Raisin uniform dewatering and drying control system based on hot air circulation technology

Abstract

The invention relates to the technical field of industrial control systems, and particularly discloses a raisin uniform dewatering and drying control system based on a hot air circulation technology, the system comprises a multi-mode sensing array, a core decision processor, a dynamic thermal field execution mechanism and a data and model library, the sensing array collects environment and material internal quality data in real time; the decision processor fuses the data and couples a drying and quality degradation kinetic model to carry out rolling optimization, and a control instruction is generated; according to the hot air circulation field multi-target collaborative optimization system and method, multi-target collaborative optimization of dehydration efficiency, uniformity and nutrition preservation is achieved, and the drying quality and uniformity of raisins are improved.

Description

Technical Field

[0001] This invention belongs to the field of industrial control system technology, specifically relating to a raisin uniform dehydration and drying control system based on hot air circulation technology. Background Technology

[0002] In the field of agricultural product processing, dehydration and drying of fruits and vegetables are key technologies for extending their shelf life, facilitating storage and transportation, and preserving their nutritional value. Among these methods, hot air drying is a widely used and cost-effective technique. Its core lies in controlling the flow of hot air to effectively remove moisture from the material.

[0003] The drying process of raisins is an important application of hot air drying technology. This process aims to ensure that grapes are dehydrated evenly and efficiently by precisely controlling the drying environment, while preserving their original flavor, color, and nutrients to the greatest extent possible.

[0004] Hot air circulation control systems based on fixed temperatures or simple staged temperature curves are often employed. These systems tend to simplify the drying process to a simple problem of moisture evaporation, neglecting the dynamic degradation and transformation of heat-sensitive nutrients such as sugars and polyphenols within raisins under continuous or inappropriate temperatures. Due to a lack of in-depth understanding of the coupling relationship between the internal biochemical reactions of the material and external drying conditions, existing control strategies struggle to achieve synergistic optimization between dehydration efficiency and nutritional quality preservation. In the pursuit of rapid dehydration, excessively high temperatures or inappropriate temperature change rates can easily lead to sugar caramelization and polyphenol oxidation, resulting in browning, flavor deterioration, and a decrease in nutritional value. Conversely, overly conservative low-temperature strategies significantly prolong the drying cycle, increase energy consumption, and negatively impact production efficiency. Summary of the Invention

[0005] The purpose of this invention is to provide a uniform dehydration and drying control system for raisins based on hot air circulation technology, so as to solve the technical contradiction in the prior art that the dehydration efficiency and nutritional quality protection are difficult to optimize in a coordinated manner, and the product is prone to browning and flavor deterioration because the dynamic degradation law of heat-sensitive nutrients inside raisins is ignored during the drying process.

[0006] This invention provides a raisin uniform dehydration and drying control system based on hot air circulation technology. The system includes a multimodal sensing array, a core decision processor, a dynamic thermal field actuator, and a data and model library.

[0007] The multimodal sensing array is used to collect multi-dimensional data characterizing the drying environment and the internal quality of the raisin material in real time during the drying process. The array includes an environmental sensing module and a material sensing module. The environmental sensing module, integrated at multiple spatial coordinate points within the drying chamber, collects real-time temperature, relative humidity, and hot air velocity data. The material sensing module is used for non-contact acquisition of internal quality data of the raisin material during the drying process; this module includes a hyperspectral imaging unit and a near-infrared spectral analysis unit. The hyperspectral imaging unit periodically scans the raisin material on the drying bed to acquire continuous spectral image data in the visible to short-wave infrared band, analyzing the uniformity of moisture distribution, sugar concentration distribution, and color change characteristics on the material surface. The near-infrared spectral analysis unit uses a fiber optic probe to perform point-to-point penetration detection on sampled raisin material, acquiring its absorption spectrum data in the near-infrared band to invert the moisture content, total phenol content, and reducing sugar content of the core area inside the material.

[0008] The core decision processor is used to perform multi-objective collaborative optimization calculations based on multi-dimensional data collected in real time by a multi-modal sensing array, combined with drying kinetic models and quality degradation kinetic models preset in the data and model library, to generate the optimal drying process parameter control instructions for the current moment. The processor includes a state fusion and feature extraction module, a multi-model coupling prediction module, and a rolling optimization control module.

[0009] The state fusion and feature extraction module receives and fuses real-time data streams from the environmental sensing module and the material sensing module. This module first preprocesses the continuous spectral image data acquired by the hyperspectral imaging unit, including dark current correction, spectral reflectance calculation, and image registration. Then, it extracts multiple feature vectors characterizing the uniformity of moisture content, sugar distribution gradient, and color spatial heterogeneity of the material surface using principal component analysis. Simultaneously, this module preprocesses the absorption spectral data acquired by the near-infrared spectral analysis unit, including spectral smoothing, derivative transformation, and standard normal variable transformation. Then, it uses partial least squares regression, combined with a pre-established near-infrared spectral quantitative analysis model in the data and model library, to calculate the internal moisture content, total phenol content, and reducing sugar content of the sampled material in real time. Finally, this module integrates environmental parameters, material surface feature vectors, and internal material quality parameters into a unified time-series state vector.

