Temperature and humidity self-adaptive control system and method for drying flower traditional Chinese medicinal materials

Abstract

The invention discloses a temperature and humidity adaptive control system and method for drying flower traditional Chinese medicinal materials, and the system comprises a humidity gradient monitoring module which is used for deploying a 12-channel capacitive humidity sensor spiral array and a laser radar, generating a three-dimensional humidity field thermodynamic diagram in combination with a Kriging interpolation method, calculating a humidity gradient change rate in real time, and carrying out the early warning of local abnormality; the heat-humidity ratio regulation and control module is used for integrating humidity gradient monitoring data and a real-time result of the evaporation latent heat calculation module based on a fuzzy PID algorithm, dynamically adjusting the heating power and the rotating speed of a dehumidification fan, and meanwhile, integrating a PVDF piezoelectric film sensor to monitor petal stress; the evaporation latent heat calculation module is used for predicting a future evaporation rate by combining an LSTM neural network based on an edge calculation unit and providing key parameters for the energy supply matching module; the energy supply matching module is used for outputting a combined control instruction of the heating power and the air door opening degree; and the air volume balance optimization module is used for dynamically adjusting air volume distribution according to the humidity gradient monitoring data.

Description

Technical Field

[0001] This invention relates to the field of drying temperature control technology, and in particular to a temperature and humidity adaptive control system and method for drying flower-type Chinese medicinal materials. Background Technology

[0002] Traditional drying methods (such as natural air drying and static drying rooms) rely on manual experience to regulate temperature and humidity, which has drawbacks such as low control precision, slow response, and numerous environmental interference factors. These methods struggle to meet the diverse temperature and humidity requirements of different varieties of floral medicinal herbs. By integrating a temperature and humidity sensor network, intelligent algorithm models, and dynamic actuators, a closed-loop control system is constructed for real-time monitoring, accurate prediction, and automatic adjustment. This system enables dynamic adaptive regulation of the drying environment's temperature and humidity, ensuring the drying process remains within the optimal range of parameters. This not only effectively preserves the active ingredients such as alkaloids, flavonoids, and volatile oils in floral medicinal herbs, maintaining their natural color and aroma characteristics and improving the quality stability and clinical efficacy of the dried product, but also reduces manual intervention through automated control, lowering energy consumption and production costs, and improving drying efficiency and adaptability to large-scale production. This promotes the modernization of medicinal herb drying technology from experience-driven to data-driven, providing key technical support for the standardization and high-quality development of the medicinal herb industry, ultimately ensuring the safety and effectiveness of clinical medication and the sustainable utilization of medicinal resources.

[0003] During the drying process of floral medicinal materials, humidity changes and temperature control need to be dynamically matched (i.e., the "heat-to-moisture ratio" needs to be adjusted according to the real-time status). Traditional systems, due to their fixed control logic, are difficult to adapt to the drying needs of different varieties and batches of medicinal materials (such as differences in moisture content and varietal characteristics), resulting in weak adaptability. Therefore, an adaptive temperature and humidity control system and method for drying floral Chinese medicinal materials is proposed. Summary of the Invention

[0004] The purpose of this invention is to address the shortcomings of existing technologies by proposing an adaptive temperature and humidity control system and method for drying flower-based Chinese medicinal materials.

[0005] To achieve the above objectives, the present invention adopts the following technical solution:

[0006] An adaptive temperature and humidity control system for drying floral medicinal materials includes:

[0007] Humidity gradient monitoring module: Deploys a 12-channel capacitive humidity sensor spiral array and lidar, and uses Kriging interpolation to generate a three-dimensional humidity field thermogram, calculates the rate of change of humidity gradient in real time and provides early warning of local anomalies;

[0008] Heat-to-moisture ratio control module: Based on the fuzzy PID algorithm, it integrates humidity gradient monitoring data and real-time results from the latent heat of vaporization calculation module to dynamically adjust the heating power and dehumidifying fan speed. At the same time, it integrates a PVDF piezoelectric film sensor to monitor petal stress. When the risk of excessive drying is detected, the heat-to-moisture ratio is reduced first.

[0009] Evaporative latent heat calculation module: Relying on edge computing units, it uses the modified formula L_eff=(40690-12.8ΔT+0.08ΔT) 2 The evaporation rate is predicted in the next 10 minutes by combining 1 + 0.03d - 0.015v (where d is the material thickness, v is the wind speed, and ΔT is the temperature change) with an LSTM neural network (error < 3%). The latent heat of vaporization is updated every 5 seconds to provide key parameters for the energy supply matching module.

[0010] Energy supply matching module: Using the DDPG reinforcement learning algorithm, with the calculation result of latent heat of vaporization as input, it outputs a joint control command of heating power and damper opening, and at the same time links the paraffin-based composite phase change energy storage system to charge during off-peak hours and release energy during the day, giving priority to the use of stored heat.

[0011] Airflow balance optimization module: dynamically adjusts airflow distribution based on humidity gradient monitoring data, and fine-tunes damper opening through PID algorithm to keep drying uniformity within ±1℃.

