TW201839640A - Model generation system and model generation method - Google Patents

Abstract

According to one embodiment, a model generation system includes a NCLM processor, a filter part, a model generator, a variable narrow-down part, a determiner, and a soundness calculator. The NCLM processor and the filter part narrows down a first input variable group to a third input variable group selected using NCLM. The model generator generates a model of a relationship between the third input variable group and an output variable. The variable narrow-down part narrows down the first input variable group to one or more of the input variables not used in the generation of the model. When the number of the models has not reached the specified number, the determiner outputting to the NCLM processor the first input variable group narrowed down by the variable narrow-down part. The soundness calculator calculates an overall soundness of the models and calculating a soundness of each of the models.

Description

Model construction system and model construction method

Embodiments of the present invention generally relate to a model construction system and a model construction method.

In order to predict an output variable (destination variable) from a plurality of input variables (description variables), a model representing the relationship between the input variable and the output variable is constructed. As an example, when the manufacturing apparatus performs the processing of the workpiece, there are cases in which various types of data obtained by the manufacturing apparatus are used as input variables, and a model for predicting the quality of the workpiece after processing is constructed. Such a model is used to detect that the workpiece is defective, for example, when the quality predicted from the data obtained in the processing of the workpiece deviates from the specific management range. In general, the model of the construction will decrease in sensitivity as the usage period is extended, and the error with respect to the predicted value of the measured value will become large. If the error with respect to the predicted value of the measured value becomes large, when the predicted value of the self-model output deviates from the management range, it is difficult to determine whether the system is caused by the sensitivity of the model or the actual workpiece defect. Therefore, it is desired to develop a technique that can alleviate the influence of the decrease in sensitivity and can judge whether the model is sound.

Embodiments of the present invention provide a model construction system and a model construction method that can alleviate the influence of a decrease in sensitivity and can determine whether a model is sound. According to the embodiment of the present invention, the model construction system includes an NCLM processing unit, a filter unit, a model construction unit, a variable restriction unit, a determination unit, and a soundness calculation unit. The NCLM processing unit limits the first input variable group including the plurality of input variables to the second input variable group selected using the latest related correlation law (NCLM). The filter unit limits the second input variable group to a third input variable group that satisfies a specific condition. The model construction unit constructs a model indicating the relationship between the third input variable group and the output variable. The variable limiting unit limits the first input variable group to the input variable that is not used in the construction of the model. The determination unit determines whether the number of the constructed models reaches a predetermined number, and when the number of the models does not reach the predetermined number, outputs the first input variable group defined by the variable limiting unit to the NCLM processing. unit. The soundness calculation unit calculates the overall soundness of the model of the predetermined number and the soundness of each of the models. According to the model construction system configured as described above, it is possible to alleviate the influence of the decrease in sensitivity and to judge whether or not the model is sound.

