Intelligent optimization control method of electric arc furnace control system based on LSTM and PSO

Abstract

The invention provides an intelligent optimization control method of an electric arc furnace control system based on LSTM and PSO, and belongs to the technical field of electric furnace control. The method comprises the following steps: acquiring historical data of an electric arc furnace system; historical data of the electric arc furnace system are preprocessed, and preprocessed data are obtained; inputting the preprocessed data into an electric arc furnace state prediction model to obtain a state parameter prediction value of the electric arc furnace at a future moment; a control instruction is obtained through a particle swarm optimization method according to the electric arc furnace state parameter predicted value at the future moment and a preset target value; the electric arc furnace control system is controlled through the control instruction; and the electric arc furnace control system outputs real-time data to feed back a control instruction, and the control instruction is optimized. The electric arc furnace intelligent optimization system integrating dynamic prediction, multi-objective optimization and real-time control is constructed, the technical bottleneck of a traditional method in non-linear, time-varying and three-phase coupling control is solved, and the energy efficiency, the production efficiency and the power grid compatibility are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of electric furnace control, and particularly to an intelligent optimization method for an electric arc furnace control system based on LSTM (Long Short-Term Memory) and PSO (Particle Swarm Optimization). Background Art

[0002] With the development of industrial automation and intelligence, higher requirements are put forward for the intelligence level of the electric arc furnace control system. In the electric arc furnace control system, traditional methods mainly adopt the way of PID controller or static neural network model. This method has the advantages of low cost, simple principle, easy operation and maintenance, and can meet some basic production requirements.

[0003] However, during the smelting process, when the load of the electric arc furnace changes violently, its key parameters such as voltage and current will show strong non-linearity and time-variability characteristics. The dynamic response speed of traditional methods is slow, resulting in large fluctuations in voltage and current, seriously affecting the stable operation of the electric arc furnace. And the large fluctuations in voltage and current will cause flicker of the grid voltage, affecting the normal operation of other electrical equipment, and the harmonic content will also increase significantly. The sharp increase in harmonic content will cause the decline of the power quality of the grid, interfere with the normal operation of other equipment in the grid, and further threaten the safe and stable operation of the grid. In addition, due to the limited ability of traditional control to suppress interference, in actual production, factors such as electromagnetic interference in the production line are likely to reduce the control accuracy, further exacerbating the adverse impact on the grid.

[0004] Therefore, an intelligent optimization method for the electric arc furnace control system is needed. Summary of the Invention

[0005] In view of this, the present invention provides an intelligent optimization control method for an electric arc furnace control system based on LSTM and PSO, applying a deep neural network and an optimization algorithm to the electric furnace control system, realizing the intelligent optimization control of the electric arc furnace, and being able to improve the production efficiency of the electric arc furnace, reduce energy consumption and mitigate the damage to the grid, thereby solving the problems of insufficient dynamic response, low optimization efficiency and grid instability existing in the electric arc furnace control system.

[0006] For this purpose, the present invention provides the following technical solutions:

[0007] An intelligent optimization control method for an electric arc furnace control system based on LSTM and PSO, comprising:

[0008] Collecting historical data of the electric arc furnace system;

[0009] Preprocessing the historical data of the electric arc furnace system to obtain preprocessed data;

[0010] Input the preprocessed data into the electric arc furnace state prediction model to obtain the predicted values of the electric arc furnace state parameters at future times;

[0011] Obtain the control command according to the predicted values of the electric arc furnace state parameters at future times and the preset target values through the particle swarm optimization method;

[0012] Control the electric arc furnace control system using the control command; and feedback the real-time data output by the electric arc furnace control system to the control command to optimize the control command.

[0013] Furthermore, the electric arc furnace state prediction model includes:

[0014] The input layer receives the preprocessed data sequence;

[0015] The hidden layer includes a double-layer LSTM unit;

[0016] The first-layer LSTM unit extracts the short-term features of the preprocessed data sequence;

[0017] The second-layer LSTM unit extracts the long-term features of the preprocessed data sequence;

[0018] The fully connected layer maps the predicted values of the electric arc furnace state parameters at future times.