[0012] The above technical solution further includes:

[0013] Preferably, a three-layer sensor array is arranged along the vertical axis inside the drying chamber, with four capacitive humidity sensors evenly distributed along the perimeter of the chamber in each layer. The lidar performs a full chamber scan once every certain period of time to generate three-dimensional point cloud data.

[0014] An adaptive temperature and humidity control method for drying flower-type Chinese medicinal materials includes the following steps;

[0015] Multimodal data acquisition and preprocessing: A 12-channel humidity sensor acquires humidity data every second, a PVDF sensor monitors petal deformation in real time, and a lidar scan generates point cloud data. The point cloud is aligned with the sensor coordinate system using the ICP algorithm, and high-frequency noise is filtered out using wavelet denoising. Temperature, humidity, and stress data are acquired and local errors are eliminated.

[0016] Three-dimensional humidity field reconstruction and gradient calculation: The humidity field model is constructed using the Kriging interpolation method, and the gradient vector and the gradient change rate of the time series are calculated by combining the central difference method. When the gradient change rate of a local area is >5%, the abnormal area is marked.

[0017] Dynamic calculation and prediction of latent heat of vaporization: Quantifying current energy demand and predicting future drying trends through real-time latent heat models and LSTM neural networks;

[0018] Dynamic control decision and execution: Based on the fuzzy PID algorithm, the heating power and dehumidifying fan speed are adjusted by combining humidity gradient and stress data. Based on the DDPG reinforcement learning algorithm, the heating power and damper opening are output by taking the latent heat of vaporization and prediction data as input. According to the abnormal area marking, the angle of the variable cross-section air duct guide plate is adjusted. The air volume balance is optimized by optimizing the air field distribution through CFD simulation.

[0019] Feedback optimization and closed-loop control: Quality detection is performed using a near-infrared spectrometer and image recognition system. The damper opening is adjusted according to the moisture content deviation. The drying data of this batch is input into the LSTM and DDPG models, and the network weights are updated through backpropagation.

[0020] Drying termination judgment and system reset: The stop signal is triggered based on the termination condition that the moisture content is ≤8% and the humidity gradient is <2% for 30 consecutive minutes. The heating element is turned off, the fan speed is reduced to cooling mode, and the high-pressure air pump is started to clean the air duct and sensor surface residue.

[0021] The above technical solution further includes:

[0022] Preferably, the specific steps for multimodal data acquisition and preprocessing are as follows:

[0023] Multimodal sensor data acquisition: A capacitive humidity sensor synchronously collects humidity values ​​every second; a PVDF piezoelectric film sensor converts the bending deformation of petals into a voltage signal and converts it into a stress value through a calibration curve; and a lidar scans the cavity to generate three-dimensional point cloud data.

[0024] Point cloud and sensor coordinate system alignment: Initial corresponding point matching: Based on the nearest neighbor principle, for each sensor point (x s ,y s ,z s Find the nearest point (x) in the point cloud p ,y p ,z p Transformation matrix calculation: Solve for the rotation matrix R and translation vector t by minimizing the objective function. Where R is a 3×3 rotation matrix and t is a 3×1 translation vector; Iterative optimization: Repeat the initial corresponding point matching and transformation matrix calculation until convergence;

[0025] Wavelet denoising and data filtering: Humidity data is decomposed into three levels using the Daubechies-4 wavelet basis, and high-frequency coefficients are processed and the signal is reconstructed using a soft thresholding function. Stress data is filtered by a moving average filter.

[0026] Data fusion and local error elimination: By combining aligned point cloud data and sensor data, local humidity values ​​are corrected using Kriging interpolation, and the deviation between sensor data and interpolation results is compared to identify abnormal sensors.

[0027] Preferably, the specific steps for the three-dimensional humidity field reconstruction and gradient calculation are as follows:

[0028] A three-dimensional humidity field is constructed using Kriging interpolation: The discrete data from a 12-channel humidity sensor are weighted and averaged using Kriging interpolation based on a Gaussian variogram. The Kriging interpolation method is expressed as follows: Wherein, λ i The weighting coefficients satisfy the unbiased estimation condition. Let Lagrange multipliers be the variables, and the variation function be expressed as follows: Where h is the distance between the two points, σ 2 Let θ be the variance and θ be the correlation length. A continuous humidity field model is constructed to visualize the humidity distribution inside the drying cavity.

[0029] Gradient vector calculation: The partial derivatives are calculated using the central difference method to generate the gradient vector and its magnitude is calculated. The partial derivatives calculated using the central difference method are expressed as follows: Similarly, calculate and Where Δx=Δy=Δz, the gradient magnitude is calculated as follows: Quantify the direction and rate of humidity change at each point, thereby identifying areas of drastic local humidity changes;

[0030] Time series gradient change rate monitoring: The gradient magnitude of time series data is calculated every second. The gradient change rate is obtained by comparing the difference between the current time and the previous time. When the gradient change rate of a local area exceeds 5%, potential risks of uneven drying are dynamically identified.