Hereinafter, each embodiment of the present invention will be described with reference to the drawings. In the present specification and the drawings, the same components as those described above are denoted by the same reference numerals, and the detailed description is omitted as appropriate. Fig. 1 is a block diagram showing the configuration of a model construction system 1 according to the first embodiment. As shown in FIG. 1 , the model construction system 1 includes an acquisition unit 100, an NCLM processing unit 102, a filter unit 104, a model construction unit 106, a model information storage unit 108, a variable restriction unit 110, a determination unit 112, and a soundness calculation unit 114. The external output unit 116, the predetermined number database 120, and the variable database 122. The prescribed number database 120 memorizes the prescribed number. The predetermined number indicates the number of models constructed in the model construction system 1. The prescribed number is input, for example, in advance by the user. The variable database 122 stores the measured values of the variables of the input variables and the output variables, that is, the variable data. The acquisition unit 100 acquires the predetermined number and the variable data from the predetermined number database 120 and the variable database 122, respectively. The acquisition unit 100 outputs the acquired information to the NCLM processing unit 102. The NCLM processing unit limits the plurality of input variables obtained by the acquisition unit 100 to a plurality of input variables selected using the Nearest Correlation Louvain Method (NCLM). NCLM is a combination of the NC (Nearest Correlation) method and the Louvain Method. By using the NC method, variables with higher correlations and more similar (strong correlations) can be found from among multiple input variables. Also, the Louvain Method is one of the best methods for dividing a weighted graph into a plurality of groups. In the Louvain Method, the weighted graph is segmented into complex arrays in such a way that the binding within the group is tight and the combination between the groups is loose. Similar variables found by the NC method are considered as a combination. Thereby, group division can be performed in such a manner that similar variables are assigned to the same group, and related lower variables are assigned to different groups. For the distribution result, use Partial Least Square (PLS) and select only the group that can well describe the output variables. Moreover, by combining these methods, a plurality of input variable arrays can be grouped in consideration of the similarity between the variables, and the input variables can be selected in groups. In other words, the NCLM processing unit 102 limits the plurality of input variables to one or more input variables that best describe the output variables. Furthermore, regarding NCLM, NC method, and Louvain Method, Uchimaru, T., Hazama, K., Fujiwara, K., and Kano, M Nearest Correlation Louvain Method for Fast and Good Selection of Input Variables of Statistical Model. 9th Detailed in International Symposium on Advanced Control of Chemical Processes. Received November 15, 2014. The filter unit 104 filters a plurality of input variables input from the NCLM processing unit 102. Thereby, it is limited to one of the plurality of input variables satisfying the predetermined condition set in advance. The filter unit 104 outputs the above-described part of the plurality of input variables to the model construction unit 106. Hereinafter, for the sake of brevity, a group that is input to the NCLM processing unit 102 and that uses a plurality of input variables defined by the NCLM may be referred to as a “first input variable group”. Further, a group of a plurality of input variables defined by the NCLM may be referred to as a "second input variable group". The group of input variables defined by the filter unit 104 from the second input variable group is referred to as a "third input variable group". The model construction unit 106 constructs a model indicating the relationship between the third input variable group and the output variable input from the filter unit 104. The model construction unit 106 constructs a model using, for example, a complex regression or a partial least square (PLA). The model construction unit 106 stores the constructed model information in the model information storage unit 108. The variable limiting unit 110 limits the first input variable group to one or more input variables that are not used in the construction of the model. That is, the input variable that has been used in the construction of the model is removed from the first input variable group. The variable limiting unit 110 outputs the limited first input variable group to the determination unit 112. The determination unit 112 determines whether or not the number of model information stored in the model storage unit 108 (the number of models constructed by the model construction unit 106) has reached a predetermined number. When the determination unit 112 determines that the number of model information has not reached the predetermined number, the first input variable group defined by the variable limiting unit 110 is input to the NCLM processing unit 102. The NCLM processing unit 102 and the filter unit 104 perform the restriction on the second input variable group and the third input variable group again. The other model is constructed by the model construction unit 106. At this time, the third input variable group input to the model construction unit 106 does not include the input variables that have been used in the construction of the model. Therefore, the model construction unit 106 constructs another model indicating the relationship between at least a part of the plurality of input variables that have not been used in the construction of the model and the output variables. The processing by the NCLM processing unit 102, the filter unit 104, the model construction unit 106, and the variable defining unit 110 is repeated until the number of models constructed reaches a predetermined number. When the number of the constructed models reaches a predetermined number, the soundness calculation unit 114 acquires the model information from the model information storage unit 108, and acquires the variable data from the variable database 122. The soundness calculation unit 114 calculates the overall soundness of the constructed model group and the soundness of each model based on the acquired data. Specifically, first, the soundness calculation unit 114 inputs the variable data to the input variables of the constructed models to obtain an output variable (predicted value). For example, the case where the output variable is within a specific range is good, and the case outside the range is a bad judgment. In this case, the soundness calculation unit 114 performs this determination on the output result of each model. Moreover, the number of judgments and the number of judgments