An intelligent integrated system for an aluminum electrolysis cell and a control method
By introducing a multi-source parameter sensing and transmission module, an intelligent algorithm unit, and a three-layer architecture integration unit into the aluminum electrolysis cell, real-time monitoring and closed-loop control under high temperature and strong magnetic environment are realized, solving the problems of measurement lag and low accuracy in the existing technology, and improving production efficiency and economic benefits.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- HENAN WANJI ALUMINUM
- Filing Date
- 2026-05-22
- Publication Date
- 2026-07-21
AI Technical Summary
Existing aluminum electrolysis cell control systems rely on manual adjustment of process parameters, which suffers from measurement lag, low accuracy, high safety risks, and a disconnect between monitoring data and the control system, making it impossible to achieve real-time and accurate control. This results in high energy consumption, weak fault early warning capabilities, easy equipment damage, and increased production costs.
Employing a multi-source parameter sensing and transmission module, an intelligent algorithm unit, and a three-layer architecture integrated unit, combined with an integrated anode current-tank temperature sensor, a dual-core heterogeneous processor, and intelligent algorithms, it achieves real-time monitoring and closed-loop control under high-temperature and strong magnetic environments. It uses LSTM neural networks and electrochemical reaction mechanism models for bias current prediction and supports data parsing, model calculation, and control command generation.
It achieves anode current measurement accuracy ≤ ±1.5%, cell temperature measurement accuracy ≤ ±2℃, closed-loop response cycle ≤ 30s, bias current warning response time ≤ 10s, warning accuracy ≥ 95%, reduces electricity consumption per ton of aluminum by 3% to 5%, and reduces anode carbon consumption by 2kg/t-Al, thereby improving production efficiency and economic benefits.
Smart Images
Figure CN122428343A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of aluminum electrolysis production technology, specifically to an intelligent integrated system and control method for aluminum electrolysis cells. Background Technology
[0002] In the production of electrolytic aluminum, 330kA and 400kA series electrolytic cells are the core production equipment, and their operational stability and control precision directly affect the quality of aluminum products, production efficiency and energy consumption.
[0003] In the existing technology, the control mode of electrolytic cell control system generally adopts "host computer + Ethernet communication + terminal", which relies on the management experience of operators to manually adjust process parameters. There are significant technical defects: (1) The detection of cell temperature and the monitoring of anode current distribution rely on manual hand-held ammeter to measure point by point. Not only is the measurement data obviously lagging, the detection speed is slow, and the measurement accuracy is greatly affected by human operation. It cannot reflect the operating status of the electrolytic cell in real time. At the same time, manual operation requires close proximity to the electrolytic cell in a high temperature and strong magnetic environment, which poses a high risk of safety. (2) The current mainstream cell control system in China only integrates basic electrical parameter monitoring functions. The coverage of automatic temperature monitoring and online anode current distribution monitoring system is less than 15%. The monitoring data is separated from the cell control system and cannot form a closed loop control, resulting in low parameter control accuracy and difficulty in adapting to the precise control requirements under complex working conditions. (3) Due to the lack of real-time and effective monitoring data support, energy consumption management is lagging and it is impossible to optimize the electrolysis process in time to reduce energy consumption. At the same time, the fault warning capability is weak and it is difficult to predict faults such as anode current deviation and depolarization in advance, which can easily cause equipment damage and production interruption, and increase production costs. Summary of the Invention
[0004] The technical problem to be solved by the present invention is to overcome the existing defects and provide an intelligent integrated system and control method for aluminum electrolysis cells. This system enables real-time and accurate monitoring of anode current and cell temperature under high temperature and strong magnetic conditions, deep integration of monitoring data with the cell control system, intelligent early warning of current deviation, and automatic optimization of operating parameters. This improves the level of intelligence and economic benefits of aluminum electrolysis production and can effectively solve the problems in the background technology.
[0005] To achieve the above objectives, the present invention provides the following technical solution: an intelligent integrated system for aluminum electrolysis cells, comprising:
[0006] The multi-source parameter sensing and transmission module includes an integrated anode current-cell temperature sensor adapted to high-temperature and strong magnetic environments. The anode current sensing element is a high-precision shunt resistor and a Hall element connected in series, while the cell temperature sensing element is a platinum-rhodium thermocouple. Both are integrated into the same high-temperature resistant ceramic base and are installed in a distributed layout at the base of each anode guide rod in the electrolytic cell, 5-8 cm below the electrolyte surface. It supports parallel sampling of 24-48 anodes in a single cell. The module integrates RS485 / LoRa dual communication units, with a sampling period ≤10s, data transmission delay ≤500ms, and a communication error rate ≤10% under strong electromagnetic interference conditions in the electrolytic cell. -6 ;
[0007] The three-layer integrated unit includes a sensing layer, a control layer, and an application layer. The sensing layer establishes a bidirectional data link with the original cell control system of the electrolytic cell through the standardized OPC UA protocol. The control layer has a built-in dual-core heterogeneous processor, model ARM Cortex-A7+M4, to realize real-time uploading of monitoring data and parallel processing of control commands. The application layer builds an integrated engine of "data parsing-model calculation-command generation" to form a closed-loop control of "monitoring-analysis-control-optimization" with a closed-loop response cycle of ≤30s.