[0019] Furthermore, the step of obtaining the control command according to the predicted values of the electric arc furnace state parameters at future times and the preset target values through the particle swarm optimization method includes:

[0020] Determine the particle position vector;

[0021] Design an optimization objective function by integrating the prediction accuracy of the electric arc furnace state prediction model, the energy consumption of the electric arc furnace, and the current volatility;

[0022] Calculate the objective function value based on the predicted values of the electric arc furnace state parameters as the particle fitness;

[0023] Iterate based on the particle fitness to obtain the optimal position corresponding to the particle with the highest fitness in the population, and use the optimal position as the control command.

[0024] Furthermore, the particle position vector includes:

[0025] x = [ΔE a , ΔE b , ΔE c , U set

[0026] where ΔE a , ΔE b , ΔE c represents the three-phase electrode adjustment amount; U​set Represents the dynamic voltage set value.

[0027] Furthermore, the optimization objective function includes:

[0028] F = α * MSE + β * energy consumption + γ * current volatility

[0029] Where α, β, and γ are preset weight coefficients; MSE represents the mean square error of the electric arc furnace state prediction model.

[0030] Furthermore, using the control instruction to control the electric arc furnace control system includes:

[0031] Converting the three-phase electrode adjustment amount into the pulse number of the servo motor, and driving the execution through the servo amplifier to complete the adjustment of the electrode position;

[0032] Adjusting the trigger angle using the voltage set value.

[0033] Furthermore, using the real-time data output by the electric arc furnace control system to feedback the control instruction and optimize the control instruction includes:

[0034] Calculating the error between the actual state parameters of the electric arc furnace control system and the predicted values of the state parameters;

[0035] Taking the error as the basis to adjust the learning rate of the electric arc furnace state prediction model and reset the particle swarm of the particle swarm optimization method.

[0036] Furthermore, taking the error as the basis to adjust the learning rate of the electric arc furnace state prediction model and reset the initial particle swarm of the particle swarm optimization method includes:

[0037] When the error is greater than 5%, adjust the learning rate of the fully connected layer of the electric arc furnace state prediction model; at the same time, retain the particle with the highest fitness in the previous round in the reset particle swarm.

[0038] Furthermore, the historical data of the electric arc furnace system includes:

[0039] Transformer primary voltage, transformer primary current, transformer secondary voltage, transformer secondary current, arc voltage, arc current, rated current, tapping and charging signal, transformer tap position, and reactor current.

[0040] Furthermore, preprocessing the historical data of the electric arc furnace system includes:

[0041] Filling the missing values in the historical data sequence in a forward filling manner;

[0042] Using wavelet denoising to filter out electromagnetic interference;

[0043] Perform min-max normalization on historical data.

[0044] Advantages and positive effects of the present invention:

[0045] The present invention uses LSTM to predict the state of the electric arc furnace in real time; PSO adjusts the optimization direction according to the prediction error to form a closed-loop control of "prediction-optimization-execution-feedback", shortening the response delay time.

[0046] The present invention introduces a three-phase collaborative optimization strategy through the PSO algorithm, and realizes the precise decoupling of electrode actions through the multi-dimensional solution space search of the particle swarm, solving the problem of incorrect electrode actions caused by single-phase control in the prior art.

