A production control method for sodium perchlorate based on electrolysis

Abstract

The application discloses a production control method for producing sodium perchlorate based on an electrolysis method, real-time collection of core parameters in an electrolytic cell is realized through a distributed sensor array, and implicit states such as an anode aging coefficient, a mass transfer efficiency and a proportion of a side reaction are analyzed through a state perception model; after the parameters and the implicit states are fused, the short-term prediction value is obtained by inputting the parameters and the implicit states into a yield-energy consumption double-target prediction model; optimal control instructions are generated through a multivariate collaborative control model, and finally, closed-loop control is realized through feedback iteration. The application solves the hysteresis and parameter coupling problems of a traditional control method, realizes multi-target optimization of yield, energy consumption and electrode life, significantly improves production efficiency and stability, and has important industrial application value.

Description

Technical Field

[0001] This invention belongs to the field of sodium perchlorate production technology, and relates to a production control method for producing sodium perchlorate by electrolysis. Background Technology

[0002] Sodium perchlorate is a chemical product with wide and indispensable applications in many key fields such as fireworks manufacturing and the electronics industry. Electrolysis has become the mainstream method of industrial production due to its advantages of high efficiency and controllability. However, in the traditional electrolytic production of sodium perchlorate, production control relies heavily on manual experience. Operators pre-set fixed parameters such as current density and electrolyte temperature based on past experience, and these parameters remain basically unchanged throughout the production process, without fully considering the complex dynamic coupling relationships between multiple parameters within the electrolysis system.

[0003] Electrolysis reactions are affected by various factors. The performance of anode materials degrades over time, the electrolyte composition may change due to raw materials and reaction byproducts, and mass transfer efficiency fluctuates due to factors such as flow rate and temperature. Traditional fixed-parameter control cannot adapt to these dynamic changes, resulting in unstable sodium perchlorate yields and large fluctuations in purity, making it difficult to meet the consistent product quality requirements of high-end applications. Furthermore, because parameters cannot be adjusted according to actual energy needs, unit energy consumption is high, and energy utilization is low. For T i New anode materials such as 4O7 cannot fully realize their performance advantages using traditional methods, resulting in suboptimal service life and increased equipment costs. Furthermore, manual adjustments rely on experience-based judgment and periodic testing data, which are highly lagging and unable to cope with the time-varying characteristics of the electrolysis system, further exacerbating the aforementioned problems.

[0004] Traditional sodium perchlorate electrolysis production control methods, which rely on manual experience to set fixed parameters, suffer from significant technical bottlenecks in yield, purity, energy consumption, material utilization, and response to system changes. These problems not only increase production costs and reduce efficiency but also limit the industry's development towards intelligent, efficient, and green production. Therefore, developing an intelligent control method capable of real-time sensing of the electrolysis system's status and dynamic optimization of control parameters is urgently needed to overcome existing technological limitations and improve the overall level and economic benefits of sodium perchlorate electrolysis production. Summary of the Invention

[0005] To address the problems existing in the background technology, this invention proposes a production control method for producing sodium perchlorate by electrolysis. Through multi-model collaboration and real-time closed-loop feedback, the method achieves precise control of the electrolysis process, improves the sodium perchlorate yield, reduces energy consumption, and extends the anode lifespan, thereby adapting to the dynamic changes of the electrolysis system.

[0006] The first aspect of this application provides a production control method for the electrolytic production of sodium perchlorate, comprising: Multiple core parameters within the electrolytic cell are collected in real time using a distributed sensor array. Multiple core parameters are input into the state-aware model to obtain the anode aging coefficient, electrolyte mass transfer efficiency, and the proportion of oxygen evolution side reaction. Multiple core parameters are integrated with the obtained anode aging coefficient, electrolyte mass transfer efficiency, and oxygen evolution side reaction ratio into a multi-dimensional feature vector, which is then input into the yield-energy consumption dual-objective prediction model. The yield-energy consumption dual-objective prediction model is an attention mechanism-BP neural network model, which outputs short-term predicted values ​​of sodium perchlorate yield and unit energy consumption. The output sodium perchlorate yield and short-term predicted values ​​of unit energy consumption are input into a multivariate collaborative control model. The multivariate collaborative control model is a hybrid algorithm combining an improved non-dominated sorting genetic algorithm and fuzzy PID. The fuzzy PID is used to convert the data into actuator instructions. The system operates according to the output actuator instructions, and feeds back the actual sodium perchlorate yield and unit energy consumption to the state-aware model and the yield-energy consumption dual-objective prediction model for online parameter updates, forming a closed-loop control.

[0007] Optionally, the state-aware model is characterized by an algorithm comprising an improved Kalman filter and support vector machine fusion model, wherein the Kalman filter is used to eliminate sensor noise as follows: ; in, The data at time k is the filtered data. Here is the state transition matrix. To control the input matrix, To control the quantity, For Kalman gain, The original monitoring data at time k, The observation matrix; Support vector machines train state evaluation models based on filtered data, using anode impedance and anode potential as core input features. The hidden states are mapped through radial basis kernel functions to obtain anode aging coefficient, electrolyte mass transfer efficiency, and the proportion of oxygen evolution side reactions.

