Smell determination system

Abstract

To provide a smell determination system that can more accurately determine whether humans sense a smell to be good or bad, and that can determine whether humans may sense an unknown smell to be good or bad.SOLUTION: A smell determination system SS is a system that determines smells in one or more spaces by a service server 31 having communication means 37. The service server 31 can receive, by the communication means 37, for each space or each smell sensor SD: measured smell change values acquired by the smell sensors SD provided in the spaces, respectively; a pleasant / unpleasant sense being a value representing pleasant / unpleasant senses about smells that are set stepwise, respectively; and a tolerance being a value representing how tolerable or intolerable each smell is. Furthermore, the service server 31 determines, through machine learning using the measured smell change values as explanatory variables, a smell determination function 108 pertaining to the pleasant / unpleasant senses and the tolerance, as an objective function.SELECTED DRAWING: Figure 1

Description

Technical Field

[0001] The present invention relates to an odor determination system that determines an odor.

Background Art

[0002] As an air conditioner capable of controlling the air purification ability according to the comfort regarding the indoor odor, one described in Japanese Patent Application Laid-Open No. 5-126391 (Patent Document 1) is known. This device has a plurality of gas sensors, a teacher signal input means, a learning means, an arithmetic means, and an operation control means. Each gas sensor detects the concentration of an odor substance in the room. The teacher signal input means receives the input of information regarding the user's comfort about the odor. The learning means learns the correlation between the detection value of each gas sensor and the user's comfort about the odor using a neural network. Based on this correlation, the arithmetic means calculates the comfort regarding the odor from the detection values of each gas sensor. The operation control means increases the air purification ability of the air purification unit of the air conditioner as the comfort is lower.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] In the above device, it is not possible to determine the comfort of human olfaction with respect to unknown odors, such as unknown odor substances that have not been learned or unknown ratios related to known odor substances. Therefore, the main object of the present invention is to provide an odor determination system that can more accurately determine the quality of an odor in human olfaction. Another main object of the present invention is to provide an odor determination system capable of determining the quality of human olfaction for an unknown odor.

Means for Solving the Problems

[0005] This specification discloses an odor determination system. This odor determination system may determine the odor in one or more spaces by a learning computer having communication means. The learning computer may receive, for each space or for each odor sensor, the odor-related value of the odor sensor, which is a value related to the odor provided in the space, or a value corresponding to the odor-related value, by means of communication. The learning computer is set step by step and may receive, for each space or for each odor sensor, the pleasantness or unpleasantness, which is a value regarding the pleasantness or unpleasantness of the odor reported for the odor. The learning computer is set step by step and may receive, for each space or for each odor sensor, the acceptability, which is a value regarding the acceptability or non-acceptability of the odor reported for the odor. The learning computer may determine an odor determination function related to pleasantness or unpleasantness as an objective function by machine learning, using the odor-related value or a value corresponding to the odor-related value as an explanatory variable. The learning computer may apply the odor-related value received after the determination or a value corresponding to the odor-related value to the determined odor determination function to derive pleasantness or unpleasantness and acceptability. The learning computer may regard the determination result of the odor as having a problem if the pleasantness or unpleasantness and the acceptability are not within predetermined ranges, respectively. In addition, the odor determination system may determine the odor in one or more spaces by a learning computer having a communication means. The learning computer may receive, for each space or for each odor sensor, the odor-related value of the odor sensor that acquires, by the communication means, the odor-related value that is a value related to the odor provided in the space or a value corresponding to the odor-related value. The learning computer is set step by step and may receive, for each space or for each odor sensor, the pleasantness or unpleasantness, which is a value regarding the pleasant or unpleasant odor reported about the odor. The learning computer is set step by step and may receive, for each space or for each odor sensor, the acceptability, which is a value regarding the acceptable or unacceptable odor reported about the odor. The learning computer may determine, by machine learning, an odor determination function related to pleasantness or unpleasantness as an objective function, using the odor-related value or a value corresponding to the odor-related value as an explanatory variable. The learning computer may apply the odor-related value received after the determination or a value corresponding to the odor-related value to the determined odor determination function to derive pleasantness or unpleasantness. The learning computer may transmit, by the communication means, the derived pleasantness or unpleasantness to the space side or the odor sensor side.

Effects of the Invention

[0006] The main effect of the present invention is to provide an odor determination system that can more accurately determine the quality of a human's sense of smell regarding an odor. In addition, another main effect of the present invention is to provide an odor determination system that can determine the quality of a human's sense of smell regarding an unknown odor.

Brief Description of the Drawings

[0007]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Figure 8

Figure 9

Figure 10

Figure 11

Mode for Carrying Out the Invention

[0008] Hereinafter, examples of embodiments according to the present invention, together with modification examples thereof, will be described as appropriate based on the drawings. Note that the embodiment is not limited to the following examples and modification examples.

[0009] FIG. 1 is an overall block diagram of an odor determination system SS and related elements according to the present invention. The odor determination system SS according to the present invention includes one or more mobile terminals 1 and one or more odor sensors SD. Also, depending on the capture method, the odor determination system SS further appropriately includes one or more service-providing server computers (service servers) 31.

