Property prediction device, property prediction program, and analysis system

Abstract

This invention provides a fuel oil properties prediction device, a properties prediction program, and an analysis system that can improve the accuracy of predicting the properties of fuel oil. [Solution] The property prediction device (14) includes a data acquisition unit 42 that acquires information about the components contained in the fuel oil (12) and measurement data (58, 62) which is a collection of measured values ​​corresponding to the information about the components; a feature generation unit 46 that generates a feature set which is a collection of features related to the measured values ​​from the acquired measurement data (58, 62); and a property prediction unit 50 that, upon input of the generated feature set, predicts property values ​​using a prediction model (PM) which outputs property values ​​indicating the properties of the fuel oil (12).

Description

Technical Field

[0001] The present disclosure relates to a property prediction apparatus, a property prediction program, and an analysis system. Background Art

[0002] Conventionally, various analysis techniques for analyzing properties of feedstock oil or fractions thereof have been known. For example, a method of predicting properties of feedstock oil or a fraction thereof by performing analysis processing on measurement data obtained through an analyzer has been proposed.

[0003] Patent Document 1 discloses a method for predicting properties of crude oil or a fraction thereof using gas chromatography and a mass spectrometer. Prior Art Documents Patent Documents

[0004] [Patent Document 1] National Publication of International Patent Application No. 11-508363 Summary of Invention Problem to be Solved by Invention

[0005] By the way, when feedstock oil is fractionally distilled (fractionated) according to a plurality of distillation temperature ranges, the obtained fraction may be used as a product or a base material. For example, regarding fuel oil containing gasoline, the ranges of distillation characteristics and vapor pressure are defined in JIS, which is the Japanese national standard, and are important indicators in terms of quality. Therefore, improvement in prediction accuracy for properties of fuel oil has been continuously desired.

[0006] The present disclosure has been made in view of the above circumstances, and an object thereof is to provide a property prediction apparatus, a property prediction program, and an analysis system capable of improving prediction accuracy of properties of fuel oil. Means for Solving Problem

[0007] To solve the above problems, a property prediction device in one aspect of the present disclosure includes: a data acquisition unit that acquires measurement data which is a collection of information about components contained in fuel oil and measured values ​​corresponding to the information about components; a feature generation unit that generates a feature set which is a collection of features related to the measured values ​​from the measurement data acquired by the data acquisition unit; and a property prediction unit that, upon input of the feature set generated by the feature generation unit, predicts the property values ​​using a prediction model that outputs property values ​​indicating the properties of the fuel oil.

[0008] A property prediction program in another aspect of the present disclosure causes one or more computers to perform the following: an acquisition process to acquire measurement data which is a collection of measurement values ​​corresponding to information about components contained in fuel oil and information about the components; a generation process to generate a feature set which is a collection of feature values ​​related to the measurement values ​​from the measurement data acquired through the acquisition process; and a prediction process to predict the property values ​​using a prediction model that outputs property values ​​indicating the properties of the fuel oil when the feature set generated through the generation process is input.

[0009] An analysis system in another aspect of the present disclosure comprises an analysis device that outputs information about components contained in fuel oil and measurement data which is a collection of measured values ​​corresponding to the information about components, and a property prediction device that performs analytical processing on the measurement data output from the analysis device and predicts the properties of the fuel oil, wherein the property prediction device performs a generation process that generates a feature set which is a collection of features related to the measured values ​​from the measurement data, and a prediction process that, upon input of the feature set generated through the generation process, predicts the property value using a prediction model that outputs a property value indicating the properties. [Effects of the Invention]

[0010] According to this disclosure, the accuracy of predicting the properties of fuel oil can be improved. [Brief explanation of the drawing]

[0011] [Figure 1] This is an overall configuration diagram of the analysis system in one embodiment of the present disclosure. [Figure 2] Figure 1 shows an example of the hardware configuration of the property prediction device. [Figure 3] This block diagram shows an example of the configuration of the property prediction device in Figure 1. [Figure 4] This figure shows an example of a combination of properties that are the target of prediction. [Figure 5] This figure shows an example of the model structure of a predictive model used to predict the properties of fuel oil. [Figure 6] Figures 1 to 3 are flowcharts illustrating an example of the learning process performed by the property prediction device. [Figure 7] This figure shows an example of the data structure of the measurement data in Figure 1. [Figure 8] This diagram schematically illustrates one example of a method for formatting a training dataset. [Figure 9] Figure 5 shows an example of a method for generating a predictive model. [Figure 10] Figure 3 is a functional block diagram relating to the predictive operation by the control unit. [Figure 11] Figures 1 to 3 are flowcharts illustrating an example of the predictive operation by the property prediction device. [Figure 12] This figure shows an example of the analysis results screen displayed on the display unit in Figure 3. [Modes for carrying out the invention]

[0012] First, we will describe some aspects of this disclosure.

[0013] The property prediction apparatus according to the first aspect of the present disclosure comprises: a data acquisition unit that acquires measurement data which is an aggregate of information on components contained in fuel oil and measurement values corresponding to the information on the components; a feature quantity generation unit that generates a feature quantity set which is an aggregate of feature quantities related to the measurement values from the measurement data acquired by the data acquisition unit; and a property prediction unit that predicts a property value indicating the property of the fuel oil using a prediction model that outputs the property value indicating the property of the fuel oil when the feature quantity set generated by the feature quantity generation unit is input.

[0014] In the property prediction apparatus according to the second aspect of the present disclosure, there are a plurality of the property values of the fuel oil predicted by the property prediction unit, and the prediction model may be provided for each of the property values predicted by the property prediction unit.

[0015] In the property prediction apparatus according to the third aspect of the present disclosure, the property includes at least one of a distillation temperature, a distillation amount, and a vapor pressure, the measurement data is data measured by a gas chromatography device, the information on the components is information based on retention time, and the measurement value may be a value based on signal intensity.

[0016] In the property prediction apparatus according to the fourth aspect of the present disclosure, the property may include at least one of a distillation temperature, a distillation amount, and a vapor pressure.

[0017] In the property prediction apparatus according to the fifth aspect of the present disclosure, the fuel oil is gasoline, and the distillation temperature may include at least one of a 10% distillation temperature, a 50% distillation temperature, and a 90% distillation temperature.

[0018] In the property prediction apparatus according to the sixth aspect of the present disclosure, the fuel oil is gasoline, there are a plurality of the distillation temperatures predicted by the property prediction unit, and the number of the prediction models that output the distillation temperatures may be equal to the number of the distillation temperatures predicted by the property prediction unit.

[0019] In the property prediction device according to the seventh aspect of this disclosure, the feature generation unit may generate the feature set by compressing the number of dimensions of the measurement data.

[0020] In the property prediction device according to the eighth aspect of this disclosure, the prediction model may be a mathematical model in which a learning process is performed on a learner having common input / output characteristics using different learning datasets for each property.

[0021] In the property prediction device according to the ninth aspect of this disclosure, the feature generation unit may generate a set of features common to two or more properties.

[0022] The property prediction device in the tenth aspect of this disclosure may further include a data processing unit that generates integrated data by associating the set of measurement data with the set of values ​​relating to the properties, and generates the learning dataset by dividing the integrated data for each property.