[0008] Optionally, the yield prediction formula of the attention mechanism-BP neural network model is: ; in, These are elements in a multidimensional feature vector composed of multiple monitoring parameters, anode aging coefficient, electrolyte mass transfer efficiency, and the proportion of oxygen evolution side reactions. For attention weights, The connection weights are from the j-th input to the i-th hidden layer neuron. Let be the weights from the i-th hidden layer neuron to the output layer. For the bias terms of the hidden layer, For output layer bias terms, The activation function used for the hidden layer. The activation function used for the output layer.

[0009] Optionally, the formula for predicting unit energy consumption is: ; Where U is the electrolytic cell voltage, I is the anode current density, t is the electrolysis time, m is the raw material input, Y is the sodium perchlorate yield, γ is the proportion of oxygen evolution side reaction, and λ is the anode aging coefficient.

[0010] Optionally, the optimization objective function of the improved non-dominated sorting genetic algorithm is: ; Where E is the predicted unit energy consumption. For the theoretical maximum yield, For the lowest theoretical energy consumption, , , The target weight.

[0011] Optionally, the constraints of the optimization objective function include anode current density ∈ [1000, 3000] A / m 2 Electrolytic cell temperature ∈ [30, 50]℃, electrolyte flow rate ∈ [0.5, 2]m / s, oxygen evolution side reaction ratio ≤15%.

[0012] Optionally, the variable collaborative control model includes an anode protection priority mechanism. When the state-aware model detects that the anode aging coefficient is less than or equal to a preset threshold, the multivariate collaborative control model increases the priority based on the preset threshold. The weight.

[0013] Optionally, after the production batch ends within a preset time period, the difference between the actual sodium perchlorate yield and the predicted sodium perchlorate yield is input into the yield-energy consumption dual-objective prediction model to update the attention weights of the attention mechanism-BP neural network using the gradient descent method. .

[0014] Compared with the prior art, the present invention has the following beneficial effects: The aforementioned production control method for producing sodium perchlorate via electrolysis firstly acquires core parameters within the electrolytic cell in real time using a distributed sensor array. A state-aware model then analyzes implicit states such as anode aging coefficient, mass transfer efficiency, and the proportion of side reactions. After fusing the core parameters with these implicit states, the results are input into a yield-energy consumption dual-objective prediction model to obtain short-term predicted values. Next, a multi-variable collaborative control model generates optimal control commands, and finally, closed-loop control is achieved through feedback iteration. This invention solves the hysteresis and parameter coupling problems of traditional control methods, achieving multi-objective optimization of yield, energy consumption, and electrode lifespan, significantly improving production efficiency and stability, and possessing significant industrial application value. Attached Figure Description

[0015] Figure 1 This is a flowchart of a production control method for producing sodium perchlorate by electrolysis in one embodiment of the present invention. Detailed Implementation

[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0017] In one embodiment, such as Figure 1 As shown, a production control method for producing sodium perchlorate by electrolysis is provided, which is applied to... Figure 1 Taking China as an example, the following specific steps will be used: S10: Real-time acquisition of multiple core parameters within the electrolytic cell via a distributed sensor array.

[0018] Specifically, a distributed sensor array can provide comprehensive, accurate, and real-time raw data support for subsequent state analysis, predictive optimization, and control decisions. Distributed deployment and multi-parameter acquisition avoid the limitations of single-point, single-parameter monitoring, and can completely reflect the overall state within the electrolyzer. Several core parameters include: reaction kinetics, material state parameters, and equipment and side reaction parameters. The reaction kinetics include: electrolyzer temperature T, which affects the reaction rate; anode current density I, which determines the electrolysis intensity; and anode / cathode potential, reflecting the electrode reaction trend. Material state parameters include: such as sodium chlorate concentration C3 (remaining raw material), sodium perchlorate concentration C4 (product generation), electrolyte pH (affecting reaction direction), and electrolyte flow rate v (affecting material mixing and mass transfer). Equipment and side reaction parameters include: such as Ti4O7 anode impedance Z, reflecting the degree of anode aging and hydrogen production rate. ，The side reaction intensity and electrolyte conductivity σ reflect ion concentration, indirectly indicating mass transfer efficiency. These parameters are directly related to the efficiency of the electrolysis reaction, product purity, energy consumption, and equipment lifespan, and serve as the primary data source for subsequent model analysis.

[0019] S20: Input multiple core parameters into the state-sensing model to obtain the anode aging coefficient, electrolyte mass transfer efficiency, and the proportion of oxygen evolution side reactions.

