Space evaluation device

Abstract

To simply and quantitatively evaluate closeness level of an evaluation object unknown space to a natural environment regardless of the unknown space being an indoor space or an outdoor space.SOLUTION: A space evaluation device 1 is a device for evaluating an object space by using naturalness having an index of closeness of the space to the natural environment. The space evaluation device 1 includes: a storage device 2 which stores VOCs data indicating a composition of a volatile organic compound contained in a sample collected from air in the object space; and a processing device 3 which predicts the naturalness of the object space. The processing device 3 predicts the naturalness of the object space from the VOCs data of the object space by using a learned prediction model 32 which is previously constructed by machine learning.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to a space evaluation device. [Background technology]

[0002] As interest in maintaining and improving human health and physical and mental functions grows, attention is being focused on creating spaces that are effective in improving labor productivity and reducing stress. For example, it is well known that humans can experience healing effects when they coexist with plants, and there are high hopes for spaces that incorporate biophilic design, making people feel as if they are in a natural forest. Biophilic design is a spatial design method based on the concept of biophilia, which states that humans instinctively seek a connection with nature. In spatial design such as biophilic design, it is important to understand how close a space is to the natural environment.

[0003] Methods for objectively evaluating the natural environment have been proposed in the past. Patent Document 1 discloses a method for evaluating forest areas by analyzing tree trunk shape images taken from the air and the results of spectral analysis of the forest areas. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Patent No. 4900356 Summary of the Invention [Problem to be solved by the invention]

[0005] However, the method disclosed in Patent Document 1 mainly analyzes image data captured from the sky, and is therefore limited to evaluation using images captured in outdoor spaces.

[0006] The present invention has been made in consideration of the above-mentioned circumstances, and aims to easily and quantitatively evaluate how close an unknown space to be evaluated is to a natural environment, regardless of whether the space is an indoor space or an outdoor space. [Means for solving the problem]

[0007] In order to solve the above-mentioned problems, the space evaluation device of the present invention is a space evaluation device that evaluates a target space based on its degree of naturalness, which is an index of how close the space is to a natural environment, and is equipped with a memory device that stores VOCs data indicating the composition of volatile organic compounds contained in samples collected from the air of the target space, and a processing device that predicts the degree of naturalness of the target space, and is characterized in that the processing device predicts the degree of naturalness of the target space from the VOCs data of the target space using a pre-trained prediction model constructed in advance by machine learning. [Effects of the Invention]

[0008] According to the present invention, it is possible to easily and quantitatively evaluate how close an unknown space to be evaluated is to a natural environment, regardless of whether the space is an indoor space or an outdoor space. [Brief explanation of the drawings]

[0009] [Figure 1] FIG. 1 is a diagram showing the configuration of a space evaluation device. [Figure 2] FIG. 4 is a diagram showing an example of environmental data. [Figure 3] FIG. 1 is a diagram illustrating a method for calculating BPS. [Figure 4] FIG. 1 shows the weights of each basis estimated by LASSO regression and the volatile organic compounds that make up each of the bases K0 and K7. [Figure 5] FIG. 10 is a diagram showing the BPS of samples in which the bases K0 and K7 are likely to be observed. [Figure 6] FIG. 10 shows the results of verifying the validity of the prediction model. DETAILED DESCRIPTION OF THE INVENTION

[0010] Hereinafter, embodiments of the present invention will be described with reference to the drawings. Components with the same reference numerals in each embodiment have similar components in each embodiment unless otherwise specified, and description thereof will be omitted.

[0011] The configuration of the space evaluation device 1 and a method for calculating the naturalness will be described with reference to Figures 1 to 3. Figure 1 is a diagram showing the configuration of the space evaluation device 1. Figure 2 is a diagram showing an example of environmental data. Figure 3 is a diagram explaining a method for calculating the BPS.