[0008] The intelligent algorithm unit includes an anode current distribution mathematical model and a bias current prediction and control algorithm. The anode current distribution mathematical model is based on the electric field distribution mechanism of the electrolytic cell and integrates multi-source data such as anode current, cell temperature, cell voltage, electrolyte temperature, and alumina concentration. It uses an improved Kalman filter algorithm to adaptively adjust the noise covariance and remove high-frequency noise and sensor drift error. The bias current prediction and control algorithm adopts a dual-model fusion architecture of "LSTM neural network data-driven model + electrochemical reaction mechanism model". The bias current intelligent early warning response time is ≤10s and the early warning accuracy is ≥95%.
[0009] The hardware and software adaptation unit, including the host computer monitoring interface and terminal control program, supports real-time rendering of anode current distribution heat map with a refresh rate of ≥10Hz, multi-dimensional historical trend query, adjustable time granularity from 1s to 1 year, fault classification alarm, and first-level alarm response time ≤1s. For pole switching interference and bus voltage fluctuation of ±10V~±30V, a dynamic data compensation algorithm and voltage adaptive adjustment mechanism are designed. After more than 1000 hours of simulation and more than 50 on-site joint debugging and optimization, it has achieved continuous stable operation time of ≥8000 hours on 330kA and 400kA series electrolytic cells.
[0010] As a preferred embodiment of the present invention, the high-temperature resistant ceramic base of the integrated anode current-tank temperature sensor is made of aluminum nitride ceramic material, with a package wall thickness of 3-5mm and a yttrium oxide anti-corrosion coating sprayed on the surface. The precious metal contact material is platinum-iridium alloy, and the contact pressure is pre-tightened by a disc spring with a pre-tightening force of 5-8N. The sensor layout satisfies the requirement that the distance between adjacent anode sampling points is 30-50cm, and the anode current measurement accuracy is ≤±1.5% and the tank temperature measurement accuracy is ≤±2℃.
[0011] As a preferred embodiment of the present invention, the RS485 / LoRa dual communication unit includes an MCU main control module, an RS485 conversion chip, and a LoRa communication module, wherein:
[0012] The RS485 unit is equipped with an optocoupler isolation circuit, a positive temperature coefficient thermistor and a transient suppression diode. The positive temperature coefficient thermistor is rated at 25℃ / 10Ω, which realizes dual isolation of power supply and signal and withstands surge voltage ≥2kV.
[0013] The LoRa unit supports adaptive frequency hopping in the 433-470MHz band, with a channel switching time of ≤50ms. It adopts an adjustable spreading factor design of SF7-SF12, with a maximum transmit power of 17dBm and a receive sensitivity of -148dBm.
[0014] The dual communication mode employs a "link quality assessment + automatic switching" mechanism. When the RS485 link bit error rate > 10, -5 If the signal-to-noise ratio of the LoRa link is less than 10dB, it will automatically switch to the backup link with a switching response time of ≤1s.
[0015] As a preferred technical solution of the present invention, the sensing layer in the three-layer architecture integrated unit uses SPI+DMA dual interface to communicate with the integrated sensor. The SPI clock frequency is ≥1MHz, and the DMA channel realizes multi-parameter parallel acquisition and buffering with acquisition delay ≤10ms.
[0016] In the dual-core heterogeneous processor of the control layer, the Cortex-A7 core is responsible for data processing and network communication, while the Cortex-M4 core is responsible for real-time control instruction issuance. Data interaction is achieved through an internal high-speed bus, and the control instruction issuance delay is ≤300ms.
[0017] The application layer has a built-in industrial-grade SD card and NAND Flash dual storage unit. The SD card capacity is ≥64GB and the NAND Flash capacity is ≥128GB. It supports ≥1 year of historical data storage, with a sampling frequency of 10 seconds / time and millisecond-level query response. The query latency is ≤50ms, and the data is stored using AES-256 encryption.
[0018] As a preferred technical solution of the present invention, the improved Kalman filtering algorithm includes: an adaptive adjustment formula for noise covariance based on the variance of the current sampled data and the variance of the historical data: Q(k) = Q0×(σ²(k) / σ0²);
[0019] Where Q0 is the initial process noise covariance, σ²(k) is the variance of the current sampled data, and σ0² is the average variance of the historical data;
[0020] The anode current sampling data is filtered in real time, and the signal-to-noise ratio of the filtered data is ≥45dB.
[0021] As a preferred embodiment of the present invention, the bias flow prediction control algorithm includes:
[0022] The data-driven model based on LSTM neural network takes the anode current and tank temperature time series data within the past 10 minutes as input, with 60 sampling points, 3 hidden layers, 128 neurons in each layer, and is trained using the Adam optimizer. The prediction step size is 5 steps, each step is 10 seconds, and the output is the bias trend prediction result.
[0023] Based on the mechanistic model of the electrochemical reaction mechanism in the electrolyzer, mathematical relationships between the bias current and the cell voltage, electrolyte composition, and anodic polarization resistance are established using Faraday's law and Ohm's law.
[0024] I_bias = f (U_cell, C_Al2O3, R_anode)
[0025] Where I_bias is the anode bias current, which is the difference between the actual current of a single anode and the average current of the anode; U_cell is the cell voltage of the electrolytic cell, which is the total voltage between the anode and cathode of the electrolytic cell; C_Al2O3 is the mass fraction of aluminum oxide in the electrolyte, with a value ranging from 2% to 5%; and R_anode is the anode polarization resistance, which is the resistance generated at the interface between the anode and the electrolyte due to the electrochemical reaction.
[0026] The dual-model fusion adopts a weighted voting mechanism. The weights ω1 of the data-driven model and ω2 of the mechanism model satisfy ω1+ω2=1. ω1 is dynamically adjusted according to the data credibility. When the credibility is ≥90%, ω1=0.7; otherwise, ω1=0.3.
