Carbon dioxide electrolysis system based on pulse voltage regulation

Abstract

This system discloses a carbon dioxide electrolysis system based on pulse voltage regulation, including a pulse voltage source module, an electrolysis reaction module, a parameter control module, a timing optimization module, a benchmark setting module, and a decision execution module. The pulse voltage source module generates a controllable pulse voltage sequence and sets parameter intervals; the electrolysis reaction module divides the reaction units and analyzes the current response to generate characteristic vectors; the parameter control module extracts key indicators and establishes voltage regulation rules; the timing optimization module identifies time-correlated patterns, dynamically calibrates indicators, and calculates voltage offsets; the benchmark setting module derives the optimal pulse threshold and generates a deviation sequence; and the decision execution module analyzes the deviation sequence to generate a control scheme. Through the collaborative efforts of multiple modules, this system achieves dynamic optimization of pulse voltage and refined management of reaction units, improving the selectivity, stability, and energy efficiency of carbon dioxide electrolysis and is suitable for the field of efficient carbon dioxide conversion.

Description

Technical Field

[0001] The present invention relates to the technical field of carbon dioxide electrolysis, and in particular to a carbon dioxide electrolysis system based on pulse voltage regulation. Background Art

[0002] Electrolysis, which can convert carbon dioxide into high-value chemicals or fuels (such as carbon monoxide, formic acid, and hydrocarbons), has become an important approach to carbon recycling. However, conventional carbon dioxide electrolysis technology faces numerous challenges in practical application, limiting its large-scale promotion and efficient application.

[0003] From a reaction mechanism perspective, carbon dioxide has a stable molecular structure, and the electrolysis process requires overcoming a high activation energy, resulting in slow reaction kinetics. Conventional constant-voltage electrolysis modes struggle to precisely control the reaction pathway and are prone to inducing side reactions, reducing the selectivity and Faraday efficiency of the target product. For example, under different voltage ranges, carbon dioxide may undergo multiple reduction reactions, generating different products. However, the constant voltage cannot be dynamically adjusted according to the reaction progress, resulting in complex product distribution and difficulty in separation and purification.

[0004] In terms of system stability, traditional electrolysis systems are slow to respond to parameter fluctuations during the reaction process. When key parameters such as current, temperature, and pressure change, the system fails to adjust in a timely manner, potentially leading to process instability and even safety hazards. For example, sudden current fluctuations can cause a sharp increase in electrode surface temperature, accelerating electrode corrosion and shortening the equipment's service life.

[0005] From the perspective of energy efficiency, under constant voltage mode, the energy input of the electrolysis process cannot be matched in real time to the reaction requirements, resulting in energy waste. In the early stages of the reaction, higher energy may be required to activate the carbon dioxide molecules, while in the later stages, lower energy can maintain the reaction. However, constant voltage cannot achieve this dynamic adjustment, resulting in low overall energy efficiency.

[0006] Traditional electrolysis systems lack the ability to fine-tune the analysis and control of reaction processes. They are unable to obtain detailed characteristic information of reaction units in real time, making it difficult to establish accurate reaction models and, consequently, to achieve optimized control of the electrolysis process. For example, different reaction stages may require different voltage waveforms and parameter settings, but traditional systems cannot tailor control to the characteristics of the reaction units.

[0007] In the prior art, although there are some studies on voltage regulation, most of them use simple pulse voltage or segmented voltage control, lacking dynamic optimization of pulse parameters and refined management of reaction units. For example, some studies only use pulse voltages with fixed frequency and amplitude to improve the reaction effect, but do not consider the differences between different reaction units, as well as the real-time correlation between pulse parameters and reaction characteristics. Therefore, there is an urgent need for a carbon dioxide electrolysis system that can realize dynamic regulation of pulse voltage, refined management of reaction units, and real-time optimization of parameters according to the reaction process, so as to improve the selectivity, stability and energy utilization efficiency of the reaction and promote the practical application of carbon dioxide electrolysis technology. Summary of the Invention

[0008] The object of the present invention is to provide a carbon dioxide electrolysis system based on pulse voltage regulation to solve the problems raised in the above background technology.

[0009] To achieve the above objectives, the present invention provides the following technical solution: a carbon dioxide electrolysis system based on pulse voltage regulation, the system comprising:

[0010] The pulse voltage source module is used to generate an adjustable pulse voltage sequence and set the pulse parameter range corresponding to the electrolysis reaction;

[0011] The electrolysis reaction module is used to divide multiple reaction units within the pulse parameter range, analyze the current response of the electrolysis process of each reaction unit, and generate the electrolysis characteristic vector corresponding to the reaction unit;

[0012] The parameter control module is used to extract key reaction indicators from the electrolysis characteristic vector, establish voltage regulation rules associated with the reaction unit, and obtain pulse correction parameters corresponding to the rules;

[0013] The timing optimization module is used to identify the time correlation pattern in the pulse correction parameters, dynamically calibrate the key reaction indicators based on the correlation pattern, and calculate the voltage offset of each reaction unit under different calibration strategies;

[0014] A benchmark setting module is used to derive the optimal pulse threshold according to the voltage offset and generate an electrolysis deviation sequence by comparing the current electrolysis characteristic value with the optimal pulse threshold;

[0015] The decision execution module is used to analyze the electrolysis deviation sequence and integrate the electrolysis deviation sequence into an electrolysis control execution plan based on the voltage offset trend of the reaction unit.

[0016] Preferably, the electrolysis reaction module is implemented by: constructing an electrolysis feature library corresponding to the reaction unit, the electrolysis feature library including current data and a reaction parameter vector of an electrolysis response mapping;

[0017] Similar response matching is performed on the reaction parameter vector, and the reaction parameter vector is divided into reaction cluster groups according to the matching results; the electrolytic focus point of the current data is extracted from the reaction cluster group, and the focus point is set as a reaction unit.

[0018] Preferably, the reaction clustering groups for dividing the reaction parameter vectors further include:

[0019] According to the time attributes and current intensity in the reaction parameter vector, the voltage distribution, reaction polarity and response density parameters are extracted, and the reaction feature label is generated based on the above parameters;

[0020] The reaction feature labels are associated with the reaction parameter vectors. By calculating the response similarity between the feature labels, the parameter vectors with similarity higher than the preset response threshold are screened to form a reaction cluster group.

[0021] Preferably, the implementation method of generating the electrolysis characteristic vector corresponding to the reaction unit includes:

[0022] For each reaction unit, according to the unit's temporal position in the pulse parameter interval, the unit's current fluctuation data within a preset window is obtained, and the unit's current fluctuation coefficient is calculated;

[0023] When the current fluctuation coefficient exceeds the first response threshold, the unit is marked as a high-response unit, and its current data is extracted to form an electrolysis feature vector; when the current fluctuation coefficient is lower than the first response threshold, the unit is marked as a steady-state unit, and the current data of the unit's adjacent units are superimposed on the response, and the superimposed data are reconstructed into an electrolysis feature vector.

[0024] Preferably, the parameter control module is implemented as follows:

[0025] Separating the positive reaction ratio, negative reaction ratio, and neutral response parameters from the electrolysis characteristic vector, and generating a voltage regulation rule for the reaction unit based on the above parameters;

[0026] If the number of reaction units covered by the current voltage regulation rule is less than the preset response threshold, the electrolysis feature vectors of adjacent reaction units are traversed, and the response indicators not included in the regulation rules of the adjacent units are added to the current rule.

