Aluminum electrolysis air compression station energy consumption optimization method based on load prediction and dynamic voltage regulation

Abstract

The invention discloses an aluminum electrolysis air compression station energy consumption optimization method based on load prediction and dynamic pressure regulation, relates to the technical field of air compression station operation optimization, and greatly reduces the fluctuation range of air supply pressure by accurately matching supply and demand through a refined control process, dynamic pressure regulation based on load prediction and smooth regulation of a frequency converter. Afterwards, a frequency conversion speed regulation mode replaces traditional frequent start and stop control, large current impact and mechanical abrasion caused by repeated start and stop of the motor are avoided, and maintenance cost and energy consumption caused by the large current impact and the mechanical abrasion are reduced. And meanwhile, the gas consumption load is accurately predicted through the LSTM model, an optimal operation strategy is planned in advance, and excessive or insufficient gas supply is avoided. And finally, through prediction of the load of the whole system and dynamic setting of the pressure, a decision basis is provided for collaborative scheduling of the air compressors, so that the comprehensive energy efficiency of the system is greatly improved, and the problems that in a traditional method, all units operate independently, and the efficiency is low are solved.

Description

Technical Field

[0001] This invention belongs to the field of air compressor station operation optimization technology, and specifically relates to an energy consumption optimization method for aluminum electrolysis air compressor stations based on load prediction and dynamic voltage regulation. Background Technology

[0002] In the aluminum electrolysis production process, the air compressor station, as a key energy supply system, provides compressed air for alumina transportation and pneumatic valve control in the electrolytic cell. Due to the periodicity and fluctuation of the electrolytic cell's operating state, the air supply demand of the air compressor station exhibits dynamic changes. However, existing air compressor stations mostly adopt fixed-frequency operation or simple pressure threshold start-stop control methods, which cannot dynamically adjust the air supply pressure and air compressor operating state according to actual air demand. This control method has the following technical defects: (1) large fluctuations in air supply pressure, affecting the stability of equipment operation; (2) frequent start-stop of air compressors, increasing mechanical wear and energy consumption; (3) inability to optimize operating strategies according to load changes, resulting in energy waste; (4) lack of consideration for the coordinated scheduling of multiple air compressors, resulting in low overall system energy efficiency. Therefore, there is an urgent need for an air compressor station energy consumption control method that can realize dynamic pressure regulation, intelligent load prediction, and multi-machine coordinated scheduling to improve energy efficiency, reduce operating costs, and enhance the reliability of system operation. Summary of the Invention

[0003] To address the shortcomings of existing technologies, this invention provides an energy consumption optimization method for aluminum electrolysis air compressor stations based on load prediction and dynamic voltage regulation, in order to solve the aforementioned technical problems.

[0004] An energy consumption optimization method for aluminum electrolysis air compressor stations based on load prediction and dynamic voltage regulation includes the following steps:

[0005] Real-time acquisition of operating data from the aluminum electrolysis air compressor station system, including pressure, flow rate, start / stop status, loading / unloading status of air-using equipment, and ambient temperature, and normalization processing of the operating data;

[0006] The future gas load of the aluminum electrolysis air compressor station is predicted based on the pre-processed operating data using a trained LSTM model, and the predicted load value is obtained.

[0007] Based on the predicted load value and combined with the preset fuzzy control rules, the output pressure of the air compressor is dynamically set.

[0008] Based on the set output pressure, a control signal is generated by the PID controller, and the frequency converter receives the control signal to adjust the operating frequency of the air compressor.

[0009] Preferably, when collecting real-time operating data of the aluminum electrolysis air compressor station system, redundant pressure and flow monitoring units are also used to collect the pressure and flow values ​​of the main pipeline tee of the gas supply network, and the pressure and flow values ​​of the inlet of the terminal electrolytic cell.

[0010] Preferably, the LSTM model also incorporates an attention mechanism during training;

[0011] The input data for the LSTM model includes: electrolytic cell duty cycle, alumina feeding frequency, historical gas flow rate sliding window, ambient temperature, and production schedule.

[0012] The output of the LSTM model is the gas load of the aluminum electrolysis air compressor station for the next 2 hours, with an output time step of 10 minutes.

