Multivariable control system for spray drying of aqueous organic solutions

Abstract

The present application relates to spray drying control technical field, disclose based on organic water solution spray drying multivariable control system. The system includes data acquisition, drying parameter determination, drying state perception, drying abnormal intervention and multivariable control module. Data acquisition module obtains the solution to be dried and collects equipment operating parameters, identifies environmental state and solution characteristics;Drying parameter determination module sets target parameters accordingly, locates the relative position of the equipment and determines the initial drying array;Drying state perception module combines the initial array and the preset rule to perceive the real-time state, analyzes the environmental change trend and sets the abnormal perception mode;Drying abnormal intervention module identifies abnormalities based on the mode, sets intervention mechanism in combination with the preset trajectory prediction network;Multivariable control module carries out multivariable control based on the above information, and obtains the result. The system can enhance the accuracy and stability of spray drying.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of spray drying control, in particular to a multivariable control system for spray drying of organic aqueous solution. BACKGROUND

[0002] In the production process of chemical, food, pharmaceutical and other industries, spray drying of organic aqueous solution is a common process. It obtains solid powder product by atomizing the solution and contacting with hot gas flow to make water evaporate quickly. However, the spray drying process of organic aqueous solution is influenced by many factors, including the concentration, viscosity, surface tension and other characteristics of the solution, as well as the inlet air temperature, outlet air temperature, air speed, atomization pressure, material flow and other operating parameters of the spray drying equipment, and also influenced by external conditions such as ambient temperature and humidity.

[0003] At present, the traditional spray drying control method mostly uses single variable control or simple multivariable correlation control, which is difficult to fully consider the interaction and dynamic change between various factors. For example, when the solution characteristics fluctuate, such as sudden increase in concentration, if only the inlet air temperature is adjusted, it may lead to over-drying or under-drying, affecting the quality of the product; and if multiple parameters are adjusted at the same time, but the precise grasp of the synergistic relationship between the parameters is lacking, new unstable factors may be caused. In addition, the traditional control system has a lag in sensing abnormal conditions during the drying process, and often intervenes only after the product quality has a significant problem, which is difficult to achieve real-time and effective adjustment, resulting in low production efficiency and serious waste of raw materials.

[0004] Different organic aqueous solutions have different characteristics in the spray drying process, and the requirements for drying parameters are also different. The existing control system lacks the ability to adaptively adjust to different solution characteristics and environmental conditions, and it is difficult to maintain stable drying effect under complex and variable working conditions. For example, for heat-sensitive organic matter, too high temperature will cause its decomposition and deterioration, and for high-viscosity solution, poor atomization effect will affect the drying efficiency. The traditional control system is difficult to automatically optimize the drying parameters according to these characteristics, and needs to rely on the experience of operators for manual adjustment, which not only increases the labor cost, but also is difficult to ensure the precision and consistency of control. SUMMARY

[0005] The present application aims to provide a multivariable control system for spray drying of organic aqueous solution to solve the problems raised in the background.

[0006] To achieve the above-mentioned purpose, the present application provides a multivariable control system for spray drying of organic aqueous solution, which comprises:

[0007] The data acquisition module is configured to acquire an organic matter aqueous solution to be spray dried and collect operation parameter data of a spray drying device, and identify a current environmental state and solution characteristics of a spray drying process according to the operation parameter data.

[0008] The drying parameter determination module is configured to set target drying parameters of spray drying, locate relative position parameters of the spray drying device in a drying process, determine initial drying patterns of spray drying based on the operation parameter data and the relative position parameters, based on the current environmental state and the solution characteristics.

[0009] The drying state sensing module is configured to sense a real-time drying state of spray drying in real time in combination with the initial drying patterns and preset drying rules, analyze an environmental change trend of the current environmental state based on the operation parameter data, and set an abnormality sensing mode of spray drying in combination with the real-time drying state and the environmental change trend.

[0010] The drying abnormality intervention module is configured to identify a drying abnormality of spray drying based on the abnormality sensing mode, and set an abnormality intervention mechanism of spray drying according to the drying abnormality and a preset trajectory prediction network.

[0011] The multivariable control module is configured to perform multivariable control processing on a spray drying process according to the abnormality sensing mode, the abnormality intervention mechanism and the preset drying rules, and obtain a spray drying control result.

[0012] Preferably, the determination of the initial drying patterns of spray drying based on the operation parameter data and the relative position parameters comprises:

[0013] identifying a drying trajectory of spray drying based on the operation parameter data;

[0014] scheduling drying road condition information and target drying parameters of spray drying according to the drying trajectory;

[0015] performing self-adaptive adjustment on the target drying parameters based on the drying road condition information and the relative position parameters to obtain self-adaptive drying parameters;

[0016] determining the initial drying patterns of spray drying in combination with the drying road condition information and the self-adaptive drying parameters.

[0017] Preferably, the identification of the drying abnormality of spray drying based on the abnormality sensing mode comprises:

[0018] collecting historical drying data of spray drying based on the abnormality sensing mode;

[0019] identifying normal drying parameters of spray drying according to the historical drying data.

[0020] identifying a current drying state of the spray drying, combining the current drying state and the historical drying data, identifying a current behavior deviation of the spray drying;

[0021] based on the normal drying parameter, setting an abnormal parameter threshold of the spray drying;

[0022] according to the abnormal parameter threshold and the current behavior deviation, identifying a drying abnormality of the spray drying.

[0023] Preferably, the environmental change trend of the current environment state is analyzed based on the operation parameter data, comprising:

[0024] performing attribute classification on the operation parameter data to obtain classified parameter data;

[0025] extracting key features of the classified parameter data, and collecting historical environment data of the current environment state based on the key features;

[0026] identifying temperature data and humidity data in the historical environment data;

[0027] according to the temperature data, analyzing a temperature change rate and an airflow behavior mode of the current environment state;

[0028] based on the humidity data, analyzing a humidity change level of the current environment state;

[0029] combining the temperature change rate, the airflow behavior mode and the humidity change level, analyzing an environmental change trend of the current environment state.