[0027] Preferably, the implementation of the timing optimization module includes: obtaining a sequence factor of the time switching frequency and a change factor of the response jump amplitude in the time correlation mode;

[0028] A weight distribution matrix associated with the sequence factor and the variation factor is constructed, and the voltage offset under different calibration strategies is determined according to the distribution probability of each element in the matrix.

[0029] Preferably, constructing the weight distribution matrix further includes:

[0030] Identify the time period characteristics of the sequence factor. If the current period characteristics completely match the preset time period, set the sequence factor as the starting index of the weight allocation matrix.

[0031] Calculate the correlation matching degree between the sequence factor and the change factor, and generate the intermediate index and the end index of the weight distribution matrix in descending order of matching degree;

[0032] The path of the termination index is backtracked, and when the matching degree of the termination index is lower than the preset matching threshold, it is output as the final distribution of the weight allocation matrix.

[0033] Preferably, the implementation of calculating the voltage offset includes:

[0034] Statistically calculate the mean and range of the sequence factor of each termination index in the weight distribution matrix, and calculate the global variance of all index factors;

[0035] The timing offset coefficient is obtained by subtracting the mean of the sequence factor of a single termination index from the mean of the sequence factors of the adjacent indexes and dividing it by the global variance. At the same time, the ratio of the variation factor range to the global variance is calculated, and the weighted sum of the two is taken as the voltage offset of the index.

[0036] Preferably, the implementation of deriving the optimal pulse threshold includes:

[0037] Extract the reaction pattern closest to the current voltage offset in the historical data, and calculate the cosine similarity of the two in time distribution as the first dynamic reference value;

[0038] Counting the difference in the number of response peaks between the current voltage offset and the historical response pattern, and using the difference as a second dynamic reference value;

[0039] Based on the linear combination of the first dynamic reference value and the second dynamic reference value, the optimal pulse threshold in the preset pulse threshold table is matched.

[0040] Preferably, the implementation of the decision execution module includes: dividing the voltage offset interval into a positive offset interval and a negative offset interval according to the voltage offset direction of each unit in the electrolysis deviation sequence;

[0041] The convergence rate of the electrolysis deviation in the positive offset interval and the diffusion rate of the electrolysis deviation in the negative offset interval are extracted, and the two are antagonistically fused according to the time weight of the reaction unit to generate the adjustment parameters of the electrolysis control execution plan.

[0042] Compared with the prior art, the present invention has the following beneficial effects:

[0043] In terms of refined management of the reaction process, the electrolysis reaction module can divide multiple reaction units within the pulse parameter range, and analyze the current response of the electrolysis process of each unit to generate the corresponding electrolysis feature vector. This process achieves in-depth analysis and classification of the reaction parameter vector by constructing an electrolysis feature library, performing similar response matching and reaction clustering group division. For example, reaction feature labels are generated based on parameters such as time attributes and current intensity, and cluster groups are formed through response similarity screening, allowing the system to clearly identify the characteristics and patterns of different reaction units. This refined management method changes the drawbacks of the traditional system's general treatment of the reaction process, and can perform targeted regulation based on the characteristics of each reaction unit, providing a solid foundation for subsequent parameter optimization and voltage adjustment.

[0044] The ability to dynamically optimize pulse parameters is a highlight of this system. The parameter control module extracts key reaction indicators from the electrolysis characteristic vector, establishes voltage adjustment rules associated with the reaction unit, and obtains pulse correction parameters. By separating the proportions of positive and negative reactions and neutral response parameters, the system can formulate personalized voltage adjustment rules based on the actual conditions of the reaction unit. When the number of reaction units covered by the current rules is insufficient, it can also automatically traverse the characteristic vectors of adjacent units, supplement the response indicators, and ensure the comprehensiveness and effectiveness of the adjustment rules. This dynamic optimization mechanism enables the pulse voltage to adapt to changes in the reaction process in real time, effectively improving the selectivity of the reaction, reducing the occurrence of side reactions, and thus improving the yield and purity of the target product.

[0045] Timing optimization and dynamic calibration functions further enhance the system's stability and adaptability. The timing optimization module constructs a weight allocation matrix by identifying the time correlation patterns in the pulse correction parameters, and determines the voltage offset based on the distribution probability of the matrix elements. This process fully considers the influence of the time switching frequency and the response jump amplitude, and can accurately capture the temporal dynamic characteristics of the reaction process. For example, by identifying the time period characteristics of the sequence factors, determining the index of the weight allocation matrix, and calculating the correlation matching degree, the system can dynamically calibrate key reaction indicators based on the time correlation pattern. This dynamic calibration mechanism can promptly respond to parameter fluctuations in the reaction process, adjust the voltage offset of each reaction unit, ensure the stability of the electrolysis process, and avoid the instability problem caused by the delayed parameter response of traditional systems.

[0046] Benchmark setting and deviation analysis provide a scientific basis for system regulation. The benchmark setting module derives the optimal pulse threshold based on the voltage offset and generates an electrolysis deviation sequence by comparing the current electrolysis characteristic value with the optimal threshold. This process combines historical data and current reaction information, and accurately determines the optimal pulse threshold by calculating dynamic reference values ​​such as cosine similarity and response peak difference. The generation of the electrolysis deviation sequence enables the system to clearly understand the gap between the current reaction state and the optimal state, providing a clear control direction for the decision-making execution module. This data- and model-based benchmark setting and deviation analysis method improves the scientific nature and accuracy of system regulation and avoids the blindness of traditional regulation methods.

[0047] The efficiency and pertinence of decision-making execution are key to achieving precise control of the system. The decision-making execution module divides the positive and negative offset intervals according to the electrolysis deviation sequence, extracts the convergence rate and diffusion rate, and generates the adjustment parameters through adversarial fusion based on the time weight. This processing method can carry out targeted control of reaction units with different offset directions and rates, ensuring the efficiency and effectiveness of the control scheme. For example, different control strategies are adopted for units with fast convergence rates in the positive offset interval and units with slow diffusion rates in the negative offset interval, allowing the system to quickly correct deviations and restore to the optimal reaction state. This efficient decision-making execution mechanism greatly improves the system's response speed and control accuracy, ensuring the efficient and stable operation of the electrolysis process.

[0048] Through the collaborative work of various modules, the system forms a complete closed loop from data acquisition and analysis to parameter optimization and control execution. This system enables dynamic management and precise control of the entire carbon dioxide electrolysis process, effectively addressing the problems of slow reaction kinetics, low selectivity, poor stability, and low energy efficiency in traditional electrolysis technology. It provides advanced technical support for the efficient conversion and utilization of carbon dioxide, with significant economic and environmental benefits, and is of great significance for promoting carbon recycling and addressing climate change. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 This is a working principle diagram of the carbon dioxide electrolysis system based on pulse voltage regulation according to the present invention;

[0050] Figure 2 Schematic diagram of the electrolysis reaction module;

[0051] Figure 3 Schematic diagram of the work for clustering reaction groups;

[0052] Figure 4 Schematic diagram of the work performed for electrolytic eigenvector generation. DETAILED DESCRIPTION

[0053] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0054] See also Figures 1-4 The present invention relates to a carbon dioxide electrolysis system based on pulse voltage control. The system includes a pulse voltage source module, an electrolysis reaction module, a parameter control module, a timing optimization module, a benchmark setting module, and a decision execution module. These modules work together to achieve dynamic control of the pulse voltage during the carbon dioxide electrolysis process. Specific implementations are as follows:

[0055] Pulse Voltage Source Module: This module is equipped with a programmable pulse signal generator to generate a controllable pulse voltage sequence. By combining hardware circuit parameters with software algorithms, the pulse parameter range corresponding to the electrolysis reaction is set. This range includes the voltage amplitude range, pulse frequency range, and duty cycle range, providing the basic excitation signal for the electrolysis reaction.

