A soybean oil leach tail gas absorption and resolution system

Abstract

The application relates to the technical field of tail gas treatment, and discloses a soybean oil leaching tail gas absorption and analysis system which comprises an adsorption unit, an analysis unit, a monitoring unit, a communication unit and a cloud platform unit, and each unit is connected in series through a data bus to form a closed-loop control. The soybean oil leaching tail gas absorption and analysis system adopts Fe3 + -TiO2 / ACF composite adsorption-photocatalysis material, and the two can synergistically improve the adsorption selectivity and efficiency; the analysis unit receives real-time data of the monitoring unit, dynamically adjusts and controls the analysis temperature, airflow speed and other parameters through a PID closed-loop control algorithm, solves the problems of low adsorption efficiency, poor selectivity, difficult regeneration and secondary pollution of single activated carbon, the sensor array of the monitoring unit is composed of specifically arranged electrochemical sensors and MOS sensors, synergistically collects component information such as hexane, fatty acid and water vapor in the tail gas, and outputs accurate concentration data through data preprocessing, hyperparameter optimization and model operation by an SSA-SVM optimization algorithm.

Description

Technical Field

[0001] This invention relates to the field of exhaust gas treatment technology, specifically to a soybean oil extraction exhaust gas absorption and desorption system. Background Technology

[0002] The soybean oil extraction process produces exhaust gas containing hexane, fatty acids, and water vapor. Hexane, as an organic solvent, is volatile and toxic, and its direct emission will cause air pollution and harm human health. At the same time, the waste of hexane will also increase production costs.

[0003] Existing technologies for treating soybean oil extraction tail gas mainly include adsorption, condensation, and combustion. Adsorption uses the surface adsorption of porous materials such as activated carbon to capture hexane in the tail gas, and after saturation, solvent recovery is achieved through thermal desorption. Condensation uses a cooling system to cool the tail gas below the dew point, causing hexane to condense into liquid for separation and recovery. Combustion uses high-temperature oxidation to decompose hexane and other organic matter into carbon dioxide and water, achieving the purpose of harmless treatment.

[0004] However, existing soybean oil extraction tail gas treatment technologies still have the following pressing problems that need to be addressed: 1. Adsorbent materials mostly use single activated carbon, which has low adsorption efficiency, poor selectivity for mixed components, and is difficult to regenerate, easily causing secondary pollution; 2. The lack of real-time and accurate monitoring methods makes it impossible to accurately grasp the changes in the concentration of exhaust gas components, resulting in fixed parameters in the adsorption and desorption processes that cannot be dynamically adjusted, leading to incomplete desorption or excessive energy consumption. 3. The lack of an integrated intelligent management platform results in fragmented data, hindering remote monitoring and intelligent early warning, and leading to low operational efficiency. Therefore, a soybean oil extraction tail gas absorption and desorption system was proposed. Summary of the Invention

[0005] (a) Technical problems to be solved To address the shortcomings of existing technologies, this invention provides a soybean oil extraction tail gas absorption and desorption system, which features the ability to absorb and desorb Fe³⁺ gas. + The TiO2 / ACF composite adsorption-photocatalytic material achieves synergistic purification through adsorption and photocatalysis, improving the exhaust gas treatment effect. Precise real-time monitoring of exhaust gas components is achieved through a sensor array and SSA-SVM optimization algorithm. Dynamic adjustment of analytical parameters is achieved through a PID closed-loop control algorithm, balancing thorough analysis with energy consumption optimization. The OneNet customized cloud platform provides advantages such as data visualization, remote monitoring, and intelligent early warning, solving the problems of low adsorption efficiency, insufficient monitoring accuracy, and low level of intelligent management in existing soybean oil leaching exhaust gas treatment technologies.

[0006] (II) Technical Solution To achieve the above through Fe³ + The TiO2 / ACF composite adsorption-photocatalytic material achieves synergistic purification through adsorption and photocatalysis, improving the exhaust gas treatment effect; precise real-time monitoring of exhaust gas components is achieved through sensor arrays and SSA-SVM optimization algorithms; dynamic adjustment of analytical parameters is achieved through PID closed-loop control algorithms, balancing thorough analytical analysis with energy consumption optimization; and data visualization, remote monitoring, and intelligent early warning are achieved through the OneNet customized cloud platform. This invention provides the following technical solution: a soybean oil leaching exhaust gas absorption and analysis system, comprising an adsorption unit, an analysis unit, a monitoring unit, a communication unit, and a cloud platform unit, with each unit connected in series via a data bus to form a closed-loop control. The adsorption unit uses Fe³ + -TiO2 / ACF composite adsorption-photocatalysis material to achieve synergistic effect of exhaust gas adsorption and photocatalytic degradation; The monitoring unit includes a sensor array and an SSA-SVM optimization algorithm, which is used to collect and process exhaust gas component concentration data in real time. The communication unit adopts a dual-mode communication architecture of LoRa and NB-IoT to ensure stable data transmission; The cloud platform unit is custom-developed based on the OneNet IoT platform and has data storage, visualization, and intelligent early warning functions. The analysis unit dynamically adjusts the analysis parameters based on the real-time concentration data output by the monitoring unit using a PID closed-loop control algorithm.

[0007] Preferably, the Fe³⁺ in the adsorption unit + The preparation method of TiO2 / ACF composite adsorption-photocatalytic material includes the following steps: Step 1: Take tetrabutyl titanate and anhydrous ethanol at a volume ratio of 1:2.0-2.5, add acetylacetone as a chelating agent, and the volume ratio of tetrabutyl titanate to acetylacetone is 1.5-1.8:1. Stir magnetically at room temperature for 1.5-2.5 hours to form TiO2 sol. Step 2: Prepare a 0.04-0.06 mol / L Fe(NO3)3·9H2O aqueous solution, according to Fe³ + With Ti 4+ Add TiO2 sol rapidly at a molar ratio of 1:180-220, cool to 0-4℃ in an ice-water bath and stir continuously for 25-35 minutes until the system gels; Step 3: Ag the wet gel at room temperature for 4-6 hours, dry it at 110-130℃ for 10-14 hours, calcine it at 480-520℃ for 2.5-3.5 hours, and cool it to obtain Fe³⁺. + -TiO2 powder; Step 4: Soak ACF in NaOH solution and dilute nitric acid solution for 12-18 min each, wash with deionized water until neutral, and dry at 90-110℃ for 0.8-1.2 h. Step 5: Take 12-18g of Fe³ + - Mix TiO2 powder with 450-550g of deionized water, add 0.8%-1.2% sodium polyacrylate as a binder, and stir to form a dispersion; After immersing the treated ACF in the dispersion for 2-4 minutes, remove it and dry it at 110-130℃ for 0.8-1.2 hours. Repeat this operation 2-4 times to obtain the finished product.

[0008] Preferably, the sensor array of the monitoring unit is arranged as follows: The three electrochemical sensors are hexane electrochemical sensor, fatty acid electrochemical sensor, and water vapor electrochemical sensor, which are evenly arranged in an equilateral triangle with a spacing of 4-6 cm between adjacent sensors; the four MOS sensors are ZnO sensor, WO3 sensor, Fe2O3 sensor, and In2O3 sensor, which are evenly distributed around the circumcircle of the equilateral triangle with a minimum distance of 4-6 cm from the electrochemical sensors. The sensor array has a sampling frequency of 8-12Hz, the ADC has a sampling accuracy of 12 bits, and the signal conditioning circuit includes an operational amplifier with an amplification factor of 90-110 times and a low-pass filter with a cutoff frequency of 0.8-1.2kHz. The fabrication and calibration process of the MOS sensors is as follows: ZnO sensors are prepared by co-precipitation method, WO3 sensors are prepared by low-temperature acid-assisted hydrothermal method, Fe2O3 sensors are prepared by hydrothermal method, and In2O3 sensors are prepared by hydrothermal synthesis of In(OH)3 precursor and calcination at 430-470℃ for 1.5-2.5h; the ceramic substrates of all MOS sensors are ultrasonically cleaned for 15-25min, rinsed with anhydrous ethanol 4-6 times, dried at 140-160℃ for 4-6min, and coated with gas-sensitive slurry with a thickness of 18-22μm; Sensor calibration uses a standard gas concentration gradient of 0-100ppm, with each gradient stabilized for 25-35 minutes. Response values ​​are recorded and calibration curves are established. Calibration error is ≤±2%.