of badness are separately counted. The soundness calculation unit 114 uses the judgment results of the more parties as the comprehensive determination result. Alternatively, the soundness calculation unit 114 calculates a median value, a weighted average value, and the like of the predicted values of the respective models. The soundness calculation unit 114 uses the calculated value as a representative value of the model group. The soundness calculation unit 114 takes the case where the representative value is within the specific range as a good condition, and the situation outside the range as a defect results in a comprehensive determination result. Further, the soundness calculation unit 114 compares the determination result of the actual output variable (actual measurement value) with the comprehensive determination result. At this time, if the comprehensive determination result matches the actual determination result, the constructed model group is judged to be comprehensively sound. Further, the soundness calculation unit 114 can directly compare the actual measurement value with the representative value of the model group. In this case, a Mean Square Error (MSE), a Root Mean Squared Error (RMSE), a coefficient of determination (R ² ), or a correlation coefficient may be used as an indicator of soundness. Next, the soundness calculation unit 114 compares the actual measurement value with the predicted value of each model, and compares the determination result of the actual measurement value with the determination result of each model. Thereby, the soundness calculation unit 114 calculates the soundness of each model. In the case of comparing the determination result of the actual measurement value with the determination result of the predicted value, a discrete value such as "sound" and "abnormal" is obtained as the soundness. Comparing the measured value with the predicted value, the indicators such as MSE, RMSE, R2 or correlation coefficient are obtained as the soundness. Alternatively, the representative value of the model group or the comprehensive determination result may be set to be positive. By comparing the predicted values or the determination results of the respective models, the soundness of each model can be calculated. Further, the methods for calculating the soundness are appropriately combined according to the application purpose of the model construction system 1, and the soundness of the model group and each model can be managed. For example, when the target frequency of the model group or the determination result thereof is positive, the soundness of each model is temporarily calculated. Moreover, the comprehensive soundness of the model group and the soundness of each model can be calculated by comparing the predicted value with the measured value or the determination result thereof at a small number of actual measurement points. The external output unit 116 displays the information on the display to the user, or outputs it in a specific document form and outputs it to the outside. This information is calculated by the sound calculation unit 114, and includes the predicted values of the models, the soundness of each model, and the soundness of the integrated model. Fig. 2 is a view for explaining the operation of the model construction system 1 of the first embodiment. The plurality of input variables (first input variable group) are limited to a plurality of input variables (second input variable groups) selected using NCLM. Filter the defined number of input variables. The first model is constructed using a plurality of input variables (the third input variable group) further defined by the NCLM and the filter. When constructing the first model, the plurality of input variables used for constructing the second and subsequent models are limited to those not used in the first model. The plurality of input variables for the limitation are further limited by the NCLM and the filter when constructing the second model. Construct a second model using a defined number of input variables. Thereafter, the same action is repeated. That is, when constructing the n-1th model, the plurality of input variables used for the nth and subsequent model construction are limited to those not used for the first, second, ..., n-1 models. The nth model is constructed using at least a portion of the plurality of defined input variables. Fig. 3 is a flow chart showing a model construction method according to the first embodiment. The acquisition unit 100 acquires the predetermined number and the variable data from the predetermined number database 120 and the variable database 122 (step S1). The NCLM processing unit 102 limits the plurality of acquired input variables to a plurality of input variables selected using the NCLM (step S2). The filter unit 104 filters a plurality of input variables defined by the NCLM, and is limited to a plurality of input variables that are set in advance to satisfy a specific condition (step S3). The model construction unit 106 constructs a model indicating the relationship between the plurality of input variables and the output variables defined by the NCLM and the filter (step S4). The model construction unit 106 stores the constructed model information in the model information storage unit 108 (step S5). The variable limiting unit 110 limits the plurality of input variables obtained in step S1 to the input variables that are not used (not used for model construction) in step S4 (step S6). The determination unit 112 determines whether or not the number of models constructed has reached a predetermined number (step S7). When the number of models does not reach the predetermined number, steps S2 to S6 are executed again based on the input variables defined in step S6. When the number of models reaches a predetermined number, the soundness calculation unit 114 acquires the model information and the variable data from the model information storage unit 108 and the variable database 122 (step S8). The soundness calculation unit 114 calculates the predicted value of each model using the acquired variable data and model information (step S9). The soundness calculation unit 114 calculates the predicted value of each model and calculates the comprehensive soundness (step S10). The soundness calculation unit 114 calculates the soundness of each model (step S11). The external output unit 116 outputs the predicted value of each model, the soundness of each model, and the soundness of the integrated model to the outside (step S12). Here, the effects of the present embodiment will be described. In order to alleviate the influence of the sensitivity reduction of the model, it is also considered to use a plurality of models to comprehensively determine whether the model is sound from a plurality of prediction values. However, when the behaviors of a plurality of models are similar, it is possible to similarly decrease the sensitivity in a plurality of models, or similarly temporarily generate a large error. Therefore, it is difficult to determine