[0010] The mobile terminal 1 as a terminal or a control device is a computer capable of executing an odor determination program. The mobile terminal 1 has portability and is, for example, a smartphone, a mobile phone, a notebook personal computer, or the like. Note that the computer does not necessarily have to have portability and may be, for example, a desktop personal computer, a server computer, or a combination of a plurality of computers appropriately connected via a network. The mobile terminal 1 is communicably connected to the odor sensor SD via a LAN (Local Area Network) 11. The LAN 11 may be wireless, wired, or a mixture of wireless and wired. Note that the mobile terminal 1 may be connected to the odor sensor SD by means other than the LAN 11, such as the Internet IN or a dedicated line. Typically, the mobile terminal 1 and the odor sensor SD are handled for each user. Each user places the odor sensor SD in the space (for example, a room) related to his or her own possession and places the mobile terminal 1 in a control room (for example, an office). One odor sensor SD may be arranged for each space, or a plurality of odor sensors SD may be arranged in one space. Hereinafter, for the sake of convenience of explanation, the case where one user determines the odor in one or more spaces using one mobile terminal 1 will be mainly described. The user also serves as an administrator of the odor determination system SS that causes the service server 31 to perform machine learning to prepare an environment for obtaining a more accurate odor determination. Note that the administrator may be a person different from the user, or may use a mobile terminal 1 different from the user's mobile terminal 1.

[0011] The mobile terminal 1 includes a terminal display unit 2, a terminal input unit 4, a terminal storage unit 6, a terminal communication unit 7, and a terminal control unit 8. The terminal display unit 2 displays information. For example, the terminal display unit 2 is a display or a lamp. The terminal input means 4 receives the input of information. For example, the terminal input means 4 is a keyboard or a pointing device. Incidentally, the terminal display means 2 and the terminal input means 4 may be integrated by a display with a touch sensor. The terminal storage means 6 stores information. For example, the terminal storage means 6 is a memory. The terminal communication means 7 communicates information through the Internet IN and the LAN 11. For example, the terminal communication means 7 is a communication device related to at least one of wired and wireless communication. The terminal control means 8 controls the terminal display means 2, the terminal input means 4, the terminal storage means 6, and the terminal communication means 7. For example, the terminal control means 8 is a CPU. The odor determination program is stored in the terminal storage means 6 and executed by the terminal control means 8. The terminal storage means 6 includes a non-transitory computer-readable storage medium. The odor determination program has a function of determining odor. The odor determination program is described with a set of instructions for exerting the functions described later.

[0012] The odor sensor SD is a sensor that quantifies the state of odor substances in the air and acquires and outputs them as odor-related values. The odor substances are chemical substances. Here, the odor sensor SD uses an electrical signal that changes when the odor components are adsorbed on the thin film F as an odor detection element. The odor sensor SD includes a plurality of thin films F, a sensor storage means 16, a sensor communication means 17, and a sensor control means 18. Each thin film F is a semiconductor here and has unique adsorption characteristics. For example, the first thin film F1 has an affinity for the first odorant substance and selectively adsorbs the first odorant substance. The sensor control means 18 of the odor sensor SD detects at least any one of the weight change, frequency change, dielectric constant change, and conductivity change of the first thin film F1, quantifies the change related to the first odorant substance, obtains it as an odor-related value, and stores it in the sensor storage means 16. Also, the second thin film F2 has an affinity for the second odorant substance and selectively adsorbs the second odorant substance. The sensor control means 18 of the odor sensor SD quantifies the change related to the second odorant substance by at least any one of the weight change, frequency change, dielectric constant change, and conductivity change of the second thin film F2, obtains it as an odor-related value, and stores it in the sensor storage means 16. The sensor control means 18 appropriately applies a measurement voltage for grasping various changes to each thin film F. The sensor control means 18 of the odor sensor SD comprehensively combines the changes of the odorant substances in the plurality of thin films F by a predetermined function, grasps the comprehensive change of each odorant substance as the detection target from the start of detection, grasps it as an odor change value, and stores it in the sensor storage means 16. The odor change value becomes a positive value when the comprehensive change is an increase, and becomes a negative value when the comprehensive change is a decrease. It is sufficient if the odor change value is grasped relatively, and the unit amount of the odor change value may be set as appropriate. The odor sensor control means 18 performs a detection operation over a predetermined time (for example, 5 minutes), and outputs, as the measurement odor change value C (see FIG. 7), the average value of the change values whose absolute values exceed a predetermined value within that time to the LAN 11 via the sensor communication means 17. The measurement odor change value C is determined based on those whose absolute values exceed a predetermined value within a predetermined time in the odor change value indicating the overall change of each odor substance from the start of detection, by a predetermined function that synthesizes the changes of a plurality of odor substances. Incidentally, the odor sensor control means 18 may output the odor change value at a predetermined timing (for example, every specific time). The grasping of the measurement odor change value C, which is the average value of the odor change values whose absolute values exceed a predetermined value within a predetermined time, may be performed on the portable terminal 1 side. The measurement odor change value C may be set to other values, for example, the average value of the odor change values within a predetermined time, or the median value of the odor change values within a predetermined time. The odor sensor SD may be capable of storing and outputting the absolute value regarding the odor substance, or may be capable of storing and outputting other values related to the odor substance. For example, there exist a first measurement odor change value C1 related to the first thin film F1 and a second measurement odor change value C2 related to the second thin film F2 in the measurement odor change value C. Incidentally, the thin film F may be other than a semiconductor, for example, it may be a conductor. Also, there may be one thin film F, or three or more thin films F. One type of measurement odor change value C may be output from a combination of some or all of the plurality of thin films F. The odor detection element may be other than the thin film F.