[0009] Preferably, the complete process of the SSA-SVM optimization algorithm for the monitoring unit is as follows: Step 1: Data Acquisition: Construct a dataset containing sensor response data of hexane, fatty acids, and water vapor components in soybean oil leaching tail gas under different concentrations and temperature conditions. The dataset size should be no less than 4,000-6,000 sets. Step 2: Data Preprocessing: The following standardized formula is used for data processing: ; in For standardized data, The original data, The mean of the original data. The number of data points; PCA dimensionality reduction reduced the original 16-dimensional data to 3-5 dimensions, with a cumulative contribution rate of ≥99.4%. Step 3: Algorithm Construction: Set the sparrow population size to 40-60, the maximum number of iterations to 80-120, the safety threshold ST = 0.7-0.9, the ratio of discoverers to joiners to be 2:8-3:7, and optimize the lower limit of the parameters. upper limit ; Step 4: Algorithm training: Search for the optimal hyperparameters using SSA. After training, the support vector machine penalty parameter c = 45-60 and kernel function parameter g = 5-7. Step 5: Algorithm Deployment: The trained model is embedded into the microcontroller of the monitoring unit, with a single data processing time of ≤25ms.

[0010] Preferably, the complete switching process for dual-mode communication of the communication unit includes: Step 1, Communication Quality Monitoring: The communication unit collects the packet loss rate and communication distance of the LoRa module in real time, with a sampling interval of ≤100ms; Step 2, Switching Condition Judgment: When the packet loss rate of 3 consecutive samplings is >1.2-1.6% or the communication distance is >1.8-2.2km, the switch to NB-IoT mode is triggered; When the LoRa module packet loss rate is ≤0.3-0.7% and the communication distance is ≤1.3-1.7km, switch back to LoRa mode; Step 3, Switching Execution: The dual-mode switching delay time is ≤50ms, and a data caching mechanism is used during the switching process to avoid data loss; Step 4: The communication protocol adopts LWM2M. The data frame format includes 1 byte node number, 1 byte function code, 2 bytes number of sensors, 6 bytes sensor address, 1 byte status bit, 4 bytes data and 1 byte checksum.

[0011] Preferably, the complete PID closed-loop control process of the analysis unit is as follows: Step 1, setting control targets: Hexane residual concentration ≤10ppm, proportionality coefficient Integral coefficient Differential coefficients ; Step 2, Parameter adjustment range: Desorption temperature 80-120℃, adjustment accuracy ±1℃; airflow velocity 0.7-1.1m / s, adjustment accuracy ±0.01m / s; Step 3, Stage Division: The analysis process is divided into a preheating stage, a constant temperature analysis stage, and a cooling stage. The heating rate in the preheating stage is 4-6℃ / min. The duration of the constant temperature analysis stage is dynamically adjusted according to the concentration data of the monitoring unit, with an adjustment step of 5-10min. The cooling stage is for natural cooling to room temperature. Step 4, Feedback Adjustment: Collect the analyzed concentration data every 200-300ms and substitute it into the following PID formula to calculate the adjustment amount: ; in For output adjustment amount, For concentration deviation, The integral time constant is... The differential time constant enables dynamic closed-loop regulation.

[0012] Preferably, the complete data processing flow of the cloud platform unit is as follows: Step 1, Data Reception: Data packets are received via NB-IoT or LoRa module, and after verification and validation, gas concentration data and device status information are obtained through parsing. Step 2, Data Storage: Store the parsed data in a MySQL database. The data table contains the fields gas_id, device_id, gas_type, gas_value, gas_unit, and gas_create_time. The gas_value field has a precision of 10 integer digits + 2 decimal digits. Step 3, Data Visualization: Generate line charts and bar charts through the visualization module to display the real-time trends of hexane, fatty acid, and water vapor concentration changes, as well as the equipment operating status. The data update frequency is 1-3 seconds / time. Step 4, Intelligent Early Warning: Set three levels of early warning thresholds. Level 1 warning: when the hexane concentration is ≥38-42ppm, an alarm message will be sent via email and SMS. Level 2 warning: when the concentration is 20-40ppm, an alarm message will be sent via email. Level 3 warning: when the concentration is 10-20ppm, an early warning message will be displayed on the cloud platform interface. The warning information includes the trigger time, sensor number, concentration value, and handling suggestions.

[0013] Preferably, the structural parameters of the adsorption unit are: The adsorption unit adopts a multi-layer honeycomb structure with 2-4 layers, each layer having a thickness of 4-6 cm and a honeycomb pore size of 0.8-1.2 cm. The honeycomb channels form an angle of 25-35° with the exhaust gas flow direction. The adsorption unit shell is made of 304 stainless steel and is equipped with a temperature sensor and a pressure sensor inside. The temperature monitoring range is -25-85℃, the pressure monitoring range is 0-0.12MPa, and the sensor data sampling frequency is 5-8Hz. The Fe³ + The BET specific surface area of ​​the TiO2 / ACF composite adsorption-photocatalytic material is 572-730 m² / g, and the total pore volume is 0.29-0.32 cm³ / g.

[0014] A method for absorbing and desorbing soybean oil leaching tail gas, based on a soybean oil leaching tail gas absorption and desorption system, includes the following steps: Step 1: Pretreatment: After particulate matter is removed from the soybean oil leaching tail gas by a pretreatment device, it enters the adsorption unit at a flow rate of 0.7-0.9 m / s, passing through Fe³⁺. + -TiO2 / ACF composite material is used for adsorption and photocatalytic degradation; Step 2, Monitoring: The monitoring unit collects the concentration data of hexane, fatty acids and water vapor in the exhaust gas at a frequency of 8-12Hz, and obtains accurate concentration values ​​after processing by the SSA-SVM algorithm; Step 3, Communication: The communication unit transmits concentration data and equipment status information to the cloud platform unit in real time, with a transmission delay of ≤1s; Step 4: Cloud Platform Processing: The cloud platform unit stores, analyzes, and visualizes the data, triggering a corresponding warning when the concentration reaches the warning threshold. Step 5, Analysis: The analysis unit, based on the control commands issued by the cloud platform, uses a PID algorithm to regulate the analysis temperature and airflow speed to regenerate the adsorbed saturated composite material. Step 6, Recovery and Discharge: The hexane vapor generated by the desorption process is recovered by a condensation recovery device, and the tail gas is discharged after meeting the standards. The entire adsorption-desorption cycle takes 25-35 minutes.

[0015] (III) Beneficial Effects Compared with the prior art, the present invention provides a soybean oil extraction tail gas absorption and desorption system, which has the following beneficial effects: 1. The soybean oil extraction tail gas absorption and desorption system uses Fe³⁺ as the adsorption unit. + -TiO2 / ACF composite adsorption-photocatalytic material, utilizing the high specific surface area of ​​ACF to adsorb mixed components in exhaust gas, while Fe³ + -TiO2 plays a photocatalytic degradation role, and the two work together to improve adsorption selectivity and efficiency; the desorption unit receives real-time data from the monitoring unit and dynamically adjusts parameters such as desorption temperature and airflow speed through a PID closed-loop control algorithm to accurately regenerate the adsorption-saturated composite material, thus completely solving the problems of low adsorption efficiency, poor selectivity, difficult regeneration, and secondary pollution of single activated carbon.

