A method for rapid analysis of key aroma components in iris essential oil using artificial intelligence

Abstract

This invention relates to the fields of fragrance and flavor analysis and artificial intelligence technology, and discloses a method for rapid analysis of the key aroma components of iris essential oil using artificial intelligence. The method acquires gas chromatography-mass spectrometry (GC-MS) data of iris essential oil, constructs compound feature vectors using an aroma knowledge base, calculates the overall aroma contribution value using an artificial intelligence fusion model, and stratifies the core components, auxiliary components, and modifying components. Simultaneously, it outputs recommended proportion ranges for aroma reconstruction. This method can be used for key aroma identification, aroma reconstruction, and fragrance design assistance in natural essential oils.

Description

Technical Field

[0001] This invention relates to the fields of fragrance and flavor analysis and artificial intelligence technology, and in particular to a method for rapidly analyzing the key aroma components of iris essential oil using artificial intelligence. Background Technology

[0002] Iris essential oil is typically derived from the rhizomes of plants in the genus *Iris*. After drying and aging, it is obtained through distillation or extraction. It possesses long-lasting, powdery, violet-like, and fixative characteristics, making it highly valuable in high-end perfumes and functional fragrances. Current techniques for analyzing the components of iris essential oil usually employ gas chromatography-mass spectrometry (GC-MS) to obtain qualitative and semi-quantitative results for volatile compounds. However, relying solely on GC-MS results only provides chemical composition information and makes it difficult to directly identify which compounds dominate the main iris fragrance, which compounds primarily contribute to the waxy base, and which low-content components have a non-linear contribution to complexity.

[0003] On the other hand, simply sorting by peak area can easily overestimate the high content of fatty acids while underestimating the low content but highly typical irisone components. Existing methods of artificially experienced perfumery or based on simple statistical sorting lack a computational path that integrates quantitative detection data, odor knowledge, and target aroma characteristics, making it difficult to achieve automatic identification, stratification, and proportion recommendation of key aroma components. Summary of the Invention

[0004] Technical problems to be solved The purpose of this invention is to provide a method for rapid analysis of the key aroma components of iris essential oil using artificial intelligence, in order to solve the problems in the prior art that it is difficult to directly analyze the overall aroma contribution structure based solely on GC-MS qualitative and quantitative results, cannot distinguish between characteristic dominant components and structural supporting components, and cannot automatically provide suggestions on aroma reconstruction ratios.

[0005] Technical solution

[0006] To achieve the above objectives, the present invention adopts the following technical solution: As attached Figure 1 As shown, step A is sample detection and data acquisition. Iris essential oil samples are obtained, and GC-MS is used to detect the samples, obtaining the retention time, retention index, peak area percentage, mass spectrometry matching degree, and molecular information of each compound, forming the original chemical composition dataset.

[0007] Step B involves the construction and retrieval of an odor knowledge base. For the compounds identified in Step A, an odor knowledge base is built or retrieved. This odor knowledge base includes at least: (1) Basic identification fields of compounds: name, CAS number, molecular formula; (2) Scent semantic fields: scent descriptors, scent category, scent hierarchy tags; (3) Aroma contribution fields: odor threshold, iris typicality weight, longevity support weight, evaporation rate label; (4) Relationship fields: Collaborative relationships, Masking relationships, Co-occurrence relationships; (5) Credibility fields: source category, literature credibility label, experiment credibility label.

[0008] Step C involves feature extraction and vectorization. Feature vectors are constructed for each compound: X_i=[P_i, O_i, T_i, B_i, S_i, Q_i, M_i] Wherein, P_i is the peak area normalized value, O_i is the odor activity term, T_i is the iris typicality weight, B_i is the lingering aroma support weight, S_i is the similarity between the compound odor description vector and the iris target aroma vector, Q_i is the synergistic / antagonistic correction term, and M_i is the qualitative confidence term.

[0009] In a preferred embodiment: P_i = A_i / ΣA_j Where A_i is the peak area of ​​the i-th compound, and ΣA_j is the sum of the peak areas of all selected compounds.

