An electrolytic cell blanking control circuit optimization method based on groove voltage and area current cooperative measurement

Abstract

The application is suitable for the technical field of industrial process control and electrical system monitoring, and provides an electrolytic cell unloading control circuit optimization method based on groove voltage and regional current collaborative measurement. The method is executed by the unloading control circuit, and includes: constructing a working condition characteristic vector representing the comprehensive running state of the electrolytic cell based on the collected groove voltage signal and regional current signal; and processing by using a called feature decoupling analysis model to separate out independent feature components representing different physical abnormal sources; diagnosing the abnormal running state of the electrolytic cell and the partition where the electrolytic cell is located through the independent feature components to generate a state evaluation signal; simulating and predicting the basic control action based on the state evaluation signal and a dynamic coupling graph to generate a collaborative control instruction package, executing the collaborative control instruction package to drive the unloading execution mechanism to perform a point adjustment operation. Based on groove voltage and multi-regional current collaborative measurement, the electrolytic cell unloading control circuit optimization and precise regulation and control are realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of industrial process control and electrical system monitoring, in particular to an electrolytic cell discharge control circuit optimization method based on groove voltage and regional current collaborative measurement. BACKGROUND

[0002] The current electrolytic cell control system is mainly based on the collection of groove voltage and series current, two global parameters, to indirectly infer the average concentration of alumina in the electrolyte by calculating the pseudo-resistance, and to make decisions on the periodic discharge of the whole cell. Some technologies also use equal-distance pressure drop method or Hall sensor to obtain partial regional current data for auxiliary judgment of groove condition trend or simple abnormal alarm. Due to the increasing size of electrolytic cell and the complex internal reaction environment, relying only on the two global signals of groove voltage and series current cannot determine the real state of each region in the cell. Abnormal alumina concentration in local area, early anode effect or cathode damage and other faults will be averaged and diluted by the overall signal, so that the control system cannot identify and locate in time. In addition, the existing technology cannot analyze the multi-channel regional current signal and the groove voltage signal collaboratively, cannot generate control instructions with spatial targeting according to the diagnosis of local state, and still uses the one-size-fits-all method to uniformly regulate the whole cell, which not only cannot effectively suppress local faults, but also can cause new disturbances due to global action, ultimately leading to difficult improvement of current efficiency and high energy consumption.

[0003] In view of this, an electrolytic cell discharge control circuit optimization method based on groove voltage and regional current collaborative measurement is proposed. SUMMARY

[0004] The present application provides an electrolytic cell discharge control circuit optimization method based on groove voltage and regional current collaborative measurement, which is used to solve the problem that the prior art cannot effectively suppress local faults, and can cause new disturbances due to global action, ultimately leading to difficult improvement of current efficiency and high energy consumption.

[0005] The present application provides an electrolytic cell discharge control circuit optimization method based on groove voltage and regional current collaborative measurement, which is executed by a discharge control circuit, comprising:

[0006] Collecting groove voltage signals of the electrolytic cell and regional current signals of a plurality of preset partitions;

[0007] Based on the groove voltage signals and the regional current signals, a working condition feature vector representing the comprehensive running state of the electrolytic cell is constructed;

[0008] According to the working condition feature vector, a corresponding feature decoupling analysis model is selected and called, and the groove voltage signals and the regional current signals are processed by using the feature decoupling analysis model to separate out independent feature components representing different physical abnormal sources.

[0009] diagnose an abnormal running state of the electrolytic cell and a partition where the abnormal running state is located, and generate a state evaluation signal including an abnormal type, a location, and a risk level;

[0010] based on the state evaluation signal and a dynamic coupling graph reflecting coupling relationships of each partition of the electrolytic cell, simulate and predict a change in a full-cell current distribution for a basic control action for the abnormal running state;

[0011] based on a simulation prediction result, generate a collaborative control instruction package, the collaborative control instruction package including a main control instruction for a target abnormal partition and at least one collaborative compensation instruction for balancing a global current distribution;

[0012] execute the collaborative control instruction package to drive a point adjustment operation of a discharging execution mechanism of the target partition and an associated partition.

[0013] Further, based on the cell voltage signal and the regional current signal, a working condition feature vector representing a comprehensive running state of the electrolytic cell is constructed, including:

[0014] based on the regional current signal, a regional current distribution imbalance degree representing a uniformity of a current spatial distribution, and a regional current change coordination degree representing a time sequence relationship of current changes of each partition are calculated;

[0015] based on the cell voltage signal, a cell voltage stability index representing a voltage fluctuation feature is calculated;

[0016] the regional current distribution imbalance degree, the regional current change coordination degree, and the cell voltage stability index constitute the working condition feature vector.

[0017] Further, the corresponding feature decoupling analysis model is selected and called according to the working condition feature vector, including:

[0018] the working condition feature vector is input into a preset model matching decision maker; the model matching decision maker matches a target decoupling model from a pre-stored feature decoupling analysis model library according to a numerical combination of each index in the working condition feature vector; wherein, the feature decoupling analysis model library stores a plurality of pre-trained decoupling network models with different structures.