[0027] When the predicted deviation value exceeds the set threshold by ±5%, the anode position adjustment command and electrolyte addition command are automatically generated, with adjustment steps of 0.1mm and 0.5kg respectively. These commands are sent to the tank control system terminal through the control layer, and the adjustment response time is ≤2s.
[0028] As a preferred technical solution of the present invention, the host computer monitoring interface adopts a B / S architecture design, supports remote access via a web browser, and includes an anode current distribution heat map, color gradient corresponding to current density 0-5A / cm², temperature trend curve, multi-curve comparison display, fault alarm pop-up window, and graded display of red first-level alarm and yellow second-level alarm functions.
[0029] The terminal control program supports local manual / remote automatic dual-mode control. Mode switching is authorized by a hardware key. When the system experiences a communication abnormality and three consecutive data transmission failures, it automatically switches to the local emergency control mode, in which the core control functions are retained.
[0030] As a preferred technical solution of the present invention, the software and hardware adaptation unit designs a dynamic data compensation algorithm for the pole-switching interference condition: when a pole-switching operation is detected, and the anode current change is determined to be ≥20%, the weighted fusion calculation of historical synchronous data and real-time sampled data is automatically enabled. The historical synchronous data is the data of the last 3 pole-switching synchronous operations, and the weight is dynamically adjusted with the pole-switching process. The weight of historical data is 0.8 in the early stage of pole-switching, and the weight of real-time data is 0.8 in the later stage of pole-switching.
[0031] For bus voltage fluctuation conditions, a voltage threshold judgment + sliding window smoothing mechanism is adopted: the voltage fluctuation threshold is set to ±10V, ±20V, and ±30V, corresponding to different smoothing window sizes of 5 sampling points, 10 sampling points, and 15 sampling points, respectively. When the fluctuation amplitude is ≤±30V, the system measurement accuracy deviation is ≤±0.5%.
[0032] An electrolytic cell control method includes the following steps:
[0033] S1: The multi-source parameter sensing and transmission module collects anode current and tank temperature data through an integrated sensor with sampling accuracy of 16-bit and 12-bit respectively. The data is transmitted to the control layer via RS485 / LoRa dual communication unit, and CRC32 check is used during the transmission process.
[0034] S2: After receiving data, the Cortex-M4 core of the control layer transmits it to the Cortex-A7 core through the internal high-speed bus. The Cortex-A7 core calls the improved Kalman filter algorithm for noise reduction, and the filtered data is uploaded to the application layer.
[0035] S3: The application layer inputs the filtered data into the mathematical model of anode current distribution, and combines it with data on tank voltage, electrolyte temperature, and alumina concentration. Then, it uses a dual-model fusion architecture of the bias current prediction and control algorithm to predict the bias current trend.
[0036] S4: Control commands are sent from the Cortex-M4 core of the control layer to the original cell control system of the electrolytic cell. The anode position and electrolysis parameters are automatically adjusted through PID regulation, and feedback data is collected in real time during the adjustment process.
[0037] S5: Feedback data is transmitted to the application layer via the communication unit. The application layer corrects the model parameters of the bias current prediction control algorithm based on the feedback data, updates the mathematical model of anode current distribution, and forms a closed-loop optimization. The closed-loop optimization cycle is once every 30 seconds.
[0038] As a preferred technical solution of the present invention, the model parameter correction of the bias flow prediction control algorithm in step S3 adopts the online gradient descent method, and the correction step size is dynamically adjusted according to the prediction error. When the error is ≥5%, the step size is 0.01, and when the error is <5%, the step size is 0.001.
[0039] In step S4, the proportional coefficient Kp, integral coefficient Ki, and derivative coefficient Kd of the PID control are pre-tuned using a particle swarm optimization algorithm, with the anode position adjustment accuracy ≤ ±0.1 mm and the electrolyte addition amount adjustment accuracy ≤ ±0.5 kg.
[0040] In step S5, during the closed-loop optimization process, the electric field distribution parameters of the mathematical model of anode current distribution are updated in real time.
[0041] Compared with the prior art, the beneficial effects of the present invention are: (1) Through the structural design and anti-interference optimization of the integrated sensor of anode current and tank temperature, the anode current measurement accuracy is ≤±1.5%, the tank temperature measurement accuracy is ≤±2%, and the sampling period is ≤10s, which solves the problem of lag and low accuracy of traditional manual measurement and provides data support for precise control; (2) By adopting a three-layer architecture and standardized OPC UA protocol, the interaction link between monitoring data and tank control system is opened up, the closed-loop response period is ≤30 seconds, and the full-process automation of "monitoring-analysis-control-optimization" is realized, which solves the defect of the separation of monitoring and control in the prior art; (3) The bias current prediction control algorithm of dual model fusion realizes the early warning response time is ≤10s and the early warning accuracy is ≥95%, the depolarization accident rate is significantly reduced, and the current efficiency is improved by 1%~2% through automatic parameter optimization, the electricity consumption per ton of aluminum is reduced by 3%~5%, and the anode carbon consumption is reduced by 2kg / t-Al, which has significant economic benefits. Attached Figure Description
[0042] Figure 1 This is a schematic diagram of the overall architecture of this intelligent integrated system;
[0043] Figure 2 This is a flowchart of the control method of the present invention. Detailed Implementation
[0044] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0045] Example 1:
[0046] Please see Figure 1-2 The present invention provides a technical solution: an intelligent integrated system for aluminum electrolysis cells includes a multi-source parameter sensing and transmission module, a three-layer architecture integration unit, an intelligent algorithm unit, and a software and hardware adaptation unit. The modules work together to achieve deep integration of monitoring and control.
