An intelligent agarwood incense system and method based on electronic nose feedback and large language model collaborative control

Abstract

The application discloses an intelligent eaglewood incense system and method based on electronic nose feedback and large language model collaborative control. The system includes an eaglewood storage module, an automatic material adding module, a heating module, an electronic nose detection module, an environment detection module, a main control module, an aroma release kinetics model module, a large language model decision module, a digital twin module, an automatic ash removal module, and a user preference learning module. The electronic nose detection module detects the aroma concentration and odor characteristic information in real time. The environment detection module collects temperature, humidity, and air flow parameters. The main control module combines the aroma release kinetics model and the digital twin prediction results to perform closed-loop control on the heating temperature and the material adding amount. The system can achieve precise incense control, natural interaction, automatic material adding, automatic ash removal, and predictive compensation control during the eaglewood incense process, reducing the frequency of manual maintenance, and improving the stability, intelligence, and personalized experience of eaglewood incense.

Description

Technical Field

[0001] This invention belongs to the field of intelligent incense equipment and smart home technology, specifically relating to an intelligent agarwood incense system and method based on electronic nasal feedback and large language model collaborative control. Background Technology

[0002] Agarwood, a precious natural fragrance, holds a central position in the incense culture of China, Japan, and Southeast Asia. During the heating and burning of agarwood, different temperature ranges release different characteristic aroma components: the low-temperature range of 120℃~150℃ releases light volatile components, mainly sesquiterpenes; the medium-temperature range of 180℃~220℃ releases the core agarwood aroma, mainly aromatic compounds; and the high-temperature range above 250℃ produces a burnt and combustible aroma. Traditional constant-temperature heating cannot reveal the multi-layered aroma changes of agarwood, while manual temperature control based on experience is difficult to guarantee consistency and repeatability.

[0003] In recent years, some technological explorations have emerged in the field of intelligent incense equipment. Regarding electronic noses and closed-loop aroma control, patent publication number "CN120631095A" discloses an adaptive temperature control method and system for incense machines based on multi-source feedback. This method relies on a pre-established knowledge base and lookup table control, lacking adaptability to unknown agarwood varieties or mixed incense materials, and does not address automatic material addition and ash removal. Patent publication number "CN107490621A" discloses an aroma reproduction device. This patent proposes an aroma reproduction closed-loop scheme based on a PID controller, but it is designed for general aroma scenarios and does not optimize for the unique multi-stage aroma release curve of agarwood, lacking mechanical linkage for automatic material addition and ash removal.

[0004] In the field of Large Language Modeling (LLM) and fragrance control, patent publication number "CN121806645A" discloses an interactive control method and device for an electric incense burner based on an AI model. However, it only maps simple voice commands to temperature adjustment and lacks a deep understanding of abstract semantics such as "sleep mode" and "meditation mode," as well as closed-loop feedback from an electronic nose and quantitative detection of fragrance concentration. Patent publication number "CN121978978A" discloses a multi-channel intelligent fragrance control method and system based on a large language model. This patent describes a multi-channel intelligent fragrance control method based on a large language model, but focuses on multi-channel fragrance mixing and pruning algorithm optimization, without involving multi-stage temperature control strategies and fragrance release kinetic modeling during solid agarwood heating.

[0005] In the field of agarwood heating and automatic cleaning, the patent with announcement number "CN204970761U" discloses an automatic incense burning machine that automatically transports incense to the combustion chamber and collects ash through a transfer mechanism. However, it is designed for traditional open flame combustion rather than electric heating, and is a step-by-step whole-stick replacement, which cannot achieve continuous micro-addition adjustment, and there is no aroma concentration detection and closed-loop feedback.

[0006] In the field of aroma release kinetic modeling, publication number "CN121541509A" discloses a method and system for controlling the release of fragrance from an electric iron. It uses the Arrhenius equation to calculate the aroma heat sensitivity coefficient, but it is for fabric fragrance rather than agarwood incense, does not consider the step release characteristics of agarwood multi-components at different temperatures, and does not involve electronic nose concentration feedback.