[0047] Through the deep integration of LSTM and PSO, the present invention constructs an intelligent optimization system for electric arc furnaces with dynamic prediction, multi-objective optimization and real-time control. It not only solves the technical bottlenecks of traditional methods in non-linearity, time-variation and three-phase coupling control, but also improves energy efficiency, production efficiency and grid compatibility, and has a wide range of application scenarios. Description of the drawings

[0048] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0049] Figure 1 It is the control framework diagram of the electric arc furnace in the embodiment of the present invention;

[0050] Figure 2 It is the flow chart of the intelligent optimization method for the electric arc furnace control system based on LSTM and PSO in the embodiment of the present invention;

[0051] Figure 3 It is the network structure diagram of the electric arc furnace state prediction model in the embodiment of the present invention;

[0052] Figure 4 It is the optimization flow chart of the PSO algorithm in the embodiment of the present invention. Detailed implementation manners

[0053] To enable those skilled in the art to better understand the solution of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0054] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and do not necessarily need to be used to describe a specific order or sequence. It should be understood that such data can be interchanged under appropriate circumstances so that the embodiments of the present invention described here can be implemented in an order other than those illustrated or described here. In addition, the terms "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device comprising a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0055] The present invention provides an intelligent optimization control method for an electric arc furnace control system based on LSTM and PSO. By utilizing the powerful ability of LSTM to process time series data, the complex dynamic characteristics during the operation of the electric arc furnace can be effectively captured. Through PSO to find the optimal solution in the complex solution space and optimize the control parameters. Applying the two to the electric arc furnace control system has high feasibility, solves the deficiencies of traditional methods in voltage and current control, improves the response speed and control accuracy of the system, enhances the adaptability to complex working conditions, and thus improves the overall performance and production efficiency of the electric arc furnace, while reducing the impact on the power grid and ensuring the stable operation of the power grid.

[0056] The method of the present invention specifically includes the following steps:

[0057] S1. Data collection;

[0058] Collect historical data during the operation of the electric arc furnace, including: primary voltage U1 of the transformer, primary current I1 of the transformer, secondary voltage U2 of the transformer, secondary current I2 of the transformer, arc voltage U k , arc current I arc , rated current I rated , tapping and charging signal, transformer tap position, and reactor current.

[0059] S2. Preprocess the collected historical data to obtain preprocessed data;

[0060] The preprocessing includes: data cleaning, normalization processing, and noise reduction processing.

[0061] S3. Calculate the active power P, arc power P h of the electric arc furnace and the arc voltage U k .

[0062] P = U2 × I2 × cosθ

[0063] P h = P - P loss

[0064]

[0065] where P loss is the power loss of the electric arc furnace system, I arc is the arc current, I rated is the rated current, and cosθ represents the power factor.

[0066] S4. Build a state prediction model for the electric arc furnace;

[0067] Build a state prediction model for the electric arc furnace based on LSTM; input the preprocessed data; output the state parameters of the electric arc furnace at the next moment, including: arc voltage U k , secondary current I2 of the transformer, and active power P.

[0068] S5. PSO optimization;

[0069] Use the PSO algorithm to generate a control command according to the error between the state parameters of the electric arc furnace at the next moment output by the state prediction model of the electric arc furnace and the preset target value;

[0070] The PSO optimization objective function is:

[0071] F = α * MSE + β * energy consumption + γ * current volatility

[0072] where α, β, and γ are preset weight coefficients.

[0073] S6. The control command includes: electrode adjustment amount and voltage setting value;

[0074] Use the control command optimized by PSO to control the electric arc furnace control system, realize intelligent optimization control, and dynamically adjust the control command through a real-time feedback mechanism.

[0075] In this embodiment, the time series modeling ability of LSTM effectively processes the time series data in the electric arc furnace control system, accurately predicts the state parameters of the electric arc furnace at the next moment, and improves the response speed and stability of the system; through its unique forgetting gate, input gate, and output gate control mechanisms, LSTM can effectively capture the long-term dependence relationships of parameters such as voltage, current, and power during the operation of the electric arc furnace, and models the continuously collected historical data through dynamic time steps. Compared with traditional feedforward neural networks, it does not rely on a fixed time window. The specific structure of LSTM in this embodiment includes:

[0076] Input layer: The input data has a feature dimension of 8: the secondary voltage U2 of the transformer, the secondary current I2 of the transformer, the arc voltage U k , the arc power P h , the active power P, the tapping and charging signal, the transformer tap position, and the reactor current; the above data features cover the core parameters of the electric arc furnace operation and external interference factors.