[0031] Abnormal area marking and feedback: Abnormal areas are marked by three-dimensional coordinate range and adjustment instructions are sent to the air volume balance optimization module and the heat-moisture ratio control module.

[0032] Preferably, the specific steps for the dynamic calculation and prediction of latent heat of vaporization are as follows:

[0033] Real-time latent heat of vaporization calculation: using a correction formula L that combines material thickness d, wind speed v, and temperature T. eff =(40690-12.8ΔT+0.08ΔT) 2 )×(1+0.03d-0.015v), where ΔT=T-25 is the temperature offset, quantifying the energy required for the evaporation of a unit mass of water;

[0034] LSTM neural network prediction: Using historical temperature and humidity RH(t), T(t) and stress data S(t) as input, the system uses two layers of LSTM units and Dropout to prevent overfitting and outputs the future evaporation rate.

[0035] Energy demand quantification: through The predicted evaporation rate and real-time latent heat are converted into future energy demand P. req (t), where η = 0.85 is the heating efficiency, generating dynamic energy supply commands;

[0036] Model self-optimization: By comparing the measured evaporation rate with the predicted value for each batch, the learning rate is dynamically adjusted and the LSTM weights are updated using the backpropagation algorithm.

[0037] Preferably, the specific steps of the dynamic control decision-making and execution are as follows:

[0038] Adjusting heating power and fan speed: The fuzzy PID algorithm maps humidity gradient deviation and its rate of change to a fuzzy set and calculates precise adjustment amounts based on a preset control rule table and the centroid method for defuzzification.

[0039] Output heating power and damper opening: The DDPG reinforcement learning algorithm takes the latent heat of vaporization and the evaporation rate predicted by LSTM as input state s. Through the deep reinforcement learning framework of actor network (output action a = [heating power P, damper opening θ]) and critic network, Ornstein-Uhlenbeck noise is added to explore the optimal action boundary, thereby completing the dynamic matching of heating power and damper opening based on long-term energy efficiency optimization.

[0040] Airflow balance: Based on the abnormal area marking, the airflow is dynamically distributed by adjusting the angle of the guide vane. The angle adjustment of the guide vane in the variable cross-section air duct is based on the formula Q=Q0·(1+0.05·α), where Q0 is the reference airflow and α is the guide vane angle.

[0041] CFD simulation optimizes wind field distribution: The uniformity of the wind field after duct adjustment is verified through computational fluid dynamics simulation, ensuring a Reynolds number Re < 2000. The Reynolds number is calculated as follows: Where ρ is air density, L is characteristic length, and μ is aerodynamic viscosity;

[0042] Multi-module collaboration and feedback: Fuzzy PID is responsible for short-term rapid adjustment, DDPG is responsible for long-term energy optimization, and after the physical adjustment of the air duct, the effect is verified by CFD simulation. If the uniformity does not meet the standard, the angle of the guide vane is further fine-tuned.

[0043] Preferably, the specific steps of the feedback optimization and closed-loop control are as follows:

[0044] Quality Inspection and Data Acquisition: Moisture content is detected in real time using a partial least squares regression model established by utilizing the absorption characteristics of moisture in the near-infrared region through a near-infrared spectrometer. This is expressed as... Among them, A i Let β be the absorbance at the i-th wavelength. i The regression coefficient is used to calculate the curl index by extracting petal edge features from an image recognition system.

[0045] PID-corrected damper opening: PID control algorithm applied based on moisture content deviation e. The damper opening θ is dynamically adjusted to achieve rapid response to moisture content deviation and air volume regulation.

[0046] Model self-optimization and weight update: Input the dry data of this batch into the LSTM and DDPG models, and optimize the mean squared error loss function of LSTM and the actor-commentator network of DDPG through backpropagation;

[0047] Data archiving and system reset: Drying process data in JSON format is stored in a cloud database, and high-pressure air pumps are used to clean residues from the sensor surface.

[0048] The present invention has the following beneficial effects:

[0049] In this invention, by real-time monitoring and analysis of the humidity gradient inside the cavity, the humidity conditions are precisely adjusted to achieve dynamic humidity control, thereby enhancing the accuracy of the drying process. Through dynamic adjustment of the heat-to-moisture ratio and coordination with the water evaporation characteristics, energy utilization efficiency is optimized, reducing energy consumption. Latent heat compensation takes into account the influence of water vapor partial pressure, improving the matching degree between heat supply and demand, ensuring drying quality. The energy supply matching and wind field balance optimization are combined with airflow direction data to improve the regional matching of wind speed and heating power, ensuring uniform distribution of the drying medium and improving overall drying efficiency and uniformity. Attached Figure Description

[0050] Figure 1 This is a system architecture diagram of a temperature and humidity adaptive control system for drying flower-type Chinese medicinal materials proposed in this invention;

[0051] Figure 2 This is a flowchart of a temperature and humidity adaptive control method for drying flower-type Chinese medicinal materials proposed in this invention. Detailed Implementation

[0052] 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.

