A method and system for intelligent diagnosis and optimization of a cathode copper electrolytic cell based on CORAL-random forest

Abstract

This invention discloses a method and system for intelligent diagnosis and optimization of cathode copper electrolytic cells based on CORAL-random forest. The method includes: S1, collecting multimodal data during the operation of the cathode copper electrolytic cell; S2, based on the multimodal data, aligning the feature distributions between different modal data using CORAL feature calibration technology to obtain calibrated feature data; S3, based on the calibrated feature data, identifying the fault types of the electrolytic cell using an intelligent diagnosis model containing a CORAL dynamic calibration layer and a weighted random forest; S4, optimizing the key operating parameters of the electrolytic cell according to the identified fault types, combining the electrolysis process mechanism model and an improved particle swarm optimization algorithm; S5, generating a health report containing fault diagnosis and parameter optimization information based on the optimization results.

Description

Technical Field

[0001] This invention relates to the field of intelligent control in non-ferrous metal smelting, and in particular to an intelligent diagnosis and optimization method and system for cathode copper electrolytic cells based on CORAL-random forest. Background Technology

[0002] Electrolytic refining of cathode copper is a crucial process for increasing copper purity to over 99.99%, and the stable operation of the electrolytic cell directly determines production efficiency and product quality. However, the operation and control of traditional cathode copper electrolytic cells have significant technical limitations: Fault diagnosis is delayed and inaccurate: relying on manual inspection makes it impossible to detect latent faults such as plate passivation and micro-leakage in a timely manner. For example, if plate passivation is not addressed in time, it can lead to a 5%-10% decrease in current efficiency, and the accuracy of manual fault type identification is less than 80%, which can easily cause the fault to escalate.

[0003] Parameter optimization lacks quantitative basis: adjustments to key parameters such as electrolyte temperature, sulfuric acid concentration, and current density are mostly based on operators' historical experience, without considering real-time fault conditions. For example, a deviation of ±5℃ in electrolyte temperature can cause the purity of cathode copper to drop below 99.95%, failing to meet the requirements of the high-purity copper industry.

[0004] Existing technologies have significant drawbacks: some existing systems only use single-modal data or ordinary machine learning algorithms, ignoring the differences in feature distribution among different modalities, resulting in poor model generalization ability and insufficient adaptability across operating conditions; moreover, most systems only implement "fault diagnosis" functions, failing to form a closed loop with parameter optimization, and thus cannot directly guide production adjustments, limiting the technology's practicality. Therefore, an intelligent system is needed that can solve the problems of "low diagnostic accuracy due to differences in feature distribution," "lack of a fault-optimization closed loop," and "insufficient data privacy protection," enabling real-time diagnosis, precise optimization, and safe operation and maintenance of cathode copper electrolytic cells. Summary of the Invention

[0005] To address the technical problems mentioned above, this invention provides a method for intelligent diagnosis and optimization of cathode copper electrolytic cells based on CORAL-random forest, comprising the following steps: S1. Collect multi-modal data during the operation of the cathode copper electrolytic cell; S2. Based on multimodal data, the feature distributions between different modal data are aligned using CORAL feature calibration technology to obtain calibrated feature data; S3. Based on the calibrated feature data, an intelligent diagnostic model including a CORAL dynamic calibration layer and a weighted random forest is used to identify the fault type of the electrolyzer. S4. Based on the identified fault type, and combining the electrolysis process mechanism model with the improved particle swarm optimization algorithm, optimize and solve the key operating parameters of the electrolyzer. S5. Based on the optimization solution results, generate a health report containing fault diagnosis and parameter optimization information.

[0006] Preferably, S1 includes: Based on a preset sampling frequency or a dynamic adaptive sampling strategy, the electrode plate image data, electrical parameter data, physicochemical parameter data, and vibration data of the electrolytic cell are collected synchronously. The electrical parameter data includes: tank voltage and current density; The physicochemical parameters include: electrolyte temperature, sulfuric acid concentration, and copper ion concentration.

[0007] Preferably, S2 includes: The collected multimodal data is preprocessed to obtain processed feature data; Based on the processed feature data, the covariance matrix between the real-time feature distribution and the preset training set feature distribution is calculated, and a distribution-calibrated feature matrix is ​​generated based on the CORAL transformation.

[0008] Preferably, the fault diagnosis model in S3 includes: CORAL dynamic calibration layer: calculates the covariance bias between input features and training set features in real time. ,when When the value is greater than 0.05, a CORAL transformation is automatically triggered to ensure consistent feature distribution; among which, This represents the covariance bias between the input features and the training set features; Covariance matrix representing input features Covariance matrix of features from the training set The Frobenius norm between them.

[0009] Weighted Random Forest Layer: To address the uneven weight distribution of key features in traditional random forests, a weighted Gini coefficient is introduced as the node splitting criterion to enhance the contribution of fault-related features. The formula is as follows: Where K is the number of fault categories, p ik Let be the probability that a sample belongs to the k-th type of fault under the i-th feature; the model is set with 100 decision trees, the feature sampling rate is 0.7, and the output is the fault type and the 95% confidence interval.

