Compressed air dynamic distribution control method and system based on electrolytic aluminum technological process

Abstract

The invention relates to the technical field of electrolytic aluminum intelligent control, and discloses a compressed air dynamic distribution control method and system based on an electrolytic aluminum technological process, and the method comprises the steps: obtaining electrolytic aluminum technological parameters, air compressor group operation data and historical compressed air demand data; predicting the initial demand quantity of compressed air; dynamically adjusting the initial demand quantity to obtain a target demand quantity; constructing a multi-target optimization control model of the air compressor group, and solving by utilizing a heuristic hierarchical search algorithm based on dynamic priority to determine target control parameters of the air compressor group; and the target demand quantity of compressed air in the electrolytic aluminum technological process is bound with the target control parameters of the air compressor group, a dynamic control strategy is generated in combination with a priority rule, and dynamic distribution control over the compressed air is achieved according to the dynamic control strategy. According to the invention, the accuracy and economical efficiency of compressed air distribution and the system stability in the electrolytic aluminum production process are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent control of electrolytic aluminum, in particular to a compressed air dynamic allocation control method and system based on an electrolytic aluminum process. BACKGROUND

[0002] With the development and expansion of electrolytic aluminum enterprises, its compressed air system has undergone several major upgrades: first, from user-side self-management to unified management, focusing on single-user demand, building air compression station, integrating and replacing higher-automation and more energy-efficient equipment to meet user demand; second, building a data acquisition and monitoring control (SCADA) system for each air compression station; the third step is to build a compressed air interconnection ring network between air compression stations, which further improves the efficiency of the compressed air system based on the overall air demand and from point to surface under the premise of meeting user air demand.

[0003] Since these upgrades and modifications are mainly based on hardware, the system's operation mode has not fundamentally changed, so there are the following pain points:

[0004] Dependence on people: manual production matching, slow response. Mainly relying on manual matching of production and production pressure, manually starting and stopping different types of units according to experience to meet production requirements, the system response is not timely.

[0005] Waste: complex air pipe network, many air points, and large pressure fluctuation range. To ensure production requirements, the system pipe network pressure must be increased, resulting in excessive pressure demand (system redundancy), and to ensure supply, more equipment must be opened to ensure pipe network pressure.

[0006] Difficult to manage: improving management requires accurate collection of energy consumption data and analysis of reasons, as well as real-time monitoring of user air consumption to help improve control levels. The existing dual-system data is not connected, manual control has reached its limit, and more advanced technologies such as industrial Internet of Things and artificial intelligence are needed to help manage production more safely and efficiently.

[0007] Control is complicated: the existing control system and energy metering system are separate, with some data at the user end and some data at the air compression station, many devices, many brands, difficult to operate, and difficult to control in a unified cluster.

[0008] The automation application of electrolytic aluminum enterprises is very extensive. Automation systems are generally divided into several levels. The most common approach is to divide them into five levels, called L1-L5. The higher the level, the larger the management scope. L1 systems are generally aimed at valve levels, L2 systems are device levels (such as continuous casting machines, rolling mills), L3 systems are workshop levels, L4 systems are company levels, and L5 systems are strategic levels.

[0009] In a sense, the role of these computer systems is to facilitate the integration of information perception, decision-making, and execution. This is in line with the original idea of the founder of cybernetics, Norbert Wiener. However, only L1 and L2 systems are called control computers or automation systems, with a relatively large proportion of computer autonomous decision-making; while L3 and L4 are called management computers or information systems, which mainly rely on human decision-making. There is an important change in the decision-making mechanism, which is essentially caused by the change in the scale of control. Generally speaking, automation systems are suitable for small-scale systems.

[0010] In reality, measures for automation systems often switch to manual mode or shut down when a fault occurs, and are handled by humans. In other words, automation is generally only suitable for objects with stable systems. For "small-scale" systems, "system normal" is a normal state, which facilitates the development of contingency plans and the use of automation technology.

[0011] However, for "large-scale" objects such as workshops and factories, the complexity of the system increases dramatically. A problem in a subsystem can cause a failure in the entire system. In particular, for factory-level systems, equipment failure, production connection, non-standard operation, and quality abnormalities often occur. When these problems occur, the entire system may not have a contingency plan, and often requires human intervention. One of the roles of information systems is to help humans make such decisions.

[0012] From the perspective of traditional automation, the compressed air system has basically achieved L1 and L2 automation levels, and L3 and above levels are still at the data acquisition level. Due to the limited computing power and architecture of existing PLCs and industrial computers, it is difficult to further increase the scope of computer control and achieve computer "cognition" of the physical world, let alone prediction and intelligent control. However, driven by industrial technology, intelligent decision-making and control can be achieved, automatic supply and demand matching can be completed according to production needs, real-time analysis, judgment, and adjustment can be made, and autonomous operation, intelligent control, improved fine management efficiency, and reduced energy waste can be achieved.

[0013] Currently, many processes in electrolytic aluminum plants and workshop compressed air demand are unstable, and this demand is often related to the production of electrolytic aluminum processes, which leads to a jump in flow changes. However, the PID control of conventional air compressor groups can only use single-point tracking control mode, and cannot achieve global optimal control based on electrolytic aluminum process data. Compressed air can significantly affect the operation of electrolytic cells, ultimately leading to energy waste and even production accidents.

[0014] To address the problems in the related art, no effective solutions have been proposed so far. SUMMARY

[0015] In view of the deficiencies of the prior art, the application provides a compressed air dynamic distribution control method and system based on an electrolytic aluminum process, which has the advantages of comprehensively improving the accuracy, economy and system stability of compressed air distribution in the electrolytic aluminum production process, thereby solving the problems in the background art.

[0016] To achieve the above-mentioned advantages of comprehensively improving the accuracy, economy and system stability of compressed air distribution in the electrolytic aluminum production process, the application adopts the following specific technical solutions:

[0017] According to one aspect of the application, a compressed air dynamic distribution control method based on an electrolytic aluminum process is provided, comprising the following steps:

[0018] S1, obtaining electrolytic aluminum process parameters, air compressor group operation data and historical compressed air demand data and preprocessing; based on a pre-constructed compressed air demand prediction model, predicting the initial demand of compressed air;

[0019] S2, based on the dynamic mapping relationship between the electrolytic cell heat balance and the compressed air consumption, dynamically adjusting the predicted initial demand of compressed air to obtain the target demand of compressed air in the electrolytic aluminum process;

[0020] S3, taking the minimization of total energy consumption, stable gas supply pressure and matching dynamic flow demand as the optimization target, constructing a multi-objective optimization control model of the air compressor group, and using a heuristic hierarchical search algorithm based on dynamic priority to solve, to determine the target control parameters of the air compressor group;

[0021] S4, binding the target demand of compressed air in the electrolytic aluminum process with the target control parameters of the air compressor group, generating a dynamic control strategy combining priority rules, and realizing dynamic distribution control of compressed air according to the dynamic control strategy.

[0022] Further, the obtaining of electrolytic aluminum process parameters, air compressor group operation data and historical compressed air demand data and preprocessing; based on a pre-constructed compressed air demand prediction model, predicting the initial demand of compressed air comprises the following steps:

[0023] S11, obtaining the current, voltage, temperature, alumina feeding amount, electrolytic cell age, cell surface temperature field distribution and furnace wall thickness data of the electrolytic aluminum production line, constructing a process parameter set; obtaining the output pressure, flow, energy consumption, start-stop state and running time data of each air compressor in the electrolytic aluminum production line, constructing an operation parameter set;

[0024] S12, extract the time series data of the total compressed air demand in the past preset time period, correlate the process parameters and air compressor operation parameters corresponding to the period, form a historical data set based on demand, process and equipment, and preprocess the data in the historical data set;

[0025] S13, based on the preprocessed historical data set, train the pre-constructed compressed air demand prediction model, and use the trained compressed air demand prediction model to output the initial demand of compressed air corresponding to the real-time electrolytic aluminum process parameters and air compressor group operation data.