[0019] Further, the feature decoupling analysis model is used to process the cell voltage signal and the regional current signal to separate independent feature components representing different physical abnormal sources, including:

[0020] the time sequence sequence of the cell voltage signal and the time sequence sequence of each regional current signal are combined to form a multi-channel input time sequence matrix;

[0021] inputting the multi-channel input time series matrix into the target decoupling model;

[0022] processing the multi-channel input time series matrix through a convolution layer included in the target decoupling model and performing feature extraction and transformation in spatial and temporal dimensions, to obtain a plurality of independent intermediate feature vectors;

[0023] mapping the plurality of independent intermediate feature vectors into independent feature components representing different physical abnormal sources through a fully connected classification layer of the target decoupling model.

[0024] Further, the different physical abnormal sources include: a global reaction intensity disturbance source, a regional concentration disturbance source, an anode effect risk source, and a cathode state abnormal source.

[0025] The independent feature components representing different physical abnormal sources include: a global feature component representing overall reaction rate deviation, a concentration disturbance feature component representing local alumina concentration abnormality, an effect warning feature component representing anode effect occurrence risk, and a cathode abnormal feature component representing cathode damage or current leakage.

[0026] Further, the diagnosis of the abnormal operating state of the electrolytic cell and the partition where the abnormal operating state is located through the independent feature components includes:

[0027] comparing the value of the global feature component with a first global threshold value; if the value of the global feature component exceeds the first global threshold value, it is determined that there is a global reaction intensity disturbance abnormality in the electrolytic cell;

[0028] For each of the concentration disturbance feature component, the effect warning feature component, and the cathode abnormal feature component, the following operations are performed:

[0029] comparing the value of the feature component in each partition corresponding sub-feature component with the corresponding threshold value; if the value of the sub-feature component of a certain partition exceeds the threshold value, it is determined that there is an abnormality corresponding to the feature component in the partition;

[0030] wherein, the partition determined to exist concentration disturbance abnormality, anode effect risk abnormality or cathode state abnormality is determined as the partition where the abnormal operating state is located.

[0031] Further, the generation of the state evaluation signal containing abnormal type, positioning and risk level includes:

[0032] if it is determined that there is a global reaction intensity disturbance abnormality, the risk level corresponding to the global reaction intensity disturbance abnormality is determined according to the amplitude that the value of the global feature component exceeds the first global threshold value;

[0033] For each partition where the concentration disturbance anomaly, anode effect risk anomaly or cathode state anomaly exists, according to the magnitude by which the numerical value of the sub-feature component corresponding to the partition exceeds the threshold value, a risk level corresponding to the anomaly existing in the partition is determined;

[0034] The abnormal type, corresponding partition positioning and risk level corresponding to each abnormal type are combined to generate a state evaluation signal.

[0035] Further, the dynamic coupling graph is constructed and updated in the following manner:

[0036] Based on the time series data of the area current signals of each partition of the electrolytic cell, a current change transmission correlation coefficient between any two partitions is calculated as a coupling weight;

[0037] A directed graph structure is established with the partitions as nodes and the coupling weights as edges, forming a dynamic coupling graph;

[0038] According to a preset update period, the current change transmission correlation coefficient is recalculated to update the coupling weights in the dynamic coupling graph.

[0039] Further, the basic control action is determined in the following manner:

[0040] According to the abnormal type in the state evaluation signal, a corresponding standard control action template is matched from a preconfigured abnormal type and control action mapping table;

[0041] According to the risk level in the state evaluation signal, the control parameters in the standard control action template are quantitatively adjusted to generate a basic control action.

[0042] Further, the simulation prediction of the change in the full-cell current distribution caused by the basic control action for the abnormal operating state includes:

[0043] The basic control action is quantified as a feedstock quantity adjustment instruction for the target abnormal partition;

[0044] Based on the coupling weights between the nodes in the dynamic coupling graph, current change prediction values of the feedstock quantity adjustment instruction for the target abnormal partition and the remaining partitions are calculated;

[0045] The current change prediction values of each partition are aggregated to obtain a simulation prediction result of the change in the full-cell current distribution.