[0030] Preferably, the abnormal perception mode of the spray drying is set by combining the real-time drying state and the environmental change trend, comprising:

[0031] based on the real-time drying state, identifying surrounding environment factors of the spray drying;

[0032] analyzing an interaction relationship between the surrounding environment factors and the spray drying;

[0033] according to the environmental change trend and the interaction relationship, identifying a potential risk factor of the spray drying, and identifying a risk level of the potential risk factor;

[0034] according to the real-time drying state and the risk level, setting a risk response mode of the spray drying;

[0035] combining the risk level and the risk response mode, setting the abnormal perception mode of the spray drying.

[0036] Preferably, the abnormal intervention mechanism of spray drying is set according to the drying abnormal situation and the preset trajectory prediction network, including:

[0037] The abnormal drying point of spray drying is located according to the drying abnormal situation;

[0038] The dilemma type of the abnormal drying point is analyzed, and a dilemma response layer of the abnormal drying point is set based on the dilemma type;

[0039] The side drying trajectory of the abnormal drying point is identified according to the preset trajectory prediction network;

[0040] The intervention path of the abnormal drying point is set based on the side drying trajectory;

[0041] The abnormal intervention mechanism of spray drying is set in combination with the dilemma response layer and the intervention path.

[0042] Preferably, the target drying parameter of spray drying is set based on the current environment state and the solution characteristics, including:

[0043] The temperature distribution data of spray drying is extracted based on the current environment state;

[0044] The influence of the temperature distribution data on drying stability is analyzed, and the correlation between temperature and drying efficiency is identified;

[0045] The concentration change rate of the solution is calculated based on the solution characteristics;

[0046] The distribution balance of drying parameters is adjusted in combination with the temperature distribution data and the concentration change rate;

[0047] The target drying parameter of spray drying is set according to the drying stability requirement and the distribution balance.

[0048] Preferably, the target drying parameter is adaptively adjusted to obtain an adaptive drying parameter, including:

[0049] The parameter offset in the drying process is identified based on the drying road condition information;

[0050] The interference degree of the parameter offset on the drying array is analyzed;

[0051] The parameter compensation mechanism is set according to the relative position parameter;

[0052] The distribution proportion of the target drying parameter is adjusted in combination with the interference degree and the parameter compensation mechanism;

[0053] The adaptive drying parameter is generated based on the adjusted distribution proportion.

[0054] Preferably, the spray drying drying track is identified, comprising:

[0055] Based on the operating parameter data, real-time data points of spray drying are collected;

[0056] The real-time data points are analyzed by using a pattern recognition algorithm, and the data points are divided into corresponding drying modes;

[0057] Based on the distribution of drying modes, the clustering characteristics of data points are identified;

[0058] According to the clustering characteristics and drying mode changes, the drying track of spray drying is determined.

[0059] Preferably, the spray drying process is subjected to multivariate control processing to obtain a spray drying control result, comprising:

[0060] Based on the abnormal perception mode, the drying dilemma of spray drying is identified;

[0061] According to the drying dilemma, the trigger condition of the abnormal intervention mechanism is set;

[0062] Based on the trigger condition and the preset drying rule, the multivariate control parameters are adjusted;

[0063] According to the adjusted multivariate control parameters, the spray drying control is executed;

[0064] A spray drying control result is generated.

[0065] Compared with the prior art, the beneficial effects of the present application are:

[0066] Through the data acquisition module, the characteristics of the solution to be dried and the operating parameters of the equipment are comprehensively obtained, the current environmental state and the solution characteristics can be accurately identified, and comprehensive and reliable basic information is provided for subsequent parameter setting and control adjustment. Compared with the traditional control method, it is no longer limited to the monitoring of a single or a few parameters, but realizes the comprehensive perception of various factors affecting the drying process, so that the system can more comprehensively understand the initial conditions and dynamic changes of the drying process.

[0067] The drying parameter determination module sets the target drying parameters and determines the initial drying array based on the obtained environmental state and solution characteristics, fully considering the individualized needs of different solution characteristics and environmental states on the drying process. This method avoids the limitations brought by fixed parameters or simple experience parameters in traditional control, and can develop a more demand-oriented initial scheme for the drying process according to the actual situation, so that the spray drying equipment is in a more reasonable running state at the start-up stage, which helps to reduce the problem of poor initial drying effect caused by parameter mismatch.

[0068] The dry state sensing module combines the initial drying array and the preset drying rule, senses the real-time state of drying in real time, analyzes the environmental change trend, and then sets an abnormal sensing mode, thereby realizing dynamic monitoring and abnormal early warning of the drying process. By tracking the drying state in real time, the system can discover subtle changes in the drying process in a timely manner, and the setting of the abnormal sensing mode in combination with the environmental change trend enhances the prediction ability of potential abnormal conditions, changes the situation of the lagging reaction of the traditional control system to abnormal conditions, and enables the system to make intervention preparations at the initial stage or even before the abnormal conditions occur.

[0069] The drying abnormal intervention module identifies abnormal conditions based on the abnormal sensing mode, and sets an abnormal intervention mechanism in combination with the preset trajectory prediction network, which can quickly take targeted intervention measures when abnormalities are found. This intervention mechanism is not a simple parameter adjustment, but a systematic adjustment scheme based on accurate identification of abnormal conditions and prediction of their development trend, which can more effectively curb the development of abnormal conditions and avoid greater impact on product quality and production process.