[0010] O_i = ln[1 + P_i / (OT_i + θ)] Where OT_i is the odor threshold of the i-th compound after normalization, and θ is a correction constant to prevent the denominator from being zero.

[0011] S_i = cos(V_i, V_t) Where cos represents the cosine similarity function, V_i is the odor description vector of the i-th compound, and V_t is the iris target aroma vector.

[0012] As attached Figure 2 As shown, step D is artificial intelligence fusion computing.

[0013] The feature vector obtained in step C is input into the artificial intelligence fusion model. The artificial intelligence fusion model includes a feature encoding module, a knowledge retrieval module, a score fusion module, a hierarchy determination module, and a text interpretation module. The overall aroma contribution value C_i is calculated according to the following formula: C_i=αP_i+βO_i+γT_i+δB_i+εS_i+ζQ_i+ηM_i Among them, α, β, γ, δ, ε, ζ, η are weight coefficients, and α+β+γ+δ+ε+ζ+η=1.

[0014] In a preferred embodiment, α=0.18, β=0.14, γ=0.22, δ=0.18, ε=0.16, ζ=0.07, and η=0.05.

[0015] As attached Figure 3 As shown, step E is the output of hierarchical classification and key aroma composition.

[0016] Based on the overall aroma contribution value C_i, each compound was stratified as follows: when C_i ≥ 0.65, it was identified as a core component; when 0.35 ≤ C_i < 0.65, it was identified as a secondary component; when 0.15 ≤ C_i < 0.35, it was identified as a modifying component; and when C_i < 0.15, it was identified as a background component. The key aroma composition of iris essential oil was then output according to the ranking of each component.

[0017] Step F: Recommended Aroma Reconstruction Ratio Range. Calculate the recommended ratio baseline value R_i = C_i / ΣC_j for the compounds entering the output results, and further provide the stratified ratio ranges: core components are 0.80–1.20 times R_i; auxiliary components are 0.60–1.00 times R_i; and modifying components are 0.20–0.60 times R_i.

[0018] Beneficial effects

[0019] Compared with existing technologies, this invention provides a method for rapid analysis of key aroma components in iris essential oil using artificial intelligence, which has the following beneficial effects: 1. Instead of sorting solely by peak area, the system integrates detection results, odor knowledge, and target aroma semantics, thereby improving the accuracy of identifying key aroma components.

[0020] 2. It can distinguish between structural support components and characteristic dominant components, and avoid the peak area fatty acids completely masking low-content, highly typical irisone components.

[0021] 3. It can identify components with low content but high typicality and strong synergistic effects, and improve the ability to analyze components that contribute to tail tone and complexity.

[0022] 4. It outputs three-layer results: core components, auxiliary components, and modifying components, which are highly interpretable.

[0023] 5. It can further provide the aroma reconstruction ratio range, shorten the R&D cycle, reduce trial and error costs, and facilitate batch consistency correction. Attached Figure Description

[0024] Figure 1 This is a flowchart illustrating the method of the present invention.

[0025] Figure 2 This is a schematic diagram of the structure of the collaborative analysis between the artificial intelligence fusion model and the odor knowledge base in this invention.

[0026] Figure 3This is a hierarchical diagram illustrating the output results of key aroma components in this invention. Detailed Implementation

[0027] Example 1: Key Aroma Analysis of Iris Flower Essential Oil Based on Disclosed Sample Data 1. Samples and testing conditions Iris essential oil was selected as the sample to be tested. An Agilent 7890B-5973C gas chromatography-mass spectrometry (GC-MS) system was used for detection. The GC column was a DB-HeavyWax polar capillary column with dimensions of 30 m × 0.25 mm × 0.25 μm. The injection volume was 1 μL, the injection port temperature was 250 °C, and split injection was used with a split ratio of 10:1. Helium was used as the carrier gas at a flow rate of 1.0 mL / min. The temperature program was as follows: 40 °C for 2 minutes, increased to 260 °C at 10 °C / min, and held for 6 minutes. The mass spectrometry conditions were: electron impact ionization, ionization energy 70 eV, transfer line temperature 230 °C, ion source temperature 230 °C, quadrupole temperature 150 °C, interface temperature 250 °C, and the scan range was 50–500 atomic mass units.