[0010] Preferably, before fault diagnosis in S3, a pre-training step of cross-cell model based on federated learning is added, including: training the fault diagnosis sub-model on local data through edge computing nodes of several electrolytic cells, and only uploading the updated model parameters to the cloud for secure aggregation to protect the data privacy of each production unit.

[0011] Preferably, S4 includes: A dual-objective optimization function is constructed with the constraints of maximizing current efficiency and cathode copper purity. Electrolyte temperature, sulfuric acid concentration, current density, and electrode spacing are used as optimization variables. An improved particle swarm optimization algorithm with linearly decreasing inertia weights and adaptive learning factors is used to solve the problem and obtain the optimal parameter combination.

[0012] Preferably, S5 includes: A Bayesian neural network is used to quantify the uncertainty of the prediction results of the optimal parameter combination and output the confidence interval.

[0013] This invention also provides an intelligent diagnosis and optimization system for cathode copper electrolytic cells based on CORAL-random forest, the system being used to implement the above method, comprising: The multimodal data acquisition module is used to simultaneously acquire electrical parameters, physicochemical parameters, electrode state, and vibration data. Edge computing nodes are deployed in the electrolysis workshop to perform real-time data preprocessing, enhancement, CORAL calibration, and emergency fault warning. The cloud-based analytics platform receives data uploaded from edge computing nodes and runs the CORAL-improved random forest diagnostic model, the PSO parameter optimization model, and the uncertainty quantification model. The parameter optimization execution module works in conjunction with the electrolytic cell control system to convert the optimal parameters output from the cloud into execution commands. A visual terminal is used to display fault location, optimization parameter trends, and health reports.

[0014] Compared with the prior art, the beneficial effects of the present invention are as follows: (1) High-precision fault diagnosis: By calibrating the differences in multimodal data distribution through CORAL feature calibration, and combining weighted random forest to improve the contribution of key features, the average diagnostic accuracy of 5 types of faults is ≥95%, which is 10 percentage points higher than that of traditional random forest (85%). The fault location accuracy is ≤0.5m, which solves the problems of lagging and low accuracy of traditional manual diagnosis.

[0015] (2) Dynamic feature adaptation: The CORAL dynamic calibration layer responds to changes in feature distribution in real time, ensuring the model's generalization ability across working conditions (such as different batches of electrolytes and electrode materials). The distribution correlation coefficient deviation is ≤5%, which is significantly better than the no-calibration scheme (deviation 12%-15%). When the model is migrated to a new electrolytic cell, there is no need to retrain on a large scale.

[0016] (3) Closed-loop parameter optimization: For the first time, fault diagnosis and parameter optimization are combined into a closed loop. The optimal parameters are output in reverse based on the fault type. The dual-objective optimization is achieved by improving the PSO algorithm. The current efficiency is increased by 4%-6%, and the purity of cathode copper is stabilized at over 99.99%. The accuracy is 10 times higher than the empirical adjustment scheme (purity fluctuation ±0.005%).

[0017] (4) Real-time security operation and maintenance: The edge-cloud collaborative architecture achieves data processing latency ≤100ms and emergency fault response ≤10s. In case of critical faults such as leakage, the power supply can be quickly cut off to reduce the risk of accidents.

[0018] (5) Data privacy protection: Federated learning pre-training realizes "data does not leave the factory and models are trained together", protecting the data privacy of each production unit, while improving the initial performance of the model (initial accuracy is improved by 20%-30%), which meets the requirements of industrial data security.

[0019] (6) Data augmentation and adaptive sampling. Through data augmentation techniques and adaptive sampling strategies, the model's adaptability to different working conditions was improved, ensuring the diversity and real-time nature of the data.

[0020] (7) Cross-modal association learning and dynamic fusion mechanism. Through cross-modal association learning and dynamic fusion mechanism, more effective multimodal data fusion is achieved, which improves the model's ability to detect the state of the electrolyzer.

[0021] (8) Multi-scale feature extraction and adaptive anomaly detection threshold. By using multi-scale feature extraction and adaptive anomaly detection threshold mechanism, the comprehensiveness of feature extraction and the accuracy of anomaly detection are improved.

[0022] (9) Real-time inversion update and multiphysics coupling effect. By considering real-time inversion update and multiphysics coupling effect, the accuracy and reliability of fault diagnosis are improved.

[0023] (10) Enhanced interpretability of multi-factor comprehensive evaluation and prediction model. By enhancing the interpretability of multi-factor comprehensive evaluation and prediction model, the comprehensiveness and scientific nature of electrolytic cell fault diagnosis are improved.

[0024] (11) Intelligent diagnosis and early warning functions and system integration and deployment optimization. Through intelligent diagnosis and early warning functions and system integration and deployment optimization, the practicality and operating efficiency of the system are improved. Attached Figure Description

[0025] To more clearly illustrate the technical solution of the present invention, the drawings used in the embodiments are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0026] Figure 1 This is a schematic diagram of the method flow according to an embodiment of the present invention; Figure 2 This is a heatmap of the design variable correlation matrix in an embodiment of the present invention. Detailed Implementation

[0027] 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.