[0013] The portable terminal 1 is communicably connected to the service server 31 as a learning computer via the Internet IN. The portable terminal 1 is disposed on the odor sensor SD side. Incidentally, the learning computer may be a personal computer or the like instead of a server computer. Also, the learning computer may be integrated with the portable terminal 1, or may be installed in the same room or building as the portable terminal 1. The service server 31 is installed here in a building belonging to an organization that provides a machine learning program. Note that the service server 31 may be installed in other locations such as in a building belonging to an organization related to the said organization, or in a building belonging to a rental server operation group, etc. The service server 31 has storage means 36 similar to the terminal storage means 6, communication means 37 similar to the terminal communication means 7, and control means 38 similar to the terminal control means 8. The storage means 36 can store machine learning information related to odor determination for each user's mobile terminal 1. Note that the storage means 36 may store the machine learning information in other storage units such as for each user. The communication means 37 is connected to the mobile terminal 1 via the Internet IN. The service server 31 can execute a machine learning program. The machine learning program is stored in the storage means 36 and executed by the control means 38. The storage means 36 includes a non - transitory computer - readable storage medium. The machine learning program has a function of performing machine learning related to odor determination. A group of instructions for realizing the functions described later is described in the machine learning program.

[0014] The mobile terminal 1 can transmit the output value of the odor sensor SD to the service server 31 via the terminal communication means 7, the Internet IN, and the communication means 37, in order to cause the service server 31 to perform machine learning using the output value of the odor sensor SD belonging to the mobile terminal 1. The output value is, here, the measured odor change value C at every predetermined time. The control means 38 stores the measured odor change value C in the storage means 36 in association with the source mobile terminal 1 and the time. Also, the mobile terminal 1 can transmit the measured odor change value C to the service server 31 together with the space identification information ID, which is information for identifying the space where the odor sensor SD is provided. The control means 38 stores the measured odor change value C in association with the space identification information ID. The mobile terminal 1 may also transmit other information such as the odor change value to the service server 31 instead of or together with the change value C in the measurement. Further, instead of or together with the space identification information ID, sensor identification information that identifies the odor sensor SD may be transmitted and received. When one odor sensor SD is provided for each space, the space identification information ID and the sensor identification information are substantially the same identification information.

[0015] FIG. 2 is a schematic diagram of a plurality of odor determination result output screens D1 displayed on the terminal display means 2 of the mobile terminal 1. Furthermore, at least any one of the size, shape, and arrangement of various display elements on various screens, and the input reception mode may be changed from those described as appropriate. Also, various screens may be displayed on the entire terminal display means 2 or on a part of the terminal display means 2. Further, various screens may be displayed in parallel with other screens or may be displayed so as to overlap other screens. Also, various screens may be displayed in combination with displays related to other programs. Furthermore, a hardware button capable of performing the same input as various buttons (input parts) displayed on the screen may be used. In addition, at least any one of the display modes of various display parts and the names of various buttons may be changed from those described below. Also, at least any one of various display parts and various buttons may be displayed on other screens.

[0016] The plurality of odor determination result output screens D1 include a mode display part 51, a current time display part 52, an odor sensor arrangement display part 53, and a message display part 54. The mode display part 51 indicates the mode currently being displayed on the terminal display means 2. In the mode display part 51 on the plurality of odor determination result output screens D1, it is displayed as "Current room situation", indicating that the odor determination result is not displayed in the current odor sensor SD arrangement part. The current time display part 52 shows the current time. In this case, the odor sensor arrangement display unit 53 has sections for each odor sensor SD. In the sections of the odor sensor arrangement display unit 53, the numbers of the rooms where the odor sensors SD are arranged are displayed. Also, in the odor sensor arrangement display unit 53, color-coded displays are made for those sections. The color of a section is yellow if it is approaching the threshold value related to the odor situation value indicating the odor situation, red if it exceeds the threshold value, and white during normal times other than these. The odor situation value becomes larger as the odor that is unpleasant to human smell is stronger. The unit value of the odor situation value (i.e., the strength of the odor when it is 1) is set as appropriate. The odor situation value only needs to be grasped relatively. The threshold value of the odor situation value is set arbitrarily. The yellow sections are indicated by vertical hatching in FIG. 2. The approach to the threshold value related to the yellow section is, for example, when the value is equal to or lower than a value that is a predetermined value lower than the threshold value. The red sections are indicated by "X" - shaped hatching in FIG. 2. In the message display unit 54, a message according to the situation is displayed. In FIG. 2, a message "The odor in Room 204 has exceeded the threshold value!" is displayed. In the multiple odor determination result output screen D1, all the sections related to all the odor sensors SD may be displayed, or some of the sections related to some of the odor sensors SD may be displayed. When some sections are displayed, at least one of a scroll bar for displaying other sections, a back button, and a forward button may be displayed. The color of the section may be other colors. The sections may be distinguished and displayed by at least any one of a change in the background pattern, the line type of the boundary line, and the shape of the boundary line, instead of or together with the change in color. There may be a plurality of odor sensors SD corresponding to a section.

[0017] FIG. 3A is a schematic diagram of the odor presence determination result single output screen D2 displayed on the terminal display means 2 of the mobile terminal 1. FIG. 3B is a schematic diagram of the odor absence determination result single output screen D3 displayed on the terminal display means 2 of the mobile terminal 1. For example, the staff member cleaning Room 101 holds the mobile terminal 1 and causes the terminal display means 2 to display the odor determination result related to Room 101 by inputting to the terminal input means 4 to perform such display. Before cleaning, the staff member checks the terminal display means 2. If the odor presence determination result single output screen D2 is displayed, the staff member can grasp the necessity of suppressing the odor. The odor presence determination result single output screen D2 is displayed when the odor status value is equal to or greater than the threshold value. The odor presence determination result single output screen D2 has an icon 61A and a message 62A. The icon 61A mimics a serious face. The message 62A reads "There may be a strong odor", indicating the possibility of the presence of an odor. After cleaning, the staff member checks the terminal display means 2. If the odor absence determination result single output screen D3 is displayed, the staff member can grasp that the odor that would be concerning to human olfaction has been suppressed. The odor absence determination result single output screen D3 has an icon 61B and a message 62B. The icon 61B mimics a smiling face. The message 62B reads "There is no odor", indicating the possibility of odor suppression.