[0053] like Figures 1-2 As shown, a temperature and humidity adaptive control system for drying flower-type Chinese medicinal materials includes:

[0054] Humidity gradient monitoring module: Deploys a 12-channel capacitive humidity sensor (Honeywell HIH-4000) spiral array and a lidar (Velodyne Puck LIDAR) for scanning, combined with Kriging interpolation to generate a 5mm resolution. 3 The three-dimensional humidity field thermogram is generated, and the humidity gradient change rate (ΔRH / min) is calculated in real time and local anomalies (>5%RH / min) are warned.

[0055] Heat-to-moisture ratio control module: Based on fuzzy PID algorithm, it integrates humidity gradient monitoring data and real-time results from the latent heat of vaporization calculation module to dynamically adjust heating power (0-5kW) and dehumidifying fan speed (0-1800rpm). At the same time, it integrates PVDF piezoelectric film sensor to monitor petal stress (deformation threshold 0.4mm). When the risk of excessive drying is detected, the heat-to-moisture ratio is reduced first to protect the morphology of the medicinal material, achieving matching and stress compensation of heat-to-moisture ratio in the optimal range of 0.8-1.2.

[0056] Evaporative latent heat calculation module: Relying on edge computing units, it uses the modified formula L_eff=(40690-12.8ΔT+0.08ΔT) 2 The evaporation rate is predicted in the next 10 minutes by combining 1 + 0.03d - 0.015v (where d is the material thickness, v is the wind speed, and ΔT is the temperature change) with an LSTM neural network (error < 3%). The latent heat of vaporization is updated every 5 seconds to provide key parameters for the energy supply matching module.

[0057] Energy supply matching module: Using the DDPG reinforcement learning algorithm, with the calculation result of latent heat of vaporization as input, it outputs a joint control command of heating power (P=0.85×(L_eff×Δm) / (ΔT×η)) and damper opening (θ=60%+20%×(ΔRH / 10%)). At the same time, it links with the paraffin-based composite phase change energy storage system (PCM) to charge during off-peak hours and release energy during the day, giving priority to the use of stored heat, thereby realizing dynamic power allocation and improving energy utilization (heat utilization rate improved by 40%).

[0058] Airflow balance optimization module: Through variable cross-section duct design (upper tapering ratio 1:3, middle honeycomb rectifier grille aperture 8mm, lower adjustable angle guide plate ±15°) and a simplified CFD model running every 5 minutes, the duct impedance coefficient is optimized to maintain laminar flow (Reynolds number <2000). The airflow distribution is dynamically adjusted according to humidity gradient monitoring data (1.5 × reference flow rate in high humidity zone, 0.8 × reference flow rate in vulnerable zone). At the same time, the damper opening is finely adjusted through PID algorithm to control the drying uniformity (temperature difference) within ±1℃.

[0059] In one embodiment, a three-layer sensor array is arranged along the vertical axis (Z-axis) inside the drying chamber. Each layer has four capacitive humidity sensors evenly distributed along the perimeter of the chamber (XY plane). The lidar performs a full chamber scan every 5 minutes to generate three-dimensional point cloud data. The point cloud data is aligned with the sensor coordinate system using the ICP (Iterative Closest Point) algorithm to correct the sensor position deviation.

[0060] An adaptive temperature and humidity control method for drying flower-type Chinese medicinal materials includes the following steps;

[0061] System initialization and parameter setting: Through sensor self-testing, parameter configuration (loading pre-trained LSTM model parameters based on the type of medicinal material, setting initial temperature, humidity, and stress thresholds) and duct calibration (low-speed operation of variable frequency fan combined with lidar scanning to verify airflow uniformity), the system achieves the functions of hardware status confirmation, control parameter initialization, and duct performance assurance.

[0062] Multimodal data acquisition and preprocessing: A 12-channel humidity sensor acquires humidity data per second, a PVDF sensor monitors petal deformation in real time (sampling frequency 1kHz), and a lidar scan generates point cloud data. The point cloud is aligned with the sensor coordinate system using the ICP algorithm, and high-frequency noise is filtered by wavelet denoising to obtain high-precision temperature, humidity, and stress data and eliminate local errors.

[0063] 3D humidity field reconstruction and gradient calculation: Kriging interpolation (based on Gaussian variation function, correlation length 0.4m) was used to construct a 5mm resolution humidity field. 3 The humidity field model is used, and the gradient vector and the gradient change rate (ΔRH / min) of the time series are calculated by combining the central difference method. When the gradient change rate of a local area is greater than 5%RH / min, the abnormal area is marked, which realizes the function of visualizing the humidity distribution in the drying chamber and identifying potential uneven drying risks in advance.

[0064] Evaporative latent heat dynamic calculation and prediction: Through a real-time latent heat model (combined with correction formulas for material thickness, wind speed, and temperature) and an LSTM neural network (inputting historical temperature, humidity, and stress data to predict the evaporation rate in the next 10 minutes with an error of <3%), the current energy demand is quantified and the future drying trend is predicted.