[0046] As can be seen from the above technical solutions, the present application has the following advantages:

[0047] The application is executed by a feeding control circuit, first, the cell voltage signal of an electrolytic cell and the area current signals of multiple partitions are collected, and a working condition feature vector representing the cell condition is constructed; a feature decoupling analysis model is adaptively selected according to the vector to process the signals to separate independent feature components corresponding to different physical abnormal sources; based on these components, the abnormal state and its position are diagnosed, and a state evaluation signal containing the abnormal type, positioning and risk level is generated; combined with a dynamic coupling graph reflecting the coupling relationship of the partitions, a simulation prediction of the full-cell current distribution for the basic control action against the abnormality is performed, and an instruction package containing the main control instruction for the target partition and the collaborative compensation instruction for balancing the global current is generated; finally, the instruction package is executed to drive the feeding mechanism of the target partition and the associated partitions to perform point adjustment. Through multi-source signal collaborative analysis and decoupling, the application realizes the abnormal identification of the local state inside the electrolytic cell; based on the simulation prediction and collaborative instruction generation of the dynamic coupling graph, the local fault is effectively suppressed and the current distribution is optimized, finally the current efficiency is significantly improved and the unit energy consumption is reduced. BRIEF DESCRIPTION OF DRAWINGS

[0048] Figure 1 An embodiment flowchart of the application is a feeding control circuit optimization method for an electrolytic cell based on the cooperative measurement of cell voltage and area current.

[0049] Figure 2 A flowchart of constructing a working condition feature vector in the application.

[0050] Figure 3 A flowchart of processing signals by a target decoupling model in the application.

[0051] Figure 4 A flowchart of constructing and updating a dynamic coupling graph in the application.

[0052] Figure 5 A flowchart of simulating and predicting the basic control action in the application. DETAILED DESCRIPTION

[0053] The terms "first", "second", "third", "fourth" and the like in the description of this application and in the claims hereof, if any, are used for distinguishing between similar elements and not necessarily for describing a particular sequential or chronological order. It is to be understood that the use of the terms so termed herein is interchangeable under appropriate circumstances such that the embodiments of the application described herein are, for example, capable of efficient implementation irrespective of the order of employment of particular steps or units. Moreover, the terms "comprises", "comprising", "corresponds" and "corresponding", as well as any variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, system, article, or apparatus that comprises a list of steps or units does not necessarily comprise only those steps or units

[0054] Embodiment one

[0055] The method implemented in this embodiment can be implemented in a system, which can be implemented in a server or in a terminal, and the specific implementation is not limited. The method in this application will be introduced from the perspective of system implementation. Please refer to Figure 1 The method provided by the embodiment of this application comprises the following steps:

[0056] S1. Collecting a cell voltage signal of the electrolytic cell and a plurality of regional current signals of a plurality of preset partitions;

[0057] The cell voltage signal refers to the direct current voltage between the anode bus and the cathode bus of the electrolytic cell, the value of which is in the range of 4V to 5V, but can be temporarily increased to tens of volts when abnormal conditions such as anode effect occur. This signal reflects the working voltage and energy input state of the entire electrolytic cell. The regional current signal is obtained by measuring the current on a plurality of anode rods in each preset partition of the electrolytic cell respectively, and then summing the current values of the plurality of rods in the same partition in real time. The plurality of preset partitions are pre-defined according to the physical structure and the distribution of the discharge port of the electrolytic cell. For example, for a 400kA electrolytic cell, the 48 anodes on the A and B sides along the length direction can be divided into 6 regions, each region corresponds to a discharge port and contains 8 anodes, thereby forming 6 independent regional current monitoring units. The collected signal data includes the instantaneous value of the cell voltage with a time stamp, the instantaneous value of each regional current and the corresponding regional identifier. In one specific embodiment of the present application, the collection frequency of the cell voltage signal and the regional current signal is set to 1Hz to ensure that typical changes in the operating state of the electrolytic cell can be captured.

[0058] S2. Based on the cell voltage signal and the regional current signal, a working condition feature vector representing the comprehensive operating state of the electrolytic cell is constructed;

[0059] This feature vector, representing a feature set, is used to quantitatively characterize the overall operating state of the electrolyzer at a specific moment, providing data for subsequent adaptive selection of the diagnostic model. Please refer to [link / reference]. Figure 2 Its construction process mainly includes the following three sub-steps:

[0060] S21. Calculate the regional current distribution unevenness, which characterizes the uniformity of the spatial distribution of current, and the regional current change coordination, which characterizes the temporal relationship of the current change in each region, based on the current signals of each region.

[0061] The regional current distribution unevenness is used to quantify the degree of unevenness in the spatial distribution of current in different zones of the electrolytic cell. One specific calculation method is to take the average current of each zone within a preset time window, and calculate the ratio of the standard deviation of these averages to the overall mean. This ratio serves as the unevenness index; the larger the value, the more uneven the current distribution. The regional current change synergy is used to quantify the synchronous or asynchronous relationship of current changes in different zones over time. One specific calculation method is to select the current time series of any two zones and calculate their Pearson correlation coefficient within a short time window. The mean or median of the correlation coefficients calculated for pairwise combinations of all zones is then used as the synergy index. The closer this value is to 1, the more consistent the current change trend in each zone. If this value decreases significantly or becomes negative, it suggests the possible existence of local disturbances.