[0047] The multi-source parameter sensing and transmission module is used to achieve accurate acquisition and stable transmission of multi-source parameters such as anode current and tank temperature, specifically including:
[0048] Integrated Anode Current-Cell Temperature Sensor: Utilizing a dual-sensitive element co-packaged structure, the anode current sensing element is a high-precision shunt resistor and Hall element connected in series, while the cell temperature sensing element is a platinum-rhodium thermocouple. Both are integrated within the same high-temperature resistant ceramic substrate, ensuring a compact structure and adaptability to high-temperature, strong magnetic environments. The sensor withstands temperatures ranging from 900℃ to 1200℃ and magnetic field strengths from 0.5T to 1.2T. It is installed in a distributed layout at the base of each anode guide rod in the electrolytic cell, 5–8 cm below the electrolyte surface, supporting parallel sampling of 24–48 anodes per cell, ensuring comprehensive sampling coverage.
[0049] High-temperature resistance and anti-interference design: The high-temperature resistant ceramic base of the sensor is made of aluminum nitride ceramic material with a wall thickness of 3-5mm. The surface is coated with yttrium oxide anti-corrosion coating to resist the high temperature and corrosive environment inside the electrolytic cell. The precious metal contact material is made of platinum-iridium alloy and is pre-tightened by disc springs with a pre-tightening force of 5-8N to ensure reliable contact. The installation process adopts a "magnetic shielding bracket + high-temperature fire-resistant fiber seal" structure. The magnetic shielding bracket is made of permalloy with a shielding effectiveness of ≥80dB, effectively reducing the impact of strong electromagnetic interference on measurement accuracy. The anode current measurement accuracy is ≤±1.5%, and the cell temperature measurement accuracy is ≤±2℃.
[0050] RS485 / LoRa Dual Communication Unit: Integrates an MCU main control module, an RS485 conversion chip, and a LoRa communication module. It employs a triple anti-interference design of "antenna impedance matching circuit + ferrite bead array filtering + differential signal transmission" to ensure stable data transmission. The RS485 unit features an optocoupler isolation circuit, a positive temperature coefficient thermistor (25℃ / 10Ω), and a transient suppression diode to achieve dual power and signal isolation, with surge voltage resistance ≥2kV. The LoRa unit supports adaptive frequency hopping in the 433~470MHz band, with a channel switching time ≤50ms. It adopts an adjustable spread spectrum factor (SF7~SF12), a maximum transmit power of 17dBm, and a receive sensitivity of -148dBm.
[0051] The dual communication mode employs a "link quality assessment + automatic switching" mechanism. When the RS485 link bit error rate > 10... -5 If the signal-to-noise ratio of the LoRa link is <10dB, it will automatically switch to the backup link, with a switching response time ≤1s, a sampling period ≤10s, a data transmission delay ≤500ms, and a communication bit error rate ≤10 in the strong electromagnetic interference environment of the electrolytic cell. -6 .
[0052] The three-layer architecture integration unit is used to connect the data links between sensing and control, enabling efficient interaction between monitoring data and control commands. Specifically, it includes:
[0053] Architecture Design: A "loosely coupled + hard real-time" integrated design is adopted, ensuring compatibility with the existing cell control system while meeting real-time control requirements. The sensing layer and the existing cell control system establish a bidirectional data link through the standardized OPC UA protocol, ensuring the universality and stability of data interaction.
[0054] The control layer has a built-in ARM Cortex-A7+M4 dual-core heterogeneous processor. The Cortex-A7 core is responsible for data processing and network communication, while the Cortex-M4 core is responsible for real-time control command issuance. The two interact with each other through an internal high-speed bus, and the control command issuance delay is ≤300ms.
[0055] The application layer constructs an integrated engine of "data parsing - model calculation - instruction generation" to form a closed-loop control of "monitoring - analysis - control - optimization" with a closed-loop response cycle of ≤30s;
[0056] Functional details of each layer: The perception layer uses SPI+DMA dual interfaces to communicate with the integrated sensor. The SPI clock frequency is ≥1MHz, and the DMA channel realizes parallel acquisition and buffering of multiple parameters with an acquisition delay of ≤10ms.
[0057] The application layer has a built-in industrial-grade SD card and NAND Flash dual storage unit, with an SD card capacity of ≥64GB and a NAND Flash capacity of ≥128GB. It supports ≥1 year of historical data storage, a sampling frequency of 10 seconds / time and millisecond-level query response, and a query latency of ≤50ms. Data is stored using AES-256 encryption to ensure data security.
[0058] Intelligent algorithm unit: used to implement data noise reduction, bias prediction, and control command generation, specifically including:
[0059] An improved Kalman filter algorithm is proposed. Based on the electric field distribution mechanism of the electrolytic cell, it integrates multi-source data such as anode current, cell temperature, cell voltage, electrolyte temperature, and alumina concentration. It achieves the removal of high-frequency noise and sensor drift error through adaptive adjustment of noise covariance. The core formula of the algorithm includes: Noise covariance adaptive adjustment formula Q(k) = Q0×(σ²(k) / σ0²).
[0060] Where Q0 is the initial process noise covariance, σ²(k) is the variance of the current sampled data, and σ0² is the average variance of the historical data. The anode current sampling data is filtered in real time, and the signal-to-noise ratio of the filtered data is ≥45dB.