[0013] Preferably, the LSTM model includes:

[0014] Input layer: has 15-dimensional feature vectors;

[0015] Two LSTM layers, each LSTM layer has 64 hidden units;

[0016] Additive attention layer: used for additive attention mechanisms

[0017] Output layer: Used to output predicted load values.

[0018] Preferably, the PID controller is configured as follows:

[0019] The input variables are pressure deviation e and its rate of change Δe, and the fuzzy set is defined as {NB, NM, NS, ZE, PS, PM, PB}.

[0020] The output variable is the pressure setpoint adjustment amount ΔPset, which is a fuzzy set of input variables;

[0021] The control rules include: if e is PB and Δe is ZE, then ΔPset is PB; if e is NS and Δe is PM, then ΔPset is NM.

[0022] When the load change rate is >0.1 MPa / min, switch to traditional PID control.

[0023] Preferably, the generation of the control signal by the PID controller specifically involves:

[0024] The frequency converter is for centrifugal air compressors: the target frequency is 55Hz when the load rate is <30%, 60Hz when the load rate is 30-60%, 65Hz when the load rate is 60-90%, and 70Hz when the load rate is >90%.

[0025] The frequency converter is for screw air compressors: the target frequency is 45Hz when the load rate is <30%, 50Hz when the load rate is 30-60%, 60Hz when the load rate is 60-90%, and 65Hz when the load rate is >90%.

[0026] Preferably, when generating control signals through the PID controller, the PID controller also provides a graded response to abnormal states, the graded response specifically including:

[0027] Level 1 warning: When the temperature exceeds 95℃, log the information and prompt the user to check the cooling system.

[0028] Level 2 alarm: Automatically switch to standby unit when exhaust pressure > 0.85 MPa;

[0029] Level 3 fault: When the motor current fluctuation is >15%, the unit will be shut down immediately and the faulty unit will be locked.

[0030] Preferably, the defuzzification of the PID controller uses a weighted average method, and the membership function is a trigonometric function.

[0031] Preferably, when multiple air compressors are operating in parallel, the following steps are used to dynamically set the output pressure of the air compressors:

[0032] Collect data on the energy efficiency ratio, operating status, and total system load of each air compressor;

[0033] Establish an objective function that aims to minimize total energy consumption while also taking into account equipment wear and switching frequency.

[0034] A genetic algorithm is used to find the operating scheme of multiple air compressors that meets the current load and has the lowest energy consumption.

[0035] Preferably, it is characterized in that,

[0036] The scheduling strategy is updated every hour and controls the start and stop of any air compressor through a digital output interface.

[0037] When the efficiency of any air compressor decreases, it is removed from the scheduling sequence.

[0038] The beneficial effects of this invention are as follows: Through a refined control process, dynamic pressure regulation based on load prediction and smooth adjustment of the frequency converter precisely match supply and demand, greatly reducing the fluctuation range of air supply pressure, thereby significantly improving the stability and reliability of downstream pneumatic equipment operation. Furthermore, variable frequency speed control replaces the traditional frequent start-stop control, avoiding the large current surges and mechanical wear caused by repeated motor starts and stops, extending equipment lifespan, and reducing maintenance costs and energy consumption. Simultaneously, accurate prediction of air load using the LSTM model allows for advance planning of optimal operating strategies, avoiding oversupply or undersupply due to delayed response. Finally, by predicting the load of the entire system and dynamically setting the pressure, a decision-making basis is provided for the coordinated scheduling of air compressors, enabling the entire air compressor station system to operate as a whole in an optimized manner, thereby significantly improving the overall energy efficiency of the system and solving the problem of independent operation and low efficiency of individual units in traditional methods. Attached Figure Description

[0039] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0040] Figure 1 This is a schematic diagram illustrating the steps of the energy consumption optimization method for aluminum electrolysis air compressor stations based on load prediction and dynamic voltage regulation provided by the present invention. Detailed Implementation

[0041] In the description of this invention, it should be understood that the terms "center," "longitudinal," "lateral," "length," "width," "thickness," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," "outer," "clockwise," "counterclockwise," "axial," "radial," and "circumferential" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0042] The following disclosure provides many different embodiments or examples for implementing different structures of the present invention. To simplify the disclosure of the present invention, the components and arrangements of specific examples are described below. Of course, these are merely examples and are not intended to limit the present invention.