[0016] 2. The soybean oil leaching tail gas absorption and desorption system has a monitoring unit whose sensor array consists of a specially arranged electrochemical sensor and a MOS sensor. These sensors work together to collect information on components such as hexane, fatty acids, and water vapor in the tail gas. The data is preprocessed, hyperparameters are optimized, and models are calculated using the SSA-SVM optimization algorithm, outputting accurate concentration data. The desorption unit dynamically adjusts the desorption parameters based on this data, replacing the fixed parameter mode. This solves the problem of incomplete desorption or excessive energy consumption caused by the lack of real-time accurate monitoring, achieving a balance between treatment effect and energy consumption.

[0017] 3. In this soybean oil extraction tail gas absorption and analysis system, each unit uses a dual-mode communication architecture of LoRa and NB-IoT in the communication unit to dynamically switch transmission modes according to communication quality, ensuring stable and real-time data transmission to the cloud platform unit. The cloud platform unit realizes centralized data storage and visualization, and sends early warning information in a timely manner through a three-level early warning mechanism, building an integrated intelligent management system to solve the problems of data dispersion, inability to remotely monitor and provide intelligent early warning, and greatly improve the convenience and efficiency of operation and maintenance. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the soybean oil leaching tail gas absorption and desorption system of the present invention; Figure 2 This is a schematic diagram of the process for absorbing and desorbing soybean oil extraction tail gas according to the present invention; Figure 3 This is a flowchart of the PID closed-loop control of the present invention. Detailed Implementation

[0019] The technical solutions of the present invention will be clearly and completely described below with reference to the embodiments and accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] Please see Figure 1-3 A soybean oil extraction tail gas absorption and desorption system includes an adsorption unit, a desorption unit, a monitoring unit, a communication unit and a cloud platform unit, with each unit connected in series via a data bus to form a closed-loop control. The adsorption unit uses Fe³ + -TiO2 / ACF composite adsorption-photocatalysis material to achieve synergistic effect of exhaust gas adsorption and photocatalytic degradation; The monitoring unit includes a sensor array and an SSA-SVM optimization algorithm, which is used to collect and process exhaust gas component concentration data in real time. The communication unit adopts a dual-mode communication architecture of LoRa and NB-IoT to ensure stable data transmission; The cloud platform unit is custom-developed based on the OneNet IoT platform and has data storage, visualization, and intelligent early warning functions. The analysis unit dynamically adjusts the analysis parameters based on the real-time concentration data output by the monitoring unit using a PID closed-loop control algorithm.

[0021] A method for absorbing and desorbing soybean oil leaching tail gas, based on a soybean oil leaching tail gas absorption and desorption system, includes the following steps: Step 1: Pretreatment: After particulate matter is removed from the soybean oil leaching tail gas by a pretreatment device, it enters the adsorption unit at a flow rate of 0.7-0.9 m / s, passing through Fe³⁺. + -TiO2 / ACF composite material is used for adsorption and photocatalytic degradation; Step 2, Monitoring: The monitoring unit collects the concentration data of hexane, fatty acids and water vapor in the exhaust gas at a frequency of 8-12Hz, and obtains accurate concentration values ​​after processing by the SSA-SVM algorithm; Step 3, Communication: The communication unit transmits concentration data and equipment status information to the cloud platform unit in real time, with a transmission delay of ≤1s; Step 4: Cloud Platform Processing: The cloud platform unit stores, analyzes, and visualizes the data, triggering a corresponding warning when the concentration reaches the warning threshold. Step 5, Analysis: The analysis unit, based on the control commands issued by the cloud platform, uses a PID algorithm to regulate the analysis temperature and airflow speed to regenerate the adsorbed saturated composite material. Step 6, Recovery and Discharge: The hexane vapor generated by the desorption process is recovered by a condensation recovery device, and the tail gas is discharged after meeting the standards. The entire adsorption-desorption cycle takes 25-35 minutes.

[0022] Example 1: This embodiment focuses on the problems of short communication distance, high power consumption, and low accuracy of mixed gas identification in traditional gas monitoring systems. It combines the OneNet cloud platform, LoRa wireless communication technology, and SSA-SVM algorithm to build a low-power, long-distance, and high-precision toxic and harmful gas monitoring and early warning system.

[0023] I. Hardware Selection and Setup Toxic and hazardous gas detector hardware design: The FM33LG048 is selected as the main control chip. This chip is a 32-bit ARM Cortex-M0+ core with ultra-low power consumption. The sensor module uses the TGS2602 gas sensor, which detects the target gas by detecting the change in current caused by the electron transfer resulting from the reaction between the gas and the electrode surface. R2 and the amplifier are combined to form a voltage follower to obtain the voltage divider signal of the internal resistor of the sensor. Zener diode D1 prevents the signal voltage from being too high and damaging the microcontroller. Capacitor C2 is used for filtering, and pull-up resistor R6 is used to prevent the signal from being masked by the high gas concentration. The audible and visual alarm circuit consists of a 555 timer, a speaker, and a light-emitting diode. The 555 timer is connected to PB9 of the FM33LG048 and attaches the pulse signal to the speaker through the spacer capacitor C2.

[0024] Hardware design of toxic and harmful gas controller: STM32F103ZET6 is used as the main control chip, based on ARM Cortex-M3 core, with low power consumption, high performance and rich interface peripherals; The power module adopts a dual power supply method: 220V AC power is converted to 24V DC power by a switching power supply, and then converted to the required voltage by the MP1584 module; battery power is converted from 24V to 3.3V by the DC-DC power chip LV2842. The voltage sampling circuit samples the main power voltage, charging voltage, and backup power voltage. When there is a 24V short circuit, the fuse blows and the MOSFET turns off. The display module is connected to the STM32F103ZET6 via RS-232 communication, and the T1IN and ROUT1 pins of the SP3232EAA-L / TR module are connected to the TX and RX pins of the USART1. The wireless module uses Quectel BC26. The hardware circuit includes a power supply circuit, a serial communication level conversion circuit, and a SIM card module circuit. It is powered by 3.3V, and parallel electrolytic capacitors and filter capacitors ensure stable power supply.

[0025] LoRa communication module design: The SX1278 chip is selected as the core. The three power supply pins VBAT_ANA, VBAT_DIG and VBAT_RF provide a stable 3.3V power supply. Filter capacitors are added to each pin to reduce ripple noise. In the radio frequency circuit, the RF switch PE4259 controls the switching between transmit and receive signals. The transmitting end uses an inductor-capacitor matching circuit and an elliptic filter to ensure signal matching and filtering. The receiving end uses an inductor-capacitor matching circuit and adds a DC blocking capacitor. The external interface brings out all IO pins through 1.27mm pitch half-holes, facilitating secondary development; LoRa networks use a star topology. Nodes broadcast data, which is then received and forwarded by the gateway. The data frame format includes a 1-byte node number, a 1-byte function code, a 2-byte number of alarms, a 6-byte alarm address, a 1-byte alarm status, 4 bytes of data, and a 1-byte checksum. The checksum is the sum of all data from the first frame to the last byte of the data field. If the sum exceeds 256, it is modulo 256.