[0070] The multivariable control module comprehensively controls the drying process by combining the abnormal sensing mode, the abnormal intervention mechanism and the preset drying rule, and realizes the coordinated adjustment of multiple influencing parameters. This multivariable control method fully considers the interaction and influence between parameters, avoids the chain problems that may be caused by single parameter adjustment, optimizes the efficiency of the drying process under the premise of ensuring product quality, and reduces energy consumption and raw material waste. At the same time, the system can adapt to the differences in characteristics of different organic aqueous solutions and changes in environmental conditions, maintain stable drying effect in various complex working conditions, and does not need to rely too much on manual operation, thereby reducing the requirement for the experience of operators and improving the automation and intelligence level of the spray drying process. BRIEF DESCRIPTION OF DRAWINGS

[0071] Figure 1 A timing diagram of the organic aqueous solution spray drying multivariable control system described in the present application;

[0072] Figure 2 A flowchart for determining the initial drying array;

[0073] Figure 3 A flowchart for analyzing the environmental change trend;

[0074] Figure 4 A flowchart for setting the target drying parameter;

[0075] Figure 5 A flowchart for generating the adaptive drying parameter. DETAILED DESCRIPTION

[0076] With reference to the accompanying drawings: there will be clearly and completely described the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by a person of ordinary skill in the art without creative work fall within the scope of the present application.

[0077] With reference to the accompanying drawings: Figure 1 The present application provides a multivariate control system for spray drying of organic aqueous solution, which comprises a data acquisition module, a drying parameter determination module, a drying state perception module, a drying abnormality intervention module and a multivariate control module.

[0078] The data acquisition module acquires the physical property parameters of the organic aqueous solution and the operating parameters of the spray drying equipment, and identifies the current environmental state and the solution characteristics. The drying parameter determination module sets the target drying parameters according to the environmental state and the solution characteristics, and determines the initial drying array in combination with the operating parameters and the relative position parameters. The drying state perception module monitors the drying state and the environmental change trend in real time, and sets the abnormality perception mode. The drying abnormality intervention module identifies the drying abnormality based on the abnormality perception mode, and formulates an intervention mechanism through a preset trajectory prediction network. The multivariate control module comprehensively controls the spray drying process based on the abnormality perception mode, the intervention mechanism and the preset drying rules, and outputs the final control result.

[0079] Embodiment 1: With reference to the accompanying drawings: Figure 2 The network comprises a temperature sensor array, a humidity detection unit, an ultrasonic flowmeter and a laser anemometer, covering all key areas of the drying tower. The temperature sensor adopts a platinum resistance probe, which collects hot air distribution data every 0.5 seconds; the humidity detection unit is based on a capacitive sensing principle, which monitors the water content of the airflow in real time; the ultrasonic flowmeter tracks the supply rate of the organic aqueous solution; and the laser anemometer records the airflow vector information. All data are transmitted to the central processing unit through the RS-485 bus, and after removing the abnormal values through the data cleaning module, the data are aligned through the sliding time window mechanism.

[0080] The three-dimensional data of temperature, humidity and flow rate are standardized, and the time offset of the time series data is eliminated through the dynamic time warping algorithm. Then, the data points are divided according to the Gaussian mixture model, and the optimal clustering number is automatically determined according to the Bayesian information criterion. Each cluster center corresponds to a drying mode, including a fast evaporation zone, a constant speed drying zone and a decreasing speed drying zone. The system identifies the main path and branch path of the drying trajectory by calculating the spatial distribution density of the cluster centers. The evolution trend of the drying trajectory is predicted through the hidden Markov model, and the model state transition probability is generated based on the historical drying data training.

[0081] The dry road condition analysis engine analyzes the airflow distribution uniformity by using computational fluid dynamics simulation results as a benchmark, quantifying the road condition fluctuation coefficient by comparing the real-time airflow velocity distribution with the ideal model. The hot air temperature gradient analysis is based on infrared thermal imaging data, establishing a three-dimensional thermal map of the temperature field. The target drying parameters, including atomization pressure set value, inlet air temperature threshold, and solution flow rate range, are dynamically corrected by an adaptive filtering algorithm. The core of the algorithm uses a Kalman filter, and its process noise covariance matrix is adjusted in real time according to the relative position parameters.

[0082] The parameter offset calculation module analyzes the deviation of actual parameters in each region from the target value, and evaluates the interference strength of the offset on the drying array through variance analysis. The parameter compensation mechanism is designed as a two-level control: the primary compensation uses a proportional-integral controller to adjust the compensation gain in real time according to the interference strength; the advanced compensation introduces a feedforward control link to correct the parameters in advance based on the drying road condition prediction model. The compensated drying parameters are normalized to form an adaptive drying parameter set.

[0083] The drying tower internal space is divided into 1 cm³ voxel units, each unit associated with an adaptive drying parameter set. A three-dimensional parameter distribution matrix is generated through tensor operations, with each element in the matrix containing temperature set value, humidity threshold, and airflow velocity range. This matrix is stored in a real-time database as the basic control framework for the drying process.

[0084] The data preprocessing stage uses median filtering smoothing processing, and calculates the probability distribution of normal drying parameters through kernel density estimation. The normal parameter range is defined as the distribution band with a confidence interval of 95%, including the atomization pressure fluctuation threshold, particle moisture content change curve, and evaporation rate standard value. The real-time monitoring system obtains the particle moisture content through an online laser particle size analyzer and calculates the drying rate through a differential pressure sensor, with a sampling frequency of 10 Hz.

[0085] The deviation quantification index includes shape difference and numerical offset, and a comprehensive deviation coefficient is generated by weighted fusion. The abnormal parameter threshold setting system is based on a reinforcement learning model, which continuously optimizes the threshold boundary through the Q-learning algorithm. When the comprehensive deviation coefficient exceeds the dynamic threshold, the system marks the drying abnormality and triggers a three-level early warning mechanism: the first level adjusts the parameter compensation strength, the second level starts the backup drying scheme, and the third level executes the emergency shutdown program.

