Perfume manufacturing process parameter dynamic optimization method and system based on production big data

Abstract

The embodiment of the invention relates to the technical field of perfume manufacturing, and discloses a perfume manufacturing process parameter dynamic optimization method based on production big data, and the method comprises the steps: receiving the production big data from a plurality of heterogeneous data sources; dynamically calculating a pollution risk evaluation value required when the production of the first perfume is switched to the production of the second perfume in the production equipment based on the perfume formula compatibility data; processing the real-time process data detected by the sensor to obtain process state data; inputting the market demand data, the pollution risk evaluation value and the process state data into a trained production optimization model, and executing the production optimization model; and transmitting the production process parameter instruction to a production execution system for perfume production. According to the scheme of the embodiment of the invention, market changes, equipment conditions and process fluctuations can be responded in real time, and a traditional static and single-target production mode is converted into a dynamic, collaborative and self-adaptive intelligent manufacturing mode.

Description

Technical Field

[0001] This invention relates to the field of production management technology, specifically to a method and system for dynamic optimization of perfume manufacturing process parameters based on big data production. Background Technology

[0002] Currently, perfume production, especially in steps such as blending and aging, heavily relies on the experience of master craftsmen. Process parameters, such as stirring speed, time, and aging environment settings, are fixed and cannot be adaptively adjusted to minor fluctuations in raw material yields, making it difficult to achieve optimal quality. Furthermore, under the current circumstances, current perfume production methods are unable to adapt to rapid market changes. Summary of the Invention

[0003] To address the aforementioned shortcomings, this invention discloses a method for dynamically optimizing perfume manufacturing process parameters based on big data from production, which enables rapid optimization and adjustment of perfume production.

[0004] The first aspect of this invention discloses a method for dynamically optimizing perfume manufacturing process parameters based on big data production data, including: Production big data is received from multiple heterogeneous data sources, including market demand data related to the perfume to be produced, current production equipment status data, perfume formula compatibility data, and real-time process data detected by sensors. Based on the perfume formula compatibility data, dynamically calculate the pollution risk assessment value required when switching from producing the first perfume to producing the second perfume in the production equipment; The real-time process data detected by the sensors is processed to obtain process status data; The market demand data, the pollution risk assessment value, and the process status data are input into the trained production optimization model. The production optimization model is executed to generate and output dynamically optimized production process parameter instructions, which include production scheduling priority, equipment cleanliness level, and maturation process parameters. The production process parameter instructions are transmitted to the production execution system for perfume production.

[0005] As an optional implementation, in a first aspect of the present invention, before transmitting the production process parameter instructions to the production execution system for perfume production, the method further includes: The production process parameter instructions generated by the production optimization model are input into the digital twin simulation model of the perfume production line; The production process parameter instructions are simulated and executed in the digital twin simulation model, and key performance indicators are predicted, including expected delivery cycle, maturation tank resource occupation conflict and equipment comprehensive utilization rate. In response to the failure of the predicted key performance indicators to reach the optimization target, the production process parameter instructions are iteratively adjusted and re-simulated in the digital twin simulation model until the optimization target is reached. The maturation process parameters include the rotation speed and time parameters of the stirrer, the temperature parameters of the mixing tank, the temperature and humidity parameters of the maturation environment, and the intensity and time parameters of the specific frequency sound waves or electromagnetic waves applied to the perfume during maturation.

[0006] As an optional implementation, in the first aspect of the present invention, the market demand data is determined through the following steps: Obtain social media sentiment analysis indicators, online search trend indicators, and inventory turnover rate indicators related to the perfume to be produced from external data interfaces; Based on a predefined weighting model, the social media sentiment analysis index, the online search trend index, and the inventory turnover rate index are weighted and fused to generate a comprehensive market popularity index. The market popularity index is used as the market demand data.

[0007] As an optional implementation, in the first aspect of the present invention, processing the real-time process data detected by the sensor to obtain process state data includes: Three-dimensional spectral data of perfume during the aging process are acquired at a set frequency using an online gas chromatography-ion mobility spectrometer, and the characteristic peak intensities of multiple key volatile organic compounds are extracted to form a real-time chemical fingerprint vector characterizing the current chemical state of the perfume. For a specific perfume formula, time-series chemical fingerprint vectors and human sensory scores are simultaneously collected during its standard maturation process. A sensory prediction model is trained to establish a predictive relationship from chemical fingerprints to sensory scores. The target chemical fingerprint corresponding to the peak sensory score and its evolution trajectory are determined, and the digital olfactory maturity curve of the perfume is determined based on the target chemical fingerprint corresponding to the peak sensory score and its evolution trajectory. The sensory prediction model is a long short-term memory network or a temporal convolutional network. The human sensory scores are obtained by a trained fragrance evaluation team after conducting multi-dimensional sensory evaluations of blind samples. The evaluation dimensions include overall harmony and characteristic fragrance intensity. The digital olfactory maturity curve includes the target chemical fingerprint vector corresponding to the peak point of the sensory score curve predicted by the model, as well as the typical chemical fingerprint evolution path from the starting point to the peak point. The real-time acquired chemical fingerprint vector is mapped to the high-dimensional chemical space where the digital olfactory maturity curve is located, and its real-time trajectory deviation and trajectory convergence speed with the target trajectory are calculated, and the remaining time to reach the optimal sensory quality is predicted. Based on the real-time trajectory deviation and trajectory convergence speed, environmental fine-tuning instructions are dynamically generated and executed. The environmental fine-tuning instructions are matched with a preset control strategy library to generate adjustment instructions.

[0008] As an optional implementation, in the first aspect of the present invention, the optimization method further includes: The maturation process is divided into multiple maturation sub-stages that are connected in sequence over time, with each sub-stage corresponding to a different molecular association process; Based on electronic nose and near-infrared spectral data, the association state index of each ripening sub-stage is evaluated, wherein the evaluation of non-initiation sub-stages is coupled with the index of upstream sub-stages. When the sensory score of the final matured product is insufficient, the system traces back the contribution of the association state index of each sub-stage to the final score. Identify the key lag sub-stages that contribute the most and match them with optimization strategies: if the environmental parameters are the main cause of the stage, then initiate a dynamic adjustment strategy for the environmental parameters; if the problem is the initial mixing uniformity of the raw materials, then feed back information to the mixing stage and trigger a calibration strategy for the mixing process parameters.

[0009] As an optional implementation, in a first aspect of the present invention, the production optimization model includes a trained process parameter generation model, which is trained through the following steps: Obtain the natural language process target text corresponding to the sensory style description of the target product in the historical successful production batches, as well as the validated optimal process parameter set finally adopted in that batch; The natural language process target text is subjected to structured parsing and feature extraction to generate a standard process target feature vector, which encodes the process focus corresponding to the corresponding sensory style; A mapping pair is established between the standard process target feature vector and the verified optimal process parameter set to form a process knowledge feature base, which is used to train the process parameter generation model, so that the model learns the mapping relationship from the semantics of the process target to the specific parameters.

[0010] As an optional implementation, in the first aspect of the present invention, the optimization method further includes: Receive natural language production requirement text input by the user, wherein the natural language production requirement text includes a description of sensory goals, a description of market positioning, or a description of artistic inspiration; The natural language production demand text is cleaned and normalized, and the professional sensory description words in it are identified and mapped to predefined standardized aroma descriptors. Complete the implicit process constraints; The complex requirements are analyzed and broken down to generate a structured, standardized process requirement vector that can be processed by the model.