[0026] II. Software Design and Development Gas detector software design: After initializing the microprocessor and each module, the LoRa module is configured and scans the channel, connecting to the gateway to establish a communication network; after the network connection is successful, the sensor continuously collects AD values ​​20 times and takes the average value to convert it into gas concentration; if the connection fails, the channel is rescanned; if the collected data is abnormal, it is immediately uploaded to the controller according to the specified protocol; if the data is normal, it waits for downlink instructions.

[0027] Gas controller software design: Before the program loop, the microprocessor and various chips are initialized, and the boot screen is displayed for 3 minutes, during which power detection, button detection, and screen detection are continuously performed; after successful boot, the LoRa gateway is powered on, scans for LoRa nodes to join, and after the node joins, the controller analyzes the received data, identifies three states: normal, fault, and alarm, and extracts the gas concentration and status to be stored in Flash; under normal conditions, data is uploaded to the cloud platform after 1 hour, and under fault or alarm conditions, data is uploaded immediately. If the upload time has not been reached, the LoRa enters a sleep state to reduce power consumption.

[0028] Complete SSA-SVM Algorithm Process: 1. Dataset Construction: Collect a public dataset of gas sensor arrays under dynamic gas mixtures, including 16 metal sensors. The gas samples are mainly methane and ethylene, with ethylene concentration of 0-20 ppm and methane concentration of 0-300 ppm. 2. Data Preprocessing: The input data undergoes PCA preprocessing, first standardizing to eliminate the influence of dimensions. The standardization formula is: ; Calculate the covariance matrix and obtain the eigenvalues ​​and eigenvectors. Select the four principal components with a cumulative contribution rate of over 99% for dimensionality reduction. Divide the dataset into training and test sets in an 8:2 ratio. 3. SSA parameter settings: The number of optimization parameters d=2, corresponding to the SVM penalty parameter c and kernel function parameter g, and the lower bound of the parameters. upper limit Population size p=50, maximum number of iterations MaxT=100, safety threshold ST=0.8, discoverer ratio 30%, joiner ratio 70%; 4. SSA optimizes SVM hyperparameters: Individual positions are updated by simulating sparrow search behavior using SSA, and the fitness function is the classification accuracy of SVM on the training set; 5. Model training: Substitute the optimal parameters c=53.248 and g=5.979 obtained from SSA optimization into SVM to construct an SSA-SVM classification model. Use the RBF kernel function and train the model using the training set. 6. Model Deployment: Compile the trained model into executable code for the controller, integrate it into the software system, receive sensor data in real time, and input the pre-processed data into the model to output the mixed gas type identification result; 7. SSA-SVR Model Construction and Deployment: Similarly, an SSA-SVR model is constructed for quantitative prediction of gas concentration. The loss function value is set to 0.01, and after optimizing the hyperparameters, it is deployed to the controller.

[0029] III. OneNet Cloud Platform Access and Configuration Product creation: Register a OneNet account, select the NB-IoT IoT kit to add products. Each product represents a gas controller. Record the product ID, user ID, and Master-API key. Add device: Add a formal device to the product. Name the device according to the installation location + number. Obtain the IMEI and IMSI through the AT commands AT+CGSN=1 and AT+CIMI of the BC26 module and fill them in. NB-IoT module network access design: Initialize the BC26 module, including port initialization, power-on operation and USART3 and DMA configuration. Send AT commands through the serial port USART3 to connect with the cloud platform. Upon successful registration, receive +MIPLEVENT:0,6. Communication protocol design: The LWM2M protocol is adopted. The ToxicGasObject object model ID is designed as 33001, which includes resource 1 (5700, toxic and harmful gas concentration value, float type, unit LEL) and resource 2 (5536, update timestamp, time type, UTC time), and the CoAP protocol is used for transmission. Data exceedance alarm design: Utilize the OneNet cloud platform trigger module to set the trigger data stream and conditions (gas concentration ≥20%LEL), select email as the information receiving method, and send alarm information via email when data exceeds the limit.

[0030] IV. Core Technology Features Detailed explanation of technical features 1. LoRa wireless communication technology: The core is linear frequency modulation spread spectrum modulation. The carrier signal is composed of linear frequency modulation, and the frequency increases or decreases with time. It operates in an unlicensed frequency band below 1GHz and is specifically designed for low-power, long-distance communication links. The communication distance can reach 2-20km, and the power consumption is only 10uA-15mA, solving the problems of short communication distance and high power consumption in traditional communication.

[0031] 2. SSA-SVM Mixed Gas Detection Algorithm: SVM achieves classification by mapping low-dimensional data to a high-dimensional space to find the optimal hyperplane, and the RBF kernel function handles nonlinearly separable data; SSA simulates the food-searching behavior of sparrows, optimizes the hyperparameters c and g of SVM, solves the problem of difficult determination of hyperparameters, and improves the mixed gas recognition rate and concentration prediction accuracy; PCA technology reduces the dimensionality of input data and eliminates redundant information.

[0032] 3. OneNet Cloud Platform Access Technology: Supports multiple access protocols and methods, providing functions such as device access management, data management and storage, application management and security management; enables hardware to communicate with the cloud platform through the NB-IoT module, completing functions such as data upload, remote real-time monitoring, and data visualization.

[0033] 4. Modular Hardware Design: The detector, controller, and LoRa communication module adopt a modular design. The detector is responsible for data acquisition, the controller is responsible for data processing and uploading, and the LoRa module is responsible for data transmission. The functions of each module are clearly defined, which facilitates debugging, maintenance, and expansion.

[0034] This embodiment utilizes the linear frequency modulation spread spectrum characteristics of LoRa technology to achieve low-power long-distance data transmission; SSA optimizes the hyperparameters of SVM, improving the accuracy of mixed gas identification and concentration prediction, and avoiding the defect of traditional algorithms being prone to getting trapped in local optima; the access to the OneNet cloud platform enables intelligent remote monitoring and early warning, allowing staff to keep abreast of gas status in real time, respond promptly to abnormal situations, and enhance the system's informatization and intelligence level.

[0035] Example 2: This embodiment addresses the problems of low efficiency and easy secondary pollution associated with single purification technologies by employing Fe³⁺. + By doping and modifying TiO2 and loading it onto ACF to form a composite material, the combined effects of adsorption and photocatalysis can achieve efficient purification of harmful gases and improve the adaptability and stability of the purification system.

[0036] I. Preparation of Composite Materials Preparation of TiO2: 50 ml of anhydrous ethanol, 13.2 ml of acetylacetone and 21.78 ml of tetrabutyl titanate were added sequentially to a 200 ml beaker. The mixture was magnetically stirred at room temperature for 2 h to obtain TiO2 colloid. The colloid was dried at 120 °C for 6 h and calcined in a muffle furnace at 500 °C for 3 h to obtain TiO2 powder.

[0037] Fe³ +Preparation of TiO2: Anhydrous ethanol, dilute nitric acid, and tetrabutyl titanate were sequentially added to a 200ml beaker and mixed. The pH was controlled at approximately 4.0. After stirring at room temperature for 30 minutes, the mixture was cooled to 2°C in an ice-water bath. Ferric nitrate was dissolved in deionized water to obtain solution B. Solution A was rapidly added under magnetic stirring. Stirring was stopped after the sol-gel was formed. The wet gel was aged at room temperature for 5 hours and dried at 120°C for 12 hours to obtain a dry gel. The dry gel was calcined in a muffle furnace at 500°C to obtain different Fe³⁺ precipitates. + :Ti 4+ Fe³⁺ in molar ratios of 1:100, 1:200, and 1:500 + -TiO2 powder.