[0062] S22. Calculate the slot voltage stability index, which characterizes voltage fluctuation, based on the slot voltage signal;

[0063] The tank voltage stability index is used to quantify the stability of tank voltage. One specific calculation method is to calculate the standard deviation of the tank voltage sequence within a preset time window. To eliminate the influence of differences in reference voltages of different tanks, the standard deviation can be divided by the average tank voltage within the time window to obtain the relative stability index. The smaller the value, the more stable the tank voltage.

[0064] S23. The operating condition characteristic vector is composed of the regional current distribution imbalance, the regional current change coordination, and the tank voltage stability index.

[0065] The calculated regional current distribution imbalance (denoted as F1), regional current variation coordination (denoted as F2), and cell voltage stability index (denoted as F3) are combined in a predetermined order to form a three-dimensional vector [F1, F2, F3]. This vector is the operating condition characteristic vector representing the overall operating status of the electrolytic cell at the current moment. In practical applications, this characteristic vector can be periodically calculated and updated by the microprocessor module in the feeding control circuit according to a preset algorithm.

[0066] S3. Select and call the corresponding feature decoupling analysis model based on the working condition feature vector, and use the feature decoupling analysis model to process the tank voltage signal and the area current signal to separate the independent feature components that characterize different physical anomaly sources.

[0067] The feature decoupling analysis model is a deep learning-based signal processing model that separates the synchronously acquired, coupled cell voltage and multi-channel region current mixed signals into multiple more independent signal components, each associated with a specific physical anomaly source. These separated signal components are the independent feature components characterizing different physical anomaly sources. The abstract features extracted from the raw data can more directly and clearly reflect different potential fault or disturbance modes inside the electrolytic cell.

[0068] Specifically, the operating condition feature vector calculated in step S2 is input into a pre-set model matching decision-maker. This model matching decision-maker has embedded explicit matching logic, which matches the target decoupling model from a pre-stored feature decoupling analysis model library based on the numerical combination of each index in the operating condition feature vector. The feature decoupling analysis model library contains multiple pre-trained decoupling network models with different network depths, convolutional kernel sizes, or structures, trained offline. For example, there are deep, wide convolutional kernel models optimized for operating conditions with extremely uneven current distribution and low coordination; models optimized for operating conditions with severe voltage fluctuations and incorporating attention mechanisms; and lightweight, fast models suitable for stable operating conditions. In practice, the matching logic can be based on preset rule thresholds: when the unevenness of regional current distribution is greater than the first threshold and the coordination of regional current changes is less than the second threshold, a deep, wide convolutional kernel model is selected; when the tank voltage stability index is greater than the third threshold, a model incorporating attention mechanisms is selected; if none of the above conditions are met, the default lightweight, fast model is selected. The first, second, and third thresholds can be calibrated and adjusted based on historical data statistics or process experience.

[0069] After selecting the target decoupling model, begin in-depth processing of the original signal. Please refer to [link to relevant documentation]. Figure 3 Specifically, it includes the following sub-steps:

[0070] S31. Combine the timing sequence of the slot voltage signal with the timing sequence of the current signal in each region to form a multi-channel input timing matrix;

[0071] Assuming one slot voltage channel is collected and The current channels in each region were continuously sampled at a frequency of 1Hz. Seconds. For each channel Each consecutive data point is treated as a time series. These sequences are stacked along the channel dimension to form a shape... A two-dimensional matrix, where rows represent time points and columns represent different signal channels. This matrix is ​​the multi-channel input timing matrix, which contains both the temporal evolution pattern and spatial (regional) distribution information of the signal.

[0072] S32. Input the multi-channel input time series matrix into the target decoupling model;

[0073] S33. By using the convolutional layers contained in the target decoupling model, which perform feature extraction and transformation in both spatial and temporal dimensions, the multi-channel input time series matrix is ​​processed to obtain multiple independent intermediate feature vectors.

[0074] The core of the target decoupling model is one or more one-dimensional convolutional layers. The convolutional kernels slide along the time dimension, capturing short-term temporal patterns of the signal. Simultaneously, since each column of the input matrix corresponds to a physical spatial location (a signal source), the convolutional operation performs weighted combinations across channels (spatial dimension), enabling the discovery of correlations and differences between signals in different regions. Through the nonlinear transformation of multiple convolutional layers, the model ultimately outputs a fixed set of intermediate feature vectors, each vector initially encoding a potential pattern in a specific aspect of the original signal.

[0075] S34. Through the fully connected classification layer of the target decoupling model, multiple independent intermediate feature vectors are mapped into independent feature components that characterize different physical anomaly sources.