[0026] Further, based on the dynamic mapping relationship between the heat balance of the electrolytic cell and the consumption of compressed air, the initial demand of compressed air is dynamically adjusted to obtain the target demand of compressed air in the electrolytic aluminum process, which includes the following steps:

[0027] S21, identify the high-temperature electrolyte area and low-temperature furnace wall area according to the temperature field distribution data of the cell surface, and determine the temperature gradient and area ratio; calculate the furnace wall thickness change rate based on the furnace wall thickness data, and combine the voltage fluctuation to construct the furnace wall stability index; calculate the heat balance gas consumption correction factor according to the temperature gradient, the furnace wall thickness change rate and the furnace wall stability index;

[0028] S22, construct a double-path correction model, determine the mechanism path correction amount and the data path correction amount using the double-path correction model, and determine the target correction amount combined with the fusion mechanism;

[0029] S23, according to the comparison result of the heat balance gas consumption correction factor and the preset threshold value, the real-time heat balance state of the electrolytic cell is divided into normal heat balance stage and heat imbalance stage, and the demand of compressed air of each electrolytic cell in the initial demand in the normal heat balance stage and the heat imbalance stage is corrected respectively by using the heat balance gas consumption correction factor and the target correction amount;

[0030] S24, use clustering algorithm to divide the type of electrolytic cell, and determine the demand of compressed air of all electrolytic cells in the workshop by weighted summation method to obtain the target demand of compressed air in the electrolytic aluminum process.

[0031] Further, the calculation formula of the heat balance gas consumption correction factor is:

[0032]

[0033] In the formula, γ(t) represents the heat balance gas consumption correction factor at time t, ReLU represents the activation function, and α and β represent two process weights respectively, represents the temperature gradient at time t, T thresrepresents a thermal imbalance temperature threshold, HIS(t) represents a furnace side stability index at time t, γ0 represents a reference correction coefficient, represents a furnace side thickness change rate at time t, d nominal represents a rated value of the furnace side thickness, U(t) represents a cell voltage at time t, η represents a voltage sensitivity coefficient, σ represents a voltage fluctuation standard deviation, U nominal represents a rated value of the cell voltage.

[0035] Further, the double-path correction model is constructed, the mechanism path correction quantity and the data path correction quantity are determined by using the double-path correction model, and the target correction quantity is determined in combination with a fusion mechanism, and the method comprises the following steps:

[0036] S221, based on an alumina dissolution kinetics equation, an additional purge requirement at the time of thermal imbalance is determined to obtain a mechanism path correction quantity;

[0037] S222, a mapping of a historical thermal balance state and a gas consumption deviation is learned by using a gated recurrent unit to obtain a data path correction quantity;

[0038] S223, based on the mechanism path correction quantity and the data path correction quantity, a target correction quantity is determined in combination with a fusion mechanism;

[0039] The calculation formula of the mechanism path correction quantity is:

[0040]

[0041] The calculation formula of the data path correction quantity is:

[0042] Q data (t) = GRU(γ(t), γ(t-1),..., Q error (t-1))

[0043] The calculation formula of the target correction quantity is:

[0044] Q adj (t) = Q mech (t) · (1 + λ · Q data (t))

[0045] In the formula, Q mech (t) represents a mechanism path correction quantity at time t, k represents a process constant reflecting the coupling relationship between current efficiency and thermal imbalance, γ(τ) represents a thermal balance gas consumption correction factor at time τ, I(τ) represents a cell current at time τ, Q data (t) represents a data path correction quantity at time t, GRU represents a gated recurrent unit, Q error (t-1) represents a historical prediction error at time t-1, and Q adj(t) represents the target correction amount at time t, and λ represents the data path weight.

[0046] Further, the real-time thermal equilibrium state of the electrolytic cell is divided into a normal thermal equilibrium stage and a thermal imbalance stage according to a comparison result of the thermal equilibrium gas consumption correction factor and the preset threshold value, and the demand amount of compressed air of the single electrolytic cell in the initial demand amount in the normal thermal equilibrium stage and the thermal imbalance stage is respectively corrected by using the thermal equilibrium gas consumption correction factor and the target correction amount, which includes the following steps:

[0047] S231, dividing the real-time thermal equilibrium state of the electrolytic cell into a normal thermal equilibrium stage and a thermal imbalance stage according to a comparison result of the thermal equilibrium gas consumption correction factor and the preset thermal equilibrium gas consumption correction factor threshold value;

[0048] S232, correcting the demand amount of compressed air of the single electrolytic cell in the initial demand amount in the normal thermal equilibrium stage by using the thermal equilibrium gas consumption correction factor;

[0049] S233, correcting the demand amount of compressed air of the single electrolytic cell in the initial demand amount in the thermal imbalance stage by using the target correction amount;

[0050] Wherein, in the normal thermal equilibrium stage, the calculation formula of the compressed air demand amount is:

[0051] Q target (t) = Q pred (t) · (1 + 0.05 · γ (t))

[0052] In the thermal imbalance stage, the calculation formula of the compressed air demand amount is:

[0053] Q target (t) = Q pred (t) + Q adj (t)

[0054] In the formula, Q target (t) represents the demand amount of compressed air at time t, Q pred (t) represents the initial demand amount of the single electrolytic cell at time t.

[0055] Further, the type of the electrolytic cell is divided by using the clustering algorithm, and the compressed air demand amount of all electrolytic cells in the workshop is determined by using the weighted summation method, so as to obtain the target demand amount of compressed air in the aluminum electrolysis process, which includes the following steps:

[0056] S241, obtaining the thermal equilibrium index, the furnace slope stability index, the temperature gradient and the cell age data of the electrolytic cell, and performing outlier rejection and standardization processing on the obtained data;

[0057] The calculation formula of the thermal equilibrium index is:

[0058]

[0059] In the formula, HE(t) represents the heat balance index at time t, ΔT(t) represents the tank temperature fluctuation value at time t, T nominal represents the rated tank temperature, Δd(t) represents the furnace wall thickness change at time t, d nominal represents the rated furnace wall thickness, and α, β, and ω respectively represent three process weights.

[0060] S242, respectively determine the contribution of the heat balance index, the furnace wall stability index, the temperature gradient, and the tank age to the heat balance state by using the variance analysis method, and determine the key feature vector according to the contribution;

[0061] S243, determine the cluster number and the initial center point by using the elbow rule, and calculate the Euclidean distance between the key feature vector of the electrolytic tank and the initial center point, wherein the cluster number includes three types of heat stable tank, heat sensitive tank, and heat imbalance tank;

[0062] S244, according to the calculation result of the Euclidean distance, the electrolytic tank is assigned to the corresponding category, and the category center point is updated repeatedly until convergence, and the clustering division is completed;

[0063] S245, according to the type and tank age of the electrolytic tank, a correction coefficient is determined, and the compressed air demand of all electrolytic tanks in the workshop is determined by using the weighted summation method, and the target demand of compressed air in the aluminum electrolysis process is obtained.

[0064] Further, the multi-objective optimization control model of the air compressor group is constructed with the optimization objectives of minimizing total energy consumption, stabilizing gas supply pressure, and matching dynamic flow demand, and the heuristic hierarchical search algorithm based on dynamic priority is used to solve, and the target control parameters of the air compressor group include the following steps:

[0065] S31, a multi-objective optimization control model of the air compressor group is constructed with the optimization objectives of minimizing total energy consumption, stabilizing gas supply pressure, and matching dynamic flow demand, and the air compressor start-stop state and the running pressure set value are used as decision variables;

[0066] S32, the heuristic hierarchical search algorithm based on dynamic priority is used to solve the multi-objective optimization control model of the air compressor group, and the target control parameters of the air compressor group are determined.

[0067] Further, the heuristic hierarchical search algorithm based on dynamic priority is used to solve the multi-objective optimization control model of the air compressor group, and the target control parameters of the air compressor group include:

[0068] According to the pressure range of the aluminum electrolysis process and the target demand of compressed air, the air compressor combination is quickly screened to ensure that the rated flow sum of the air compressor combination meets the process demand;

[0069] With the minimization of total energy consumption as the core, global optimization is carried out in combination with a tabu search algorithm, local search is carried out by using a simulated annealing algorithm, and a running time balancing mechanism is used to generate a shutdown scheme of high-load equipment preferentially, so as to balance multi-objective optimization;

[0070] When the standard deviation of pipe network pressure fluctuation continuously exceeds a preset value, the current optimization is immediately interrupted, the priority of the stability of the pressure is raised, and a solution set with stability as the core is regenerated; when the continuous running time of the air compressor exceeds a maximum allowable time threshold, a candidate scheme containing only the shutdown of the equipment and meeting the total flow demand is generated;

[0071] The start-stop state and the running pressure set value of each air compressor are determined by comprehensively searching and adjusting the results, and a target control parameter combination of the air compressor group is obtained.