[0023] The property prediction device in the eleventh aspect of this disclosure may further include a learning processing unit that generates a prediction model for each property by applying a learning process to a learner having common input / output characteristics using the learning dataset generated for each property by the data processing unit.

[0024] In the property prediction device according to the twelfth aspect of this disclosure, the information relating to the component is identification information of the component, and the data processing unit may perform a name matching process to unify the notation of the identification information before generating the integrated data.

[0025] The property prediction device in the 13th aspect of this disclosure may further include an output processing unit that instructs an output means to output a property analysis table including the property values ​​predicted by the property prediction unit.

[0026] A property prediction program in a 14th aspect of this disclosure causes one or more computers to execute: an acquisition process that acquires measurement data which is a collection of measured values ​​corresponding to information about components contained in fuel oil; a generation process that generates a feature set which is a collection of features related to the measured values ​​from the measurement data acquired through the acquisition process; and a prediction process that, upon input of the feature set generated through the generation process, predicts the property values ​​using a prediction model that outputs property values ​​indicating the properties of the fuel oil.

[0027] An analysis system according to a 15th aspect of this disclosure comprises an analysis device that outputs information about components contained in fuel oil and measurement data which is a collection of measured values ​​corresponding to the information about components, and a property prediction device that performs analysis processing on the measurement data output from the analysis device and predicts the properties of the fuel oil, wherein the property prediction device performs a generation process that generates a feature set which is a collection of feature quantities related to the measured values ​​from the measurement data, and a prediction process that, upon input of the feature set generated through the generation process, predicts the property value using a prediction model that outputs a property value indicating the properties.

[0028] Embodiments of this disclosure will be described below with reference to the accompanying drawings. To facilitate understanding of the description, the same reference numerals are used for identical components and steps in each drawing whenever possible, and redundant descriptions are omitted. Where terms such as “first,” “second,” etc. are used in this specification or claims, unless otherwise specified, they do not indicate any order or importance, but are used to distinguish one configuration from another. The word “or” is interpreted in its broadest sense, i.e., “at least one,” unless otherwise specified. The word “part” may be replaced with other words such as, for example, unit, module, device, or element.

[0029] [Configuration of Analysis System 10] <Overall Structure> Figure 1 is an overall diagram of an analysis system 10 in one embodiment of the present disclosure. The analysis system 10 is provided for, for example, to analyze raw oil or fractions of raw oil and to manage the analysis results. The analysis system 10 analyzes fuel oil 12 produced from petroleum (i.e., crude oil) as extracted from an oil well, for example, and predicts the properties of the fuel oil 12. In this embodiment, the fuel oil 12 is gasoline, but it may be replaced with, for example, kerosene, light oil, heavy oil, naphtha, jet fuel, fractions obtained by fractionation from crude oil, etc.

[0030] The analysis system 10 specifically comprises a property prediction device 14, an analysis device 16, a measurement terminal 18, and a file server 20. The property prediction device 14, the measurement terminal 18, and the file server 20 are configured to communicate with each other via the NT network.

[0031] The property prediction device 14 is a device that predicts the properties of fuel oil 12. The property prediction device 14 is a computer used by users (e.g., analysts) involved in the analysis of fuel oil 12. The property prediction device 14 consists of, for example, a stationary device including a personal computer, or a portable device including a tablet, laptop, or smartphone.

[0032] The analytical apparatus 16 is, for example, an apparatus for analyzing fuel oil 12. The analytical apparatus 16 is, for example, a chromatograph that uses the chromatographic method. Examples of chromatographic methods include gas chromatography, surface liquid chromatography, liquid chromatography, or size exclusion chromatography.

[0033] The measurement terminal 18 is a computer that controls the measurement operation by the analyzer 16. The measurement terminal 18 acquires the measurement data output from the analyzer 16 and links it with various supplementary information related to the measurement (for example, identification information of the fuel oil 12 and at least one of the measurement conditions) and supplies measurement result information 22, including the measurement data, to the file server 20. The measurement data is obtained, for example, through instrumental analysis of the fuel oil 12 by the analyzer 16. The measurement data is a collection of information about the components contained in the fuel oil 12 and measured values ​​corresponding to the information about the components. The measurement data includes, for example, qualitative information about the components contained in the fuel oil 12 and quantitative information about the components contained in the fuel oil 12.

[0034] If the analytical instrument 16 is a gas chromatograph, information about the components is, for example, the retention time. The retention time is the time from when the sample (fuel oil 12) is injected into the chromatograph until the components contained in the fuel oil 12 are detected. Each component contained in the sample injected into the gas chromatograph moves along with the mobile phase in the column of the chromatograph, and as a result, differences in the speed at which they move within the column occur, leading to separation. Therefore, there is a correlation between the retention time and the types of components contained in the fuel oil 12. Thus, the types of components contained in the fuel oil 12 can be determined based on the retention time. Hence, the retention time is an example of qualitative information about the components contained in the fuel oil 12.

[0035] If the analyzer 16 is a gas chromatograph, the information regarding the components may include at least one of the following as identification information: the name of the component, the abbreviation of the component, the chemical formula of the component, and an identifier corresponding to the component. The identification information is an example of qualitative information regarding the components contained in the fuel oil 12. The identification information is an example of information regarding the components contained in the fuel oil 12. The identification information is an example of information based on retention time. The measurement terminal 18 may have, for example, a database of identification information in which a specific retention time is associated with a specific identification information (identification information database). The measurement terminal 18 may, for example, use the database of retention time and identification information contained in the measurement data measured by the analyzer 16 to add identification information corresponding to the retention time to the measurement data, or it may replace the retention time contained in the measurement data with the identification information corresponding to that retention time. Note that the identification information database may be located on the file server 20 instead of the measurement terminal 18.

[0036] If the analyzer 16 is a gas chromatograph, the measured value corresponding to the information about the components is, for example, the signal intensity value. The detector of the gas chromatograph is, for example, a flame ionization detector (hereinafter referred to as FID). The FID is a detector that can detect organic compounds, which are an example of components, and is commonly used as a detector for gas chromatographs. The FID sequentially detects the components that come out of the column of the gas chromatograph. The signal intensity value (measured value) of the FID is correlated with the concentration of the organic compound (component). Therefore, the amount of components contained in the fuel oil 12 can be determined based on the signal intensity value corresponding to a certain retention time. Hence, the signal intensity value is an example of quantitative information about the components contained in the fuel oil 12. Here, the signal intensity value includes, for example, at least one of the following: peak value, peak area, and ratio of peak areas. The peak value is, for example, the signal intensity value corresponding to one specific retention time. The peak area is, for example, the sum of the signal intensity values ​​corresponding to multiple retention times, including the retention time in which the signal intensity value is maximum. The peak area ratio is, for example, the proportion of a specific peak area included in the measurement data to the total peak areas included in the measurement data. The measurement terminal 18 may, for example, calculate at least one of the peak area and the peak area ratio based on the retention time and peak value included in the measurement data, and either add the calculated value to the measurement data or replace the peak value included in the measurement data with the calculated value.