[0010] The multi-model coupled prediction module is used to perform joint simulation prediction of the drying process and quality evolution in a future prediction time domain based on the current state vector. This module incorporates two core dynamic models: one is a thin-layer drying dynamic model for raisins based on unsteady-state heat and mass transfer theory, used to predict the change trajectory of material moisture content over time under different drying conditions; the other is a quality deterioration dynamic model based on the Arrhenius equation and chemical reaction kinetics, used to predict the change trajectory of key nutritional indicators such as total phenol content and reducing sugar content in the material over time under different temperature and humidity conditions. The coupled prediction process of this module is as follows: using the current state vector as the initial condition, a set of drying process parameters to be evaluated is input into the drying dynamic model to simulate and obtain the predicted change curve of material moisture content in the prediction time domain; simultaneously, the same process parameter sequence, along with intermediate variables such as material temperature and water activity output from the drying dynamic model, are input into the quality deterioration dynamic model to simulate and obtain the predicted change curves of total phenol content and reducing sugar content in the prediction time domain. The rolling optimization control module is used to perform multi-objective optimization calculations to generate control commands. This module defines an objective function containing three sub-objectives: the first sub-objective is the squared deviation between the predicted end-time material moisture content and the target moisture content; the second sub-objective is the variance of the feature vector of uniformity of surface moisture distribution of the material within the predicted time domain, calculated from the hyperspectral feature extraction results; and the third sub-objective is the weighted sum of the retention rates of total phenol content and reducing sugar content within the predicted time domain. Using a constrained nonlinear programming algorithm, under constraints set by the physical limits of the drying equipment, this module continuously solves for the optimal drying process parameter sequence within several future control time domains that minimizes the aforementioned multi-objective weighted sum function. The first set of parameters in this sequence, i.e., the optimal setpoint for the current control time domain, is then output as the control command.

[0011] The dynamic thermal field actuator is used to precisely execute control commands issued by the core decision processor to finely regulate the hot air circulation field inside the drying chamber. This mechanism includes a zoned air supply unit, a precise temperature and humidity control unit, and a circulation power and airflow guiding unit.

[0012] The zoned air supply unit consists of an array of multiple independent electric air valves arranged in the air supply static pressure box at the top of the drying chamber. Each air valve corresponds to an independent air supply area below the drying bed. The unit receives data on the target air volume of each air supply area from the control command and achieves independent and precise control of the air volume of different areas of the drying bed by adjusting the opening of the corresponding electric air valve. This allows for proactive intervention and correction of the uneven distribution of moisture on the material surface identified by hyperspectral imaging.

[0013] The precise temperature and humidity control unit includes a gas proportional control heater, a steam humidifier, and a surface-cooled dehumidifier. This unit receives data on the target temperature and target relative humidity from the control command, and dynamically adjusts the opening of the gas valve, the steam solenoid valve, and the refrigerant expansion valve through a proportional-integral-derivative control algorithm, so that the air at the return air vent of the drying chamber reaches the temperature and humidity parameters set in the command after mixing and processing.

[0014] The circulating power and flow guiding unit includes a variable frequency centrifugal fan and an adjustable angle guide plate system. The variable frequency centrifugal fan is used to adjust its operating frequency according to the total circulating air volume set in the control command. The adjustable angle guide plate system is installed in the side wall of the drying chamber and the return air channel. It is used to adjust its angle according to the flow field mode pre-optimized by computational fluid dynamics simulation, guide the hot air to form a specific circulation path in the drying chamber, and ensure that the treated hot air can penetrate the material layer evenly and fully.

[0015] The data and model library supports the system's offline knowledge base and online learning module. This library stores historical drying process datasets, hyperspectral and near-infrared spectral feature databases, drying kinetic model parameter sets, quality degradation kinetic model parameter sets, and optimization algorithm parameters. Furthermore, the system includes an online model update module; after each drying batch, this module compares and analyzes the actual collected complete process data and final product quality inspection data with the prediction results of the multi-model coupled prediction module to calculate the prediction error. Then, it uses the recursive least squares method to fine-tune and update the key parameters in the drying kinetic model and the quality degradation kinetic model, enabling the model to adapt to the differences in characteristics of grape raw materials from different origins and varieties.

[0016] As one embodiment of the present invention, the construction process of the quality degradation kinetic model in the multi-model coupling prediction module is as follows: First, by designing constant temperature and humidity drying experiments under different temperature and humidity conditions, measured data on the changes of total phenol content and reducing sugar content of raisin samples over time during the drying process are obtained; then, based on the theory of chemical reaction kinetics, rate equations for total phenol degradation and reducing sugar conversion are established respectively, with the rate constants described by the Arrhenius equation as a function of temperature, and water activity is introduced as a correction factor affecting the reaction rate; finally, using experimental data, the apparent activation energy, pre-exponential factor, and water activity influence coefficient of each reaction are determined by fitting a nonlinear regression algorithm, thereby completing the parameterization of the model.