[0061] Side current prediction and control algorithm: A dual-model fusion architecture of "LSTM neural network data-driven model + electrochemical reaction mechanism model" is adopted to improve the accuracy and generalization ability of side current prediction, as detailed below:
[0062] Data-driven model: Based on LSTM neural network, the input is the time series data of anode current and tank temperature within the past 10 minutes, with 60 sampling points, 3 hidden layers, 128 neurons in each layer, trained with Adam optimizer, prediction step size of 5 steps, 10 seconds per step, and output the bias trend prediction result.
[0063] Mechanism model: The mathematical relationship between bias current and cell voltage, electrolyte composition, and anodic polarization resistance is established through Faraday's law and Ohm's law: I_bias = f (U_cell, C_Al2O3, R_anode);
[0064] Where I_bias is the anode bias current, which is the difference between the actual current of a single anode and the average current of the anode; U_cell is the cell voltage, which is the total voltage between the anode and cathode of the electrolytic cell; C_Al2O3 is the mass fraction of alumina in the electrolyte, ranging from 2% to 5%; and R_anode is the anode polarization resistance, which is the resistance generated at the interface between the anode and the electrolyte due to the electrochemical reaction. This mathematical relationship establishes a quantitative correlation between bias current and the above three core influencing factors by quantifying the influence of cell voltage on the driving force of electrochemical reaction, the influence of alumina concentration on the conductivity of electrolyte, and the hindering effect of anode polarization resistance on current distribution.
[0065] Dual-model fusion: A weighted voting mechanism is adopted, and the weights ω1 of the data-driven model and ω2 of the mechanism model satisfy ω1+ω1=1. ω1 is dynamically adjusted according to the data credibility. When the credibility is ≥90%, ω1=0.7; otherwise, ω1=0.3.
[0066] When the predicted flow deviation exceeds the set threshold by ±5%, the anode position adjustment command and electrolyte addition command are automatically generated, with adjustment steps of 0.1mm and 0.5kg respectively. These commands are sent to the tank control system terminal through the control layer. The adjustment response time is ≤2s, the flow deviation intelligent early warning response time is ≤10s, and the early warning accuracy is ≥95%.
[0067] Hardware and software adaptation unit: used to realize data visualization, operation control and adaptation to complex working conditions, specifically including:
[0068] The host computer monitoring interface and terminal control program: The host computer monitoring interface adopts a B / S architecture design, supports remote access via a web browser, and includes an anode current distribution heatmap (color gradient corresponds to current density 0~5A / cm²), temperature trend curve (multi-curve comparison display), and fault alarm pop-up window (tiered display: red first-level alarm, yellow second-level alarm). The refresh rate is ≥10Hz, the time granularity is adjustable from 1s to 1 year, and the first-level alarm response time is ≤1s. The terminal control program supports local manual / remote automatic dual-mode control. Mode switching is authorized by a hardware key. When the system experiences a communication anomaly of three consecutive data transmission failures, it automatically switches to the local emergency control mode. In emergency mode, the core control functions are retained to ensure production safety.
[0069] Adaptation to complex operating conditions: For pole switching interference conditions, a dynamic data compensation algorithm is designed. When a pole switching operation is determined by a sudden change in anode current of ≥20%, the weighted fusion calculation of the data from the three most recent pole switching operations and the real-time sampled data is automatically activated. The weight is dynamically adjusted with the pole switching process, i.e., the weight of historical data is 0.8 in the early stage of pole switching and the weight of real-time data is 0.8 in the later stage of pole switching.
[0070] For bus voltage fluctuations of ±10V to ±30V, a "voltage threshold judgment + sliding window smoothing" mechanism is adopted. Voltage fluctuation thresholds are set to ±10V, ±20V, and ±30V, with corresponding smoothing window sizes of 5, 10, and 15 sampling points, respectively. When the fluctuation amplitude is ≤±30V, the system measurement accuracy deviation is ≤±0.5%. The system has undergone more than 1000 hours of simulation and more than 50 on-site joint debugging and optimization, and has achieved continuous stable operation time of ≥8000 hours on 330kA and 400kA series electrolytic cells.
[0071] Taking the 330kA series electrolytic cell as an example, the model selection and parameter settings are as follows:
[0072] Multi-source parameter sensing and transmission module: The integrated sensor has a high-temperature resistant ceramic base with a wall thickness of 4mm, an adjacent anode sampling point spacing of 40cm, an RS485 conversion chip of ADM2483, a LoRa communication module of SX1278, an MCU main control module of STM32H7 series, a positive temperature coefficient thermistor of 25℃ / 10Ω, a transient suppression diode of SMBJ6.5CA, and an optocoupler isolation circuit of 6N137.
[0073] The three-layer integrated unit consists of a control layer processor that is an ARM Cortex-A7+M4 dual-core heterogeneous chip with an SPI clock frequency of 2MHz and a DMA channel cache capacity of 1KB; and an application layer SD card with a capacity of 128GB, a NAND Flash capacity of 256GB, and data storage using AES-256 encryption.
[0074] Intelligent Algorithm Unit: The initial process noise covariance of the improved Kalman filter algorithm is Q0=[[1e-6,0],[0,1e-8]], the sliding window size is N=20, and the number of historical statistical samples is M=100; the learning rate of the LSTM neural network is 0.001, the decay rate is β1=0.9, β2=0.999, and the loss function is mean square error; the bias current reference value I_bias_ref is set according to the number of anodes and the rated current of the 330kA electrolytic cell, the average current of a single anode is 330kA / 30=11kA, and the bias current threshold is set to ±550A (i.e. ±5%×11kA).