[0077] Hidden layer: A double-layer LSTM unit is adopted, with 128 nodes in each layer, and the activation function is tanh. The extraction ability of non-linear features is enhanced through the stacked structure. The first layer of LSTM captures short-term fluctuation features, such as the instantaneous current impact caused by electrode contact, and the second layer of LSTM further extracts long-term trends, such as the power gradual change during the furnace charge melting process.

[0078] Output layer: The output layer is a fully connected layer, which directly maps to the arc voltage, current, and power at the next moment, providing a high-precision prediction benchmark for PSO optimization, and reducing the error by more than 30% compared with traditional methods.

[0079] In this embodiment, PSO optimization takes the predicted value of the electric arc furnace state as the input and combines it with the preset target value to minimize energy consumption, stabilize the arc, and suppress grid flicker, and quickly searches for the optimal control parameters in the continuous space; the control parameters include: the electrode adjustment amount and the voltage setting value.

[0080] The effects of the present invention are further illustrated by specific application embodiments:

[0081] In this embodiment, an intelligent optimization control model of the electric arc furnace based on LSTM and PSO is deployed inside the industrial control computer. Through the LSTM model and the PSO optimization algorithm, accurate prediction and intelligent control of the operation state of the electric arc furnace are realized. The specific steps include:

[0082] S1. Data acquisition: Historical data is collected every 50 ms, including: the primary voltage U1 of the transformer, the primary current I1 of the transformer, the secondary voltage U2 of the transformer, the secondary current I2 of the transformer, the arc voltage U k , the arc current I arc , the rated current I rated, tapping charging signal, transformer tap position, and reactor current.

[0083] Transmitted to the industrial control computer via Ethernet, the data cache queue retains the most recent 10 s of data (200 sampling points) for processing in a dynamic time window. In this embodiment, high-frequency sampling is used to ensure the capture of transient times such as arc short circuits.

[0084] S2. Data preprocessing:

[0085] 1) Fill in missing values in a forward filling manner;

[0086] In this embodiment, the collected data is a continuous time series. Since the subsequent data is correlated with the previous data, the forward filling method is adopted to complete the missing values and reduce the interference to the data.

[0087] 2) Use wavelet denoising to filter out electromagnetic interference and retain the main features of the arc signal;

[0088] Preferably, select the db4 wavelet basis for 5-layer decomposition and perform soft threshold processing on the high-frequency coefficients; the threshold calculation formula:

[0089]

[0090] where σ is the noise standard deviation and N is the signal length.

[0091] 3) Normalization processing;

[0092] Perform Min-Max normalization on each feature to ensure that each data feature is mapped to the interval (0, 1). The normalization formula is:

[0093]

[0094] where x is the original data, x min is the minimum value in the dataset where x is located, and x max are the maximum values in the dataset where x is located, respectively. In this embodiment, x min and x max are statistically calculated based on a large dataset to avoid statistical deviations caused by too small data volume. The various data collected during the operation of the electric arc furnace have different physical meanings and dimensions, and directly using them for model training will lead to slow model convergence or even non-convergence. Normalization processing can make the data all in the interval (0, 1), eliminate the dimension, and accelerate the model convergence speed.

[0095] The input data is denoised by wavelet transform and normalized, mapped to the interval (0, 1), and the interference of sensor noise to the prediction model is reduced.

[0096] S3. Calculate the active power P, arc power P h and arc voltage U k of the electric arc furnace according to the collected transformer voltage and current signals:

[0097] P = U2 × I2 × cosθ

[0098] P h = P - P loss

[0099]

[0100] where P loss is the power loss of the electric arc furnace system, I arc is the arc current, I rated is the rated current, and cosθ represents the power factor.