[0017] In one embodiment of the present invention, the weight coefficients of the multi-objective weighted sum function in the rolling optimization control module are not fixed, but dynamically adjusted according to different stages of the drying process. In the initial drying stage, when the material moisture content is higher than 50%, the optimization weights tend to prioritize rapid dehydration and suppression of surface crusting, thus assigning higher weights to the first and second sub-objectives. In the middle drying stage, when the material moisture content is between 20% and 50%, the optimization weights shift towards a balance between dehydration uniformity and nutrient preservation, thus the weights of the three sub-objectives tend to be more balanced. In the later drying stage, when the material moisture content is lower than 20%, the optimization weights strongly favor nutrient preservation and prevention of overheating, thus assigning the highest weight to the third sub-objective. This phased dynamic weighting strategy is automatically switched by the core decision processor based on the real-time calculated average material moisture content.

[0018] In one embodiment of the present invention, the control logic of the zoned air supply unit is as follows: the core decision processor calculates the average moisture value corresponding to each zone of the drying bed based on the current moment's material surface moisture distribution map extracted by the hyperspectral imaging unit; then, using the average moisture value of all zones as a benchmark, the moisture deviation of each zone is calculated; for zones with moisture values ​​higher than the benchmark, the system increases the opening of the corresponding air supply valve to improve the local wind speed and ventilation efficiency of that zone, thereby accelerating its dehydration rate; for zones with moisture values ​​lower than the benchmark, the air supply volume is reduced accordingly to avoid over-drying. This adjustment process is performed in a fixed control cycle until the moisture uniformity index of each zone reaches a preset threshold.

[0019] In one embodiment of the present invention, the system operates within a hierarchical control framework. This framework includes a batch planning layer, a process optimization layer, and an equipment execution layer. The batch planning layer operates before the start of the drying batch, and based on the initial moisture content of the grape raw materials, varietal characteristics, and the target product quality grade, it retrieves or generates initial drying baseline curves and optimization weight strategy templates from a data and model library. The process optimization layer, which is the part of the core decision processor that operates in real-time during the drying process, receives sensing data, performs rolling optimization, and issues control commands on a minute-by-minute basis. The equipment execution layer, which is the dynamic thermal field actuator, responds to the commands of the process optimization layer on a second-by-second basis, completing precise closed-loop control of the execution terminals such as temperature, humidity, wind speed, and airflow.

[0020] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0021] 1. This invention breaks through the limitations of traditional drying control, which only focuses on external environmental parameters and overall moisture content. By integrating hyperspectral imaging and near-infrared spectral analysis technologies, it achieves for the first time real-time, non-destructive, online monitoring of the core nutritional quality components inside raisin materials during the drying process.

[0022] 2. This invention creatively couples a drying kinetics model with a quality degradation kinetics model, constructing a digital twin simulation environment capable of simultaneously predicting moisture removal and nutrient changes. The core decision processor performs rolling optimization based on this coupled model, generating process parameter commands that are no longer isolated temperature or humidity setpoints, but rather globally optimal solutions that comprehensively consider dehydration efficiency, dehydration uniformity, and the retention rates of various nutrient indicators. This fundamentally transforms the drying process from a purely thermophysical process into a controlled physicochemical synergistic process, achieving a fundamental shift in the technical solution.

[0023] 3. This invention provides a physical means to actively intervene in the uniformity of material drying through the synergistic effect of zoned air supply units and dynamic thermal field control. The system can adjust the local air volume in real time and direction based on the uneven surface moisture distribution identified by hyperspectral imaging, thereby actively smoothing the drying gradient in the spatial dimension. This effectively solves the problem of localized over-drying or under-drying caused by uneven airflow distribution or material accumulation characteristics, ensuring the achievement of the uniform dehydration target from the system execution level.

[0024] 4. The built-in online model update module of this invention endows the system with self-learning and adaptive capabilities. By continuously calibrating the core prediction model using production data from each batch, the system can gradually adapt to the natural fluctuations in raw material characteristics, ensuring that the control strategy remains in an optimal or near-optimal state. This continuous evolution mechanism significantly improves the system's robustness and universality in different production scenarios, extending its technological lifecycle. Attached Figure Description

[0025] Figure 1 This is a schematic diagram of the overall technical solution architecture of the present invention;

[0026] Figure 2 This is a schematic diagram of the core principle framework of the core decision processor in this invention for multi-objective rolling optimization;

[0027] Figure 3 This is a flowchart illustrating the logical process of data acquisition and feature extraction using a multimodal sensing array in this invention.

[0028] Figure 4 This is a schematic diagram of the multi-level interaction relationship and data flow of the dynamic thermal field actuator for fine-tuning in this invention;

[0029] Figure 5 This is a logical flow diagram of the hierarchical control framework in this invention. Detailed Implementation

[0030] Example 1: The overall technical architecture of the raisin uniform dehydration and drying control system based on hot air circulation technology proposed in this invention is shown in the attached figure. Figures 1 to 5As shown in the attached diagram. This system is centered around four core functional modules: a multimodal sensing array, a core decision processor, a dynamic thermal field actuator, and a data and model library. It constructs a closed-loop control system integrating real-time sensing, intelligent decision-making, precise execution, and continuous learning. The entire system operates within a hierarchical control framework, as illustrated in the attached diagram. Figure 5 As shown, the framework includes a batch planning layer, a process optimization layer, and an equipment execution layer, which correspond to three levels: process pre-setting before drying, rolling optimization during drying, and physical control of drying equipment, respectively, ensuring seamless connection from macro strategy to micro execution.