[0043] The embodiments of the invention will now be described in detail with reference to the accompanying drawings.

[0044] like Figure 1 As shown, the energy consumption optimization method for aluminum electrolysis air compressor stations based on load prediction and dynamic voltage regulation includes the following steps:

[0045] Real-time acquisition of operating data from the aluminum electrolysis air compressor station system, including pressure, flow rate, start / stop status, loading / unloading status of air-using equipment, and ambient temperature, and normalization processing of the operating data;

[0046] The future gas load of the aluminum electrolysis air compressor station is predicted based on the pre-processed operating data using a trained LSTM model, and the predicted load value is obtained.

[0047] Based on the predicted load value and combined with the preset fuzzy control rules, the output pressure of the air compressor is dynamically set.

[0048] Based on the set output pressure, a control signal is generated by the PID controller, and the frequency converter receives the control signal to adjust the operating frequency of the air compressor.

[0049] Before normalizing historical data, the mean, variance, and kurtosis of the running data are pre-calculated:

[0050] mean

[0051] variance

[0052] Kurtosis

[0053] in, The mean of the running data, The total number of running data points, For the first The value of each running data point, The variance of the running data, The kurtosis value of the running data is 3. In the kurtosis calculation, subtracting 3 is to make the kurtosis of the normal distribution 0, which is called excess kurtosis.

[0054] The Isolation Forest algorithm is used to monitor anomalies, such as when the temperature data of a certain air compressor meets the following conditions: If it is, it is judged as an anomaly, triggering the protection mechanism, where For data points The path length in the decision tree is a preset threshold.

[0055] During the data acquisition phase, the system collects key operational data from the air compressor station in real time and at high frequency. This includes the pressure, flow rate, start / stop and loading / unloading status of each gas-consuming device, as well as ambient temperature. The raw data collected is not used directly but first enters the preprocessing stage. The system first performs statistical analysis on historical operational data, calculating its mean, variance, and kurtosis. These statistical characteristics help to better understand the data distribution and volatility. Subsequently, all data is normalized to eliminate the influence of different physical dimensions, preparing for subsequent machine learning model predictions. Simultaneously, to ensure system operational safety, a real-time anomaly monitoring mechanism based on the isolated forest algorithm is introduced. For example, the system continuously monitors key indicators such as the temperature of a particular air compressor. If the path length of a data point in the algorithm's decision tree is less than a preset threshold, it is immediately identified as an abnormal state, and the corresponding protection mechanism is automatically triggered, effectively preventing equipment failure. After data preprocessing, a pre-trained long short-term memory network model is invoked, using the processed real-time operational data to accurately predict gas load changes over a future period, generating precise predicted load values. Next, based on this predicted load value and a set of pre-set expert fuzzy control rules, the optimal output pressure value that the air compressor needs to achieve is dynamically set. Finally, the dynamically set pressure target value is transmitted to the PID controller. The PID controller accurately generates a control signal based on the deviation between the target value and the actual pressure value. This signal is sent to the frequency converter, which adjusts the operating frequency of the air compressor motor in real time, thereby precisely controlling the output of compressed air and ensuring that the air supply is highly matched with the actual demand.

[0056] Compared to existing technologies, this method, through a refined control process, precisely matches supply and demand based on dynamic pressure regulation according to load prediction and smooth adjustment of the frequency converter, greatly reducing the fluctuation range of air supply pressure and thus significantly improving the stability and reliability of downstream pneumatic equipment. Furthermore, variable frequency speed control replaces the traditional frequent start-stop control, avoiding the high current surges and mechanical wear caused by repeated motor starts and stops, extending equipment lifespan and reducing maintenance costs and energy consumption. Simultaneously, accurate prediction of air load using an LSTM model allows for advance planning of optimal operating strategies, preventing oversupply or undersupply due to delayed response. Finally, by predicting the overall system load and dynamically setting the pressure, a decision-making basis is provided for the coordinated scheduling of air compressors, enabling the entire air compressor station system to operate as a whole in an optimized manner, thereby significantly improving the overall energy efficiency of the system and solving the problem of independent operation and low efficiency of individual units in traditional methods.