[0028] 2. Sample composition dataset Based on the detection results, a sample composition dataset containing 43 compounds was established. Table 1 lists the GC-MS qualitative and quantitative results of Example 1.

[0029] Table 1. Qualitative and quantitative results of GC-MS analysis of iris essential oil in Example 1 1 10.48 Limonene Limonene C10H16 138-86-3 0.09 2 27.93 Tridecane, 2-methyl- 2-Methyltridecane C14H30 1560-96-9 0.06 3 28.04 3-Carene,4-acetyl- 4-Acetyl-3-carene C12H18O 1000156-14-1 0.04 4 29.81 Hexadecanoic acid, methylester Methyl hexadecanoate C17H34O2 112-39-0 0.13 5 30.37 Hexadecanoic acid, ethyl ester Ethyl hexadecanoate C18H36O2 628-97-7 2.07 6 30.66 n-Decanoicacid n-decanoic acid C10H20O2 334-48-5 2.20 7 30.77 Ethyl9-hexadecenoate ethyl 9-hexadecenoate C18H34O2 54546-22-4 0.26 8 31.00 Tricosane Tridecane C23H48 638-67-5 0.52 9 33.31 Octadecanoic acid, ethyl ester Ethyl octadecanoate C20H40O2 111-61-5 0.20 10 33.68 Dodecanoic acid dodecanoic acid C12H24O2 143-07-7 11.43 11 33.82 Pentacosane Pentadecane C25H52 629-99-2 2.57 12 34.26 Linoleicacidethylester Ethyl linoleate C20H36O2 544-35-4 2.42 13 34.48 Benzene,1,2,3-trimethoxy-5-methyl- 1,2,3-Trimethoxy-5-methylbenzene C10H14O3 6443-69-2 0.27 14 35.14 Linolenicacidethylester Ethyl linolenic acid C20H34O2 1191-41-9 0.51 15 35.37 γ-Tetradecalactone γ-Tetradecyl lactone C14H26O2 2721-23-5 0.24 16 35.82 Apocynin Apuxining C9H10O3 498-02-2 4.92 17 36.50 Tetradecanoic acid Tetradecanoic acid C14H28O2 544-63-8 35.08 18 36.81 Oleic Acid Oleic acid C18H34O2 112-80-1 0.55 19 37.06 Z-7-Tetradecenoic acid Z-7-Tetradecenoic acid C14H26O2 1000130-98-4 0.54 20 38.41 9,12-Octadecadienoic acid(Z,Z)- 9,12-Octadecadienoic acid (Z,Z) C18H32O2 60-33-3 3.58 21 27.81 Ethyl9-tetradecenoate ethyl 9-tetradecenoate C16H30O2 1000336-60-8 0.28 22 27.41 Octanoicacid bitter C8H16O2 124-07-2 0.55 23 27.23 Tetradecanoic acid, ethyl ester Ethyl tetradecanoate C16H32O2 124-06-1 10.59 24 26.82 γ-irone γ-Irisone C14H22O 79-68-5 4.59 25 10.60 1H-Pyrazole,4,5-dihydro-3-methyl- 1H-pyrazole 4,5-dihydro-3-methyl- C4H8N2 1911-30-4 0.01 26 15.79 Octanoic acid, ethyl ester Ethyl octanoate C10H20O2 106-32-1 0.23 27 18.09 Linalool Linalool C10H18O 78-70-6 0.05 28 19.12 Decanoic acid, methylester Methyl decanoate C11H22O2 110-42-9 0.06 29 19.95 Decanoic acid, ethyl ester Ethyl decanoate C12H24O2 110-38-3 0.86 30 21.63 Methylmethacrylate Methyl methacrylate C5H8O2 80-62-6 0.05 31 21.80 Aceticacid, phenylmethylester Benzyl acetate C9H10O2 140-11-4 0.06 32 39.34 n-Hexadecanoic acid hexadecanoic acid C16H32O2 57-10-3 5.14 33 23.00 Dodecanoic acid, methylester Methyl dodecanoate C13H26O2 111-82-0 0.15 34 23.74 Dodecanoic acid, ethyl ester Ethyl dodecanoate C14H28O2 106-33-2 3.32 35 25.39 Spiro[2.5]octane,5,5-dimethyl-4-(3-oxobutyl)- Spiro[2.5]octane, 5,5-dimethyl-4-(3-oxobutyl)- C14H24O 77143-32-9 0.69 36 25.68 α-Irone α-Irisone C14H22O 79-69-6 0.15 37 25.83 2-Methyl-3-methoxy-4H-pyran-4-one 2-Methyl-3-methoxy-4H-pyran-4-one C7H8O3 4780-14-7 0.04 38 26.03 Ethanone, 1-(1H-pyrrol-2-yl)- Ethyl ketone, 1-(1H-pyrrolo-2-yl)- C6H7NO 1072-83-9 0.08 39 26.24 (E)-4-((1R,5S)-2,5,6,6-Tetramethylcyclohex-2-en-1-yl)but-3-en-2-one-rel- (E)-4-((1R,5S)-2,5,6,6-tetramethylcyclohexyl-2-en-1-yl)but-3-en-2-one-rel- C14H22O 472-46-8 1.37 40 26.37 4-Cyclohexenyl-3-methylbutene 4-Cyclohexenyl-3-methylbutene C11H18 1000432-70-6 0.14 41 26.56 Methyltetradecanoate Methyltetradecanoate C15H30O2 124-10-7 0.47 42 22.86 4-methyl-1- 4-Methyl-1-heptane C8H16O 1186-31-8 0.02 Hepten-4-ol En-4-ol 43 39.97 Palmitoleicacid Palm oil acid C16H30O2 373-49-9 0.18 3. Construction of an odor knowledge base Establish a dedicated aroma knowledge base for iris essential oil analysis. For each compound, record at least the following fields: basic identifier, aroma description, aroma category, aroma threshold, iris typicality weight, longevity support weight, synergistic / antagonistic relationship, and credibility tag.