[0031] During the system startup phase, the batch planning layer first retrieves or generates an initial drying baseline curve and an optimized weighting strategy template from the data and model library based on the initial moisture content, varietal characteristics, and target product quality grade of the grape raw materials used in this drying task. The baseline curve defines the ideal trajectory of key parameters such as temperature, humidity, and wind speed changing over time during the drying process; while the optimized weighting strategy template specifies the dynamic adjustment rules for the trade-off between dehydration efficiency, moisture uniformity, and nutrient retention rate at different drying stages. After completing the above initialization, the system enters the real-time operation phase of the process optimization layer.

[0032] A multimodal sensing array, serving as the system's sensory center, is deployed within the drying chamber to simultaneously collect multi-dimensional data on environmental conditions and the internal quality of the material throughout the drying process. (See attached image) Figure 3 As shown, the array consists of an environmental sensing module and a material sensing module. The environmental sensing module contains multiple distributed sensor nodes, which are precisely positioned at key spatial coordinates within the drying chamber, including air inlets, outlets, above the bed layers, and in the gaps between the bed layers. These nodes are used to collect real-time temperature, relative humidity, and hot air velocity data within the drying chamber. All environmental parameters are sampled at a rate down to the second and transmitted to the core decision processor via industrial Ethernet.

[0033] The material sensing module employs non-contact optical detection methods for non-destructive, online monitoring of the raisin material. This module includes a hyperspectral imaging unit and a near-infrared spectral analysis unit. The hyperspectral imaging unit is mounted on a track-sliding mechanism directly above the drying bed, capable of performing a complete scan of the entire bed surface within each control cycle (typically 2 minutes). Its operating wavelength covers the visible to short-wave infrared range from 400 nm to 2500 nm, with a spectral resolution of 5 nm and a spatial resolution of 0.5 mm / pixel. Each scan generates a three-dimensional data cube containing two-dimensional spatial information and one-dimensional spectral information. This raw data is first corrected for dark current to eliminate the influence of the detector's own thermal noise; then, spectral reflectance is calculated to convert the raw digital values ​​into standard reflectance values; finally, an image registration algorithm is used to align images from different time points to the same spatial coordinate system to support subsequent time-series analysis.

[0034] The preprocessed hyperspectral image data is fed into the state fusion and feature extraction module. This module uses principal component analysis (PCA) to reduce the dimensionality of the high-dimensional spectral data, extracting the first five principal components as core feature vectors characterizing the material's state. The first principal component primarily reflects the spatial distribution of the overall moisture content of the material; the second principal component characterizes the lateral gradient of sugar concentration; and the third principal component is highly correlated with the degree of browning. By performing regional statistical analysis on the images of each principal component, the average moisture content, sugar variation coefficient, and color difference ΔE value corresponding to each zone of the drying bed can be calculated, thereby quantifying the uniformity of moisture on the material surface, the heterogeneity of sugar distribution, and the degree of color degradation.

[0035] Meanwhile, the near-infrared spectroscopy analysis unit uses a fiber optic probe installed on the sidewall of the drying bed to perform point-to-point penetration detection on several randomly selected raisin samples. This unit operates in the 900 nm to 1700 nm wavelength range, with a sampling frequency of once every 5 minutes. Each detection yields a complete near-infrared absorption spectrum curve. This raw spectrum is first smoothed by Savitzky-Golay filtering to suppress high-frequency noise; then, a first-order derivative transformation is performed to eliminate baseline drift; and finally, a standard normal variable transformation is used to eliminate scattering effects caused by particle size and packing density. The processed spectral data is input into a pre-established partial least squares regression model, stored in a data and model library and trained through extensive offline calibration experiments. The model outputs three key internal quality parameters: internal moisture content, total phenol content, and reducing sugar content, expressed as percentages, milligrams per gram, and percentages, respectively.

[0036] The state fusion and feature extraction module integrates all the above information—including temperature, humidity, and wind speed data provided by the environmental perception module, surface feature vectors extracted by the hyperspectral imaging unit, and internal quality parameters calculated by the near-infrared spectroscopy analysis unit—into a unified time-series state vector. This state vector has 18 dimensions and specifically includes: average temperature of the drying chamber, maximum temperature difference, average relative humidity, average wind speed, average moisture content of bed zones 1 to 6, sugar distribution gradient index, color space heterogeneity index, internal moisture content, total phenol content, and reducing sugar content. This state vector, as a complete representation of the current system state, is transmitted in real-time to the multi-model coupled prediction module.

[0037] The multi-model coupled prediction module is the simulation brain of the core decision processor. Its function is to jointly simulate the drying process and quality evolution within a future prediction time domain (usually set to 30 minutes), given the current state and a set of candidate control instructions. This module has two built-in core dynamic models: a raisin thin-layer drying dynamic model and a quality degradation dynamic model.

[0038] The drying kinetics model is based on unsteady-state heat and mass transfer theory, and its basic form is as follows:

[0039]

[0040] in, This represents the local moisture content of the material (dry basis). For time, The effective moisture diffusion coefficient, For the Laplace operator, The convective mass transfer coefficient, To balance the moisture content, this model considers the effects of temperature, humidity, and wind speed. and The model mitigates the nonlinear effects of airflow in different zones of the bed by introducing zoned airflow as a boundary condition, thus enabling differentiated simulation of dehydration rates in different zones. In each prediction iteration, the model uses the internal moisture content and environmental parameters in the current state vector as initial conditions, inputs a set of six candidate process parameters (including airflow in each zone, total airflow, temperature, and humidity) divided into 5-minute intervals for the next 30 minutes, and outputs the predicted moisture content distribution of each zone at each future time.