[0007] In summary, existing technologies have conducted independent research in areas such as "electronic nose closed-loop control", "large language model interaction", "automatic material addition and ash removal", and "heating temperature control", but there is still no solution for an agarwood incense system that organically integrates the five elements of aroma release kinetic model, electronic nose closed-loop feedback, large language model semantic understanding, automatic material addition and ash removal and digital twin predictive control. Furthermore, existing agarwood incense equipment generally suffers from the following shortcomings: (1) Low fragrance control accuracy, most of which are timed and temperature-controlled heating, or rely solely on temperature feedback, lacking direct quantitative detection and closed-loop control of fragrance concentration; (2) Unnatural interaction, the operation of buttons or remote controls cannot understand abstract semantic requirements such as "sleep mode", "fragrance lighter", "suitable for tea tasting"; (3) Frequent manual maintenance, requiring manual addition of materials and cleaning of ash residue, making it impossible to achieve long-term unattended continuous operation; (4) Lack of predictive ability, the fragrance release has a large lag characteristic, existing systems can only passively respond to concentration changes, and cannot predict the fragrance decay trend in advance and actively compensate; (5) Insufficient agarwood adaptability, the fragrance release of agarwood has significant multi-component step release characteristics, and existing general temperature control strategies do not consider this special law.

[0008] In view of the problems existing in the prior art, how to provide an intelligent agarwood incense system and method based on electronic nose feedback and large language model collaborative control, so as to solve the technical problems of low fragrance control accuracy, unnatural interaction, frequent manual maintenance, lack of predictive ability and insufficient agarwood adaptability of existing agarwood incense equipment, is the problem that this invention urgently needs to solve. Summary of the Invention

[0009] The purpose of this invention is to propose an intelligent agarwood incense system and method based on electronic nose feedback and large language model collaborative control. This system uses a multi-sensor electronic nose array to detect the concentration of volatile organic compounds (VOCs) in real time as a closed-loop feedback signal. It combines this with an aroma release kinetic model for state estimation and trend prediction, uses a large language model as a semantic understanding engine to achieve natural language interaction, and automates operation and maintenance through automatic material feeding and ash removal mechanisms. A composite control strategy of "predictive feedforward + real-time feedback" overcomes the large lag in aroma release, ultimately forming a complete intelligent closed loop of "perception → prediction → decision → execution → learning." This solves the technical problems of existing agarwood incense equipment, such as low aroma control accuracy, unnatural interaction, frequent manual maintenance, lack of predictive ability, and insufficient agarwood compatibility.

[0010] To achieve the above objectives, the technical solution adopted by the present invention is as follows: The present invention provides an intelligent agarwood incense system based on electronic nose feedback and large language model collaborative control, including an agarwood storage module, an automatic feeding module, a heating module, an electronic nose detection module, an environmental detection module, a main control module, an aroma release kinetic model module, a large language model decision-making module, a digital twin module, an automatic ash removal module, and a user preference learning module.

[0011] The agarwood storage module is used to store agarwood raw materials and to supply materials to the automatic feeding module;

[0012] The automatic feeding module is used to quantitatively feed agarwood raw materials to the heating module according to the control instructions of the main control module;

[0013] The heating module is used to heat the agarwood raw material, causing the agarwood raw material to release its aroma;

[0014] The electronic nose detection module is used to detect aroma concentration and odor characteristic information, and feeds the detection results back to the main control module.

[0015] The environmental detection module is used to detect ambient temperature, humidity, and air circulation.

[0016] The main control module is used to control the operation of the automatic feeding module, heating module and automatic dust removal module based on information from the electronic nose detection module, environmental detection module, aroma release kinetic model module, large language model decision module, digital twin module and user preference learning module;

[0017] The aroma release kinetics model module is used to estimate the current aroma state and predict the future trend of aroma concentration change.

[0018] The large language model decision module is used to parse user natural language commands into structured target aroma parameters;

[0019] The digital twin module is used to construct a virtual incense system and predict its future operating status;

[0020] The automatic ash removal module is used to clean the ash and slag produced after heating.

[0021] The user preference learning module is used to record user usage data and correct aroma control parameters.

[0022] As a further improvement, the agarwood storage module includes a hopper, a discharge port, and a vibrator. The discharge port is connected to an automatic feeding module, and the vibrator is used to prevent the agarwood raw materials from clogging.

[0023] As a further improvement, the automatic feeding module includes a screw conveyor mechanism, a stepper motor, and a weight sensor. The stepper motor is used to drive the screw conveyor mechanism to feed the material quantitatively, and the weight sensor is used to detect the remaining material. The automatic feeding module uses the screw conveyor mechanism as the quantitative feeding execution element, which is driven by the stepper motor. The quantitative feeding is achieved by controlling the motor speed, and the control accuracy can reach ±0.05g. The built-in weight sensor detects the remaining material in real time.

[0024] As a further improvement, the heating module includes a heating plate, a ceramic heater, and a temperature sensor, and performs phased heating control according to the preheating stage, the aroma release stage, the aroma maintenance stage, and the aroma preservation stage.