[0012] The space evaluation device 1 is a device for evaluating the degree to which various spaces, including outdoor spaces such as forests or urban areas, and indoor spaces such as offices or residences, resemble natural environments. The space evaluation device 1 is effective in realizing spaces that incorporate the aforementioned biophilic design. In spatial design, such as biophilic design, which creates a symbiotic space with plants that evokes a sense of nature, it is important to grasp the "degree of naturalness" as an index of how similar the space is to the natural environment. Furthermore, people are affected not only by sensory stimuli such as sight and hearing, but also by the quality of the air present in a space (hereinafter also referred to as "air quality"). The inventors have discovered that the degree of naturalness of a space is influenced by the air quality of the space. In particular, the inventors have found that the degree of naturalness of a space is significantly influenced by volatile organic compounds (hereinafter also referred to as "VOCs") present in the air of the space.

[0013] The space evaluation device 1 predicts the naturalness of a target space from VOCs data indicating the composition of volatile organic compounds contained in a sample collected from the air of the target space. The target space can be any space that can be determined as an indoor or outdoor space. In this embodiment, the biophilic score (hereinafter also referred to as "BPS") is introduced as the naturalness index. The BPS is an index in which a higher value indicates a space that is closer to a natural environment, and a lower value indicates a space that is closer to an artificial environment. The BPS is an index calculated from "environmental data" acquired by sensors and sensory evaluation in multiple reference spaces that differ in the amount of natural objects, such as forests, or artificial objects, such as concrete buildings.

[0014] Environmental data is data that indicates spatial conditions such as temperature or humidity. As shown in FIG. 2, the environmental data includes a plurality of quantitative data acquired by various sensors in each reference space and a plurality of qualitative data acquired by sensory evaluation such as a questionnaire survey in the reference space. The acquired environmental data is associated with samples collected in the reference space in which the environmental data was acquired and stored in a table such as that shown in the upper part of FIG. 3. As shown in the upper part of FIG. 3, this table stores the quantitative data and the qualitative data separately.

[0015] BPS is calculated by performing multifactor analysis (MFA) on environmental data. Specifically, principal component analysis (PCA) is performed on the quantitative data contained in the environmental data, and multiple correspondence analysis (MCA) is performed on the qualitative data contained in the environmental data, followed by singular value decomposition (SVDE) on each. To unify the scales of the data, the quantitative data is divided by the first singular value obtained by SVDE on the quantitative data, and the qualitative data is divided by the first singular value obtained by SVDE on the qualitative data. A table containing the scaled quantitative data and a table containing the scaled qualitative data are then merged. Principal component analysis is then performed on all data stored in the merged table. As a result, the high-dimensional environmental data, which includes multiple quantitative and qualitative data, is dimensionally compressed into one-dimensional continuous-value data, as shown by the number line in the bottom row of Figure 3.

[0016] Looking at the environmental data items plotted below the number line in Figure 3, we can see that "artificial" items appear as the number moves in the negative direction, and "natural" items appear as the number moves in the positive direction. Therefore, the number line in Figure 3 can be said to express how close a space is to an artificial or natural environment.

[0017] Additionally, samples collected in each reference space are plotted above the number line shown in FIG. 3 . In other words, a numerical value on the number line is assigned to each reference space from which each sample was collected. Samples from reference spaces of artificial environments, such as laboratories or factories, are plotted on the negative side of the number line, and samples from reference spaces of natural environments, such as forests, are plotted on the positive side of the number line. While this may seem obvious, it is important to be able to express the perceived "naturalness" numerically. In this embodiment, the one-dimensional continuous value data shown on the number line in FIG. 3 is defined as BPS. In this manner, BPS is calculated from environmental data acquired in each of multiple reference spaces.

[0018] 1, the space evaluation device 1 includes a storage device 2 and a processing device 3. Although not shown, the space evaluation device 1 may also include an input device that inputs data and the like to the storage device 2, an output device that outputs the processing results of the processing device 3, and a communication device that communicates with external devices.

[0019] VOCs data of samples collected in the target space is stored in the storage device 2. The VOCs data of the target space is input to the space evaluation device 1 by a user and stored in the storage device 2.