[0041] The quality degradation kinetic model, based on the Arrhenius equation and chemical reaction kinetics, describes the degradation behavior of total phenols and reducing sugars under hot and humid conditions. This model measures the degradation rate of total phenols. With reducing sugar conversion rate Expressed as temperature With water activity Functions:

[0042]

[0043] in, Pre-exponential factor, As the apparent activation energy, This is the universal gas constant. The water activity correction function is typically expressed as a quadratic polynomial. Model parameters. , and The coefficients were all determined by fitting offline constant temperature and humidity experimental data and stored in the data and model library. In the coupled prediction process, the model receives the material temperature and water activity (converted from moisture content) output by the drying kinetic model as input, and combines the same process parameter sequence to calculate the predicted change trajectory of total phenol content and reducing sugar content in the next 30 minutes.

[0044] The rolling optimization control module receives multiple sets of simulation results output by the multi-model coupled prediction module and performs multi-objective optimization calculations. The objective function defined by this module is... The weighted sum of the three sub-objectives:

[0045]

[0046] The first item is the predicted endpoint moisture content. With target moisture content The square of the deviation; the second term To predict the variance of the eigenvector representing the uniformity of moisture distribution on the material surface at different times in the time domain, it was calculated using the results of hyperspectral principal component analysis. The first term represents the feature vector of uniformity of moisture distribution on the material surface; the third term is the weighted sum of total phenol retention rate and reducing sugar retention rate. To predict the final total phenol retention rate, This represents the initial total phenol retention rate. To predict the endpoint reducing sugar retention rate, The initial reducing sugar retention rate. This is the weighting coefficient for nutritional indicators, typically set to 0.6 to reflect the higher heat sensitivity of total phenols. Weighting coefficient , , It is not fixed, but dynamically adjusted based on the current average moisture content of the material: when the moisture content is higher than 50%, =0.5、 =0.4、 =0.1; when the moisture content is between 20% and 50%, all three are 0.33; when the moisture content is below 20%, =0.1、 =0.2、 =0.7. This dynamic weighting strategy is automatically switched by the core decision processor.

[0047] The optimization problem is solved under the physical constraints of the drying equipment. These constraints include: a temperature range of 40°C to 70°C, a relative humidity range of 10% to 60%, a total air volume range of 5000 cubic meters per hour to 20000 cubic meters per hour, and a deviation in air volume for each zone not exceeding ±30% of the average. The rolling optimization control module employs a sequential quadratic programming algorithm to search for the objective function within the next six control time domains (i.e., 30 minutes). The minimum optimal process parameter sequence is obtained, and the first set of parameters in the sequence is output as the current control command to the dynamic thermal field actuator.

[0048] The dynamic thermal field actuator, as the executing element of the system, is as follows: Figure 4 As shown, it is responsible for translating abstract control commands into concrete physical actions. This mechanism consists of three parts: a zoned air supply unit, a precise temperature and humidity control unit, and a circulation power and airflow guiding unit.

[0049] The zoned air supply unit consists of six independent electrically operated air valves installed in the air supply static pressure box at the top of the drying chamber. Each valve corresponds to a rectangular air supply area below the bed. The air valves use proportionally adjustable electric actuators. Upon receiving the target air volume data for each zone from the control command, the control system first converts the air volume value into a corresponding air valve opening command, and then drives the actuator through a pulse width modulation signal. Specifically, the control logic is as follows: if the current moisture value of a zone is higher than the average moisture value of the bed, the opening of the air valve in that zone is increased to increase the local air velocity and accelerate dehydration; conversely, the opening is decreased to prevent over-drying. This adjustment is performed in a 2-minute cycle until the moisture distribution uniformity index detected by hyperspectral imaging (defined as the ratio of the standard deviation of moisture content in each zone to the mean) is lower than a preset threshold of 3%.

[0050] The precise temperature and humidity control unit is integrated into the air handling unit, including a gas-fired proportional heater, a steam humidifier, and a surface-cooled dehumidifier. The heater uses a proportional gas valve, the opening of which is adjusted in real time by a PID controller based on the deviation between the return air temperature and the target temperature. The steam humidifier controls the amount of saturated steam injected through a high-speed solenoid valve to increase air humidity. The surface-cooled dehumidifier regulates the refrigerant flow through an electronic expansion valve to achieve deep dehumidification. These three components work together to ensure that the mixed air reaches the set temperature and humidity parameters before entering the drying chamber, with control accuracies of ±0.5 degrees Celsius and ±2% relative humidity, respectively.

[0051] The circulating power and flow guiding unit consists of a variable frequency centrifugal fan and an adjustable angle baffle system. The variable frequency fan has a power of 15 kW, a frequency adjustment range of 20 Hz to 50 Hz, and a corresponding airflow adjustment range of 5000 to 20000 cubic meters per hour. The baffle system includes eight sets of baffle blades, installed on both sides of the drying chamber and at the inlet of the return air duct. The angle of each set of blades can be continuously adjusted from 0 to 60 degrees by a servo motor. The initial angle configuration of the baffles is pre-optimized and determined through computational fluid dynamics simulation, aiming to form a forced circulation path that flows from top to bottom, penetrates the material layer, and then returns along the sidewall. In actual operation, the baffle angle can be fine-tuned according to the bed stacking height or material type to maintain optimal airflow penetration.