[0072] According to another aspect of the present application, a compressed air dynamic allocation control system based on an electrolytic aluminum process is provided, comprising a demand prediction module, a demand optimization module, a control parameter determination module and a dynamic allocation control module;

[0073] The demand prediction module is used to obtain electrolytic aluminum process parameters, air compressor group running data and historical compressed air demand data and perform preprocessing; based on a pre-constructed compressed air demand prediction model, the initial demand of compressed air is predicted.

[0074] The demand optimization module is used to dynamically adjust the predicted initial demand of compressed air based on the dynamic mapping relationship between the heat balance of the electrolytic cell and the compressed air consumption, to obtain the target demand of compressed air in the electrolytic aluminum process;

[0075] The control parameter determination module is used to construct a multi-objective optimization control model of the air compressor group with the optimization objectives of minimizing total energy consumption, stabilizing the supply pressure and matching the dynamic flow demand, and to determine the target control parameters of the air compressor group by using a heuristic hierarchical search algorithm based on dynamic priority.

[0076] The dynamic allocation control module is used to bind the target demand of compressed air in the electrolytic aluminum process with the target control parameters of the air compressor group, generate a dynamic control strategy in combination with priority rules, and realize dynamic allocation control of compressed air according to the dynamic control strategy.

[0077] Compared with the prior art, the present application provides a compressed air dynamic allocation control method and system based on an electrolytic aluminum process, which has the following beneficial effects:

[0078] (1) The present application is based on machine learning and multi-objective optimization technology, deeply integrates the electrolytic cell heat balance state and compressed air demand, and builds a complete technical system of data collection and analysis, demand accurate prediction, dynamic optimization control and feedback iteration upgrade. Compared with the traditional technical scheme, the present application significantly improves the accuracy of compressed air demand prediction, greatly reduces energy consumption, effectively stabilizes the pipe network pressure, reduces the abnormal situation of electrolytic cell caused by heat imbalance, prolongs the service life of air compressor equipment, and comprehensively improves the accuracy, economy and system stability of compressed air distribution in the electrolytic aluminum production process.

[0079] (2) The present application collects electrolytic cell heat balance related parameters, combines advanced prediction model and feature engineering, and comprehensively captures the influence of electrolytic cell heat state on gas consumption. Compared with the traditional method, the accuracy of initial demand prediction of compressed air is significantly improved, especially in complex working conditions, which lays a solid foundation for subsequent dynamic adjustment.

[0080] (3) The present application can dynamically adjust the demand for compressed air according to the heat balance state of the electrolytic cell through the double-path correction model and the stage-by-stage correction strategy, realize the accurate compensation of gas demand under the heat imbalance state, and effectively avoid the problem of uneven gas supply by using the multi-cell collaborative clustering mechanism, reduce invalid blow-by gas consumption, and optimize the supply and demand balance of compressed air in the whole workshop.

[0081] (4) The present application is based on a heuristic hierarchical search algorithm based on dynamic priority, which takes into account multi-objective optimization such as energy consumption and pressure stability. Under the premise of ensuring the feasibility of gas supply, it effectively reduces the energy consumption of the air compressor group, significantly reduces the pipe network pressure fluctuation, and based on the dynamic priority adjustment mechanism, it can quickly respond to pressure abnormalities, equipment overload and other sudden situations, greatly improving the stability and reliability of the system under complex working conditions. Through real-time binding of target demand and control parameters, combined with priority scheduling rules, efficient matching of compressed air flow is realized.

[0082] (5) The present application not only can balance each control link in the air compression station, monitor and control the air compressor and each related equipment in the station, but also can based on the continuous pressure feedback of the air compressor, gas terminal and pipe network multiple pressure sensors, real-time continuous control of the entire compressed air system to balance the supply and demand of compressed air. In addition, based on the continuous collection of air compression system operation data on site, the running parameters of each air compressor are optimized in real time to ensure the best energy efficiency of the system and ensure the safety of the system. BRIEF DESCRIPTION OF DRAWINGS

[0083] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed in the embodiments. Obviously, the drawings in the following description only constitute some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained from these drawings without creative labor.

[0084] Figure 1 is a flow chart of a compressed air dynamic allocation control method based on an electrolytic aluminum process according to an embodiment of the present application;

[0085] Figure 2 is a principle schematic diagram of a compressed air dynamic allocation control method based on an electrolytic aluminum process according to an embodiment of the present application. DETAILED DESCRIPTION

[0086] In order to further illustrate the embodiments, the present application provides drawings which are part of the disclosure of the present application, mainly used to illustrate the embodiments, and can be used to explain the operating principle of the embodiments in conjunction with the related description of the specification. Those skilled in the art should be able to understand other possible embodiments and advantages of the present application by referring to these contents. The components in the drawings are not drawn to scale, and similar component symbols are generally used to represent similar components.

[0087] According to an embodiment of the present application, a compressed air dynamic allocation control method and system based on an electrolytic aluminum process are provided.

[0088] The present application will be further described in conjunction with the drawings and specific embodiments. As shown in Figures 1-2 According to an embodiment of the present application, a compressed air dynamic allocation control method based on an electrolytic aluminum process is provided, including the following steps:

[0089] S1, obtaining electrolytic aluminum process parameters, air compressor group operation data and historical compressed air demand data and preprocessing; based on a pre-constructed compressed air demand prediction model, predicting the initial demand of compressed air;

[0090] Among them, the step of obtaining electrolytic aluminum process parameters, air compressor group operation data and historical compressed air demand data and preprocessing; based on a pre-constructed compressed air demand prediction model, predicting the initial demand of compressed air includes the following steps:

[0091] S11, obtaining current, voltage, temperature, alumina feed quantity, electrolytic cell age, cell surface temperature field distribution and furnace wall thickness and other data of the electrolytic aluminum production line, constructing a process parameter set; obtaining the output pressure, flow, energy consumption, start-stop state and running time of each air compressor in the electrolytic aluminum production line, constructing an operation parameter set;

[0092] S12, extract the time series data of the total compressed air demand in the past preset time period from the industrial database, associate the process parameters and air compressor operation parameters corresponding to the period, form a historical data set based on demand, process and equipment, and preprocess the data in the historical data set;

[0093] The preprocessing includes: data cleaning (i.e. removing outliers such as sensor failure data, and filling missing values such as using linear interpolation or process parameter association), time alignment (i.e. aligning multi-source heterogeneous data according to a unified timestamp, time resolution ≤ 1 second), normalization processing (i.e. standardizing process parameters and operation data such as z-score, eliminating dimension difference);

[0094] S13, based on the preprocessed historical data set, train the pre-constructed compressed air demand prediction model, and use the trained compressed air demand prediction model to output the initial demand of compressed air corresponding to the real-time electrolytic aluminum process parameters and air compressor group operation data;

[0095] Specifically, the input features of the compressed air demand prediction model include: electrolytic aluminum process parameters (such as current, voltage, temperature, alumina feed quantity, electrolytic cell age, cell surface temperature field distribution and furnace side thickness, etc.), time features (hours, weekdays / holidays), historical demand sequence; The output data is the compressed air demand at the future k time points.

[0096] The model architecture adopts a LSTM-Transformer hybrid network:

[0097] LSTM (Long Short-Term Memory, i.e. Long Short-Term Memory Network) layer: capture short-term demand fluctuations (window ≤ 2h), memory unit stores the time sequence dependence of process parameter changes;

[0098] Transformer layer: process long-period patterns (such as weekly load fluctuations), self-attention mechanism focuses on key process parameters (such as current mutations during electrode replacement period);

[0099] The loss function adopts MAE (Mean Absolute Error) and RMSE (Root Mean Square Error) joint optimization.

[0100] The above steps comprehensively collect electrolytic aluminum process parameters, air compressor operation data and historical demand data, and after preprocessing such as cleaning, alignment and normalization, use the compressed air demand prediction model constructed by the LSTM-Transformer hybrid network to effectively capture short-term demand fluctuations and long-period load patterns, realize accurate prediction of compressed air demand at future k time points, and provide reliable initial demand data support for subsequent dynamic adjustment and optimization control.