the soundness. In the model construction system 1 of the present embodiment, the NCLM processing unit 102 defines a plurality of input variables. The plurality of input variables included in the second input variable group defined by the NCLM processing unit 102 have a strong correlation with each other. In other words, the correlation between the input variable group defined by the NCLM processing unit 102 and the undefined input variable group becomes weak. Therefore, the behavior of models constructed using these input variable groups tends to be different from each other. Further, in the model construction system 1, the second input variable group is filtered by the filter unit 104. Therefore, the accuracy of the model relative to the output variable can also be improved. In other words, according to the model construction system 1, the NCLM processing unit 102, the filter unit 104, the model construction unit 106, and the variable defining unit 110 can improve the accuracy of the output variables for each model and construct a plurality of models having different behaviors. . Further, the constructed model calculates the overall soundness of the model of the soundness and the predetermined number of each model by the soundness calculation unit 114. The comprehensive soundness of these and the soundness of each model can be confirmed. Thereby, even when the output variable of a part of the model deviates from the management range, it is possible to more accurately determine whether the system is caused by the decrease in the sensitivity of the model or whether the change actually occurs. Further, in the model construction system 1 of the present embodiment, the first input variable group defined by the variable limiting unit 110 is again input to the NCLM processing unit 102. For example, the condition of the NCLM executed in the NCLM processing unit 102 can be adjusted in accordance with the increase in the number of models constructed. By this method, it is possible to further improve the accuracy of the output variables of each model while constructing a plurality of models having different behaviors. (Second Embodiment) Fig. 4 is a block diagram showing the configuration of a model construction system 2 according to a second embodiment. In the model construction system 1 of the first embodiment, the variable limiting unit 110 limits the first input variable group to an input variable that is not used in the construction of the model. Then, the limited first input variable group is input to the NCLM processing unit 102. In the model construction system 2 of the second embodiment shown in FIG. 4, the variable limiting unit 110 limits the second input variable group to the input variables that are not used in the construction of the model. Then, the limited second input variable group is input to the filter unit 104. Therefore, in the model construction system 2, the first input variable group other than the second input variable group that is not limited by the NCLM processing unit 102 is not used for construction of the model. Fig. 5 is a view for explaining the operation of the model construction system 2 of the second embodiment. As shown in FIG. 5, the first model is constructed using the input variables defined by the NCLM and the filter, similarly to the model construction system 1 of the first embodiment. The following model is defined by NCLM, but is constructed using input variables that are not defined by filters. Fig. 6 is a flow chart showing a model construction method in the second embodiment. Steps S21 to S25 are executed in the same manner as steps S1 to S5 of the flowchart shown in Fig. 3, respectively. The variable defining unit 110 limits the variable defined in step S22 to the variable not used in step S24 (step S26). The determination unit 112 determines whether or not the number of models constructed has reached a predetermined number (step S27). When the number of models does not reach the predetermined number, steps S23 to S26 are executed again based on the input variables defined in step S26. After the number of models reaches a predetermined number, steps S28 to S32 are executed in the same manner as steps S8 to S12 of the flowchart shown in FIG. 3, respectively. As described in the description of the first embodiment, the NCLM processing unit 102 limits the first input variable group to the second input variable group including a plurality of input variables, which have a strong correlation and a high accuracy for output variables. Further, in the model construction system 2 of the present embodiment, the construction of the model and the limitation of the input variables are repeatedly performed based on the second input variable group. Therefore, according to the model construction system of the present embodiment, the behavior of the model constructed is likely to be relatively similar, but the accuracy of each model can be improved. (Third Embodiment) Fig. 7 is a block diagram showing the configuration of a model construction system 3 according to a third embodiment. In the model construction system 3 of the third embodiment shown in FIG. 7, the variable limiting unit 110 limits the first input variable group to the input variables that are not used in the construction of the model. Further, the limited first input variable group is input to the filter unit 104 as the second input variable group. That is, in the model construction system 1, when each model is constructed, the input variable by the NCLM processing unit 102 is limited each time. In the model construction system 3, the limitation of the input variables by the NCLM processing unit 102 is performed only at the time of the initial model construction. Fig. 8 is a view for explaining the operation of the model construction system 3 of the third embodiment. As shown in Fig. 8, the first model is constructed using the input variables defined by the NCLM and the filter, similarly to the model construction system 1 of the first embodiment. The second model is constructed with a filter qualifier using input variables that are not defined by the NCLM and the filter. The following model uses a filter qualifier to construct input variables that are not defined by the filter. Fig. 9 is a flow chart showing a model construction method according to a third embodiment. Steps S41 to S45 are executed in the same manner as steps S1 to S5 of the flowchart shown in Fig. 3, respectively. The variable limiting unit 110 limits the plurality of input variables obtained in step S41 to the variables that are not used in step S44 (step S46). The determination unit 112 determines whether or not the number of models constructed has reached a predetermined number (step S47). When the number of models does not reach the predetermined number, steps S43 to S46 are executed again based on the input variables defined in step S46. After the number of models reaches a predetermined number, steps S48 to S52 are executed in the same manner as steps S8 to S12 of the flowchart shown in FIG. 3, respectively. As