[0037] If the analyzer 16 is a gas chromatograph, the measured value corresponding to the information about the components may be, for example, a component ratio. The component ratio may be, for example, a mass fraction or a volume fraction. A mass fraction is an index that shows the concentration as the ratio of the mass of a specific component to the total mass. A volume fraction is an index that shows the concentration as the ratio of the volume of a specific component to the total volume. Therefore, the component ratio is an example of quantitative information about the components contained in the fuel oil 12. The component ratio is an example of a measured value corresponding to the information about the components. The component ratio is an example of a value based on signal intensity. Here, the component ratio can be calculated based on the value of the signal intensity. The value of the signal intensity used to calculate the component ratio is preferably, for example, a peak area or a ratio of peak areas. The measurement terminal 18 may have, for example, a database of calibration curves in which the value of the signal intensity is associated with the concentration corresponding to the value of the signal intensity. The measurement terminal 18 may calculate the concentration of a specific component (e.g., mass fraction) based, for example, the peak area included in the measurement data and the database of calibration curves. The measuring terminal 18 may, for example, add the calculated concentration to the measurement data, or replace it with the signal intensity value included in the measurement data.

[0038] As described above, if the analyzer 16 is a gas chromatograph and the measurement data includes an identification number, component ratio, signal intensity value, and retention time, then the data will contain information that allows for the analysis of the types of components and the amounts of each component contained in the fuel oil 12.

[0039] The property data includes, for example, values ​​indicating the properties of the fuel oil 12 (hereinafter referred to as property values). The property values ​​include, for example, at least one of the following: distillation temperature, distillation volume, vapor pressure, density, sulfur content, octane number, copper plate corrosion, and actual gum. The property values ​​included in the property data can be obtained for the fuel oil 12 by using a measurement method corresponding to the desired properties. The property data is, for example, data in which the type of properties and the property values ​​are associated for each fuel oil 12.

[0040] The distillation temperature is the temperature (in °C) corresponding to the amount of distillate at a certain point in time during distillation. The distillation temperature can be measured, for example, in accordance with JIS K2254 (2018) "Petroleum Products - Method for Determining Distillation Properties," and can be measured using, for example, a simple distillation method. The distillation temperature includes, for example, the initial boiling point temperature (IPB), a volume % distillation temperature (Ta), and the final boiling point temperature (EP). The volume % distillation temperature a includes, for example, at least one of the following: a=5 "5% distillation temperature" (T5), a=10 "10% distillation temperature" (T10), a=20 "20% distillation temperature" (T20), a=30 "30% distillation temperature" (T30), a=40 "40% distillation temperature" (T40), a=50 "50% distillation temperature" (T50), a=60 "60% distillation temperature" (T60), a=70 "70% distillation temperature" (T70), a=80 "80% distillation temperature" (T80), a=90 "90% distillation temperature" (T90), a=95 "95% distillation temperature" (T95), and a=97 "97% distillation temperature" (T97).

[0041] When the fuel oil 12 is gasoline, it is preferable that the above-mentioned a volume% distillation temperatures include T10, T50, and T90. T10 is correlated with the starting performance of the gasoline engine. When T10 is below a certain value, for example, it can improve the starting performance of the gasoline engine in low-temperature environments. In this respect, it is preferable that T10 is below 70°C. T50 is correlated with the acceleration performance or warm-up performance of the gasoline engine. When T50 is below a certain value, for example, it can improve the acceleration performance of a vehicle equipped with a gasoline engine and the warm-up performance of the gasoline engine after starting. In this respect, it is preferable that T50 is below 105°C. T90 is correlated with the dilution of lubricating oil (e.g., engine oil) in the gasoline engine. When T90 is below a certain value, for example, it can suppress the dilution of lubricating oil present in the gasoline engine by gasoline. In this respect, it is preferable that T90 is below 180°C. Therefore, when the fuel oil 12 is gasoline, it is preferable that the properties to be predicted by the property prediction device 14 include at least one of T10, T50, and T90. Although the above example illustrates the case where the fuel oil 12 is gasoline, the same applies when the fuel oil 12 is an oil other than gasoline (for example, diesel fuel), although the type of a volume% distillation temperature (Ta) of interest and the preferred a volume% distillation temperature (Ta) may vary depending on the type of fuel oil 12 and the characteristics of the engine in which the fuel oil 12 is used.

[0042] The distillate volume is the amount of sample distilled at a given point in time during distillation, corresponding to the temperature. The distillate volume can be measured, for example, in accordance with JIS K2254 (2018) "Petroleum Products - Method for Determining Distillation Properties," and can be measured using, for example, a simplified distillation method. The distillate volume is expressed, for example, as a mass fraction or a volume fraction. The distillate volume includes, for example, the b°C distillate volume (Eb). The b°C distillate volume includes, for example, at least one of the following: "70°C distillate volume" (E70) where b=70, "100°C distillate volume" (E100) where b=100, and "150°C distillate volume" (E150) where b=150.

[0043] When the fuel oil 12 is gasoline, it is preferable that the b°C distillate amounts described above include E70 and E150. E70 correlates with the starting performance of a gasoline engine and the acceleration performance of a vehicle equipped with a gasoline engine. E150 correlates with the starting performance of a gasoline engine, the acceleration performance of a vehicle equipped with a gasoline engine, and the amount of carbon deposits adhering to the intake valves of a gasoline engine. Therefore, when the fuel oil 12 is gasoline, it is preferable that the properties to be predicted by the property prediction device 14 include at least one of E70 and E150. Although the above example illustrates the case where the fuel oil 12 is gasoline, the same applies when the fuel oil 12 is an oil other than gasoline (for example, diesel fuel), although the types of b°C distillate amounts of interest may vary depending on the type of fuel oil 12 and the characteristics of the engine in which the fuel oil 12 is used.

[0044] Vapor pressure is the pressure (in kPa) of the vapor of a sample present in the air. In the case of gasoline, 37.8°C is set as the reference temperature. Vapor pressure can be measured, for example, in accordance with JIS K2258-2 (2009) "Crude oil and petroleum products - Method for determining vapor pressure," and can be measured using, for example, the triple expansion method. Vapor pressure needs to be adjusted depending on the season and region in which gasoline is used, but from the viewpoint of suppressing evaporative emissions at high temperatures such as in summer, it is preferable to have a vapor pressure of, for example, 65 kPa or less. From the viewpoint of ensuring starting performance at low temperatures such as in winter, it is preferable to have a vapor pressure of, for example, 93 kPa or less. Therefore, when the fuel oil 12 is gasoline, it is preferable that the properties predicted by the property prediction device 14 include vapor pressure. In the above example, the case where the fuel oil 12 is gasoline was used, but the same applies when the fuel oil 12 is an oil other than gasoline (for example, diesel fuel), although the temperature of the vapor pressure of interest and the preferred vapor pressure value may vary depending on the type of fuel oil 12 and the characteristics of the engine in which the fuel oil 12 is used.