[0101] S2, dynamically adjusting the initial demand of compressed air according to the dynamic mapping relationship between the heat balance of the electrolytic cell and the consumption of compressed air, to obtain the target demand of compressed air in the aluminum electrolysis process;

[0102] The dynamic adjustment of the initial demand of compressed air according to the dynamic mapping relationship between the heat balance of the electrolytic cell and the consumption of compressed air to obtain the target demand of compressed air in the aluminum electrolysis process includes the following steps:

[0103] S21, identifying the high-temperature electrolyte area (temperature greater than 950℃) and the low-temperature furnace side area (temperature less than 850℃) according to the cell surface temperature field distribution data, and determining the temperature gradient and area ratio; Based on the thickness data of the furnace side, the thickness change rate of the furnace side is calculated, and the stability index of the furnace side is constructed combined with the voltage fluctuation; According to the temperature gradient, the thickness change rate of the furnace side and the stability index of the furnace side, the heat balance gas consumption correction factor is calculated;

[0104] Specifically, the calculation formula of the heat balance gas consumption correction factor is:

[0105]

[0106] In the formula, γ(t) represents the heat balance gas consumption correction factor at time t, ReLU represents the activation function, and α and β represent two process weights, respectively, represents the temperature gradient at time t, T thres represents the heat imbalance temperature threshold, HIS(t) represents the stability index of the furnace side at time t, γ0 represents the reference correction coefficient, represents the thickness change rate of the furnace side at time t, d nominal represents the rated value of the furnace side thickness, U(t) represents the voltage of the electrolytic cell at time t, η represents the voltage sensitivity coefficient, σ represents the voltage fluctuation standard deviation, U nominal represents the rated value of the electrolytic cell voltage.

[0108] S22, constructing a double-path correction model, determining the mechanism path correction amount and the data path correction amount by using the double-path correction model, and determining the target correction amount combined with the fusion mechanism;

[0109] Specifically, the double-path correction model is constructed, the mechanism path correction amount and the data path correction amount are determined by using the double-path correction model, and the target correction amount is determined combined with the fusion mechanism. The following steps include:

[0110] S221, based on the alumina dissolution kinetics equation, the additional purge demand when the heat is unbalanced is determined to obtain the mechanism path correction amount;

[0111] The calculation formula of the mechanism path correction amount is:

[0112]

[0113] wherein Q mech (t) represents the mechanism path correction amount at time t, k represents a process constant reflecting the coupling relationship between current efficiency and thermal imbalance, γ(τ) represents a thermal equilibrium gas consumption correction factor at time τ, and I(τ) represents the electrolytic cell current at time τ;

[0114] S222, learning the mapping of historical thermal equilibrium states and gas consumption deviations by using a gated recurrent unit to obtain a data path correction amount;

[0115] The calculation formula of the data path correction amount is:

[0116] Q data (t) = GRU(γ(t), γ(t-1),..., Q error (t-1))

[0117] wherein Q data (t) represents the data path correction amount at time t, GRU represents a gated recurrent unit, and Q error (t-1) represents the historical prediction error at time t-1;

[0118] S223, determining a target correction amount based on the mechanism path correction amount and the data path correction amount in combination with a fusion mechanism;

[0119] The calculation formula of the target correction amount is:

[0120] Q adj (t) = Q mech (t) · (1 + λ · Q data (t))

[0121] Setting a lower limit of thermal imbalance gas consumption, when γ(τ) > 0.3, a minimum compensation amount is forcibly triggered Avoiding insufficient adjustment of the model due to data loss and ensuring process safety:

[0122]

[0123] wherein Q data (t) represents the data path correction amount at time t, GRU represents a gated recurrent unit, and Q error (t-1) represents the historical prediction error at time t-1, and Q adj (t) represents the target correction amount at time t, and λ represents a data path weight.

[0124] S23, according to the comparison result of the heat balance gas consumption correction factor and the preset threshold value, the real-time heat balance state of the electrolytic cell is divided into a normal heat balance stage and a heat imbalance stage, and the initial demand amount of compressed air of a single electrolytic cell in the normal heat balance stage and the heat imbalance stage is corrected respectively by using the heat balance gas consumption correction factor and the target correction amount;

[0125] Specifically, the real-time heat balance state of the electrolytic cell is divided into a normal heat balance stage and a heat imbalance stage according to the comparison result of the heat balance gas consumption correction factor and the preset threshold value, and the initial demand amount of compressed air of a single electrolytic cell in the normal heat balance stage and the heat imbalance stage is corrected respectively by using the heat balance gas consumption correction factor and the target correction amount, which includes the following steps:

[0126] S231, according to the comparison result of the heat balance gas consumption correction factor and the preset heat balance gas consumption correction factor threshold value, the real-time heat balance state of the electrolytic cell is divided into a normal heat balance stage and a heat imbalance stage;

[0127] When γ(τ)≤0.2, the real-time heat balance state of the electrolytic cell is divided into a normal heat balance stage, and when γ(τ)>0.2, the real-time heat balance state of the electrolytic cell is divided into a heat imbalance stage;

[0128] S232, the initial demand amount of compressed air of a single electrolytic cell in the normal heat balance stage is corrected by using the heat balance gas consumption correction factor;

[0129] In the normal heat balance stage (small correction, to prevent over-adjustment), the calculation formula of the compressed air demand amount is:

[0130] Q target (t)=Q pred (t)·(1+0.05·γ(t))

[0131] In the formula, Q target (t) represents the demand amount of compressed air at time t, and Q pred (t) represents the initial demand amount of a single electrolytic cell at time t;

[0132] S233, the initial demand amount of compressed air of a single electrolytic cell in the heat imbalance stage is corrected by using the target correction amount;

[0133] In the heat imbalance stage (superposition mechanism-data fusion correction amount, to respond to the increase of gas consumption caused by heat imbalance), the calculation formula of the compressed air demand amount is:

[0134] Q target (t)=Q pred (t)+Q adj (t)。

[0135] S24, using a clustering algorithm to divide the type of electrolytic cell, and determining the compressed air demand of all electrolytic cells in the workshop by weighted summation method, to obtain the target demand of compressed air in the aluminum electrolysis process;

[0136] Specifically, the use of clustering algorithm to divide the type of electrolytic cell, and determining the compressed air demand of all electrolytic cells in the workshop by weighted summation method, to obtain the target demand of compressed air in the aluminum electrolysis process includes the following steps:

[0137] S241, obtaining the heat balance index, the stability index of the furnace slope, the temperature gradient and the cell age data of the electrolytic cell, and performing outlier rejection and standardization processing on the obtained data;

[0138] The calculation formula of the heat balance index is:

[0139]

[0140] In the formula, HE(t) represents the heat balance index at time t, ΔT(t) represents the cell temperature fluctuation value at time t, T nominal represents the rated cell temperature, Δd(t) represents the change amount of the furnace slope thickness at time t, d nominal represents the rated furnace slope thickness, and α, β and ω represent three process weights, respectively;

[0141] S242, using variance analysis method to determine the contribution degree of heat balance index, furnace slope stability index, temperature gradient and cell age to heat balance state respectively, and determining the key feature vector according to the contribution degree;

[0142] In this embodiment, the contribution degree of the heat balance index is ≥40%, which is the core index, the contribution degree of the furnace slope stability index is ≥30%, the contribution degree of the temperature gradient is ≥20%, and the contribution degree of the cell age is ≤10%, which is an auxiliary index;

[0143] In addition, when the feature dimension is ≥5, the first three principal components are extracted by principal component analysis (PCA), the cumulative variance contribution rate is ≥85%, and the calculation complexity is reduced;

[0144] S243, using elbow rule to determine the clustering number and initial center point, and calculating the Euclidean distance between the key feature vector of the electrolytic cell and the initial center point, wherein the clustering number includes three types of heat stable cell, heat sensitive cell and heat imbalance cell;

[0145] Specifically, the optimal clustering number is determined as follows:

[0146] The Silhouette Score of different K values is calculated by the Elbow Method: draw the "K value-Silhouette Score" curve, and select K=3 corresponding to the peak of the Silhouette Score (three types of thermal stability, thermal sensitivity, and thermal imbalance);

[0147] The clustering parameters are set as follows:

[0148] The initial center points are selected: thermal stability tank (thermal balance index 0.1, furnace slope stability index 0.2, temperature gradient 10℃ / m), thermal sensitive tank (thermal balance index 0.25, furnace slope stability index 0.4, temperature gradient 30℃ / m), thermal imbalance tank (thermal balance index 0.4, furnace slope stability index 0.6, temperature gradient 50℃ / m); the iteration termination condition is that the center point update amplitude is less than 0.01 or the iteration number is greater than or equal to 50 times;

[0149] S244, according to the calculation results of the Euclidean distance, the electrolytic tank is assigned to the corresponding category, and the category center point is updated repeatedly until convergence, and the clustering division is completed;

[0150] S245, according to the type and tank age of the electrolytic tank, the correction coefficient is determined, and the weighted summation method is used to determine the compressed air demand of all electrolytic tanks in the workshop, and the target demand of compressed air in the aluminum electrolysis process is obtained;

[0151] Specifically, the electrolytic tank type division and the reference correction coefficient are as follows:

[0152] The electrolytic tank is divided into three categories by K-means clustering algorithm, and the reference correction coefficient is set according to the type and tank age of the electrolytic tank:

[0153] The reference correction coefficient of the thermal stability tank is 1.0 (representing the standard gas supply demand);

[0154] The reference correction coefficient of the thermal sensitive tank is 1.2 (additional 20% gas supply is needed to maintain thermal balance);

[0155] The reference correction coefficient of the thermal imbalance tank is 1.5 (additional 50% gas supply is needed for thermal balance adjustment);

[0156] The incremental adjustment of tank age to correction coefficient is as follows:

[0157] In order to reflect the change of heat dissipation efficiency caused by equipment aging, the correction coefficient of different types of electrolytic tank is increased by section according to tank age:

[0158] Thermal stability tank: tank age ≤12 months, correction coefficient=1.0 (reference value); 12 months< tank age ≤24 months, correction coefficient=1.0×1.05 (increased by 5%); tank age> 24 months, correction coefficient=1.0×1.1 (increased by 10%);

[0159] Thermal sensitive cell: tank age ≤ 12 months, correction factor = 1.2 (reference value); 12 months < tank age ≤ 24 months, correction factor = 1.2 x 1.1 (increased by 10%); tank age > 24 months, correction factor = 1.2 x 1.2 (increased by 20%);

[0160] Thermal imbalance cell: tank age ≤ 12 months, correction factor = 1.5 (reference value); 12 months < tank age ≤ 24 months, correction factor = 1.5 x 1.15 (increased by 15%); tank age > 24 months, correction factor = 1.5 x 1.3 (increased by 30%);

[0161] The example calculation of the comprehensive correction factor is as follows: assuming that electrolytic cell A belongs to a thermal sensitive cell, the tank age is 18 months, and the correction factor calculation thereof = 1.2 x 1.1 = 1.32.

[0162] The total demand of the workshop is calculated as follows: if there are 10 electrolytic cells in the workshop, of which 4 are thermal stable cells (tank ages are 6, 15, 20, and 30 months respectively), 3 are thermal sensitive cells (tank ages are 8, 18, and 25 months respectively), and 3 are thermal imbalance cells (tank ages are 10, 16, and 28 months respectively), and the corrected demand of each cell is Q1, Q2, …, Q 10 , respectively, then the total demand of the workshop is (Q i x correction factor).

[0163] In addition, the embodiment also includes defining a thermal balance correction effectiveness index, evaluating the correction effect by comparing the prediction errors before and after correction, and triggering the correction model retraining when the index at consecutive multiple time points is not ideal; a dynamic parameter updating mechanism combining short-term and long-term is adopted, the short-term adjusts the data path weight based on real-time error, and the long-term recalibrates the mechanism path parameters according to the whole workshop thermal balance-air consumption data, to adapt to seasonal changes and equipment aging.

[0164] The above steps realize the precise dynamic adjustment of the initial demand of compressed air by constructing the dynamic mapping relationship between the thermal balance of the electrolytic cell and the consumption of compressed air, using the thermal balance air consumption correction factor, the double path correction model, and the multi-cell collaborative clustering mechanism, not only effectively identifying the thermal imbalance state and providing targeted compensation, but also optimizing the overall gas distribution of the workshop through the type-tank age correction coefficient system, and the introduced effectiveness evaluation and dynamic parameter updating mechanism ensures the long-term stability of the correction effect, significantly improving the matching degree of compressed air distribution and electrolytic process demand.

[0165] S3, a multi-objective optimization control model of the air compressor group is constructed with the optimization objectives of minimizing the total energy consumption, stabilizing the gas supply pressure, and matching the dynamic flow demand, and a heuristic hierarchical search algorithm based on dynamic priority is used to solve the model to determine the target control parameters of the air compressor group;

[0166] Wherein, the air compressor group multi-objective optimization control model is constructed with the optimization objectives of minimizing total energy consumption, stabilizing gas supply pressure and matching dynamic flow demand, and a heuristic hierarchical search algorithm based on dynamic priority is used for solving to determine the target control parameters of the air compressor group, including the following steps:

[0167] S31, with the optimization objectives of minimizing total energy consumption, stabilizing gas supply pressure and matching dynamic flow demand, the air compressor group multi-objective optimization control model is constructed with the start-stop state and operating pressure set value of the air compressor as the decision variable;

[0168] Specifically, the total energy consumption is minimized: Wherein, P i represents the power of the i-th air compressor, N represents the number of air compressors, η i represents the efficiency coefficient, S i represents the start-stop times, C start represents the single start-stop cost;

[0169] Stabilizing gas supply pressure: min |P actual -P target |, wherein P actual represents the actual pressure of the pipe network, P target represents the target pressure, which is related to the electrolytic aluminum process, and the required pressure and flow of each process are shown in the following table:

[0170] Table 1 Required pressure and flow of each process

[0171]

[0172]

[0173] Matching dynamic flow demand: Wherein, Q i represents the flow output capacity of the i-th air compressor in state S i , Q process (t) represents the process flow demand at time t, i.e. the target demand obtained after correction;

[0174] Device running stability: when switching between adjacent process stages, the running device is preferentially reused; when it is detected that it is in the transition stage of the electrolytic aluminum process, the transition is carried out according to the stepped flow scheme to ensure smooth change of flow and improve the service life of the device;

[0175] Decision variable: start-stop state U i of each air compressor ∈ {0, 1} (0-off, 1-running), operating pressure set value P i (continuous variable, needs to satisfy P min <P i <Pmax );

[0176] S32, a heuristic hierarchical search algorithm based on dynamic priority is used to solve the multi-objective optimization control model of the air compressor group, and target control parameters of the air compressor group are determined.

[0177] Specifically, the heuristic hierarchical search algorithm based on dynamic priority is used to solve the multi-objective optimization control model of the air compressor group, and target control parameters of the air compressor group are determined, including:

[0178] According to the pressure range of the electrolytic aluminum process and the target demand of compressed air, the air compressor combination is quickly screened to ensure that the sum of the rated flow of the air compressor combination meets the process demand;

[0179] Taking the minimization of total energy consumption as the core, a global optimization is performed in combination with a tabu search algorithm, a local search is performed by using a simulated annealing algorithm, and a shutdown scheme of a high-load device is preferentially generated through a running time balancing mechanism to balance multi-objective optimization.

[0180] When the standard deviation of the pipe network pressure fluctuation continuously exceeds the preset value, the current optimization is immediately interrupted, the priority of the stability of the pressure is increased, and a solution set with stability as the core is regenerated; when the continuous running time of the air compressor exceeds the maximum allowable time threshold, a candidate scheme containing only the shutdown of the device and meeting the total flow demand is generated.

[0181] The hierarchical search and dynamic adjustment results are combined to determine the start-stop state and running pressure set value of each air compressor, and a target control parameter combination of the air compressor group is obtained.

[0182] Specifically, to adapt to the scenario where discrete and continuous variables coexist, the present application innovatively proposes a heuristic hierarchical search algorithm based on dynamic priority:

[0183] 1) Hierarchical strategy:

[0184] The first layer (hard constraint layer): priority is given to meeting the non-violable constraints (such as the pressure range of different electrolytic aluminum processes, whether the flow meets the maximum continuous running time of the device);

[0185] A fast filtering method is used to eliminate obviously infeasible solutions (air compressor combinations whose pressure set exceeds the intersection range of the current electrolytic aluminum process scenario), that is, the rated flow of each air compressor combination is summed up, and the flow demand can be met, and the rated exhaust pressure of the air compressor can meet the pressure demand range of the process stage.