[0086] After each drying process, the system automatically adds the current parameter data to the training set, re-trains the Gaussian mixture model and the hidden Markov model. The model update adopts an incremental learning method, retaining the historical model characteristics while incorporating new data patterns. The false alarm rate control of the anomaly detection module is achieved through confusion matrix analysis, continuously optimizing the split threshold of the isolation forest algorithm. The entire perception process forms a closed-loop control system, allowing the accuracy of anomaly recognition to continuously evolve over time.

[0087] Embodiment 2: refer to Figure 3 The attribute classification engine uses a random forest-based classification model to dynamically divide the original data stream into temperature data sets, humidity data sets, airflow velocity data sets, and solution physical property data sets. The temperature data set includes multi-point temperature records at the inlet, constant temperature zone, and outlet of the drying tower; the humidity data set covers the inlet air humidity, exhaust air humidity, and relative humidity gradient in the tower; the airflow velocity data set records the axial wind speed, radial wind speed, and vortex intensity; and the solution physical property data set includes concentration, viscosity, and surface tension parameters. The classified data sets enter the feature extraction link, which uses principal component analysis technology to reduce the data dimension and extract key feature vectors representing the environmental state. The key feature vectors include temperature distribution dispersion coefficient, humidity change slope, airflow uniformity index, and solution concentration fluctuation amplitude.

[0088] After receiving the key feature vectors, the environmental change trend analysis module starts the historical environment database retrieval. The database stores snapshots of the environmental state for the past 72 hours, organized in a ring buffer structure in chronological order. The temperature change rate analysis uses a sliding window difference method to calculate the change amplitude of the temperature gradient within adjacent time windows. Meanwhile, by calculating the fluid mechanics simulation model, the real-time airflow data are compared with the standard turbulent flow model to identify the characteristic parameters of the airflow behavior pattern, including the vortex core position, recirculation zone range, and boundary layer separation point. The humidity change level analysis uses an adaptive threshold method to establish a correlation curve of humidity change rate and time variable, and judges the acceleration trend of humidity change through the second derivative of the curve.

[0089] The temperature change rate, airflow behavior pattern characteristic parameters, and humidity change level are input into the time series prediction model, which is based on a Gated Recurrent Unit network. The network input layer receives the standardized feature vectors, the hidden layer contains 128 neurons, and the output layer generates a comprehensive score value of the environmental change trend. The score value is mapped to the 0-100 interval, and a score above 60 indicates a significant change risk in the environmental state. The score result is transmitted in real time to the anomaly perception mode configuration unit as the basis for adjusting the mode sensitivity.

[0090] The laser dust sensor is installed at the inlet of the drying tower to detect the concentration of PM2.5 in real time. The silicon photodiode array monitors the intensity of ambient light, with a resolution of 0.1 lux. The ultrasonic anemometer is deployed in the periphery of the equipment to capture external airflow interference vectors. The interaction analysis between environmental factors and spray drying uses a multiple regression model to establish the correlation matrix of dust concentration and airflow cleanliness, the response surface of light intensity and solution photosensitivity, and the coupling coefficient of external airflow and flow field in the drying tower. The quantified results of the interaction are stored as a correlation coefficient tensor, with the tensor dimensions corresponding to the combined influence weights of different environmental factors.

[0091] The identification engine of potential risk factors loads the correlation coefficient tensor and calculates the risk value of each factor through a risk probability model. The model input is the real-time environmental factor readings and the interaction tensor, and the output is a risk level matrix. The risk level is divided into four levels: first-level risk corresponds to slight fluctuations in environmental factors, second-level risk represents significant changes in a single factor, third-level risk represents the coordinated deterioration of multiple factors, and fourth-level risk indicates a systemic environmental failure. The configuration of risk response mode uses a fuzzy logic controller, with real-time drying state parameters and risk levels as input variables and drying parameter adjustment strategies as output variables. The controller defines 49 fuzzy rules, such as triggering the inlet temperature increase strategy when detecting a third-level risk and an increase in particle moisture content.

[0092] The first-level decision is based on the risk level matrix to generate the basic monitoring frequency: first-level risk corresponds to regular monitoring (1 Hz sampling), second-level risk activates enhanced monitoring (5 Hz sampling), and third-level risk and above activates high-frequency monitoring (20 Hz sampling). The second-level decision combines the risk response mode to configure the monitoring parameter combination: temperature-related risks focus on hot air distribution, humidity risks focus on dew point detection, and airflow risks intensify vortex monitoring. The third-level decision dynamically adjusts the weight coefficients of each sensor through a neural network prediction model. The final abnormal perception mode contains three levels of early warning mechanisms: primary early warning prompts parameter adjustment, intermediate early warning requires intervention verification, and high-level early warning triggers system review.

[0093] After each drying process, the system compares the predicted environmental trend with the actual recorded data and calculates the trend prediction error. The error data is input into the training algorithm of the gated recurrent unit network, which updates the network weights using the backpropagation method. At the same time, the interaction analysis model performs offline optimization once a month using accumulated environmental data to recalculate the correlation coefficient tensor. The risk probability model is continuously updated through online learning, and the risk value calculation parameters are automatically adjusted whenever a new drying abnormality case is added.

[0094] The system builds a virtual model of the spray drying process and compares real-time running data with the virtual model simulation results regularly. When a significant deviation is found between the perceived mode and the actual working condition, the mode reconstruction program is started. The reconstruction process retains the effective parameters of the historical configuration and optimizes the deviation link locally to ensure that the perceived mode is always synchronized with the actual state of the equipment. The entire system establishes a configuration version management mechanism to record the timestamp and change content of each mode update, forming a complete mode evolution archive.

[0095] Embodiment 3: refer to Figure 4 The positioning system loads the real-time drying state data stream and uses an improved isolation forest algorithm to build an anomaly detection model. The model sets 100 isolation trees, and each tree randomly selects 32 feature subsets. By calculating the path length difference of the data points, an anomaly score matrix is generated. When the anomaly score of a data point exceeds the dynamic threshold, the system marks it as a candidate abnormal point. After spatial clustering analysis of the candidate points, the three-dimensional coordinates of the abnormal drying points are confirmed. The coordinate data includes the axial position, radial distance and height information of the drying tower, with an accuracy of millimeters.