[0020] Specifically, the state-aware model is an algorithm that combines an improved Kalman filter and a support vector machine fusion model. The Kalman filter eliminates sensor noise by: ; in, The data at time k is the filtered data. Here is the state transition matrix. To control the input matrix, To control the quantity, For Kalman gain, The original monitoring data at time k, The observation matrix; Support vector machines train state evaluation models based on filtered data, using anode impedance and anode potential as core input features. The hidden states are mapped through radial basis kernel functions to obtain anode aging coefficient, electrolyte mass transfer efficiency, and the proportion of oxygen evolution side reactions.

[0021] The anodic aging factor is an indicator used to quantify the degree of performance degradation of anodes, such as Ti4O7 anodes, during the electrolytic production of sodium perchlorate. Its value ranges from 0 to 1, where 1 represents a brand-new anode (i.e., no performance degradation) and 0 represents a completely failed anode (i.e., unusable). The following explanation uses Ti4O7 anodes as an example.

[0022] The anodic aging coefficient is calculated using a state-aware model, with core input parameters including Ti4O7 anode impedance Z and anode potential E. a Anode surface temperature T a And electrolyte conductivity σ, etc. These parameters directly reflect changes in the physical and electrochemical properties of the anode: for example, with long-term use of the anode, the thickening of the surface oxide layer leads to an increase in impedance Z, and the intensification of the oxygen evolution side reaction increases the anode potential E. a These changes, after being analyzed by the model, are converted into the decay value of the anodic aging coefficient.

[0023] In this application, the core function of the anode aging coefficient is to provide equipment status information for subsequent control decisions: when the anode aging coefficient is slightly aged, the control model can prioritize yield optimization; when the anode aging coefficient is moderately aged, the model will balance yield and energy consumption, appropriately reducing current density to delay aging; when the anode aging coefficient is severely aged, the system will activate the anode protection mechanism, extending its service life by reducing operating temperature and optimizing electrolyte flow rate. The dynamic control based on quantitative indicators in this application solves the problem of lag in traditional experience-based judgment of anode status, achieving precise matching between equipment performance and production efficiency.

[0024] Electrolyte mass transfer efficiency η is a key indicator for measuring the efficiency of ion diffusion and migration in the electrolyte within an electrolytic cell, such as sodium chlorate ions and perchlorate ions. A higher value indicates smoother ion transport in the electrolyte, enabling more efficient participation in the electrolytic reaction at the electrode surface. This indicator is calculated using a state-aware model, specifically a modified Kalman filter-support vector machine fusion model. Core input parameters include electrolyte flow rate (v), electrolyte conductivity (σ), sodium chlorate concentration (C3), and sodium perchlorate concentration (C4). These parameters directly reflect the dynamic characteristics of the mass transfer process: for example, increasing the electrolyte flow rate enhances convection and diffusion, and increasing conductivity indicates an increase in ion concentration, both of which promote improved mass transfer efficiency. Conversely, excessively large concentration gradients, such as low sodium chlorate concentration near the anode, inhibit ion migration, leading to a decrease in electrolyte mass transfer efficiency.

[0025] Electrolyte mass transfer efficiency is crucial to the uniformity and efficiency of the electrolysis reaction: when the electrolyte mass transfer efficiency is greater than or equal to 90%, ions can be rapidly replenished to the electrode surface, avoiding a decrease in reaction rate or an increase in the proportion of oxygen evolution side reactions due to local ion depletion; when the electrolyte mass transfer efficiency is less than 70%, uneven ion distribution on the electrode surface occurs, leading to an increase in local overpotential, increasing energy consumption and reducing product purity. Therefore, electrolyte mass transfer efficiency, as a core state parameter, is incorporated into the yield-energy consumption prediction model, providing a basis for multivariate coordinated control. For example, when the yield-energy consumption prediction model detects a low electrolyte mass transfer efficiency, it will increase the electrolyte flow rate to enhance convection or adjust the electrolyte circulation path to optimize mass transfer, ensuring efficient electrolysis.

[0026] The oxygen evolution side reaction ratio is a key indicator in the quantitative electrolytic production of sodium perchlorate, representing the proportion of electricity consumed to generate oxygen from the oxygen evolution side reaction on the anode surface to the total electrolytic electricity. It is expressed as a percentage and directly reflects the selectivity of the main reaction, i.e., the selectivity of sodium chlorate to sodium perchlorate.

[0027] The proportion of oxygen evolution side reactions was calculated using a state-aware model, with key input parameters including anode potential, cathode potential, hydrogen production rate, and electrolyte pH. In an ideal electrolysis process, electrons should preferentially participate in the main reaction (ClO3).- →ClO4 - However, when the anode potential is too high or the electrolyte condition is abnormal, the oxygen evolution side reaction (2H2O→O2↑+4H2O) will be triggered. + +4e - This leads to electron waste. The model accurately calculates the proportion of oxygen evolution side reactions by analyzing the deviation between the anode potential and the theoretical oxygen evolution potential, and the correlation between the hydrogen production rate and the main reaction. In the hydrogen production rate, hydrogen is generated by the main reaction, while the side reactions do not directly produce hydrogen.