[0028] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0029] Example 1 This embodiment provides a method for intelligent diagnosis and optimization of cathode copper electrolytic cells based on CORAL-random forest. Figure 1 As shown, the steps include: S1. Collect multimodal data during the operation of the cathode copper electrolytic cell.

[0030] Using a high-resolution industrial camera and an integrated multi-sensor array, the electrolytic cell's electrical parameters (cell voltage, current density), physicochemical parameters (electrolyte temperature, sulfuric acid concentration, copper ion concentration), electrode condition data (surface image, passivation degree), and cell vibration data are simultaneously acquired under a unified spatiotemporal reference. The specific acquisition method is as follows: Electrode status acquisition: A high-resolution industrial camera is used to acquire images of the electrode surface at fixed intervals; a laser rangefinder is used to simultaneously measure the electrode spacing, with the sampling frequency consistent with that of the camera.

[0031] Electrical Parameter Acquisition: Electrical parameters are acquired using cell voltage sensors and current density sensors. The cell voltage sensor is connected in parallel to the positive and negative electrodes of the electrolytic cell via insulated terminals; the current density sensor is connected in series to the copper busbar of the main current circuit of the electrolytic cell (using a through-hole installation method to reduce current loss). The sampling frequency of both types of sensors is set to 10Hz to ensure the capture of instantaneous fluctuations in electrical parameters.

[0032] Physicochemical parameter acquisition: Temperature monitoring uses a PT100 platinum resistance sensor, inserted 5-10 cm below the electrolyte level through a pre-drilled temperature measuring hole on the side wall of the tank; sulfuric acid concentration monitoring sensor is installed in the electrolyte circulation outlet pipe; copper ion concentration monitoring sensor is fixed in the electrolyte flow area in the middle of the tank. The sampling interval for the three types of sensors is set to 5 minutes to balance real-time performance and power consumption.

[0033] Vibration data acquisition: An accelerometer is deployed in the middle of the side wall of the tank. It is attached to the middle of the side wall of the tank with high-temperature epoxy adhesive, avoiding the electrolyte inlet and outlet pipes (vibration of the pipes will cause the sensor to be falsely triggered. The vibration signal-to-noise ratio at this location is measured to be 40% higher). It is used to determine the stability of the tank structure.

[0034] All sensors are connected to a unified clock module, and data is uploaded to an edge server (deployed in the workshop control room) via the Profinet protocol.

[0035] S2. Based on multimodal data, the feature distributions between different modal data are aligned using CORAL feature calibration technology to obtain calibrated feature data.

[0036] S21. After the multimodal data acquisition is completed, preprocess it.

[0037] (1) Data cleaning Outlier removal: A combined strategy of "3σ rule + sliding window filtering" is used to process the raw data—for each sensor, the mean μ and standard deviation σ are calculated based on 30 days of historical data, and outliers exceeding these values ​​are removed. The values ​​are marked as outliers and replaced with the moving average of the previous 5 normal data (e.g., if the temperature sensor has a history of μ=60℃ and σ=2℃, then data <54℃ or >66℃ will be replaced) to ensure data continuity.

[0038] A sliding window averaging filter with a length of 10 was used to remove instantaneous pulse interference (such as voltage spikes reaching 5.5V), resulting in a 65% improvement in data smoothness after filtering. Data normalization: After removing outliers, perform Min-Max normalization on the data to map the feature values ​​to the [0,1] interval. The formula is as follows: Where x represents the original data, and xmin and xmax represent the historical minimum and maximum values ​​of the feature, respectively (based on statistics of normal data over the past 90 days), to avoid interference from features of different magnitudes on the model.

[0039] The normalization of key parameters is shown in Table 1: Table 1 .

[0040] (2) Data Augmentation Image enhancement: Performing spatial geometric transformations, including scaling transformations, on electrode images acquired by industrial cameras: Scaling ratio Randomly selected between 0.9 and 1.1; I represents the original electrode image to be scaled.

[0041] Noise injection: Add Gaussian noise with an intensity of 0.5%-1% to improve the model's robustness to image noise.

[0042] Time-series data enhancement: Gaussian white noise is added to the time-series signals of electrical and physicochemical parameters. noise intensity Dynamically adjusted based on real-time signal-to-noise ratio (if the signal-to-noise ratio is high). Take 2%, if the signal-to-noise ratio is low... (Take 0.5%) to increase data diversity and avoid model overfitting. This represents the Gaussian white noise component to be added to the original time-series signals containing electrical and physicochemical parameters, which follows a mean of 0 and a variance of . The normal distribution; N represents a Gaussian white noise sequence that follows a standard normal distribution N(0,1), with its dimension consistent with the original time-series signal x, used to correlate noise intensity. Combine the generated noise components to be added.