[0096] The network input layer receives temperature gradient, humidity distribution and airflow vector data in a 5cm³ area around the abnormal point. The hidden layer includes three special analysis channels: the temperature channel analyzes the thermal inertia coefficient and heat transfer efficiency, the humidity channel calculates the moisture diffusion rate and latent heat of phase change, and the airflow channel evaluates the Reynolds stress and vorticity intensity. The output layer generates a probability distribution of the dilemma type, and when the probability of a certain type exceeds 85%, it is determined as an effective diagnosis. The system presets three main dilemmas: temperature imbalance corresponds to a decrease in heat transfer efficiency of more than 30% of the baseline value, humidity exceeds the local relative humidity continuously above the set threshold of 15%, and airflow turbulence is characterized by a sudden change in vortex intensity exceeding 2 times the standard deviation of the historical average.

[0097] The temperature imbalance response unit includes three levels of response: the primary response adjusts the heater power of the partition where the abnormal point is located, with a change range of ±10%; the intermediate response starts thermal compensation in adjacent areas to balance the temperature field through hot air redistribution; the advanced response triggers intermittent spraying of cooling nozzles to prevent material overheating degradation. The humidity exceeding unit deploys a humidity gradient control system to automatically select a dehumidification scheme according to the degree of exceeding: mild exceeding activates the molecular sieve adsorption module, moderate exceeding starts the condensing dehumidification unit, and severe exceeding executes the airflow replacement program. The airflow turbulence response unit integrates a matrix of guide plates, with 128 intelligent guide plates controlled independently by servo motors. The guide plate angle adjustment algorithm is based on fluid mechanics simulation to generate the optimal angle combination to restore the airflow to a laminar state.

[0098] The network input contains device geometric topology data, real-time operating parameters and historical drying trajectories. The graph convolution layer constructs a three-dimensional grid model of the drying tower, with nodes representing spatial positions and edge weights representing mass transfer correlations between positions. The convolution kernel extracts local features in the spatial dimension, and the temporal recurrent layer captures the evolution law of the trajectory. The network output is a set of drying trajectories on both sides of the abnormal drying point, each trajectory containing a sequence of coordinates and a confidence score. The trajectory screening mechanism uses non-dominated sorting to retain the top 5 candidate trajectories in the Pareto optimal set.

[0099] Intervention path optimization introduces multi-objective function :

[0100]

[0101] wherein: E(t) represents the energy consumption function at time t, E(t) represents the drying efficiency function, and are dynamic weight coefficients. The path optimization uses an improved ant colony algorithm, with 200 virtual ants performing parallel search. The pheromone update rule introduces a simulated annealing mechanism to avoid local optimal solutions. The algorithm outputs the key node sequence of the intervention path, and the node spacing is adjusted adaptively according to the drying rate.

[0102] The model defines seven states: standby monitoring, anomaly identification, dilemma classification, trajectory generation, path optimization, intervention execution, and effect evaluation. The state transition conditions are dynamically set based on real-time sensor data, for example, the condition for migrating from the dilemma classification state to the trajectory generation state is that the dilemma diagnosis confidence exceeds 90%. The intervention execution stage uses a distributed actuator network, with 32 control terminals synchronously receiving path instructions. The temperature intervention terminal controls 128 heating plates, the humidity intervention terminal manages 24 dehumidification units, and the airflow intervention terminal operates the guide plate matrix. The execution process uses a gradual adjustment strategy, adjusting the intervention intensity by 10% every 5 seconds to avoid parameter mutations.

[0103] The infrared thermal imaging system scans the drying tower at a frequency of 10Hz, generating temperature distribution point cloud data. After voxelization, the point cloud is reconstructed into a three-dimensional temperature field using a heat conduction inversion algorithm. The solution property monitoring unit analyzes the concentration change through an online ultraviolet-visible spectrometer, with a spectral range of 200-800nm. The concentration change rate is calculated using the sliding window derivative method, with the window width adaptively adjusted according to the solution viscosity.

[0104] The coupling matrix of temperature field and concentration field is established, and the optimal parameter distribution is solved by Jacobi iteration method. The objective function considers the drying stability coefficient and the distribution uniformity index, and the constraint condition includes the equipment safety threshold and the material heat sensitivity. The final generated target drying parameter set contains 256 control variables, which are distributed to each execution unit through the control bus. The parameter update mechanism adopts the version rolling strategy, which retains the last three groups of parameter configurations for quick rollback.

[0105] The primary verification collects temperature recovery rate, humidity drop slope and airflow stability through the micro sensor network installed in the intervention area in real time. The secondary verification starts the parallel simulation of the digital twin system, compares the actual intervention effect with the virtual model prediction, and triggers the online optimization cycle of intervention parameters when the actual effect deviates from the predicted value by more than 15%. Each intervention process forms a complete operation log, including timestamp, intervention type, execution parameter and effect index, which is used to continuously improve the prediction network accuracy.

[0106] The state space is defined as the drying abnormal feature vector, the action space corresponds to the intervention strategy combination, and the reward function is based on the comprehensive evaluation of intervention effect and energy consumption. The intelligent agent uses a double deep Q network architecture, and updates the network parameters every 10 interventions. The experience replay buffer stores the last 1000 intervention records, and preferentially samples the cases with abnormal effects for training. The network update adopts a soft synchronization strategy, and the target network parameters are synchronized every 50 updates.