[0052] The data and model library, serving as the system's knowledge hub, not only stores historical drying process datasets, hyperspectral and near-infrared spectral feature databases, kinetic model parameter sets, and optimization algorithm parameters, but also includes an online model update module. This module automatically activates after each drying batch, comparing the actual collected full-process state vectors, control command sequences, and final product quality test data (including laboratory-measured final moisture content, total phenols, reducing sugars, and sensory scores) with the prediction results of the multi-model coupled prediction module point by point to calculate the prediction error. Subsequently, the recursive least squares method is used to update the effective diffusion coefficient in the drying kinetic model. With convective mass transfer coefficient And the pre-exponential factor in the quality degradation model With activation energy Fine-tuning was performed. The updated model parameters were written back to the data and model library to guide the drying control of subsequent batches, thereby enabling the system to adapt to grape raw materials from different origins and at different ripeness levels.

[0053] Throughout the drying process, the system executes a closed-loop sensing-decision-execution cycle with a control period of 2 minutes. Within each cycle, the multimodal sensing array completes a comprehensive data acquisition, the core decision processor performs a rolling optimization calculation and issues new instructions, and the dynamic thermal field actuator completes physical control with a response speed of seconds. Through this high-frequency closed-loop control, the system can respond in real time to dynamic deviations caused by changes in material properties, environmental disturbances, or equipment aging during the drying process, always maintaining the drying process at the optimal balance between dehydration efficiency and nutrient preservation.

[0054] Example 2: In another embodiment, the multimodal sensing array of this system can further expand its sensing dimensions to enhance its adaptability to complex drying scenarios. Specifically, in addition to the hyperspectral imaging unit and the near-infrared spectral analysis unit, the material sensing module can also integrate a thermal imaging unit. This thermal imaging unit uses an uncooled infrared focal plane array detector, operating in the 8-14 micrometer band, with a spatial resolution of 1 millimeter and a frame rate of 1 Hz. Its installation position is adjacent to the hyperspectral imaging unit, allowing for simultaneous acquisition of real-time temperature distribution maps of the raisin material surface. This temperature distribution data is directly input into the state fusion and feature extraction module to correct the boundary heat exchange conditions in the drying kinetics model.

[0055] Because a semi-permeable membrane forms on the surface of raisins during the drying process, hindering internal moisture migration, traditional models relying solely on ambient temperature struggle to accurately predict internal heat conduction. By incorporating measured surface temperature fields, the state fusion module can calculate the actual convective heat transfer coefficient between the material surface and the ambient air, using it as a dynamic input parameter for the drying kinetics model. This significantly improves the model's ability to simulate surface crusting, especially during the high-humidity initial drying stage. Experiments show that after introducing thermal imaging data, the model's prediction error for the material's core temperature decreased from ±3 degrees Celsius to within ±1 degree Celsius.

[0056] Furthermore, in this embodiment, the objective function of the rolling optimization control module is also adjusted accordingly. A fourth sub-objective is added, namely, predicting the square of the maximum temperature difference on the material surface within the time domain, to suppress the risk of coking caused by local overheating. The weight of this sub-objective is... Activated in the later stages of drying (moisture content below 15%), the initial value is 0.2, increasing linearly to 0.4 as the moisture content decreases. Simultaneously, the moisture uniformity index in the original second sub-objective has been upgraded from a variance based solely on the hyperspectral moisture map to a moisture-temperature coupled uniformity index, calculated as the variance of the weighted Euclidean distance between the moisture values ​​and surface temperature values ​​of each zone. This improvement allows the system to consider not only moisture differences but also temperature distribution when adjusting airflow distribution, thus more comprehensively ensuring drying uniformity.

[0057] At the execution level, the control logic of the zoned air supply unit has also been upgraded. In addition to adjusting the opening of the air valve based on the moisture content deviation, temperature deviation is also introduced as an auxiliary adjustment factor. The specific rule is as follows: if the moisture content of a zone is high and the temperature is low, the air supply volume is increased significantly to simultaneously promote dehydration and heating; if the moisture content is low but the temperature is high, the air supply volume is reduced first to prevent coking, even if the moisture content is slightly lower than the benchmark. This composite criterion is implemented through a fuzzy logic controller, whose membership function is trained based on historical drying data to ensure that reasonable decisions can be made under various operating conditions.

[0058] This method is particularly suitable for processing grape varieties with thick skins that are prone to forming pods, such as seedless White Chicken Heart or Red Globe.

[0059] Example 3: In another embodiment, the hierarchical control framework of this system can further enhance its intelligence level, especially by introducing an initial strategy generation mechanism based on deep reinforcement learning at the batch planning layer. Traditional batch planning layers rely on pre-stored drying baseline curves, which may lead to suboptimal initial control strategies due to insufficient prior knowledge when dealing with raw materials from entirely new varieties or years with extreme climates. Therefore, this embodiment adds an artificial intelligence strategy engine to the data and model library, which is trained based on a deep Q-network architecture.