[0028] The proportion of oxygen evolution reaction (OER) side reactions has a significant impact on energy consumption and yield: when the OER proportion is less than or equal to 10%, the main reaction selectivity is high, and electrical energy is mainly used to generate sodium perchlorate, resulting in low unit energy consumption; when the OER proportion is greater than or equal to 20%, a large amount of electrical energy is consumed by the side reactions, leading not only to a surge in unit energy consumption but also potentially exacerbating the oxidation and corrosion of the Ti4O7 anode due to excessively high local oxygen concentrations, thus reducing the OER proportion. Therefore, the OER proportion is one of the core parameters of the yield-energy consumption prediction model. When the model detects that the OER proportion exceeds the standard, the multivariate collaborative control model will suppress the OER by reducing the anode current density and adjusting the electrolyte pH, thereby improving the efficiency of the main reaction and ensuring effective energy utilization.

[0029] S30: Multiple core parameters are integrated with the obtained anode aging coefficient, electrolyte mass transfer efficiency, and oxygen evolution side reaction ratio into a multi-dimensional feature vector, which is then input into the yield-energy consumption dual-objective prediction model. The yield-energy consumption dual-objective prediction model is an attention mechanism-BP neural network model, which outputs short-term predicted values ​​of sodium perchlorate yield and unit energy consumption.

[0030] The formula for predicting unit energy consumption is: ; The formula for predicting unit energy consumption precisely quantifies the energy consumption during the electrolytic production of sodium perchlorate through multi-dimensional factor coupling. It includes energy calculations for the basic electrolytic reaction and incorporates the impact of side reactions and equipment status on energy consumption, making it a core indicator reflecting production efficiency. U is the electrolytic cell voltage, calculated in real-time from the difference between the anode and cathode potentials, directly reflecting the driving potential of the electrolytic reaction; I is the anode current density, multiplied by the anode area to obtain the total current, a key parameter characterizing electrolysis intensity; t is the electrolysis time, reflecting the reaction duration; m is the raw material input, representing the total amount of sodium perchlorate participating in the reaction; Y is the sodium perchlorate yield, reflecting the proportion of raw materials converted into the target product. This is a fundamental term in energy consumption calculations, representing the ratio of total electrical energy (U*I*t) to effective product quantity (m*Y), directly reflecting the energy efficiency of the basic reaction. The formula (1+0.5γ) is a correction term for the oxygen evolution side reaction, where γ represents the proportion of the oxygen evolution side reaction. The oxygen evolution side reaction (2H₂O→O₂↑+4H₂) + +4e - The side reaction consumes some electrical energy but does not generate the target product, resulting in an artificially high energy consumption. When γ increases, the value of this correction term increases (e.g., when γ=20%, the correction term is 1+0.5×0.2=1.1), which means that the impact of the side reaction on energy consumption is quantified and amplified, making the prediction results more consistent with reality. (1+0.3(1-λ)) is the anode aging correction term, where λ is the anode aging coefficient (range 0-1). The smaller λ is, the more severe the Ti4O7 anode aging is, that is, the impedance increases, and a higher voltage is required to maintain the same electrolysis intensity, indirectly increasing energy consumption. For example, when λ=0.5, that is, moderate aging, the correction term is 1+0.3×(1-λ)). 0.5)=1.15, which amplifies the additional contribution of aging to energy consumption.

[0031] S40: Input the short-term predicted values ​​of sodium perchlorate yield and unit energy consumption into the multivariate collaborative control model. The multivariate collaborative control model is a hybrid algorithm combining an improved non-dominated sorting genetic algorithm and fuzzy PID. The fuzzy PID is used to convert the data into actuator instructions.

[0032] S50: Operates according to the output actuator instructions, feeding back the actual sodium perchlorate yield and unit energy consumption to the state-aware model and the yield-energy consumption dual-objective prediction model for online parameter updates, forming a closed-loop control.

[0033] Specifically, this application fully presents the closed-loop control logic of decision-making, execution, and feedback in the intelligent production of sodium perchlorate. Dynamic optimization is achieved through a multi-variable collaborative control model, and accuracy is continuously improved based on real-time feedback, ultimately achieving high efficiency and self-adaptation in the production process. This closed-loop system breaks through the limitations of traditional fixed-parameter control and can accurately cope with complex changes in the electrolysis environment.

[0034] The decision-making process of the multivariate collaborative control model is dominated by an improved non-dominated sorting genetic algorithm. It receives short-term predictions of sodium perchlorate yield, unit energy consumption, and anode aging coefficient. Within preset constraints, such as current density, temperature, and flow rate, it generates multiple sets of candidate control parameters through selection, crossover, and mutation operations simulating biological evolution. These parameters are then screened using non-dominated sorting to find the Pareto optimal solution. Finally, the optimal combination of control parameters is determined by combining the weights of the anode aging coefficient, achieving a multi-objective balance of maximizing yield, minimizing energy consumption, and maximizing anode lifetime.