[0020] VOCs data is data obtained by analyzing the composition of volatile organic compounds present in the air. In this embodiment, VOCs data was obtained using the following equipment: a sampling device used to collect samples from the air and an analyzer for the collected samples. Specifically, during sample collection, a glass collection tube (manufactured by GL Science) filled with Tenax-TA, a porous polymer adsorbent, was connected to a mini-pump MP-W5P (manufactured by Shibata Scientific), and 10 L of air was passed through the glass collection tube at a rate of 300 mL / min to collect the sample. During sample analysis, the collected sample was injected into a gas chromatograph mass spectrometer (Agilent) equipped with a thermal desorption device TD-100 (manufactured by Markes), and the composition of the volatile organic compounds present in the sample was analyzed. The analytical conditions for the gas chromatograph mass spectrometer (GC / MS) were as shown in Table 1. The mass spectrometry data obtained was deconvoluted using AMDIS (Automated Mass Spectral Deconvolution & Identification System), and then statistical processing was performed using Mass Profiler Professional (Agilent) to identify the compounds. The data obtained in this way was used as VOCs data.

[0021] [Table 1]

[0022] The storage device 2 also stores VOCs data for a negative control (hereinafter also referred to as "NC") sample. The NC sample is a substance that is not inherently present in the air of the target space and that may be mixed in during the process of collecting a sample from the air of the target space and acquiring VOCs data. The NC sample is a substance that is present in a sample collection device, an analysis device, or the like. In this embodiment, either or both of the VOCs data for substances present in the collection device before the sample is collected or the VOCs data for substances present in the analysis device before the sample is analyzed are stored in advance in the storage device 2 as VOCs data for the NC sample.

[0023] The processing device 3 includes a processor such as a CPU and memories such as a ROM and a RAM. The processing device 3 realizes a function of predicting BPS by the processor executing a program stored in the memory. As this function, the processing device 3 includes an NC elimination unit 31 and a prediction model 32.

[0024] The NC removal unit 31 removes the VOCs data of the NC samples stored in the storage device 2 from the VOCs data of the target space stored in the storage device 2. The NC removal unit 31 outputs the VOCs data of the target space from which the VOCs data of the NC samples has been removed to the prediction model 32.

[0025] The prediction model 32 is a model that predicts the BPS (naturalness) of the target space from the VOCs data of the target space. The prediction model 32 receives the VOCs data of the target space output from the NC removal unit 31, and outputs the BPS corresponding to the input VOCs data. The prediction model 32 is a trained model that has been constructed in advance by machine learning. A method for constructing the prediction model 32 by machine learning will be described later.

[0026] The prediction model 32 includes an extraction unit 33 and a prediction unit 34. The extraction unit 33 extracts, from the VOCs data of the target space input to the prediction model 32, feature quantities that represent characteristics of the composition of volatile organic compounds in the VOCs data. Specifically, the extraction unit 33 extracts, as feature quantities, bases generated by performing nonnegative matrix factorization (NMF) on the VOCs data of the target space. The prediction unit 34 predicts the BPS of the target space based on the feature quantities extracted by the extraction unit 33. Specifically, the prediction unit 34 predicts the BPS of the target space from the basis quantities extracted by the extraction unit 33. The basis quantities represent characteristic compositions of volatile organic compounds that represent the VOCs data.

[0027] A method for constructing a prediction model 32 by machine learning will be described with reference to Figures 1 and 4 to 6. Figure 4 is a diagram showing the weights of each of the bases K0 to K14 estimated by LASSO regression and the volatile organic compounds that make up each of the bases K0 and K7.

[0028] In the learning stage of the prediction model 32, as shown in FIG. 1, first, a learning sample is collected from the air in each of a plurality of predetermined reference spaces. The composition of volatile organic compounds contained in each collected sample is analyzed to obtain VOCs data for each of the plurality of reference spaces. Similarly, VOCs data for the NC sample is obtained. Then, the VOCs data for the NC sample is removed from the VOCs data for each of the plurality of reference spaces. Furthermore, environmental data is obtained for each of the plurality of reference spaces. The BPS is calculated based on the obtained environmental data. Then, a dataset is created by linking the VOCs data for each of the plurality of reference spaces from which the VOCs data for the NC sample has been removed with the BPS corresponding to each of the plurality of reference spaces.

[0029] The prediction model 32 is constructed by machine learning using a dataset in which the VOCs data of each of the multiple reference spaces from which the VOCs data of the NC sample has been removed is linked to the BPS of each of the multiple reference spaces.