[0075] Hardware and software adaptation unit: The heat map refresh rate of the host computer monitoring interface is 15Hz, and the fault alarm pop-up response time is 0.5s; the local emergency control mode of the terminal control program retains the functions of anode position adjustment, electrolyte addition control and emergency shutdown.
[0076] Its control method includes the following steps:
[0077] S1: Multi-source parameter acquisition: An integrated sensor is installed at the base of 30 anode guide rods and 6cm below the electrolyte level in a 330kA electrolytic cell to acquire anode current and cell temperature data in parallel. The sampling accuracy is 16-bit and 12-bit, respectively. The data is transmitted to the control layer via RS485 / LoRa dual communication units, and CRC32 verification is used during the transmission process.
[0078] S2: Data noise reduction: After receiving data, the Cortex-M4 core in the control layer transmits it to the Cortex-A7 core through the AHB3 high-speed bus, calls the improved Kalman filter algorithm to process the anode current data, and the signal-to-noise ratio of the filtered data is 48dB before being uploaded to the application layer;
[0079] S3: Bias Current Prediction: The application layer inputs the filtered data along with the cell voltage (set value 4.2V), electrolyte temperature (set value 960℃), and alumina concentration (set value 3.5%) into the anode current distribution mathematical model, and predicts the bias current value by fusing the LSTM data-driven model with the mechanism model;
[0080] When the predicted bias current value is 600A (exceeding the threshold of 550A), an audible and visual alarm is triggered, and a red pop-up window prompt is displayed on the host computer. An anode position adjustment command is generated with an adjustment amount of 0.012mm, corresponding to 600A×0.02mm / A, and an electrolyte addition command with an adjustment amount of 0.05kg, corresponding to 50A×0.1kg / A.
[0081] S4: Parameter Adjustment: Control commands are sent to the existing tank control system via the Cortex-M4 core. The anode position and electrolyte addition amount are adjusted through PID control. The PID parameters are pre-tuned by the particle swarm optimization algorithm, with Kp=2.5, Ki=0.1, and Kd=0.5. The anode position adjustment accuracy is ±0.08mm, and the electrolyte addition amount adjustment accuracy is ±0.3kg.
[0082] S5: Closed-loop optimization: Feedback data shows that the adjusted bias current value is 520A (within the threshold range). The application layer uses the online gradient descent method to correct the LSTM model parameters with a correction step size of 0.005 (error 4.5% < 5%). The anode polarization resistance parameter is updated, and the closed-loop optimization cycle is 30s.
[0083] Example 2: Module selection and parameter settings for the 400kA series electrolytic cell intelligent integrated system:
[0084] Multi-source parameter sensing and transmission module: The high-temperature resistant ceramic base of the integrated sensor has a wall thickness of 5mm, and the spacing between adjacent anode sampling points is 35cm; the LoRa frequency band of the RS485 / LoRa dual communication unit is set to 470MHz, the spreading factor SF=10, and the maximum transmit power is 17dBm.
[0085] The three-layer integrated unit has an SPI clock frequency of 1.5MHz and an acquisition latency of 8ms; the application layer data storage time is 1.5 years and the query latency is 40ms.
[0086] Intelligent algorithm unit: bias reference value I_bias_ref=400kA / 36≈11.11kA, bias threshold set to ±555.5A; the prediction step size of the LSTM neural network is 5 steps, each step is 10s, and it outputs the bias trend for the next 50 seconds.
[0087] Hardware and software adaptation unit: For the bus voltage fluctuation characteristics of the 400kA electrolytic cell, the voltage fluctuation threshold is extended to ±30V, and the corresponding smoothing window size is 15 sampling points.
[0088] The specific control method is as follows:
[0089] S1: Multi-source parameter acquisition: 36 integrated sensors acquire anode current and cell temperature data of 400kA electrolytic cell in parallel, and transmit the data to the control layer via LoRa communication module (RS485 link backup).
[0090] S2: Data noise reduction: After processing with the improved Kalman filter algorithm, the data signal-to-noise ratio is 46dB before being uploaded to the application layer;
[0091] S3: Bias Current Prediction: The fusion model predicts a bias current value of -600A (negative bias current indicates insufficient current), triggering a yellow alarm pop-up window and generating an anode position adjustment command and an electrolyte addition command;
[0092] S4: Parameter adjustment: After PID adjustment, the anode position adjustment accuracy is ±0.09mm, and the electrolyte addition amount adjustment accuracy is ±0.4kg;
[0093] S5: Closed-loop optimization: The corrected bias current value is -530A, the model parameter correction step size is 0.006, and the closed-loop optimization is completed.
[0094] Examples 1 and 2 verify that the intelligent integrated system and control method of the present invention can be adapted to 330kA and 400kA series electrolytic cells. Under different parameter settings, it can achieve accurate monitoring, intelligent early warning and parameter optimization, and operate stably and reliably, meeting the intelligent and green requirements of electrolytic aluminum production.
[0095] The parts of the invention not described in detail are prior art. Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.