[0025] As a further improvement, the heating module includes a heating plate, a ceramic heater, and a temperature sensor. The heating temperature range is 50℃~350℃. The controller controls the heating power according to a four-stage temperature curve, namely a preheating stage of 120℃~150℃, an aroma release stage of 180℃~220℃, a constant aroma stage of 190℃±5℃, and an aroma preservation stage of 140℃~170℃.

[0026] As a further improvement, the electronic nose detection module includes multiple VOC gas sensors for collecting VOC concentration, CO2 concentration, temperature and humidity, and odor feature vectors. The main control module identifies the current heating stage or judges the risk of scorching based on the odor feature vectors.

[0027] As a further improvement, the electronic nose detection module is installed in the aromatherapy chamber or exhaust channel, and includes multiple VOC gas sensors of different types, such as BME688, SGP30, CCS811, and MQ135. It samples every 10 seconds, collecting total VOC concentration, CO2 concentration, temperature and humidity, and odor feature vectors, denoted as... ,in Let be the normalized response value of the i-th sensor channel.

[0028] As a further improvement, the aroma release kinetic model module outputs the current aroma concentration estimate, the future predicted concentration, the recommended amount of additive, and the recommended temperature adjustment amount based on the heating temperature, the remaining mass of agarwood, the ambient humidity, and historical VOC concentration data.

[0029] As a further improvement, the digital twin module is used to predict the future trend of aroma concentration change. When the predicted concentration deviates from the target concentration by more than a preset range, the main control module controls the automatic feeding module to replenish the material or controls the heating module to adjust the temperature in advance.

[0030] The environmental monitoring module is used to detect the temperature, humidity, and air circulation parameters of the environment where the incense burner is located, as compensation inputs for the control model.

[0031] The main control module, as the decision-making center of the system, receives real-time data from the electronic nose detection module to calculate the current aroma concentration, runs the aroma release kinetic model to estimate the current state and predict future trends, receives the target aroma parameters generated by the large language model decision module, compares the deviation between the actual concentration and the target concentration to generate control commands, and interrupts the normal control process and forces cooling when S(t) triggers a scorching warning.

[0032] The aroma release kinetics model module includes release kinetics and concentration prediction. The release kinetics formula describes the volatility of agarwood fragrance itself, with the core formula being:

[0033] ;

[0034] Where t is the cumulative running time of the spices from the start of heating. VOC concentration released from the surface of the fragrance , Current heating temperature , For the remaining mass of agarwood , This is the temperature influence coefficient. This is the quality influence coefficient. This is a proportionality coefficient, related to the type and particle size of agarwood. The pyrolysis attenuation coefficient is... This indicates that the aroma release rate decreases over time under constant temperature conditions.

[0035] Based on the release kinetic formula, the process is divided into three steps. Predicted concentration converted to sensor location Step 1: Humidity correction. ,in , The relative humidity of the environment (%) For reference humidity, This is a humidity correction factor, approximately Higher humidity means higher water vapor content in the air, slower VOC molecule diffusion, and higher local concentration near the sensor. Follow Monotonically increasing; Step 2 introduces the rate of temperature change. Calculate the rate of concentration change This reflects the dynamic impact of temperature rises and falls on concentration trends. The current concentration is after humidity correction, T is the current heating temperature, and M is the current remaining material quantity. For quality influencing factors, For the rate of temperature change, The rate of change of material quantity, where Directly determined by the feeding action; Step 3: Extrapolation prediction: To obtain the future Predicted concentration at time, The time difference between the target concentration prediction time and the current time is used; simultaneously, historical VOC concentration data is continuously compared with the electronic nose's measured values ​​to... , , , The four parameters are calibrated using least squares, and the model's final output is: the current concentration estimate. rate of concentration change Future predicted concentration Recommended feed amount and recommended temperature adjustment .

[0036] Recommended amount of feed and recommended temperature adjustment Calculations were made by working backwards from the concentration gap: That is, when the predicted concentration is lower than the target, the amount of material to be added is calculated based on the ratio. This means that when the predicted concentration is lower than the target, the required temperature increase is calculated based on the ratio.

[0037] For the amount of material added, This represents the current remaining material quantity. For the target concentration, To predict concentration, For factors affecting quality; For temperature adjustment amount, The current temperature. This is the temperature influence coefficient.