[0056] Electrolysis Reaction Module: Within the pulse parameter range, the continuous electrolysis process is divided into multiple discrete reaction units. For each reaction unit, a current sensor collects real-time current response data during the electrolysis process. Signal processing algorithms such as Fourier transform and wavelet analysis are used to analyze the current response characteristics. An electrolysis feature vector containing parameters such as current peak value, rise time, and energy integral is generated to characterize the reaction state of the unit.

[0057] Parameter Control Module: This module extracts key reaction indicators (such as the proportion of positive reactions, the proportion of negative reactions, and neutral response parameters) from the electrolysis feature vector and establishes voltage regulation rules associated with the reaction units based on preset fuzzy logic rules or machine learning models. For example, when the proportion of positive reactions falls below a threshold, a regulation rule is generated to increase the pulse amplitude and obtain the corresponding pulse correction parameters (such as voltage increment and frequency adjustment).

[0058] Timing Optimization Module: This module uses timing analysis algorithms to identify temporal correlation patterns in pulse correction parameters, such as the time intervals between pulse parameter adjustments at different reaction stages. Based on these correlation patterns, it dynamically calibrates key reaction indicators and uses algorithms such as Kalman filtering to calculate the voltage offset of each reaction unit under different calibration strategies.

[0059] Benchmark Setting Module: Based on the voltage offset, the optimal pulse threshold is derived through a historical data matching algorithm. Specifically, the current electrolysis characteristic value is compared with the characteristic value of the historical optimal reaction mode, and an electrolysis deviation sequence containing the voltage deviation of each reaction unit is generated to reflect the difference between the current reaction state and the ideal state.

[0060] Decision-making execution module: Utilizes the deviation analysis algorithm to analyze the electrolysis deviation sequence, combines the voltage offset trend of the reaction unit, integrates the discrete deviation data into a continuous electrolysis control execution plan, and adjusts the output parameters of the pulse voltage source module in real time through the drive circuit to achieve closed-loop control of the electrolysis process.

[0061] The present invention will be further described below in conjunction with Examples 1 to 5:

[0062] Example 1:

[0063] In the electrolysis reaction module, building an electrolysis feature library corresponding to the reaction unit is a basic link in the implementation method. The electrolysis feature library stores the current data of each reaction unit in real time through the data acquisition system. The current data is a waveform sequence of continuous sampling, and the sampling frequency is synchronized with the output frequency of the pulse voltage source module to ensure the consistency of the data time dimension. For example, when the output frequency of the pulse voltage source module is 1000Hz, the sampling frequency of the current data is set to not less than 2000Hz to meet the Nyquist sampling theorem and avoid signal aliasing. The reaction parameter vector is generated through multi-sensor fusion technology. In addition to current data, it also integrates real-time data collected by equipment such as temperature sensors, pressure sensors and gas composition analyzers. Feature mapping is performed through a neural network model to convert multi-dimensional physical quantities into low-dimensional reaction parameter vectors. The vector contains parameters that indirectly reflect the reaction state, such as temperature gradient, pressure fluctuation coefficient, and product selectivity coefficient, to achieve multi-dimensional characterization of the electrolysis process.

[0064] When matching reaction parameter vectors for similar responses, a clustering algorithm from machine learning is used to group the data. Specifically, the reaction parameter vectors are first standardized to eliminate the influence of different parameter dimensions on the clustering results. Z-score normalization can be used as a standard normalization method. This involves calculating the mean and standard deviation of each parameter to transform the raw data into a standard normal distribution with a mean of 0 and a standard deviation of 1. The preprocessed data is then input into a clustering algorithm. Common algorithms include K-means clustering, hierarchical clustering, or DBSCAN density clustering. Taking the K-means clustering algorithm as an example, K cluster centers are first randomly initialized. The distance between each reaction parameter vector and each cluster center (using either Euclidean or cosine distance) is then calculated. The vector is then assigned to the cluster with the closest cluster center. The centroid of each cluster is then recalculated as the new cluster center. This assignment and updating process is repeated until the cluster center no longer changes significantly or the preset number of iterations is reached. This completes the clustering of the reaction parameter vectors, forming distinct reaction cluster groups.

[0065] Extracting electrolytic focusing points from current data within reaction clusters is a key step in identifying reaction units. This process relies on a feature analysis of the current data, aiming to identify key data points that represent the typical reaction characteristics of each cluster. Specifically, the current waveforms within each cluster are first superimposed and averaged to generate a typical current curve for that cluster. This typical current curve is then differentiated to calculate the slope change, identifying characteristic locations such as the rising and falling edges of the current and the peak point. Simultaneously, the energy integral of the current curve is calculated to identify extreme points of energy consumption. These characteristic locations and extreme points are selected as candidate focusing points. Further statistical analysis is performed to select points within the cluster whose occurrence frequency exceeds a preset threshold as the final electrolytic focusing points. For example, if 80% of the current waveforms within a cluster exhibit a current peak 2ms after the rising edge of the pulse voltage, this time point is identified as the electrolytic focusing point for that cluster.

[0066] When setting a focal point as a reaction unit, it must be defined in conjunction with the time dimension of the pulse parameter interval. A reaction unit can be the time point or time period corresponding to the focal point, depending on the time resolution requirements of the electrolysis process. For example, if the focal point is the time point when the current peak occurs, the reaction unit can be defined as a short time period centered on that time point, such as a time window of 1ms before and after, to ensure that the unit contains the complete peak characteristics. After dividing the reaction units, each unit corresponds to a specific reaction stage in the electrolysis process, facilitating the subsequent refined analysis of the current response at each stage.

[0067] In building the electrolysis characteristic library, the data storage structure was implemented using a database management system, such as MySQL or MongoDB, to support efficient storage and fast querying of massive amounts of data. The database table structure includes fields for timestamps, current data, reaction parameter vectors, and clustering labels. Data indexing is established using timestamps to ensure real-time data access across modules. Furthermore, to avoid data redundancy, the reaction parameter vectors are serialized, converting multidimensional vectors into binary data streams for storage, reducing storage space.

[0068] During the similar response matching process, the clustering algorithm parameters need to be adjusted based on the actual reaction characteristics. For example, in scenarios where the reaction state changes rapidly, the number of clusters (K) can be appropriately increased to improve clustering precision; in scenarios where the reaction process is relatively stable, the K value can be reduced to reduce computational complexity. Furthermore, to prevent the K-means algorithm from falling into a local optimum, a strategy of randomly initializing the cluster centers multiple times and selecting the optimal result can be employed, or the cluster center initialization can be combined with an optimization algorithm such as a genetic algorithm.