[0043] S22. Use CORAL feature calibration technology to perform feature alignment on the processed multimodal data to obtain calibrated feature data.

[0044] Because there are distribution differences between real-time acquired data (such as electrolyte temperature baseline shift caused by seasonal changes) and historical training data, the CORAL (correlation alignment) algorithm is used to achieve cross-domain feature adaptation: 1000 normal operating condition samples are collected in real time (covering different time periods and different tanks) to construct feature matrix X (1000×8, 8-dimensional features as described above); 10000 samples are selected from the historical training set (including 6000 normal operating conditions and 800 samples of each of the 5 types of faults) to construct feature matrix Y (10000×8).

[0045] (1) Calculation of covariance matrix To address the distribution differences between different modal data and low-frequency concentration data, a CORAL transformation is used to align the feature distribution. Let the original feature matrix be... (where n is the number of samples and d is the feature dimension), the target distribution feature matrix is: (where m is the number of samples in the training set), then: Where 1 represents a vector consisting entirely of 1s, and its dimension is the same as the number of samples.

[0046] (2) Eigenvalue decomposition Eigenvalue decomposition of ΣX and ΣY yields: Where UX and UY are orthogonal eigenvector matrices. , It is a diagonal eigenvalue matrix.

[0047] (3) Calculation of the inverse square root and square root of the covariance matrix in, , It is a diagonal matrix. , These are the inverse square root and square root of the corresponding elements.

[0048] (4) Feature calibration The generated calibrated feature matrix is ​​mathematically expressed as follows: Where ΣX is the original feature covariance matrix, and ΣY is the target distribution covariance matrix. , The correlation coefficient of the characteristic distribution after calibration is ≤5% after solving by eigenvalue decomposition.

[0049] After calibration, the deviation of the distribution correlation coefficient between the real-time features and the training set features is ≤5%, which significantly improves the model's generalization ability.

[0050] (5) Dynamic sampling strategy The sampling frequency is adjusted based on the real-time status of the electrolytic cell, and the calculation formula is as follows: Among them, the fundamental frequency Adjustment factor Based on the current slot voltage With electrolyte temperature calculate: ( , (This is the average value under normal operating conditions); when When the voltage fluctuation is greater than 0.1 (i.e., voltage fluctuation > 10% or temperature fluctuation > 10%), the sampling frequency is increased to 10Hz to capture the fault evolution process more densely; when When the value is less than 0.1, the base frequency is maintained to reduce energy consumption.

[0051] S3. Based on the calibrated feature data, an intelligent diagnostic model including a CORAL dynamic calibration layer and a weighted random forest is used to identify the fault type of the electrolyzer.

[0052] The intelligent diagnostic model in this embodiment achieves high-precision identification and localization of five typical fault types through "CORAL dynamic calibration + weighted random forest". The specific technical solution is as follows: (1) Model structure design The model consists of a CORAL dynamic calibration layer and a weighted random forest layer, which work together to ensure the accuracy of the diagnosis.

[0053] ①CORAL Dynamic Calibration Layer: Calculates the covariance bias between input features and training set features in real time. ,when When the value is greater than 0.05, the CORAL transformation is automatically triggered to ensure the consistency of feature distribution. ② Weighted Random Forest Layer: To address the shortcomings of traditional random forests in "uneven weight distribution of key features," a weighted Gini coefficient is introduced as a node splitting criterion to enhance the contribution of fault-related features. The formula is as follows: Where K is the number of fault categories, p ik Let be the probability that a sample belongs to the k-th type of fault under the i-th feature; the model is set with 100 decision trees, the feature sampling rate is 0.7, and the output is the fault type and the 95% confidence interval.

[0054] (2) Calculation of feature weights The Gini coefficient importance assessment method based on random forests is used to calculate the contribution weights of 8-dimensional features. Specifically, this is achieved through mutual information. The contribution of the i-th feature to the fault category fi is calculated using the following formula: Where Vfi is the set of values ​​for feature fi. As a probability distribution, mutual information is normalized into weights. ,make sure v represents a specific value of feature fi (an element in the set of values ​​of feature fi, Vfi); c represents a specific category of fault category (an element in the set of fault categories, C).

[0055] The sorting results are shown below: Voltage (0.25, directly reflects the change in internal resistance of the tank) > Plate image features (0.22, intuitively captures the oxide layer / damage state) > Temperature (0.18, affects the reaction rate and electrolyte stability) > Sulfuric acid concentration (0.15, related to electrolyte conductivity) > Current density (0.10, reflects the uniformity of energy input) > Vibration acceleration (0.06, related to the structural stability of the tank) > Copper ion concentration (0.03, has a lagging effect on purity) > Plate spacing (0.01, small fluctuation under normal operating conditions).

[0056] (3) Model parameter settings The system uses 100 decision trees, with a feature sampling rate of 0.7 for each tree (70% of features are randomly selected), and a minimum number of samples per leaf node of 5 to avoid overfitting. Five-fold cross-validation is used to optimize model parameters and ensure generalization ability.