[0045] Density (15°C) can be measured, for example, in accordance with JIS K2249 (2011) "Petroleum products - Method for determining density". Sulfur content can be measured, for example, in accordance with JIS K2541 (2003) "Crude oil and petroleum products - Test method for sulfur content". Octane number can be measured, for example, in accordance with JIS K2280 (2018) "Petroleum products - Method for determining octane number, cetane number and cetane index". Copper plate corrosion (50°C) can be measured, for example, in accordance with JIS K2513 (2000) "Petroleum products - Test method for copper plate corrosion". Actual gum can be measured, for example, in accordance with JIS K2261 (2000) "Petroleum products - Automotive gasoline and aviation fuel oil - Test method for actual gum - Injection evaporation method".

[0046] File server 20 is an on-premises server computer that manages data files related to the analysis process of fuel oil 12. File server 20 may be a cloud-based server computer instead of an on-premises one. Although Figure 1 shows file server 20 as a single computer, file server 20 may also be a group of computers forming a distributed system.

[0047] In the example shown in Figure 1, the file server 20 has databases for property analysis tables (hereinafter referred to as "analysis table DB24"), measurement result information 22 (hereinafter referred to as "measurement result DB26"), and property prediction information (hereinafter referred to as "prediction information DB28"). The file server 20 may omit at least one of the analysis table DB24, measurement result DB26, and prediction information DB28. The file server 20 may also have databases other than the analysis table DB24, measurement result DB26, and prediction information DB28.

[0048] The property analysis table includes, for example, at least one of the following: [1] "Sample Information" including the name and type of fuel oil 12; [2] "Property Information" including the analysis items, whether or not results were obtained, the units of the indicators, and numerical values; [3] "Judgment Information" including the type of judgment criteria applied and whether the judgment criteria were passed or failed; and [4] Evidence Information including the source name and calculation method of the numerical values.

[0049] The measurement result information 22 includes, for example, at least one of the following: [1] raw values ​​of measurement data, [2] "sample information" including the name and type of fuel oil 12, [3] "analysis work information" including the name of the analyst, date and time of analysis, location of analysis, and model of the analytical instrument 16, and [4] "measurement information" including value definitions of measurement data and measurement conditions.

[0050] The prediction information includes, for example, at least one of the following: [1] a group of model parameters 56 used for predicting characteristics (Figure 3), [2] prediction result information 60 showing the prediction results of characteristics, and [3] "prediction evaluation information" including the accuracy and error of the prediction.

[0051] <Hardware configuration of the property prediction device 14> Figure 2 is a block diagram showing an example of the hardware configuration of the property prediction device 14 according to this embodiment.

[0052] The property prediction device 14 includes a processor 101, a non-volatile memory 102, a volatile memory 103, an interface (hereinafter also referred to as "IF") 104, an input device 105, and a display device 106. The processor 101, the non-volatile memory 102, the volatile memory 103, the IF 104, the input device 105, and the display device 106 are each connected to one another by a bus (data bus). The processor 101, the non-volatile memory 102, the volatile memory 103, the IF 104, the input device 105, and the display device 106 can each transmit data to one another via the bus. The property prediction device 14 functions in various functional configurations described later by the processor 101 executing a predetermined program stored in the volatile memory 103 or the non-volatile memory 102.

[0053] The volatile memory 103 is, for example, a semiconductor memory such as SRAM (Static Random Access Memory) or DRAM (Dynamic Random Access Memory). The volatile memory 103 stores programs and various data necessary for executing the processing in the property prediction device 14. The non-volatile memory 102 is, for example, a rewritable storage device such as flash memory or a hard disk. The non-volatile memory 102 stores programs and various data necessary for executing the processing in the property prediction device 14.

[0054] The processor 101 is, for example, a CPU (Central Processing Unit). However, the processor 101 is not limited to a CPU. The processor 101 may also be a GPU (Graphics Processing Unit). In a specific example, the processor 101 is a multi-core processor. The processor 101 may also be a single-core processor. The processor 101 may include multiple processors or cores and be capable of performing parallel processing. The processor 101 is configured to execute computer programs. The processor 101 may include, for example, an ASIC (Application Specific Integrated Circuit) as part, or programmable hardware such as an FPGA (Field Programmable Gate Array) or a CPLD (Complex Programmable Logic Device) as part.

[0055] IF104 is a communication interface for communication with the measurement terminal 18 or the file server 20. For example, IF104 is an Ethernet interface ("Ethernet" is a registered trademark).

[0056] The input device 105 is, for example, an input device that receives input from an external source. The input device 105 receives user operations and inputs those operations to the property prediction device 14. The input device 105 consists of, for example, a mouse, keyboard, touch sensor, and microphone.

[0057] The display device 106 is comprised of a device including, for example, a liquid crystal display or an organic EL (Electro-Luminescence) display.

[0058] The property prediction device 14 may consist of a single information processing device or multiple information processing devices. Furthermore, Figure 2 only shows a part of the main hardware configuration of the property prediction device 14, and the property prediction device 14 may have other configurations. In addition, the property prediction device 14 may omit some of the hardware exemplified; for example, the input device 105 and the display device 106 may be integrated as a touch panel. The measurement terminal 18 and the file server 20 have hardware components similar to those of the property prediction device 14.

[0059] <Functional configuration of the property prediction device 14> Figure 3 is a block diagram showing an example of the functional configuration of the property prediction device 14 shown in Figure 1. Specifically, this property prediction device 14 is a computer comprising a communication unit 32, an input unit 34, a display unit 36, a control unit 38, and a storage unit 40. Furthermore, the functions of each functional block shown in Figure 3 may be executed by a single computer or by multiple computers in a distributed manner. When the functions of each functional block shown in Figure 3 are executed by multiple computers in a distributed manner, these multiple computers may send and receive data via a communication network including a LAN (Local Area Network), a WAN (Wide Area Network), or the Internet.

[0060] The communication unit 32 is an interface for sending and receiving electrical signals to and from external devices. This allows the property prediction device 14 to, for example, acquire prediction target data 58 from the file server 20 and supply the prediction result information 60 it has generated to the file server 20.

[0061] The input unit 34 accepts information input from the user, for example. The display unit 36 ​​provides the user with, for example, prediction result information 60 generated by the property prediction device 14 in a visually recognizable manner. The property prediction device 14 may construct a graphical user interface (GUI) by combining, for example, the input function of the input unit 34 and the display function of the display unit 36.

[0062] The control unit 38 reads and executes the programs and data stored in the memory unit 40, thereby functioning as a data acquisition unit 42, a data processing unit 44, a feature generation unit 46, a learning processing unit 48, a characteristic prediction unit 50, and an output processing unit 52.

[0063] The data acquisition unit 42 acquires various data related to the learning process or prediction process. For example, the data acquisition unit 42 acquires a learning dataset 54, a group of model parameters 56, and data to be predicted 58. The data acquisition unit 42 may acquire data via communication from an external device including a file server 20, or it may acquire data via various parts of the characteristic prediction device 14 (for example, the input unit 34 or the storage unit 40).

[0064] The data processing unit 44 performs processing to modify the data acquired by the data acquisition unit 42. This processing includes, for example, at least one of the following: name matching, integration, and splitting.