[0035] The execution phase of the multivariable collaborative control model is completed by fuzzy PID, which transforms the abstract optimal parameters output by the improved non-dominated sorting genetic algorithm into specific action commands for the actuators. Fuzzy PID combines the empirical nature of fuzzy control with the precision of PID. Through preset rules, such as increasing the heating power by 25% when the temperature difference exceeds 3℃, it quickly reduces the deviation. Then, through proportional, integral, and derivative adjustments, it eliminates steady-state errors, ensuring that equipment such as rectifiers, heating devices, and infusion pumps operate precisely according to commands, such as linearly adjusting the current density from 2500A / m² to 2800A / m², avoiding shock fluctuations.

[0036] The objective function of the improved non-dominated sorting genetic algorithm is: ; in, For the theoretical maximum yield, For the lowest theoretical energy consumption, , , represents the target weight. Wherein, , , These are used to quantify the priority of sodium perchlorate yield (Y), predicted unit energy consumption (E), and anode aging coefficient (λ) in the overall optimization. The sum of the three is 1 ( + + =1), which is the core parameter for balancing multi-objective conflicts by dynamically adjusting the value to emphasize different objectives.

[0037] The yield target weight reflects the priority of improving sodium perchlorate yield in the optimization. A larger value indicates that the algorithm is more inclined to sacrifice some energy consumption or anode lifetime to improve Y. For example, when... When the value is 0.4, it indicates that the yield is the highest priority among the three objectives; if the production task requires prioritizing output, such as urgent orders, the yield can be temporarily adjusted. When the value is increased to 0.5, the algorithm will prioritize the control parameters that can improve Y, such as appropriately increasing the current density, even if it may lead to a slight increase in energy consumption.

[0038] This is the weight assigned to the energy consumption target, used to adjust the priority of reducing unit energy consumption. A larger value indicates that the algorithm focuses more on controlling energy consumption, and may even moderately reduce yield or allow for faster anode aging. For example, =0.3 is the benchmark value in normal production. If the enterprise is in a period of energy shortage, it can be adjusted accordingly. When the value is increased to 0.4, the algorithm will prioritize the combination of parameters such as low current density and optimized flow rate to ensure that E is controlled within the target range, such as ≤0.85kWh / kg.

[0039] This represents the target weight for anode lifetime, reflecting the importance placed on extending the lifespan of Ti4O7 anodes. It is directly related to the anode aging coefficient λ; the higher the value, the more the algorithm tends to mitigate λ decay by reducing current density and controlling anode surface temperature. For example, in conventional production... =0.3; when the state-aware model detects λ≤0.6, i.e., moderate aging of the anode, It will automatically increase to 0.5. At this point, the algorithm will prioritize low load parameters, such as reducing the current density to below 2000A / m², in order to avoid further aging of the anode and extend the replacement cycle.

[0040] In this embodiment, the constraints of the optimization objective function include anode current density ∈ [1000, 3000] A / m2, electrolytic cell temperature ∈ [30, 50] ℃, electrolyte flow rate ∈ [0.5, 2] m / s, and oxygen evolution side reaction ratio ≤ 15%.

[0041] Specifically, the constraints of the optimization objective function described in this invention are boundary restrictions to ensure the safe, efficient, and stable operation of the electrolysis process. Their function is to limit the optimization decisions of the improved non-dominated sorting genetic algorithm within a technologically feasible range, avoiding production accidents, product defects, or equipment damage due to parameter runaway. These constraints are formulated based on the chemical characteristics of the sodium perchlorate electrolysis reaction, equipment performance, and safety production regulations.

[0042] The anolyte current density is limited to the range of [1000, 3000] A / m², taking into account both reaction efficiency and equipment load-bearing capacity. Too low a current density (<1000 A / m²) will result in insufficient sodium chlorate oxidation rate and a significant decrease in yield; while too high a current density (>3000 A / m²) will cause a sharp increase in the overpotential on the anolyte surface, exacerbating the oxygen evolution side reaction (increased γ), which not only increases energy consumption but also accelerates the oxidation and corrosion of the Ti4O7 anolyte due to localized high temperatures, shortening its service life. This range represents a balance between efficiency and safety verified through extensive experiments.

[0043] The electrolyzer temperature is constrained within the range of [30, 50]℃, primarily based on considerations of both reaction kinetics and product stability. Below 30℃, the ion diffusion rate in the electrolyte slows down, reducing the electrolyte mass transfer efficiency and thus decreasing the reaction rate. Above 50℃, while the reaction can be accelerated, it exacerbates water evaporation in the electrolyte, leading to abnormal fluctuations in sodium chlorate / perchlorate concentrations and even causing localized crystallization that blocks the flow channels. Furthermore, high temperatures increase the activity of the oxygen evolution side reaction, which is detrimental to energy consumption control. This temperature range ensures a balance between reaction rate and product stability.