[0038] The large language model decision module is equipped with a microphone, speaker, and speech recognition engine. Users express their aroma preferences in natural language, such as "a lighter aroma," "I want a meditation mode," "suitable for tea," "last for two hours," or "a bit strong now." After speech recognition, the text is sent to the large language model inference engine, which parses the fuzzy semantics into structured target aroma concentration, aroma duration, heating rate, temperature range, and addition frequency. For example, "sleep mode" is mapped to: target temperature 150℃, target concentration approximately 30ppm, duration 8 hours; "meeting mode" is mapped to: target temperature 210℃, target concentration approximately 80ppm; and "a lighter aroma" is mapped to: at the current target concentration... Reduce by 10% to 20% on the basis.

[0039] The digital twin module constructs a virtual replica of the incense system in real time, mapping the remaining amount of agarwood. Temperature status VOC concentration Ambient humidity Based on user preference parameters and running synchronously with the physical system, the digital twin model uses a three-step process based on a concentration prediction model to predict concentration change trends over the next 30 minutes, 1 hour, and 2 hours. When the predicted concentration deviates from the target concentration When the deviation exceeds the allowable range of ±30%, the system automatically triggers feedforward compensation. Intervene before the deviation actually occurs.

[0040] The automatic ash removal module is located at the bottom of the combustion chamber and includes an ash pan, scraper, drive motor, and ash collection box. It automatically triggers the ash removal operation when the continuous operating time exceeds 4 hours or the aroma release efficiency decreases by more than 20%. The determination method for the decrease in aroma release efficiency is as follows: when… Time-triggered dust cleaning At the same temperature and remaining mass The theoretical concentration below this value indicates that the effective volatilization of the spice is hindered due to the ash covering the spice surface.

[0041] The user preference learning module continuously records user commands, operating parameters, aroma concentration, and adjustment records to build a personal aroma database for each user, automatically corrects the control parameters of each mode, and achieves personalized adaptive adjustment.

[0042] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0043] (1) Precise aroma control: The VOC concentration is detected in real time by the electronic nose detection module as a closed-loop feedback signal. Combined with the incremental PID controller, the heating temperature and the amount of feed are dynamically adjusted so that the actual aroma concentration is continuously stable within the target concentration range. The aroma control accuracy is significantly better than the traditional open-loop timed temperature control method.

[0044] (2) Natural interaction: By introducing a large language model as a semantic understanding engine, users only need to express their needs in natural language to complete the automatic mapping of abstract semantics such as "sleep mode", "tea tasting mode" and "lighter aroma" to multi-dimensional control parameters, which greatly reduces the user's operating threshold.

[0045] (3) Long-term unattended operation: When the running time is long or the aroma efficiency decreases, the automatic feeding mechanism and the automatic ash removal mechanism are automatically triggered. Under the unified scheduling of the closed-loop control algorithm, continuous operation is achieved without manual intervention.

[0046] (4) Predictive control to avoid overshoot: The digital twin module predicts the trend of aroma change in advance and adopts a combined feedforward and feedback control strategy. The compensation operation is started before the aroma concentration deviates from the target, which effectively overcomes the large lag characteristic of aroma release and eliminates the problem of control overshoot.

[0047] (5) Agarwood-specific adaptation: The four-stage heating strategy based on the aroma release kinetic model fully demonstrates the multi-component step release characteristics of agarwood in different temperature ranges, and the user experience is better than constant temperature heating.

[0048] (6) Personalized Adaptation: The user preference learning module continuously accumulates operational data and automatically corrects the parameters of each mode to achieve an adaptive and personalized experience. Attached Figure Description

[0049] Figure 1 This is a schematic diagram of the structure of an intelligent agarwood incense system based on the collaborative control of electronic nasal feedback and a large language model according to the present invention.

[0050] Figure 2 This is a block diagram of the overall structure of an intelligent agarwood incense system based on the collaborative control of electronic nose feedback and large language model according to the present invention.

[0051] Figure 3 This is a closed-loop control flowchart of an intelligent agarwood incense burning method based on electronic nose feedback and large language model collaborative control according to the present invention.

[0052] In the diagram, 1-Agarwood storage module, 2-Automatic feeding module, 3-Heating module, 4-Electronic nose detection module, 5-Environmental detection module, 6-Main control module, 7-Aroma release kinetic model module, 8-Large language model decision module, 9-Digital twin module, 10-Automatic dust removal module, 11-User preference learning module. Detailed Implementation

[0053] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments:

[0054] like Figure 1-2 As shown, the intelligent agarwood incense system based on electronic nose feedback and large language model collaborative control of the present invention consists of eleven parts: an agarwood storage module 1, an automatic feeding module 2, a heating module 3, an electronic nose detection module 4, an environmental detection module 5, a main control module 6, an aroma release kinetic model module 7, a large language model decision module 8, a digital twin module 9, an automatic ash removal module 10, and a user preference learning module 11.