described in the description of the first embodiment, the correlation between the input variable group defined by the NCLM processing unit 102 and the undefined output variable group is weak. Therefore, the input variable group not limited by the NCLM processing unit 102 is not limited by the NCLM processing unit 102, and a model of behavior different from the first model can be constructed. Further, in the construction of the second or lower model, since the input variable group is limited by the filter unit 104, the accuracy with respect to the output variable can be improved. Therefore, according to the model construction system 3 of the present embodiment, it is possible to construct a plurality of models having different behaviors as compared with the first embodiment while improving the accuracy of the output variables of the respective models. (Fourth Embodiment) Fig. 10 is a block diagram showing the configuration of a model construction system 4 according to a fourth embodiment. In the model construction system 4 of the fourth embodiment, for example, the model construction system 4 is different from the model construction systems of the first to third embodiments. In the model construction system 4, the NCLM processing unit 102 generates a plurality of sets of input variables. In other words, the NCLM processing unit 102 generates a plurality of second input variable groups from the first input variable group. When the number of the second input variable groups generated is larger than the predetermined number, for example, the NCLM processing unit 102 limits the plurality of second input variable groups to the second input variable group that can better explain the predetermined number of output variables. . The filter unit 104 limits each of the plurality of second input variable groups to a plurality of third input variable groups that satisfy a specific condition. The model construction unit 106 constructs a model indicating the relationship between the third input variable group and the output variable for each of the third input variable groups. The soundness calculation unit 114 calculates the overall soundness of the plurality of models and the soundness of each model. Fig. 11 is a view for explaining the operation of the model construction system 4 of the fourth embodiment. As shown in FIG. 11, a plurality of sets of input variables (a plurality of second input variable groups) are generated using the NCLM. Next, each of the second input variable groups is filtered by a filter. Further, a model is constructed for each of the second input variable groups. Fig. 12 is a flow chart showing a model construction method in the fourth embodiment. The acquisition unit 100 acquires the predetermined number and the variable data from the predetermined number database 120 and the variable database 122 (step S61). The NCLM processing unit 102 generates a plurality of sets of input variables (a plurality of second input variable groups) selected by the NCLM from the plurality of input variables obtained (step S62). The filter unit limits each of the plurality of second input variable groups to a third input variable group that satisfies a specific condition (step S63). The model construction unit 106 constructs a plurality of models (step S64). A plurality of models respectively represent the relationship between a plurality of third input variable groups and output variables. The model construction unit 106 stores a plurality of models in the model information storage unit 108 (step S65). The following steps S66 to S70 are executed in the same manner as steps S8 to S12 of the flowchart shown in Fig. 3, respectively. (First Embodiment) Fig. 13 is a view showing an output variable and an input variable of the first embodiment. Fig. 14 is a view showing an output variable of the first embodiment and an input variable selected by the NCLM. Fig. 15 is a view showing an output variable of the first embodiment and an input variable selected by a filter. Fig. 16 is a graph showing the measured values of the first embodiment and the predicted values of the respective models. In the first embodiment, the model building system 2 of the second embodiment is used. Here, as shown in FIG. 13, an example in which a plurality of models are constructed based on the variable data of the output variable Y and the 34 input variables X will be described. The output variable Y is the quality characteristic of the workpiece. The input variable X is the quality of the workpiece after machining in each step. The quality is based on at least one of the size of the processed workpiece and the processing rate of the workpiece. First, 34 input variables (first input variable group) shown in FIG. 13 are input to the NCLM processing unit 102. As a result, the NCLM processing unit 102 is limited to 15 input variables (second input variable group) as shown in FIG. 14 . Next, the 15 input variables are input to the filter unit 104. In the example shown in FIG. 15, the filter unit 104 calculates an estimated value, a standard error, a t value, a p value, and a VIF (Variance Inflation Factor) for each input variable. Further, the filter unit 104 limits, for example, the input explanatory variable to an input variable (the third input variable group) that satisfies the P value of <0.0001. In the example shown in FIG. 15, the filter unit 104 is limited to four input variables. The first model that shows the relationship between the four input variables and the output variable is constructed by the model construction unit 106. Next, the variable limiting unit 110 limits the 15 input variables (the second input variable group) defined by the NCLM processing unit 102 to the 11 input variables that are not used in the first model. Here, simply, the model construction unit 106 constructs the second model indicating the relationship with the output variable using the eleven input variables (the second input variable that is limited). Fig. 16 is a graph showing the measured values of the output variable Y and the predicted values of the first model and the second model. As can be seen from Fig. 16, the first model can predict the measured value with higher accuracy than the second model. On the other hand, although the accuracy of the second model is inferior to the first model, it does not significantly deviate from the measured value. Furthermore, it is known that the first model and the second model exhibit different behaviors with respect to changes in the measured values. That is, according to the model construction system according to the embodiment of the present invention, it is possible to construct a plurality of models having different behaviors while improving the accuracy of the output variables. Further, by using the predicted values obtained by the respective models, the soundness calculation unit 114 calculates the overall soundness of the model group and the soundness of each model, and can more accurately determine whether or not the constructed model is sound. (Second Embodiment) Fig. 