[0069] The extraction of electrolysis focusing points can also be manually manipulated by incorporating domain knowledge. For example, based on the chemical reaction mechanism of carbon dioxide electrolysis, the current characteristic ranges corresponding to certain key reaction stages (such as carbon dioxide adsorption, electron transfer, and product desorption) can be predefined to guide the focusing point extraction process to prioritize the characteristic data of these stages. This combination of data-driven and knowledge-driven approaches can improve the scientific and practical nature of reaction unit delineation.

[0070] After the reaction units are divided, the electrolytic feature vectors of each unit need to be checked for consistency to ensure that the reaction units within the same cluster have similar reaction characteristics. This verification method can be used to calculate the intra-cluster distance of the electrolytic feature vectors of each unit within the cluster. If the intra-cluster distance exceeds a preset threshold, the clustering process is reviewed, and clustering parameters are adjusted or focal points are re-extracted until consistency requirements are met.

[0071] Example 2:

[0072] When dividing the reaction parameter vector into reaction cluster groups, it is necessary to combine the time attribute and current intensity in the reaction parameter vector, further extract the voltage distribution, reaction polarity and response density parameters to generate reaction feature labels, and achieve accurate division of cluster groups by calculating the similarity of the feature labels. The specific implementation method is as follows:

[0073] The extraction of time attributes is based on the timestamp information corresponding to the reaction parameter vector, analyzing the temporal position of the parameter vector in the electrolysis process, such as the rising edge, falling edge, plateau period, or intermittent period of the pulse voltage. The current intensity is directly taken from the real-time current data in the electrolysis feature library, reflecting the activity level of the electrolysis reaction at that moment. By analyzing the correlation between time attributes and current intensity, the differences in reaction characteristics at different stages can be identified. For example, the plateau period may correspond to a stable electrolysis reaction stage, while the rising or falling edge may be accompanied by violent current fluctuations, corresponding to key processes such as reactant adsorption or product desorption.

[0074] Extracting voltage distribution parameters requires combining the real-time voltage sequence output by the pulse voltage source module to calculate the fluctuation range of parameters such as voltage amplitude, pulse width, and duty cycle at the corresponding time point for each reaction parameter vector. For example, the maximum, minimum, and average voltage amplitudes over multiple consecutive pulse cycles are statistically analyzed to form a characteristic description of the voltage distribution. The reaction polarity parameter is determined by the correlation between current direction and voltage polarity. When the current direction matches the pulse voltage polarity, a positive reaction is considered dominant (e.g., a reduction reaction); otherwise, a negative reaction is considered dominant (e.g., an oxidation reaction). If the current fluctuation is small and lacks clear directionality, a neutral response state is defined. The response density parameter is calculated by counting valid reaction events per unit time. A valid reaction event can be defined as a fluctuation event whose current change amplitude exceeds a preset threshold. This threshold is dynamically adjusted based on the noise level of the electrolysis reaction, for example, 2-3 times the standard deviation of the current baseline is used as the discrimination threshold.

[0075] When generating reaction feature labels, the extracted voltage distribution, reaction polarity, and response density parameters are semantically converted. For example, if the voltage amplitude corresponding to a reaction parameter vector fluctuates between 30-40V (voltage distribution characteristics), the current direction and voltage polarity are consistent (positive reaction predominance), and the number of valid reaction events per unit time is 15 (high-density response), then the feature label "High Voltage Fluctuation - Positive Predominance - High-Density Response" is generated. The semantic structure of the label is composed of "attribute-value" pairs, ensuring that each label uniquely represents a combination of reaction features, facilitating subsequent similarity calculations and cluster analysis.

[0076] After associating the reaction feature labels with the reaction parameter vectors, the response similarity between the feature labels needs to be calculated to screen cluster group members. This similarity calculation method can use a metric based on semantic distance. For example, a semantic network of feature labels is constructed, and each label is decomposed into independent feature dimensions (such as voltage distribution, reaction polarity, and response density). Each dimension corresponds to a different value range (e.g., "high voltage fluctuation" corresponds to a voltage amplitude fluctuation range >20V, "medium voltage fluctuation" corresponds to 10-20V, and "low voltage fluctuation" corresponds to <10V). For each dimension, if the values ​​of two labels fall into the same or adjacent intervals, a higher similarity score (e.g., 0.8-1.0) is assigned; if they fall into non-adjacent intervals, the score decreases according to the distance between them (e.g., a score of 0.5 for a one-interval interval and 0.2 for two or more intervals). Finally, the similarity scores of each dimension are weighted and averaged to obtain the comprehensive similarity value of the two labels. The weighting coefficient is preset according to the degree of influence of each dimension on the reaction mechanism. For example, the weight of the reaction polarity dimension can be set to 0.4, and the weights of the voltage distribution and response density dimensions can be set to 0.3 each.

[0077] When selecting parameter vectors with similarities exceeding a preset response threshold (e.g., 0.7) to form response clusters, a hierarchical clustering merging strategy is employed. Each parameter vector is first treated as an independent cluster unit. Label similarities are calculated between all units. Unit pairs with the highest similarity exceeding the threshold are merged into a new cluster. The similarity between this new cluster and other units (or clusters) is then recalculated, and the merging process is repeated until no similarity pairs meet the threshold. During this process, the similarity values ​​and cluster composition of each merge are recorded to facilitate subsequent interpretable analysis of the clustering results.

[0078] To extract temporal attributes, to avoid high-frequency noise interference, the timestamp sequence can be filtered using a sliding window. The window width is set based on the pulse voltage cycle length. For example, when the pulse frequency is 500 Hz, the window width is set to 5 cycles (i.e., 10 ms). The smoothed temporal features are obtained by calculating the median or mean of the temporal attributes within the window. Current intensity is also processed using filtering algorithms, such as Gaussian filtering or Savitzky-Golay filtering, to eliminate the impact of sampling noise on feature extraction.

[0079] Determining reaction polarity requires considering the electrode configuration of the electrolysis reaction and the mechanism of carbon dioxide reduction. For example, in a typical carbon dioxide electrolysis cell, oxidation reactions occur at the anode (such as oxygen evolution from water) and reduction reactions occur at the cathode (conversion of carbon dioxide to hydrocarbons or carbon monoxide). Therefore, a current flowing from anode to cathode corresponds to a forward reaction (where the reduction reaction dominates), while a reverse current flow may indicate abnormal electrode polarization or the occurrence of a side reaction. By establishing a mapping between current direction and reaction polarity, abstract current data can be converted into physically meaningful reaction characteristics.

[0080] The calculation of the response density parameter requires defining a time interval constraint for valid reaction events to avoid splitting continuous fluctuations within the same reaction process into multiple independent events. For example, if the event interval threshold is set to 0.5ms, two current fluctuation events whose time interval is less than this threshold are considered to be continuations of the same event and are not counted repeatedly. This constraint can be adjusted based on the time constant of the electrolysis reaction to ensure that the count of valid reaction events truly reflects the independent chemical reaction steps.

[0081] When generating reaction feature labels, a unified label dictionary must be established to clearly define the value range and semantic definition of each feature dimension to avoid clustering bias caused by terminology ambiguity. For example, in the voltage distribution dimension, the classification of "high," "medium," and "low" must be linked to the voltage amplitude range of the pulse parameter interval. When the pulse voltage amplitude range is 0-50V, high voltage fluctuations can be defined as >35V, medium voltage fluctuations as 15-35V, and low voltage fluctuations as <15V, ensuring that the label definitions are consistent with the actual operating parameters of the system.