Claims
1. An intelligent integrated system for aluminum electrolysis cells, characterized in that: include: The multi-source parameter sensing and transmission module includes an integrated anode current-cell temperature sensor adapted to high-temperature and strong magnetic environments. The anode current sensing element is a high-precision shunt resistor and a Hall element connected in series, while the cell temperature sensing element is a platinum-rhodium thermocouple. Both are integrated into the same high-temperature resistant ceramic base and are installed in a distributed layout at the base of each anode guide rod in the electrolytic cell, 5-8 cm below the electrolyte surface. It supports parallel sampling of 24-48 anodes in a single cell. The module integrates RS485 / LoRa dual communication units, with a sampling period ≤10s, data transmission delay ≤500ms, and a communication error rate ≤10% under strong electromagnetic interference conditions in the electrolytic cell. -6 ; The three-layer integrated unit includes a sensing layer, a control layer, and an application layer. The sensing layer establishes a bidirectional data link with the original cell control system of the electrolytic cell through the standardized OPC UA protocol. The control layer has a built-in dual-core heterogeneous processor, model ARM Cortex-A7+M4, to realize real-time uploading of monitoring data and parallel processing of control commands. The application layer builds an integrated engine of "data parsing-model calculation-command generation" to form a closed-loop control of "monitoring-analysis-control-optimization" with a closed-loop response cycle of ≤30s. The intelligent algorithm unit includes an anode current distribution mathematical model and a bias current prediction and control algorithm. The anode current distribution mathematical model is based on the electric field distribution mechanism of the electrolytic cell, integrating multi-source data such as anode current, cell temperature, cell voltage, electrolyte temperature, and alumina concentration. It uses an improved Kalman filter algorithm to adaptively adjust the noise covariance and remove high-frequency noise and sensor drift errors. The bias current prediction and control algorithm adopts a dual-model fusion architecture of "LSTM neural network data-driven model + electrochemical reaction mechanism model", with a bias current intelligent early warning response time ≤10s and an early warning accuracy ≥95%. The hardware and software adaptation unit, including the host computer monitoring interface and terminal control program, supports real-time rendering of anode current distribution heat map with a refresh rate of ≥10Hz, multi-dimensional historical trend query with adjustable time granularity from 1s to 1 year, fault classification alarm with a first-level alarm response time of ≤1s. For pole switching interference and bus voltage fluctuations of ±10V to ±30V, a dynamic data compensation algorithm and voltage adaptive adjustment mechanism are designed. After more than 1000 hours of simulation and more than 50 on-site joint debugging and optimization, it has achieved continuous stable operation time of ≥8000 hours on 330kA and 400kA series electrolytic cells.
2. The intelligent integrated system for aluminum electrolysis cells according to claim 1, characterized in that: The high-temperature resistant ceramic base of the integrated anode current-tank temperature sensor is made of aluminum nitride ceramic material, with a package wall thickness of 3-5mm and a yttrium oxide anti-corrosion coating sprayed on the surface. The precious metal contact material is platinum-iridium alloy, and the contact pressure is pre-tightened by a disc spring with a pre-tightening force of 5-8N. The sensor layout meets the requirement that the distance between adjacent anode sampling points is 30-50cm, and the anode current measurement accuracy is ≤±1.5% and the tank temperature measurement accuracy is ≤±2℃.
3. The intelligent integrated system for aluminum electrolysis cells according to claim 1, characterized in that: The RS485 / LoRa dual communication unit includes an MCU main control module, an RS485 conversion chip, and a LoRa communication module, wherein: The RS485 unit is equipped with an optocoupler isolation circuit, a positive temperature coefficient thermistor and a transient suppression diode. The positive temperature coefficient thermistor is rated at 25℃ / 10Ω, which realizes dual isolation of power supply and signal and withstands surge voltage ≥2kV. The LoRa unit supports adaptive frequency hopping in the 433-470MHz band, with a channel switching time of ≤50ms. It adopts an adjustable spreading factor design of SF7-SF12, with a maximum transmit power of 17dBm and a receive sensitivity of -148dBm. The dual communication mode utilizes a "link quality assessment + automatic switching" mechanism. When the RS485 link bit error rate > 10, -5 If the signal-to-noise ratio of the LoRa link is less than 10dB, it will automatically switch to the backup link with a switching response time of ≤1s.
4. The intelligent integrated system for aluminum electrolysis cells according to claim 1, characterized in that: In the three-layer architecture integrated unit, the sensing layer uses SPI+DMA dual interfaces to communicate with the integrated sensor. The SPI clock frequency is ≥1MHz, and the DMA channel realizes parallel acquisition and buffering of multiple parameters with an acquisition delay of ≤10ms. In the dual-core heterogeneous processor of the control layer, the Cortex-A7 core is responsible for data processing and network communication, while the Cortex-M4 core is responsible for real-time control instruction issuance. Data interaction is achieved through an internal high-speed bus, and the control instruction issuance delay is ≤300ms. The application layer has a built-in industrial-grade SD card and NAND Flash dual storage unit. The SD card capacity is ≥64GB and the NAND Flash capacity is ≥128GB. It supports ≥1 year of historical data storage, with a sampling frequency of 10 seconds / time and millisecond-level query response. The query latency is ≤50ms, and the data is stored using AES-256 encryption.
5. The intelligent integrated system for aluminum electrolysis cells according to claim 1, characterized in that: The improved Kalman filter algorithm includes: an adaptive adjustment formula for noise covariance based on the variance of the current sampled data and the variance of the historical data: Q(k) = Q0×(σ²(k) / σ0²); Where Q0 is the initial process noise covariance, σ²(k) is the variance of the current sampled data, and σ0² is the average variance of the historical data; The anode current sampling data is filtered in real time, and the signal-to-noise ratio of the filtered data is ≥45dB.