[0101] S4. Construction and training of the electric arc furnace state prediction model;

[0102] As shown in combination Figure 3 , in this embodiment, the LSTM model includes: an input layer, a hidden layer, and an output layer. The input layer is used to introduce key data collected during the operation of the electric arc furnace into the model. The data covers the core parameters of the electric arc furnace operation and external interference factors that may affect its operation, providing a comprehensive information basis for subsequent model analysis. The hidden layer is the core of the model. In this embodiment, a double-layer LSTM unit is adopted. The output layer receives the information processed by the hidden layer and converts it into the final prediction result for output; directly maps to the arc voltage, current, and power at the next moment; the prediction result provides a high-precision prediction benchmark for multi-objective particle swarm optimization and is the key data for realizing the intelligent optimization control of the electric arc furnace.

[0103] Specifically, at time T, the system adjustment control quantities of each electrical parameter of the electric arc furnace are obtained through the electric arc furnace state prediction model:

[0104] Input layer: The input signals are the secondary voltage U2 of the transformer, the secondary current I2 of the transformer, the arc voltage U k , the arc power P h , the active power P, the tapping and charging signal, the transformer tap position, and the reactor current at time T and the previous 9 moments (time window length = 10).

[0105] Hidden layer: Double-layer LSTM unit, with 128 neuron nodes in each layer. This structure in this embodiment achieves a good balance between extracting complex non-linear features and computational efficiency. The first layer of LSTM captures short-term dynamic features, and the second layer of LSTM extracts long-term trends. The activation function is selected as the hyperbolic tangent function because it can map data to the [-1, 1] interval when processing data with positive and negative values, which is beneficial to the training and convergence of the model.

[0106] Output layer: Output U at time T+1 k ′ , I2 ′ and P ′ .

[0107] In this embodiment, a dynamic time window mechanism is adopted for the input signal. The industrial control computer maintains a sliding window with a length of 10. Each time new data arrives, the oldest data is removed to form a new input sequence. The window covers 0.5 seconds of historical data (50ms×10), which is sufficient to capture the dynamics of the electric arc. The time step of the input layer is fixed at 10, covering the data of the past 10 sampling periods (about 0.5 seconds), effectively capturing the dynamic characteristics of emergencies such as electric arc short circuits.

[0108] In the first layer of LSTM in the hidden layer of this embodiment, there are 128 nodes, activated by tanh, and the complete sequence is returned; the second layer of LSTM: 128 nodes, activated by tanh, and the output of the last time step is returned; a Dropout layer is added, and the Dropout rate is set to 0.2 to prevent overfitting.

[0109] For the training of the electric arc furnace state prediction model, the training data uses the historical data of the past 3 months to form a dataset to train the LSTM model. The loss function is set to mean square error + L2 regularization (λ = 0.001), the optimizer is selected as Adam (learning rate = 0.001, decay rate = 1e-6), the training period is 100, and the batch size is 32.

[0110] S5. Take the output of the electric arc furnace state parameters by the electric arc furnace prediction model as the input of the PSO, and perform iterative search within the defined search space. The search space is defined as: the range of electrode adjustment amount is ±5mm, and the dynamic adjustment value of the voltage setting value is 90%-110%. In each iteration, each particle calculates a new position and velocity according to its own velocity and position update formula. The specific steps are as follows:

[0111] 1) Parameter configuration and initialization: The particle swarm size is 50, the number of iterations is 300, the inertia weight decreases linearly (0.9 - 0.4), and the learning factors c1 = c2 = 2.0.

[0112] In this embodiment, 50 particles achieve an optimal balance between search efficiency (<3 seconds) and solution space coverage; 300 iterations can stabilize the objective function value within a fluctuation range of ±1%, meeting the real-time requirements. Initially, a high weight of 0.9 is set to enhance the global search ability, and later a low weight of 0.4 is used to achieve local optimization.