[0082] During the similarity calculation process, for parameter vectors with missing feature dimensions (e.g., missing data for a dimension due to a sensor failure), nearest neighbor interpolation or a feature correlation-based filling algorithm can be used to fill in the gaps using corresponding dimension data from other vectors in the same clustering group to ensure the integrity of the similarity calculation. If a vector is missing multiple key dimensions (e.g., missing both voltage distribution and reaction polarity), it will not be included in the clustering until the data is complete or marked as an anomaly.

[0083] After forming the reaction clustering group, the rationality of the clustering results needs to be verified. The verification methods include: (1) Visual analysis, using principal component analysis (PCA) to reduce the high-dimensional reaction parameter vector to a two-dimensional or three-dimensional space, draw a scatter plot of the cluster group distribution, and observe whether each cluster forms a clear cluster structure; (2) Statistical test, using the ANOVA analysis of variance method to test whether there are significant differences in the mean values ​​of different cluster groups in each feature dimension (significance level α = 0.05). If the mean difference in a dimension is not significant, it is necessary to re-examine the feature extraction process of that dimension or adjust the clustering parameters.

[0084] Example 3:

[0085] When generating the electrolysis eigenvector corresponding to the reaction unit, it is necessary to dynamically analyze the current fluctuation characteristics of each reaction unit at its temporal position within the pulse parameter interval and adopt a differentiated eigenvector generation strategy. The specific implementation is as follows:

[0086] Determine the temporal position of the reaction unit within the pulse parameter interval. This position is determined by the unit's corresponding timestamp and the periodic characteristics of the pulse voltage sequence. For example, a pulse voltage sequence consists of continuous rectangular pulses, each pulse cycle consisting of four phases: rising edge, plateau, falling edge, and rest period. The reaction unit may be located in any of these phases. By synchronously collecting the trigger signal and current data from the pulse voltage source module, a mapping relationship between the reaction unit's temporal position and the pulse phase is established, providing a time reference for subsequent current fluctuation analysis.

[0087] For each reaction unit, obtain its current fluctuation data within the preset window. The preset window is centered on the timestamp of the reaction unit and extends forward and backward for a certain length of time. The window length is determined by the pulse period and the reaction time constant. For example, when the pulse frequency is 1000Hz (period 1ms), the preset window can be set to 0.5ms before and after, with a total length of 1ms, to ensure that the window covers the pulse phase corresponding to the unit and partial data of the adjacent phase to capture the complete dynamic process of the current response. The current fluctuation data is a sequence of current sampling values ​​within the window. The sampling frequency must meet the Nyquist criterion and is usually more than twice the pulse frequency to avoid frequency aliasing.

[0088] When calculating the current fluctuation coefficient of a unit, a statistical method is used to quantify the degree of fluctuation of the current data. The calculation formula for the current fluctuation coefficient is:

[0089]

[0090] in, is the current fluctuation coefficient, which is used to characterize the relative amplitude of current fluctuation; is the number of sampling points within the preset window; For the The current value of each sampling point; is the average current value within the preset window. This formula eliminates the influence of the current mean on the fluctuation amplitude by calculating the coefficient of variation of the current data (the ratio of the standard deviation to the mean), facilitating horizontal comparison of the fluctuation degree between different reaction units.

[0091] When the current fluctuation coefficient Exceeding the first response threshold When , the unit is determined to be a high response unit. The first response threshold The current coefficient of variation is pre-set based on the noise level and historical data of the electrolysis reaction. For example, it is twice the current coefficient of variation during normal steady-state operation to ensure that the high-response unit corresponds to the stage of intense reaction or state mutation. For high-response units, their current data is directly extracted to form the electrolysis feature vector. The extracted current data includes but is not limited to: peak current , average current , current rising edge slope , falling edge slope , current duration These parameters directly reflect the kinetic characteristics of the electrolysis reaction in the unit. For example, the peak current corresponds to the utilization rate of the reaction active sites, and the rising edge slope reflects the reaction start-up speed.

[0092] When the current fluctuation coefficient Below the first response threshold When the cell is marked as a steady-state cell. The steady-state cell corresponds to the stable stage of the electrolysis reaction, and the current fluctuation is small, but the current data of the adjacent cells may contain gradual trend information or weak characteristic changes. The current data of each unit is superimposed. The superposition method uses weighted summation, and the weight coefficient decreases according to the time interval from the current unit. For example, exponential decay weight is used:

[0093]

[0094] in, For the The weight coefficients of adjacent cells, Take the integer, and ; is the time constant, which is set according to the pulse period and is usually Through this weight distribution, the adjacent units closer to the current unit contribute more to the superposition result, reflecting the temporal locality of the current response.

[0095] The current data after response superposition needs to be reconstructed to generate the electrolysis feature vector. The reconstruction process includes: firstly arranging the current data of adjacent units in time sequence to form a length of The sequence is then smoothed and filtered to eliminate high-frequency noise that may have been introduced during the superposition process. Finally, the statistical and trend characteristics of the filtered data, such as mean, variance, linear fit slope, and trend change points, are extracted as the electrolysis feature vectors of the steady-state unit. These features reflect the overall level and slowly changing trend of the steady-state current, avoiding the loss of valuable information due to small fluctuations in the data of a single steady-state unit.

[0096] When determining the preset window length, the pulse width of the pulse voltage and the rate of reaction product formation must be comprehensively considered. For example, for a pulse with a pulse width of 0.8 ms, the preset window length can be set to 1.2 ms to ensure that the window covers the entire pulse width and part of the rest period to capture any afterglow effects or current responses during product desorption after the pulse ends. For fast reaction systems (such as those involving highly active catalysts), the window length can be appropriately shortened to focus on transient response characteristics; for slow reaction systems, the window length can be extended to obtain more steady-state data.

[0097] First response threshold The setting needs to be completed through statistical analysis of historical data. The specific steps are: collect the current data of continuous operation under normal working conditions, calculate the current fluctuation coefficient of all reaction units , draw its probability density distribution curve. According to the curve, determine the steady-state operation The distribution range is usually taken as the mean plus 2 times the standard deviation. , so that the threshold can distinguish more than 95% of steady-state cells from high-response cells. After the threshold is set, it must be dynamically adjusted according to actual operating conditions. For example, when the catalyst is replaced or the electrolyte composition is adjusted, the statistical data is re-analyzed to update the threshold to ensure the adaptability of the classification standard.

[0098] When extracting current data from high-response units, attention must be paid to outlier handling. Sampling points that significantly deviate from the normal fluctuation range (e.g., current values ​​exceeding ±3 standard deviations of the preset range) are identified as outliers and corrected using interpolation methods (such as linear or cubic spline interpolation) to prevent outliers from interfering with the eigenvector. Outlier detection and correction are performed in real time during the data acquisition phase to ensure the accuracy of subsequent analysis.

[0099] The number of neighboring units whose responses are superimposed It is an adjustable parameter, and its value range is usually 1-5 units. When , only one adjacent unit before and after is superimposed, which is suitable for scenarios where the current trend changes rapidly; when , 5 units are superimposed before and after, which is suitable for scenarios where slow trend changes need to be captured. The value of can be determined by experimental debugging, for example, by comparing different The ability of the eigenvector to characterize the reaction state is selected to make the eigenvector with the highest classification accuracy. value.