[0057] (4) Scope and characteristics of fault diagnosis The model supports the identification of five types of typical faults in cathode copper electrolytic cells. The core characteristics and diagnostic logic of each type of fault are as follows: Electrode passivation: The core characteristics are a rise in tank voltage >0.2V and the appearance of a black oxide layer on the electrode surface image (the area of ​​the oxide layer after image segmentation is >5%), with a diagnostic accuracy of ≥96%; Abnormal electrolyte concentration: sulfuric acid concentration <180g / L or >220g / L, or copper ion concentration <3g / L or >8g / L, combined with changes in conductivity (conductivity fluctuation >10% when concentration is abnormal), the diagnostic accuracy is ≥97%; Uneven cathode deposition: Copper layer thickness deviation after image segmentation of electrode surface > 0.5 mm, electrode spacing deviation measured by laser rangefinder > 2 mm, diagnostic accuracy ≥ 95%; Leakage current: The tank voltage drops suddenly by more than 5% and the current does not change significantly. The vibration sensor detects abnormal vibration (acceleration > 1g), and the diagnostic accuracy is ≥ 98%. Anodic dissolution anomaly: current density fluctuation >10A / m², irregular corrosion areas appear on the anode surface image (corrosion area ratio >8%), diagnostic accuracy ≥94%.

[0058] (5) Fault location and uncertainty quantification ① Fault localization: Combining sensor deployment coordinates with image segmentation technology, spatial fault localization is achieved. An improved U-Net network was used to perform semantic segmentation on the electrode surface image to segment fault regions (such as passivation layer and corrosion area), with a segmentation accuracy of ≥92%. Based on the installation location and lens parameters (focal length, angle of view) of the industrial camera, the image pixel coordinates are mapped to the actual physical coordinates; For non-image-related faults (such as leakage current), the fault location can be determined based on the sensor deployment location (such as the voltage sensor installation end).

[0059] ② Uncertainty Quantification: The confidence interval of the fault diagnosis results is calculated by using the Bootstrap method (resampled 100 times), and a 95% confidence level is output to provide risk reference for maintenance decisions.

[0060] To protect the data privacy of each production unit, federated learning is introduced during the model initialization phase: Multiple electrolytic cells (e.g., 10) are used as edge computing nodes to train a fault diagnosis sub-model on local data, uploading only the model parameters (weights, biases) to the cloud. The cloud then uses a federated averaging algorithm to aggregate the parameters and generate a global initial model, using the following formula: Where M is the number of edge nodes, and ni is the number of local samples of the i-th node. These are local model parameters; after the global model is distributed to each edge node, it is fine-tuned by combining new local data to achieve "global generalization + local adaptation", without the need to transmit the original data, thus protecting privacy.

[0061] S4. Based on the identified fault type, and combining the electrolysis process mechanism model with the improved particle swarm optimization algorithm, optimize and solve the key operating parameters of the electrolyzer.

[0062] This step, based on fault diagnosis results and combined with the electrolysis process mechanism model, constructs a dual-objective parameter optimization model to achieve closed-loop control of "fault repair + production indicator improvement". The specific technical solution is as follows: S41. Construct a dual-objective optimization function with the constraints of maximizing current efficiency and cathode copper purity.

[0063] (1) With maximizing current efficiency η as the main objective and the purity of cathode copper as the constraint objective, the mathematical expression is: Where T is the electrolyte temperature. Where J is the sulfuric acid concentration, J is the current density, and D is the electrode spacing (optimization variable). The constraint condition is that the performance index P (which, depending on the scenario, usually refers to the performance compliance rate of the electrode plate, the finished product qualification rate, etc.) is not less than 99.99%.

[0064] (2) Constraints: ① Hard constraints: P≥99.99% (standard for electronic grade electrolytic copper), temperature 50~70℃ (to avoid electrolyte crystallization / evaporation), electrode spacing 100~150mm (to prevent short circuit / excessive voltage drop), circulation pump frequency 30~60Hz (to ensure uniform electrolyte).

[0065] ② Soft constraints: current efficiency η≥90% (considering production capacity) and energy consumption ≤300kWh / t (controlling costs).

[0066] (3) Customizable target weight: Supports adjusting the weight of current efficiency and purity according to production needs. For example, in high-purity scenarios (such as electronic-grade copper), the purity weight is increased from 0.4 to 0.6 to adapt to different application scenarios.

[0067] S42. Using electrolyte temperature, sulfuric acid concentration, current density, and electrode spacing as optimization variables, an improved particle swarm optimization algorithm incorporating linearly decreasing inertia weights and adaptive learning factors is employed to obtain the optimal parameter combination.

[0068] An improved PSO algorithm is used to solve the bi-objective optimization model. To address the problem that traditional PSO is prone to getting trapped in local optima, a linearly decreasing inertial weight and an adaptive learning factor are introduced to improve convergence speed and optimization accuracy.