[0076] The intermediate feature vector obtained in the previous step is input into the fully connected classification layer. This layer learns to map the intermediate features to preset feature component dimensions corresponding to specific physical anomaly sources. In this embodiment, the preset physical anomaly sources mainly include four types: global reaction intensity disturbance sources that cause overall fluctuations in the overall tank reaction rate, regional concentration disturbance sources that lead to excessively high or low local alumina concentrations, anode effect risk sources that indicate an impending anode effect, and cathode state anomaly sources that reflect cathode damage or abnormal current leakage. Therefore, the fully connected layer outputs four independent feature components accordingly: a global feature component characterizing the deviation of the overall reaction rate, a concentration disturbance feature component characterizing local alumina concentration anomalies, an effect warning feature component characterizing the risk of anode effect occurrence, and a cathode anomaly feature component characterizing cathode damage or current leakage. Thus, the original mixed voltage and current signal is decoupled into multiple feature components that are clearly oriented and can be analyzed independently.

[0077] S4. Diagnose the abnormal operating status of the electrolytic cell and the zone where the abnormal operating status is located by using independent feature components, and generate a status assessment signal containing the abnormality type, location and risk level;

[0078] This step, by comparing and analyzing the values ​​of each feature component with preset thresholds, transforms the independent feature components obtained from the decoupling in step S3 into actionable diagnostic conclusions. Specifically:

[0079] 1. Compare the values ​​of the global feature components with the first global threshold; if the values ​​of the global feature components exceed the first global threshold, it is determined that there is an abnormal global reaction intensity disturbance in the electrolytic cell;

[0080] 2. For each of the concentration disturbance characteristic component, effect warning characteristic component, and cathode anomaly characteristic component, perform the following operations:

[0081] The values ​​of the sub-features corresponding to the feature component in each partition are compared with the corresponding thresholds; if the value of a sub-feature component in a certain partition exceeds the threshold, it is determined that there is an anomaly corresponding to the feature component in that partition.

[0082] Among them, the zones that are found to have abnormal concentration disturbances, abnormal anode effect risk, or abnormal cathode condition are identified as the zones where abnormal operating conditions exist.

[0083] Specifically, the first global threshold is the upper limit of the global characteristic component distribution obtained based on a large amount of historical normal operating data. If the value of the global characteristic component exceeds this threshold, it is determined that there is an abnormal global reaction intensity disturbance in the electrolyzer, indicating that the overall reaction rate of the entire electrolyzer has deviated significantly. For the concentration disturbance characteristic component, effect warning characteristic component, and cathode abnormal characteristic component, the following operations are performed respectively, with the core logic being the same but the specific thresholds and diagnostic objects being different:

[0084] Concentration disturbance diagnosis: The values ​​of the sub-features corresponding to the concentration disturbance feature components in each partition are compared with a preset concentration disturbance threshold. This preset concentration disturbance threshold is set based on the allowable range of local concentration fluctuations during normal production. If the value of the concentration disturbance feature component is greater than the preset concentration disturbance threshold, it is determined that there is an abnormal concentration disturbance in that partition.

[0085] Anode effect risk diagnosis: The values ​​of the sub-features corresponding to the effect warning feature components in each partition are compared with a preset effect warning threshold. This preset effect warning threshold is set based on the critical values ​​of current and voltage characteristics before the occurrence of the anode effect, and is lower than the characteristic values ​​when the effect is fully generated; if the value of the effect warning feature component is greater than the preset effect warning threshold, it is determined that there is an anode effect risk anomaly in that partition.

[0086] Cathode condition diagnosis: The values ​​of the sub-features corresponding to the cathode abnormality feature components in each zone are compared with a preset cathode abnormality threshold. This preset cathode abnormality threshold is a safety boundary value set based on the symmetry of current distribution and leakage current level under intact cathode conditions; if the value of the cathode abnormality feature component is greater than the preset cathode abnormality threshold, it is determined that there is a cathode condition abnormality in that zone.

[0087] All zones that are determined to have abnormal concentration disturbances, abnormal anode effect risk, or abnormal cathode condition are identified as the specific zones where abnormal operating conditions occur during this diagnostic cycle.

[0088] After identifying the anomaly, its severity needs to be further assessed. Specifically:

[0089] 1. If it is determined that there is an anomaly in the global response intensity perturbation, the risk level corresponding to the anomaly in the global response intensity perturbation is determined according to the magnitude by which the value of the global feature component exceeds the first global threshold.

[0090] 2. For each partition that is determined to have abnormal concentration disturbance, abnormal anode effect risk, or abnormal cathode state, the risk level corresponding to the abnormality in that partition is determined based on the extent to which the value of the sub-feature component corresponding to that partition exceeds the threshold.

[0091] 3. Combine each anomaly type, its corresponding partition location, and the risk level corresponding to each anomaly type to generate a status assessment signal.

[0092] Finally, the system integrates all the diagnostic and assessment information in a structured manner to generate the final status assessment signal. This signal is a machine-readable, fixed-format data object, including a diagnostic timestamp accurate to milliseconds to mark the effective time of this assessment; and a detailed list of anomalies, which includes all anomaly instances identified during this diagnostic cycle. For each item in the list, three key pieces of information are clearly recorded: first, the anomaly type, indicating whether it is a global response intensity perturbation, a local concentration perturbation, an anodic effect risk, or a cathodic state anomaly; second, the zonal location, recording the zonal number for zonal-level anomalies and marking them accordingly for global anomalies; and third, the risk level, clearly marked as low, medium, high, or other predefined levels based on the exceedance magnitude calculated in the preceding steps.

