameter data in each set of data; wherein, Data cleaning includes at least one of removing outliers, removing lanthanide elements, removing actinide elements, deleting abnormal values and duplicate values, and missing value processing methods; Feature engineering includes increasing or decreasing features by drawing a correlation heat map, combining different features based on theory, and standardization.

10. The method for predicting the thermophysical properties of a hybrid molten salt based on machine learning according to claim 1, wherein: The melting point prediction model of the molten salt is an artificial neural network; The density prediction model of the molten salt, the specific heat prediction model of the molten salt, and the upper limit temperature prediction model of the molten salt are all support vector machines.

Images