corresponding to the partial structures “CH₃” and “CCC” of the first raw material, and three partial structure input data pieces D_2,1, D_2,2, and D_2,3corresponding to the partial structures “OH”, “CH₃”, and “CCC” of the second raw material, the synthesis unit 14 generates the synthetic input data F in which five vectors V_Mcorresponding to such five partial structure input data pieces D_1,1, D_1,2, D_2,1, D_2,2, and D_2,3are compiled.

In this case, the synthesis unit 14 reflects the blending ratio data and the number data corresponding to each of the partial structures on five vectors V_Mcorresponding to such five partial structure input data pieces D_1,1, D_1,2, D_2,1, D_2,2, and D_2,3to obtain the weighted average of five vectors V_Mand generate the synthetic input data F. More specifically, the synthesis unit 14 multiplies the element of each of five vectors V_Mcorresponding to the partial structure input data pieces D_1,1, D_1,2, D_2,1, D_2,2, and D_2,3, by the ratio r and the number n corresponding to each of the partial structures, and then, adds (or averages) the element of each of five vectors V_Mto generate the synthetic input data F. For example, the synthesis unit 14 multiplies the vector V_Mcorresponding to the partial structure of the first raw material by a value obtained by multiplying the ratio r₁of the first raw material and the number n of the partial structure together, and multiplies the vector V_Mcorresponding to the partial structure of the second raw material by a value obtained by multiplying the ratio r₂of the second raw material and the number n of the partial structure together. As an example, the vector V_Mcorresponding to the partial structure “CH₃” of the first raw material illustrated in FIG. 3A is multiplied by a value (0.5×2=1.0) obtained by multiplying Ratio r₁=0.5 and “2”, which is the number of the partial structure, together, and the vector V_Mcorresponding to the partial structure “OH” of the second raw material illustrated in FIG. 3B is multiplied by a value (0.5×1=0.5) obtained by multiplying Ratio r₂=0.5 and “1”, which is the number of the partial structure, together. Here, the reflection of the ratio and the number data may be performed by adding the value obtained by multiplying the ratio r and the number n together to each of the elements of the vector V_M, or may be performed by concatenating the value obtained by multiplying the ratio r and the number n together to the element of the vector. In addition, more generally, the synthesis unit 14 may generate a vector by using a function of outputting a single vector in accordance with a certain rule, with the vector, the blending ratio, and the number relevant to all the partial structures as input, or may generate a single vector in one processing without dividing the processing into a step of reflecting the blending ratio and a step of adding the vector.

Further, the synthesis unit 14 outputs the generated synthetic input data F to the outside such that the data is input to an external machine learning model. That is, the output synthetic input data F is read by the training unit 20 in a computer connected to the outside of the input data generation system 10. Then, in the training unit 20, a learned model is generated by inputting the synthetic input data F to the machine learning model as an explanatory variable, together with any training label. Further, a machine learning model in the predictor 30 is set on the basis of the learned model generated by the training unit 20. Here, the training unit 20 and the predictor 30 may be the same functional unit. Then, by inputting the synthetic input data F generated by the input data generation system 10 to the machine learning model in the predictor 30, the predictor 30 generates and outputs a prediction result of the properties of the multi-component substance. Note that, the training unit 20 and the predictor 30 may be configured in the same computer as the computer 100 configuring the input data generation system 10, or may be configured in a computer different from the computer 100.

In an example, the training unit 20 generates a learned model using a neural network. The learned model is generated by a computer processing training data including a plurality of combinations of input data and output data. The computer calculates the output data by inputting the input data to the machine learning model, and obtains an error between the calculated output data and the output data indicated by the training data (that is, an error between an estimation result and a ground truth). Then, the computer updates given parameters of the neural network, which is the machine learning model, on the basis of the error. The computer generates the learned model by repeating such learning. The processing of generating the learned model can be referred to as a learning phase, and the processing of the predictor 30 using the learned model can be referred to as an operation phase.

[Operation of System]

The operation of the property prediction system 1 will be described, and a property prediction method according to this embodiment will be described, with reference to FIG. 4. FIG. 4 is a flowchart illustrating an example of the operation of the property prediction system 1.

First, in a case where input data generation processing is started in accordance with the instruction input of the user of the input data generation system 10, the acquisition unit 11 is allowed to acquire the raw material structure data and the blending ratio data for each of the plurality of raw materials, and acquire the partial structure data indicating the plurality of known partial structures from the partial structure database 16 (step S1). Next, the generation unit 12 is allowed to search for the partial structure in the molecular structure of the raw material indicated by each of the raw material structure data pieces to specify the partial structure in the molecular structure and the number thereof (step S2). Further, the generation unit 12 is allowed to generate the partial structure input data D0 for each of the partial structures in the molecular structures of the plurality of raw materials (step S3). After that, the vector conversion unit 13 is allowed to convert each of all the partial structure input data pieces D0 into one vector V_Min a vector format, and combine the data of the ratio r relevant to the raw material having the partial structure, and the number data of the partial structure in the raw material, with the vector V_M, to generate the partial structure input data D (step S4).