17 is a view showing an output variable and an input variable of the second embodiment. Fig. 18 is a view showing an output variable of the second embodiment and an input variable selected by the NCLM. Fig. 19 is a table showing the characteristics of each model of the second embodiment. Fig. 20 is a graph showing the measured values of the second embodiment and the predicted values of the respective models. In the second embodiment, as shown in FIG. 17, an example in which a plurality of models are constructed based on the variable data of the output variable Y and the 270 input variables X will be described. The output variable Y is the quality of the workpiece. The input variable X is the data of the sensor obtained in each step (temperature or pressure during processing, etc.). The quality is based on at least one of the size of the processed workpiece and the processing rate of the workpiece. In the present embodiment, the model construction system 4 of the fourth embodiment is used for the construction of the model. First, 270 input variables (first input variable group) shown in FIG. 17 are input to the NCLM processing unit 102. As a result, a set of three input variables (three second input variable groups) shown in FIG. 18 is generated. One second input variable group G1 includes 132 input variables. The other second input variable group G2 contains 62 input variables. Yet another input variable group G3 contains nine input variables. Each of the second input variable groups G1 and G2 is filtered by a filter and is limited to the third input variable group. Since the number of variables in the second input variable group G3 is sufficiently small, it is not filtered by the filter. In the filter unit 104, the input variable is defined using a stepwise method. Thereby, the second input variable group G1 is limited to the 22 third input variable groups G4. The second input variable group G2 is limited to the 13 third input variable groups G5. The first model, the second model, and the third model are constructed by the model construction unit 106 using the third input variable group G4, the third input variable group G5, and the second input variable group G3. The characteristics of each of the first model to the third model were evaluated. Fig. 19 shows the characteristics of each of the first model and the second model. The third model has been abandoned due to its low prediction accuracy. As shown in Fig. 19, with respect to the first model, R ² was about 0.64, and the accuracy of the first model was shown to be good. On the second model, R ² is about 0.42, although slightly lower, but the accuracy is tolerable. Fig. 20 is a view showing the measured values of the output variable Y and the predicted values of the first model and the second model. In the graph of Fig. 20, the horizontal axis represents time T. In the graph of Fig. 20, the average value A of the measured values is shown. Further, an example of the lower limit value L1 and the upper limit value L2 is shown. The lower limit value L1 and the upper limit value L2 respectively indicate the lower limit and the upper limit of the allowable output variable in manufacturing. As can be seen from Fig. 20, the first model can predict the measured value more accurately than the second model. Further, although the accuracy of the second model is inferior to that of the first model, it does not largely deviate from the measured value. The first model and the second model exhibit different behaviors with respect to changes in measured values. Further, at time T126, the measured value is between the lower limit value L1 and the upper limit value L2, and the predicted value of the first model is lower than the lower limit value L1. Therefore, in the prediction using only the first model, at time T126, it may be erroneously determined that the predicted output is in an unacceptable range. On the other hand, the predicted value of the second model is between the lower limit value L1 and the upper limit value L2. Therefore, with respect to the predicted value at time T126, by constructing more models, the comprehensive soundness of the model group can be calculated, and the determination can be accurately performed. Fig. 21 is a block diagram showing the configuration of a model construction device 5 for realizing the model construction system of each embodiment. The model construction device 5 includes, for example, an input device 200, an output device 202, and a computer 204. The computer 204 includes, for example, a ROM (Read Only Memory) 206, a RAM (Random Access Memory) 208, a CPU (Central Processing Unit) 210, and a memory device HDD (Hard Disc Drive). : Hard disk drive) 212. The input device 200 is a user who inputs information to the model building device 5. The input device 200 is a keyboard, a touch panel, or the like. The output device 202 is for outputting the output obtained by the model building system 1 to the user. The output device 202 is a display or a printer or the like. The ROM 206 stores a program for controlling the operation of the model construction device 5. In the ROM 206, the computer 204 is stored as the acquisition unit 100, the NCLM processing unit 102, the filter unit 104, the basic model construction unit 106, the similarity calculation unit 108, the variable restriction unit 110, and the determination unit 112 shown in FIG. The soundness calculation unit 114 and the external output unit 116 function as a program. The RAM 208 functions as a memory area in which the program stored in the ROM 206 is developed. The CPU 210 reads the control program stored in the ROM 103, and controls the operation of the computer 204 based on the control program. Further, the CPU 210 expands various materials obtained by the operation of the computer 204 in the RAM 208. The HDD 212 is stored in the predetermined number database 120 and the variable data library 122 shown in FIG. Further, the HDD 212 also functions as a model information storage unit and a similarity information storage unit 108 that store a model of construction or a similar degree of similarity. The embodiments of the present invention have been described above by way of example only, and are not intended to limit the scope of the present invention. The present invention can be implemented in various other forms, and various omissions, substitutions and changes can be made without departing from the spirit of the invention. The scope of the invention and the scope of the invention are included in the scope of the invention and the scope of the invention as set forth in the appended claims. Further, each of the above embodiments can be implemented in combination with each other. This application is based on Japanese Patent Application No. 2017-064333 (application date: March 29, 2017) and Japanese Patent Application No. 2017-249763 (application date: December 26, 2017), from which the applications are enjoyed. Priority interest. The entire contents of the applications are hereby incorporated by reference in their entirety.