[0057] More specifically, when collecting real-time operating data of the aluminum electrolysis air compressor station system, redundant pressure and flow monitoring units are also used to collect the pressure and flow values ​​of the main pipeline tee of the gas supply network, as well as the pressure and flow values ​​of the inlet of the terminal electrolytic cell.

[0058] To improve the accuracy and reliability of the data, additional pressure and flow monitoring units were deployed at two key locations: the tee of the main gas supply pipeline and the inlet of the electrolytic cell at the end of the process, to enable cross-validation and supplementation.

[0059] More specifically, the LSTM model also incorporates an attention mechanism during training;

[0060] The input data for the LSTM model includes: electrolytic cell duty cycle, alumina feeding frequency, historical gas flow rate sliding window, ambient temperature, and production schedule.

[0061] The output of the LSTM model is the gas load of the aluminum electrolysis air compressor station for the next 2 hours, with an output time step of 10 minutes.

[0062] The model's prediction extension integrates multi-dimensional, heterogeneous production process data directly related to gas demand, and incorporates an attention mechanism during training and prediction. Specifically, after encoding the sequence data and extracting the hidden state at each time step using an LSTM layer, an attention layer is introduced. This attention layer learns and dynamically assigns different weights to the hidden states at different time points and with different input features. Finally, after attention weighting and LSTM processing, the model outputs a refined time-series prediction result, specifically the gas load prediction value every 10 minutes for the next 2 hours, forming a future load curve containing 12 data points. The alumina feeding frequency is 0-3 times / hour, the historical gas flow rate sliding window is 2 hours, and the ambient temperature is -10~40℃.

[0063] More specifically, the LSTM model includes:

[0064] Input layer: has 15-dimensional feature vectors;

[0065] Two LSTM layers, each LSTM layer has 64 hidden units;

[0066] Additive attention layer: used for additive attention mechanisms

[0067] Output layer: Used to output predicted load values.

[0068] The additive attention mechanism is shown in the following equation:

[0069]

[0070] in, For the first The attention weights corresponding to each input. Weight vector transpose, This is the weight matrix. For the first A vector of inputs, This is the bias vector.

[0071] First, at the input layer, the system integrates the aforementioned production and environmental data into a standardized 15-dimensional feature vector. This vector is updated and input into the model at each time step. Next, this feature vector sequence passes sequentially through two stacked LSTM layers, each containing 64 hidden units. The first LSTM layer is responsible for initially extracting and encoding temporal dependencies and short-term patterns from the original input sequence; its output serves as the input to the second LSTM layer. The second LSTM layer performs deeper abstraction and learning, capturing longer-term and more complex temporal dynamic features, thus building a deep understanding of future behavior. Then, the additive attention layer receives the hidden state sequence from all time steps of the output of the previous second LSTM layer and learns an alignment score through a small feedforward network, determining which past time points should receive more attention when predicting future load. Finally, the information, weighted and fused by the attention mechanism, is fed into the output layer, which decodes and transforms the highly condensed feature information learned by all previous layers into a specific prediction of future gas load.

[0072] More specifically, the PID controller is configured as follows:

[0073] The input variables are pressure deviation e and its rate of change Δe, and the fuzzy set is defined as {NB, NM, NS, ZE, PS, PM, PB}.

[0074] The output variable is the pressure setpoint adjustment amount ΔPset, which is a fuzzy set of input variables;

[0075] The control rules include: if e is PB and Δe is ZE, then ΔPset is PB; if e is NS and Δe is PM, then ΔPset is NM.

[0076] When the load change rate is >0.1 MPa / min, switch to traditional PID control.

[0077] The pressure deviation is calculated using the following formula:

[0078]

[0079] Where is the pressure deviation, is the target pressure value, and is the measured pressure value.

[0080] The rate of change of deviation is calculated using the following formula:

[0081]

[0082] Where is the deviation change rate, is the deviation value at the current time, and is the deviation value at the previous time. The allowable range of the deviation rate is [-0.02, +0.02] MPa / s.