[0065] Dynamic control decision-making and execution: Based on the fuzzy PID algorithm, the heating power and dehumidifying fan speed are adjusted by combining humidity gradient and stress data. Based on the DDPG reinforcement learning algorithm, the heating power and damper opening are output by taking the latent heat of vaporization and prediction data as input. According to the abnormal area marking, the angle of the variable cross-section air duct guide plate is adjusted (±15°). The air volume balance is optimized by optimizing the air field distribution through CFD simulation.

[0066] Feedback optimization and closed-loop control: Quality inspection is carried out using a near-infrared spectrometer and image recognition system (to evaluate petal curling). The damper opening is adjusted according to the moisture content deviation (PID correction). The drying data (temperature, humidity / stress / moisture content) of this batch is input into the LSTM and DDPG models, and the network weights are updated through backpropagation.

[0067] Drying termination judgment and system reset: Based on the termination condition of moisture content ≤8% and humidity gradient <2%RH / min for 30 consecutive minutes, a stop signal is triggered, and the heating element is turned off, the fan speed is reduced to cooling mode, and the high-pressure air pump is started to clean the air duct and sensor surface residue.

[0068] In one embodiment, the specific steps of the multimodal data acquisition and preprocessing are as follows:

[0069] Multimodal sensor data acquisition: A capacitive humidity sensor synchronously collects humidity values ​​every second; a PVDF piezoelectric film sensor converts the bending deformation of petals into a voltage signal at a sampling frequency of 1kHz and converts it into a stress value through a calibration curve; and a lidar scans the cavity at a density of 1 million points / second to generate three-dimensional point cloud data.

[0070] Point cloud and sensor coordinate system alignment: Initial corresponding point matching: Based on the nearest neighbor principle, for each sensor point (x s ,y s ,z s Find the nearest point (x) in the point cloud p ,y p ,z p Transformation matrix calculation: Solve for the rotation matrix R and translation vector t by minimizing the objective function. Where R is a 3×3 rotation matrix and t is a 3×1 translation vector; Iterative optimization: Repeat the initial corresponding point matching and transformation matrix calculation until convergence (residual < 0.1 mm or number of iterations > 50);

[0071] Wavelet denoising and data filtering: Humidity data is decomposed into three levels using the Daubechies-4 wavelet basis. High-frequency coefficients are processed and the signal is reconstructed using a soft thresholding function (η(c)=sign(c)·max(|c|-λ,0), threshold λ=σ√(2lnN), σ=0.5%RH). A moving average filter with a window length of 5 is applied to the stress data. k = -2 to 2);

[0072] Data fusion and local error elimination: Combining aligned point cloud data with sensor data, kriging interpolation is used ( weight λ i (Generated under point cloud guidance) Corrects local humidity values ​​and compares the deviation between sensor data and interpolation results (ΔRH=|RH) sensor -RH interpolated |) Mark the abnormal sensor.

[0073] In one embodiment, the specific steps of the three-dimensional humidity field reconstruction and gradient calculation are as follows:

[0074] A three-dimensional humidity field is constructed using Kriging interpolation: The discrete data from a 12-channel humidity sensor are weighted and averaged using Kriging interpolation based on a Gaussian variogram. The Kriging interpolation method is expressed as follows: Wherein, λ i The weighting coefficients satisfy the unbiased estimation condition. Let Lagrange multipliers be the variables, and the variation function be expressed as follows: Where h is the distance between the two points, σ 2 Let θ be the variance and θ be the correlation length, and construct a resolution of 5 mm. 3 A continuous humidity field model was used to visualize the humidity distribution within the drying chamber.

[0075] Gradient vector calculation: using the central difference method at a depth of 5 mm. 3 Calculate the partial derivatives on the grid, generate the gradient vector and calculate its magnitude. The partial derivatives calculated using the central difference method are expressed as follows: Similarly, calculate and Where Δx=Δy=Δz, the gradient magnitude is calculated as follows: Quantify the direction and rate of humidity change at each point, thereby identifying areas of drastic local humidity changes;

[0076] Time-series gradient change rate monitoring: The time-series data of gradient magnitude is calculated every second. The gradient change rate (ΔRH / min) is obtained by comparing the difference between the current time and the previous time. When the gradient change rate in a local area exceeds 5%RH / min, potential risks of uneven drying are dynamically identified.

[0077] Abnormal area marking and feedback: Abnormal areas are marked by three-dimensional coordinate range and adjustment instructions are sent to the air volume balance optimization module and the heat-humidity ratio control module to increase the air volume in the abnormal area to 1.5 times the reference flow rate and reduce the heating power by 10%, thereby eliminating uneven drying and achieving precise adaptive control of temperature and humidity parameters.