[0060] The training data for this strategy engine comes from thousands of batches of drying process data accumulated over long-term system operation. Each data point includes initial raw material characteristics (such as initial moisture content, sugar-acid ratio, and variety code), initial environmental conditions, the entire process control command sequence, final product quality indicators, and energy consumption data. The training objective is to learn a strategy function. Such that, given an initial state Under conditions including raw material and environmental characteristics, it can output a set of initial drying baseline curves and optimized weight templates to maximize the comprehensive reward function R, where R is defined as the product quality score minus the unit energy consumption cost.

[0061] In actual operation, when the batch planning layer receives a new batch task, it first determines whether the characteristics of the raw material are within the historical data coverage range. If so, it directly calls the baseline curve of the nearest batch; if not, it activates the artificial intelligence strategy engine, which generates a customized initial strategy. This strategy includes not only the baseline trajectory of temperature, humidity, and wind speed, but also exclusive dynamic weight adjustment rules for the characteristics of the raw material, such as assigning a higher reducing sugar retention weight to high-sugar varieties and a lower initial temperature upper limit to thin-skinned varieties.

[0062] Furthermore, the online model update module in this implementation also incorporates a Bayesian optimization mechanism to guide subsequent batches of exploratory experiments. When the system detects that the model prediction error for a certain type of raw material consistently exceeds a threshold, it automatically inserts a small number of exploratory control points in the next batch—that is, applies small perturbations near the normal control trajectory and records the system response. This perturbation data is used to update the Gaussian process surrogate model, thereby efficiently searching for a better combination of model parameters through the Bayesian optimization algorithm, avoiding getting trapped in local optima.

[0063] This implementation method has been validated when processing Cabernet Sauvignon raisins from two production areas: Turpan in Xinjiang and Helan Mountain in Ningxia. Given the significant differences in skin thickness and initial phenolic content between the two regions, the system automatically generates differentiated control schemes using an artificial intelligence strategy engine: Turpan raisins are treated with a high-temperature, rapid dehydration strategy in the early stages, while Helan Mountain raisins are treated with a low-temperature, slow-drying strategy throughout the process.

[0064] In summary, this invention constructs a highly intelligent and adaptive raisin drying control system by deeply integrating multimodal perception, multi-model coupled prediction, rolling optimization control, and dynamic actuators. Its core technology not only resolves the inherent contradiction between dehydration uniformity and nutrient preservation but also achieves long-term adaptation to diverse production scenarios through a continuous learning mechanism, providing a complete technical solution for the standardized and intelligent production of high-quality raisins.

Claims

1. A raisin uniform dehydration and drying control system based on hot air circulation technology, characterized in that, include: A multimodal sensing array is used to collect multi-dimensional data in real time during the drying process, which characterizes the state of the drying environment and the internal quality state of the raisin material. The core decision processor is used to perform multi-objective collaborative optimization calculations based on multi-dimensional data collected in real time by the multi-modal sensing array, combined with the preset drying kinetics model and quality degradation kinetics model, to generate the optimal drying process parameter control command for the current moment. The dynamic thermal field actuator is used to precisely execute the control commands issued by the core decision processor and to finely regulate the hot air circulation field inside the drying chamber; The data and model library stores historical drying process datasets, hyperspectral and near-infrared spectral feature databases, drying kinetic model parameter sets, quality degradation kinetic model parameter sets, and optimization algorithm parameters.

2. The raisin uniform dehydration and drying control system based on hot air circulation technology according to claim 1, characterized in that, The multimodal sensing array includes an environmental sensing module and a material sensing module; the environmental sensing module is integrated and arranged at multiple spatial coordinate points inside the drying chamber, and is used to collect real-time temperature data, real-time relative humidity data and real-time hot air flow rate data inside the drying chamber. The material sensing module is used to collect internal quality status data of raisins during the drying process in a non-contact manner; the material sensing module includes a hyperspectral imaging unit and a near-infrared spectral analysis unit. The hyperspectral imaging unit is used to periodically scan the raisin material on the drying bed to acquire its continuous spectral image data in the visible to short-wave infrared band. The near-infrared spectroscopy analysis unit is used to perform point-to-point penetration detection on the sampled raisin material through an optical fiber probe to obtain its absorption spectrum data in the near-infrared band.

3. The raisin uniform dehydration and drying control system based on hot air circulation technology according to claim 2, characterized in that, The core decision processor includes a state fusion and feature extraction module, which receives and fuses real-time data streams from the environmental sensing module and the material sensing module. This module preprocesses the continuous spectral image data acquired by the hyperspectral imaging unit, including dark current correction, spectral reflectance calculation, and image registration. Then, it extracts multiple feature vectors characterizing the uniformity of moisture content, sugar distribution gradient, and color spatial heterogeneity of the material surface using principal component analysis. Simultaneously, this module preprocesses the absorption spectral data acquired by the near-infrared spectral analysis unit, including spectral smoothing, derivative transformation, and standard normal variable transformation. Then, it uses partial least squares regression, combined with a pre-established near-infrared spectral quantitative analysis model in the data and model library, to calculate the internal moisture content, total phenol content, and reducing sugar content of the sampled material in real time. Finally, this module integrates environmental parameters, material surface feature vectors, and internal material quality parameters into a unified time-series state vector.