Next, the synthesis unit 14 is allowed to compile the vectors V_Mcorresponding to all the partial structure input data pieces D for each of the plurality of raw materials to generate the synthetic input data F (step S5). In this case, the synthesis unit 14 is allowed to compute the weighted average of the vectors V_Mwhile reflecting the blending ratio data and the number data on each of the vectors V_Mto generate the synthetic input data F. After that, the synthesis unit 14 is allowed to output the synthetic input data F to the training unit 20 as input data for machine learning (step S6). In this case, the reflection of the ratio and the number on the vector V_Mis performed by multiplying, adding, or concatenating the value obtained by multiplying the ratio and the number together with respect to each of the vectors V_M. In addition, more generally, the synthesis unit 14 may generate the vector by using the function of outputting the single vector in accordance with a certain rule, with the vector, the blending ratio, and the number relevant to all the partial structures as input, or may generate the single vector in one processing without dividing the processing into the step of reflecting the blending ratio and the step of adding the vector.

Next, in the training unit 20, the learning phase is executed, and the learned model is generated by learning using the input data and the training data (step S7). Then, the generated learned model is set in the predictor 30, and the predictor 30 is allowed to execute the operation phase by using the input data newly acquired from the input data generation system 10, and generate and output the prediction result of the properties of the multi-component substance (step S8).

[Program]

A property prediction program for allowing a computer or a computer system to function as the property prediction system 1 includes a program code for allowing the computer system to function as the acquisition unit 11, the generation unit 12, the vector conversion unit 13, the synthesis unit 14, the partial structure database 16, the training unit 20, and the predictor 30. Such a property prediction program may be provided by being regularly recorded in a physical recording medium such as a CD-ROM, a DVD-ROM, and a semiconductor memory. Alternatively, the property prediction program may be provided via a communication network as a data signal superimposed on a carrier wave. The provided property prediction program, for example, is stored in the auxiliary storage unit 103. By the processor 101 reading and executing the property prediction program from the auxiliary storage unit 103, each of the functional constituents described above is attained.

[Effect]

As described above, according to the embodiment described above, the partial structure input data D expressing the known partial structure in the structure of the raw material is generated on the basis of the structure data of the raw material and the structure data of the partial structure acquired from the database, and the input data is generated by reflecting the ratio of the raw material on the partial structure input data D, and compiling the partial structure input data D. Then, the generated input data is input to the machine learning model. As a result thereof, the input data narrowed down to the data of the partial structure data in advance from the entire structure of the raw material is generated, and by processing the input data with the machine learning, it is possible to efficiently predict the properties of the material.

In addition, in the embodiment described above, the input of the raw material structure data relevant to the plurality of raw materials, and the blending ratio data indicating the ratio of the blending of each of the plurality of raw materials is received, the partial structure input data is generated for each of the plurality of raw materials, and the input data is generated by reflecting the blending ratio data relevant to the plurality of raw materials on the partial structure input data of the plurality of raw materials. In this case, the partial structure input data D0 expressing the known partial structure is generated on the basis of the raw material structure data for each of the plurality of raw materials, and the synthetic input data F is generated by reflecting the ratio of each of the plurality of raw materials on the partial structure input data D0 for each of the plurality of raw materials. Then, the generated synthetic input data F is input to the machine learning model. As a result thereof, by processing the input data with the machine learning for the multi-component substance to be produced on the basis of the plurality of raw materials, it is possible to efficiently predict the properties of the substance.

In addition, in the embodiment described above, the number of the partial structure existing in the structure of the raw material is specified, and the input data is generated by reflecting the value obtained by multiplying the blending ratio data and the number of the partial structure together on the partial structure input data of the raw material. In this case, the number of partial structures in the molecules of the raw material is specified, and the ratio of the raw material and the number of partial structures can be reflected on the partial structure input data D0. As a result thereof, it is possible to accurately predict the properties of the multi-component substance based on the raw material.

Further, in the embodiment described above, the input data to be input to the machine learning model is generated by multiplying, adding, or concatenating a value based on the blending ratio data with respect to the vector included in the partial structure input data D, and compiling the multiplied, added, or concatenated vector on one vector. Accordingly, it is possible to effectively and simply reflect the ratio of the raw material on the partial structure input data D relevant to the partial structure of the raw material. As a result thereof, the prediction accuracy of the properties of the multi-component substance is improved.

Modification Example

The present invention has been described in detail, on the basis of the embodiment. However, the present invention is not limited to the embodiment described above. The present invention can be variously modified within a range not departing from the gist thereof.

In the embodiment described above, an example has been described in which the input data generation system 10 generates the synthetic input data F by combining the partial structure input data D0 of two raw materials, but the partial structure input data D0 of three or more raw materials may be combined together with the ratio thereof, or only the partial structure input data D of one raw material may be generated as the input data and used in the property prediction of the single-component substance.

In addition, a certain conversion rule provided in the vector conversion unit 13 of the input data generation system 10 may be other rules.

FIG. 5 illustrates the configuration of a property prediction system 1A according to a Modification Example. In this Modification Example, the functions of the vector conversion unit 13, the synthesis unit 14, and the predictor 30 are provided in a training unit 20A. Then, in the training unit 20A, a vectorization function, a data synthesizing function, and a reflection function using blending ratio data and number data of such functional units are attained by using a machine learning model of a neural network integrated with the predictor 30. In this case, the training unit 20A inputs the partial structure input data D0 for each of the partial structures generated by the generation unit 12 of an input data generation system 10A to the machine learning model.