[0100] The dimensionality of the electrolysis eigenvector varies depending on the type of reaction unit. For highly responsive units, the eigenvector may contain 5-8 dimensions (such as peak current, average current, and slope parameters). For steady-state units, after response superposition and reconstruction, the eigenvector may contain 3-5 dimensions (such as mean, variance, and trend slope). The dimensionality of the eigenvector must be kept within a reasonable range to avoid excessively high dimensionality, which can lead to increased computational complexity in subsequent parameter control modules, while ensuring sufficient information capacity to distinguish between different reaction states.

[0101] In the time series position analysis, if the reaction unit is located in the rising edge stage of the pulse sequence, its current fluctuation is usually more intense, and the proportion of high-response units is higher. At this time, it is necessary to focus on extracting parameters such as peak current and rising edge slope; if it is located in the plateau stage, the current tends to be stable and the proportion of steady-state units increases. It is necessary to capture the subtle change trend of adjacent units through response superposition; if it is located in the intermittent stage, the current may be close to the baseline level. At this time, the characteristic vector can reflect the recovery process of the electrode or the background current characteristics.

[0102] Example 4:

[0103] In the parameter control module, the process of separating the positive reaction ratio, negative reaction ratio, and neutral response parameters from the electrolysis feature vector and generating voltage regulation rules requires combining the electrochemical mechanism of the electrolysis reaction with data feature analysis. The following detailed implementation is described using specific examples:

[0104] Assume that the electrolysis characteristic vector for a reaction unit contains the following parameters: peak current of 2.5A, average current of 1.8A, rising slope of 0.5A / ms, falling slope of -0.3A / ms, and reaction duration of 4ms. Based on the correlation between current direction and pulse voltage polarity (assuming a positive voltage polarity and current flow from cathode to anode is defined as a forward reaction), the current direction of this unit is positive, and the forward reaction is determined to be dominant. Using preset thresholds (e.g., a forward reaction contribution of ≥60% is considered high, 30%-60% is considered medium, and <30% is considered low), combined with current parameters and reaction product analysis (e.g., a high CO generation rate detected by a gas sensor), the forward reaction contribution of this unit is determined to be 70%, the negative reaction contribution is 20%, and the neutral response parameter (e.g., side reaction or background current contribution) is 10%.

[0105] Based on the above parameters, the logic for generating voltage regulation rules is as follows: If the proportion of positive reactions is higher than a threshold (e.g., a preset threshold of 50%) and the proportion of negative reactions is lower than a threshold (e.g., 30%), the current voltage parameters are determined to be appropriate and the existing pulse voltage sequence is maintained. If the proportion of positive reactions is lower than a threshold and the proportion of negative reactions is higher than a threshold (e.g., if the positive proportion is 25% and the negative proportion is 55%), a regulation rule of "increasing the pulse amplitude by 10% and reducing the frequency by 20%" is generated to enhance the driving force of the positive reaction and prolong the reaction time. If the neutral response parameter exceeds a preset threshold (e.g., 20%), monitoring of the electrolyte composition or electrode status is triggered. For example, impedance spectroscopy analysis is used to determine whether there is electrode passivation or electrolyte contamination, and then auxiliary regulation rules are generated (e.g., starting electrolyte circulation or adjusting the electrode spacing).

[0106] When the number of reaction units covered by the current voltage regulation rule is less than the preset response threshold (for example, if the total number of reaction units is 100 and the current rule only covers 15, which is lower than the preset 20), the rule needs to be expanded by traversing the electrolysis feature vectors of adjacent reaction units. For example, adjacent unit A, its feature vector shows a 35% positive reaction contribution, a 45% negative reaction contribution, and a 20% neutral response parameter. It also includes parameters not covered by the current rule, such as pulse rise time (2ms) and reaction product selectivity (CO selectivity 60%). Feature difference analysis revealed that the pulse rise time of this unit is too long, which may lead to insufficient reactant adsorption and, in turn, affect the positive reaction contribution. Therefore, the response metric "reducing the pulse rise time to less than 1.5ms" is added to the current rule, forming a new regulation rule combination: "increase the pulse amplitude by 10%, reduce the frequency by 20%, and shorten the rise time by 30%."

[0107] In another scenario, if a reaction unit is in the intermittent period of a pulse sequence, its electrolysis eigenvector indicates a current close to the baseline level (0.1A), with a positive reaction ratio of 10%, a negative reaction ratio of 15%, and a neutral response parameter of 75%. The parameter control module identifies the unit as being in a low-activity state and generates a regulation rule to superimpose a small pulse (5V amplitude, 100Hz frequency) during the intermittent period to activate residual reactants on the electrode surface and reduce the neutral response ratio. If this rule only covers the five units in the intermittent period (less than the preset threshold of 10), the module then iterates through the adjacent plateau units and finds a negative correlation between the average current of the units in the plateau period and the pulse duty cycle (the higher the duty cycle, the slower the average current increase). This leads to the inclusion of a metric to dynamically adjust the duty cycle during the plateau period (reducing the duty cycle by 5% every 10ms) in the rule, forming a coordinated cross-stage regulation strategy.

[0108] The generation of voltage regulation rules relies on a pre-defined rule base, built based on the fundamental theory and historical data of electrolysis reactions. For example, the rule base includes the empirical rule that "for every 1V increase in pulse amplitude, the proportion of positive reactions increases by approximately 2%-5% (within the 30-40V range)" and the statistical rule that "for every 100Hz decrease in frequency, the proportion of negative reactions decreases by approximately 3%-6%." The parameter ranges in the rule base must match the adjustable range of the pulse voltage source module, for example, the amplitude adjustment step size is 1V and the frequency adjustment step size is 50Hz, to ensure the enforceability of the rules.

[0109] When traversing adjacent reaction units, feature difference analysis uses principal component analysis (PCA) to reduce the high-dimensional electrolysis feature vector to 2-3 principal components. By observing the distribution distance of each unit in the principal component space, samples that differ significantly from the units covered by the current rule are identified. For example, the units covered by the current rule are mainly distributed in the positive intervals of principal component 1 (reflecting current intensity) and principal component 2 (reflecting time characteristics), while the adjacent unit A is located in the negative interval of principal component 2. This indicates that its time characteristics (such as pulse rise time) are significantly different from the characteristics of the current rule, and time-related indicators need to be incorporated into the rule.

[0110] For electrolysis systems with multi-electrode arrays, reaction units at different electrode sites may have different characteristic parameters. For example, reaction units at the edge of the electrode generally have a higher proportion of forward reactions than those in the center due to their higher mass transfer efficiency. The parameter control module needs to generate differentiated rules for units in different regions, such as using lower pulse amplitudes for edge region units (to avoid side reactions caused by overpotential) and higher frequencies for center region units (to enhance mass transfer). At this time, if the number of units covered by the rules in a certain area is insufficient, it is necessary to prioritize traversing adjacent units in the same area to avoid mixing cross-regional features, which may lead to a decrease in rule generalization ability.

[0111] The update frequency of voltage regulation rules is related to the temporal resolution of the reaction units. When the reaction unit granularity is 10ms, the rule update cycle can be set to perform a global evaluation every 100ms (i.e., 10 units). This ensures that the control actions can respond promptly to reaction changes while avoiding system instability caused by frequent adjustments. During each rule update, the number of units currently covered by each rule is first counted. If the number is below a threshold, the traversal mechanism is triggered; otherwise, the existing rules are maintained.