[0078] In one embodiment, the specific steps for the dynamic calculation and prediction of latent heat of vaporization are as follows:

[0079] Real-time latent heat of vaporization calculation: using a correction formula L that combines material thickness d, wind speed v, and temperature T. eff =(40690-12.8ΔT+0.08ΔT) 2 )×(1+0.03d-0.015v), where ΔT=T-25 is the temperature offset. The formula is calibrated by experiments with an error of <2%, quantifying the energy required for the evaporation of a unit mass of water.

[0080] LSTM neural network prediction: Using historical temperature and humidity RH(t), T(t) and stress data S(t) as input, a two-layer LSTM unit (128 neurons per layer) and Dropout (ratio 0.2) are used to prevent overfitting. The future evaporation rate is output. The mean squared error (MSE) loss function and Adam optimizer (learning rate 0.001) are combined to achieve a prediction error of <3%.

[0081] Energy demand quantification: through The predicted evaporation rate and real-time latent heat are converted into future energy demand P. req (t), where η = 0.85 is the heating efficiency, generating dynamic energy supply commands;

[0082] Model self-optimization: By comparing the measured evaporation rate (verified by weighing method) with the predicted value for each batch, the learning rate is dynamically adjusted using the backpropagation algorithm (0.001 when the error is >2%, and 0.0001 when the error is <1%) to update the LSTM weights.

[0083] In one embodiment, the specific steps of the dynamic control decision-making and execution are as follows:

[0084] Adjusting heating power and fan speed: The fuzzy PID algorithm adjusts the humidity gradient deviation ( The rate of change (Δe) is mapped to a fuzzy set and the precise adjustment amount is calculated by combining the centroid method to defuzzify according to the preset control rule table.

[0085] Output heating power and damper opening: The DDPG reinforcement learning algorithm takes the latent heat of vaporization and the evaporation rate predicted by LSTM as input state s. Through a deep reinforcement learning framework of actor network (output action a = [heating power P, damper opening θ]) and critic network (evaluate action value Q(s,a)), Ornstein-Uhlenbeck noise is added to explore the optimal action boundary, thereby completing the dynamic matching of heating power and damper opening based on long-term energy efficiency optimization.

[0086] Airflow balance: Based on the abnormal area marking, the airflow is dynamically distributed by adjusting the angle of the guide vane. The angle adjustment of the guide vane in the variable cross-section air duct is based on the formula Q=Q0·(1+0.05·α), where Q0 is the reference airflow and α is the guide vane angle.

[0087] CFD simulation optimizes wind field distribution: Computational fluid dynamics (CFD) simulation verifies the uniformity of the wind field after duct adjustment, ensuring a Reynolds number Re < 2000 (laminar state). The Reynolds number is calculated as follows: Where ρ is air density, L is characteristic length, and μ is aerodynamic viscosity;

[0088] Multi-module collaboration and feedback: Fuzzy PID is responsible for short-term rapid adjustment, DDPG is responsible for long-term energy optimization, and after the physical adjustment of the air duct, the effect is verified by CFD simulation. If the uniformity does not meet the standard, the angle of the guide vane is further fine-tuned.

[0089] In one embodiment, the specific steps of the feedback optimization and closed-loop control are as follows:

[0090] Quality Inspection and Data Acquisition: Moisture content is detected in real time using a partial least squares (PLS) regression model established by utilizing the absorption characteristics of moisture in the near-infrared region through a near-infrared spectrometer (NIR). This is expressed as... Among them, A i Let β be the absorbance at the i-th wavelength. i The regression coefficients (determined through calibration experiments) were used to extract petal edge features using an image recognition system (CNN) to calculate the curl index, thereby achieving precise monitoring and data acquisition of drying endpoint quality parameters (moisture content ≤8%, curl <5%).

[0091] PID-corrected damper opening: PID control algorithm applied based on moisture content deviation e. The damper opening θ is dynamically adjusted to achieve rapid response to moisture content deviation and air volume regulation.

[0092] Model self-optimization and weight update: Input the drying data (temperature, humidity, stress, moisture content, control commands) of this batch into the LSTM and DDPG models, and optimize the mean squared error (MSE) loss function of LSTM and the actor-commentator network of DDPG through backpropagation;

[0093] Data archiving and system reset: Drying process data (temperature and humidity / stress / moisture content / control instructions / model weights) in JSON format are stored in a cloud database (AWS S3), and high-pressure air pump (0.6MPa, 5 minutes) is used to clean the sensor surface residue.