[0055] Agarwood storage module 1: Used to store granular, powdered, flake or block agarwood raw materials. The hopper volume is selected from 100g to 2000g. The bottom is equipped with a feeding port connected to the automatic feeding module. A vibrator is configured to prevent the agarwood raw materials from clogging. This module is the material source of the entire system. Its capacity determines the maximum unattended continuous operation time of the system.

[0056] Automatic feeding module 2: Connected to the storage module, it uses a screw conveyor mechanism as the quantitative feeding actuator, driven by a stepper motor. Quantitative feeding is achieved by controlling the motor's rotation speed, with a control accuracy of ±0.05g. A built-in weight sensor monitors the remaining material level in real time. This module automatically determines the feeding timing, frequency, and amount based on aroma release kinetics models and digital twin predictions, making it a key actuator for achieving long-term unmanned operation.

[0057] Heating Module 3: Includes a heating plate, ceramic heater, temperature sensor, air duct, and fan. Heating temperature range: 50℃~350℃, normal operating range: 120℃~260℃. Based on the aroma release pattern of agarwood, the controller controls the heating power according to a four-stage temperature curve: Preheating stage: 120℃~150℃, 5~15 minutes, releasing light volatile components mainly composed of sesquiterpenes; Aroma release stage: 180℃~220℃, releasing the core agarwood aroma mainly composed of aromatic compounds; Constant aroma stage: 190℃±5℃, maintaining stable aroma output; Aroma preservation stage: 140℃~170℃, reducing the rate of fragrance consumption. Stage switching is controlled by... The cosine similarity matching results with the calibrated quantities at each stage are automatically triggered, rather than executed according to a fixed schedule. When the combustion-sensitive channel exceeds the safety threshold, the heating module will be forced to cool down to a safe temperature below 140°C, regardless of the current stage.

[0058] Electronic nose detection module 4: Located in the aromatherapy chamber or exhaust channel, it contains multiple VOC gas sensors of different types, such as BME688, SGP30, CCS811, and MQ135. It samples every 10 seconds, collecting parameters including: total VOC concentration, CO2 concentration, temperature and humidity, and an odor feature vector formed by the fusion of multiple sensors. The real-time aroma concentration data output by this module serves as the feedback signal source for closed-loop control.

[0059] Different sensors have different sensitivities to different VOC components. For example, BME688 is sensitive to alcohols and aldehydes, SGP30 is sensitive to total VOCs and CO2, and MQ135 is sensitive to CO and nitrogen-containing compounds. It has two core uses:

[0060] (a) The VOC components released at different heating stages are different, resulting in The sensor response ratio exhibits phased fingerprint characteristics. The current... Calculate the cosine similarity with each of the four stage calibration vectors. , This is the feature vector of the current odor. For the first The system identifies the odor-labeled feature vectors for each stage, and the stage with the highest cosine similarity exceeding a threshold is the current stage of the system. Stage switching is automatically triggered by similarity matching results, eliminating the need for manual scheduling.

[0061] (b) When Channels sensitive to combustion products, such as the CO response component of MQ135. When the preset safety threshold is exceeded, regardless of Whether it's normal or not, all tests indicate a risk of scorching. This solves the problem of relying solely on total VOC levels. The core issue is the inability to distinguish between "normal high concentration" and "abnormal charring"; during high-concentration aroma release... It can reach over 80ppm but The ratios of all channels are normal, but there is scorching. It may only be 40ppm, but the MQ135 channel spiked abnormally.

[0062] Environmental monitoring module 5: Monitors parameters such as temperature, humidity, and air circulation in the environment where the incense diffuser is located. These environmental parameters directly affect the diffusion and dilution rate of the aroma in the space and are key compensation inputs for the control model.

[0063] Main control module 6: As the decision-making center of the system, it has the following core functions: (1) receiving real-time data from the electronic nose detection module and calculating the current aroma concentration; (2) running the aroma release kinetic model, estimating the current state and predicting future trends; (3) receiving the target aroma parameters generated by the large language model decision module; (4) when When a scorching warning is triggered, a forced cooling protection procedure is executed; (5) the deviation between the actual concentration and the target concentration is compared. , For the target concentration, Based on the current actual concentration, the heating power adjustment is generated using incremental PID control. Simultaneously, based on the digital twin feedforward compensation amount Generate feeding and temperature adjustment instructions; (6) Call the user preference learning module to continuously optimize control parameters. The control method adopts PID and closed-loop control strategy with predictive feedforward, that is, a composite control algorithm that combines real-time feedback and model prediction.