[0011] As an optional implementation, in the first aspect of the present invention, the optimization method further includes: The market popularity index of odor characteristics is calculated from multi-source data, and the future odor demand spectrum is predicted based on the popularity index; The odor demand spectrum is converted into a corresponding state feature vector, and then matched with the standard process target feature vector that is most similar to the target feature vector. The corresponding production process parameter instructions are determined based on the matched standard feature vectors.

[0012] As an optional implementation, in a first aspect of the present invention, the matching of the standard process target feature vector most similar to the state feature vector includes: Determine the state feature vector of the odor demand spectrum; The state feature vector is matched with multiple pre-classified successful production pattern clusters in the historical database; each successful production pattern cluster is formed by clustering the state vectors of multiple historical successful batches in the corresponding odor state, and is characterized by the cluster's center vector and distribution covariance matrix. Calculate the first matching distance between the state feature vector and each successful pattern cluster; the first matching distance is the Mahalanobis distance. Based on the first matching distance, the candidate successful pattern cluster with the smallest distance is selected; Within the selected candidate successful pattern clusters, calculate the second matching distance between the state feature vector and the state vector of each historical successful batch in the cluster. The second matching distance is Euclidean distance or dynamic time warping distance. The corresponding standard process target feature vector is determined based on the two matching results.

[0013] A second aspect of the present invention discloses a computer-readable storage medium storing a computer program, wherein the computer program causes a computer to execute the method for dynamic optimization of perfume manufacturing process parameters based on production big data disclosed in the first aspect of the present invention.

[0014] Compared with the prior art, the embodiments of the present invention have the following beneficial effects: The method in this invention integrates and processes heterogeneous big data from multiple sources, including market demand, equipment status, formula compatibility, and real-time processes, which were originally isolated or relied on human experience. It then uses a trained model for comprehensive analysis, ultimately outputting joint optimization instructions for multiple key process parameters such as production scheduling, cleanliness levels, and maturation processes. This enables perfume production to respond in real-time to market changes, equipment conditions, and process fluctuations, transforming it from a traditional static, single-objective production model into a dynamic, collaborative, and adaptive intelligent manufacturing model. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a flowchart illustrating the method for dynamic optimization of perfume manufacturing process parameters based on production big data disclosed in an embodiment of the present invention. Figure 2 This is a schematic diagram of the simulation process disclosed in the embodiments of the present invention; Figure 3 This is a schematic diagram of the process for obtaining market demand data disclosed in an embodiment of the present invention; Figure 4 This is a schematic diagram of the vehicle passage status processing disclosed in an embodiment of the present invention; Figure 5 This is a schematic diagram of the structure of a perfume manufacturing process parameter dynamic optimization system based on production big data provided in an embodiment of the present invention; Figure 6 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] It should be noted that the terms first, second, third, fourth, etc., in the specification and claims of this invention are used to distinguish different objects, not to describe a specific order. The terms used in the embodiments of this invention include and have, and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to these processes, methods, products, or devices.

[0019] Example 1 Please see Figure 1 , Figure 1 This is a flowchart illustrating the dynamic optimization method for perfume manufacturing process parameters based on big data production, as disclosed in this embodiment of the invention. The execution entity of the method described in this embodiment is an execution entity composed of software and / or hardware. This execution entity can receive relevant information via wired or / or wireless means and can send certain instructions. It may also have certain processing and storage functions. This execution entity can control multiple devices, such as remote physical servers or cloud servers and related software, or local hosts or servers and related software that perform related operations on devices located in a certain location. In some scenarios, multiple storage devices can also be controlled; these storage devices may be placed in the same location as the devices or in different locations. Figure 1 As shown, this method for dynamically optimizing perfume manufacturing process parameters based on big data production includes the following steps: S101: Receive production big data from multiple heterogeneous data sources, wherein the production big data includes market demand data related to the perfume to be produced, current production equipment status data, perfume formula compatibility data, and real-time process data detected by sensors; S102: Dynamically calculate the pollution risk assessment value required when switching from producing the first perfume to producing the second perfume in the production equipment based on the perfume formula compatibility data; S103: Process the real-time process data detected by the sensor to obtain process status data; S104: Input the market demand data, the pollution risk assessment value and the process status data into the trained production optimization model, execute the production optimization model to generate and output dynamically optimized production process parameter instructions, the production process parameter instructions including production scheduling priority, equipment cleanliness level and maturation process parameters; S105: Transmit the production process parameter instructions to the production execution system to produce perfume.

[0020] Specifically, the solution in this invention introduces and quantifies perfume formula compatibility data and pollution risk assessment values. This addresses the industry pain point of cross-contamination caused by residues in perfume manufacturing. By dynamically assessing risks through a model and precisely matching equipment cleanliness levels, it is possible to minimize over-cleaning while ensuring product quality and safety, achieving an optimal balance between quality risk control and the conservation of production resources (time, materials, and energy).

[0021] The production optimization model in this embodiment of the invention is a reinforcement learning model, wherein: the state space of the reinforcement learning model includes feature vectors representing production micro-states extracted from the real-time process data; The reward function of the reinforcement learning model is a multi-objective weighted function based on the predicted final product sensory attribute score, physicochemical index compliance, and production energy consumption cost. Executing the production optimization model includes: based on the current state, the agent of the reinforcement learning model selects an action that maximizes the expected cumulative reward, the action corresponding to the dynamically optimized production process parameter instructions.

[0022] When calculating the pollution risk value, the following method is used: First, an odor molecule database is consulted to obtain the adsorption coefficient data of the first perfume's formulation ingredients on the inner wall material of the production equipment. Based on the adsorption coefficient data and the concentration of the first perfume's components, the theoretical residual concentration is calculated. Then, a sensory threshold database is consulted to obtain the human olfactory threshold of the target ingredient in the second perfume. Finally, based on the ratio of the theoretical residual concentration to the human olfactory threshold, the pollution risk assessment value is calculated. This method achieves a concrete quantitative assessment.

[0023] Furthermore, after outputting instructions for adjusting the equipment cleaning level, the method further includes: Receive post-cleaning detection data from an online odor detection sensor connected to the production equipment; The post-cleaning test data is compared with a preset cleanliness standard threshold. In response to the determination that the post-cleaning test data does not meet the cleanliness standard threshold, an upgraded cleaning instruction is automatically triggered and output until the post-cleaning test data meets the cleanliness standard threshold. This method significantly improves management efficiency, and in practical implementation, advance formula compatibility analysis allows the production line to more quickly quantify the differences in production using different formulas.

[0024] More preferably, such as Figure 2 As shown, before transmitting the production process parameter instructions to the production execution system for perfume production, the process further includes: S1041: Input the production process parameter instructions generated by the production optimization model into the digital twin simulation model of the perfume production line; S1042: Simulate the execution of the production process parameter instructions in the digital twin simulation model and predict key performance indicators, including expected delivery cycle, maturation tank resource occupation conflict and equipment comprehensive utilization rate; S1043: In response to the failure of the predicted key performance indicators to reach the optimization target, the production process parameter instructions are iteratively adjusted and re-simulated in the digital twin simulation model until the optimization target is reached. The maturation process parameters include the rotation speed and time parameters of the stirrer, the temperature parameters of the mixing tank, the temperature and humidity parameters of the maturation environment, and the intensity and time parameters of the specific frequency sound wave or electromagnetic wave applied to the perfume during maturation.