[0030] In a preferred embodiment, the iris target aroma vector V_t is defined as: [0.28, 0.22, 0.16, 0.12, 0.10, 0.07, 0.05], corresponding to violet / iris characteristics, powdery, waxy, woody, herbal, fruity, and sweet aromas, respectively. For each compound, a vector V_i of the same dimension is constructed based on its aroma description, and then S_i = cos(V_i, V_t) is calculated.

[0031] 4. Artificial Intelligence Fusion Model Setup The artificial intelligence fusion model in this embodiment consists of a feature encoding module, a knowledge retrieval module, a score fusion module, a hierarchy determination module, and a text interpretation module. In a preferred embodiment, the model is implemented by combining a semantic embedding model and a rule fusion engine, wherein the semantic embedding model is responsible for mapping compound odor descriptors into vectors, and the rule fusion engine is responsible for calculating the comprehensive aroma contribution value according to the formula and performing threshold determination.

[0032] 5. Overall Aroma Contribution and Hierarchical Output For each compound in the sample composition dataset, P_i, O_i, T_i, B_i, S_i, Q_i, and M_i were calculated. Based on the calculation results, the hierarchical output and overall aroma contribution value of representative compounds were obtained, as shown in Table 2.

[0033] Table 2. Hierarchical output results of representative compounds core ingredients γ-Irisone 4.59 0.86 Provides a fragrance with violet / iris as the main characteristics. Low proportion but highly typical core ingredients Tetradecanoic acid 35.08 0.89 Provides a waxy base and a long-lasting fragrance framework High proportion of structural support components core ingredients Apuxining 4.92 0.73 Enhance the herbal aroma and natural feel Improve naturalness Auxiliary ingredients Ethyl tetradecanoate 10.59 0.58 Provides fruity flavor complement and softens the aroma. Balanced wax base Auxiliary ingredients dodecanoic acid 11.43 0.51 Strengthening the continuity of wax Enhance structural balance Auxiliary ingredients (E)-4-((1R,5S)-2,5,6,6-tetramethylcyclohexyl-2-en-1-yl)but-3-en-2-one 1.37 0.44 Provides wood taillight support Enhance the depth of the tail tone Repairing components α-Irisone 0.15 0.28 Increase the complexity of floral fragrance Family Synergy Enhancement Repairing components Ethyl linoleate 2.42 0.22 Provides a sweet touch Improve softness Repairing components Ethyl dodecanoate 3.32 0.20 Provides fruity balance Modify the tail tone As shown in Table 2, although γ-iridin has a low peak area, it was identified as a core component due to its high similarity to the target fragrance and its typical iris characteristics. Tetradecanoic acid had the highest peak area, primarily serving as a waxy base and longevity skeleton, and also entered the core layer. α-iridin had a lower peak area, but was still identified as a modifying component due to its family synergistic effect. These results demonstrate that the method of this invention can distinguish between high-content structural support components and low-content highly typical components.

[0034] 6. Aroma reconstruction ratio range output Recommended proportion benchmarks for each output compound were calculated based on the overall aroma contribution value, and proportion ranges were given according to level. For core components, the recommended proportion range is 0.80 to 1.20 times the benchmark value; for auxiliary components, the recommended proportion range is 0.60 to 1.00 times the benchmark value; and for modifying components, the recommended proportion range is 0.20 to 0.60 times the benchmark value.