[0107] Embodiment 4: refer to Figure 5 The drying road condition information is collected through a multi-sensor fusion system. In a spray drying tower with a diameter of 2 meters and a height of 5 meters, 32 wireless sensor nodes are deployed to form a monitoring network. The node spacing is arranged every 0.5 meters along the axial direction, and 8 nodes are evenly distributed on the circumference of each layer. Each node integrates temperature, humidity, and airflow three-in-one sensors with a sampling frequency of 20Hz. The drying road condition analysis engine processes sensor data in real time to generate a three-dimensional road condition heat map. The heat map divides the drying tower into 8 independent control areas, and each area is labeled with airflow uniformity level (1-5) and temperature stability index (0-1.0).

[0108] Parameter offset measurement uses ultra-wideband positioning technology. Positioning tags are installed at the atomizing nozzles, and 4 base stations are placed at the four corners of the drying tower. The system records the spatial coordinates of the nozzles with a time resolution of 100ms, and calculates the Euclidean distance difference between the actual position and the theoretical trajectory. The offset analysis module converts the distance difference into a parameter influence factor: when the offset is less than 5mm, the influence factor is 0.1, 5-10mm is 0.3, 10-20mm is 0.6, and more than 20mm is 1.0. The interference degree evaluation uses a regional correlation model to establish a weight relationship matrix between the offset position and each control area.

[0109] The parameter compensation mechanism is based on an adaptive neural network. The input layer of the network contains three types of data: real-time position coordinates (x, y, z), interference influence factors (δ), and regional weight matrices (W). The hidden layer is designed as a dual-channel structure: the spatial channel processes the relationship between the position and the weight, and the interference channel analyzes the transmission characteristics of the influence factors. The output layer generates a compensation coefficient matrix C for 8 control regions, with a coefficient range of 0.8-1.2. The network training uses real-time incremental learning, and after each drying process is completed, the network weights are adjusted in reverse according to the actual drying effect.

[0110] The distribution ratio adjustment of the target drying parameters is realized through tensor operations. The initial target parameters are stored as an 8x3 matrix: each row represents a region, and the three columns represent the temperature set value, humidity threshold, and air flow speed range. The compensation coefficient matrix C is subjected to Hadamard product operation with the target parameter matrix to generate a compensation parameter matrix. The adjusted parameters are normalized: the temperature parameters are scaled by equal ratio, the humidity parameters take the regional average, and the air flow parameters keep the range unchanged. The final adaptive drying parameter set is written to the control executor and updated every 30ms.

[0111] The drying trajectory recognition system uses an improved K-means clustering algorithm. The algorithm sets a dynamic number of cluster centers, with an initial value of 3 core modes (atomization zone, heat transfer zone, and settling zone). The real-time data point collection frequency is 50Hz, and each data point contains four-dimensional features of spatial coordinates, temperature, humidity, and air flow speed. The data preprocessing uses a sliding window standardization, and the window width is dynamically adjusted according to the drying stage: 10 seconds for the atomization stage, 20 seconds for the heat transfer stage, and 30 seconds for the settling stage.

[0112] The clustering process introduces a dynamic time warping technique. The algorithm establishes two similarity measurement pools: a spatial distance pool to calculate the Euclidean distance, and a mode similarity pool to evaluate the matching degree of drying characteristics. When a new data point arrives, the two distances between it and each cluster center are calculated simultaneously, and the weighted fusion determines the belonging category. The weight distribution follows the spatial priority principle: the spatial weight is 0.7 and the mode weight is 0.3 in the atomization stage, 0.5 / 0.5 in the heat transfer stage, and 0.3 / 0.7 in the settling stage.

[0113] The clustering feature extraction uses the density peak detection method. The system scans the clustering results every 5 seconds, and calculates the core density, boundary gradient, and stability index of each category: the number of data points per unit volume, the steepness of the inter-class boundary, and the inverse of the class center movement distance.

[0114] The system constructs a transition matrix of cluster centers, recording the migration path of each cluster center in consecutive time slices. When a new cluster center is detected or an old center disappears, a trajectory reconstruction program is started. The reconstruction algorithm is based on the Bayesian filtering principle, which integrates the current cluster distribution and historical transition probability to generate the optimal drying trajectory prediction. Trajectory data is stored as an ordered coordinate sequence, with 10 trajectory point coordinates output per second. See Table 1.

[0115] Table 1: Example of area parameter compensation effect.

[0116]

[0117] The system maintenance module implements periodic self-checking. The sensor calibration program is started at midnight every day: heat the standard temperature source to 200°C, introduce standard humidity gas, and record the measurement deviation of 32 nodes. The calibration data is used to compensate for subsequent measurement values. The clustering algorithm parameters are optimized every week, and the initial cluster centers are recalculated using the drying data of the week. The neural network is trained offline every month, and historical data is loaded to enhance the model's generalization ability. All maintenance operations are recorded in the blockchain log, forming an unalterable device health record.

[0118] Visualization of trajectory recognition results is achieved through an augmented reality interface. The operator wearing AR glasses can observe the three-dimensional trajectory cloud chart inside the drying tower, with different colors marking the atomization, heat transfer, and sedimentation trajectory segments. The trajectory thickness represents the data point density, and dynamic flickering indicates abnormal fluctuation areas. The interface supports gesture operation to scale and rotate the model, and key parameters are displayed in real time on the corresponding spatial position.

[0119] Example 5: The anomaly perception mode identifies drying difficulties through a multi-dimensional scanning mechanism. The system loads real-time drying state data streams and uses a convolutional neural network to analyze three-dimensional parameter distribution maps. The network input layer receives the fusion data of temperature field heat maps, humidity gradient cloud maps, and airflow vector fields. The feature extraction layer contains eight parallel convolution channels, which capture spatial patterns of different scales. The classification output layer generates a difficulty probability distribution: local overheating corresponds to a sustained expansion pattern in high-temperature regions, drying unevenness is characterized by a standard deviation exceeding the humidity distribution, and particle agglomeration is characterized by bimodal particle size distribution. The difficulty recognition result is accompanied by a confidence score, and when the score exceeds 90%, the subsequent control process is triggered.