[0025] Before production orders are actually issued to the physical production line, simulations are performed in a digital twin model. This allows for the early exposure and assessment of potential conflicts and risks that optimized solutions may cause in complex production environments (such as multi-tasking parallelism and resource sharing), such as conflicts over the use of aging tank resources and equipment overload. This effectively avoids on-site production interruptions, scheduling chaos, or quality accidents caused by model prediction biases or unconsidered constraints, transferring decision-making risks from the physical world to the virtual space. The digital twin model can simulate the dynamic operation of the entire perfume production line, thus predicting system-level comprehensive performance indicators (such as expected delivery time and equipment utilization rate). This makes the optimization objective no longer simply the optimization of single process parameters, but rather a comprehensive optimization that ensures that global objectives such as meeting delivery time, maximizing equipment utilization, and avoiding resource conflicts are met. Through simulation, the impact of different parameter combinations on global objectives can be explored in a virtual environment, leading to the finding of the optimal solution for the system.

[0026] Simulation models can accurately predict expected delivery cycles, enabling production planning to be more forward-looking and better meet market demands. Simultaneously, through the collaborative simulation and optimization of multi-dimensional and refined parameters such as stirrer speed and time, temperature and humidity, and sound / electromagnetic wave intensity and time, the coupling mechanism of these parameters' influence on the aging effect of perfume can be deeply revealed. This allows for more precise process control than simply relying on historical data models, improving the reproducibility of product quality.

[0027] In addition to the methods mentioned above, since there are corresponding recognition models for odors, data twins can be used to efficiently simulate a large number of different odor combinations to obtain different optimization results, thereby improving the corresponding knowledge base and the final result prediction.

[0028] More preferably, such as Figure 3 As shown, the market demand data is determined through the following steps: S1011: Obtain social media sentiment analysis indicators, online search trend indicators, and inventory turnover rate indicators related to the perfume to be produced from an external data interface; S1012: Based on a predefined weighting model, the social media sentiment analysis index, the online search trend index, and the inventory turnover rate index are weighted and fused to generate a comprehensive market popularity index; S1013: Use the market popularity index as the market demand data.

[0029] Traditional market demand data typically relies on historical sales orders or channel inventory, which is significantly lagging. The solution in this invention introduces social media sentiment analysis indicators and online search trend indicators, which are digital representations of real-time consumer behavior and potential interests. This enables the system to capture early signals and future trends in market demand, allowing for advance planning in production arrangements, predictive production scheduling, and better response to a rapidly changing market.

[0030] Specifically, the embodiments of the present invention construct a multi-dimensional, highly sensitive quantitative perception system for market heat.

[0031] Social media sentiment analysis reflects consumers' emotional inclinations (positive, negative, and level of discussion) towards perfume brands, fragrance types, and concepts. It is a soft indicator of product reputation and potential acceptance.

[0032] Online search trends reflect the intensity of consumers' interest in actively seeking information and are a direct quantitative indicator of the emergence and growth potential of market demand.

[0033] Inventory turnover rate reflects the actual sales speed of existing products in the sales channels and is the most direct and hard verification indicator of market demand. By weighting and integrating these three types of indicators (sentiment, interest, and actual sales), the resulting market popularity index can more comprehensively, three-dimensionally, and sensitively depict the real and dynamic state of market demand than a single indicator (such as sales volume).

[0034] Market demand data determined by this market heat index is of higher quality and more timely. When it is used as a key input into subsequent production optimization models, it can drive the models to adjust production scheduling priorities more quickly and accurately. For example, it can prioritize the production of perfumes whose market demand is rapidly increasing, or reserve capacity for upcoming hot trends, thereby shortening the reaction time from market signals to production actions and improving supply chain agility.

[0035] More preferably, the process of processing the real-time process data detected by the sensor to obtain process state data includes: S1031: At a set frequency, the three-dimensional spectral data of the perfume during the aging process are acquired by an online gas chromatograph-ion mobility spectrometer, and the characteristic peak intensities of multiple key volatile organic compounds are extracted to form a real-time chemical fingerprint vector characterizing the current chemical state of the perfume. S1032: For a specific perfume formula, the time-series chemical fingerprint vector and human sensory scores during its standard maturation process are collected simultaneously. A sensory prediction model is trained to establish a predictive relationship from chemical fingerprint to sensory score. The target chemical fingerprint corresponding to the peak sensory score and its evolution trajectory are determined, and the digital olfactory maturity curve of the perfume is determined based on the target chemical fingerprint corresponding to the peak sensory score and its evolution trajectory. The sensory prediction model is a long short-term memory network or a temporal convolutional network. The human sensory scores are obtained by a trained fragrance evaluation team after conducting multi-dimensional sensory evaluations of blind samples. The evaluation dimensions include overall harmony and characteristic fragrance intensity. The digital olfactory maturity curve includes the target chemical fingerprint vector corresponding to the peak point of the sensory score curve predicted by the model, as well as the typical chemical fingerprint evolution path from the starting point to the peak point. S1033: Map the real-time acquired chemical fingerprint vector to the high-dimensional chemical space where the digital olfactory maturity curve is located, calculate its real-time trajectory deviation and trajectory convergence speed from the target trajectory, and predict the remaining time to reach the optimal sensory quality. S1034: Based on the real-time trajectory deviation and trajectory convergence speed, dynamically generate and execute environmental fine-tuning instructions; S1035: Match the environmental fine-tuning command with a preset control strategy library to generate an adjustment command.

[0036] The solution presented in this invention establishes a predictive bridge from chemical composition to sensory quality, achieving an optimized closed loop aimed at the end-user experience. By training sensory prediction models (such as LSTM and TCN), the system can directly predict theoretical human sensory scores from real-time chemical fingerprints. This solves the core industry problem of the disconnect between objective instrument data and subjective quality evaluation in the perfume aging process, which is highly dependent on sensory assessment. The optimization objective is thus clearly defined as approaching the peak sensory score, rather than simply achieving a certain chemical indicator or maintaining it for a certain physical time, truly realizing process control guided by the sensory quality of the final product.

[0037] The digital olfactory maturity curve proposed in this embodiment of the invention provides a dynamic and personalized optimal reference trajectory for the maturation process. This curve is not a fixed function of time, but rather the optimal evolutionary path of a specific perfume formula in chemical space. It includes the target chemical fingerprint at which sensory peaks are reached, as well as the typical evolutionary path to that target. This allows the system to provide a tailored optimal maturation navigation map for each perfume and each batch of production.

[0038] The solution in this invention calculates the real-time trajectory deviation and convergence speed between the real-time chemical fingerprint vector and the target trajectory of the digital olfactory maturity curve. This not only determines whether the current state is off track but also predicts the remaining time to reach optimal quality. Based on this, the system can dynamically generate and execute environmental fine-tuning instructions, thereby correcting deviations and accelerating or stabilizing the convergence process in real time, ensuring that each batch of products can efficiently and consistently reach peak sensory quality.

[0039] The above methods significantly improve the consistency, efficiency, and traceability of the aging process. Traditional aging relies on fixed durations and periodic sampling by perfumers, which carries risks of large batch variations, unstable efficiency, and over- or under-aging. This solution, through real-time, objective chemical fingerprint monitoring and closed-loop control based on predictive models, can significantly reduce batch-to-batch variations and terminate aging immediately when optimal quality is achieved, avoiding unnecessary wasted aging time (improving efficiency) or quality defects caused by premature termination (ensuring consistency). All chemical fingerprint data and control instructions are recorded, achieving a fully digitalized and traceable quality archive.

[0040] The solutions presented in this invention provide a powerful data-driven tool for process innovation and new product development. The accumulated chemical fingerprint-sensory score mapping data and the effectiveness data of different regulatory strategies (regulation strategy library) constitute a valuable process knowledge base. This can be used to accelerate the development of maturation processes for new formulations, rapidly determining their digital olfactory maturity curves and optimal regulatory parameters through simulation and optimization.