[0083] For most of the stable operation period, the controller does not use fixed PID parameters but enters fuzzy control mode. In this mode, two core input variables are monitored in real time: the current pressure deviation *e*, which is the difference between the actual supply pressure and the dynamic target pressure; and the rate of change of this deviation *Δe*, which reflects the speed and trend of the pressure deviating from the target value. These two precise values ​​are fuzzified using a preset membership function. Subsequently, the fuzzy inference engine applies these fuzzified inputs to a series of preset control rule bases. For example, the rule "If *e* is PB and *Δe* is ZE, then *ΔPset* is PB" is triggered to quickly increase the pressure and eliminate large deviations; while the rule "If *e* is NS and *Δe* is PM, then *ΔPset* is NM" can predictively prevent overshoot or oscillation caused by excessively rapid pressure drops. The inference engine synthesizes the results of all triggered rules and calculates a precise pressure setpoint adjustment *ΔPset* through defuzzification. This adjustment is added to the original setpoint to form a new, better pressure target, which is then executed by the underlying PID controller, thereby achieving smooth and intelligent fine-tuning of the pressure. However, when the load change rate is detected to exceed the threshold of 0.1 MPa / min, which is a sharp fluctuation, it means that a sudden change in the production conditions may have occurred. At this time, the fuzzy control may not respond quickly enough, and the system will automatically switch to the traditional PID control mode to ensure the stability of the system by utilizing its classic and fast response characteristics. After the pressure returns to a stable state, it will switch back to the more efficient fuzzy control mode.

[0084] More specifically, the generation of control signals through the PID controller specifically involves:

[0085] The frequency converter is for centrifugal air compressors: the target frequency is 55Hz when the load rate is <30%, 60Hz when the load rate is 30-60%, 65Hz when the load rate is 60-90%, and 70Hz when the load rate is >90%.

[0086] The frequency converter is for screw air compressors: the target frequency is 45Hz when the load rate is <30%, 50Hz when the load rate is 30-60%, 60Hz when the load rate is 60-90%, and 65Hz when the load rate is >90%.

[0087] In addition, a limit on the rate of change of frequency is introduced, which is calculated by the following formula:

[0088]

[0089] in, The maximum rate of change of frequency. For the target frequency, For the current frequency, This is the system response time constant, which is generally set to 5 seconds.

[0090] Based on the real-time load rate of the air compressor, the mapping rules fully consider the differences in operating characteristics of different types of air compressors when mapping it to the optimal operating frequency target range: For centrifugal air compressors with high speed and narrow high-efficiency range, the target frequency is set at an efficient 55Hz when the load rate is below 30%; as the load rate climbs to 30-60%, 60-90%, and above 90%, the target frequency will correspondingly increase in steps to 60Hz, 65Hz, and 70Hz, ensuring that it always operates at or near its optimal energy efficiency point. For screw air compressors with relatively lower speed and a wider operating range, the system uses a lower and smoother frequency mapping table, with target frequencies set at 45Hz, 50Hz, 60Hz, and 65Hz respectively within the same load rate range. At the same time, the design of the frequency change rate limit ensures that the frequency adjustment process is both efficient and stable, avoiding any form of abrupt change.

[0091] More specifically, when generating control signals through a PID controller, the PID controller also provides a graded response to abnormal states, and the graded response specifically includes:

[0092] Level 1 warning: When the temperature exceeds 95℃, log the information and prompt the user to check the cooling system.

[0093] Level 2 alarm: Automatically switch to standby unit when exhaust pressure > 0.85 MPa;

[0094] Level 3 fault: When the motor current fluctuation is >15%, the unit will be shut down immediately and the faulty unit will be locked.

[0095] More specifically, the defuzzification of the PID controller adopts a weighted average method, and the membership function is a trigonometric function.