[0035] Example 2: Implementation of Multi-Source Data Fusion Building upon Example 1, in addition to GC-MS data, electronic nose data and gas chromatography-olfaction (GC-O) data were further introduced. GC-MS data was used as the chemical composition input, electronic nose data as the overall odor response input, and GC-O data as the local sensory intensity input. The data from different sources were normalized before being input into the artificial intelligence fusion model. Compared to using only GC-MS data, this implementation method can further improve the ability to identify substances with low concentrations of strong odors.

[0036] Example 3: Alternative Implementation Methods for Different Artificial Intelligence Models Based on Example 1, the artificial intelligence fusion model can be replaced by any of the following models or combinations thereof: a combination of a pre-trained semantic embedding model and a rule engine, a classification model based on gradient boosting trees, a classification model based on multilayer perceptrons, or a relational reasoning model based on graph neural networks. As long as it possesses the ability to fuse sample component data with an odor knowledge base and can output a comprehensive aroma contribution value and hierarchical classification results, it can be used to implement this invention.

[0037] Example 4: Simulation Comparison Verification Example To enhance the completeness of the explanation of the technical effects of this invention, a set of theoretically reasonable simulated comparative verification data was constructed without changing the sample composition logic of Example 1. This set of data is used to illustrate the potential advantages of the method of this invention compared with the simple peak area ranking method and the manual expert stratification method, and does not represent the results of actual experiments.

[0038] In this embodiment, the reference key component set is defined as consisting of GC-O results, odor activity analysis results, and literature-recognized typical iris components, containing eight representative compounds. A sensory panel of 12 trained olfactory evaluators is set up to score the similarity between the reconstructed sample and the target sample on a 10-point scale. Each sample is evaluated three times in parallel, and the average value is taken. The comparison indicators include: the first-pass yield rate with the reference key component set, the hierarchical consistency coefficient κ, the similarity between the reconstructed sample and the target sample, the number of blending rounds required to achieve a similarity score of not less than 8.0, the average resolution time for a single sample, and the coefficient of variation (CV) of the three batches.

[0039] Table 3 Simulation Comparison Verification Results Fatal rate / % compared to reference key component set 50.0 75.0 87.5 Hierarchical consistency coefficient κ 0.46 0.68 0.82 Similarity between the reconstructed sample fragrance and the target sample fragrance / score (out of 10) 6.4 7.9 8.8 Number of trial mating rounds / times required to achieve a similarity score of no less than 8.0 7 4 2 Average resolution time per sample / 95 60 18 Comparison indicators Simple peak area sorting method Stratification by human experts Method of the present invention min (minutes) Coefficient of variation (CV) of three batches of results 18.6 11.4 6.9 As shown in Table 3, under simulated conditions, the method of this invention outperforms both the simple peak area ranking method and the manual expert stratification method in terms of the one-fatal hit rate with the reference key component set, the hierarchical consistency coefficient κ, and the similarity between the reconstructed sample and the target sample. Furthermore, it is significantly superior in terms of the number of trial blending rounds required to achieve a similarity score of at least 8.0 and the average analysis time per sample. These results are consistent with the technical approach of this invention: this invention does not simply improve the ranking speed, but rather improves the accuracy of identifying key aroma components by integrating peak area, aroma activity, iris typicality, longevity support, and synergistic relationships.