[0065] "Name unification processing" is information processing for unifying the notation of component identification information when notation fluctuations occur in component identification information. Notation fluctuations can occur, for example, due to differences in the model of the analyzer 16 or the firmware version. The name unification processing is performed by referring to a data aggregate (that is, an identification information database) composed of words after name unification (for example, standard names) and word groups before name unification (for example, similar names). For example, when the component is "2-methyl-2-butene", the standard name is set as "2M2C4=" and the similar name is set as "2MC4=2" respectively, the notation can be unified to "2M2C4=" through name unification processing.

[0066] "Integration processing" is information processing for integrating a plurality of pieces of data from different sources prior to learning processing. This integration processing includes [1] "aggregation processing" for aggregating a plurality of measurement data into one location, or [2] association processing for associating input-output relationships of learning data. Examples of the aggregation processing include [1] aggregating N (N≧2) data files into n (1≦n<N) data files, and [2] aggregating a plurality of data sheets into one data sheet. An example of the association processing includes arranging 1 series of learning target data 62 and one or more property values corresponding to the learning target data 62 (that is, correct answer values 64) in a predetermined positional relationship. Through this integration processing, multidimensional data (integrated data D3 in Fig. 8) in which an aggregate of measurement data (measurement data group D1 in Fig. 8) and an aggregate of property values (property data group D2 in Fig. 8) are associated with each other is generated. In this multidimensional data, measurement values and property values are arranged in a matrix form.

[0067] "Division processing" is information processing for dividing the multidimensional data obtained through the integration processing into predetermined groups. These groups are classified, for example, by each property of the fuel oil 12. For example, a 10% distillation temperature and a 20% distillation temperature are classified into different categories. Through this division processing, a learning data set 54 is generated for each property of the fuel oil 12.

[0068] The feature quantity generation unit 46 performs generation processing for generating an aggregate of feature quantities related to the aforementioned measured values (hereinafter referred to as a "feature quantity set") from data processed by the data processing unit 44 (for example, prediction target data 58 or learning target data 62). This generation processing includes: [1] arithmetic processing for calculating an arithmetic value corresponding to a measured value in accordance with a predetermined arithmetic rule, [2] statistical processing for calculating a statistic related to a population of measured values using a statistical method, or [3] dimension reduction processing for generating M (1≦M<N) feature quantities from N (N≧2) measured values or arithmetic values.

[0069] Examples of the arithmetic rule include four arithmetic operations of addition, subtraction, multiplication, and division, function operation, or LUT (Lookup Table) operation. Examples of the statistic include a maximum value, a minimum value, an average value, a median value, a standard deviation, and a variance. Examples of the dimension reduction processing include principal component analysis (PCA), independent component analysis (ICA), latent semantic analysis (LSA), and linear discriminant analysis (LDA).

[0070] When there are a plurality of prediction models PM, the feature quantity generation unit 46 may generate a different feature quantity set for each prediction model PM, or generate a feature quantity set common to two or more prediction models PM. For example, when at least a part of input / output characteristics is common among a plurality of prediction models PM, the feature quantity generation unit 46 may generate a feature quantity set common to two or more properties. When the prediction target data 58 or the learning target data 62 includes metadata related to measurement, the feature quantity generation unit 46 may generate the feature quantity set using the metadata, or may generate the feature quantity by excluding the metadata. Examples of the metadata include: [1] measurement conditions for the fuel oil 12, [2] the type of the fuel oil 12, or [3] apparatus information related to the analysis apparatus 16.

[0071] The learning processing unit 48 performs learning processing on the learner using the training dataset 54. Specifically, the learning processing unit 48 takes a set of features generated from the training data 62 as input values ​​and performs supervised learning with one or more correct values ​​64 as output values. Examples of rules for updating the learning parameters include stochastic gradient descent, momentum method, AdaGrad method, or Adam method. The learning error may be either the mean absolute value error (MAE) or the root mean square error (RMSE).

[0072] Through this learning process, the values ​​of each of the model parameter group 56 are determined, thereby generating a predictive model PM regarding the properties of fuel oil 12. The predictive model PM is a mathematical model that, when input from the feature set generated by the feature generation unit 46, outputs property values ​​for fuel oil 12. The predictive model PM is, for example, a regression model that uses the feature set as explanatory variables and the property values ​​as the dependent variable. Examples of regression models include linear regression, ridge regression, lasso regression, elastic network regression, logistic regression, random forest, gradient boosting decision tree, support vector machine, or neural network regression.

[0073] A predictive model PM is provided for each property or type of fuel oil 12. In this case, the input / output characteristics of the learner may be the same (or uniform) regardless of the property, or they may be different. Similarly, the input / output characteristics of the learner may be the same regardless of the type of fuel oil 12, or they may be different. The learning processing unit 48 generates a predictive model PM for each property by, for example, applying a learning process to learners with common input / output characteristics using a learning dataset 54 generated for each property of fuel oil 12.

[0074] Examples of conditions for terminating the learning process (hereinafter referred to as "learning termination conditions") include: [Condition 1] the learning error becoming smaller than a threshold; [Condition 2] the number of learning iterations reaching the upper limit; or [3] a combination of (Condition 1) and (Condition 2) being met. The learning termination conditions can be set by the user as appropriate. The learning termination conditions may be the same regardless of the properties of the fuel oil 12, or they may differ depending on the properties. Specifically, the learning termination conditions for the first property value (e.g., vapor pressure) may be set to be stricter than the learning termination conditions for the second property value (e.g., distillation volume or distillation temperature).

[0075] The property prediction unit 50 performs prediction processing to predict the properties of the fuel oil 12 using the prediction model PM that has been trained by the learning processing unit 48. Prior to this prediction processing, the property prediction unit 50 selects the corresponding model parameter group 56 from among multiple types of model parameter groups 56 and sets it in the learner, thereby making the prediction model PM corresponding to the properties of the fuel oil 12 available. Based on the feature set and the prediction model PM, the property prediction unit 50 generates prediction result information 60 that includes predicted values ​​of the properties.

[0076] The output processing unit 52 performs output processing to cause the prediction results (e.g., property values) from the property prediction unit 50 to be output to the output means. The output means may be provided in the property prediction device 14 or in an external device different from the property prediction device 14. Examples of output include displaying visual information, outputting sound, or transmitting signals. Specifically, the output processing unit 52 either [1] supplies display data showing the analysis result screen 70 (Figure 12) to the display unit 36, or [2] transmits data including the prediction result information 60 to the file server 20.

[0077] The memory unit 40 stores the programs and data necessary for the control unit 38 to control each component. In the example shown in Figure 3, the memory unit 40 stores the training dataset 54, the model parameter group 56, the data to be predicted 58, and the prediction result information 60.

[0078] The training dataset 54 is a collection of training data used for training the learner. Each training dataset consists of a set of training data 62 and ground truth values ​​64. The training data 62 corresponds to the data used for training from the measurement data output from the analysis device 16. In addition to this measurement data, the training data 62 may also include the metadata described above (e.g., measurement conditions, type of fuel oil 12, etc.). The training data 62 may also include, for example, a set of features generated by the feature generation unit 46. The ground truth values ​​64 correspond to one or more property values ​​corresponding to the training data 62.