[0030] VOCs data is expressed as a non-negative matrix. Non-negative matrix VOCs data can be decomposed into the product of two non-negative matrices with an intervening basis by non-negative matrix factorization. That is, when VOCs data is expressed as a P×N matrix X, non-negative matrix factorization can decompose the matrix X into the product of a P×K (= basis) non-negative matrix W and a K (= basis)×N non-negative matrix H. The extraction unit 33 of the prediction model 32 is optimized in advance by machine learning to generate a basis that minimizes the difference between the matrix X and the product WH of the matrix W and the matrix H. The number of bases is optimized in advance as the number that can most accurately represent the original VOCs data. In this embodiment, 15 bases K0 to K14 are generated by non-negative matrix factorization of the VOCs data. The VOCs data, which is high-dimensional data, is dimensionally reduced by non-negative matrix factorization. Each dimension compressed by the non-negative matrix factorization corresponds to a basis that is generated.

[0031] The prediction unit 34 of the prediction model 32 is configured as a regression model that uses each basis generated by performing nonnegative matrix factorization on the VOCs data as an explanatory variable and BPS as a response variable. In this embodiment, the regression model is constructed using LASSO regression (least absolute shrinkage and selection operator regression), a machine learning technique. In LASSO regression, weights for each of the bases K0 to K14, which are explanatory variables, are estimated. The weights for each of the bases K0 to K14 indicate the degree of influence (contribution rate) on BPS, which is the response variable.

[0032] As shown on the number line in the upper part of Figure 4, the weights of each base K0 to K14 indicate a composition of volatile organic compounds that are more abundant in natural environments, as the bases have a larger weight value, and a composition of volatile organic compounds that are more abundant in artificial environments, as the bases have a smaller weight value. In fact, samples in which bases K7 or K8, which have a larger weight value, are likely to be observed, are often samples collected from campsites with lots of greenery, while samples in which bases K11 or K0, which have a smaller weight value, are likely to be observed, are often samples collected from ordinary offices.

[0033] The volatile organic compounds that make up the base K7, which has a high weight value, are rich in plant-derived volatile organic compounds, such as (1R)-2,6,6-Trimethylbicyclo[3.1.1]hept-2-ene, Bicyclo[3.1.0]hexan-2-one, 5-(1-methylethyl)-, and Bicyclo[3.1.1]heptane, 6,6-dimethyl-2-methylene-, (1S)-, as shown in the lower right of Figure 4. On the other hand, the volatile organic compounds that make up the base KO, which has a low weight value, are rich in artificial volatile organic compounds, such as 1,3-Butanediol and 2-Propanol, 1-methoxy-, as shown in the lower left of Figure 4. Note that each table in the lower part of Figure 4 lists only the top 20 volatile organic compounds that make up each base; there are also volatile organic compounds not listed in each table.

[0034] Fig. 5 is a diagram showing the BPS of samples in which the bases K0 and K7 are likely to be observed. In the upper graph of Fig. 5, the horizontal axis shows the BPS of the sample, and the vertical axis shows the proportion of the bases K7 observed in one sample among the bases K0 to K14. In the lower graph of Fig. 5, the horizontal axis shows the BPS of the sample, and the vertical axis shows the proportion of the bases K0 observed in one sample among the bases K0 to K14. The BPS of a sample refers to the BPS calculated from environmental data acquired in the reference space in which the sample was collected.

[0035] As shown in the upper part of Figure 5, samples with a large proportion of basis K7 with a large weight value tend to have positive BPS values. As shown in the lower part of Figure 5, samples with a large proportion of basis K0 with a small weight value tend to have negative BPS values. In other words, it can be seen that bases with large weight values ​​have a high contribution rate to BPS, and bases with small weight values ​​have a low contribution rate to BPS.

[0036] Fig. 6 is a diagram showing the results of verifying the validity of the prediction model 32. Fig. 6 shows the correlation between the predicted value of BPS by the prediction model 32 and the actual value (correct value). The vertical axis of Fig. 6 represents the predicted value of BPS by the prediction model 32, and the horizontal axis of Fig. 6 represents the actual value of BPS.