[0018] FIG. 4 is a flowchart regarding the processing related to machine learning and odor determination executed in the service server 31 of FIG. 1. The service server 31 includes a machine learning process (step S1) and an odor determination process (step S2) after a predetermined machine learning.

[0019] In the machine learning process, the user inputs various information to the service server 31 via the terminal input means 4, the terminal communication means 7, the Internet IN, and the communication means 37 in order to cause the service server 31 to perform the machine learning process related to odor. Note that the mobile terminal 1 may receive an input to a learning start button or the like for starting machine learning for confirmation. In this case, the learning start button is displayed in a pop-up window on the terminal display means 2 and may receive an input as the terminal input means 4. A cancel button may also be displayed in the pop-up window. Also, during the machine learning process, information related to the machine learning may be displayed. In this case, the information may be displayed within a pop-up window. The information may be something like "Learning in progress (currently 11 hours / 20 hours)". Also, at least any one of the above messages may be displayed together. Furthermore, during the machine learning process, a reset button for resetting the machine learning may be displayed and an input thereto may be accepted. The reset button may be displayed on the plurality of output screens D1 for the odor determination results, may be displayed in a pop-up window, or may be displayed within another screen that transitions by an input to the setting button. When an input is made to the reset button during the machine learning process, the machine learning is stopped and various information related to the machine learning is deleted. Note that after an input to the reset button, similar to the case of the above-described learning start button, a confirmation screen for further confirming whether to really reset may be displayed and a confirmation input may be accepted.

[0020] FIG. 5 is a block diagram related to the machine learning process. The machine learning process is performed in a machine learning processing unit 101 formed by a control means 38 that executes a functional learning program and a storage means 36 controlled by the control means 38. The machine learning processing unit 101 includes a storage unit 104 that receives an input of teacher data 102, a plurality (here, n) of models, namely, a first model M1, a second model M2,..., and an nth model Mn, each connected to the storage unit 104, and a function estimation unit 110 that is connected to these models and obtains an odor determination function 108 as a model function by fusing the individual processing results in these models as ensemble learning. Hereinafter, the first model M1, the second model M2,..., and the nth model Mn may be collectively referred to as the ith model Mi. i is i = 1, 2,..., n. The ith model Mi and the function estimation unit 110 can appropriately refer to the storage unit 104 in which data related to operations related to the teacher data 102 and the odor determination function 108 is stored.

[0021] The i-th model Mi is a machine learning classifier. The algorithm of the i-th model Mi is, for example, at least any one of the k-nearest neighbor method, ADABoost, CATBoost, XGBoost, Gradient Boosting Machine (GBM), LightGBM, random forest, neural network, decision tree, support vector machine, and logistic regression. Part or all of the algorithms of the i-th model Mi may be the same. From the perspective of obtaining better processing accuracy, it is preferable that part or all of the algorithms of the i-th model Mi are different from each other, and it is preferable that the number of types of the i-th model Mi is larger. However, if the number of types of the i-th model Mi is too large, the processing amount increases, the cost increases, and the response deteriorates. Therefore, it is preferable that the number of types of the i-th model Mi is increased while the processing amount is suppressed.

[0022] The function estimation unit 110 integrates the processing results of the i-th model Mi, in other words, the prediction results of the i-th model Mi, and estimates the odor determination function 108 as the final prediction result in the machine learning processing unit 101. The integration of the prediction results of the i-th model Mi can be performed in various ways. For example, the function estimation unit 110 may integrate the results obtained by applying all the target teacher data 102 to the i-th model Mi at the time of processing by voting, or may integrate the results obtained by dividing the teacher data 102 into a plurality of groups and applying different groups to the i-th model Mi, or may generate an integration model from the i-th model Mi and treat the processing result of the integration model as the integrated processing result. The integration by voting may use the maximum value of the processing results of the i-th model Mi as the maximum value voting, or may use the weighted average voting, which is the weighted value determined by the coefficients for the processing results of the i-th model Mi. The integration by processing each group of teacher data 102 may be performed by averaging the respective processing results as a Bagging Ensemble, or as a Boosting Ensemble, where the processing result of the first model M1 is used as the teacher data 102 of the second model M2, the processing result of the second model M2 is used as the teacher data 102 of the third model M3, and this is repeated up to the nth model Mn to obtain the final result. Alternatively, as a method using an integrated model, a Stacking Ensemble in which the ith model Mi is linearly combined may be used. Note that the machine learning processing unit 101 may perform other machine learning instead of ensemble learning, or may perform multivariate analysis such as linear regression.