6. The intelligent integrated system for aluminum electrolysis cells according to claim 1, characterized in that: The eccentricity prediction control algorithm includes: The data-driven model based on LSTM neural network takes the anode current and tank temperature time series data within the past 10 minutes as input, with 60 sampling points, 3 hidden layers, 128 neurons in each layer, and is trained using the Adam optimizer. The prediction step size is 5 steps, each step is 10 seconds, and the output is the bias trend prediction result. Based on the mechanistic model of the electrochemical reaction mechanism in the electrolyzer, mathematical relationships between the bias current and cell voltage, electrolyte composition, and anodic polarization resistance are established using Faraday's law and Ohm's law. I_bias = f (U_cell, C_Al2O3, R_anode) Where I_bias is the anode bias current, which is the difference between the actual current of a single anode and the average current of the anode; U_cell is the cell voltage of the electrolytic cell, which is the total voltage between the anode and cathode of the electrolytic cell; C_Al2O3 is the mass fraction of aluminum oxide in the electrolyte, with a value ranging from 2% to 5%; and R_anode is the anode polarization resistance, which is the resistance generated at the interface between the anode and the electrolyte due to the electrochemical reaction. The dual-model fusion adopts a weighted voting mechanism. The weights ω1 of the data-driven model and ω2 of the mechanism model satisfy ω1+ω2=1. ω1 is dynamically adjusted according to the data credibility. When the credibility is ≥90%, ω1=0.7; otherwise, ω1=0.
3. When the predicted deviation value exceeds the set threshold by ±5%, the anode position adjustment command and electrolyte addition command are automatically generated, with adjustment steps of 0.1mm and 0.5kg respectively. These commands are sent to the tank control system terminal through the control layer, and the adjustment response time is ≤2s.
7. The intelligent integrated system for aluminum electrolysis cells according to claim 1, characterized in that: The host computer monitoring interface adopts a B / S architecture design, supports remote access via a web browser, and includes an anode current distribution heat map, color gradient corresponding to current density 0-5A / cm², temperature trend curve, multi-curve comparison display, fault alarm pop-up window, and graded display of red first-level alarm and yellow second-level alarm functions. The terminal control program supports local manual / remote automatic dual-mode control. Mode switching is authorized by a hardware key. When the system experiences a communication abnormality and three consecutive data transmission failures, it automatically switches to the local emergency control mode, in which the core control functions are retained.
8. The intelligent integrated system for aluminum electrolysis cells according to claim 1, characterized in that: The software and hardware adaptation unit is designed with a dynamic data compensation algorithm for pole switching interference conditions: when a pole switching operation is detected, and the anode current change is ≥20%, the weighted fusion calculation of historical synchronous data and real-time sampled data is automatically enabled. The historical synchronous data is the data of the last 3 pole switching synchronous operations. The weight is dynamically adjusted with the pole switching process. The weight of historical data is 0.8 in the early stage of pole switching and the weight of real-time data is 0.8 in the later stage of pole switching. For bus voltage fluctuation conditions, a voltage threshold judgment + sliding window smoothing mechanism is adopted: the voltage fluctuation threshold is set to ±10V, ±20V, and ±30V, corresponding to different smoothing window sizes of 5 sampling points, 10 sampling points, and 15 sampling points, respectively. When the fluctuation amplitude is ≤±30V, the system measurement accuracy deviation is ≤±0.5%.
9. A method for controlling an electrolytic cell based on the system described in any one of claims 1-8, characterized in that: Includes the following steps: S1: The multi-source parameter sensing and transmission module collects anode current and tank temperature data through an integrated sensor with sampling accuracy of 16-bit and 12-bit respectively. The data is transmitted to the control layer via RS485 / LoRa dual communication unit, and CRC32 check is used during the transmission process. S2: After receiving data, the Cortex-M4 core of the control layer transmits it to the Cortex-A7 core through the internal high-speed bus. The Cortex-A7 core calls the improved Kalman filter algorithm for noise reduction, and the filtered data is uploaded to the application layer. S3: The application layer inputs the filtered data into the mathematical model of anode current distribution, and combines it with data on tank voltage, electrolyte temperature, and alumina concentration. Then, it uses a dual-model fusion architecture of the bias current prediction and control algorithm to predict the bias current trend. S4: Control commands are sent from the Cortex-M4 core of the control layer to the original cell control system of the electrolytic cell. The anode position and electrolysis parameters are automatically adjusted through PID regulation, and feedback data is collected in real time during the adjustment process. S5: Feedback data is transmitted to the application layer via the communication unit. The application layer corrects the model parameters of the bias current prediction control algorithm based on the feedback data, updates the mathematical model of anode current distribution, and forms a closed-loop optimization. The closed-loop optimization cycle is once every 30 seconds.
10. The electrolytic cell control method according to claim 9, characterized in that: In step S3, the model parameter correction of the bias flow prediction control algorithm adopts the online gradient descent method. The correction step size is dynamically adjusted according to the prediction error. When the error is ≥5%, the step size is 0.01, and when the error is <5%, the step size is 0.
001. In step S4, the proportional coefficient Kp, integral coefficient Ki, and derivative coefficient Kd of the PID control are pre-tuned using a particle swarm optimization algorithm, with the anode position adjustment accuracy ≤ ±0.1 mm and the electrolyte addition amount adjustment accuracy ≤ ±0.5 kg. In step S5, during the closed-loop optimization process, the electric field distribution parameters of the mathematical model of anode current distribution are updated in real time.