[0041] Specifically, in the baseline learning phase, the solution of this invention automatically samples and analyzes samples from a standard curing tank with high temporal resolution.

[0042] From raw spectra to feature vectors: For GC-IMS data, after preprocessing (baseline correction, peak alignment, normalization), all detectable peaks are identified. Each peak is defined by its coordinates in the two-dimensional space of retention time and drift time (i.e., the identity feature of the compound) and its intensity (concentration characterization).

[0043] Construct a fingerprint vector by selecting n key feature peaks (e.g., n = 50-200) that appear stably and significantly throughout the maturation process. The chemical fingerprint at each time point is an n-dimensional vector F(t): F(t) = [I1(t), I2(t), ..., I... n [(t)], where I x (t) is the normalized intensity of the x-th characteristic peak at time t.

[0044] In addition to single-peak intensity, the intensity ratios of key peaks (e.g., esters / alcohols, aldehydes / acids) can also be calculated during implementation. These ratios reflect states such as esterification, hydrolysis, and oxidation, and are more chemically significant than single concentrations.

[0045] A high-dimensional odor space is constructed, treating the chemical fingerprint vector F(t) at each time point as a point in an n-dimensional space (each feature peak being one dimension). Dimensionality reduction techniques (such as principal component analysis (PCA) or t-SNE) can project this n-dimensional space onto a 2- or 3-dimensional visualization space. In this projected space, the entire ripening process appears as a continuous trajectory, which is the evolution trajectory of the chemical fingerprint.

[0046] The modeling of the digital olfactory maturity curve links the evolution of chemical fingerprints with the optimal maturity point of human senses. The specific implementation involves sensory scoring data collection, where a rigorously trained olfactory evaluation team conducts blind evaluations of the same batch of samples at the same time point as the chemical fingerprint collection. Evaluation dimensions include overall harmony, characteristic aroma intensity, and intensity of unpleasant odors, ultimately summarizing into a comprehensive maturity score S(t).

[0047] Establish a correlation model: The core issue is that the relationship between chemical fingerprint F(t) and sensory score S(t) is nonlinear and has a lag.

[0048] Model selection: A supervised machine learning model is used to establish the mapping between F(t) and S(t).

[0049] Specifically, recurrent neural networks (RNNs, such as LSTM) or temporal convolutional networks (TCNs) are used for mapping. These models are better able to capture the temporal dynamic dependencies in chemical fingerprint sequences, meaning that the current odor depends not only on the current chemical composition but also on previous evolutionary paths.

[0050] Define a target maturity curve. After model training, for a standard maturity process, the system can generate a curve showing the predicted sensory rating over time. The peak point (or peak plateau region) on this curve is defined as the optimal maturity point.

[0051] In the odor space of chemical fingerprints, the chemical fingerprint F corresponding to this optimal maturity point is... optimal and to reach F optimal The typical chemical evolution trajectory it has undergone collectively constitutes the digital olfactory maturity curve of this fragrance. This curve is both a target point and an ideal reference path.

[0052] In practice, real-time chemical fingerprints (F) are acquired at the same frequency for each new batch of aging tanks. now .

[0053] F now Projecting onto the inherent scent space of the perfume, calculate F. now Reference point F corresponding to the development stage on the target maturity curve (based on trajectory similarity matching) ref The distance between them. This distance can be Euclidean distance or a weighted distance considering the chemical importance of each dimension. Using a time series model, based on F(t) at the most recent time points, the trajectory in the near future and the estimated remaining time to reach the optimal point are predicted. A perturbation strategy library is built into the system, which triggers different physical field fine-tuning based on the type and degree of prediction deviation.

[0054] Specifically, an example of control logic: Scenario 1: Deviation from the target trajectory (e.g., esterification reaction is too slow, resulting in a high acidity). Judgment: The key ester / acid ratio is lower than the reference trajectory; Action: Slightly reduce ambient humidity (e.g., reduce RH by 2%). This slightly shifts the hydrolysis / esterification balance towards esterification.

[0055] Scenario 2: Slow convergence (sluggish overall response kinetics); Judgment: The predicted remaining time is significantly higher than the historical baseline; Action: Activate low-frequency, low-intensity ultrasound in pulsed mode. Utilize its cavitation effect to promote molecular mixing and mass transfer without causing a significant temperature rise.

[0056] Scenario 3: An unexpected oxidation pathway occurs (abnormally high aldehyde levels); Judgment: The peak intensity of a specific aldehyde deviates abnormally; Action: Adjust the LED light of a specific wavelength inside the curing tank. This method enables efficient control.

[0057] More preferably, the optimization method further includes: The maturation process is divided into multiple maturation sub-stages that are connected in sequence over time, with each sub-stage corresponding to a different molecular association process; Based on electronic nose and near-infrared spectral data, the association state index of each ripening sub-stage is evaluated, wherein the evaluation of non-initiation sub-stages is coupled with the index of upstream sub-stages. When the sensory score of the final matured product is insufficient, the system traces back the contribution of the association state index of each sub-stage to the final score. Identify the key lag sub-stages that contribute the most and match them with optimization strategies: if the environmental parameters are the main cause of the stage, then initiate a dynamic adjustment strategy for the environmental parameters; if the problem is the initial mixing uniformity of the raw materials, then feed back information to the mixing stage and trigger a calibration strategy for the mixing process parameters.

[0058] The solution of this invention divides the maturation process, which was originally considered a black box or a single stage, into multiple sequential sub-stages based on the theoretical process of molecular association. This aligns with the physicochemical nature of perfume maturation (involving volatilization, oxidation, esterification, changes in intermolecular forces, etc.), enabling the system to move from monitoring the overall state to monitoring the stages of the process, providing a more refined dimension for understanding and controlling the process.

[0059] The system employs a data fusion approach, combining electronic nose (for sensing the overall odor profile) and near-infrared spectroscopy (for analyzing chemical components and intermolecular interactions), to assess the association state index of each sub-stage. This fusion integrates the overall odor signal with deep chemical structure information, making the state assessment more comprehensive and reliable. The design of coupling upstream indices with the assessment of non-initial sub-stages reflects the transitivity and cumulative effects between stages, making the model more consistent with physical reality. When the final product's sensory score fails to meet the standards, the system does not simply make overall adjustments but can reverse-analyze the contribution of the association state index of each sub-stage to the final score. This is similar to establishing a digital quality fault tree for the production process, enabling rapid and quantitative location of key weak links (critical lag sub-stages) leading to quality problems, pinpointing the problem from the curing process to the Xth sub-stage of the curing process.

[0060] This system implements differentiated and precise compensation strategies based on root cause diagnosis, improving the targeting and effectiveness of optimization. It can not only identify key lag stages but also match differentiated optimization strategies according to the characteristics of each stage. If the problem mainly stems from poor control of environmental parameters (such as temperature, humidity, sound waves / electromagnetic fields) in this stage, then a dynamic adjustment strategy for this stage will be initiated for compensation. If the root cause of the problem is insufficient initial uniformity of raw materials in the upstream mixing stage, then diagnostic information will be fed back to the mixing process to trigger the calibration of the mixing process.