[0044] Limiting the electrolyte flow rate to [0.5, 2] m / s optimizes mass transfer efficiency and controls energy consumption. Below 0.5 m / s, the electrolyte turnover rate on the electrode surface is slow, easily leading to concentration gradients. For example, if the sodium chlorate concentration near the anode is too low, it can cause uneven reaction and trigger side reactions. While a flow rate above 2 m / s enhances mass transfer, it significantly increases the energy consumption of the circulating pump, and the high-speed turbulence may scour the anode surface, damaging the stability of the Ti4O7 coating. This range ensures mass transfer efficiency while avoiding unnecessary energy consumption and equipment wear.

[0045] The oxygen evolution reaction (γ) ratio must be ≤15%, which is a key constraint to ensure the selectivity of the main reaction. When γ exceeds 15%, it means that more than 15% of the electrical energy is consumed by the oxygen evolution reaction. This not only directly leads to a surge in unit energy consumption (E) (from 0.8 kWh / kg to 1.2 kWh / kg), but also accelerates the oxidative aging of Ti4O7 due to excessively high oxygen concentration on the anode surface, causing γ to drop rapidly. Simultaneously, it may cause abnormal fluctuations in the electrolyte pH, affecting the purity of sodium perchlorate. By limiting the upper limit of γ, it is ensured that electrical energy is mainly used for the main reaction and the production of sodium perchlorate, maintaining the economic efficiency of production and product quality.

[0046] The variable collaborative control model includes an anode protection priority mechanism. When the state-aware model detects that the anode aging coefficient is less than or equal to a preset threshold, the multi-variable collaborative control model then prioritizes protection based on the preset threshold. The weights are determined by the production batch. After the production batch is completed within the preset time period, the difference between the actual sodium perchlorate yield and the predicted sodium perchlorate yield is input into the yield-energy consumption dual-objective prediction model to update the attention weights of the attention mechanism-BP neural network through gradient descent.

[0047] The anode protection priority mechanism is a dynamic adjustment strategy designed for the aging state of Ti4O7 anodes in a multivariate collaborative control model. Its core function is to prioritize the protection of anode life by adjusting and optimizing the target weights when the anode performance degrades to a critical point, so as to avoid production interruption or cost surge due to excessive wear.

[0048] The triggering condition for the anode protection priority mechanism is determined by the state-aware model: when the model detects that the anode aging coefficient is ≤ a preset threshold, the system automatically activates the protection mechanism. The preset threshold is typically set to 0.6, indicating moderate anode aging. This threshold setting is based on the performance degradation curve of the Ti4O7 anode. When λ ≤ 0.6, the anode impedance (Z) has significantly increased. If high-load operation continues, such as high current density and high temperature, it will lead to an increase in the anode potential (E). a The rapid increase in oxygen evolution rate accelerates the oxygen evolution side reaction and material oxidation and corrosion, which may reduce the anode life from the expected 18 months to less than 12 months, seriously affecting production continuity.

[0049] After the mechanism is activated, the multivariate collaborative control model achieves protection by adjusting the weights of the objective function: increasing the anode lifetime target weight ω3 from the usual 0.3 to 0.5 (while reducing the yield weight ω1 to 0.3 and the energy consumption weight ω2 to 0.2). After the weight adjustment, the improved non-dominated sorting genetic algorithm will prioritize combinations that reduce anode load when selecting optimal control parameters, for example, reducing the current density (I) from the usual 2500 A / m 2 Reduced to 2000A / m 2 The following describes how to reduce the anode surface temperature (T). a The temperature is lowered to below 40°C, and the electrolyte flow rate (v) is appropriately increased to enhance heat dissipation. Although these parameter adjustments may slightly reduce the short-term yield (e.g., from 95% to 93%), they can significantly slow down the decay rate of λ (e.g., from a decrease of 0.05 per month to 0.03 per month), ensuring stable operation of the anode until the preset replacement cycle.

[0050] In the production process, the process of updating the attention mechanism-BP neural network weights based on the difference between actual and predicted yields is the core link in achieving closed-loop optimization of "prediction, control, and feedback." Essentially, it uses production practice data to correct the prediction model, improving the model's fitting accuracy to complex process dynamics and providing a more reliable decision-making basis for subsequent optimization control. The specific process of updating the attention mechanism-BP neural network weights based on the difference between actual and predicted yields is as follows: First, choosing the right time to update is crucial, as it coincides with the end of the production batch within a preset time period. A production batch is a unit with relatively stable process parameters and high data integrity, such as a cycle of continuous operation for 24 hours or achieving a specific output. Within a batch, fluctuations in key parameters such as electrolyzer temperature, current density, electrolyte flow rate, and the proportion of oxygen evolution side reactions are controlled within constraints. Yield data, i.e., the hourly sodium perchlorate production, reflects the true reaction efficiency under the process conditions of that batch. Updating after the batch ends avoids noise interference caused by instantaneous fluctuations, such as short-term flow rate anomalies, in real-time updates. This ensures that the error signal used to correct the model—the difference between the actual and predicted yields—is statistically significant, rather than a random disturbance.