[0096] During the fuzzification input phase, the pressure deviation *e* and the rate of change of deviation *Δe* are mapped using a set of preset triangular membership functions. Each fuzzy set corresponds to a triangular function graph, with the vertices and bases defining the numerical range of the fuzzy concept. For example, the ZE fuzzy set of pressure deviation *e* might be defined as a triangle with a vertex of 0 MPa and a base ranging from -0.02 MPa to +0.02 MPa. When *e* is 0, its membership degree is 1; when *e* is 0.01 MPa, its membership degree might be 0.5. Substituting the current values ​​of *e* and *Δe* into all relevant triangular membership functions, the degree to which they belong to each fuzzy set is calculated. Then, the fuzzy inference engine activates a preset control rule base based on these membership degrees, obtaining the trigger strength of each rule. Finally, during defuzzification, the system uses a weighted average method to integrate the output results of all triggered rules. The process involves multiplying the center value of the output fuzzy set corresponding to each activated rule by the trigger strength of that rule, then summing all these weighted center values, and finally dividing by the sum of the trigger strengths of all rules. Mathematically, this calculation is equivalent to finding the x-coordinate of the geometric centroid of a composite graph formed by the membership functions of all activated outputs. This final, precise value is the pressure setpoint adjustment ΔPset that the PID controller needs to perform, and it will be immediately used to adjust the operating target of the air compressor.

[0097] More specifically, when multiple air compressors are operating in parallel, the following steps are used to dynamically set the output pressure of the air compressors:

[0098] Collect data on the energy efficiency ratio, operating status, and total system load of each air compressor;

[0099] Establish an objective function that aims to minimize total energy consumption while also taking into account equipment wear and switching frequency.

[0100] A genetic algorithm is used to find the operating scheme of multiple air compressors that meets the current load and has the lowest energy consumption.

[0101] The objective function is shown in the following formula:

[0102]

[0103] Its constraints are:

[0104]

[0105] (0 = shutdown, 1 = running)

[0106] in, The total number of air compressors. For the first The power of the air compressor. For the first The air supply capacity of the air compressor. For the first Decision variables for an air compressor This represents the total gas demand.

[0107] When using a genetic algorithm to solve the problem, the population size is 50, the mutation rate is 0.1, and the scheduling strategy is updated every hour.

[0108] First, in the data acquisition and status awareness phase, the energy efficiency ratio of each air compressor under current operating conditions is calculated using precise electricity meters and flow meters; the operating status of each piece of equipment is acquired in real time; and the total demand of all gas consumption points in the entire gas supply network at the current moment is measured. Next, a complex multi-objective optimization function is constructed. The core objective of this function is to minimize the total energy consumption of the entire air compressor station, while incorporating constraints and penalties: one is to achieve balanced equipment wear, which is typically quantified as the variance of the cumulative operating time or start-stop count of each air compressor. The smaller the variance, the more balanced the wear, which is reflected as a reward term in the objective function; the other is to minimize the number of switching operations, i.e., to minimize the number of unit start-stop or loading / unloading actions while meeting load changes, to avoid the impact and additional energy consumption caused by frequent switching, which is reflected as a penalty term in the objective function. By assigning different weight coefficients to these three sub-objectives, a complex multi-dimensional scheduling problem is transformed into a solvable mathematical model. Finally, in the intelligent decision-making and solution phase, a genetic algorithm is invoked to find the optimal solution to the objective function. The system can determine the optimal air compressor operating scheme that maximizes the overall objective function value while meeting the current total system load requirements. This final scheme will then be issued as an instruction to each air compressor for execution.

[0109] More specifically, the scheduling strategy is updated every hour and controls the start and stop of any air compressor through a digital output interface;

[0110] When the efficiency of any air compressor decreases, it is removed from the scheduling sequence.

[0111] When the efficiency drops by 10% or more, it is removed from the scheduling sequence.

[0112] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be covered within the scope of the claims and specification of the present invention.

Claims

1. An energy consumption optimization method for aluminum electrolysis air compressor stations based on load prediction and dynamic voltage regulation, characterized in that, Includes the following steps: The system collects real-time operational data from the aluminum electrolysis air compressor station system. This data includes the pressure, flow rate, start / stop status, loading / unloading status of the air-consuming equipment, and ambient temperature. The operational data is then normalized. A trained LSTM model is used to predict the future air load of the aluminum electrolysis air compressor station based on the preprocessed operational data, yielding a predicted load value. Based on the predicted load value and pre-defined fuzzy control rules, the output pressure of the air compressor is dynamically set. Based on the set output pressure, a PID controller generates a control signal, and the frequency converter receives this control signal to adjust the operating frequency of the air compressor.