[0037] The prediction accuracy of prediction model 32, which predicts BPS from VOCs data, was verified using five-fold cross-validation. As shown in Figure 6, the predicted BPS values ​​by prediction model 32 exhibit a high correlation coefficient (Pearson's ρ) with the actual values, at 0.684. It can be seen that the predicted BPS values ​​by prediction model 32 have a positive correlation with the actual values. In other words, it can be seen that prediction model 32 is capable of accurately predicting the naturalness of a space (BPS) from VOCs data. Therefore, it can be seen that prediction model 32 is valid.

[0038] As described above, the space evaluation device 1 of this embodiment is a device that evaluates a target space based on its naturalness, which is an index of how close the space is to a natural environment. The space evaluation device 1 includes a storage device 2 that stores VOCs data indicating the composition of volatile organic compounds contained in samples collected from the air of the target space, and a processing device 3 that predicts the naturalness of the target space. The processing device 3 predicts the naturalness of the target space from the VOCs data of the target space using a trained prediction model 32 that has been constructed in advance by machine learning.

[0039] As a result, the space evaluation device 1 can predict the naturalness level solely from the VOCs data by simply collecting a sample from the air of a target space, which can be determined arbitrarily, and acquiring the VOCs data of the collected sample. In other words, the space evaluation device 1 can predict the naturalness level solely from the VOCs data, without having to photograph the target space from the air or perform sensory evaluations each time. Additionally, the space evaluation device 1 is applicable to both indoor spaces, which lack soil, and outdoor spaces, which resemble natural environments. Therefore, the space evaluation device 1 can easily and quantitatively evaluate the degree to which an unknown space to be evaluated resembles a natural environment, regardless of whether it is an indoor space or an outdoor space.

[0040] Furthermore, in the space evaluation device 1 of this embodiment, the prediction model 32 includes an extraction unit 33 that extracts feature quantities representing compositional characteristics in the VOCs data of the target space, and a prediction unit 34 that predicts the naturalness of the target space based on the feature quantities extracted by the extraction unit 33. The extraction unit 33 extracts, as feature quantities, bases generated by performing non-negative matrix factorization on the VOCs data of the target space. The prediction unit 34 predicts the naturalness of the target space from the bases extracted by the extraction unit 33.

[0041] This allows the space evaluation device 1 to easily extract, for example, the characteristics (i.e., the basis) of the composition of volatile organic compounds that affect the naturalness of the VOCs data. Therefore, the space evaluation device 1 can more easily and accurately predict the naturalness of the space using the prediction model 32. Therefore, the space evaluation device 1 can more easily and quantitatively evaluate how close the unknown space to be evaluated is to a natural environment, regardless of whether it is an indoor or outdoor space. Moreover, since the space evaluation device 1 can extract the basis that affects the naturalness as described above, it can clearly indicate which basis should be added or removed to change the naturalness. Therefore, the space evaluation device 1 can easily and quantitatively grasp the composition of volatile organic compounds required to achieve a desired naturalness. Therefore, the space evaluation device 1 can significantly contribute to easily and quantitatively formulating design guidelines for spaces that achieve a desired naturalness.

[0042] Furthermore, in the space evaluation device 1 of this embodiment, the naturalness is an index calculated from environmental data acquired by sensors and sensory evaluation in each of a plurality of reference spaces that differ in the amount of natural or artificial objects, with a larger naturalness value indicating a space closer to a natural environment and a smaller naturalness value indicating a space closer to an artificial environment. The prediction model 32 is constructed by machine learning using a dataset that links VOCs data acquired by analyzing learning samples collected from the air in each of the plurality of reference spaces with the naturalness calculated from the environmental data acquired in each of the plurality of reference spaces.

[0043] As a result, the space evaluation device 1 can calculate and set the degree of naturalness by combining various data from different perspectives, such as data acquired by sensors and data acquired by sensory evaluation, thereby establishing the degree of naturalness as a highly probable index that can be used for comprehensive evaluation from various perspectives. In particular, by including data acquired by sensory evaluation in the environmental data, the space evaluation device 1 can establish the degree of naturalness as an index that is close to the results of human sensory evaluation. Therefore, the space evaluation device 1 can accurately predict the degree of naturalness using the prediction model 32. Therefore, the space evaluation device 1 can accurately, easily, and quantitatively evaluate how close an unknown space to be evaluated is to a natural environment, regardless of whether it is an indoor space or an outdoor space.