[0040] Among them, the "simple peak area sorting method" is prone to missing low-content but highly typical substances such as γ-irisone and α-irisone; the "artificial expert stratification method" is highly dependent on experience and its efficiency and batch stability are relatively average; the method of this invention, by explicitly introducing iris typicality weights and synergistic correction terms, makes the determination of core components and modified components closer to the actual aroma structure of iris essential oil.

[0041] Example 5: Simulated Batch Robustness Example To verify the adaptability of the method of the present invention to the fluctuations of raw material batches, three theoretically reasonable simulated batches were constructed based on the peak areas of the key compounds in Example 1, where the peak areas of each key compound fluctuated within approximately ±10% of the results in Example 1. Keeping the knowledge base, weight coefficients, and hierarchical thresholds unchanged, the methods of the present invention were input, and the results are shown in Table 4.

[0042] Table 4. Robustness results of simulated batches γ-Irisone 4.59 4.21 4.84 All were identified as core ingredients Tetradecanoic acid 35.08 32.91 37.46 All were identified as core ingredients Apuxining 4.92 4.63 5.08 All were identified as core ingredients Ethyl tetradecanoate 10.59 9.88 11.14 All were identified as auxiliary ingredients. dodecanoic acid 11.43 10.96 12.10 All were identified as auxiliary ingredients. α-Irisone 0.15 0.13 0.17 All were identified as modifying ingredients. As shown in Table 4, although there are some fluctuations in peak area among different batches for γ-iridinone, tetradecanoic acid, apoxinine, ethyl tetradecanoate, dodecanoic acid, and α-iridinone, the hierarchical results output by the method of this invention remain stable. This indicates that the method of this invention is not sensitive to a single numerical ranking, but rather possesses a certain degree of batch robustness, which is beneficial for automatically correcting batch differences.

[0043] The above results are consistent with the theoretical mechanism of action of this invention: when the peak area fluctuates slightly, if the iris typicality, longevity support properties and synergistic relationship of the relevant compounds do not change substantially, the ranking of their overall aroma contribution values ​​will not be significantly reversed.

[0044] This invention provides a standardized, automated, and interpretable pathway from chemical detection results to aroma structure results, particularly suitable for complex natural fragrance systems such as iris essential oil, which contain both high-area fatty acids and low-content, highly typical irisones. Through the coupling of an aroma knowledge base and an artificial intelligence fusion model, this invention can not only output key aroma components but also further provide aroma reconstruction ratio ranges.