1~4‧‧‧Model Construction System

5‧‧‧Model building device

100‧‧‧Acquisition Department

102‧‧‧NCLM Processing Department

104‧‧‧Filter Department

106‧‧‧Model Construction Department

108‧‧‧Model Information Saving Department

110‧‧‧Variable Limits

112‧‧‧Decision Department

114‧‧‧Surability calculation department

116‧‧‧External output

120‧‧‧Specified database

122‧‧‧Variable database

200‧‧‧ input device

202‧‧‧Output device

204‧‧‧ computer

206‧‧‧ROM

208‧‧‧RAM

210‧‧‧CPU

212‧‧‧HDD

S1～S12‧‧‧Steps

S21～S32‧‧‧Steps

S41～S52‧‧‧Steps

S61～S70‧‧‧Steps

Fig. 1 is a block diagram showing the configuration of a model construction system according to the first embodiment. Fig. 2 is a view showing the operation of the model construction system of the first embodiment. Fig. 3 is a flow chart showing a model construction method according to the first embodiment. Fig. 4 is a block diagram showing the configuration of a model construction system according to a second embodiment. Fig. 5 is a view for explaining the operation of the model building system of the second embodiment. Fig. 6 is a flow chart showing a model construction method in the second embodiment. Fig. 7 is a block diagram showing the configuration of a model construction system according to a third embodiment. Fig. 8 is a view for explaining the operation of the model construction system of the third embodiment. Fig. 9 is a flow chart showing a model construction method according to a third embodiment. Fig. 10 is a block diagram showing the configuration of the model construction system 4 of the fourth embodiment. Fig. 11 is a view for explaining the operation of the model construction system 4 of the fourth embodiment. Fig. 12 is a flow chart showing a model construction method in the fourth embodiment. Fig. 13 is a view showing an output variable and an input variable of the first embodiment. Fig. 14 is a view showing an output variable of the first embodiment and an input variable selected by the NCLM. Fig. 15 is a view showing an output variable of the first embodiment and an input variable selected by a filter. Fig. 16 is a graph showing the measured values of the first embodiment and the predicted values of the respective models. Fig. 17 is a view showing an output variable and an input variable of the second embodiment. Fig. 18 is a view showing an output variable of the second embodiment and an input variable selected by the NCLM. Fig. 19 is a table showing the characteristics of each model of the second embodiment. Fig. 20 is a graph showing the measured values of the second embodiment and the predicted values of the respective models. Fig. 21 is a block diagram showing the configuration of a model building device for realizing the model building system of each embodiment.