[0079] The model parameter group 56 is a collection of model parameters identified through the learning process of the prediction model PM. This model parameter group 56 is defined for each property of the fuel oil 12, or for each type of fuel oil 12. Examples of model parameters include [1] "hyperparameters" for identifying the model structure of the prediction model PM, or [2] "variable parameters" whose optimal values ​​change depending on the population of training data. Examples of variable parameters include weight coefficients between computation units and threshold values ​​for activation functions.

[0080] The data to be predicted 58 corresponds to the data used for predicting properties among the measurement data obtained through the measurement of the analysis device 16. After the properties have been predicted, the data to be predicted 58 may be used as the data to be trained 62. In addition to this measurement data, the data to be predicted 58 may also include the metadata described above (for example, measurement conditions, type of fuel oil 12, etc.). The data to be predicted 58 may also include, for example, a set of features generated by the feature generation unit 46.

[0081] The prediction result information 60 includes at least one of the following: information related to the property analysis table described above, for example, [1] "sample information" including the name or type of fuel oil 12, [2] "prediction information" including predicted values ​​for each analysis item, [3] "judgment information" including the type of judgment criteria applied and whether the judgment criteria are passed or failed, and [4] evidence information such as the method for predicting properties.

[0082] <Structure of the Predictive Model PM> Figure 4 shows an example of a combination of properties to be predicted. In the example in Figure 4, the properties to be predicted are IBP, T10, T20, T30, T40, T50, T60, T70, T80, T90, EP (all in °C), E70, E100, E150 (all in volume fraction %), and vapor pressure (in kPa). In particular, when the fuel oil 12 is gasoline, it is more preferable to predict the combination of 11 types of distillation temperatures and 3 types of distillation volumes shown in Figure 4 as the distillation properties.

[0083] Figure 5 shows an example of the structure of a predictive model PM for predicting the properties of fuel oil 12. This predictive model PM consists of a learner that takes a set of features as input and outputs a single property value (or a predicted property value). In the example shown in this figure, multiple types of predictive models PM have a common model structure. By selectively setting multiple types of model parameter groups 56 in a predetermined memory area (indicated as "LP"), at least one predictive model PM for each property of fuel oil 12 is constructed. For example, a predictive model PM for predicting the "10% distillation temperature" of gasoline is constructed.

[0084] [Operation of Analysis System 10] The analysis system 10 in this embodiment is configured as described above. Next, the operation of the analysis system 10 (in particular, the property prediction device 14) will be explained with reference to Figures 6 to 12.

[0085] <Learning operation by the property prediction device 14> Figure 6 is a flowchart showing an example of the learning operation by the property prediction device 14 shown in Figures 1 to 3.

[0086] In step SP10, the data acquisition unit 42 refers to the analysis table DB24 and measurement results DB26 of the file server 20 and acquires the data necessary for the learning process (in this case, the learning target data 62 and the correct answer value 64). This acquires the learning dataset 54.

[0087] Figure 7 shows an example of measurement data obtained by a gas chromatography system. The measurement data is, for example, in a table format that shows the correspondence between an identification number, information about the component, and the measured value corresponding to the information about the component. In the example in Figure 7, the information about the component is the component name (an example of identification information) and the retention time. In the example in Figure 7, the measured values ​​corresponding to the information about the component are the peak area value (an example of peak area), the area % (an example of the ratio of peak area), the mass fraction (an example of the component ratio), and the volume fraction (an example of the component ratio).

[0088] In step SP12 of Figure 6, the data processing unit 44 performs data matching on the training dataset 54 (in this case, the training target data 62) acquired in step SP10, as needed.

[0089] In step SP14, the data processing unit 44 performs formatting on the training dataset 54 that underwent data matching in step SP12.

[0090] Figure 8 shows an example of a method for formatting the training dataset 54. The measurement data group D1 is a collection of data in which information about fuel oil 12 (hereinafter referred to as fuel oil information) and measurement data are associated. The property data group D2 is a collection of data in which fuel oil information and property values ​​are associated. Fuel oil information is, for example, the metadata described above. First, the data processing unit 44 generates a single integrated data D3 in which measurement data and property values ​​are integrated by associating the measurement data group D1 and the property data group D2 using the fuel oil information as a key. Next, the data processing unit 44 generates a divided data group D4 consisting of multiple (15 in the example in Figure 4) divided data by property by dividing the integrated data D3. Each divided data consists of a common explanatory variable and one type of target variable.

[0091] In step SP16 of Figure 6, the feature generation unit 46 generates feature sets for each property of the fuel oil 12 from the training dataset 54 that was formatted in step SP14.

[0092] In step SP18, the learning processing unit 48 performs learning processing on the learner using the feature set generated in step SP16.

[0093] Figure 9 shows an example of how to generate the prediction model PM in Figure 5. The feature generation unit 46 selects the first divided data from the divided data group D4 and generates the first set of features using principal component analysis (PCA). The learning processing unit 48 performs learning on the learner using the combination of the first set of features and property values. This generates the first prediction model PM. The same operation is repeated thereafter to sequentially generate the nth (1 ≤ n ≤ N) prediction model PMs. In this way, N prediction model PMs are generated.

[0094] In step SP20 of Figure 6, the learning processing unit 48 checks whether the termination condition for the learning process is met. If the termination condition is not met (step SP20: NO), the control unit 38 returns to step SP10 and repeats steps SP10 to SP20 sequentially until the termination condition is met. On the other hand, if the termination condition is met (step SP20: YES), the learning processing unit 48 proceeds to the next step SP22.

[0095] In step SP22, the learning processing unit 48 saves the values ​​of each of the model parameter group 56 at the time the termination condition in step SP20 was met.

[0096] In this way, the property prediction device 14 completes the learning operation shown in the flowchart of Figure 6. As a result, the property prediction unit 50 can utilize the prediction model PM (Figure 5) through the set of model parameter group 56.

[0097] <Predictive operation by the property prediction device 14> Figure 10 is a functional block diagram relating to the predictive operation by the control unit 38 in Figure 3. Figure 11 is a flowchart showing an example of the predictive operation by the property prediction device 14 in Figures 1 to 3. This predictive operation will be explained below with reference to Figures 10 and 11.

[0098] In step SP30, the data acquisition unit 42 acquires information designated as the target of prediction (i.e., designated information) via the input unit 34 through input operations by the analysis operator. This designated information includes, for example, the sample name or type of properties of the fuel oil 12.

[0099] In step SP32, the data acquisition unit 42 refers to the specified information acquired in step SP30 and acquires the measurement data to be predicted (i.e., the prediction target data 58) from the measurement result DB26.

[0100] In step SP34, the property prediction unit 50 refers to the specified information obtained in step SP30 and retrieves a group of model parameters 56 to be used for prediction from the prediction information DB28. As a result, one or more prediction models PM are selected.

[0101] In step SP36, the control unit 38 performs preprocessing on the prediction target data 58 acquired in step SP32. This preprocessing includes [1] data matching processing by the data processing unit 44 (SP36A), [2] data formatting processing by the data processing unit 44 (SP36B), and [3] feature generation processing by the feature generation unit 46 (SP36C). Each information processing is performed in the same manner as in the learning operation described above (Figures 6 to 9).