4. The raisin uniform dehydration and drying control system based on hot air circulation technology according to claim 3, characterized in that, The core decision processor also includes a multi-model coupled prediction module, which is used to perform joint simulation prediction of the drying process and quality evolution in a future prediction time domain based on the current state vector. This module incorporates two core kinetic models: one is a thin-layer drying kinetic model for raisins based on unsteady-state heat and mass transfer theory, used to predict the change trajectory of material moisture content over time under different drying conditions; the other is a quality deterioration kinetic model based on the Arrhenius equation and chemical reaction kinetics, used to predict the change trajectory of total phenol content and reducing sugar content in the material over time under different temperature and humidity conditions.

5. The raisin uniform dehydration and drying control system based on hot air circulation technology according to claim 4, characterized in that, The core decision processor also includes a rolling optimization control module, which performs multi-objective optimization calculations to generate control commands. This module defines an objective function with three sub-objectives: the first sub-objective is the squared deviation between the predicted end-time material moisture content and the target moisture content; the second sub-objective is the variance of the eigenvector representing the uniformity of surface moisture distribution within the predicted time domain; and the third sub-objective is the weighted sum of the retention rates of total phenol content and reducing sugar content within the predicted time domain. This module uses a constrained nonlinear programming algorithm to continuously solve for the optimal drying process parameter sequence within several future control time domains that minimizes the aforementioned multi-objective weighted sum function, and outputs the first set of parameters in this sequence as control commands.

6. The raisin uniform dehydration and drying control system based on hot air circulation technology according to claim 1, characterized in that, The dynamic thermal field actuator includes a zoned air supply unit, a precise temperature and humidity control unit, and a circulation power and flow guiding unit. The zoned air supply unit consists of an array of multiple independent electric air valves arranged in the air supply static pressure box at the top of the drying chamber. Each air valve corresponds to an independent air supply area below the drying bed. The unit receives data on the target air volume of each air supply area from the control command and achieves independent control of the air supply volume of different areas of the drying bed by adjusting the opening of the corresponding electric air valve. The precise temperature and humidity control unit includes a gas proportional control heater, a steam humidifier, and a surface-cooled dehumidifier. This unit receives data on the target temperature and target relative humidity from the control command and dynamically adjusts the opening degree of the gas valve, the opening degree of the steam solenoid valve, and the opening degree of the refrigerant expansion valve through a proportional-integral-derivative control algorithm. The circulating power and flow guiding unit includes a variable frequency centrifugal fan and an adjustable angle guide plate system; the variable frequency centrifugal fan is used to adjust its operating frequency according to the total circulating air volume set in the control command; the adjustable angle guide plate system is installed in the side wall of the drying chamber and in the return air channel, and is used to adjust its angle according to the flow field mode pre-optimized by computational fluid dynamics simulation.

7. The raisin uniform dehydration and drying control system based on hot air circulation technology according to claim 1, characterized in that, The coupling prediction process of the multi-model coupled prediction module is as follows: taking the current state vector as the initial condition, a set of drying process parameter sequences to be evaluated is input into the drying kinetic model to simulate and obtain the predicted change curve of material moisture content in the prediction time domain; at the same time, the same process parameter sequence and the intermediate variables of material temperature and water activity output from the drying kinetic model are input together into the quality deterioration kinetic model to simulate and obtain the predicted change curves of total phenol content and reducing sugar content in the prediction time domain.

8. The raisin uniform dehydration and drying control system based on hot air circulation technology according to claim 5, characterized in that, The weight coefficients of the multi-objective weighted sum function in the rolling optimization control module are dynamically adjusted according to different stages of the drying process. In the early stage of drying, when the material moisture content is higher than 50%, the first and second sub-objectives are given higher weights. In the middle stage of drying, when the material moisture content is between 20% and 50%, the weights of the three sub-objectives tend to be balanced. In the later stage of drying, when the material moisture content is lower than 20%, the third sub-objective is given the highest weight.

9. The raisin uniform dehydration and drying control system based on hot air circulation technology according to claim 6, characterized in that, The control logic of the zoned air supply unit is as follows: the core decision processor calculates the average moisture value of each zone of the drying bed based on the current moisture distribution map of the material surface extracted by the hyperspectral imaging unit; then, using the average moisture value of all zones as a benchmark, the deviation of moisture in each zone is calculated; for zones with moisture values ​​higher than the benchmark, the system increases the opening of the corresponding air supply valve to improve the local wind speed and ventilation efficiency in that zone; for zones with moisture values ​​lower than the benchmark, the air supply volume is reduced accordingly; this adjustment process is carried out in a fixed control cycle until the moisture uniformity index of each zone reaches the preset threshold.

10. The raisin uniform dehydration and drying control system based on hot air circulation technology according to claim 7, characterized in that, It also includes an online model update module; after each drying batch is completed, the online model update module compares and analyzes the complete process data actually collected for this batch with the final product quality inspection data and the prediction results of the multi-model coupled prediction module to calculate the prediction error; then, it uses the recursive least squares method to fine-tune and update the key parameters in the drying kinetic model and the quality deterioration kinetic model.

Images