[0023] Figure 6 is a flowchart regarding machine learning processing (step S1). Figure 7 is a schematic diagram of the teacher data 102 stored in the storage unit 104 of Figure 5. The machine learning processing unit 101 that executes the machine learning processing stores, in the storage unit 104, combinations of the following plurality of items as teacher data 102 (step S11). As groups of items of the teacher data 102, there are an item group G1 related to the odor sensor SD and an item group G2 related to the reported values based on human olfaction. The item group G1 related to the odor sensor SD is, here, the measured odor change value C for each thin film F (each odor substance) detected by the odor sensor SD. Also, the item group G2 related to the reported values is, here, the pleasantness-unpleasantness PU and the acceptability AL. The evaluation of the reported values by the odor evaluator is performed, for example, in a laboratory simulating the living room of a hotel or a residence before and after cleaning, in a place where various fragrances are volatilized.

[0024] The pleasantness-unpleasantness PU is set in nine levels here. When the odor evaluator's sense of smell is extremely unpleasant in the room where the odor may exist, the pleasantness-unpleasantness PU is evaluated and set to -4 and reported. When the odor evaluator is very unpleasant, the pleasantness-unpleasantness PU is evaluated to -3. When the odor evaluator is unpleasant, the pleasantness-unpleasantness PU is evaluated to -2. When the odor evaluator is slightly unpleasant, the pleasantness-unpleasantness PU is evaluated to -1. When the odor evaluator is neither pleasant nor unpleasant, the pleasantness-unpleasantness PU is evaluated to 0. When the odor evaluator is slightly comfortable, the pleasantness-unpleasantness PU is evaluated to 1. When the odor evaluator is comfortable, the pleasantness-unpleasantness PU is evaluated to 2. When the odor evaluator is very comfortable, the pleasantness-unpleasantness PU is evaluated to 3. When the odor evaluator is extremely comfortable, the pleasantness-unpleasantness PU is evaluated to 4. The pleasantness-unpleasantness PU is associated with the change value C in the odor measurement at the time when the evaluation is made and stored as part of the teacher data 102.

[0025] The acceptability AL is set in two levels here. When the odor evaluator can accept and tolerate the odor with their sense of smell in the room where the odor may exist, the acceptability AL is evaluated and set to 0 and reported. Also, when the odor evaluator cannot accept and tolerate the odor, the acceptability AL is evaluated and set to 1. Furthermore, the pleasantness-unpleasantness PU and the acceptability AL can be evaluated independently. That is, if the odor evaluator is comfortable and acceptable with the odor (including no odor), the pleasantness-unpleasantness PU becomes larger and the acceptability AL becomes 1. Also, if the odor evaluator is unpleasant and unacceptable with the odor, the pleasantness-unpleasantness PU becomes smaller and the acceptability AL becomes 0. On the other hand, if the odor evaluator is comfortable with the odor but unacceptable, the pleasantness-unpleasantness PU becomes larger and the acceptability AL becomes 0. For example, it is assumed that a favorable type of fragrance wafts strongly. Also, if the odor evaluator is unpleasant with the odor but acceptable, the pleasantness-unpleasantness PU becomes smaller and the acceptability AL becomes 1. For example, it is assumed that a type of fragrance that is not to be avoided but not preferred wafts slightly.

[0026] Furthermore, the odor intensity OI was also tested as an item group G2 related to the declared value. Here, the odor intensity OI is set in six levels. A human being in the same environment as the odor sensor SD, in other words, an odor evaluator in the same room as the odor sensor SD, declares and evaluates the odor intensity OI as 0 when they smell no odor with their own sense of smell. Also, when there is an odor that can just be detected, the odor evaluator evaluates the odor intensity OI as 1. When there is a weak odor such that the odor evaluator can tell what kind of odor it is, the odor evaluator evaluates the odor intensity OI as 3. When there is an odor that can be easily detected, the odor evaluator evaluates the odor intensity OI as 4. When there is a strong odor, the odor evaluator evaluates the odor intensity OI as 5. When there is an intense odor, the odor evaluator evaluates the odor intensity OI as 6. The odor intensity OI is associated with the measurement odor change value C at the time when the evaluation is made and stored as part of the teacher data 102. Here, the evaluation by the odor evaluator is done by filling out a questionnaire, and the same applies to the item group G2 related to the declared value below. Furthermore, the odor intensity OI may be five levels or less, or may be seven levels or more. At least any one of the setting contents of each level may be changed from the above. Similarly, at least any one of the number of levels and the setting contents may be changed in other items in the item group G2 related to the declared value.

[0027] Then, a threshold for evaluating a human's sense of smell was set for the item group G2 related to the declared value. Figure 8 is a graph showing the relationship between the pleasantness / unpleasantness PU and the acceptability AL. Figure 9 is a graph showing the relationship between the odor intensity OI and the pleasantness / unpleasantness PU. Figure 10 is a graph showing the relationship between the odor intensity OI and the acceptability AL. In Figures 8 to 10, one plot shows the values at one location where a set of the item group G1 related to the odor sensor SD and the item group G2 related to the declared value was obtained. That is, as shown in FIG. 8, when the pleasure-unpleasure feeling PU, which becomes more comfortable as the value increases, is equal to or greater than the first predetermined threshold value T1 (here, -0.38), and the acceptability AL, which is more acceptable as the value decreases, is equal to or less than the second predetermined threshold value T2 (here, 0.29), the odor is in a state without problems and is set to be within the allowable range PR (GOOD). When the pleasure-unpleasure feeling PU is below the first predetermined threshold value T1 or the acceptability AL is above the second predetermined threshold value T2, the odor is in a problematic state and is set to be outside the allowable range PR (BAD). In addition, at least one of the values of the first predetermined threshold value T1 and the second predetermined threshold value T2 may be other than the above values. Also, for example, when the acceptability AL is acceptable, 1 is set, and when it is not acceptable, 0 is set. When the acceptability AL is equal to or greater than the second predetermined threshold value T2 and the pleasure-unpleasure feeling PU is equal to or greater than the first predetermined threshold value T1, the allowable range PR is within the allowable range. The setting format regarding the allowable range PR may be changed.