[0061] More preferably, the production optimization model includes a trained process parameter generation model, which is trained through the following steps: Obtain the natural language process target text corresponding to the sensory style description of the target product in the historical successful production batches, as well as the validated optimal process parameter set finally adopted in that batch; The natural language process target text is subjected to structured parsing and feature extraction to generate a standard process target feature vector, which encodes the process focus corresponding to the corresponding sensory style; A mapping pair is established between the standard process target feature vector and the verified optimal process parameter set to form a process knowledge feature base, which is used to train the process parameter generation model, so that the model learns the mapping relationship from the semantics of the process target to the specific parameters.

[0062] Traditional process parameter settings rely heavily on the personal experience of the craftsman, making them difficult to convey precisely using structured language. The solution in this invention parses natural language process target text—for example, the desired fresher top notes while maintaining a rich woody finish—and transforms this into a standard process target feature vector. This allows the model to understand the artistic and ambiguous sensory style appeals described by humans in natural language. This significantly lowers the barrier to entry, enabling non-technical experts (such as perfumers and marketing personnel) to directly participate in setting production goals through natural language.

[0063] When a perfume needs to be produced with a new sensory style description, the system can quickly generate a set of initial process parameters based on a trained model and the new natural language target text, drawing on successful precedents. This is equivalent to providing an intelligent process parameter generator for new product development, significantly shortening the cycle from new product design to determining the production process, and accelerating product innovation and time-to-market.

[0064] Specifically, GC-MS data standardization involves performing the same preprocessing on all batches of data: peak alignment, baseline correction, and normalization (e.g., using an internal standard as a reference).

[0065] The selection of key characteristic peaks does not involve using the entire spectrum (curse of dimensionality), but rather uses statistical analysis (such as PCA and ANOVA) to identify 50-200 characteristic peaks that exhibit significant batch-to-batch variation and have a known contribution to the final aroma. For example, for rose essential oil, the peak areas of 20-30 key molecules, such as phenylethyl alcohol, citronellol, geraniol, rose ether, and daumatones, are tracked.

[0066] Vectorization uses the relative percentage content (or ratio to the internal standard) of each characteristic peak as one dimension of the vector. Ultimately, the chemical vector P of a batch of raw materials is an n-dimensional vector: P = [C1, C2, ..., C...]. n ].

[0067] Metadata embedding uses a lightweight text encoder (such as BERT) to process unstructured text, extract key information such as year, production area, and extraction method, and encode it into an independent metadata vector M.

[0068] The chemical vector P is concatenated with the metadata vector M or fused through a neural network to form the final chemical passport vector P containing multimodal information. final .

[0069] Raw material-formulation-process knowledge graph construction, graph definition: Node type: Raw material batch (Attribute: Chemical Passport P) finalFormula version (attributes: percentage of each component), process parameter set (attributes: mixing speed, temperature, time, etc.), product batch (attributes: final sensory score, chemical fingerprint).

[0070] Relationship type: Used for (raw materials, formula), Adopted for (formula, process), Output for (process, product), Similar to (raw material batches, based on P) final Cosine similarity.

[0071] Construction method: Structured data is extracted from the historical production management system and imported into a graph database (such as Neo4j or Amazon Neptune) using ETL (Extract, Transform, Load) tools. Each historical production record constitutes a connected subgraph in the graph.

[0072] Generative recipe compensation (a professional LLM implementation) is the brain of the system. It is not a general-purpose LLM, but a highly specialized and finely tuned domain model. It is based on open-source foundational LLMs (such as Llama3 and Qwen) for domain-adaptive pre-training and instruction fine-tuning.

[0073] Training data construction: Classic perfumery works and fragrance chemistry literature were transformed into text corpora for further pre-training, enabling the model to grasp basic associations such as phenylethanol bringing a sweet rose scent and patchouli adding an earthy feel.

[0074] Construct instruction-response pairs from knowledge graphs.

[0075] Instructions: Raw material chemical passport: Phenylacetyl alcohol content = baseline value * 0.85, turkeyone content = baseline value * 1.08. Requirements: Full, dark rose hue with woody and amber undertones.

[0076] Ideal response (from historical successful adjustment records): Analysis shows that the decrease in phenylethyl alcohol results in insufficient sweetness, while the increase in rufatone brings a stronger rose character and fruity aroma. To maintain fullness, sweetness and volume need to be added; to enhance darkness, woody and animalic notes need to be strengthened.

[0077] Calculate the percentage difference between the target fragrance and the standard ingredient passport across key dimensions. Combine this difference, the target fragrance description, and the current baseline formula into structured prompts, which are then fed into a professional LLM (Liquidity, Mechanics, and Management) system. The LLM utilizes its internally learned chemistry-sensory-formulation association knowledge to output multiple fine-tuning schemes described in natural language. A post-processor parses these natural language schemes (e.g., adding 0.05% patchouli oil) into executable formula adjustment instructions and automatically calculates the total amount and cost of the new formula.

[0078] In the knowledge graph, using the property vectors and raw material passports of the new formulation as query criteria, the most similar formulation-process node pairs in history are sought. Similarity calculation can combine Euclidean distance of property data and raw material passport similarity. Process parameters (such as mixing rate, homogenization pressure, and aging start temperature) used in the most similar historical records are extracted as recommended initial values ​​for this production run. After the batch production is completed, its final product chemical fingerprint, sensory evaluation, and actual process parameters are added to the knowledge graph as new nodes and relationships.

[0079] Meanwhile, the complete decision-making chain of this successful raw material difference, formula adjustment, process adjustment, and good results has been constructed into a new instruction-response pair, added to the fine-tuning training set of professional LLM, and the model is updated regularly to make it increasingly intelligent.

[0080] The process parameter generation model includes: maintaining an expandable process case map internally, with nodes including: sensory description features, raw material chemical features, equipment status features, process parameter set, and final quality score; When a new standardized process requirement vector is received, the model searches for the historical case node with the most similar sensory description features in the case graph. Instead of directly copying the process parameter set of the historical case, it analyzes the relationships around the case node and infers them by analogy to the current new raw material and equipment context to generate a new set of adapted process parameters.

[0081] Specifically, input: sensory evaluation of the new batch, chemical fingerprint of the batch, and process log of the batch.

[0082] Process: The system automatically converts chemical fingerprints and process logs into descriptive text, combines it with sensory evaluations, and inputs it into the trained LLM.

[0083] Output: LLM directly generates a natural language report: Analysis: The insufficient burst of citrus flavor may stem from two factors: First, the chemical spectrum shows a low peak area for top-tone monoterpenes (such as limonene); second, the process log shows that the mixing stage temperature was 2°C higher than standard, which may have caused these volatile components to dissipate prematurely. Recommendations: For the next batch: Lower the mixing start temperature from 25°C to 23°C; check the batch consistency of top-tone flavor ingredients; add some citrus essential oils in the later stages of mixing to reduce heat exposure time.

[0084] Predictive patterns (sensory perception based on chemistry and process): Input: Chemical fingerprint of a certain batch, and planned process parameters.

[0085] Process: After converting to descriptive text, ask the LLM: Based on this chemical composition and process setting, predict what main sensory evaluations the tasting panel might give? Output: LLM generation prediction: Predicted sensory characteristics: May have a strong sweet fruity aroma (due to high ester peaks), but may be accompanied by a cooked taste or a lack of freshness (due to the high-temperature rapid stirring process potentially causing some esters to hydrolyze and produce a sour taste, or a heat-deteriorated aroma). Risk warning: It is recommended to review the stirring speed and temperature settings.

[0086] Creative Optimization Model (Deriving Solutions from Objectives): Input: Target sensory description (e.g., to add a touch of damp forest scent to this fragrance while maintaining its existing elegance).