[0064] The complete process of aroma release kinetics model module 7 is as follows:

[0065] 1. The release kinetic formula describes the volatilization pattern of agarwood fragrance itself:

[0066] ;

[0067] in, VOC concentration released from the surface of the fragrance , Current heating temperature , For the remaining mass of agarwood , This is the temperature influence coefficient. This is the quality influence coefficient. This is a proportionality coefficient, related to the type and particle size of agarwood. The pyrolysis attenuation coefficient is... This indicates that the aroma release rate decreases over time under constant temperature conditions.

[0068] 2. Concentration prediction model in Based on this, the conversion to predicted concentration at sensor location is performed in three steps: Step 1: Humidity correction is applied. ,in , The relative humidity of the environment (%) For reference humidity, This is a humidity correction factor, approximately Higher humidity means higher water vapor content in the air, slower VOC molecule diffusion, and higher local concentration near the sensor. Follow Monotonically increasing. Step 2 introduces the rate of temperature change. Calculate the rate of concentration change This reflects the dynamic impact of temperature rises and falls on concentration trends, among which... Determined directly by the material feeding action. Step 3: Extrapolation prediction. To obtain the future The predicted concentration at any given time. Simultaneously, historical VOC concentration data is continuously compared with the electronic nose's measured values ​​to... , , , The five parameters are calibrated using least squares. The model's final output is: the current concentration estimate. rate of concentration change Future predicted concentration Recommended feed amount and recommended temperature adjustment Recommended feed amount and recommended temperature adjustment Calculations were made by working backwards from the concentration gap: That is, when the predicted concentration is lower than the target, the amount of material to be added is calculated based on the ratio. This means that when the predicted concentration is lower than the target, the required temperature increase is calculated based on the ratio. For the amount of material added, This represents the current remaining material quantity. For the target concentration, To predict concentration, For factors affecting quality; For temperature adjustment amount, The current temperature. This is the temperature influence coefficient.

[0069] Large Language Model (LLM) Decision Module 8: The system is equipped with a microphone, speaker, and speech recognition engine. Users express their aroma preferences in natural language, such as "lighter aroma," "I want meditation mode," "suitable for tea," "last for two hours," "a bit strong now," etc. After speech recognition, the text is sent to the Large Language Model inference engine. LLM parses the fuzzy semantics into structured control parameters: target aroma concentration, aroma duration, heating rate, temperature range, addition frequency, airflow control parameters, etc. For example, "sleep mode" is mapped to: target temperature 150℃, target concentration 30%, duration 8 hours; "meeting mode" is mapped to: target temperature 210℃, target concentration 80%; "lighter aroma" is mapped to a 10%–20% reduction from the current target concentration.

[0070] Digital Twin Module 9: Constructs a real-time replica of the virtual incense system, mapping the remaining agarwood quantity, temperature, VOC concentration, environmental parameters, and user preference parameters. Based on the current state, the system predicts aroma changes over the next 30 minutes, 1 hour, and 2 hours. For example, if the current aroma score is 85 and it's predicted to drop to 72 in 30 minutes, the system automatically initiates replenishment in advance. The digital twin module is also used for offline optimization—testing different control strategies in the virtual environment, selecting the optimal solution, and then deploying it to the physical system for execution.

[0071] Automatic ash removal module 10: Located at the bottom of the combustion chamber, it includes an ash pan, scraper, drive motor, and ash collection box. The system automatically triggers the ash removal operation when one of the following conditions is met: (1) continuous operation time exceeds 4 hours; (2) aroma release efficiency decreases by more than 20%, the criterion being the ratio of the actual VOC concentration to the theoretical concentration of the release kinetic formula at the same temperature T(t) and remaining mass M(t). The decrease in this ratio indicates that the effective volatilization of the spices is hindered due to the ash covering the surface. The ash removal process is as follows: reduce heating power → rotate the ash pan → scrape off the residue with a scraper → the ash falls into the ash collection box → resume operation. The entire process requires no manual intervention.

[0072] User Preference Learning Module 11: The system continuously records user commands, operating parameters, aroma concentration, and adjustment records to build a personalized aroma database for each user. For example, if a user reports "too strong" five times consecutively in the system-generated "meditation mode," the system will automatically lower the target concentration for that mode by 10%. Through continuous learning, personalized adaptation is achieved.

[0073] like Figure 3 As shown, the closed-loop control process of the present invention includes the following steps:

[0074] S1: Users input their fragrance preferences via voice input in natural language.