[0094] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A temperature and humidity adaptive control system for drying floral Chinese medicinal materials, characterized in that, include: Humidity gradient monitoring module: Deploys a 12-channel capacitive humidity sensor spiral array and lidar, and uses Kriging interpolation to generate a three-dimensional humidity field thermogram, calculates the rate of change of humidity gradient in real time and provides early warning of local anomalies; Heat-to-moisture ratio control module: Based on the fuzzy PID algorithm, it integrates humidity gradient monitoring data and real-time results from the latent heat of vaporization calculation module to dynamically adjust the heating power and dehumidifying fan speed. At the same time, it integrates a PVDF piezoelectric film sensor to monitor petal stress. When the risk of excessive drying is detected, the heat-to-moisture ratio is reduced first. Evaporative latent heat calculation module: Relying on the edge computing unit and combined with the LSTM neural network to predict the future evaporation rate, update the latent heat of vaporization value once, and provide key parameters for the energy supply matching module; Energy supply matching module: Using the DDPG reinforcement learning algorithm, with the calculation result of latent heat of vaporization as input, it outputs a joint control command of heating power and damper opening, and at the same time links the paraffin-based composite phase change energy storage system to charge during off-peak hours and release energy during the day, giving priority to the use of stored heat. Airflow balance optimization module: dynamically adjusts airflow distribution based on humidity gradient monitoring data, and fine-tunes damper opening through PID algorithm to keep drying uniformity within ±1℃.

2. The temperature and humidity adaptive control system for drying flower-type Chinese medicinal materials according to claim 1, characterized in that, Inside the drying chamber, a three-layer sensor array is arranged along the vertical axis. Each layer has four capacitive humidity sensors evenly distributed along the perimeter of the chamber. The lidar performs a full chamber scan once every certain period of time to generate three-dimensional point cloud data.

3. A method for adaptive temperature and humidity control in the drying of floral medicinal materials, using the temperature and humidity adaptive control system for drying floral medicinal materials as described in claim 1, characterized in that... Includes the following steps; Multimodal data acquisition and preprocessing: A 12-channel humidity sensor acquires humidity data every second, a PVDF sensor monitors petal deformation in real time, and a lidar scan generates point cloud data. The point cloud is aligned with the sensor coordinate system using the ICP algorithm, and high-frequency noise is filtered out using wavelet denoising. Temperature, humidity, and stress data are acquired and local errors are eliminated. Three-dimensional humidity field reconstruction and gradient calculation: The humidity field model is constructed using the Kriging interpolation method, and the gradient vector and the gradient change rate of the time series are calculated by combining the central difference method. When the gradient change rate of a local area is >5%, the abnormal area is marked. Dynamic calculation and prediction of latent heat of vaporization: Quantifying current energy demand and predicting future drying trends through real-time latent heat models and LSTM neural networks; Dynamic control decision and execution: Based on the fuzzy PID algorithm, the heating power and dehumidifying fan speed are adjusted by combining humidity gradient and stress data. Based on the DDPG reinforcement learning algorithm, the heating power and damper opening are output by taking the latent heat of vaporization and prediction data as input. According to the abnormal area marking, the angle of the variable cross-section air duct guide plate is adjusted. The air volume balance is optimized by optimizing the air field distribution through CFD simulation. Feedback optimization and closed-loop control: Quality detection is performed using a near-infrared spectrometer and image recognition system. The damper opening is adjusted according to the moisture content deviation. The drying data of this batch is input into the LSTM and DDPG models, and the network weights are updated through backpropagation. Drying termination judgment and system reset: The stop signal is triggered based on the termination condition that the moisture content is ≤8% and the humidity gradient is <2% for 30 consecutive minutes. The heating element is turned off, the fan speed is reduced to cooling mode, and the high-pressure air pump is started to clean the air duct and sensor surface residue.

4. The temperature and humidity adaptive control method for drying flower-type Chinese medicinal materials according to claim 3, characterized in that, The specific steps of multimodal data acquisition and preprocessing are as follows: Multimodal sensor data acquisition: A capacitive humidity sensor synchronously collects humidity values ​​every second; a PVDF piezoelectric film sensor converts the bending deformation of petals into a voltage signal and converts it into a stress value through a calibration curve; and a lidar scans the cavity to generate three-dimensional point cloud data. Point cloud and sensor coordinate system alignment: Initial corresponding point matching: Based on the nearest neighbor principle, for each sensor point (x s ,y s ,z s Find the nearest point (x) in the point cloud p ,y p ,z p Transformation matrix calculation: Solve for the rotation matrix R and translation vector t by minimizing the objective function. Where R is a 3×3 rotation matrix and t is a 3×1 translation vector; Iterative optimization: Repeat the initial corresponding point matching and transformation matrix calculation until convergence; Wavelet denoising and data filtering: Humidity data is decomposed into three levels using the Daubechies-4 wavelet basis, and high-frequency coefficients are processed and the signal is reconstructed using a soft thresholding function. Stress data is filtered by a moving average filter. Data fusion and local error elimination: By combining aligned point cloud data and sensor data, local humidity values ​​are corrected using Kriging interpolation, and the deviation between sensor data and interpolation results is compared to identify abnormal sensors.