[0186] Second layer (core target layer): Optimize total energy consumption, combined with Tabu Search to avoid repeated search for inefficient solutions. Start-stop switching: Preferentially select air compressors with running time close to half the average service life of the device for start-stop state switching. Pressure fine-tuning: Small adjustments to pressure settings within the allowed range (e.g. ±0.05 bar), while controlling the pressure difference between different air compressors. Short-term taboos: Prohibit repeated start-stop or pressure adjustment operations on the same device within 5 steps. Long-term balance: Record the historical switching frequency of the device, and preferentially select devices with low switching frequency for adjustment.

[0187] Third layer (secondary target layer): Fine-tune pressure stability and device life, balance multiple objectives using local search (simulated annealing). Introduce flow demand prediction: According to the process calendar, predict Q process (t) for the next 1 hour. For air compressors with excessively large pressure setting differences, perform forced synchronization (e.g. lock two devices with the largest pressure difference, and adjust their settings to be consistent). Introduce running time equalization mechanism: When the running time of a device is significantly higher than that of other devices, the algorithm preferentially generates solutions that include the shutdown of that device.

[0188] 2) Dynamic priority adjustment:

[0189] A. If the standard deviation of pressure fluctuations exceeds 0.3 bar for 30 consecutive seconds, it is determined to be an abnormal state, and the current optimization process needs to be immediately interrupted, the control parameters are frozen, the weight of the pressure stability target is increased to 3 times the regular value, and a new solution set is generated with stability as the core. For the two air compressors with the largest pressure setting difference, perform forced synchronization to quickly reduce the pipe network pressure fluctuation. When the pipe network pressure fluctuation exceeds the threshold, temporarily increase the priority of the pressure stability target, triggering re-optimization.

[0190] B. If the continuous running time of an air compressor exceeds 85% of its maximum allowed time, a forced shutdown candidate set needs to be generated: only solutions that meet the demand and include the shutdown of that device are retained. At the same time, select the device that has the least impact on the system after shutdown (e.g. the device with the lowest energy consumption change rate) to shut down first. To prevent frequent start-stop, the device that is shut down needs to be added to the short-term operation taboo list. If the continuous running time of an air compressor approaches the upper limit, temporarily insert a device protection layer to force the generation of a shutdown solution.

[0191] 3) Key role:

[0192] Under the premise of meeting dynamic demand, achieve global optimal balance of energy efficiency, stability and device wear. At the same time, compared with traditional MIP / NSGA-II, the calculation speed is improved by 30%-50% (through hierarchical pruning).

[0193] The above steps achieve global optimization of the air compressor group operation by constructing an air compressor group multi-objective optimization control model targeting at minimizing total energy consumption, stabilizing gas supply pressure and matching dynamic flow demand, and solving the model by combining a heuristic hierarchical search algorithm based on dynamic priority: the hard constraint layer ensures gas supply feasibility, the core target layer reduces energy consumption through tabu search, the secondary target layer optimizes pressure stability using simulated annealing, and the dynamic priority adjustment mechanism quickly responds to pressure abnormalities and equipment overload, not only significantly reducing the total energy consumption of the air compressor group, effectively controlling pressure fluctuations in the pipe network, and greatly improving the matching degree of flow demand, but also prolonging the service life of the equipment through the operation time balancing mechanism, and significantly improving the calculation speed compared to traditional algorithms, while meeting the dynamic demand of the aluminum electrolysis process, achieving comprehensive optimization of energy efficiency, stability and equipment wear.

[0194] S4, binding the target demand of compressed air in the aluminum electrolysis process with the target control parameters of the air compressor group, generating a dynamic control strategy combining priority rules, and realizing dynamic allocation control of compressed air according to the dynamic control strategy, specifically including:

[0195] Real-time matching: binding the predicted target demand with the air compressor group control parameters (start-stop state, pressure set value) output by the optimization model;

[0196] Priority rules: high demand transition period (preferentially starting high-efficiency variable-frequency air compressors and gradually loading fixed-speed air compressors), low demand period (turning off redundant air compressors and retaining one variable-frequency machine as a pressure buffer), and command issuance (sending control instructions to air compressor PLCs through Modbus TCP protocol).

[0197] Key role: converting optimization results into executable equipment control instructions to ensure that gas supply pressure fluctuation is ≤±1%.

[0198] In addition, the embodiment also includes the following: first, collect the actual gas supply pressure and flow rate, calculate the prediction error ε = | predicted flow rate - collected flow rate |, then update the prediction model, if the prediction error > threshold value (such as 10%) of the prediction error, trigger the incremental learning mechanism to fine-tune the prediction model parameters with the latest data, at the same time, periodically (such as every week) retrain the optimization model to adapt to factors such as equipment aging and process changes, improve system robustness through closed-loop feedback, and ensure long-term prediction and control accuracy.

[0199] In addition, in order to better understand the above technical solutions of the present application, the embodiment also includes the following specific application scenarios:

[0200] Scenario 1: alumina delivery → electrolytic cell cleaning

[0201] 1) The process characteristics are shown in the following table:

[0202] Table 2 Process Property Comparison

[0203]

[0204] 2) Equipment Reuse Strategy:

[0205] 1. Pressure Compatibility Check: The running 3# air compressor is currently set at 0.55 MPa (satisfies the two-stage intersection 0.4-0.6 MPa) → keep running;

[0206] 2. Flow Matching:

[0207] Alumina Delivery Stage: 3# machine (1.0Q) + 5# machine (0.3Q) running;

[0208] Switch to cleaning stage: calculate the demand flow drops to 0.9Q → only 3# machine (1.0Q) can cover; turn off 5# machine (because of its poor flow regulation ability, prefer to keep variable frequency unit 3#);

[0209] 3. Transition Action: Slowly reduce 3# machine flow to 0.95Q (through variable frequency speed regulation) 5 minutes in advance, and adjust the pressure to 0.5 MPa (satisfies the intermediate value in the cleaning stage); 5# machine stops after completing the current delivery period;

[0210] 4. Effect: Avoid starting a new unit, reduce 1 start-stop action, pressure fluctuation <0.03 MPa.

[0211] Scenario 2: Anode Replacement → Casting Cooling

[0212] 1) Process Property Comparison as shown in the following table:

[0213] Table 3 Process Property Comparison

[0214] Parameters Anode replacement phase Casting cooling phase Pressure demand 0.6-0.8 MPa 0.4-0.7 MPa Flow demand 0.6-0.8Q 0.7-0.9Q Equipment requirements High pressure screw machine Variable frequency centrifuge

[0215] 2) Challenges of Equipment Reuse:

[0216] The running 2# high-pressure screw machine is set at 0.7 MPa;

[0217] Pressure lower limit conflict: Casting cooling allows a minimum of 0.4 MPa, but the minimum stable running pressure of 2# machine is 0.55 MPa;

[0218] Efficiency loss: If forced to reuse 2# machine, need to maintain 0.55 MPa (higher than the ideal cooling pressure 0.5 MPa) → energy consumption increases by 8%;

[0219] 3) Optimization Solution:

[0220] 1. Step-by-step transition:

[0221] Step1: Start 4# variable frequency centrifuge (pressure range 0.3-0.7MPa), set to 0.5MPa / 0.7Q;

[0222] Step2: Reduce the pressure of 2# machine to 0.55MPa, and the flow to 0.2Q (minimum load);

[0223] Step3: After confirming that the pipe network pressure is stable at 0.52MPa, turn off 2# machine;

[0224] 2. Decision basis:

[0225] Although 2# machine needs to be turned off, the pressure jump is avoided by starting 4# machine in advance;

[0226] Opt for the suboptimal energy consumption but optimal life solution: the remaining life loss rate of 2# machine high-pressure operation is 2.3 times that of normal pressure;

[0227] 3. Effect: Although the start-stop is increased once, the high-pressure unit is prevented from running in the low-efficiency zone, and it is expected to extend the 2# machine overhaul cycle by 2000 hours.