[0044] Furthermore, in the space evaluation device 1 of this embodiment, VOCs data is acquired by analyzing a sample collected by a collection device with an analysis device. The storage device 2 stores, as VOCs data of an NC sample, either or both of VOCs data of substances present in the collection device before the sample is collected or VOCs data of substances present in the analysis device before the sample is analyzed. The processing device 3 further includes an NC removal unit 31 that removes the VOCs data of the NC sample from the VOCs data of the target space stored in the storage device 2. The prediction model 32 is constructed by machine learning using a dataset that links the VOCs data of each of a plurality of reference spaces from which the VOCs data of the NC sample has been removed with the degree of naturalness of each of the plurality of reference spaces.

[0045] This allows the space evaluation device 1 to predict the degree of naturalness from the original VOCs data of the collected sample. Therefore, the space evaluation device 1 can more accurately predict the degree of naturalness using the prediction model 32. Therefore, the space evaluation device 1 can more accurately, simply, and quantitatively evaluate how close the unknown space to be evaluated is to a natural environment, regardless of whether it is an indoor space or an outdoor space.

[0046] Although the embodiments of the present invention have been described in detail above, the present invention is not limited to these embodiments and various modifications can be made without departing from the spirit of the present invention. In the present invention, elements of one embodiment can be added to elements of another embodiment, elements of one embodiment can be replaced with elements of another embodiment, or some of the elements of one embodiment can be deleted. [Explanation of symbols]

[0047] 1... space evaluation device, 2... storage device, 3... processing device, 31... NC removal unit, 32... prediction model, 33... extraction unit, 34... prediction unit

Claims

1. A space evaluation device that evaluates a target space based on the degree of naturalness, which is an index of how close the space is to a natural environment, a storage device that stores VOCs data indicating the composition of volatile organic compounds contained in a sample collected from the air of the target space; a processing device for predicting the degree of naturalness of the target space, The processing device predicts the degree of naturalness of the target space from the VOCs data of the target space using a trained prediction model that has been constructed in advance by machine learning. A space evaluation device characterized by:

2. The predictive model is an extraction unit that extracts a feature quantity that represents a feature of the composition in the VOCs data of the target space; a prediction unit that predicts the naturalness of the target space based on the feature amount extracted by the extraction unit, the extraction unit extracts, as the feature, a basis generated by performing non-negative matrix factorization on the VOCs data in the target space; The prediction unit predicts the degree of naturalness of the target space from the basis extracted by the extraction unit.

2. The space evaluation device according to claim 1.

3. The degree of naturalness is an index calculated from environmental data acquired by sensors and sensory evaluation in each of a plurality of reference spaces having different abundances of natural or artificial objects, and a larger value of the degree of naturalness indicates a space closer to the natural environment, and a smaller value of the degree of naturalness indicates a space closer to an artificial environment, The prediction model is constructed by the machine learning using a data set in which the VOCs data obtained by analyzing learning samples collected from the air in each of the plurality of reference spaces is linked to the naturalness calculated from the environmental data obtained in each of the plurality of reference spaces.

3. The space evaluation device according to claim 2.

4. the VOCs data is obtained by analyzing a sample collected by a collecting device with an analyzing device; the storage device stores, as the VOCs data of a negative control sample, either the VOCs data of a substance present in the collection device before the sample is collected or the VOCs data of a substance present in the analysis device before the sample is analyzed, or both of these; the processing device further includes a negative control removal unit that removes the VOCs data of the negative control sample from the VOCs data of the target space stored in the storage device; The prediction model is constructed by the machine learning using the dataset in which the VOC data of each of the plurality of reference spaces from which the VOC data of the negative control sample has been removed is associated with the degree of naturalness of each of the plurality of reference spaces.

4. The space evaluation device according to claim 3.

Images