[0069] (1) Particle initialization Particles represent combinations of optimization variables. Population size 50, particle positions randomly initialized within constraints, velocity initialization range: .

[0070] Fitness function design: A fitness function is constructed by combining current efficiency and purity, and the formula is as follows: in, ( As the current efficiency weight, (For purity weighting), when the product purity index P < 99.99, the fitness function value is multiplied by 0.5 to penalize solutions that do not meet the constraints.

[0071] (2) Particle velocity and position update in, The inertial weights decrease linearly from 0.9 to 0.4 (for early-stage global search and later-stage local optimization); c1 and c2 are learning factors that are adaptively adjusted: when particle aggregation is high, c1=2.5 and c2=1.5 (enhancing individual exploration); when particle dispersion is high, c1=1.5 and c2=2.5 (enhancing global convergence); r1 and r2 are random numbers in the range [0,1]. This represents the historical best position of the i-th particle. This is the globally optimal position for the population.

[0072] (3) Convergence conditions The algorithm stops when the fitness function changes by less than 0.1% over 10 consecutive generations, or when the number of iterations reaches 100, and outputs the optimal parameter combination. The optimization time is ≤30s.

[0073] S5. Based on the optimization solution results, generate a health report containing fault diagnosis and parameter optimization information.

[0074] To evaluate the reliability of the optimization parameters, this embodiment introduces a Bayesian neural network (BNN) to quantify the uncertainty, specifically including: (1) BNN model structure The input layer contains the optimization variables, there are two hidden layers (64 neurons per layer, ReLU activation function), and the output layer contains the predicted values ​​and variances of current efficiency and purity.

[0075] (2) Model training The model parameters are sampled using the Markov Chain Monte Carlo (MCMC) method, trained based on historical optimization data (1000+ sets), and the 95% confidence interval of the predicted value is output (e.g., current efficiency 92.5% ± 1.2%).

[0076] (3) Uncertainty visualization A heatmap showing the confidence intervals of optimized parameters and prediction results is displayed on the visualization terminal, such as... Figure 2 As shown, this helps operators determine the risk level of parameters (e.g., a narrow confidence interval indicates low risk and can be applied directly; a wide confidence interval suggests small-batch verification). The flowchart of this embodiment is shown below. Figure 1 As shown.

[0077] Example 2 This embodiment also provides an intelligent diagnosis and optimization system for cathode copper electrolytic cells based on CORAL-random forest, the system comprising: The multimodal data acquisition module is used to simultaneously acquire electrical parameters, physicochemical parameters, electrode state, and vibration data.

[0078] Edge computing nodes are deployed on-site in the electrolysis workshop to perform real-time data preprocessing, enhancement, CORAL calibration, and emergency fault warning.

[0079] The cloud-based analytics platform receives data uploaded from edge nodes and runs the CORAL-improved random forest diagnostic model, the PSO parameter optimization model, and the uncertainty quantification model.

[0080] The parameter optimization execution module works in conjunction with the electrolytic cell control system to convert the optimal parameters output from the cloud into execution commands.

[0081] A visual terminal is used to display fault location, optimization parameter trends, and health reports.

[0082] The multimodal data acquisition module includes: a plate image acquisition device; a high-resolution industrial camera; and electrical parameter sensors: a slot voltage sensor and a current density sensor.

[0083] Physicochemical parameter sensors: platinum resistance temperature sensor, sulfuric acid concentration sensor, copper ion concentration sensor; ① Vibration sensor: Accelerometer.

[0084] ② Edge computing nodes are configured to perform the following tasks: The sensor sampling frequency is dynamically adjusted based on the built-in algorithm. When a leakage fault is detected (sudden drop in cell voltage >5% and lasts >1s), an audible and visual alarm is immediately triggered and the power supply to the branch of the electrolytic cell is cut off, with a response delay of ≤10s.

[0085] The cloud-based analytics platform includes: a CORAL-Random Forest diagnostic engine for performing feature calibration and fault classification; a dual-objective parameter optimization solver using a GPU-accelerated (CUDA-based) improved PSO algorithm; and an uncertainty quantification module that integrates a Bayesian neural network to output the 95% confidence interval of the optimization results.

[0086] Integrated parameter optimization solver: Electrolysis mechanism database, storing physicochemical parameter thresholds, Faraday law coefficients, etc. for cathode copper of different purities; Multi-objective weight customization interface, supporting adjustment of the weights of current efficiency and purity according to production needs (e.g., increasing the purity weight to 0.6 in high-purity scenarios).

[0087] The visual terminal provides the following functions: ① Electrolytic cell real-time status dashboard: Displays real-time curves of cell voltage, temperature, current efficiency, and abnormal indicators are highlighted in red; ② 3D Fault Visualization: Based on the CAD model of the tank, the fault area is superimposed, and the color depth corresponds to the fault confidence level (red ≥90%, yellow 70%-90%). ③ Optimization effect comparison and analysis: Display the predicted difference in current efficiency and purity before and after optimization, as well as historical data backtracking (supports Excel export).