[0038] Fe³ + Preparation of TiO2 / ACF: Cut ACF SY-1000 type was soaked in NaOH solution for 15 min, repeatedly washed with deionized water and dried, then soaked in HNO3 solution for 15 min, washed until neutral, and dried at 100℃ for 1 h; 5 g, 10 g, and 15 g of photocatalyst Fe³⁺ were taken respectively. + - TiO2 and sodium polyacrylate were used to prepare dispersions with mass fractions of 1%, 2%, and 3% in 500g of deionized water. The treated ACF was immersed in the dispersion for 2-3 minutes and then pulled out. It was dried at 120℃ for 1 hour. The operation was repeated 3 times to obtain the adsorption-photocatalytic composite material.

[0039] II. Core Technology Features 1.Fe³ + Doping and modification technology of TiO2: Fe³ + Ionic radius 0.064 nm and Ti 4+ Ions with radii close to 0.068 nm can easily enter the TiO2 crystal structure without changing the phase composition of TiO2, but can reduce the electron-hole recombination rate and improve photocatalytic activity; when Fe³⁺ + :Ti 4+ The highest photocatalytic activity was observed at a molar ratio of 1:200, with an average TiO2 grain size of 19.7 nm.

[0040] 2. ACF loading technology: The surface of ACF contains a large number of mesopores, with a compact structure and various voids, which is conducive to the loading of Fe³⁺. + -TiO2 has a good loading effect, enabling Fe³ + -TiO2 is uniformly distributed on its surface; the high specific surface area of ​​ACF can adsorb gas molecules, allowing the gas to pass through Fe³⁺. + - A local high-concentration region is formed on the TiO2 surface, which promotes the photocatalytic reaction. At the same time, the degradation gas of the photocatalytic reaction can clear the active sites of ACF, realizing the in-situ regeneration of ACF.

[0041] 3. Adsorption-Photocatalysis Synergistic Technology: During adsorption, ACF physically adsorbs gas molecules onto the surface, reducing the gas diffusion distance; during photocatalysis, Fe³⁺… + TiO2 generates electron-hole pairs under light irradiation, which react with H2O and O2 in the air to produce hydroxyl radicals (·OH) and superoxide radicals (O2). - It oxidizes and decomposes harmful gases into harmless substances such as CO2 and H2O, and the two work together to improve the gas degradation rate.

[0042] 4. Multi-factor optimization technology: By controlling key factors such as initial concentration, oncoming wind speed, light intensity, and photocatalyst loading, the optimal reaction conditions for each gas are determined to achieve the best degradation effect.

[0043] This embodiment uses Fe³ + Doping modification enhances the photocatalytic activity of TiO2 and reduces the electron-hole recombination rate; the high specific surface area of ​​ACF enhances the gas adsorption capacity, and the synergistic effect of the two solves the problem of low treatment efficiency of single adsorption or photocatalysis technologies; multi-factor optimization enables the system to adapt to different environmental conditions, while the good thermal stability of the composite material ensures stable operation in most indoor temperature environments, avoiding secondary pollution.

[0044] Example 3: This embodiment addresses the problems of severe cross-response, low monitoring accuracy, and inability to conduct remote real-time monitoring with a single sensor. It constructs an array composed of electrochemical and MOS sensors, combines a random forest algorithm for data fusion, and connects to a cloud platform to achieve accurate identification, remote monitoring, and automatic early warning of harmful mixed gases.

[0045] I. Sensor Array Setup Electrochemical sensor selection: ZE-H2, ZE-CO, and ME-H2S electrochemical sensors were selected to measure H2, CO, and H2S, respectively. ZE-H2 has a detection range of 0-2000ppm, a response time of <30s, and a sensitivity of 2ppm; ZE-CO has a detection range of 0-500ppm, a response time of <30s, and a sensitivity of 0.1ppm; ME-H2S has a detection range of 0-100ppm, a response time of <30s, and a sensitivity of 0.1ppm. The electrochemical sensor driving circuit includes a constant potential circuit, an I / V conversion circuit, an amplification circuit, and a filtering circuit. The constant potential circuit fixes the working electrode voltage, the I / V conversion circuit converts the weak current signal into a V-level voltage signal, the amplification circuit adjusts the signal to the range that the microcontroller can acquire, and the filtering circuit smooths the signal.

[0046] MOS sensor fabrication: Four oxides—WO3, ZnO, In2O3, and Fe2O3—were selected and doped with noble metals as raw materials. Fe2O3 was prepared via a hydrothermal method, ZnO via a simple co-precipitation method, In2O3 via a hydrothermal method to synthesize the In(OH)3 precursor followed by high-temperature calcination, and WO3 via a low-temperature acid-assisted hydrothermal method and annealing treatment. MOS sensor fabrication steps: The ceramic sheet was ultrasonically cleaned for 20 minutes, rinsed 5 times with anhydrous ethanol, and then dried. The ceramic sheet was then welded to the base, and no incomplete or missing welds were found. Add the prepared material to an agate mortar, add deionized water and grind into a uniform slurry. Apply the slurry evenly to the surface of the ceramic sheet with a brush, repeat 5-6 times, add a protective shell and calcine at 150℃ for 5 minutes. The MOS sensor driving circuit includes a filter circuit, a voltage divider circuit, and an amplifier circuit. After the heating resistor on the back of the ceramic plate is preheated, the sensor adsorbs the target gas, changing its resistance. The voltage divider circuit detects the voltage across the load resistor, and the sensor resistance is calculated. The resistance calculation formula is as follows: .

[0047] The sensor array consists of three electrochemical sensors (ZE-H2, ZE-CO, ME-H2S) and four MOS sensors (WO3, ZnO, In2O3, Fe2O3) to collect multi-dimensional information about the mixed gas.

[0048] II. Embedded Minimum System Design Microcontroller selection: The STM32F103VET6 chip is selected, with a main frequency of up to 72MHz and 64 pins, including peripherals such as serial communication, general-purpose timer, DMA controller, SPI interface, RTC, and ADC. It has a built-in 12-bit ADC with a conversion accuracy of 0.8mV, which meets the system's data acquisition requirements.

[0049] Peripheral circuit design: The program download circuit uses an ST-LINK interface and is connected to the microcontroller's 3.3V, GND, SWDIO, and SWCLK pins; the 3.3V voltage regulator circuit uses an LM1117-3.3 chip with an output voltage error of <1%; the clock circuit consists of an 8MHz crystal oscillator, a feedback resistor, and a start-up capacitor, and is multiplied to 72MHz by a PLL phase-locked loop; the hardware reset circuit consists of a 10KΩ resistor, a 0.1μF capacitor, and a button, and the reset signal is held for 1.1ms; the boot mode selection circuit grounds two pins and selects the main flash memory for booting; other pins are brought out for easy expansion of peripherals.

[0050] Power management module design: The externally input 12V DC power is converted to 5V through the LM2596 chip, and a π-type filter circuit is added to ensure the stability of the DC power source. The Zener diode is SS34.

[0051] RTC Real-Time Clock Circuit Design: A 32.768kHz crystal oscillator is selected. After 15 frequency divisions, the counting period of 32768Hz is 1S, ensuring accurate clocking. The RTC power supply circuit adopts dual power supply of 3.3V external power supply and button battery. When the external power supply is interrupted, the button battery provides power to ensure the normal operation of the chip. The RTC module records the sensor running time and sends a replacement reminder after the set working time is reached.

[0052] III. Communication Module Design WiFi communication module: The ESP8266 module is selected, powered by 3.3V, and connected to the USART2_TX and USART2_RX pins of the microcontroller. The baud rate is set to 115200. It establishes a connection with the host computer hotspot through AT commands to build a local area network for data transmission and reception.