[0112] When generating pulse correction parameters, the response delay of the hardware module must be considered. For example, the amplitude adjustment of the pulse voltage source module requires a 0.5ms stabilization time. Therefore, the amplitude adjustment amount in the rule must compensate for this delay in advance to ensure that the correction parameter takes effect in the target reaction unit. The value range of the correction parameter must also be limited by the hardware capabilities. For example, the amplitude cannot exceed the upper limit of 50V, and the frequency cannot fall below the lower limit of 10Hz to avoid issuing invalid commands or damaging the equipment.

[0113] Analysis of neutral response parameters can also be combined with thermodynamic calculations. For example, by calculating the difference between the theoretical decomposition voltage and the actual applied voltage (overpotential) under the current pulse parameters, it can be determined whether the neutral response is caused by insufficient energy. If the overpotential is less than 80% of the theoretical value, a rule is generated to "increase the amplitude to increase the overpotential." If the overpotential is sufficient but the neutral response remains high, it may indicate a kinetic obstacle (such as insufficient catalyst active sites) and require coordinated adjustment through other modules (such as the electrode pretreatment module).

[0114] Example 5:

[0115] In the timing optimization module, the process of identifying the temporal correlation pattern in the pulse correction parameters and calculating the voltage offset requires combining the temporal characteristics of the pulse sequence with the dynamic correlation of the parameter adjustment. The following is a detailed implementation through a specific example:

[0116] Suppose that the pulse correction parameter sequence for a certain stage includes the following adjustment records: at 10ms, the pulse amplitude is adjusted by +5% and the frequency by -10%; at 20ms, the amplitude is adjusted by +3% and the frequency by -5%; and at 30ms, the amplitude is adjusted by +2% and the frequency by -3%. By analyzing the time intervals and parameter change amplitudes of these records, we can extract the sequence factor of the time switching frequency (e.g., an adjustment every 10ms) and the variation factor of the response jump amplitude (e.g., the amplitude adjustment amplitude decreases from 5% to 2%, and the frequency adjustment amplitude decreases from 10% to 3%). The sequence factor reflects the temporal density of parameter adjustments, while the variation factor reflects the amplitude difference between adjacent adjustments. Together, these two factors constitute the core characteristics of the temporal correlation pattern.

[0117] When constructing the weight assignment matrix associated with the sequence factor and the variation factor, the time period characteristics of the sequence factor are first identified. If the time interval of the adjustment records is 10ms, which exactly matches the preset pulse period (10ms), the sequence factor is determined to be strictly periodic and is set as the starting index of the weight assignment matrix. The starting index corresponds to the first row or column of the matrix and is used to anchor the time dimension. Next, the correlation between the sequence factor and the variation factor is calculated. For example, by observing whether the amplitude adjustment decreases with increasing time interval, a negative correlation can be determined. If statistics show that the amplitude adjustment decreases by an average of 2% for every 5ms increase in time interval, the correlation can be defined as a high level of matching (e.g., a matching score of 0.8). The intermediate and ending indices of the matrix are generated in descending order of matching. The intermediate indices correspond to the sequence factors at different time intervals, and the ending indices correspond to the amplitudes of the variation factors at each time interval.

[0118] When backtracking the path of the termination index, it is necessary to check whether its matching degree meets a preset threshold (such as 0.5). For example, if the time interval corresponding to a termination index is 15ms and the amplitude adjustment is 1%, and analysis shows that this adjustment has no obvious correlation with the previous and next adjustments, the matching degree is calculated to be 0.4, which is below the threshold. Therefore, it is output as an independent item in the final distribution of the weight allocation matrix to avoid errors caused by forced association. The structure of the weight allocation matrix can be designed as a two-dimensional table, with rows representing sequence factors (time intervals), columns representing change factors (adjustment amplitudes), and cell values ​​representing the probability or weight of the occurrence of each time-amplitude combination. For example, at a 10ms interval, the probability of an amplitude adjustment of 5% is 0.6, the probability of an adjustment of 3% is 0.3, and the probability of an adjustment of 2% is 0.1.

[0119] To calculate voltage offset, first calculate the mean and range of the sequence factor for each termination index in the weight distribution matrix. For example, the time interval sequence corresponding to a given termination index is 10ms, 10ms, and 20ms, with a mean of 13.3ms; the variation factors (amplitude adjustment ranges) are 5%, 3%, and 2%, with a range of 3%. The global variance of all index factors is also calculated to reflect the overall fluctuation level of the time intervals and adjustment ranges. The mean of the sequence factor for a single termination index is subtracted from the mean of the adjacent indexes to obtain the timing offset (for example, the difference between 13.3ms and the adjacent index mean of 10ms is 3.3ms). Dividing this by the global variance yields the dimensionless timing offset coefficient, which normalizes the impact of time differences. The ratio of the variation factor range to the global variance is also calculated to reflect the relative change in the adjustment range. The timing offset coefficient and the amplitude change ratio are weighted and summed according to preset weights (such as a time weight of 0.6 and an amplitude weight of 0.4) to obtain the voltage offset for the index. For example, if the timing offset coefficient is 0.5 and the amplitude change ratio is 0.3, the weighted offset is 0.5 × 0.6 + 0.3 × 0.4 = 0.42, corresponding to a voltage adjustment of +0.42% (or -0.42%, the sign depends on the offset direction).

[0120] In another scenario, if the temporal correlation pattern of the pulse correction parameters exhibits non-periodic characteristics, such as adjustment intervals of 8ms, 12ms, and 9ms, and the sequence factor lacks obvious periodic characteristics, it is not set as the starting index. Instead, the matrix is ​​constructed based on the amplitude characteristics of the change factor. For example, records with amplitude adjustments of 4% or greater are classified as "high amplitude," 2%-4% as "medium amplitude," and <2% as "low amplitude." The time interval distribution within each category is then calculated. If the time intervals in the "high amplitude" category are concentrated between 8-10ms, the "medium amplitude" category is concentrated between 10-12ms, and the "low amplitude" category is concentrated between 12-14ms, then the weight distribution matrix is ​​composed of amplitude categories as rows and time interval ranges as columns. The cell values ​​are the frequency of occurrence of time intervals within each category. For example, the frequency of "high amplitude - 8-10ms" is 0.7, indicating that high amplitude adjustments tend to occur within shorter intervals.

[0121] When identifying time-periodic characteristics, autocorrelation analysis can be used to calculate the autocorrelation function of the sequence factor and detect whether there are significant peaks (for example, an autocorrelation coefficient of 0.9 at a 10ms lag) to determine whether periodicity exists. If periodicity exists, further verification is conducted to verify whether the period length is consistent with the fundamental period of the pulse voltage source module (for example, the period corresponding to the user-set pulse frequency) to ensure that the time-correlation pattern is synchronized with the inherent rhythm of the system.

[0122] The construction of the weight allocation matrix also needs to consider the stage characteristics of the electrolysis reaction. For example, during the reaction startup phase (the first 100ms), the pulse correction parameters are adjusted more frequently (with a smaller sequence factor) and with larger amplitudes (with a higher variation factor). Therefore, the first few rows of the weight matrix can prioritize assigning high weights to combinations with short intervals and high amplitudes. During the stable operation phase, the adjustment frequency and amplitude decrease, and the weight matrix shifts to combinations with long intervals and low amplitudes. By constructing the matrix in stages, the timing optimization can be more closely aligned with the evolving reaction kinetics.