Claims (9)

A model construction system including: an NCLM processing unit that limits a first input variable group including a plurality of input variables to a second input variable group selected using a latest related correlation law (NCLM); a filter unit that limits the second input variable group to a third input variable group that satisfies a specific condition; the model construction unit constructs a model that indicates a relationship between the third input variable group and an output variable; and a variable limiting unit that The first input variable group is limited to one or more of the input variables that are not used in the construction of the model; and the determining unit determines whether the number of the constructed models reaches a predetermined number, and the number of the models does not meet the above requirement. In the case of a number, the first input variable group defined by the variable limiting unit is output to the NCLM processing unit, and the soundness calculating unit calculates a comprehensive soundness of the model of the predetermined number and each of the models. The soundness. The model construction system of claim 1, further comprising an external output unit that outputs the soundness of each of the models, the output variable of each of the models, and the integrated soundness to the outside. A model construction system including: an NCLM processing unit that limits a first input variable group including a plurality of input variables to a second input variable group selected using a latest related correlation law (NCLM); a filter unit that limits the second input variable group to a third input variable group that satisfies a specific condition; the model construction unit constructs a model that indicates a relationship between the third input variable group and an output variable; and a variable limiting unit that The second input variable group is limited to one or more input variables that are not used in the construction of the model; and the determining unit determines whether the number of the constructed models reaches a predetermined number, and the number of the models does not reach the above requirement In the case of a number, the second input variable group defined by the variable limiting unit is output to the filter unit; and the soundness calculation unit calculates a comprehensive soundness of the model of the predetermined number and each of the models Soundness. The model construction system of claim 3, further comprising an external output unit that outputs the soundness of each of the models, the output variable of each of the models, and the integrated soundness to the outside. A model construction system including: an NCLM processing unit that limits a first input variable group including a plurality of input variables to a second input variable group selected using a latest related correlation law (NCLM); a filter unit that limits the second input variable group to a third input variable group that satisfies a specific condition; the model construction unit constructs a model that indicates a relationship between the third input variable group and an output variable; and a variable limiting unit that The first input variable group is limited to one or more input variables that are not used in the construction of the model; and the determining unit determines whether the number of the constructed models reaches a predetermined number, and the number of the models does not reach the predetermined number In the case of the first input variable group defined by the variable limiting unit, the second input variable group is output to the filter unit, and the soundness calculation unit calculates a comprehensive sound of the model of the predetermined number. Degree and the soundness of each of the above models. The model construction system of claim 5, further comprising an external output unit that outputs the soundness of each of the models, the output variable of each of the models, and the integrated soundness to the outside. A model construction system includes: an NCLM processing unit that generates a plurality of second input variable groups selected using a latest related Leuven method (NCLM) from a first input variable group including a plurality of input variables And a filter unit that limits the plurality of second input variable groups to a plurality of third input variable groups that satisfy a specific condition; and the model construction unit constructs each of the plurality of third input variable groups and an output variable A model of the relationship; and a soundness calculation unit that calculates the overall soundness of the plurality of models and the soundness of each of the models. The model construction system of claim 7, further comprising an external output unit that outputs the soundness of each of the models, the output variable of each of the models, and the integrated soundness to the outside. A model construction method comprising: a first step of limiting a first input variable group including a plurality of input variables to a second input variable group selected using a Nearest Correlation Louvain Method (NCLM); In the second step, the second input variable group is limited to a third input variable group that satisfies a specific condition; and the third step is a model that represents a relationship between the third input variable group and an output variable; and a fourth step Limiting the first input variable group to the input variable not used in the construction of the model in the third step; and repeating the first step to the first step based on the first input variable group defined in the fourth step In the fourth step, the number of the models to be constructed reaches a predetermined number, and further includes: a fifth step of calculating the overall soundness of the model of the predetermined number and the soundness of each of the models.