5. The temperature and humidity adaptive control method for drying flower-type Chinese medicinal materials according to claim 3, characterized in that, The specific steps for reconstructing the three-dimensional humidity field and calculating the gradient are as follows: A three-dimensional humidity field is constructed using Kriging interpolation: The discrete data from a 12-channel humidity sensor are weighted and averaged using Kriging interpolation based on a Gaussian variogram. The Kriging interpolation method is expressed as follows: Wherein, λ i The weighting coefficients satisfy the unbiased estimation condition. Let Lagrange multipliers be the variables, and the variation function be expressed as follows: Where h is the distance between the two points, σ 2 Let θ be the variance and θ be the correlation length. A continuous humidity field model is constructed to visualize the humidity distribution inside the drying cavity. Gradient vector calculation: The partial derivatives are calculated using the central difference method to generate the gradient vector and its magnitude is calculated. The partial derivatives calculated using the central difference method are expressed as follows: Similarly, calculate and Where Δx=Δy=Δz, the gradient magnitude is calculated as follows: Quantify the direction and rate of humidity change at each point, thereby identifying areas of drastic local humidity changes; Time series gradient change rate monitoring: The gradient magnitude of time series data is calculated every second. The gradient change rate is obtained by comparing the difference between the current time and the previous time. When the gradient change rate of a local area exceeds 5%, potential risks of uneven drying are dynamically identified. Abnormal area marking and feedback: Abnormal areas are marked by three-dimensional coordinate range and adjustment instructions are sent to the air volume balance optimization module and the heat-moisture ratio control module.

6. The temperature and humidity adaptive control method for drying flower-type Chinese medicinal materials according to claim 3, characterized in that, The specific steps for dynamic calculation and prediction of latent heat of vaporization are as follows: Real-time latent heat of vaporization calculation: using a correction formula L that combines material thickness d, wind speed v, and temperature T. eff =(40690-12.8ΔT+0.08ΔT) 2 )×(1+0.03d-0.015v), where ΔT=T-25 is the temperature offset, quantifying the energy required for the evaporation of a unit mass of water; LSTM neural network prediction: Using historical temperature and humidity RH(t), T(t) and stress data S(t) as input, the system uses two layers of LSTM units and Dropout to prevent overfitting and outputs the future evaporation rate. Energy demand quantification: through The predicted evaporation rate and real-time latent heat are converted into future energy demand P. req (t), where η = 0.85 is the heating efficiency, generating dynamic energy supply commands; Model self-optimization: By comparing the measured evaporation rate with the predicted value for each batch, the learning rate is dynamically adjusted and the LSTM weights are updated using the backpropagation algorithm.

7. The temperature and humidity adaptive control method for drying flower-type Chinese medicinal materials according to claim 3, characterized in that, The specific steps of dynamic regulation decision-making and execution are as follows: Adjusting heating power and fan speed: The fuzzy PID algorithm maps humidity gradient deviation and its rate of change to a fuzzy set and calculates precise adjustment amounts based on a preset control rule table and the centroid method for defuzzification. Output heating power and damper opening: The DDPG reinforcement learning algorithm takes the latent heat of vaporization and the evaporation rate predicted by LSTM as input state s. Through the deep reinforcement learning framework of actor network (output action a = [heating power P, damper opening θ]) and critic network, Ornstein-Uhlenbeck noise is added to explore the optimal action boundary, thereby completing the dynamic matching of heating power and damper opening based on long-term energy efficiency optimization. Airflow balance: Based on the abnormal area marking, the airflow is dynamically distributed by adjusting the angle of the guide vane. The angle adjustment of the guide vane in the variable cross-section air duct is based on the formula Q=Q0·(1+0.05·α), where Q0 is the reference airflow and α is the guide vane angle. CFD simulation optimizes wind field distribution: The uniformity of the wind field after duct adjustment is verified through computational fluid dynamics simulation, ensuring a Reynolds number Re < 2000. The Reynolds number is calculated as follows: Where ρ is air density, L is characteristic length, and μ is aerodynamic viscosity; Multi-module collaboration and feedback: Fuzzy PID is responsible for short-term rapid adjustment, DDPG is responsible for long-term energy optimization, and after the physical adjustment of the air duct, the effect is verified by CFD simulation. If the uniformity does not meet the standard, the angle of the guide vane is further fine-tuned.

8. The temperature and humidity adaptive control method for drying flower-type Chinese medicinal materials according to claim 3, characterized in that, The specific steps of the feedback optimization and closed-loop control are as follows: Quality Inspection and Data Acquisition: Moisture content is detected in real time using a partial least squares regression model established by utilizing the absorption characteristics of moisture in the near-infrared region through a near-infrared spectrometer. This is expressed as... Among them, A i Let β be the absorbance at the i-th wavelength. i The regression coefficient is used to calculate the curl index by extracting petal edge features from an image recognition system. PID-corrected damper opening: PID control algorithm applied based on moisture content deviation e. The damper opening θ is dynamically adjusted to achieve rapid response to moisture content deviation and air volume regulation. Model self-optimization and weight update: Input the dry data of this batch into the LSTM and DDPG models, and optimize the mean squared error loss function of LSTM and the actor-commentator network of DDPG through backpropagation; Data archiving and system reset: Drying process data in JSON format is stored in a cloud database, and high-pressure air pumps are used to clean residues from the sensor surface.

Images