2. The energy consumption optimization method for aluminum electrolysis air compressor stations based on load prediction and dynamic voltage regulation according to claim 1, characterized in that, When collecting real-time operating data of the aluminum electrolysis air compressor station system, redundant pressure and flow monitoring units are also used to collect the pressure and flow values ​​of the main pipeline tee of the gas supply network, as well as the pressure and flow values ​​of the inlet of the terminal electrolytic cell.

3. The energy consumption optimization method for aluminum electrolysis air compressor stations based on load prediction and dynamic voltage regulation according to claim 1, characterized in that, The LSTM model incorporates an attention mechanism during training. The input data of the LSTM model includes: the duty cycle of the electrolytic cell, the frequency of alumina feeding, the historical gas flow rate sliding window, the ambient temperature, and the production schedule. The output of the LSTM model is the gas load of the aluminum electrolysis air compressor station for the next 2 hours, with an output time step of 10 minutes.

4. The energy consumption optimization method for aluminum electrolysis air compressor stations based on load prediction and dynamic voltage regulation according to claim 3, characterized in that, The LSTM model includes: an input layer with a 15-dimensional feature vector; a two-layer LSTM layer, each with 64 hidden units; an additive attention layer for additive attention mechanism; and an output layer for outputting the prediction load value.

5. The energy consumption optimization method for aluminum electrolysis air compressor stations based on load prediction and dynamic voltage regulation according to claim 1, characterized in that, The specific settings of the PID controller are as follows: the input variables are the pressure deviation e and its rate of change Δe, and the fuzzy set is defined as {NB, NM, NS, ZE, PS, PM, PB}; the output variable is the pressure setpoint adjustment ΔPset, and the fuzzy set is the same as the input variable; the control rules include: if e is PB and Δe is ZE, then ΔPset is PB; if e is NS and Δe is PM, then ΔPset is NM; when the load change rate is >0.1 MPa / min, it switches to traditional PID control.

6. The energy consumption optimization method for aluminum electrolysis air compressor stations based on load prediction and dynamic voltage regulation according to claim 5, characterized in that, The control signals generated by the PID controller are as follows: For centrifugal air compressors: the target frequency is 55Hz when the load rate is <30%, 60Hz when the load rate is 30-60%, 65Hz when the load rate is 60-90%, and 70Hz when the load rate is >90%; For screw air compressors: the target frequency is 45Hz when the load rate is <30%, 50Hz when the load rate is 30-60%, 60Hz when the load rate is 60-90%, and 65Hz when the load rate is >90%.

7. The energy consumption optimization method for aluminum electrolysis air compressor stations based on load prediction and dynamic voltage regulation according to claim 6, characterized in that, When generating control signals through the PID controller, the PID controller also provides graded responses to abnormal states. The graded responses specifically include: Level 1 warning: when the temperature is >95℃, log is recorded and a prompt is made to check the cooling system; Level 2 alarm: when the exhaust pressure is >0.85 MPa, the standby unit is automatically switched; Level 3 fault: when the motor current fluctuation is >15%, the unit is shut down urgently and the faulty unit is locked.

8. The energy consumption optimization method for aluminum electrolysis air compressor stations based on load prediction and dynamic voltage regulation according to claim 7, characterized in that, The defuzzification of the PID controller uses a weighted average method, and the membership function is a trigonometric function.

9. The energy consumption optimization method for aluminum electrolysis air compressor stations based on load prediction and dynamic voltage regulation according to claim 1, characterized in that, When multiple air compressors are operating in parallel, the following steps are used to dynamically set the output pressure of the air compressors: collect the energy efficiency ratio, operating status and total system load data of each air compressor; establish an objective function with the goal of minimizing total energy consumption while taking into account equipment wear and switching frequency; and use a genetic algorithm to solve for the operating scheme of multiple air compressors that meets the current load and has the lowest energy consumption.

10. The energy consumption optimization method for aluminum electrolysis air compressor stations based on load prediction and dynamic voltage regulation according to claim 9, characterized in that, The scheduling strategy is updated every hour and controls the start and stop of any air compressor through a digital output interface; when the efficiency of any air compressor decreases, it is removed from the scheduling sequence.

Images