[0113] Initialize the particle position vector:

[0114] x = [ΔE a , ΔEb , ΔE c , I set

[0115] Among them, ΔE a , ΔE b , ΔE c ∈[-5, +5] mm represents the adjustment amount of the three-phase electrode; U set ∈[0.9U rated , 1.1U rated represents the dynamic voltage setting value; U rated represents the rated voltage.

[0116] 2) The optimization of the electric arc furnace control system needs to consider multiple interrelated objectives, and there are certain conflicts among these objectives. The design of the multi-objective function can find a balance among different objectives, so as to achieve global optimal control. The mean square error is used to measure the prediction accuracy of the electric arc furnace state prediction model. The state parameters of the electric arc furnace have strong nonlinearity and time-variability. By minimizing the MSE (Mean Squared Error), the dynamic response ability of the control system can be significantly improved; the energy consumption is the total electric energy during the operation of the electric arc furnace, and the proportion of the total electric energy consumption in the total cost is very large. By optimizing the energy consumption, the economic benefit can be significantly improved; the current volatility reflects that strong current fluctuations (flicker) and high-order harmonics will be generated during the operation of the electric arc furnace. By minimizing the current volatility, the damage to the power grid can be reduced.

[0117] In this embodiment, the objective function is designed as:

[0118] F = α * MSE + βEV + γ * σI

[0119] Among them:

[0120]

[0121] In the formula, N = 3 (three-phase current, voltage, power), Y actaul represents the actual state parameters of the electric arc furnace, Y pred represents the predicted value of the electric arc furnace state parameters;

[0122]

[0123] In the formula, EV represents the energy consumption, Δt = 50 ms, P arc represents the arc power, that is, P h , reflecting the power consumption during the operation of the electric arc furnace;

[0124]

[0125] In the formula, σI represents the current volatility, which is used to measure the current stability of the electric arc furnace, I​k represents the current value at the k-th sampling point represents the average current; in this embodiment, the secondary current of the transformer is used to calculate the current volatility.

[0126] According to historical data, perform weight allocation in the objective function; α takes values from 0.4 to 0.6, β takes values from 0.2 to 0.4, and γ takes values from 0.3 to 0.5; preferably, α = 0.5, β = 0.3, γ = 0.2 to ensure control accuracy.

[0127] 3) Combine Figure 4 As shown, the optimization process in this embodiment includes:

[0128] First, initialize the particle swarm, randomly generate 50 particles, uniformly distribute them in the search space, set the initial velocity to 0 to avoid premature convergence, and decode the particle position into control instructions (electrode position, voltage value);

[0129] Use the electric arc furnace state prediction model to predict the electrical parameters U k ′ 、I2 ′ and P ′ for the next 50 ms, and calculate the objective function value F as the particle fitness through the predicted electric arc furnace state parameters;

[0130] Compare the fitness of the current particle with the fitness of its historical best position. If the current fitness is better, then use the current position as its best position;

[0131] Find the particle with the highest fitness in the population and use its position as the global best position;

[0132] The formulas for updating the velocity and position are as follows:

[0133]

[0134] Where is the velocity of particle i at the (k + 1)-th iteration; ω is the inertia weight; c1 and c2 are learning factors; r1 and r2 are random numbers between [0, 1], pBest i is the historical best position searched by particle i up to the k-th generation; gBest is the best position searched by the entire particle swarm so far; is the position of particle i at the (k + 1)-th iteration, is the position of particle i at the k-th iteration.

[0135] When the number of iterations reaches 20 and the change in gBest < 1% or the number of iterations reaches 200 times, terminate the optimization and output the optimal solution as

[0136] S6. Control Instruction Issuance and Execution:

[0137] 1) First, generate an electrode adjustment instruction: Convert the finally output electrode adjustment amount into the pulse number of the servo motor, and drive and execute it through the servo amplifier to complete the position adjustment of the electrode;

[0138] 2) Then issue the voltage setting value, and write it into the OPC UA (Open Platform Communications Unified Architecture) server of the thyristor voltage regulator to adjust the trigger angle and complete the voltage setting. The trigger angle calculation formula is as follows:

[0139]

[0140] where U max is the maximum allowable voltage.