[0053] DTU Communication Module: The selected DTU device is based on the EC200S module and supports 4G IoT cards. It sends data to the cloud platform via TCP / MQTT protocol. The DTU communicates with the microcontroller via RS232 level. The microcontroller's TTL level needs to be converted to RS232 level, and the baud rate needs to be set to 115200 to ensure normal communication.

[0054] IV. Software Design and Development Data acquisition module: The microcontroller's built-in ADC uses an independent multi-channel scanning mode to acquire sensor array signals, with a sampling time of 71.5 clock cycles and a single conversion time of approximately 7µs. A general-purpose timer is used to control the acquisition cycle, with arr=9999, psc=7199, a counting cycle of 1 second, and the timer starts ADC acquisition after 10 seconds. A DMA dual-buffer peripheral is used, with the two buffers working alternately to avoid data acquisition interruption. After acquisition, the data is preprocessed using the bubble sort method, and the mean is calculated after removing extreme values ​​to improve data accuracy.

[0055] Complete process of random forest algorithm: 1. Dataset Construction: A dataset of 216 mixed gas samples was constructed. CO, H2, and H2S were mixed at six concentration gradients of 0, 20, 40, 60, 80, and 100 ppm. The sensor array collected the multi-dimensional response signal of each sample as feature data, and the gas concentration was used as the label. 2. Data Preprocessing: The host computer uses the MinMaxScalar() function from the sklearn library to normalize the data, and the transformation equation is as follows: After normalization, the data range is between [0,1]. 3. Random Forest Model Construction: The number of decision trees is set to 100, the maximum depth of each tree is None, and the number of features randomly selected when splitting a node is the square root of the total number of features. The training dataset for each tree is constructed using random sampling with replacement. Each decision tree is constructed using the CART algorithm, and the optimal splitting feature is selected by calculating the Gini coefficient. 4. Model Training: The preprocessed dataset is divided into training and test sets in an 8:2 ratio. The random forest model is trained using the training set. Each decision tree is trained independently, and no pruning is performed during training. The classification results of all decision trees are aggregated through a voting mechanism as the final output. 5. Model Deployment: The trained random forest model is deployed on the host computer. It receives real-time data collected by the sensor array via WiFi module, and after preprocessing, it is input into the model to output the concentration identification results of each component of the mixed gas.

[0056] Data Cloud Upload Module: OneNet cloud platform is selected. After registering an account, a new project is created, project information is filled in, the data transmission protocol MQTT is selected, devices are added, and the product ID and Access_key are recorded. The DTU device establishes a connection with the cloud platform via the MQTT protocol, writes the product ID and Access_key into the message, sends a login request to bind to the virtual device on the cloud platform, and after successful binding, uploads the analyzed gas concentration data to the cloud platform.

[0057] Early warning function module: Add triggers to the OneNet cloud platform, set the data streams H2_ppm, CO_ppm, H2S_ppm and the trigger threshold of 40ppm, select email as the information receiving method, and the platform will automatically send an early warning message when the gas concentration is ≥40ppm; add a working time trigger, set the threshold to 365 days, and send a sensor replacement reminder when the maximum working time is reached.

[0058] V. Core Technology Features 1. Sensor array technology: Combining the advantages of high sensitivity and good anti-interference of electrochemical sensors with the fast response speed and easy integration of MOS sensors, a multi-sensor array is constructed to collect multi-dimensional response signals of mixed gases, make up for the defects of cross-response of single sensors, increase the amount of effective information collected, and provide a data foundation for qualitative and quantitative analysis of mixed gases.

[0059] 2. Random Forest Data Fusion Technology: Random forest is an ensemble learning algorithm that uses random sampling with replacement to form multiple datasets. Each dataset is used to train a decision tree. The decision tree calculates the Gini coefficient from randomly selected features, selects the smallest Gini coefficient for node splitting, and finally aggregates the classification results of all decision trees through a voting mechanism. This algorithm can eliminate data redundancy and contradictory information, extract effective features, avoid overfitting, and improve the accuracy of mixed gas recognition.

[0060] 3. Cloud Platform Remote Monitoring Technology: The sensor array communicates with the OneNet cloud platform through DTU devices and the MQTT protocol. The DTU devices support 4G IoT cards, enabling data transmission in areas with mobile network coverage, without being limited by WiFi coverage. The cloud platform provides functions such as data storage, data visualization, historical data query, and automatic early warning, enabling remote real-time monitoring of mixed gases.

[0061] 4. RTC timing reminder technology: The STM32's built-in RTC module is independently powered and records the sensor's running time. When the set maximum working time is reached, an interrupt is triggered, and a replacement reminder is sent to ensure that the sensor operates within the normal working range and avoids a decrease in monitoring accuracy due to sensor aging.

[0062] This embodiment collects multi-dimensional information about the mixed gas through a sensor array, providing data support for accurate identification; the ensemble learning characteristics of the random forest algorithm effectively handle the sensor cross-response problem and improve the accuracy of mixed gas identification; the combination of DTU equipment and cloud platform enables remote real-time monitoring without geographical restrictions, reducing labor costs; the RTC timing reminder function ensures stable sensor operation, extends system lifespan, and enhances overall reliability.

[0063] Example 4: This embodiment addresses the issues of single monitoring or purification systems having limited functionality and poor coordination. It integrates the core technologies of the previous three embodiments to construct an integrated monitoring-purification system, thereby improving monitoring accuracy, enabling timely purification response, and providing integrated management to meet the gas treatment needs in complex scenarios.

[0064] I. System Implementation Steps 1. System Integration Design: Integrating the monitoring and early warning system based on OneNet+LoRa+SSA-SVM from Example 1 with the Fe³ system from Example 2. + The TiO2 / ACF adsorption-photocatalytic purification system and the monitoring and analysis system based on sensor array + random forest + cloud platform in Example 3 are integrated; the system in Example 1 is used as the core monitoring unit, and the sensor array in Example 3 is used as the auxiliary monitoring unit to improve the accuracy of monitoring data; the purification system in Example 2 is used as the execution unit to receive instructions from the monitoring system to start or stop the purification operation; the OneNet cloud platform is used as the central control unit to integrate monitoring data and purification system status data to achieve unified management and control.

[0065] 2. Linkage Control Logic Design: Linkage rules are set on the OneNet cloud platform. When the monitoring systems of Example 1 and Example 3 simultaneously detect that the gas concentration exceeds the set threshold (e.g., H2S concentration ≥ 40ppm) and the duration exceeds 30s, the cloud platform sends a start command to the purification system of Example 2. The purification system automatically adjusts the oncoming wind speed (set according to the optimal wind speed for each gas), light intensity (54W UV lamp), and photocatalyst load (15g) according to the detected gas type and concentration, and starts the adsorption-photocatalytic purification process. When the monitoring system detects that the gas concentration is lower than the safety threshold and the duration exceeds 10min, the cloud platform sends a stop command, and the purification system stops working.

[0066] 3. Data Interaction Design: The monitoring system uploads gas concentration data to the OneNet cloud platform via LoRa and DTU communication. The purification system connects to the cloud platform via an NB-IoT module and uploads real-time operating status data (operating mode, current wind speed, and light intensity). The cloud platform integrates and analyzes the data to generate an integrated monitoring-purification report, which includes information such as gas concentration change curves, purification efficiency, and equipment operating status.

[0067] 4. System Debugging and Operation: Simulate a gas leak environment by introducing mixed gases of different concentrations, such as CO, H2, H2S, NH3, and formaldehyde, to test the system's linkage response speed, purification efficiency, and data interaction stability; debug the linkage control logic to ensure that the purification system can start and stop in a timely manner based on monitoring data; test the data interaction function to ensure that monitoring data and purification status data are accurately uploaded to the cloud platform; test the purification efficiency to verify the purification effect of the fusion system on gases of different concentrations.