[0087] Process: The LLM calls upon its internal knowledge (from the training data) to perform the following operations: Deconstruction objective: To understand the odor molecules corresponding to the scent of a damp forest (such as earthy xylene musk, damp ozone or marine aldehydes, vegetal leaf alcohol, and woody oakmoss / cedar extract).

[0088] Constraint Analysis: Maintain the existing elegance and avoid using materials that are too rough or disrupt the balance.

[0089] Production Scheme: Output: To achieve a moist forest scent, consider the following safe additions: add oakmoss absolute to the base notes (to enhance the moist woody and earthy feel), and add gabbonate to the middle notes (to bring a green, moist forest scent). Processing Warning: Both of these substances are sensitive to light and heat; ensure strict protection from light during aging and storage, and it is recommended to mix them at low temperatures.

[0090] More preferably, the optimization method further includes: Receive natural language production requirement text input by the user, wherein the natural language production requirement text includes a description of sensory goals, a description of market positioning, or a description of artistic inspiration; The natural language production demand text is cleaned and normalized, and the professional sensory description words in it are identified and mapped to predefined standardized aroma descriptors. Complete the implicit process constraints; specifically, when described as maintaining a natural feel, automatically associate process rules to avoid high-temperature processing; The complex requirements are analyzed and broken down to generate a structured, standardized process requirement vector that can be processed by the model.

[0091] The solution of this invention allows users (such as perfumers, brand managers, and marketing personnel) to directly express their production needs using natural language (e.g., "I need a limited-edition summer perfume with the scent of fresh grass after rain and a touch of mysterious smokiness"). This enables artistic creations and business ideas to be transformed into production instructions in the most direct and expressive way, significantly improving the system's usability and human-machine collaboration efficiency.

[0092] Different users have subjective differences in their understanding of terms such as "fresh," "rich," and "fruity." The solution in this invention establishes a unified sensory vocabulary by identifying and mapping professional sensory terms to predefined standardized aroma descriptors. This ensures that regardless of how the user expresses their meaning, their core intent can be interpreted unambiguously and consistently by the system, laying the foundation for subsequent precise process matching and guaranteeing product quality and design intent. Figure 1 A key aspect of consistency.

[0093] Users may only focus on sensory goals, ignoring production feasibility or cost constraints. The system can automatically complete implicit and reasonable process constraints based on the knowledge base (such as inferring the volatility of fresh fragrances based on summer limited editions, thus implying the need for fragrance enhancement processes).

[0094] Complex natural language descriptions are often a mixture of multiple requirements. The system can parse and decompose these requirements into independent, structured sub-requirements (such as decomposing the grassy scent intensity, smoky characteristics, and overall freshness from the example above), and generate a standardized process requirement vector. This allows subsequent model processing and parameter optimization to be performed on each specific dimension, achieving a refined response to complex requirements.

[0095] More preferably, the optimization method further includes: The market popularity index of odor characteristics is calculated from multi-source data, and the future odor demand spectrum is predicted based on the popularity index; The odor demand spectrum is converted into a corresponding state feature vector, and then matched with the standard process target feature vector that is most similar to the target feature vector. The corresponding production process parameter instructions are determined based on the matched standard feature vectors.

[0096] Traditional market forecasting typically focuses on the sales trends of specific products (such as a particular perfume). The solution presented in this invention delves into the scent characteristics of perfumes (such as fragrance notes, floral notes, woody notes, freshness, sweetness, and longevity), calculating market popularity indices for these dimensions using multi-source data and predicting future scent demand patterns. This allows for the capture of more fundamental and abstract changes in market preferences; for example, predicting the increasing popularity of blends of aquatic and leathery notes, rather than focusing on which specific perfumes combining these two notes will be bestsellers.

[0097] By converting the predicted abstract odor demand spectrum into a state feature vector and matching it with a standard process target feature vector library, the known production process path that best achieves the odor characteristic can be quickly identified. This allows companies to rapidly translate predicted market trends into specific, executable production process parameter instructions, thereby gaining a first-mover advantage by launching products that align with future trends to the market.

[0098] The odor demand spectrum predicted by the system may represent an emerging market preference that has not yet been fully met or combined by existing products. By matching it to known processes, process solutions for new products can be proactively generated, which not only meets the predicted demand but also guides and creates new demand.

[0099] By predicting future scent trends, the system can provide early warnings and pre-planning for key raw materials (such as specific synthetic fragrances or natural extracts) and production resources (such as production lines specializing in a particular fragrance) required for specific scent characteristics at an earlier stage. This helps companies prepare their supply chains and optimize production schedules in advance, smoothing out production fluctuations and reducing inventory or supply shortage risks caused by sudden market shifts.

[0100] Through the above method, the solution of this invention no longer passively waits for sales data, but actively predicts the market demand changes for specific scent characteristics (such as tranquil wood, summer fruit tea, retro powder) from real-time information such as social media and search trends, and automatically generates or adjusts perfume formulas and production processes to match the predicted demand.

[0101] Specifically, construct an aroma profile: define a multi-dimensional sensory description space, which may include: floral-fruity-woody fragrance families, fresh-rich (intensity), sweet-spicy (emotion), simple-complex (structure), etc. Each dimension has a quantifiable scale.

[0102] Training the predictive model: Using a time series predictive model (such as LSTM, Transformer), the input is a sequence of market popularity indexes for each dimension over the past N periods, and the output is the predicted scent demand spectrum for the next M periods—that is, the expected intensity of each dimension in the sensory description space. For example, the model might predict: In the next quarter, the demand for tranquil woody tea will increase (woody, 2; fresh, 1; complex, -1), and the demand for energizing fruit tea will remain high (fruity, 2; sweet, 1; fresh, 2).

[0103] Demand Matching: The predicted scent demand spectrum is matched against the digital sensory profiles (defined by historical reviews and chemical fingerprints) of each fragrance in the company's existing fragrance database. The 1-3 base fragrances that best match are identified.

[0104] Intelligent Fine-tuning: Invokes a professional LLM (Liquid Management Model). The instruction is: Using perfume A as a base, make its scent spectrum closer to the target desired spectrum (e.g., increase woodiness from 7 to 9, decrease sweetness from 5 to 3). Please generate a formula adjustment plan. The LLM will output specific formula modifications based on its chemical knowledge.

[0105] The solution in this invention combines market demand forecasts with the production preparation time and raw material inventory of new formulas to provide dynamic prioritization suggestions for the production planning system. For example, if the demand for Tranquil Wood is predicted to peak in 4 weeks, and the production cycle is 2 weeks, the system suggests starting production of the formula 2 weeks later to ensure sufficient supply during peak demand.

[0106] The specific working method of the solution in this embodiment of the invention is as follows: Monday morning: The system automatically generated a weekly report: Based on data from the past week, discussions about topics related to fresh grass after rain and the scent of ink and books increased by 40% week-on-week, with extremely positive sentiment. It is predicted that demand for the fresh, green scent and the aroma of dry paper will rise in the coming month.

[0107] Perfumer / Product Manager: Confirm the predicted trend in the system interface and click to generate a concept formula.

[0108] The system searched the solution library and found an existing green tea perfume that matched the fresh green notes, but lacked the dry paper texture. It then invoked a generative model to generate a draft new formula that added violet leaf absolute (greenish notes) and a small amount of iris balsam (powdery, papery texture) to the original green tea base, along with process adjustment suggestions (due to the difficulty in dissolving iris balsam, it is recommended to extend the stirring time).

[0109] Perfumer: Review and fine-tune the AI-generated formula to ensure it meets artistic standards.

[0110] Production system: Upon receiving confirmed new formulas and process packages, it automatically checks raw material inventory and suggests arranging small-batch trial production 10 days later to catch up with the predicted demand window.