[0141] 3) Real-time feedback;

[0142] Calculate the dynamic error through the actual arc furnace state parameters Y actaul = [U k , I2, P] collected every 50ms and the predicted values of the arc furnace state parameters Y pred [U k ′ , I2 ′ , P ′ . The formula is as follows:

[0143] Error = ||Y pred - Y actaul ||2

[0144] Implement dynamic optimization according to the Error value;

[0145] If Error > 5%, trigger the online update of the arc furnace state prediction model, without changing the weights of the LSTM layer, only fine-tune the learning rate of the fully connected layer, and set it to 0.0001; and reset the PSO particle swarm: retain gBest as one of the initial particles, and randomly initialize the rest again; to achieve the purpose of dynamic optimization.

[0146] Collect the production data of a 40t arc furnace in a steel plant using the S1 - S6 methods in this embodiment and the traditional PID control method for a period of production to obtain Production Statistics Table 1;

[0147] Table 1

[0148]

[0149] Among them, from January to March 2024, traditional PID control was used, and the furnace output fluctuated less, but the power consumption was relatively high (455 - 465 kWh / t), and the smelting time was relatively long (51 - 53 minutes). From April to June 2024, intelligent control using the method of the present invention was adopted. The output of a single furnace increased to nearly 40 tons, the power consumption decreased significantly (385 - 405 kWh / t), the smelting time was shortened to 43 - 46 minutes, and the total output increased by more than 30%. Table 2 further quantifies the effect of the method of the present invention.

[0150] Table 2

[0151] Index Traditional PID method Method of the present invention Improvement range Power consumption per ton of steel (kWh / t) 460 395 14.1％ Electrode consumption 2.5 2.0 20.0％ <![CDATA[Flicker factor P st > 1.2 0.8 33.3％ Smelting cycle 52 44.7 14.0％

[0152] Benefit calculation: As can be seen from Table 1 and Table 2, when the method of the present invention is adopted, the average power consumption per ton of electricity decreases by 65 degrees. Calculated based on 13,000 tons per month, each electric furnace can save 10.14 million degrees of electricity annually. The electrode consumption decreases by 0.5 kg / t, and 78 tons of electrode consumption can be saved annually. Therefore, the method of the present invention has engineering practical value.

[0153] The method of the present invention fully utilizes the advantages of the LSTM and PSO algorithms, and realizes the efficient coordination of dynamic prediction and global optimization for the complex industrial process of the electric arc furnace with strong nonlinearity, time-variation and multi-interference characteristics. And through the method of the present invention, the production efficiency of the electric arc furnace is improved, the power grid stability is enhanced, and the production cost is reduced. Specifically: through the high-precision prediction of LSTM and the rapid optimization of PSO, the effective power utilization rate of the electric arc furnace is improved, the smelting cycle is shortened, the output is increased, and the molten steel production capacity of a single furnace is increased; the power loss is reduced compared with the traditional system, the electrode consumption decreases, and it is expected to save more than one million yuan in annual production costs; the current volatility decreases, the harmonic distortion rate is controlled, and the impact on the power grid is significantly reduced; by stabilizing the arc length and reducing the short-circuit frequency, the electrode life is extended and the operation and maintenance cost is reduced.

[0154] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. An intelligent optimization control method for an electric arc furnace control system based on LSTM and PSO, characterized in that, Including: Collecting historical data of the electric arc furnace system; Preprocessing the historical data of the electric arc furnace system to obtain preprocessed data; Inputting the preprocessed data into the electric arc furnace state prediction model to obtain predicted values of the electric arc furnace state parameters at future moments; Obtaining a control instruction according to the predicted values of the electric arc furnace state parameters at future moments and the preset target values through the particle swarm optimization method; Controlling the electric arc furnace control system by using the control instruction; and feeding back the real-time data output by the electric arc furnace control system to the control instruction to optimize the control instruction.