[0102] In step SP38, the property prediction unit 50 uses the prediction model PM selected in step SP34 to predict the properties of the fuel oil 12 from the set of features generated through the preprocessing in step SP36. This yields one or more predicted values.

[0103] In step SP40, the output processing unit 52 outputs prediction result information 60, which includes the property values ​​predicted in step SP38. For example, the output processing unit 52 generates display data for displaying the analysis result screen 70 and supplies the display data to the display unit 36. As a result, the analysis result screen 70, which will be described later, is displayed on the display unit 36 ​​of the property prediction device 14.

[0104] Figure 12 shows an example of the analysis results screen 70 displayed on the display unit 36 ​​in Figure 3. The analysis results screen 70 includes an analysis results section 72 showing the analysis results of the fuel oil 12. The analysis results section 72 includes a property analysis table showing the correspondence between the sample name and the analysis items. In the corresponding column of the property analysis table, the predicted property values ​​74 (in the example in Figure 12, IBP, T10, T20, T30, T40, ... for "Gasoline Z") are entered. The analyst can grasp the properties of the fuel oil 12 at a glance through this property analysis table.

[0105] In this way, the property prediction device 14 completes the prediction operation shown in the flowchart of Figure 11. As a result, users, including the analysis operator, can use the property analysis table that reflects the prediction results by updating the analysis table DB 24.

[0106] [Summary of Embodiments] As described above, the analysis system 10 in this embodiment includes an analysis device 16 that outputs measurement data (here, prediction target data 58 or learning target data 62), which is a collection of measured values ​​related to fuel oil 12, and a property prediction device 14 that performs analytical processing on the measurement data output from the analysis device 16 to predict the properties of fuel oil 12. The property prediction device 14 predicts the properties of fuel oil 12 using a prediction model PM that outputs property values ​​when a collection of features related to measured values ​​(i.e., a feature set) is input, thereby improving the accuracy of predicting the properties of fuel oil 12.

[0107] The property prediction device 14 according to this embodiment includes a data acquisition unit 42 that acquires measurement data which is a collection of information about the components contained in the fuel oil 12 and measured values ​​corresponding to the information about the components; a feature quantity generation unit 46 that generates a feature quantity set which is a collection of feature quantities related to the measured values ​​from the measurement data acquired by the data acquisition unit 42; and a property prediction unit 50 that, upon input of the feature quantity set generated by the feature quantity generation unit 46, predicts property values ​​using a prediction model PM that outputs property values ​​indicating the properties of the fuel oil 12. Since the properties of the fuel oil 12 are predicted using the prediction model PM, the accuracy of predicting the properties of the fuel oil 12 can be improved.

[0108] Furthermore, in the property prediction device 14, there are multiple properties of the fuel oil 12 that the property prediction unit 50 predicts, and a prediction model PM may be provided for each property value predicted by the property prediction unit 50. Therefore, by using different prediction models PM according to the properties of the fuel oil 12 predicted by the property prediction unit 50, the prediction accuracy of the property values ​​of the fuel oil 12 can be improved.

[0109] Furthermore, in the property prediction device 14, the properties include at least one of distillation temperature, distillation volume, and vapor pressure, the measurement data is data measured by a gas chromatograph, the information regarding the components is based on retention time, and the measured value may be a value based on signal intensity. The gas chromatograph can separate, qualitatively and quantitatively analyze each component contained in the fuel oil 12 by heating and vaporizing the fuel oil 12. Therefore, the gas chromatograph can be considered a device that simulates distillation of the fuel oil 12 and measures the boiling point distribution of the fuel oil 12. Thus, the prediction accuracy of the properties of the fuel oil 12 (at least one of distillation temperature, distillation volume, and vapor pressure) can be further improved.

[0110] Furthermore, in the property prediction device 14, the properties of the fuel oil 12 may include at least one of the following: distillation temperature, distillation volume, and vapor pressure. These properties, for example, can affect engine starting performance when the fuel oil 12 is used to power an engine. Therefore, users can accurately predict the properties required when using the fuel oil 12. In addition, when predicting multiple properties of the fuel oil 12, each required property can be predicted from common measurement data, thus enabling efficient prediction.

[0111] Furthermore, in the property prediction device 14, the fuel oil 12 is gasoline, and the distillation temperature may include at least one of 10% distillation temperature (T10), 50% distillation temperature (T50), and 90% distillation temperature (T90). T10 correlates with the starting performance of the gasoline engine. T50 correlates with the acceleration or warm-up performance of the gasoline engine. T90 correlates with the dilution properties to lubricating oil (e.g., engine oil) in the gasoline engine. Therefore, the user can accurately predict the properties required when using the fuel oil 12.

[0112] Furthermore, in the property prediction device 14, the fuel oil 12 is gasoline, and there are multiple distillation temperatures predicted by the property prediction unit 50. The number of prediction models PM that output distillation temperatures may be equal to the number of distillation temperatures predicted by the property prediction unit 50. Therefore, by using different prediction models PM according to the properties of the fuel oil 12 predicted by the property prediction unit 50, the prediction accuracy of the property values ​​of the fuel oil 12 can be improved.

[0113] Furthermore, in the property prediction device 14, the feature generation unit 46 may generate a feature set by compressing the number of dimensions of the measurement data. This makes it possible to extract features that have a higher correlation with the properties of the fuel oil 12 while reducing the amount of information contained in the measurement data.

[0114] Furthermore, in the property prediction device 14, the prediction model PM may be a mathematical model in which a learning process is performed using different learning datasets 54 for each property of the fuel oil 12, with a learner having common input / output characteristics. By standardizing the input / output characteristics, the regularity of the learning process is increased, and the amount of computation can be reduced accordingly.

[0115] Furthermore, in the characteristic prediction device 14, the feature generation unit 46 may generate a set of features common to two or more characteristics. By commonizing the feature set, the amount of computation in the learning process or prediction process can be reduced.

[0116] Furthermore, in the property prediction device 14, the data processing unit 44 may generate integrated data D3 by associating a collection of measurement data (measurement data group D1) with a collection of property values ​​(property data group D2), and then divide the integrated data D3 according to the properties of the fuel oil 12 to generate a training dataset 54. Through data integration and division, the training dataset 54 can be generated more efficiently.

[0117] Furthermore, in the property prediction device 14, the learning processing unit 48 may generate a prediction model PM for each property by performing a learning process on a learner with common input / output characteristics using a learning dataset 54 generated by the data processing unit 44 for each property of fuel oil 12. The regularity of the learning process is increased by the common input / output characteristics, so the amount of computation can be reduced accordingly.

[0118] Furthermore, in the property prediction device 14, the information regarding components is component identification information, and the data processing unit 44 may perform a name matching process to unify the notation of the identification information before generating integrated data. Differences in component identification information arising from the type of analytical device 16, etc., can be reconciled so that they are treated as the same chemical species in the property prediction device 14. Therefore, the training dataset 54 can be generated more efficiently.