[0028] As shown in FIG. 9, in the relationship between the odor intensity OI and the pleasure-unpleasure feeling PU, a plurality of plots corresponding to the evaluation of the odor are distributed throughout the graph. More specifically, in addition to the plots located on the diagonal line DL1 from the point (0, 4) where the odor intensity OI is small and the pleasure-unpleasure feeling PU is large to the point (5, -4) where the odor intensity OI is large and the pleasure-unpleasure feeling PU is small and its adjacent part, there are also many plots within the region SP1 where the odor intensity OI ranges from 2 to 4, which is wide, but the pleasure-unpleasure feeling PU is large and the odor is comfortable. That is, there are many plots within the region SP1 where the pleasure-unpleasure feeling PU is large and is independent of the odor intensity OI. Therefore, in the relationship between the odor intensity OI and the pleasure-unpleasure feeling PU, the allowable range PR is not set.

[0029] Also, as shown in FIG. 10, in the relationship between the odor intensity OI and the acceptability AL, a plurality of plots corresponding to the evaluation of the odor are distributed throughout the graph. More specifically, in addition to the plots located on the diagonal line DL2 from the point (0, 0) where the odor intensity OI is small and the acceptability AL is small to the point (5, 1.0) where the odor intensity OI is large and the acceptability AL is large and in its adjacent part, there are also many plots within the region SP2 where the odor intensity OI ranges widely from 1 to 4 but the acceptability AL is small and the odor is acceptable. That is, there are many plots within the region SP2 where the acceptability AL is small and is independent of the odor intensity OI. Therefore, in the relationship between the odor intensity OI and the acceptability AL, the allowable range PR is not set.

[0030] The machine learning processing unit 101 first performs step S11 until a predetermined amount or more of the teacher data 102 is accumulated (No in step S12). The condition for once finishing the accumulation of the teacher data 102 may be that a predetermined period (for example, 7 days) has elapsed since the start of the accumulation, or that a predetermined number of plots or more have been obtained, or a composite condition such as after a predetermined period has elapsed or after the number of plots has exceeded a predetermined number. Also, the machine learning processing unit 101 may proceed to the next step S13 and perform the calculation (update) of machine learning each time the teacher data 102 is once obtained.

[0031] When the machine learning processing unit 101 accumulates a predetermined amount or more of teacher data 102 (Yes in step S12), it applies part or all of the teacher data 102 to the i-th model Mi to obtain individual processing results, fuses them as ensemble learning, and derives an odor determination function 108 (step S13). That is, the machine learning processing unit 101 performs machine learning by referring to the teacher data 102 in the storage unit 104 in the function estimation unit 110, and determines the odor determination function 108 as a function of the item group G2 related to the declared value based on human olfaction, with the item group G1 related to the odor sensor SD (i.e., the change value C in the measurement of odor) as the explanatory variable. The machine learning processing unit 101 stores the obtained odor determination function 108 in the storage means 36 for each user, completes the current machine learning, and ends the machine learning processing. Note that the explanatory variable may be a part of the item group G1 related to the odor sensor SD, or may be one with other elements added. Also, the objective variable may be a part of the item group G2 related to the declared value based on human olfaction. In particular, when only the determination of whether it is within the allowable range PR (GOOD) or outside the allowable range PR (BAD) is performed, only the pleasantness / unpleasantness PU and the acceptability AL may be used. Furthermore, the objective variable may be one with other elements added.

[0032] After the machine learning processing of the machine learning processing unit 101 ends, the control means 38 performs an odor determination process (step S2). FIG. 11 is a flowchart regarding the odor determination process. In the odor determination process, the control means 38 provides an odor determination result based on the odor determination function 108.

[0033] That is, in response to a smell determination request from the mobile terminal 1 in which a space is specified and the current measurement change value C and space identification information ID of the smell sensor SD related to the space are attached, the control means 38 stores the measurement change value C and the space identification information ID in the storage means 36 (step S21). At the same time, the measurement change value C is applied to the smell determination function 108 to obtain the pleasantness / unpleasantness feeling PU and the acceptability AL, and it is derived whether these are within the allowable range PR (GOOD) or not (BAD) (step S22). Note that step S21 may be omitted. Further, the control means 38 transmits any one of the pleasantness / unpleasantness feeling PU, the acceptability AL, and GOOD or BAD related to the allowable range PR to the mobile terminal 1 (step S23). Note that only the result related to the allowable range PR may be transmitted, and the pleasantness / unpleasantness feeling PU and the acceptability AL may not be transmitted. Alternatively, the pleasantness / unpleasantness feeling PU and the acceptability AL may be transmitted, the result related to the allowable range PR may not be transmitted, and the determination of GOOD or BAD related to the allowable range PR may be made on the mobile terminal 1 side. Further, the control means 38 (the machine learning processing unit 101) may also perform machine learning using at least any one of the measurement change value C obtained from the mobile terminal 1, the pleasantness / unpleasantness feeling PU, and the acceptability AL as the teacher data 102 during the smell determination process. In this case, although the processing amount increases, the machine learning regarding the smell determination function 108 will be deepened while performing the smell determination.