[0123] During path backtracking, if the matching degree of a termination index is found to be below a threshold, in addition to being output as an independent item, it can also be marked as an "abnormal adjustment," triggering backtracking analysis of the reaction unit corresponding to the adjustment. For example, the electrolysis eigenvector of the unit is checked for abnormal values ​​(such as a sudden increase in the current fluctuation coefficient) to determine whether the adjustment is caused by noise data. If so, the record is removed to prevent abnormal data from contaminating the weight matrix.

[0124] The calculated voltage offset must be mapped to the adjustable parameter range of the pulse voltage source module. For example, if the calculated offset is +0.42%, and the module's amplitude adjustment step size is 1%, the result must be rounded to the nearest integer, resulting in an actual adjustment of +1%. The adjustment error must also be recorded for subsequent optimization. If the offset is negative, such as -0.35%, the amplitude is reduced by 0% (due to step size limitations), but the error must be accumulated until the next adjustment to ensure the system's continuous control capabilities.

[0125] The identification of time-correlated patterns can also be integrated with domain knowledge, such as the mass transfer lag effect in carbon dioxide electrolysis. If, after adjusting pulse parameters, the formation of reaction products requires a certain amount of time (e.g., 5ms) to be reflected in the current data, a time delay factor should be introduced when analyzing time-correlated patterns. This allows the adjusted records to be correlated with the reaction unit characteristics after a 5ms lag, thus avoiding pattern misidentification due to timing misalignment.

[0126] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "includes," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0127] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A carbon dioxide electrolysis system based on pulse voltage regulation, characterized in that: include: The pulse voltage source module is used to generate an adjustable pulse voltage sequence and set the pulse parameter range corresponding to the electrolysis reaction; The electrolysis reaction module is used to divide multiple reaction units within the pulse parameter range, analyze the current response of the electrolysis process of each reaction unit, and generate the electrolysis characteristic vector corresponding to the reaction unit; The parameter control module is used to extract key reaction indicators from the electrolysis characteristic vector, establish voltage regulation rules associated with the reaction unit, and obtain pulse correction parameters corresponding to the rules; The timing optimization module is used to identify the time correlation pattern in the pulse correction parameters, dynamically calibrate the key reaction indicators based on the correlation pattern, and calculate the voltage offset of each reaction unit under different calibration strategies; A benchmark setting module is used to derive the optimal pulse threshold according to the voltage offset and generate an electrolysis deviation sequence by comparing the current electrolysis characteristic value with the optimal pulse threshold; The decision-making execution module is used to analyze the electrolysis deviation sequence and integrate the electrolysis deviation sequence into an electrolysis control execution plan based on the voltage offset trend of the reaction unit; The electrolysis reaction module is implemented by: constructing an electrolysis feature library corresponding to the reaction unit, the electrolysis feature library including current data and reaction parameter vectors mapped by electrolysis response; Performing similar response matching on the reaction parameter vectors and dividing the reaction parameter vectors into reaction cluster groups according to the matching results; extracting electrolytic focus points of the current data from the reaction cluster groups and setting the focus points as reaction units; The implementation methods for generating the electrolysis feature vector corresponding to the reaction unit include: For each reaction unit, according to the unit's temporal position in the pulse parameter interval, the unit's current fluctuation data within a preset window is obtained, and the unit's current fluctuation coefficient is calculated; When the current fluctuation coefficient exceeds the first response threshold, the unit is marked as a high-response unit, and its current data is extracted to form an electrolysis feature vector; when the current fluctuation coefficient is lower than the first response threshold, the unit is marked as a steady-state unit, and the current data of the unit's adjacent units are superimposed on the response, and the superimposed data are reconstructed into an electrolysis feature vector; The implementation methods of the parameter control module include: Separating the positive reaction ratio, negative reaction ratio, and neutral response parameters from the electrolysis characteristic vector, and generating a voltage regulation rule for the reaction unit based on the above parameters; If the number of reaction units covered by the current voltage regulation rule is less than the preset response threshold, the electrolysis feature vectors of adjacent reaction units are traversed, and the response indicators not included in the regulation rules of the adjacent units are added to the current rule.

2. A carbon dioxide electrolysis system based on pulse voltage control according to claim 1, characterized in that: The reaction cluster groups that partition the reaction parameter vector also include: According to the time attributes and current intensity in the reaction parameter vector, the voltage distribution, reaction polarity and response density parameters are extracted, and the reaction feature label is generated based on the above parameters; The reaction feature labels are associated with the reaction parameter vectors. By calculating the response similarity between the feature labels, the parameter vectors with similarity higher than the preset response threshold are screened to form a reaction cluster group.

3. The carbon dioxide electrolysis system based on pulse voltage control according to claim 1, characterized in that: The implementation method of the timing optimization module includes: obtaining a sequence factor of a time switching frequency and a change factor of a response jump amplitude in a time correlation pattern; A weight distribution matrix associated with the sequence factor and the variation factor is constructed, and the voltage offset under different calibration strategies is determined according to the distribution probability of each element in the matrix.

4. A carbon dioxide electrolysis system based on pulse voltage control according to claim 3, characterized in that: Constructing the weight distribution matrix also includes: Identify the time period characteristics of the sequence factor. If the current period characteristics completely match the preset time period, set the sequence factor as the starting index of the weight allocation matrix. Calculate the correlation matching degree between the sequence factor and the change factor, and generate the intermediate index and the end index of the weight distribution matrix in descending order of matching degree; The path of the termination index is backtracked, and when the matching degree of the termination index is lower than the preset matching threshold, it is output as the final distribution of the weight allocation matrix.

5. The carbon dioxide electrolysis system based on pulse voltage control according to claim 4, characterized in that: The implementation methods for calculating voltage offset include: Calculate the mean and range of the sequence factors of each termination index in the weight distribution matrix, and calculate the global variance of all index factors; The timing offset coefficient is obtained by subtracting the mean of the sequence factor of a single termination index from the mean of the sequence factors of the adjacent indexes and dividing it by the global variance. At the same time, the ratio of the variation factor range to the global variance is calculated, and the weighted sum of the two is taken as the voltage offset of the index.

6. The carbon dioxide electrolysis system based on pulse voltage control according to claim 1, characterized in that: The implementation of deriving the optimal pulse threshold includes: Extract the reaction pattern closest to the current voltage offset in the historical data, and calculate the cosine similarity of the two in time distribution as the first dynamic reference value; Counting the difference in the number of response peaks between the current voltage offset and the historical response pattern, and using the difference as a second dynamic reference value; Based on the linear combination of the first dynamic reference value and the second dynamic reference value, the optimal pulse threshold in the preset pulse threshold table is matched.

7. The carbon dioxide electrolysis system based on pulse voltage control according to claim 1, characterized in that: The implementation method of the decision execution module includes: dividing the voltage offset interval and the negative offset interval according to the voltage offset direction of each unit in the electrolysis deviation sequence; The convergence rate of the electrolysis deviation in the positive offset interval and the diffusion rate of the electrolysis deviation in the negative offset interval are extracted, and the two are antagonistically fused according to the time weight of the reaction unit to generate the adjustment parameters of the electrolysis control execution plan.

Images