[0075] S2: The large language model parses the semantics and generates structured target aroma parameters, such as temperature range, duration and addition frequency;

[0076] S3: Calculate the target concentration ;

[0077] S4: Electronic nose detection module 4 detects the current actual VOC concentration at a sampling interval of 10 seconds. and odor feature vector ;

[0078] S5: The main controller will With calibration vectors at each stage Perform cosine similarity matching to automatically identify the current heating stage and determine the target temperature range accordingly;

[0079] S6: Inspection If the combustion-sensitive channel exceeds the safety threshold, proceed to the next step. Forced cooling;

[0080] S7: Main controller calculates concentration deviation Incremental PID control law is adopted:

[0081] ;

[0082] Calculate the heating power adjustment amount; This is the amount of heating power adjustment. This is the proportionality coefficient. The integral coefficient is... These are the differential coefficients. For target concentration and the current concentration at time t The difference, Let be the dummy variable for integration, representing the integration interval. Any historical moment within the timeframe is used to distinguish between past times in the integration process and the current system time t. For a historic moment The corresponding concentration deviation is calculated by integration to accumulate the deviation over the entire time period. For concentration deviation The first derivative with respect to time t represents the instantaneous rate of change of the deviation at the current moment;

[0083] S8: If Then increase the heating temperature or increase the amount of material added;

[0084] S9: If Then lower the heating temperature or reduce the amount of material added;

[0085] S10: Digital Twin Module 9 predicts the future based on the current state. Concentration trend over time ;

[0086] S11: Determine the predicted concentration Does it exceed the permitted range? );

[0087] S12: If the range is exceeded, feedforward compensation control will be activated in advance. ,in Forward gain coefficient ( ); The amount of heating power adjustment provided for feedforward compensation Predicted concentrations for future moments;

[0088] S13: According to Control the material replenishment or pre-adjust the temperature, and run continuously in a cycle. At the same time, write the running data and user feedback of this round into the user preference learning module for the next round of parameter optimization.

[0089] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any other way. Any modifications or equivalent changes made based on the technical essence of the present invention shall still fall within the scope of protection claimed by the present invention.

Claims

1. An intelligent agarwood incense system based on electronic nasal feedback and large language model collaborative control, characterized in that, The intelligent agarwood incense system based on electronic nose feedback and large language model collaborative control includes an agarwood storage module (1), an automatic feeding module (2), a heating module (3), an electronic nose detection module (4), an environmental detection module (5), a main control module (6), an aroma release dynamics model module (7), a large language model decision module (8), a digital twin module (9), an automatic ash removal module (10), and a user preference learning module (11). The agarwood storage module (1) is used to store agarwood raw materials and supply them to the automatic feeding module; The automatic feeding module (2) is used to quantitatively feed agarwood raw materials to the heating module; The heating module (3) is used to heat the agarwood raw material so that the agarwood raw material releases its aroma; The electronic nose detection module (4) is used to detect aroma concentration and odor characteristic information, and to feed the detection results back to the main control module; The environmental detection module (5) is used to detect ambient temperature, humidity and air circulation. The aroma release kinetics model module (7) is used to estimate the current aroma state and predict the future trend of aroma concentration change; The large language model decision module (8) is used to parse the user's natural language instructions into structured target aroma parameters; The digital twin module (9) is used to construct a virtual incense system and predict its future operating status; The automatic ash removal module (10) is used to clean the ash residue generated after heating; The user preference learning module (11) is used to record user usage data and correct aroma control parameters; The main control module (6) is used to control the operation of the automatic feeding module (2), heating module (3) and automatic dust removal module (10) based on the information from the electronic nose detection module (4), environmental detection module (5), aroma release dynamics model module (7), large language model decision module (8), digital twin module (9) and user preference learning module (11).

2. The intelligent agarwood incense system based on electronic nasal feedback and large language model collaborative control as described in claim 1, characterized in that, The agarwood storage module (1) includes a hopper, a discharge port and a vibrator. The discharge port is connected to an automatic feeding module. The vibrator is used to prevent the agarwood raw materials from clogging. The hopper is used to store the agarwood raw materials.

3. The intelligent agarwood incense system based on electronic nasal feedback and large language model collaborative control as described in claim 1, characterized in that, The automatic feeding module (2) includes a screw conveyor, a stepper motor and a weight sensor. The stepper motor is used to drive the screw conveyor to feed the material in a quantitative manner, and the weight sensor is used to detect the amount of remaining material.