[0228] The above steps generate a dynamic control strategy by binding the target demand of compressed air with the target control parameters of the air compressor group in real time, and combining the priority rules of high and low demand periods, use Modbus TCP protocol to realize accurate instruction issuing, effectively guarantee the supply pressure fluctuation control within ±1%; At the same time, by collecting the actual supply data to calculate the prediction error, triggering the incremental learning and periodic retraining mechanism, forming a closed-loop feedback optimization, enhancing the adaptability of the system to equipment aging and process changes; In the typical process scenarios such as alumina transportation and electrolytic cell cleaning, anode replacement and casting cooling, with the help of equipment reuse and step-by-step transition strategy, not only the number of air compressor start-stop is reduced, the energy consumption is reduced, but also the pressure jump and low-efficiency operation of equipment are avoided, realizing the double improvement of supply stability and equipment life.

[0229] According to another embodiment of the application, a compressed air dynamic allocation control system based on an aluminum electrolysis process is provided, comprising a demand prediction module, a demand optimization module, a control parameter determination module and a dynamic allocation control module;

[0230] The demand prediction module is configured to obtain aluminum electrolysis process parameters, air compressor group operation data and historical compressed air demand data and preprocess them; based on a pre-constructed compressed air demand prediction model, predict the initial demand of compressed air;

[0231] The demand optimization module is configured to dynamically adjust the predicted initial demand of compressed air based on the dynamic mapping relationship between the electrolytic cell heat balance and the compressed air consumption, to obtain the target demand of compressed air in the aluminum electrolysis process.

[0232] The control parameter determination module is configured to construct a multi-objective optimization control model of the air compressor group with the optimization objectives of minimizing total energy consumption, stabilizing gas supply pressure and matching dynamic flow demand, and determine target control parameters of the air compressor group by using a heuristic hierarchical search algorithm based on dynamic priority.

[0233] The dynamic allocation control module is configured to bind the target demand quantity of compressed air in the aluminum electrolysis process with the target control parameters of the air compressor group, generate a dynamic control strategy in combination with priority rules, and realize dynamic allocation control of compressed air according to the dynamic control strategy.

[0234] In summary, by means of the technical solutions of the present application, the present application deeply integrates the thermal balance state of the electrolytic cell and the compressed air demand based on machine learning and multi-objective optimization technology, and constructs a complete technical system of data collection and analysis, accurate demand prediction, dynamic optimization control and feedback iteration and upgrading. Compared with the traditional technical solutions, the present application significantly improves the accuracy of compressed air demand prediction, greatly reduces energy consumption, effectively stabilizes the pipe network pressure, reduces the abnormal situation of the electrolytic cell caused by thermal imbalance, prolongs the service life of the air compressor equipment, and comprehensively improves the accuracy, economy and system stability of compressed air allocation in the aluminum electrolysis production process.

[0235] In addition, by collecting electrolytic cell thermal balance related parameters and combining advanced prediction models and feature engineering, the present application comprehensively captures the influence of the thermal state of the electrolytic cell on gas consumption, significantly improves the accuracy of compressed air initial demand prediction compared with traditional methods, and lays a solid foundation for subsequent dynamic adjustment, especially in complex working conditions.

[0236] In addition, by means of the double-path correction model and the staged correction strategy, the present application can dynamically adjust the compressed air demand according to the thermal balance state of the electrolytic cell, realize accurate compensation of the gas demand in the thermal imbalance state, and effectively avoid the problem of uneven gas supply by using the multi-cell collaborative clustering mechanism, reduce invalid blow-by gas consumption, and optimize the supply and demand balance of compressed air in the whole workshop.

[0237] In addition, the heuristic hierarchical search algorithm based on dynamic priority of the present application takes into account multi-objective optimization of energy consumption, pressure stability and the like, effectively reduces the energy consumption of the air compressor group under the premise of ensuring the feasibility of gas supply, significantly reduces the pipe network pressure fluctuation, and based on the dynamic priority adjustment mechanism, can quickly respond to sudden situations such as pressure abnormalities and equipment overload, greatly improves the stability and reliability of the system in complex working conditions, and realizes efficient matching of compressed air flow by real-time binding of the target demand quantity and the control parameters and cooperating with the priority scheduling rules.

[0238] In addition, the application can not only balance each control link in the air compression station, monitor and control the air compressor and each related equipment in the station, but also can continuously control the entire compressed air system in real time based on the continuous pressure feedback of the air compressor, the gas terminal and the pipe network pressure sensor, so that the compressed air supply and demand is balanced. In addition, the running data of the air compression system can be continuously collected on site, and the running parameters of each air compressor can be optimized in real time to ensure that the system has the best energy efficiency and safety.

[0239] The above only describes the preferred embodiments of the application and is not intended to limit the application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the application shall be included in the protection scope of the application.

Claims

1. A dynamic compressed air distribution control method based on the electrolytic aluminum process, characterized in that, Includes the following steps: S1. Obtain electrolytic aluminum process parameters, air compressor group operation data, and historical compressed air demand data and perform preprocessing; predict the initial demand for compressed air based on the pre-built compressed air demand prediction model. S2. Based on the dynamic mapping relationship between the heat balance of the electrolytic cell and the consumption of compressed air, the predicted initial demand for compressed air is dynamically adjusted to obtain the target demand for compressed air in the electrolytic aluminum process. S3. With minimizing total energy consumption, stabilizing air supply pressure, and matching dynamic flow demand as optimization objectives, a multi-objective optimization control model for the air compressor group is constructed, and the target control parameters of the air compressor group are determined by using a heuristic hierarchical search algorithm based on dynamic priority. S4. Bind the target demand for compressed air in the electrolytic aluminum process with the target control parameters of the air compressor group, generate a dynamic control strategy based on priority rules, and realize the dynamic allocation control of compressed air according to the dynamic control strategy.

2. The compressed air dynamic distribution control method based on the electrolytic aluminum process flow according to claim 1, characterized in that, The process of acquiring electrolytic aluminum process parameters, air compressor group operating data, and historical compressed air demand data, and preprocessing them; and predicting the initial demand for compressed air based on a pre-built compressed air demand prediction model, includes the following steps: S11. Obtain data on current, voltage, temperature, alumina feed rate, cell age, cell surface temperature field distribution, and furnace wall thickness of the electrolytic aluminum production line, and construct a set of process parameters; obtain data on output pressure, flow rate, energy consumption, start / stop status, and running time of each air compressor in the electrolytic aluminum production line, and construct a set of operating parameters. S12. Extract the time series data of the total compressed air demand in the past preset time period, associate the process parameters and air compressor operating parameters of the corresponding time period to form a historical dataset based on demand, process and equipment, and preprocess the data in the historical dataset. S13. Based on the preprocessed historical dataset, train the pre-built compressed air demand prediction model, and use the trained compressed air demand prediction model to output the initial demand for compressed air corresponding to the real-time electrolytic aluminum process parameters and the air compressor group operation data.

3. The compressed air dynamic distribution control method based on the electrolytic aluminum process flow according to claim 1, characterized in that, The method of dynamically adjusting the predicted initial demand for compressed air based on the dynamic mapping relationship between the thermal balance of the electrolytic cell and the consumption of compressed air, to obtain the target demand for compressed air in the electrolytic aluminum process, includes the following steps: S21. Identify the high-temperature electrolyte zone and the low-temperature furnace side zone based on the temperature field distribution data of the tank surface, and determine the temperature gradient and area ratio; calculate the furnace side thickness change rate based on the furnace side thickness data, and construct the furnace side stability index in combination with voltage fluctuations; calculate the heat balance gas consumption correction factor based on the temperature gradient, furnace side thickness change rate and furnace side stability index. S22. Construct a dual-path correction model, use the dual-path correction model to determine the mechanism path correction amount and the data path correction amount, and combine the fusion mechanism to determine the target correction amount; S23. Based on the comparison results between the heat balance gas consumption correction factor and the preset threshold, the real-time heat balance state of the electrolyzer is divided into a normal heat balance stage and a heat imbalance stage. The heat balance gas consumption correction factor and the target correction amount are used to correct the demand for compressed air in a single electrolyzer in the initial demand under the normal heat balance stage and the heat imbalance stage, respectively. S24. Use clustering algorithm to classify the types of electrolytic cells, and use weighted summation method to determine the compressed air demand of all electrolytic cells in the workshop, so as to obtain the target demand of compressed air in the electrolytic aluminum process.