[0051] Secondly, the difference between the actual and predicted sodium perchlorate yield serves as a feedback signal for model correction. Before the start of a production batch within a preset time period, the attention mechanism-backpropagation neural network predicts the yield of the production batch within that preset time period based on current process parameters (such as initial electrolyte concentration and equipment status). After the production batch ends within the preset time period, the actual measured yield (obtained through online detection or offline analysis) is compared with the predicted value. The resulting difference, i.e., the prediction error, directly reflects the model's shortcomings. For example, if the actual yield of a batch is 5% lower than the predicted value, it may be because the model underestimated the impact of decreased mass transfer efficiency at low electrolyte flow rates on the yield, or it failed to fully capture the inhibitory effect of small changes in the anode aging coefficient on reactivity. This error becomes the core basis for driving model optimization.

[0052] Furthermore, the yield-energy consumption dual-objective prediction model uses an attention mechanism-BP neural network as its core structure. The BP neural network is responsible for fitting the nonlinear relationship between process parameters (such as current density and electrolyzer temperature) and sodium perchlorate yield and unit energy consumption. The attention mechanism is the feature filter of this network, adjusting the influence of different input features, such as current density and the proportion of oxygen evolution side reactions, on the prediction results through weight adjustments. For example, if the attention weights show that the weight of current density is 0.3 and the weight of electrolyzer temperature is 0.2, it indicates that the yield-energy consumption dual-objective prediction model considers the current density to have a more significant impact on the yield. This invention inputs the sodium perchlorate yield error into the model and transmits the "error signal" to the attention mechanism layer of the network to determine which unreasonable weight allocation of the input features leads to the prediction deviation.

[0053] Finally, this invention utilizes gradient descent to update attention weights as a technical means to correct the model. Gradient descent is an optimization algorithm that minimizes error by iteratively adjusting parameters. In this process, the yield-energy consumption dual-objective prediction model uses the yield prediction error as the loss function, such as the sum of squared errors, and calculates the gradient of the loss function with respect to each attention weight, i.e., the degree of influence of small changes in weights on the error. Subsequently, the weights are adjusted in the opposite direction of the gradient. For example, if the error indicates that "the model underestimates the influence of electrolyte flow rate," then gradient calculation will reveal that "the attention weight of electrolyte flow rate is too low." In this case, the yield-energy consumption dual-objective prediction model will increase the weight of electrolyte flow rate, such as from 0.15 to 0.22, making subsequent predictions more focused on the impact of changes in electrolyte flow rate on yield. Through multiple rounds of batch iterative updates, the attention weights gradually approach the "importance of true features," such as the fact that the influence of current density on yield is indeed greater than that of temperature, thus making the model's prediction results closer to actual production patterns.

[0054] The yield prediction formula for the attention mechanism-BP neural network model is as follows: ; in, These are elements in a multidimensional feature vector composed of multiple monitoring parameters, anode aging coefficient, electrolyte mass transfer efficiency, and the proportion of oxygen evolution side reactions. For attention weights, The connection weights are from the j-th input to the i-th hidden layer neuron. Let be the weights from the i-th hidden layer neuron to the output layer. For the bias terms of the hidden layer, For output layer bias terms, The activation function used for the hidden layer. The activation function used for the output layer.

[0055] The attention mechanism-BP neural network model's yield prediction formula achieves accurate prediction of sodium perchlorate yield by integrating dynamic allocation of feature weights with nonlinear fitting capabilities. The BP neural network maps weighted features to predicted yield values ​​through layer-by-layer calculations in the input, hidden, and output layers. The attention weights... The feature scoring function is used to calculate the correlation strength between each input feature and the yield (e.g., the weight of current density may be four times the electrolyte flow rate), and it is dynamically adjusted according to production conditions (e.g., its weight is increased during anode aging). The weighted feature vector focuses on key information, and then passes through the hidden layer (handling nonlinear relationships through activation functions) and output layer of a BP neural network to calculate the final yield prediction. This design not only solves the limitation of traditional models that treat all features equally, but also captures complex process laws such as the surge in side reactions caused by excessively high current density, accurately adapting to the dynamic characteristics of electrolysis production and providing a reliable basis for optimized control.

[0056] Since closed-loop feedback is crucial for maintaining the long-term stability of the system, after the actuator operates, the sensors collect the actual yield and energy consumption in real time, feeding the data back into the state-aware model and the yield-energy consumption prediction model. The state-aware model uses actual data to correct the calculation logic of parameters such as the anode aging coefficient and the proportion of oxygen evolution side reactions; the prediction model updates the neural network weights through deviation (ΔY = Y_actual - Y_predicted), such as increasing the weight of the influence of "anode impedance" on the yield, continuously reducing the prediction error.