[0093] S5. Based on the state assessment signal and the dynamic coupling spectrum reflecting the coupling relationship of each zone of the electrolyzer, the simulation prediction of the change in the current distribution of the whole cell is performed on the basic control actions for abnormal operating conditions.

[0094] The dynamic coupling graph here is used to quantify the network model describing the electrical and material interactions between different physical zones within the electrolyzer, demonstrating how changes in the state of one zone will be transmitted through physical and chemical processes and affect other zones. This graph complements the feature decoupling analysis model in step S3. The feature decoupling model separates different anomaly sources from mixed signals, while the dynamic coupling graph is used to predict new changes in the overall current distribution of the electrolyzer that may be triggered by an actively applied control action. Simulation prediction refers to simulating and calculating a anticipated control action based on this graph, and deducing its impact on the overall current distribution of the electrolyzer in the near future. The principle is that the thermal balance, material balance, and current field within the electrolyzer are closely coupled.

[0095] Please see Figure 4 The dynamic coupling map is constructed and updated in the following ways:

[0096] S511. Based on the time-series data of the regional current signals of each partition of the electrolytic cell, calculate the correlation coefficient of current change transmission between any two partitions as the coupling weight;

[0097] S512. Establish a directed graph structure with partitions as nodes and coupling weights as edges to form a dynamic coupling graph.

[0098] S513. Recalculate the current change transmission correlation coefficient according to the preset update cycle to update the coupling weight in the dynamic coupling spectrum.

[0099] Specifically, based on the time-series data of the regional current signals of each zone of the electrolytic cell within a certain time window, the correlation coefficient of current change propagation between any two zones is calculated. The calculation method for this coefficient includes determining the time delay: calculating the zone... With partitions Given a current change rate sequence, within a pre-defined reasonable time delay search range, calculate the partition. The current change rate sequence and partitions under different time delays The correlation coefficients between the current rate of change sequences are used, and the value with the largest absolute value among the correlation coefficients calculated under all time delays is selected as the final coupling weight. The corresponding time delay is the partition. For partitions The influence propagation time is determined by the weight, with the magnitude of the weight representing the influence intensity, and positive or negative indicating a promoting or inhibiting relationship. Secondly, a directed graph structure is established, with each partition as a node and the calculated coupling weights as directed edges. This graph structure constitutes the initial or updated dynamic coupling graph, reflecting the topology of the influence network within the cell. Finally, to ensure the graph reflects the latest changes in the electrolyzer's operating conditions, the system re-collects the latest current time-series data according to a preset update cycle, repeats the above correlation coefficient calculation, and updates the weights of each edge in the graph, thereby achieving dynamic updates to the graph and ensuring it always tracks the latest state of coupling relationships within the cell.

[0100] In this embodiment, the basic control actions are determined in the following way:

[0101] 1. Based on the anomaly type in the status assessment signal, match the corresponding standard control action template from the preset anomaly type and control action mapping table;

[0102] 2. Based on the risk level in the status assessment signal, the control parameters in the standard control action template are quantitatively adjusted to generate basic control actions.

[0103] The basic control actions are determined as follows: First, the system queries a preset mapping table of anomaly types and control actions based on the anomaly type in the state assessment signal. This mapping table is built based on domain knowledge; for example, a concentration disturbance anomaly corresponds to the action template of increasing the feed rate in the target zone. The system then quantifies and adjusts the control parameters in the matched standard control action template based on the risk level in the state assessment signal. For example, the template action of increasing the feed rate, set to a medium risk level, is specifically quantified as increasing the feed rate of the target zone by 50% over the next two minutes.

[0104] Please see Figure 5 Simulation prediction of full-slot current distribution changes for basic control actions under abnormal operating conditions, including:

[0105] S521. Quantify the basic control actions into feeding quantity adjustment commands for the target abnormal zone;

[0106] S522. Based on the coupling weights between nodes in the dynamic coupling graph, calculate the predicted current changes generated by the material feeding adjustment command on the target abnormal zone and other zones.

[0107] S523. Aggregate the predicted current changes of each zone to obtain the simulation prediction results of the current distribution change of the entire tank.

[0108] Specifically, the basic control action determined in the previous step is quantified into a specific feed rate adjustment command. This command can be represented as an input vector acting on a specific partition and having amplitude and duration dimensions. This adjustment command is used as a disturbance applied to the corresponding node (partition A) in the dynamic coupling graph. Based on the directed coupling weights between nodes in the dynamic coupling graph, the propagation of this disturbance in the influence network is simulated. For example, the direct impact of this disturbance on the current of the target partition A itself is calculated. Then, based on the coupling weights from A to other partitions, the predicted indirect current changes to the remaining partitions are calculated level by level. Finally, the calculated predicted current changes of all partitions, including the target partition and the remaining partitions, are aggregated to form a vector describing the predicted current values ​​of all partitions in the entire tank at a future moment. This vector is the simulation prediction result of the overall tank current distribution change.