2. The intelligent optimization control method of an electric arc furnace control system based on LSTM and PSO according to claim 1, characterized in that, The electric arc furnace state prediction model includes: The input layer receives the preprocessed data sequence; The hidden layer includes a double-layer LSTM unit; The first-layer LSTM unit extracts the short-term features of the preprocessed data sequence; The second-layer LSTM unit extracts the long-term features of the preprocessed data sequence; The fully connected layer maps the predicted values of the electric arc furnace state parameters at future moments.

3. The intelligent optimization control method of an electric arc furnace control system based on LSTM and PSO according to claim 1, characterized in that The obtaining of the control instruction according to the predicted values of the electric arc furnace state parameters at future moments and the preset target values through the particle swarm optimization method includes: Determining the particle position vector; Designing an optimization objective function by integrating the prediction accuracy of the electric arc furnace state prediction model, the energy consumption of the electric arc furnace, and the current volatility; Calculating the objective function value according to the predicted values of the electric arc furnace state parameters as the particle fitness; Performing iteration based on the particle fitness to obtain the optimal position corresponding to the particle with the highest fitness in the population, and taking the optimal position as the control instruction.

4. The intelligent optimization control method of an electric arc furnace control system based on LSTM and PSO according to claim 3, characterized in that, The particle position vector includes: x = [ΔE a , ΔE b , ΔE c , U set ​ Among them, ΔE a , ΔE b , ΔE c represents the three-phase electrode adjustment amount; U set represents the dynamic voltage set value.

5. The intelligent optimization control method of an electric arc furnace control system based on LSTM and PSO according to claim 3, characterized in that, The optimization objective function includes: F = α * MSE + β * energy consumption + γ * current volatility Wherein, α, β, and γ are preset weight coefficients; MSE represents the mean square error of the electric arc furnace state prediction model.

6. The intelligent optimization control method of an electric arc furnace control system based on LSTM and PSO according to claim 1, characterized in that, The controlling of the electric arc furnace control system by using the control instruction includes: Converting the three-phase electrode adjustment amount into the servo motor pulse number and driving the execution through the servo amplifier to complete the adjustment of the electrode position; Adjusting the trigger angle by using the voltage set value.

7. The intelligent optimization control method of an electric arc furnace control system based on LSTM and PSO according to claim 1, characterized in that, The feeding back of the real-time data output by the electric arc furnace control system to the control instruction to optimize the control instruction includes: Calculating the error between the actual state parameters of the electric arc furnace control system and the predicted values of the state parameters; Adjusting the learning rate of the electric arc furnace state prediction model based on the error and resetting the particle swarm of the particle swarm optimization method at the same time.

8. The intelligent optimization control method of an electric arc furnace control system based on LSTM and PSO according to claim 7, characterized in that, Adjusting the learning rate of the electric arc furnace state prediction model based on the error and resetting the initial particle swarm of the particle swarm optimization method at the same time includes: When the error is greater than 5%, adjusting the learning rate of the fully connected layer of the electric arc furnace state prediction model; at the same time, retaining the particle with the highest fitness in the previous round in the reset particle swarm.

9. The intelligent optimization control method of an electric arc furnace control system based on LSTM and PSO according to claim 1, characterized in that, The historical data of the electric arc furnace system includes: Transformer primary voltage, transformer primary current, transformer secondary voltage, transformer secondary current, arc voltage, arc current, rated current, tapping and charging signal, transformer tap position, and reactor current.

10. The intelligent optimization control method of an electric arc furnace control system based on LSTM and PSO according to claim 1, characterized in that The preprocessing of the historical data of the electric arc furnace system includes: Filling the missing values in the historical data sequence in a forward filling manner; Filtering out electromagnetic interference by using wavelet denoising; Performing min-max normalization processing on the historical data.

Images