[0111] More preferably, the matching of the standard process target feature vector most similar to the state feature vector includes: Determine the state feature vector of the odor demand spectrum; The state feature vector is matched with multiple pre-classified successful production pattern clusters in the historical database; each successful production pattern cluster is formed by clustering the state vectors of multiple historical successful batches in the corresponding odor state, and is characterized by the cluster's center vector and distribution covariance matrix. Calculate the first matching distance between the state feature vector and each successful pattern cluster; the first matching distance is the Mahalanobis distance. Based on the first matching distance, the candidate successful pattern cluster with the smallest distance is selected; Within the selected candidate successful pattern clusters, calculate the second matching distance between the state feature vector and the state vector of each historical successful batch in the cluster. The second matching distance is Euclidean distance or dynamic time warping distance. The corresponding standard process target feature vector is determined based on the two matching results.

[0112] The solution in this invention pre-clusters historical successful batches into multiple successful production mode clusters based on their aroma state vectors, and characterizes these clusters using a center vector and a covariance matrix (representing the shape and direction of data distribution within the cluster). The system first performs screening at the macro-mode level. Mahalanobis distance is used to calculate the matching degree, considering not only the distance to the cluster center but, more importantly, the directionality of the data distribution. This makes the matching insensitive to correlated feature dimensions (e.g., sweetness and fruit aroma intensity often change synergistically) but sensitive to independently changing key dimensions, thus enabling more accurate and stable identification of historical experience clusters belonging to the same production mode or process paradigm as the target aroma spectrum.

[0113] After identifying the macroscopic clusters of candidate successful patterns, instead of simply using the average parameters of this cluster, we further calculate a second matching distance (using Euclidean distance or Dynamic Time Warping (DTW) distance) between the state vectors of each specific historical successful batch and the cluster itself. Euclidean distance is suitable for precise comparison of static vectors, while DTW distance is particularly suitable for comparing vectors with temporal evolution sequence characteristics (if the state vectors contain time series information), and can better match the dynamic ripening process trajectory, thereby finding the specific successful batch with the most similar evolutionary path.

[0114] The aforementioned two-layer matching mechanism is more refined and reliable than a simple global nearest neighbor search. It avoids the risks of accidental matches or being misled by abnormal batches that may occur when directly comparing with all historical batches. By searching for the most similar individual cases in the most relevant experience pool, it ensures that the recommended process parameters are based on the most relevant and similar successful precedents, thereby improving the reliability and success rate of the generated parameters.

[0115] Specifically, the similarity between vectors in the vector set is pre-calculated, and the similarity relationships between vectors are maintained in the form of a search tree. Each standard process target feature vector serves as a leaf of the search tree, i.e., a node; leaves that are close together are connected to form subtrees. The first matching distance is calculated by measuring the distance between the current node and all subtrees. Then, the distance between the node and the currently stored leaves of each subtree is further calculated, which yields the second matching distance. Based on the weighted distance, the features (nodes) collected this time are assigned to a specific subtree.

[0116] In this embodiment of the invention, the subtree is a set of similar feature vectors as described in the embodiment, and the number of subtrees may be one or more. If multiple subtrees exist, the closest set of similar feature vectors is selected based on the calculated weighted distance to form a new chemical fingerprint vector feature classification search tree. Subsequently, the entry root node of the subtree is updated according to the second matching distance.

[0117] The method in this invention integrates and processes heterogeneous big data from multiple sources, including market demand, equipment status, formula compatibility, and real-time processes, which were originally isolated or relied on human experience. It then uses a trained model for comprehensive analysis, ultimately outputting joint optimization instructions for multiple key process parameters such as production scheduling, cleanliness levels, and maturation processes. This enables perfume production to respond in real-time to market changes, equipment conditions, and process fluctuations, transforming it from a traditional static, single-objective production model into a dynamic, collaborative, and adaptive intelligent manufacturing model.

[0118] Example 2 Please see Figure 5 , Figure 5 This is a schematic diagram of the structure of the perfume manufacturing process parameter dynamic optimization system based on production big data disclosed in an embodiment of the present invention. Figure 5 As shown, this perfume manufacturing process parameter dynamic optimization system based on big data production can include: Receiving module 21: used to receive production big data from multiple heterogeneous data sources, wherein the production big data includes market demand data related to the perfume to be produced, current production equipment status data, perfume formula compatibility data, and real-time process data detected by sensors; Risk assessment module 22: used to dynamically calculate the pollution risk assessment value required when switching from producing the first perfume to producing the second perfume in the production equipment based on the perfume formula compatibility data; Processing module 23: Used to process the real-time process data detected by the sensor to obtain process status data; Generation module 24: is used to input the market demand data, the pollution risk assessment value and the process status data into the trained production optimization model, execute the production optimization model to generate and output dynamically optimized production process parameter instructions, the production process parameter instructions including production scheduling priority, equipment cleanliness level and maturation process parameters; Execution module 25: used to transmit the production process parameter instructions to the production execution system for perfume production.

[0119] Example 3 Please see Figure 6 , Figure 6This is a schematic diagram of the structure of an electronic device disclosed in an embodiment of the present invention. The electronic device can be a computer, a server, etc. Of course, in certain cases, it can also be a mobile phone, tablet computer, monitoring terminal, or other smart device, as well as an image acquisition device with processing capabilities. Figure 6 As shown, the electronic device may include: Memory 510 storing executable program code; Processor 520 coupled to memory 510; The processor 520 calls the executable program code stored in the memory 510 to execute some or all of the steps in the method for dynamic optimization of perfume manufacturing process parameters based on production big data in Embodiment 1.

[0120] This invention discloses a computer-readable storage medium storing a computer program that enables a computer to execute some or all of the steps in the method for dynamically optimizing perfume manufacturing process parameters based on production big data in Embodiment 1.

[0121] This invention also discloses a computer program product, wherein when the computer program product is run on a computer, the computer executes some or all of the steps in the method for dynamic optimization of perfume manufacturing process parameters based on production big data in Embodiment 1.

[0122] This invention also discloses an application publishing platform, which is used to publish computer program products. When the computer program products are run on a computer, the computer executes some or all of the steps in the method for dynamically optimizing perfume manufacturing process parameters based on production big data in Embodiment 1.

[0123] In various embodiments of the present invention, it should be understood that the sequence number of each process does not necessarily imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0124] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; they can be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0125] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0126] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-accessible memory. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a memory and includes several requests to cause a computer device (which can be a personal computer, server, or network device, specifically a processor in the computer device) to execute some or all of the steps of the methods described in the various embodiments of the present invention.

[0127] In the embodiments provided by this invention, it should be understood that B corresponding to A means that B is associated with A, and B can be determined based on A. However, it should also be understood that determining B based on A does not mean determining B solely based on A; B can also be determined based on A and / or other information.

[0128] Those skilled in the art will understand that some or all of the steps in the various methods of the embodiments described can be implemented by a program instructing related hardware. This program can be stored in a computer-readable storage medium, including read-only memory (ROM), random access memory (RAM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), one-time programmable read-only memory (OTPROM), electrically-Erasable Programmable Read-Only Memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disc storage, disk storage, magnetic tape storage, or any other computer-readable medium capable of carrying or storing data.