[0109] S6. Generate a cooperative control instruction package based on the simulation prediction results. The cooperative control instruction package includes a main control instruction for the target abnormal partition and at least one cooperative compensation instruction for balancing the global current distribution.

[0110] Based on the simulation prediction results obtained in step S5, the system generates a final collaborative control instruction package comprising two parts: first, a main control instruction directly targeting the diagnosed abnormal zone, the content of which is consistent with the basic control action determined in S5, used to correct the abnormal state of that region; and second, at least one collaborative compensation instruction, the generation logic of which is: analyzing the undesirable current offsets that may occur in other zones besides the target zone as shown in the simulation prediction results, i.e., secondary disturbances caused by the main control action through the dynamic coupling spectrum, and for these affected zones, calculating and generating corresponding feed rate adjustment instructions with calculated intensity and direction as compensation, so that the predicted current of these zones can return to or approach the normal range. Finally, the main control instruction and the collaborative compensation instruction are encapsulated into an instruction package with a unified execution timing, thereby ensuring that while correcting local anomalies, the overall stability and balance of the current distribution across the entire tank are actively maintained.

[0111] S7. Execute the collaborative control instruction package to drive the unloading actuators of the target partition and associated partitions to perform fixed-point adjustment operations.

[0112] Specifically, the feeding control circuit first parses the received collaborative control instruction packet, extracting the main control instruction and various collaborative compensation instructions. Each instruction contains a clear target zone identifier, feeding adjustment parameters, and optional execution timing information. Second, the circuit's multi-channel independent output control module converts the corresponding feeding adjustment parameters into drive signals based on the zone identifier in the instruction and sends them to the feeding actuators, such as pneumatic or electric valves installed at the feeding ports of the target zone and associated zones. Under the precise control of the drive signals, these actuators perform fixed-point and quantitative feeding actions according to the instruction requirements. For example, they perform enhanced feeding to replenish alumina in abnormal zones, and perform compensatory feeding to stabilize the material balance in zones affected by coupling, thereby completing an adjustment operation from intelligent decision-making to physical execution.

[0113] The above embodiments enable early and accurate identification and spatial positioning of local abnormal states inside the electrolytic cell, and can generate systematic collaborative instructions for proactive intervention; the feeding control is transformed from a rough mode based on global average to a precise collaborative mode based on local state perception and global impact optimization, thereby significantly improving current efficiency, reducing energy consumption, and enhancing the overall stability and adaptability of the electrolytic cell operation.

[0114] It is understood that those skilled in the art can combine various implementation methods in the above embodiments under the guidance of the above examples to obtain technical solutions with multiple implementation methods.

[0115] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for optimizing a circuit for controlling the feed of an electrolytic cell based on a combined measurement of the cell voltage and the area current, characterized in that, The method is executed by a feeding control circuit, and comprises the following steps: Collecting a cell voltage signal of an electrolytic cell and a plurality of preset partition area current signals; Based on the cell voltage signal and each area current signal, a working condition feature vector representing the comprehensive running state of the electrolytic cell is constructed; According to the working condition feature vector, a corresponding feature decoupling analysis model is selected and called, and the feature decoupling analysis model is used to process the cell voltage signal and the area current signal to separate out independent feature components representing different physical abnormal sources; Through the independent feature components, the abnormal running state of the electrolytic cell and the partition area where the abnormal running state is located are diagnosed, and a state evaluation signal containing the abnormal type, positioning and risk level is generated; Based on the state evaluation signal and a dynamic coupling graph reflecting the coupling relationship of each partition area of the electrolytic cell, a simulation prediction of the change of the whole cell current distribution for the basic control action of the abnormal running state is performed; Based on the simulation prediction result, a cooperative control instruction package is generated, which includes a main control instruction for the target abnormal partition area and at least one cooperative compensation instruction for balancing the global current distribution; The cooperative control instruction package is executed to drive the feeding execution mechanism of the target partition area and the associated partition area to perform a point adjustment operation.

2. The electrolytic cell underflow control circuit optimization method based on the synergistic measurement of cell voltage and area current of claim 1, wherein, The method comprises the following steps: Based on each area current signal, an area current distribution imbalance degree representing the uniformity of the current spatial distribution and an area current change coordination degree representing the time sequence relationship of the current change of each partition area are calculated; Based on the cell voltage signal, a cell voltage stability index representing the voltage fluctuation characteristics is calculated; The area current distribution imbalance degree, the area current change coordination degree and the cell voltage stability index constitute a working condition feature vector.