4. The intelligent agarwood incense system based on electronic nasal feedback and large language model collaborative control as described in claim 1, characterized in that, The heating module (3) includes a heating plate, a ceramic heater and a temperature sensor, and performs phased heating control according to the preheating stage, the aroma release stage, the aroma retention stage and the aroma preservation stage.

5. The intelligent agarwood incense system based on electronic nasal feedback and large language model collaborative control as described in claim 4, characterized in that, The electronic nose detection module (4) includes multiple VOC gas sensors for collecting VOC concentration, CO2 concentration, temperature and humidity and odor feature vectors. The main control module identifies the current heating stage or judges the risk of scorching based on the odor feature vectors.

6. The intelligent agarwood incense system based on electronic nasal feedback and large language model collaborative control as described in claim 4, characterized in that, The aroma release kinetic model module (7) outputs the current aroma concentration estimate, future predicted concentration, recommended amount of additive and recommended temperature adjustment based on the heating temperature, remaining mass of agarwood, ambient humidity and historical VOC concentration data.

7. The intelligent agarwood incense system based on electronic nasal feedback and large language model collaborative control as described in claim 4, characterized in that, The digital twin module (9) is used to predict the future trend of aroma concentration change. When the predicted concentration deviates from the target concentration by more than the preset range, the main control module controls the automatic feeding module to replenish the material or controls the heating module to adjust the temperature.

8. The intelligent agarwood incense system based on electronic nasal feedback and large language model collaborative control according to claim 4, characterized in that, The temperature during the preheating stage is 120℃~150℃, the temperature during the aroma release stage is 180℃~220℃, the temperature during the aroma maintenance stage is 185℃~195℃, and the temperature during the aroma preservation stage is 140℃~170℃.

9. The intelligent agarwood incense system based on electronic nasal feedback and large language model collaborative control as described in claim 4, characterized in that, The aroma release kinetics model module (7) includes release kinetics and concentration prediction. The release kinetics formula describes the volatility of agarwood fragrance itself, and the core formula is: ； in, VOC concentration released from the surface of the fragrance , Current heating temperature , For the remaining mass of agarwood , This is the temperature influence coefficient. This is the quality influence coefficient. This is a proportionality coefficient, related to the type and particle size of agarwood. The pyrolysis attenuation coefficient is... This indicates that the aroma release rate decreases over time under constant temperature conditions.

10. A smart agarwood incense burning method based on the collaborative control of electronic nasal feedback and large language model as described in any one of claims 1-9, characterized in that: Includes the following steps: S1: Users input their fragrance preferences via voice input in natural language. S2: The large language model parses the semantics and generates structured target aroma parameters, such as temperature range, duration and addition frequency; S3: Calculate the target concentration ; S4: The electronic nose detection module (4) detects the current actual VOC concentration at a sampling interval of 10 seconds. and odor feature vector ; S5: The main controller will With calibration vectors at each stage Perform cosine similarity matching to automatically identify the current heating stage and determine the target temperature range accordingly; S6: Inspection If the combustion-sensitive channel exceeds the safety threshold, proceed to the next step. Forced cooling; S7: Main controller calculates concentration deviation Incremental PID control law is adopted. Calculate the heating power adjustment amount; This is the amount of heating power adjustment. This is the proportionality coefficient. The integral coefficient is... These are the differential coefficients. For target concentration and the current concentration at time t The difference, Let be the dummy variable for integration, representing the integration interval. Any historical moment within the timeframe is used to distinguish between past times in the integration process and the current system time t. For a historic moment The corresponding concentration deviation is calculated by integration to accumulate the deviation over the entire time period. For concentration deviation The first derivative with respect to time t represents the instantaneous rate of change of the deviation at the current moment; S8: If Then increase the heating temperature or increase the amount of material added; S9: If Then lower the heating temperature or reduce the amount of material added; S10: Digital Twin Module (9) predicts the future based on the current state. Concentration trend over time ; S11: Determine the predicted concentration Does it exceed the permitted range? S12: If the range is exceeded, feedforward compensation control will be activated in advance. ,in, This is the feedforward gain coefficient. ; The amount of heating power adjustment provided for feedforward compensation; S13: According to The system controls advance material replenishment or pre-adjusts temperature, continuously cycling through the system. Simultaneously, it writes the current cycle's operational data and user feedback into a user preference learning module for parameter optimization in the next round. ; ; For the amount of material added, This represents the current remaining material quantity. For the target concentration, To predict concentration, For factors affecting quality; For temperature adjustment amount, The current temperature. This is the temperature influence coefficient.

Images