[0068] 5. Algorithm Co-operation Design: Data fusion: The SSA-SVM model and the random forest model process the collected gas data separately, outputting identification results and concentration prediction values. The cloud platform uses a weighted average method to fuse the two sets of results. The weights are determined based on the historical accuracy of the two models (e.g., SSA-SVM with 96% accuracy has a weight of 0.45, and random forest with 98% accuracy has a weight of 0.55). The fused concentration value is... ; Adaptive adjustment of purification parameters: Based on the fused gas type and concentration data, the multi-factor optimization results of Example 2 are called to automatically match the best purification parameters. For example, when formaldehyde is detected, the oncoming wind speed is adjusted to 0.8m / s, the light intensity is 54W, and the load is 15g.

[0069] II. Core Technology Features 1. Integration of multiple monitoring technologies: The monitoring technology of LoRa communication + SSA-SVM algorithm is integrated with the monitoring technology of sensor array + random forest algorithm to form a dual monitoring mechanism. The two monitoring technologies verify each other, improve the accuracy of gas concentration detection and type identification, and avoid misjudgment or omission by a single monitoring technology.

[0070] 2. Monitoring-Purification Linkage Technology: Through the linkage rules of the OneNet cloud platform, a communication link is established between the monitoring system and the purification system. The monitoring system acts as the trigger condition, and the purification system acts as the actuator, realizing closed-loop control of concentration exceeding the limit - automatic purification - concentration reaching the standard - automatic shutdown, without manual intervention, improving the timeliness and efficiency of gas treatment.

[0071] 3. Centralized cloud platform management technology: The OneNet cloud platform integrates data from multiple systems to achieve unified storage, visualization, and analysis of monitoring data and purification status data. Staff can grasp the gas environment and equipment operating status through a single platform, simplifying management processes and improving management efficiency.

[0072] In summary, this soybean oil extraction tail gas absorption and desorption system uses Fe³⁺ as its adsorption unit. + -TiO2 / ACF composite adsorption-photocatalytic material, utilizing the high specific surface area of ​​ACF to adsorb mixed components in exhaust gas, while Fe³ + -TiO2 plays a photocatalytic degradation role, and the two work together to improve adsorption selectivity and efficiency; the desorption unit receives real-time data from the monitoring unit and dynamically adjusts parameters such as desorption temperature and airflow speed through a PID closed-loop control algorithm to accurately regenerate the adsorption-saturated composite material, thus completely solving the problems of low adsorption efficiency, poor selectivity, difficult regeneration, and secondary pollution of single activated carbon.

[0073] Furthermore, the soybean oil extraction tail gas absorption and desorption system's monitoring unit's sensor array consists of a specifically arranged array of electrochemical and MOS sensors. These sensors work together to collect information on components such as hexane, fatty acids, and water vapor in the tail gas. The data is preprocessed, hyperparameters optimized, and modeled using the SSA-SVM optimization algorithm, outputting accurate concentration data. The desorption unit dynamically adjusts the desorption parameters based on this data, replacing the fixed parameter mode. This solves the problem of incomplete desorption or excessive energy consumption caused by a lack of real-time accurate monitoring, achieving a balance between treatment effectiveness and energy consumption.

[0074] Furthermore, in this soybean oil leaching tail gas absorption and analysis system, each unit uses a dual-mode communication architecture of LoRa and NB-IoT in the communication unit to dynamically switch transmission modes based on communication quality, ensuring stable and real-time data transmission to the cloud platform unit. The cloud platform unit enables centralized data storage and visualization, and sends early warning information in a timely manner through a three-level early warning mechanism, constructing an integrated intelligent management system. This solves the problems of data dispersion, inability to remotely monitor, and lack of intelligent early warning, significantly improving the convenience and efficiency of operation and maintenance. It also addresses the issues of low adsorption efficiency, insufficient monitoring accuracy, and low level of intelligent management found in existing soybean oil leaching tail gas treatment technologies.

[0075] The relevant modules involved in this system are all hardware system modules or functional modules that combine computer software programs or protocols with hardware in the prior art. The computer software programs or protocols involved in these functional modules are technologies known to those skilled in the art and are not improvements to this system. The improvement of this system lies in the interaction or connection between the modules, that is, in improving the overall structure of the system to solve the corresponding technical problems that this system aims to address.

[0076] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A soybean oil extraction tail gas absorption and desorption system, characterized in that, It includes an adsorption unit, a desorption unit, a monitoring unit, a communication unit, and a cloud platform unit. Each unit is connected in series via a data bus to form a closed-loop control. The adsorption unit uses Fe³ + -TiO2 / ACF composite adsorption-photocatalytic material to achieve synergistic effect of exhaust gas adsorption and photocatalytic degradation; The monitoring unit includes a sensor array and an SSA-SVM optimization algorithm, which is used to collect and process exhaust gas component concentration data in real time. The communication unit adopts a dual-mode communication architecture of LoRa and NB-IoT to ensure stable data transmission; The cloud platform unit is custom-developed based on the OneNet IoT platform and has data storage, visualization, and intelligent early warning functions. The analysis unit dynamically adjusts the analysis parameters based on the real-time concentration data output by the monitoring unit using a PID closed-loop control algorithm. Among them, the Fe³ of the adsorption unit + The preparation method of TiO2 / ACF composite adsorption-photocatalytic material includes the following steps: Step 1: Take tetrabutyl titanate and anhydrous ethanol at a volume ratio of 1:2.0-2.5, add acetylacetone as a chelating agent, and the volume ratio of tetrabutyl titanate to acetylacetone is 1.5-1.8:

1. Stir magnetically at room temperature for 1.5-2.5 hours to form TiO2 sol. Step 2: Prepare a 0.04-0.06 mol / L Fe(NO3)3·9H2O aqueous solution, according to Fe³ + With Ti 4+ Add TiO2 sol rapidly at a molar ratio of 1:180-220, cool to 0-4℃ in an ice-water bath and stir continuously for 25-35 minutes until the system gels; Step 3: Ag the wet gel at room temperature for 4-6 hours, dry it at 110-130℃ for 10-14 hours, calcine it at 480-520℃ for 2.5-3.5 hours, and cool it to obtain Fe³⁺. + -TiO2 powder; Step 4: Soak ACF in NaOH solution and dilute nitric acid solution for 12-18 min each, wash with deionized water until neutral, and dry at 90-110℃ for 0.8-1.2 h. Step 5: Take 12-18g of Fe³ + - Mix TiO2 powder with 450-550g of deionized water, add 0.8%-1.2% sodium polyacrylate as a binder, and stir to form a dispersion; After immersing the treated ACF in the dispersion for 2-4 minutes, remove it and dry it at 110-130℃ for 0.8-1.2 hours. Repeat this operation 2-4 times to obtain the finished product. The sensor array of the monitoring unit is arranged as follows: The three electrochemical sensors are hexane electrochemical sensor, fatty acid electrochemical sensor, and water vapor electrochemical sensor, which are evenly arranged in an equilateral triangle with a spacing of 4-6 cm between adjacent sensors; the four MOS sensors are ZnO sensor, WO3 sensor, Fe2O3 sensor, and In2O3 sensor, which are evenly distributed around the circumcircle of the equilateral triangle with a minimum distance of 4-6 cm from the electrochemical sensors.