[0034] The mobile terminal 1 that has received any one of the pleasantness / unpleasantness feeling PU, the acceptability AL, and the result (GOOD or BAD) related to the allowable range PR stores these in the terminal storage means 6 (step S24). At the same time, according to the type of the result related to the allowable range PR, the smell present determination result single output screen D2 (BAD) or the smell absent determination result single output screen D3 (GOOD) is displayed on the terminal display means 2, and further, according to the space identification information ID, the display of the corresponding section in the smell determination result multiple output screen D1 is updated (step S25).

[0035] Such a smell determination system SS has the following operational effects. That is, the odor determination system SS is a system for determining an odor in one or more spaces by a service server 31 having a communication means 37. The service server 31, via the communication means 37, obtains a measurement odor change value C which is a value corresponding to the odor-related value of an odor sensor SD that obtains an odor-related value which is a value related to the odor provided in the space, a pleasantness-unpleasantness feeling PU which is a value for the pleasantness or unpleasantness of the odor, respectively set step by step and reported for each space's odor, and an acceptability AL which is a value for the acceptability or non-acceptability of the odor, for each space or for each odor sensor SD. Further, the service server 31 uses the measurement odor change value C as an explanatory variable, and by machine learning, determines an odor determination function 108 related to the pleasantness-unpleasantness feeling PU and the acceptability AL as an objective function, applies the measurement odor change value C received after the determination to the determined odor determination function 108, derives the pleasantness-unpleasantness feeling PU and the acceptability AL, and if the pleasantness-unpleasantness feeling PU and the acceptability AL are not within a tolerance range PR which is a predetermined range respectively, determines that there is a problem with the odor determination result. Therefore, by deriving the odor determination function 108 using machine learning related to the measurement odor change value C, the pleasantness-unpleasantness feeling PU, and the acceptability AL, and by determining the odor using the tolerance range PR, an odor determination system SS is provided that can more accurately determine the quality of a human's sense of smell regarding an odor. Also, by deriving the odor determination function 108 using machine learning related to the measurement odor change value C, the pleasantness-unpleasantness feeling PU, and the acceptability AL, the quality of a human's sense of smell regarding an unknown odor can be determined.

[0036] Furthermore, the odor determination system SS is a system that determines the odor in one or more spaces by the service server 31 having the communication means 37. The service server 31 uses the communication means 37 to obtain the measured odor change value C, which is the value corresponding to the odor-related value of the odor sensor SD that is a value related to the odor provided in the space, the pleasure-displeasure feeling PU, which is a value for the pleasure or displeasure of the odor and is set step by step and reported for the odor in the space respectively, and the acceptability AL, which is a value for the acceptable or unacceptable of the odor, for each space or for each odor sensor SD. Also, the service server 31 uses the measured odor change value C as an explanatory variable, determines the odor determination function 108 related to the pleasure-displeasure feeling PU and the acceptability AL as an objective function by machine learning, applies the measured odor change value C received after the determination to the determined odor determination function 108, derives the pleasure-displeasure feeling PU and the acceptability AL, and transmits the derived pleasure-displeasure feeling PU and acceptability AL to the space side or the odor sensor side SD by the communication means 37. Therefore, by deriving the odor determination function 108 using machine learning related to the measured odor change value C and the pleasure-displeasure feeling PU and the acceptability AL, an odor determination system SS is provided that outputs the pleasure-displeasure feeling PU and the acceptability AL for more accurately determining the quality of human olfaction regarding the odor. Also, by deriving the odor determination function 108 using machine learning related to the measured odor change value C and the pleasure-displeasure feeling PU and the acceptability AL, the pleasure-displeasure feeling PU and the acceptability AL for determining the quality of human olfaction regarding unknown odors can be output.

[0037] Also, the value corresponding to the odor-related value is the measured odor change value C determined based on those whose absolute value exceeds a predetermined value within a predetermined time in the odor change value indicating the overall change of each odor substance as the detection target from the start of detection by a predetermined function that synthesizes the changes of a plurality of odor substances. Therefore, while making the accuracy of odor determination sufficient, the amount of information related to the value corresponding to the odor-related value is suppressed, and the calculation amount in the service server 31 is suppressed. Furthermore, the service server 31 can receive a plurality of sets of the change value C, as well as the pleasantness-unpleasantness feeling PU and the acceptability AL, in states where the times are different from each other in the measurement. Therefore, these values can be obtained more efficiently in a state where the accuracy of the determination of the odor is sufficient. Furthermore, the predetermined range is a range in which the pleasantness-unpleasantness feeling PU is equal to or less than a first predetermined threshold value T1 and the acceptability AL is equal to or less than a second predetermined threshold value T2. Therefore, the determination of the odor is performed more appropriately.

[0038] In addition, when the service server 31 outputs the determination result of the odor, the odor determination system SS further includes a mobile terminal 1 having a terminal communication unit 7 and a terminal display unit 2. The terminal communication unit 7 receives the determination result of the odor from the communication unit 37 of the service server 31. The terminal display unit 2 displays an odor presence determination result single output screen D2, which is information regarding the determination result of the odor. Therefore, in the mobile terminal 1 on the space side where the odor may exist, the determination result of the odor is shown, and an appropriate determination result of the odor is provided to the user.

[0039] Also, when the service server 31 outputs the pleasantness-unpleasantness feeling PU and the acceptability AL, the odor determination system SS further includes a mobile terminal 1 having a terminal communication unit 7, a terminal display unit 2, and a terminal control unit 8. The terminal communication unit 7 receives the pleasantness-unpleasantness feeling PU and the acceptability AL from the communication unit 37 of the service server 31. The terminal control unit 8 determines that there is a problem with the determination result of the odor when the pleasantness-unpleasantness feeling PU and the acceptability AL are not within a permissible range PR that is a predetermined range. The terminal display unit 2 displays information regarding the determination result of the odor. Therefore, in the mobile terminal 1 on the space side where the odor may exist, the determination result of the odor is calculated based on the pleasantness-unpleasantness feeling PU and the acceptability AL, and an appropriate determination result of the odor is provided to the user.