3. The electrolytic cell underflow control circuit optimization method based on the synergistic measurement of cell voltage and area current of claim 2, wherein, The method comprises the following steps: The working condition feature vector is input into a preset model matching decision maker; the model matching decision maker matches a target decoupling model from a pre-stored feature decoupling analysis model library according to the numerical combination of each index in the working condition feature vector; wherein the feature decoupling analysis model library stores a plurality of pre-trained decoupling network models with different structures.

4. The electrolytic cell underflow control circuit optimization method based on the synergistic measurement of cell voltage and area current of claim 3, wherein, The method comprises the following steps: The time sequence sequence of the cell voltage signal and the time sequence sequence of each area current signal are combined to form a multi-channel input time sequence matrix; The multi-channel input time sequence matrix is input into the target decoupling model; Through the convolution layer included in the target decoupling model and performing feature extraction and transformation in the spatial dimension and the time dimension, the multi-channel input time sequence matrix is processed to obtain a plurality of independent intermediate feature vectors; Through the full-connection classification layer of the target decoupling model, the plurality of independent intermediate feature vectors are mapped into independent feature components representing different physical abnormal sources.

5. The electrolytic cell underflow control circuit optimization method based on the synergistic measurement of cell voltage and area current of claim 4, wherein, The different physical abnormal sources include: a global reaction intensity disturbance source, a regional concentration disturbance source, an anode effect risk source, and a cathode state abnormal source; The independent feature components characterizing the different physical abnormal sources include: a global feature component characterizing the overall reaction rate deviation, a concentration disturbance feature component characterizing the local alumina concentration abnormality, an effect early warning feature component characterizing the anode effect occurrence risk, and a cathode abnormal feature component characterizing the cathode damage or current leakage.

6. The electrolytic cell underflow control circuit optimization method based on the synergistic measurement of cell voltage and area current of claim 5, wherein, The diagnosis of the abnormal running state of the electrolytic cell and the partition where the abnormal running state is located by the independent feature components includes: Comparing the value of the global feature component with a first global threshold value; if the value of the global feature component exceeds the first global threshold value, it is determined that the electrolytic cell has a global reaction intensity disturbance abnormality; For each of the concentration disturbance feature component, the effect early warning feature component, and the cathode abnormal feature component, the following operations are performed: Compare the value of the feature component in each partition corresponding to the sub-feature component with the corresponding threshold value; if the value of the sub-feature component of a certain partition exceeds the threshold value, it is determined that the partition has an abnormality corresponding to the feature component; Wherein, the partition where the concentration disturbance abnormality, the anode effect risk abnormality, or the cathode state abnormality is determined to be the partition where the abnormal running state is located.

7. The electrolytic cell underflow control circuit optimization method based on the synergistic measurement of cell voltage and area current of claim 6, wherein, The generation of the state evaluation signal containing the abnormal type, the location, and the risk level includes: If it is determined that there is a global reaction intensity disturbance abnormality, the risk level corresponding to the global reaction intensity disturbance abnormality is determined according to the amplitude by which the value of the global feature component exceeds the first global threshold value; For each partition where it is determined that there is a concentration disturbance abnormality, an anode effect risk abnormality, or a cathode state abnormality, the risk level corresponding to the abnormality existing in the partition is determined according to the amplitude by which the value of the sub-feature component corresponding to the partition exceeds the threshold value; The abnormal type, the corresponding partition location, and the risk level corresponding to each abnormal type are combined to generate a state evaluation signal.

8. The electrolytic cell underflow control circuit optimization method based on the synergistic measurement of cell voltage and area current of claim 1, wherein, The dynamic coupling graph is constructed and updated in the following way: Based on the time series data of the regional current signals of each partition of the electrolytic cell, the current change transmission correlation coefficient between any two partitions is calculated as the coupling weight; A directed graph structure is established with the partitions as nodes and the coupling weights as edges, forming a dynamic coupling graph; According to a preset update period, the current change transmission correlation coefficient is recalculated to update the coupling weights in the dynamic coupling graph.

9. The electrolytic cell underflow control circuit optimization method based on the synergistic measurement of cell voltage and area current of claim 1, wherein, The basic control action is determined in the following way: According to the abnormal type in the state evaluation signal, the corresponding standard control action template is matched from the pre-set abnormal type and control action mapping table; According to the risk level in the state evaluation signal, the control parameters in the standard control action template are quantitatively adjusted to generate a basic control action.

10. The electrolytic cell underflow control circuit optimization method based on the synergistic measurement of cell voltage and area current of claim 9, wherein, The simulation prediction of the change of the full-cell current distribution of the basic control action for the abnormal running state includes: The basic control action is quantified as a feeding amount adjustment instruction for the target abnormal partition; Based on the coupling weights between the nodes in the dynamic coupling graph, a current change prediction value of the target abnormal partition and the remaining partitions caused by the strip length adjustment instruction is calculated; The current change prediction values of the partitions are aggregated to obtain a simulation prediction result of the full-slot current distribution change.

Images