4. The dynamic compressed air distribution control method based on the electrolytic aluminum process flow according to claim 3, characterized in that, The formula for calculating the heat balance gas consumption correction factor is as follows: In the formula, γ(t) represents the heat balance gas consumption correction factor at time t, ReLU represents the activation function, and α and β represent the two process weights, respectively. T represents the temperature gradient at time t. thres The threshold temperature for thermal imbalance is represented by HIS(t), which represents the furnace side stability index at time t, and γ0 represents the baseline correction factor. d represents the rate of change of furnace wall thickness at time t. nominal U(t) represents the rated value of the furnace wall thickness, U(t) represents the electrolytic cell voltage at time t, η represents the voltage sensitivity coefficient, and σ represents the standard deviation of voltage fluctuation. nominal This indicates the rated value of the electrolytic cell voltage.

5. The compressed air dynamic distribution control method based on the electrolytic aluminum process flow according to claim 4, characterized in that, The construction of the dual-path correction model, the determination of the mechanism path correction amount and the data path correction amount using the dual-path correction model, and the determination of the target correction amount by combining the fusion mechanism include the following steps: S221. Based on the alumina dissolution kinetics equation, determine the additional purging requirement during thermal imbalance and obtain the mechanism path correction amount; S222. Use the gated loop unit to learn the mapping between historical thermal balance state and gas consumption deviation to obtain the data path correction amount. S223. Based on the mechanism path correction amount and the data path correction amount, the target correction amount is determined by combining the fusion mechanism; The formula for calculating the mechanism path correction is as follows: The formula for calculating the data path correction amount is: Q data (t)=GRU(γ(t),γ(t-1),...,Q error (t-1)) The formula for calculating the target correction amount is: Q adj (t)=Q mech (t)·(1+λ·Q data (t)) In the formula, Q mech (t) represents the mechanism path correction at time t, k represents the process constant reflecting the coupling relationship between current efficiency and thermal imbalance, γ(τ) represents the thermal balance gas consumption correction factor at time τ, I(τ) represents the electrolytic cell current at time τ, and Q data (t) represents the data path correction at time t, GRU represents the gated loop unit, and Q... error (t-1) represents the historical prediction error at time t-1, Q adj (t) represents the target correction amount at time t, and λ represents the data path weight.

6. The dynamic compressed air distribution control method based on the electrolytic aluminum process flow according to claim 5, characterized in that, The process of dividing the real-time thermal balance state of the electrolyzer into a normal thermal balance stage and a thermal imbalance stage based on the comparison result between the thermal balance gas consumption correction factor and the preset threshold, and then correcting the demand for compressed air in a single electrolyzer in the initial demand under the normal thermal balance stage and the thermal imbalance stage using the thermal balance gas consumption correction factor and the target correction amount, includes the following steps: S231. Based on the comparison results between the heat balance gas consumption correction factor and the preset heat balance gas consumption correction factor threshold, the real-time heat balance state of the electrolyzer is divided into a normal heat balance stage and a heat imbalance stage. S232. The demand for compressed air in a single electrolytic cell in the initial demand under normal thermal balance stage is corrected by using a thermal balance air consumption correction factor. S233. Use the target correction amount to correct the demand for compressed air in a single electrolytic cell in the initial demand during the thermal imbalance stage; The formula for calculating the compressed air demand under normal thermal equilibrium conditions is as follows: Q target (t)=Q pred (t)·(1+0.05·γ(t)) The formula for calculating the compressed air demand during the thermal imbalance stage is: Q target (t)=Q pred (t)+Q adj (t) In the formula, Q target (t) represents the demand for compressed air at time t, Q pred (t) represents the initial demand of a single electrolyzer at time t.

7. The dynamic compressed air distribution control method based on the electrolytic aluminum process flow according to claim 4, characterized in that, The process of classifying electrolytic cells by clustering algorithm and determining the compressed air demand of all electrolytic cells in the workshop by weighted summation method to obtain the target compressed air demand in the electrolytic aluminum process includes the following steps: S241. Obtain the thermal balance index, furnace side stability index, temperature gradient and cell age data of the electrolytic cell, and perform outlier removal and standardization on the obtained data. The formula for calculating the thermal balance index is: In the formula, HE(t) represents the thermal balance index at time t, ΔT(t) represents the temperature fluctuation of the bath at time t, and T nominal The value represents the rated bath temperature, and Δd(t) represents the change in furnace wall thickness at time t. nominal The rated furnace side thickness is represented by α, β, and ω, which represent the three process weights, respectively. S242. Use the analysis of variance method to determine the contribution of the heat balance index, furnace side stability index, temperature gradient and furnace age to the heat balance state, and determine the key feature vector based on the contribution. S243. The elbow rule is used to determine the number of clusters and the initial center point, and the Euclidean distance between the key feature vector of the electrolytic cell and the initial center point is calculated. The number of clusters includes three types: thermally stable cell, thermally sensitive cell, and thermally unbalanced cell. S244. Based on the Euclidean distance calculation results, assign the electrolytic cells to the corresponding categories and update the category center points repeatedly until convergence is achieved, thus completing the clustering partitioning. S245. Determine the correction coefficient based on the type and age of the electrolytic cell, and use the weighted summation method to determine the compressed air demand of all electrolytic cells in the workshop, thereby obtaining the target demand of compressed air in the electrolytic aluminum process.

8. The method for dynamic distribution control of compressed air based on the electrolytic aluminum process flow according to claim 1, characterized in that, The process involves constructing a multi-objective optimization control model for an air compressor group, with the optimization objectives of minimizing total energy consumption, stabilizing air supply pressure, and matching dynamic flow demand. This model is then solved using a heuristic hierarchical search algorithm based on dynamic priority. The determination of the target control parameters for the air compressor group includes the following steps: S31. With minimizing total energy consumption, stabilizing air supply pressure and matching dynamic flow demand as optimization objectives, and with air compressor start-stop status and operating pressure setpoint as decision variables, a multi-objective optimization control model for air compressor group is constructed. S32. A heuristic hierarchical search algorithm based on dynamic priority is used to solve the multi-objective optimization control model of the air compressor group and determine the target control parameters of the air compressor group.

9. The dynamic compressed air distribution control method based on the electrolytic aluminum process flow according to claim 8, characterized in that, A heuristic hierarchical search algorithm based on dynamic priority is used to solve the multi-objective optimization control model of the air compressor group, and the target control parameters of the air compressor group are determined as follows: Based on the pressure range and target demand for compressed air in the electrolytic aluminum process, quickly select air compressor combinations to ensure that the total rated flow of the air compressor combination meets the process requirements. With the minimization of total energy consumption as the core, the algorithm combines tabu search for global optimization, simulated annealing for local search, and a runtime balancing mechanism to prioritize the generation of shutdown schemes for high-load equipment, thus balancing multi-objective optimization. When the standard deviation of pipeline pressure fluctuation exceeds the preset value continuously, the current optimization and pressure stability target priority is immediately interrupted, and a solution set with stability as the core is regenerated; when the continuous running time of the air compressor exceeds the maximum allowable time threshold, a candidate solution is generated that only includes the shutdown of the equipment and the total flow meets the requirements. By combining the results of hierarchical search and dynamic adjustment, the start-stop status and operating pressure setpoint of each air compressor are determined, and the target control parameter combination of the air compressor group is obtained.

10. A dynamic compressed air distribution control system based on an electrolytic aluminum process, used to implement the steps of the dynamic compressed air distribution control method based on an electrolytic aluminum process as described in any one of claims 1-9, characterized in that, The system includes a demand forecasting module, a demand optimization module, a control parameter determination module, and a dynamic allocation control module; The demand prediction module is used to acquire and preprocess electrolytic aluminum process parameters, air compressor group operation data and historical compressed air demand data; and predict the initial demand for compressed air based on a pre-built compressed air demand prediction model. The demand optimization module is used to dynamically adjust the predicted initial demand for compressed air based on the dynamic mapping relationship between the heat balance of the electrolytic cell and the consumption of compressed air, so as to obtain the target demand for compressed air in the electrolytic aluminum process. The control parameter determination module is used to construct a multi-objective optimization control model for the air compressor group with the optimization objectives of minimizing total energy consumption, stabilizing air supply pressure, and matching dynamic flow demand, and to solve the model using a heuristic hierarchical search algorithm based on dynamic priority to determine the target control parameters of the air compressor group. The dynamic allocation control module is used to bind the target demand for compressed air in the electrolytic aluminum process with the target control parameters of the air compressor group, generate a dynamic control strategy based on priority rules, and realize the dynamic allocation control of compressed air according to the dynamic control strategy.

Images