[0040] In addition, the odor determination system SS and the odor determination method may appropriately have the following modification examples in addition to the above-described modification examples. That is, the service server 31 (learning computer) and the mobile terminal 1 in the odor determination system SS may be communicably connected via a dedicated line or the like instead of the Internet IN. In short, it suffices that they are connected by a communication network. Furthermore, the mobile terminal 1 may control other devices, such as switching the on / off state of the deodorizer to on when the determination result of the odor is BAD and to off when it is GOOD, according to the determination result of the odor. In addition, when the mobile terminal 1 has a low cost and high processing power, etc., it may play the role of the service server 31 that determines the odor in the surrounding space and the like by machine learning. Furthermore, each step in the odor determination method (each flowchart) may be replaced with steps of substantially the same other content, may be divided into a plurality of steps with substantially the same content, at least any two of them may be combined with each other, and the order may be appropriately changed. Furthermore, the machine learning may be performed for each segment including a plurality of spaces instead of for each space. By applying the odor machine learning result of a part of the space in the segment to the spaces in the same segment by such machine learning for each segment, the machine learning can be performed more efficiently. Also, for a new space, by determining which segment it belongs to and applying the machine learning result in the same segment, the machine learning can be performed even more efficiently. Alternatively, by performing machine learning on a plurality of spaces in the same segment together, it is also possible to improve the accuracy.

Explanation of Reference Numerals

[0041] C1... Control system (of air conditioner), 1... Mobile terminal (terminal), 11... Remote controller (RC), 31... Service server (service - providing server computer, learning computer), 37... Communication means, 108... Air - conditioning set temperature function, C... Measured odor change value (value corresponding to odor - related value), SD... Odor sensor, SS... Odor determination system.

Claims

1. An odor determination system for determining an odor in one or more spaces by a learning computer having a communication means, wherein the learning computer, by the communication means, the odor-related value of an odor sensor provided in the space for obtaining an odor-related value that is a value related to the odor, or a value corresponding to the odor-related value, and the pleasantness-unpleasantness, which are values for the pleasantness or unpleasantness of the odor respectively reported for the odor and are set step by step, and the acceptability, which is a value for the acceptable or unacceptable of the odor, can be received for each space or for each odor sensor, and using the odor-related value or the value corresponding to the odor-related value as an explanatory variable, an odor determination function related to the pleasantness-unpleasantness and the acceptability is determined by machine learning as an objective function, the odor-related value or the value corresponding to the odor-related value received after the determination is applied to the determined odor determination function to derive the pleasantness-unpleasantness and the acceptability, if the pleasantness-unpleasantness and the acceptability are not within predetermined ranges respectively, the determination result of the odor is regarded as having a problem characterizing the odor determination system.

2. An odor determination system for determining an odor in one or more spaces by a learning computer having a communication means, wherein the learning computer, by the communication means, the odor-related value of an odor sensor provided in the space for obtaining an odor-related value that is a value related to the odor, or a value corresponding to the odor-related value, and the pleasantness-unpleasantness, which are values for the pleasantness or unpleasantness of the odor respectively reported for the odor and are set step by step, and the acceptability, which is a value for the acceptable or unacceptable of the odor, can be received for each space or for each odor sensor, and using the odor-related value or the value corresponding to the odor-related value as an explanatory variable, an odor determination function related to the pleasantness-unpleasantness and the acceptability is determined by machine learning as an objective function, the odor-related value or the value corresponding to the odor-related value received after the determination is applied to the determined odor determination function to derive the pleasantness-unpleasantness and the acceptability, and the derived pleasantness-unpleasantness and acceptability are transmitted to the space side or the odor sensor side by the communication means characterizing the odor determination system.

3. The value corresponding to the odor-related value described above is a measurement odor change value determined based on those whose absolute value exceeds a predetermined value within a predetermined time in the odor change value indicating the comprehensive change of each odor substance as the detection target from the start of detection, by a predetermined function that synthesizes the changes of a plurality of odor substances. The odor determination system according to claim 1 or claim 2, characterized in that.

4. The learning computer can receive a plurality of sets of the odor-related value or the value corresponding to the odor-related value, and the pleasantness / unpleasantness and the acceptability, in states where the times are different from each other. The odor determination system according to claim 1 or claim 2, characterized in that.

5. The predetermined range is a range in which the pleasantness / unpleasantness is equal to or less than a first predetermined threshold value and the acceptability is equal to or less than a second predetermined threshold value. The odor determination system according to claim 1, characterized in that.

6. Furthermore, it is provided with a terminal having terminal communication means and terminal display means, The terminal communication means receives the determination result of the odor from the communication means of the learning computer, The terminal display means displays information regarding the determination result of the odor. The odor determination system according to claim 1, characterized in that.

7. Furthermore, it is provided with a terminal having terminal communication means, terminal display means, and terminal control means, The terminal communication means receives the pleasantness / unpleasantness and the acceptability from the communication means of the learning computer, When the pleasantness / unpleasantness and the acceptability are not respectively within a predetermined range, the terminal control means determines that there is a problem with the determination result of the odor, The terminal display means displays information regarding the determination result of the odor. The odor determination system according to claim 2, characterized in that.

Images