Claims

1. A method for rapid analysis of key aroma components of iris essential oil using artificial intelligence, characterized in that, Includes the following steps: S1. Obtain an iris essential oil sample and perform gas chromatography-mass spectrometry on the iris essential oil sample to obtain the retention time, retention index, peak area percentage, mass spectrometry matching degree and molecular information of the compound to be analyzed. S2 constructs a sample component dataset based on the detection results obtained in step S1, and establishes an odor knowledge base corresponding to each compound to be analyzed. The odor knowledge base includes at least the compound name, Chemical Abstracts Registry number, odor descriptor, fragrance category, odor threshold, iris typicality weight, longevity support weight, and synergistic / antagonistic relationship. S3 performs feature fusion between the sample component dataset and the odor knowledge base data to obtain the feature vector of each compound to be analyzed. S4 inputs the feature vector into the artificial intelligence fusion model, calculates the comprehensive aroma contribution value of each compound to be analyzed, and classifies each compound to be analyzed into core components, auxiliary components, modifying components or background components according to a preset threshold. S5 outputs key aroma composition results for iris essential oil; S6 outputs a recommended range of proportions for reconstructing the aroma of iris essential oil based on the comprehensive aroma contribution value of each compound to be analyzed. The following detection conditions were used in step S1: the gas chromatography column was a polar capillary column DB-HeavyWax; the injection volume was 1 μL; the injection port temperature was 250°C; split injection was used with a split ratio of 10:1; the carrier gas was helium with a flow rate of 1.0 mL / min; the temperature program was 40°C for 2 minutes, then increased to 260°C at 10°C / min and held for 6 minutes; the mass spectrometry detection used electron impact ionization with an ionization energy of 70 eV, a transfer line temperature of 230°C, an ion source temperature of 230°C, a quadrupole temperature of 150°C, an interface temperature of 250°C, and a mass-to-charge ratio scan range of 50–500 atomic mass units. In step S3, the feature vector X_i of each compound i to be analyzed is represented as: X_i=[P_i, O_i, T_i, B_i, S_i, Q_i, M_i]; Wherein, P_i is the peak area normalization value, O_i is the odor activity term, T_i is the iris typicality weight, B_i is the lingering aroma support weight, S_i is the similarity between the compound odor description vector and the iris target aroma vector, Q_i is the synergistic / antagonistic correction term, and M_i is the qualitative confidence term. The odor activity term O_i is calculated according to the following formula: O_i = ln[1 + P_i / (OT_i + θ)] Where O_i represents the odor activity term of the i-th compound, ln represents the natural logarithm function, P_i represents the peak area normalized value of the i-th compound, OT_i represents the odor threshold of the i-th compound after normalization, and θ represents the correction constant to prevent the denominator from being zero. In step S4, the overall aroma contribution value C_i of each compound i to be analyzed is calculated according to the following formula: C_i=αP_i+βO_i+γT_i+δB_i+εS_i+ζQ_i+ηM_i; Where C_i represents the overall aroma contribution value of the i-th compound, α, β, γ, δ, ε, ζ, and η are weighting coefficients, and α+β+γ+δ+ε+ζ+η=1.

2. The method for rapid analysis of key aroma components of iris essential oil using artificial intelligence according to claim 1, characterized in that, In step S1, retention index correction and qualitative screening are performed on the compounds; a retention index correction sequence is established using n-alkane standard samples, and compounds with mass spectrometry matching scores of not less than 80 points and retention index deviations not greater than a preset threshold are included in the sample composition dataset.

3. The method for rapid analysis of key aroma components of iris essential oil using artificial intelligence according to claim 1, characterized in that, In step S4, the compounds to be analyzed are hierarchically classified according to their comprehensive aroma contribution value C_i: When C_i ≥ 0.65, it is determined to be a core component; When 0.35 ≤ Ci < 0.65, it is determined to be an auxiliary component; When 0.15 ≤ C_i < 0.35, it is determined to be a modifying component; when C_i < 0.15, it is determined to be a background component.

4. The method for rapid analysis of key aroma components of iris essential oil using artificial intelligence according to claim 1, characterized in that, In step S6, the recommended proportion benchmark value R_i for each compound i to be analyzed is determined according to the following formula: R_i=C_i / ΣC_j Where R_i represents the recommended proportion baseline value of the i-th compound, and ΣC_j represents the sum of the comprehensive aroma contribution values ​​of all compounds selected for the output results; For core ingredients, the recommended ratio range is 0.80 to 1.20 times R_i; for auxiliary ingredients, the recommended ratio range is 0.60 to 1.00 times R_i; and for modifying ingredients, the recommended ratio range is 0.20 to 0.60 times R_i.

5. The method for rapid analysis of key aroma components of iris essential oil using artificial intelligence according to claim 1, characterized in that, In addition to GC-MS detection data, the input data in step S1 also includes one or more of the following: Data from electronic nose testing, gas chromatography-olfactometry (GC-OL) data, gas chromatography-time-of-flight mass spectrometry (GC-TOF-MS) data, and human sensory evaluation data.

6. A computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the method for rapid analysis of key aroma components of iris essential oil using artificial intelligence as described in any one of claims 1 to 5.

Images