[0119] Furthermore, in the property prediction device 14, the output processing unit 52 may instruct the output means (in this case, the communication unit 32 or the display unit 36) to output a property analysis table that includes the property values ​​predicted by the property prediction unit 50. The property analysis table can be supplemented by predicting the properties of the fuel oil 12.

[0120] [Differentiation] This disclosure is not limited to the embodiments described above, and can be freely modified without departing from the spirit of this disclosure. Alternatively, the respective configurations may be combined as they see fit, without creating any technical inconsistencies. Alternatively, the execution status or execution order of each step constituting the flowchart may be changed, without creating any technical inconsistencies.

[0121] In the embodiments described above, the case where the analyzer 16 is a chromatograph was used as an example, but the type of analyzer 16 is not limited to this. For example, the analyzer 16 may be an "infrared spectrometer" (hereinafter referred to as an IR analyzer) using infrared spectroscopy, or a "nuclear magnetic resonance analyzer" (hereinafter referred to as an NMR analyzer) using nuclear magnetic resonance. The analyzer 16 is not particularly limited as long as it is a device that can acquire measurement data which is a collection of information about the components contained in the fuel oil 12 and measurement values ​​corresponding to the information about the components. For example, when using an IR analyzer instead of a gas chromatograph, "absorbance" may be used instead of "signal intensity" as the measurement value, and "wavenumber" may be used instead of "retention time" as the information about the components.

[0122] In the embodiment described above, the case in which the property prediction device 14 performs a learning process was used as an example, but the device configuration is not limited to this. For example, the property prediction device 14 does not need to be provided with a learning processing unit 48. In this case, an external device other than the property prediction device 14 stores the model parameter group 56 obtained through the learning process in the prediction information DB 28, and then the property prediction device 14 can perform property prediction processing by reading and using the model parameter group 56.

[0123] In this specification or in the claims, information, physical quantities, features, sample values, indicators, parameters, etc., may be expressed using absolute values, relative values ​​from a given value, or corresponding other information.

[0124] Where expressions such as "acquiring / setting information / using / based on / with / as input" (including similar expressions) are used in this specification or claims, unless otherwise specified, this includes using the information itself or using information that has been processed in some way (e.g., noise-added, normalized, features extracted from the information, intermediate representation of the information, etc.). Furthermore, where it is stated that some result is obtained by "acquiring / setting information / using / based on / as input" (including similar expressions), unless otherwise specified, this includes cases where the result is obtained based solely on the information in question or where the result is influenced by other information, factors, conditions, and / or states other than the information in question. Furthermore, where it is stated that "information is output" (including similar expressions), unless otherwise specified, this includes cases where the information itself is used as output or where information that has been processed in some way (e.g., noise-added, normalized, features extracted from the information, intermediate representation of various types of information, etc.) is used as output. [Explanation of Symbols]

[0125] 10...Analysis system, 12...Fuel oil, 14...Property prediction device, 16...Analysis device, 22...Measurement result information, 32...Communication unit (output means), 34...Input unit, 36...Display unit (output means), 38...Control unit, 40...Storage unit, 42...Data acquisition unit, 44...Data processing unit, 46...Feature generation unit, 48...Learning processing unit, 50...Property prediction unit, 52...Output processing unit, 54...Training dataset, 56...Model parameter group, 58...Prediction target data (measurement data), 60...Prediction result information, 62...Training target data (measurement data), 64...Correct value, 70...Analysis result screen, D1...Measurement data group, D2...Property data group, D3...Integrated data, D4...Divided data group, PM...Prediction model

Claims

1. A data acquisition unit acquires measurement data which is a collection of information about the components contained in fuel oil and measurement values ​​corresponding to the information about the components, A feature generation unit generates a feature set, which is a collection of features related to the measured values, from the measurement data acquired by the data acquisition unit, A property prediction unit, upon receiving the feature set generated by the feature generation unit, predicts the property values ​​using a prediction model that outputs property values ​​indicating the properties of the fuel oil, A property prediction device equipped with the following features.

2. The property values ​​of the fuel oil predicted by the property prediction unit are multiple, The prediction model is provided for each of the property values ​​predicted by the property prediction unit. The property prediction device according to claim 1.

3. The aforementioned properties include at least one of the distillation temperature, distillation volume, and vapor pressure. The aforementioned measurement data is data measured by a gas chromatography device. The information regarding the aforementioned components is based on retention time, The aforementioned measurement is a value based on signal intensity. The property prediction device according to claim 1.

4. The aforementioned properties include at least one of the following: distillation temperature, distillation volume, and vapor pressure. The property prediction device according to claim 1.

5. The aforementioned fuel oil is gasoline. The aforementioned distillation temperature includes at least one of the following: 10% distillation temperature, 50% distillation temperature, and 90% distillation temperature. The property prediction device according to claim 1.

6. The aforementioned fuel oil is gasoline. The property prediction unit predicts multiple distillation temperatures, The number of prediction models that output the distillation temperature is equal to the number of distillation temperatures predicted by the property prediction unit. The property prediction device according to claim 5.

7. The feature generation unit compresses the number of dimensions of the measurement data to generate the feature set. The property prediction device according to claim 1.

8. The aforementioned prediction model is a mathematical model in which a learner with common input / output characteristics is trained using different training datasets for each characteristic. The property prediction device according to claim 1.

9. The feature generation unit generates a set of features common to two or more of the characteristics. The property prediction device according to claim 6.

10. The system further includes a data processing unit that generates integrated data by associating the set of measurement data with the set of values ​​relating to the properties, and generates the learning dataset by dividing the integrated data according to the properties. The property prediction device according to claim 8.

11. The learning processing unit further comprises a learning unit that generates a predictive model for each characteristic by performing a learning process on learners having common input / output characteristics using the learning dataset generated for each characteristic by the data processing unit. The property prediction device according to claim 10.

12. The information relating to the said component is identification information of the said component, The data processing unit performs a name matching process to standardize the notation of the identification information before generating the integrated data. The property prediction device according to claim 10.

13. The system further includes an output processing unit that instructs an output means to output a property analysis table including the property values ​​predicted by the property prediction unit. The property prediction device according to claim 1.

14. An acquisition process that acquires information about the components contained in fuel oil and measurement data which is a collection of measured values ​​corresponding to the information about the components, A generation process that generates a feature set, which is a collection of features related to the measured values, from the measurement data obtained through the acquisition process, A prediction process that takes the set of features generated through the above generation process as input and predicts the property values ​​using a prediction model that outputs property values ​​indicating the properties of the fuel oil, A property prediction program that runs on one or more computers.

15. An analytical device that outputs information about the components contained in fuel oil and measurement data which is a collection of measured values ​​corresponding to the information about the components, A property prediction device that performs analytical processing on the measurement data output from the analytical device to predict the properties of the fuel oil, Equipped with, The property prediction device, A generation process that generates a feature set, which is a collection of features related to the measured values, from the aforementioned measurement data, A prediction process that takes the set of features generated through the above generation process as input and predicts the property values ​​using a prediction model that outputs property values ​​indicating the properties, An analysis system that performs this task.

Images