[0088] An adaptive data compression and encrypted transmission protocol is used between edge computing nodes and the cloud analytics platform: for critical data with abnormal scores exceeding the threshold (such as leakage faults), lossless compression and strong encryption are used for immediate transmission; for normal data, lossy compression (compression ratio 20:1) based on compression sensing theory and TLS1.3 standard encryption are used for periodic batch transmission, and the communication protocol is industrial Ethernet.

[0089] The parameter optimization execution module is linked with the electrolyzer control system via the Modbus protocol to drive the following actuators: ① Electrolyte temperature control: The temperature is adjusted by heating / cooling devices; ② Plate spacing adjustment: The spacing is adjusted via a servo lifting motor; ③ Electrolyte circulation rate control: The rate is adjusted by a variable frequency circulation pump.

[0090] The visualization terminal integrates a display module, which can overlay the fault area map, optimal parameters and maintenance suggestions generated in the cloud onto the operator's field of vision through the display device, so as to realize monitoring and maintenance guidance.

[0091] Edge computing nodes have a built-in lightweight anomaly detection model (pruned CORAL-random forest) for initial screening of real-time streaming data. Once a suspected anomaly is detected, the complete model (including CORAL-random forest with image segmentation module) downloaded from the cloud is immediately activated for confirmatory analysis, forming a "cloud-edge collaboration, hierarchical early warning" mechanism.

[0092] Example 3 The system in Example 2 adopts an "edge-cloud collaborative" architecture, which balances real-time performance and computing power to achieve a closed loop for the entire process of data acquisition, processing, diagnosis, and optimization. The specific architecture design is as follows: Edge computing nodes: Deployed on-site in the electrolysis workshop, responsible for real-time data processing and emergency fault early warning.

[0093] Real-time data processing: Performs data cleaning, enhancement, and CORAL calibration with a processing latency of ≤100ms; Employs parallel computing (multi-threaded) to process multi-sensor data and supports parallel data preprocessing for multiple electrolytic cells.

[0094] Emergency fault warning: The built-in lightweight anomaly detection model (pruned CORAL-random forest, with a 60% reduction in parameter size) immediately triggers an audible and visual alarm (workshop audible and visual alarm) and cuts off the branch power supply of the electrolytic cell when an emergency fault such as leakage or sudden temperature rise is detected.

[0095] Data transmission strategy: Adaptive data compression and encrypted transmission are adopted: lossy compression batch transmission based on compression awareness reduces bandwidth usage.

[0096] Cloud-based analytics platform: Deployed on the enterprise's private cloud, responsible for core algorithm operation and big data storage.

[0097] Core algorithm module: CORAL - Improved Random Forest Diagnostic Engine: GPU accelerated (based on CUDA 11.0), single-cell fault diagnosis time ≤2s, supports parallel computing of 100+ electrolytic cells.

[0098] Dual-objective parameter optimization solver: GPU-accelerated PSO algorithm, population iteration speed improved by 5 times, supports simultaneous optimization of multiple slots.

[0099] Uncertainty quantification module: BNN model inference uses GPU parallel computing.

[0100] Data storage and management: Time-series databases are used to store multimodal data, supporting data backtracking and trend analysis.

[0101] Model update mechanism: The diagnostic model is incrementally trained based on newly collected fault data to update the model parameters; through model version management, historical version backtracking is supported to ensure model stability.

[0102] Parameter optimization execution module: This module works in conjunction with the electrolytic cell control system to automatically execute optimization parameters. Its specific functions are as follows: Protocol integration: Communicates with the electrolytic cell PLC control system via Modbus-RTU protocol, supporting parameter distribution and execution status feedback.

[0103] Executive control: ① Electrolyte temperature: Adjust the temperature to the optimal value by controlling the heating rod power (0-100%) or the cooling water pump speed (0-3000rpm), with a control accuracy of ±0.5℃.

[0104] ② Plate spacing: The plate spacing is adjusted by a servo lifting motor (accuracy ±0.1mm), with a motor response speed of 5mm / s.

[0105] ③ Electrolyte circulation rate: The rate is adjusted by a variable frequency circulation pump (0-50Hz), and the circulation volume control accuracy is ±1L / min.

[0106] ④ Execution status monitoring: Real-time acquisition of the operating parameters of the actuator (such as motor speed and pump frequency). When the execution deviation is greater than 5%, a secondary adjustment is triggered to ensure the accuracy of parameter execution.

[0107] Visual terminal: Developed based on a web platform, it provides an intuitive human-computer interaction interface. Its core functions are as follows: ① Real-time status dashboard: Displays real-time curves of key indicators of the electrolytic cell (cell voltage, temperature, current efficiency, purity), provides alerts for abnormal indicators (such as voltage exceeding the normal range), and supports custom threshold values ​​for indicators.

[0108] ② Fault visualization: Fault areas are overlaid based on the tank CAD model, and the color depth corresponds to the fault confidence level (e.g., red ≥90%, yellow 70%-90%, green <70%). Clicking on the fault area can view detailed information (fault type, occurrence time, characteristic data).