[0120] The trigger condition configuration of the abnormal intervention mechanism adopts a fuzzy rule engine. The engine input variables include the dilemma type, severity index, and diffusion trend coefficient. The severity index is calculated by weighting indicators such as the proportion of high-temperature areas and the standard deviation multiple of humidity; the diffusion trend coefficient is based on the output of a time series prediction model. The rule base is preset with 128 trigger logics, such as when the local overheating type is accompanied by a severity index > 0.7 and a diffusion trend coefficient > 1.2, the highest level of intervention response is activated. The trigger threshold setting has an adaptive mechanism that dynamically adjusts the threshold boundary based on historical intervention effects to avoid over-response.

[0121] The adjustment of the multivariable control parameters implements a model predictive control strategy. The controller establishes a state space model containing 256 variables, including temperature setpoint, air flow rate, atomization pressure, and other key parameters. A rolling horizon optimization is performed every 50 milliseconds: within the current control period, based on the drying dilemma characteristics and the preset drying rules, the optimal parameter trajectory for the next 10 seconds is solved. The optimization objective function balances drying efficiency and energy consumption indicators, and the constraint conditions include device physical limits and material safety thresholds. The solving process uses a parallel gradient descent algorithm, which iterates synchronously on 8 computing cores.

[0122] The execution of spray drying control adopts a partitioned collaborative strategy. The drying tower is divided into 8 control zones along the axial direction, and each zone is equipped with an independent execution unit. The central controller decomposes the adjusted parameter set into regional instruction sets and distributes them through real-time industrial Ethernet. The temperature control unit uses pulse width modulation technology to accurately adjust the power output of 128 groups of heating elements; the air flow control unit operates 64 variable frequency fans to achieve wind speed precision adjustment at the level of 0.1 m / s; the atomization system controls the nozzle aperture through a piezoelectric ceramic actuator, with a regulation range of 50-200 μm. The execution process implements a cross-checking mechanism, in which sensors in adjacent regions monitor each other's execution effects.

[0123] The generation of spray drying control results fuses multi-source evaluation data. The online laser particle size analyzer scans 1000 particles per second to generate a particle size distribution spectrum; the near-infrared moisture analyzer measures the moisture content of the particles in real time, with a sampling depth of 3 mm; the high-speed camera captures the particle motion trajectory to calculate the drying uniformity index. The data fusion engine uses evidence theory to handle uncertain information, and when different sensor data conflict, it starts a confidence weighted arbitration mechanism. The final control results are output as a structured report, including three-dimensional spatiotemporal distribution graphs and statistical characteristic values of key indicators.

[0124] The visual monitoring interface constructs an immersive operation environment. A 55-inch curved display presents a transparent model of the drying tower, with color streamlines showing airflow movement, a thermal map superimposed on the temperature field, and a semi-transparent cloud effect rendering the humidity distribution. The operator can control the rotation viewing angle by gestures, and hover at any position to display the real-time parameters at that point. The alarm system uses a hierarchical visual prompt: a yellow flashing border for first-level warning, a 3D arrow pointing to the abnormal area for second-level warning, and a full-screen red pulsating warning for third-level warning. The historical data playback function supports 50 times speed simulation of the drying process evolution.

[0125] The self-optimization mechanism of the control system is implemented through a double-loop. The inner loop monitors the deviation of the control result from the expected target in real time, and automatically fine-tunes the weight parameters of the model predictive control when the key indicators deviate from the set range by more than 5%. The outer loop starts after each drying process is completed, analyzes the control effect throughout the process, and updates the structural parameters of the state-space model. Model updating uses transfer learning technology, preserving the general feature layer weights and only fine-tuning the specific working condition adaptation layer.

[0126] The operation log system implements holographic recording. The timestamp, decision basis, execution parameters, and feedback data of each control decision are written into a time series database. The log adopts a hierarchical storage strategy: real-time data is retained for 7 days, feature data is retained for 3 months, and statistical summaries are permanently saved. The log analysis engine automatically detects abnormal operation patterns, and triggers the system self-check program when it finds that the parameter adjustment frequency is abnormally high or the actuator is running at full capacity.

[0127] The version control module manages the iterative updates of the control algorithm. Each algorithm modification generates an independent version number, recording the change content and test cases. The system retains three parallel versions during operation: the stable version is used in the production environment, the test version is verified in the virtual drying tower, and the development version is provided for algorithm engineers to debug. Version switching uses a gray release mechanism, first running for 24 hours in a single control zone, and then promoting to the whole tower after confirmation. All version changes form a timeline map, which intuitively shows the evolution path of the control strategy.

[0128] The fault safety mechanism designs multiple protections. When the main control loop is abnormal, it automatically switches to the backup PLC system; in the case of communication interruption, each partition execution unit maintains operation according to the last valid parameters; when the emergency stop button is triggered, the system performs a three-step safety procedure: cutting off the heating power, starting the cooling fan, and injecting inert gas. The safety state monitoring board displays the working status of 16 key components in real time, with each status verified by dual-channel independent sensors.

[0129] It is to be understood that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting; it is not intended to exclude myriad other embodiments of the present application that other inventors can develop based on the same general inventive concepts embodied by the described embodiments. That is, although the present application is described in terms of particular embodiments and illustrative figures, it should be apparent that the scope of the present application is not limited to these specific embodiments.

[0130] While the embodiments of the application have been shown and described herein, it will be understood by those of ordinary skill in the art that various changes, modifications, alternatives, and variations can be made to the embodiments without departing from the spirit and scope of the application, which is defined by the appended claims and their equivalents.