2. The soybean oil extraction tail gas absorption and desorption system according to claim 1, characterized in that, The sensor array of the monitoring unit also includes: The sensor array has a sampling frequency of 8-12Hz, the ADC has a sampling accuracy of 12 bits, and the signal conditioning circuit includes an operational amplifier with an amplification factor of 90-110 times and a low-pass filter with a cutoff frequency of 0.8-1.2kHz. The fabrication and calibration process of the MOS sensors is as follows: ZnO sensors are prepared by co-precipitation method, WO3 sensors are prepared by low-temperature acid-assisted hydrothermal method, Fe2O3 sensors are prepared by hydrothermal method, and In2O3 sensors are prepared by hydrothermal synthesis of In(OH)3 precursor and calcination at 430-470℃ for 1.5-2.5h; the ceramic substrates of all MOS sensors are ultrasonically cleaned for 15-25min, rinsed with anhydrous ethanol 4-6 times, dried at 140-160℃ for 4-6min, and coated with gas-sensitive slurry with a thickness of 18-22μm; Sensor calibration uses a standard gas concentration gradient of 0-100ppm, with each gradient stabilized for 25-35 minutes. Response values ​​are recorded and calibration curves are established. Calibration error is ≤±2%.

3. The soybean oil extraction tail gas absorption and desorption system according to claim 1, characterized in that, The complete process of the SSA-SVM optimization algorithm for the monitoring unit is as follows: Step 1: Data Acquisition: Construct a dataset containing sensor response data of hexane, fatty acids, and water vapor components in soybean oil leaching tail gas under different concentrations and temperature conditions. The dataset size should be no less than 4,000-6,000 sets. Step 2: Data Preprocessing: The following standardized formula is used for data processing: ； in For standardized data, This is the original data. The mean of the original data. The number of data points; PCA dimensionality reduction reduced the original 16-dimensional data to 3-5 dimensions, with a cumulative contribution rate of ≥99.4%. Step 3: Algorithm Construction: Set the sparrow population size to 40-60, the maximum number of iterations to 80-120, the safety threshold ST = 0.7-0.9, the ratio of discoverers to joiners to be 2:8-3:7, and optimize the lower limit of the parameters. upper limit ; Step 4: Algorithm training: Search for the optimal hyperparameters using SSA. After training, the support vector machine penalty parameter c = 45-60 and kernel function parameter g = 5-7. Step 5: Algorithm Deployment: The trained model is embedded into the microcontroller of the monitoring unit, with a single data processing time of ≤25ms.

4. The soybean oil extraction tail gas absorption and desorption system according to claim 1, characterized in that, The complete switching process for the dual-mode communication of the communication unit includes: Step 1, Communication Quality Monitoring: The communication unit collects the packet loss rate and communication distance of the LoRa module in real time, with a sampling interval of ≤100ms; Step 2, Switching Condition Judgment: When the packet loss rate of 3 consecutive samples is >1.2% or the communication distance is >1.8km, the switch to NB-IoT mode is triggered; When the LoRa module packet loss rate is ≤0.7% and the communication distance is ≤1.7km, switch back to LoRa mode; Step 3, Switching Execution: The dual-mode switching delay time is ≤50ms, and a data caching mechanism is used during the switching process to avoid data loss; Step 4: The communication protocol adopts LWM2M. The data frame format includes 1 byte node number, 1 byte function code, 2 bytes number of sensors, 6 bytes sensor address, 1 byte status bit, 4 bytes data and 1 byte checksum.

5. The soybean oil extraction tail gas absorption and desorption system according to claim 1, characterized in that, The complete PID closed-loop control process of the analysis unit is as follows: Step 1, setting control targets: Hexane residual concentration ≤10ppm, proportionality coefficient Integral coefficient Differential coefficients ; Step 2, Parameter adjustment range: Desorption temperature 80-120℃, adjustment accuracy ±1℃; airflow velocity 0.7-1.1m / s, adjustment accuracy ±0.01m / s; Step 3, Stage Division: The analysis process is divided into a preheating stage, a constant temperature analysis stage, and a cooling stage. The heating rate in the preheating stage is 4-6℃ / min. The duration of the constant temperature analysis stage is dynamically adjusted according to the concentration data of the monitoring unit, with an adjustment step of 5-10min. The cooling stage is for natural cooling to room temperature. Step 4, Feedback Adjustment: Collect the analyzed concentration data every 200-300ms and substitute it into the following PID formula to calculate the adjustment amount: ； in For output adjustment amount, For concentration deviation, The integral time constant is... The differential time constant enables dynamic closed-loop regulation.

6. The soybean oil extraction tail gas absorption and desorption system according to claim 1, characterized in that, The complete data processing flow of the cloud platform unit is as follows: Step 1, Data Reception: Data packets are received via NB-IoT or LoRa module, and after verification and validation, gas concentration data and device status information are obtained through parsing. Step 2, Data Storage: Store the parsed data in a MySQL database. The data table contains the fields gas_id, device_id, gas_type, gas_value, gas_unit, and gas_create_time. The gas_value field has a precision of 10 integer digits + 2 decimal digits. Step 3, Data Visualization: Generate line charts and bar charts through the visualization module to display the real-time trends of hexane, fatty acid, and water vapor concentration changes, as well as the equipment operating status. The data update frequency is 1-3 seconds / time. Step 4, Intelligent Early Warning: Set three levels of early warning thresholds. Level 1 warning: if the hexane concentration is >42ppm, send an alarm message via email and SMS. Level 2 warning: if the hexane concentration is 20-40ppm, send an alarm message via email. Level 3 warning: if the hexane concentration is 10-20ppm, display an early warning prompt on the cloud platform interface. The warning information includes the trigger time, sensor number, concentration value, and handling suggestions.

7. The soybean oil extraction tail gas absorption and desorption system according to claim 1, characterized in that, The structural parameters of the adsorption unit are: The adsorption unit adopts a multi-layer honeycomb structure with 2-4 layers, each layer having a thickness of 4-6 cm and a honeycomb pore size of 0.8-1.2 cm. The honeycomb channels form an angle of 25-35° with the exhaust gas flow direction. The adsorption unit shell is made of 304 stainless steel and is equipped with a temperature sensor and a pressure sensor inside. The temperature monitoring range is -25-85℃, the pressure monitoring range is 0-0.12MPa, and the sensor data sampling frequency is 5-8Hz. The Fe³ + The BET specific surface area of ​​the TiO2 / ACF composite adsorption-photocatalytic material is 572-730 m² / g, and the total pore volume is 0.29-0.32 cm³ / g.

8. A method for absorbing and desorbing soybean oil extraction tail gas, characterized in that, The soybean oil extraction tail gas absorption and desorption system based on any one of claims 1-7 includes the following steps: Step 1: Pretreatment: After particulate matter is removed from the soybean oil leaching tail gas by a pretreatment device, it enters the adsorption unit at a flow rate of 0.7-0.9 m / s, passing through Fe³⁺. + -TiO2 / ACF composite material is used for adsorption and photocatalytic degradation; Step 2, Monitoring: The monitoring unit collects the concentration data of hexane, fatty acids and water vapor in the exhaust gas at a frequency of 8-12Hz, and obtains accurate concentration values ​​after processing by the SSA-SVM algorithm; Step 3, Communication: The communication unit transmits concentration data and equipment status information to the cloud platform unit in real time, with a transmission delay of ≤1s; Step 4: Cloud Platform Processing: The cloud platform unit stores, analyzes, and visualizes the data, triggering a corresponding warning when the concentration reaches the warning threshold. Step 5, Analysis: The analysis unit, based on the control commands issued by the cloud platform, uses a PID algorithm to regulate the analysis temperature and airflow speed to regenerate the adsorbed saturated composite material. Step 6, Recovery and Discharge: The hexane vapor generated by the desorption process is recovered by a condensation recovery device, and the tail gas is discharged after meeting the standards. The entire adsorption-desorption cycle takes 25-35 minutes.

Images