[0109] ③ Optimization effect comparison: Displays the predicted difference in current efficiency and purity before and after optimization, and supports querying historical optimization effects (search by time and slot number).

[0110] ④ Health Report Generation: Automatically generates daily / weekly health reports, including fault statistics, optimization parameter summaries, and maintenance suggestions, and supports Excel / PDF export.

[0111] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made to the technical solutions of the present invention by those skilled in the art without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. A CORAL-random forest-based intelligent diagnosis and optimization method for cathode copper electrolytic cell, characterized in that, Includes the following steps: S1. Collect multi-modal data during the operation of the cathode copper electrolytic cell; S2. Based on multimodal data, the feature distributions between different modal data are aligned using CORAL feature calibration technology to obtain calibrated feature data; S3. Based on the calibrated feature data, an intelligent diagnostic model including a CORAL dynamic calibration layer and a weighted random forest is used to identify the fault type of the electrolyzer. S4. Based on the identified fault type, and combining the electrolysis process mechanism model with the improved particle swarm optimization algorithm, optimize and solve the key operating parameters of the electrolyzer. S5. Based on the optimization solution results, generate a health report containing fault diagnosis and parameter optimization information.

2. The CORAL-Random Forest based cathode copper cell intelligent diagnosis and optimization method according to claim 1, characterized in that, S1 includes: Based on a preset sampling frequency or a dynamic adaptive sampling strategy, the electrode plate image data, electrical parameter data, physicochemical parameter data, and vibration data of the electrolytic cell are collected synchronously. The electrical parameter data includes: tank voltage and current density; The physicochemical parameters include: electrolyte temperature, sulfuric acid concentration, and copper ion concentration. 3.The CORAL-Random Forest based cathode copper cell intelligent diagnosis and optimization method according to claim 1, characterized in that, S2 includes: The collected multimodal data is preprocessed to obtain processed feature data; Based on the processed feature data, the covariance matrix between the real-time feature distribution and the preset training set feature distribution is calculated, and a distribution-calibrated feature matrix is ​​generated based on the CORAL transformation.

4. The CORAL-Random Forest based cathode copper cell intelligent diagnosis and optimization method according to claim 1, characterized in that, The fault diagnosis model in S3 includes: CORAL dynamic calibration layer: real-time compute covariance shift between input features and training set features, When > 0.05, automatically trigger CORAL transform to ensure feature distribution consistency; where, covariance shift between input features and training set features; covariance matrix of input features covariance matrix of training set features Frobenius norm between Weighted Random Forest Layer: To address the shortcomings of traditional random forests in "uneven weight distribution of key features," a weighted Gini coefficient is introduced as a node splitting criterion to enhance the contribution of fault-related features. The formula is as follows: where K is the number of failure categories, p ik is the probability of the sample belonging to the kth failure category under the ith feature; the model sets 100 decision trees, the feature sampling rate is 0.7, and the output failure type and 95% confidence interval.

5. The CORAL-Random Forest based cathode copper cell intelligent diagnosis and optimization method according to claim 4, characterized in that, Before fault diagnosis in S3, a pre-training step for a cross-cell model based on federated learning is added, including: training a fault diagnosis sub-model on local data through edge computing nodes of several electrolytic cells, and only uploading model parameter updates to the cloud for secure aggregation to protect the data privacy of each production unit.

6. The CORAL-Random Forest based cathode copper electrolysis cell intelligent diagnosis and optimization method according to claim 1, characterized in that S4 include: A dual-objective optimization function is constructed with the constraints of maximizing current efficiency and cathode copper purity. Electrolyte temperature, sulfuric acid concentration, current density, and electrode spacing are used as optimization variables. An improved particle swarm optimization algorithm with linearly decreasing inertia weights and adaptive learning factors is used to solve the problem and obtain the optimal parameter combination.

7. The CORAL-Random Forest based cathode copper cell intelligent diagnosis and optimization method according to claim 6, characterized in that S5 include: A Bayesian neural network is used to quantify the uncertainty of the prediction results of the optimal parameter combination and output the confidence interval.

8. A CORAL-Random Forest based cathode copper electrolysis cell intelligent diagnosis and optimization system, the system is used to implement the method of any one of claims 1-7, characterized in that, include: The multimodal data acquisition module is used to simultaneously acquire electrical parameters, physicochemical parameters, electrode state, and vibration data. Edge computing nodes are deployed in the electrolysis workshop to perform real-time data preprocessing, enhancement, CORAL calibration, and emergency fault warning. The cloud-based analytics platform receives data uploaded from edge computing nodes and runs the CORAL-improved random forest diagnostic model, the PSO parameter optimization model, and the uncertainty quantification model. The parameter optimization execution module works in conjunction with the electrolytic cell control system to convert the optimal parameters output from the cloud into execution commands. A visual terminal is used to display fault location, optimization parameter trends, and health reports.

Images