Claims

1. A multivariate control system for spray drying an aqueous organic solution, comprising: The method comprises the following steps: A data acquisition module is used to acquire an organic matter aqueous solution to be spray dried and collect operation parameter data of a spray drying device, and to identify a current environmental state and solution characteristics of a spray drying process according to the operation parameter data; A drying parameter determination module is used to set target drying parameters of spray drying based on the current environmental state and the solution characteristics, to locate relative position parameters of the spray drying device in a drying process, and to determine initial drying patterns of spray drying based on the operation parameter data and the relative position parameters; A drying state sensing module is used to combine the initial drying patterns and preset drying rules to sense a real-time drying state of spray drying in real time, to analyze an environmental change trend of the current environmental state based on the operation parameter data, and to set an abnormal sensing mode of spray drying in combination with the real-time drying state and the environmental change trend; A drying abnormality intervention module is used to identify a drying abnormality of spray drying based on the abnormal sensing mode, and to set an abnormal intervention mechanism of spray drying according to the drying abnormality and a preset trajectory prediction network; A multivariable control module is used to perform multivariable control processing on a spray drying process according to the abnormal sensing mode, the abnormal intervention mechanism and the preset drying rules, and to obtain a spray drying control result; The setting of the target drying parameters of spray drying based on the current environmental state and the solution characteristics comprises the following steps: Temperature distribution data of spray drying are extracted based on the current environmental state; The influence of the temperature distribution data on drying stability is analyzed, and the correlation between temperature and drying efficiency is identified; The concentration change rate of the solution is calculated based on the solution characteristics; The drying parameter distribution uniformity is adjusted in combination with the temperature distribution data and the concentration change rate; The target drying parameters of spray drying are set according to drying stability requirements and distribution uniformity; The self-adaptive adjustment of the target drying parameters to obtain self-adaptive drying parameters comprises the following steps: Parameter offsets in a drying process are identified based on the drying road condition information; The interference degree of the parameter offsets on drying patterns is analyzed; A parameter compensation mechanism is set according to the relative position parameters; The distribution proportion of the target drying parameters is adjusted in combination with the interference degree and the parameter compensation mechanism; Self-adaptive drying parameters are generated based on the adjusted distribution proportion; The identification of the drying trajectory of spray drying comprises the following steps: Real-time data points of spray drying are collected based on the operation parameter data; The real-time data points are analyzed by using a pattern recognition algorithm, and the data points are divided into corresponding drying modes; The clustering characteristics of the data points are identified based on the distribution of the drying modes; The drying trajectory of spray drying is determined according to the clustering characteristics and the change of the drying modes.

2. The multi-variable control system for spray drying of an aqueous organic solution of claim 1, wherein, The determination of the initial drying patterns of spray drying based on the operation parameter data and the relative position parameters comprises the following steps: The drying trajectory of spray drying is identified based on the operation parameter data; The drying road condition information and the target drying parameters of spray drying are scheduled according to the drying trajectory. Adaptively adjusting the target drying parameter based on the drying road condition information and the relative position parameter to obtain an adaptive drying parameter; Determining an initial drying array of spray drying in combination with the drying road condition information and the adaptive drying parameter.

3. The multi-variable control system for spray drying of an aqueous organic solution of claim 1, wherein, The abnormal perception mode is based on the identification of the drying abnormality of spray drying, including: Based on the abnormal perception mode, the historical drying data of spray drying is collected; According to the historical drying data, the normal drying parameter of spray drying is identified; Identify the current drying state of spray drying, and identify the current behavior deviation of spray drying in combination with the current drying state and the historical drying data; Based on the normal drying parameter, set the abnormal parameter threshold of spray drying; According to the abnormal parameter threshold and the current behavior deviation, the drying abnormality of spray drying is identified.

4. The organic-based aqueous solution spray-drying multivariable control system of claim 1, wherein, The environmental change trend of the current environment state is analyzed based on the running parameter data, including: Attribute classification is performed on the running parameter data to obtain classified parameter data; Extract the key features of the classified parameter data, and collect the historical environment data of the current environment state based on the key features; Identify the temperature data and humidity data in the historical environment data; According to the temperature data, analyze the temperature change rate and airflow behavior mode of the current environment state; Based on the humidity data, analyze the humidity change level of the current environment state; In combination with the temperature change rate, the airflow behavior mode and the humidity change level, the environmental change trend of the current environment state is analyzed.

5. The organic-based aqueous solution spray-drying multivariable control system of claim 1, wherein, The abnormal perception mode of spray drying is set in combination with the real-time drying state and the environmental change trend, including: Based on the real-time drying state, identify the surrounding environmental factors of spray drying; Analyze the interaction between the surrounding environmental factors and spray drying; According to the environmental change trend and the interaction, identify the potential risk factors of spray drying, and identify the risk level of the potential risk factors; According to the real-time drying state and the risk level, set the risk response mode of spray drying; In combination with the risk level and the risk response mode, set the abnormal perception mode of spray drying.

6. The organic-based aqueous solution spray-drying multivariable control system of claim 1, wherein, The abnormal intervention mechanism of spray drying is set according to the drying abnormality and the preset trajectory prediction network, including: According to the drying abnormality, locate the abnormal drying point of spray drying; Analyze the dilemma type of the abnormal drying point, and set the dilemma response layer of the abnormal drying point based on the dilemma type; According to the preset trajectory prediction network, identify the side drying track of the abnormal drying point; Based on the side drying track, set the intervention path of the abnormal drying point; In combination with the dilemma response layer and the intervention path, set the abnormal intervention mechanism of spray drying.

7. The organic-based aqueous solution spray-drying multivariable control system of claim 1, wherein, The multivariate control process of spray drying is processed to obtain the spray drying control result, including: Based on the abnormal perception mode, identify the drying dilemma of spray drying; According to the drying dilemma, set the trigger condition of the abnormal intervention mechanism; Based on the trigger condition and the preset drying rule, adjust the multivariate control parameter; performing spray drying control according to the adjusted multivariate control parameters; generating a spray drying control result.

Images