[0057] For example, taking a production line as an example, when the predicted Y=92% and E=0.95kWh / kg (the target values ​​are ≥95% and ≤0.85kWh / kg respectively), the improved non-dominated sorting genetic algorithm outputs I*=2800A / m 2With T*=44℃ and v*=1.6m / s, after the fuzzy PID controller is converted into specific instructions, the actuator achieves Yactual=96% and Eactual=0.82kWh / kg within 10 minutes. Feedback data further optimizes the model, reducing the prediction deviation for the next round to ΔY=1.2% and ΔE=0.03kWh / kg. Continuous operation verification shows that this closed-loop system can increase the average yield to 95%, reduce energy consumption to 0.8kWh / kg, and extend anode life by 4 months, fully demonstrating its technological advantages.

[0058] Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A production control method for producing sodium perchlorate by electrolysis, characterized in that, include: Multiple core parameters within the electrolytic cell are collected in real time using a distributed sensor array. Multiple core parameters are input into the state-aware model to obtain the anode aging coefficient, electrolyte mass transfer efficiency, and the proportion of oxygen evolution side reaction. Multiple core parameters are integrated with the obtained anode aging coefficient, electrolyte mass transfer efficiency, and oxygen evolution side reaction ratio into a multi-dimensional feature vector, which is then input into the yield-energy consumption dual-objective prediction model. The yield-energy consumption dual-objective prediction model is an attention mechanism-BP neural network model, which outputs short-term predicted values ​​of sodium perchlorate yield and unit energy consumption. The output sodium perchlorate yield and short-term predicted values ​​of unit energy consumption are input into a multivariate collaborative control model. The multivariate collaborative control model is a hybrid algorithm combining an improved non-dominated sorting genetic algorithm and fuzzy PID. The fuzzy PID is used to convert the data into actuator instructions. The system operates according to the output actuator instructions, and feeds back the actual sodium perchlorate yield and unit energy consumption to the state-aware model and the yield-energy consumption dual-objective prediction model for online parameter updates, forming a closed-loop control.

2. The production control method for producing sodium perchlorate by electrolysis according to claim 1, characterized in that, The state-aware model is an improved Kalman filter and support vector machine fusion model. The algorithm includes Kalman filtering and support vector machine. The process of Kalman filtering to eliminate sensor noise is as follows: ; in, The data at time k is the filtered data. Here is the state transition matrix. To control the input matrix, To control the quantity, For Kalman gain, The original monitoring data at time k, The observation matrix; Support vector machines train state evaluation models based on filtered data, using anode impedance and anode potential as core input features. The hidden states are mapped through radial basis kernel functions to obtain anode aging coefficient, electrolyte mass transfer efficiency, and the proportion of oxygen evolution side reactions.

3. The production control method for producing sodium perchlorate by electrolysis according to claim 1, characterized in that, The yield prediction formula for the attention mechanism-BP neural network model is as follows: ； in, These are elements in a multidimensional feature vector composed of multiple monitoring parameters, anode aging coefficient, electrolyte mass transfer efficiency, and the proportion of oxygen evolution side reactions. For attention weights, The connection weights are from the j-th input to the i-th hidden layer neuron. Let be the weights from the i-th hidden layer neuron to the output layer. For the bias terms of the hidden layer, For output layer bias terms, The activation function used for the hidden layer. The activation function used for the output layer.

4. The production control method for producing sodium perchlorate by electrolysis according to claim 1, characterized in that, The formula for predicting unit energy consumption is: ； Where U is the electrolytic cell voltage, I is the anode current density, t is the electrolysis time, m is the raw material input, Y is the sodium perchlorate yield, γ is the proportion of oxygen evolution side reaction, and λ is the anode aging coefficient.

5. The production control method for producing sodium perchlorate by electrolysis according to claim 1, characterized in that, The objective function of the improved non-dominated sorting genetic algorithm is: ； Where E is the predicted unit energy consumption. For the theoretical maximum yield, For the lowest theoretical energy consumption, , , The target weight.

6. The production control method for producing sodium perchlorate by electrolysis according to claim 5, characterized in that, The constraints of the objective function include the anode current density ∈ [1000, 3000] A / m 2 Electrolytic cell temperature ∈ [30, 50]℃, electrolyte flow rate ∈ [0.5, 2]m / s, oxygen evolution side reaction ratio ≤15%.

7. The production control method for producing sodium perchlorate by electrolysis according to claim 5, characterized in that, The variable collaborative control model includes an anode protection priority mechanism. When the state-aware model detects that the anode aging coefficient is less than or equal to a preset threshold, the multi-variable collaborative control model then prioritizes protection based on the preset threshold. The weight.

8. The production control method for producing sodium perchlorate by electrolysis according to claim 3, characterized in that, After each production batch is completed within a preset time period, the difference between the actual and predicted sodium perchlorate yield is input into the yield-energy consumption dual-objective prediction model to update the attention weights of the attention mechanism-BP neural network using the gradient descent method. .

Images