[0129] The foregoing has provided a detailed description of the method, system, electronic device, and storage medium for dynamic optimization of perfume manufacturing process parameters based on production big data disclosed in the embodiments of the present invention. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for dynamic optimization of perfume manufacturing process parameters based on big data production data, characterized in that, include: Production big data is received from multiple heterogeneous data sources, including market demand data related to the perfume to be produced, current production equipment status data, perfume formula compatibility data, and real-time process data detected by sensors. Based on the perfume formula compatibility data, dynamically calculate the pollution risk assessment value required when switching from producing the first perfume to producing the second perfume in the production equipment; The real-time process data detected by the sensors is processed to obtain process status data; The market demand data, the pollution risk assessment value, and the process status data are input into the trained production optimization model. The production optimization model is executed to generate and output dynamically optimized production process parameter instructions, which include production scheduling priority, equipment cleanliness level, and maturation process parameters. The production process parameter instructions are transmitted to the production execution system for perfume production.

2. The method for dynamic optimization of perfume manufacturing process parameters based on production big data as described in claim 1, characterized in that, Before transmitting the production process parameter instructions to the production execution system for perfume production, the method further includes: The production process parameter instructions generated by the production optimization model are input into the digital twin simulation model of the perfume production line; The production process parameter instructions are simulated and executed in the digital twin simulation model, and key performance indicators are predicted, including expected delivery cycle, maturation tank resource occupation conflict and equipment comprehensive utilization rate. In response to the failure of the predicted key performance indicators to reach the optimization target, the production process parameter instructions are iteratively adjusted and re-simulated in the digital twin simulation model until the optimization target is reached. The maturation process parameters include the rotation speed and time parameters of the stirrer, the temperature parameters of the mixing tank, the temperature and humidity parameters of the maturation environment, and the intensity and time parameters of the specific frequency sound waves or electromagnetic waves applied to the perfume during maturation.

3. The method for dynamic optimization of perfume manufacturing process parameters based on production big data as described in claim 1, characterized in that, The market demand data was determined through the following steps: Obtain social media sentiment analysis indicators, online search trend indicators, and inventory turnover rate indicators related to the perfume to be produced from external data interfaces; Based on a predefined weighting model, the social media sentiment analysis index, the online search trend index, and the inventory turnover rate index are weighted and fused to generate a comprehensive market popularity index. The market popularity index is used as the market demand data.

4. The method for dynamic optimization of perfume manufacturing process parameters based on production big data as described in claim 1, characterized in that, The process of processing real-time process data detected by the sensors to obtain process status data includes: Three-dimensional spectral data of perfume during the aging process are acquired at a set frequency using an online gas chromatography-ion mobility spectrometer, and the characteristic peak intensities of multiple key volatile organic compounds are extracted to form a real-time chemical fingerprint vector characterizing the current chemical state of the perfume. For a specific perfume formula, time-series chemical fingerprint vectors and human sensory scores are simultaneously collected during its standard maturation process. A sensory prediction model is trained to establish a predictive relationship from chemical fingerprints to sensory scores. The target chemical fingerprint corresponding to the peak sensory score and its evolution trajectory are determined, and the digital olfactory maturity curve of the perfume is determined based on the target chemical fingerprint corresponding to the peak sensory score and its evolution trajectory. The sensory prediction model is a long short-term memory network or a temporal convolutional network. The human sensory scores are obtained by a trained fragrance evaluation team after conducting multi-dimensional sensory evaluations of blind samples. The evaluation dimensions include overall harmony and characteristic fragrance intensity. The digital olfactory maturity curve includes the target chemical fingerprint vector corresponding to the peak point of the sensory score curve predicted by the model, as well as the typical chemical fingerprint evolution path from the starting point to the peak point. The real-time acquired chemical fingerprint vector is mapped to the high-dimensional chemical space where the digital olfactory maturity curve is located, and its real-time trajectory deviation and trajectory convergence speed with the target trajectory are calculated, and the remaining time to reach the optimal sensory quality is predicted. Based on the real-time trajectory deviation and trajectory convergence speed, environmental fine-tuning instructions are dynamically generated and executed. The environmental fine-tuning instructions are matched with a preset control strategy library to generate adjustment instructions.

5. The method for dynamic optimization of perfume manufacturing process parameters based on big data production as described in claim 4, characterized in that, The optimization method further includes: The maturation process is divided into multiple maturation sub-stages that are connected in sequence over time, with each sub-stage corresponding to a different molecular association process; Based on electronic nose and near-infrared spectral data, the association state index of each ripening sub-stage is evaluated, wherein the evaluation of non-initiation sub-stages is coupled with the index of upstream sub-stages. When the sensory score of the final matured product is insufficient, the system traces back the contribution of the association state index of each sub-stage to the final score. Identify the key lag sub-stages that contribute the most and match them with optimization strategies: if the environmental parameters are the main cause of the stage, then initiate a dynamic adjustment strategy for the environmental parameters; if the problem is the initial mixing uniformity of the raw materials, then feed back information to the mixing stage and trigger a calibration strategy for the mixing process parameters.

6. The method for dynamic optimization of perfume manufacturing process parameters based on big data production as described in claim 1, characterized in that, The production optimization model includes a trained process parameter generation model, which is trained through the following steps: Obtain the natural language process target text corresponding to the sensory style description of the target product in the historical successful production batches, as well as the validated optimal process parameter set finally adopted in that batch; The natural language process target text is subjected to structured parsing and feature extraction to generate a standard process target feature vector, which encodes the process focus corresponding to the corresponding sensory style; A mapping pair is established between the standard process target feature vector and the verified optimal process parameter set to form a process knowledge feature base, which is used to train the process parameter generation model, so that the model learns the mapping relationship from the semantics of the process target to the specific parameters.

7. The method for dynamic optimization of perfume manufacturing process parameters based on production big data as described in claim 6, characterized in that, The optimization method further includes: Receive natural language production requirement text input by the user, wherein the natural language production requirement text includes a description of sensory goals, a description of market positioning, or a description of artistic inspiration; The natural language production demand text is cleaned and normalized, and the professional sensory description words in it are identified and mapped to predefined standardized aroma descriptors. Complete the implicit process constraints; The complex requirements are analyzed and broken down to generate a structured, standardized process requirement vector that can be processed by the model.

8. The method for dynamic optimization of perfume manufacturing process parameters based on production big data as described in claim 6, characterized in that, The optimization method further includes: The market popularity index of odor characteristics is calculated from multi-source data, and the future odor demand spectrum is predicted based on the popularity index; The odor demand spectrum is converted into a corresponding state feature vector, and then matched with the standard process target feature vector that is most similar to the target feature vector. The corresponding production process parameter instructions are determined based on the matched standard feature vectors.

9. The method for dynamic optimization of perfume manufacturing process parameters based on production big data as described in claim 8, characterized in that, The matching of the standard process target feature vector most similar to the state feature vector includes: Determine the state feature vector of the odor demand spectrum; The state feature vector is matched with multiple pre-classified successful production pattern clusters in the historical database; each successful production pattern cluster is formed by clustering the state vectors of multiple historical successful batches in the corresponding odor state, and is characterized by the cluster's center vector and distribution covariance matrix. Calculate the first matching distance between the state feature vector and each successful pattern cluster; the first matching distance is the Mahalanobis distance. Based on the first matching distance, the candidate successful pattern cluster with the smallest distance is selected; Within the selected candidate successful pattern clusters, calculate the second matching distance between the state feature vector and the state vector of each historical successful batch in the cluster. The second matching distance is Euclidean distance or dynamic time warping distance. The corresponding standard process target feature vector is determined based on the two matching results.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, wherein the computer program causes a computer to execute the method for dynamic optimization of perfume manufacturing process parameters based on big data production, as described in any one of claims 1 to 9.

Images