Intelligent control method, system and equipment for kiln atmosphere

Abstract

The invention relates to the technical field of kilns, and provides a kiln atmosphere intelligent control method, system and equipment, and the method comprises the steps: obtaining a temperature mean square error term, a pressure fluctuation variance term, an air volume adjustment regular term, a gas consumption penalty term, a carbon monoxide concentration penalty term and an oxygen concentration penalty term; constructing an LSTM-CNN hybrid neural network model, and constructing a loss function of the LSTM-CNN hybrid neural network model according to a temperature mean square error term, a pressure fluctuation variance term, an air volume adjustment regular term, a gas consumption penalty term, a carbon monoxide concentration penalty term and an oxygen concentration penalty term; the LSTM-CNN hybrid neural network model is trained, and an optimized control parameter group is acquired according to the trained LSTM-CNN hybrid neural network model; wherein the control parameter group at least comprises air volume control parameters of the target area. According to the invention, the atmosphere of the kiln can be automatically controlled, and the stability of atmosphere control and the product quality are ensured.

Description

Technical Field

[0001] The present invention relates to the technical field of kilns, and in particular, to an intelligent control method, system and device for kiln atmosphere. Background Art

[0002] The roller hearth kiln is an essential device for tile production. There are many factors affecting the product firing, among which the three major environmental factors of temperature field, pressure field and atmosphere field are the most important. These three are interrelated and interact with each other, and the temperature field is the most core factor among the key points. Different intervals have different requirements for the temperature field, thus forming a longitudinal distribution curve. For the same interval cross-section of the roller hearth kiln, the less the differences in temperature, pressure and atmosphere, the better, so as to reduce or avoid differences in the product calcination effect (size, deformation, color, etc.), which is also one of the important indicators for measuring the technical level of the kiln.

[0003] At present, the industry generally automatically controls the "temperature in the heating / holding interval" by adjusting the fuel flow of the burners, but cannot automatically control the atmosphere (CO, O2 concentration). Although some kilns are equipped with instruments such as flue gas composition analyzers, they can only be used for display and cannot achieve automatic control, and manual operation is required.

[0004] Therefore, it is necessary to develop an intelligent control technology for kiln atmosphere to automatically control the kiln atmosphere and improve the product calcination quality. Summary of the Invention

[0005] Based on this, in order to automatically control the kiln atmosphere to improve the product calcination quality, the present invention provides an intelligent control method, system and device for kiln atmosphere. By constructing a loss function for multi-objective combination, it integrates multi-dimensional parameter information such as temperature, pressure and atmosphere, and uses the carbon monoxide concentration and oxygen concentration as penalty terms of the loss function, so as to automatically control the kiln atmosphere and improve the product calcination quality. The specific technical solutions are as follows:

[0006] An intelligent control method for kiln atmosphere, which includes:

[0007] Obtain the temperature information, pressure information, air volume information, gas consumption information, carbon monoxide concentration information and oxygen concentration information of the target area;

[0008] Respectively obtain the temperature mean square error term, pressure fluctuation variance term, air volume adjustment regularization term, gas consumption penalty term, carbon monoxide concentration penalty term and oxygen concentration penalty term according to the temperature information, pressure information, air volume information, gas consumption information, carbon monoxide concentration information and oxygen concentration information;

[0009] Build an LSTM-CNN hybrid neural network model, and construct a loss function for the LSTM-CNN hybrid neural network model based on the temperature mean square error term, the pressure fluctuation variance term, the air volume adjustment regularization term, the gas consumption penalty term, the carbon monoxide concentration penalty term, and the oxygen concentration penalty term;

[0010] Train the LSTM-CNN hybrid neural network model, and obtain an optimized control parameter group according to the trained LSTM-CNN hybrid neural network model;

[0011] Among them, the control parameter group includes at least the air volume control parameter of the target area.

[0012] In the intelligent control method for the kiln furnace atmosphere, a loss function of the LSTM-CNN hybrid neural network model is constructed through the temperature mean square error term, the pressure fluctuation variance term, the air volume adjustment regularization term, the gas consumption penalty term, the carbon monoxide concentration penalty term, and the oxygen concentration penalty term, which integrates multi-dimensional information such as the temperature, pressure, atmosphere, and gas consumption of the kiln furnace, can reduce the differences in temperature, pressure, and atmosphere during the kiln furnace calcination process, and improve the product calcination quality and overall production efficiency; by creatively adding the carbon monoxide concentration penalty term and the oxygen concentration penalty term to the loss function to inhibit and constrain the carbon monoxide concentration and oxygen concentration during the product calcination process, the stability of the atmosphere control can be ensured and the product quality can be guaranteed; based on the trained hybrid neural network model, an optimized control parameter group is obtained, and the control parameter group includes at least the air volume control parameter of the target area, which can automatically control the kiln furnace atmosphere.

[0013] Preferably, the specific method for obtaining the air volume adjustment regularization term according to the air volume information includes:

[0014] Obtain the real-time air volume distribution value at the t-th time step and the air volume distribution value at the (t-1)-th time step, and calculate the absolute value of the air volume change between adjacent time steps;

[0015] Obtain the preset length of the continuous monitoring time window, and calculate the cumulative adjusted air volume according to the time window length and the absolute value of the air volume change between adjacent time steps;

[0016] Obtain the air volume adjustment L1 regularization term according to the cumulative adjusted air volume.

[0017] Preferably, the specific method for obtaining the gas consumption penalty term according to the gas consumption information includes:

[0018] Obtain the current gas consumption and the historical optimal gas consumption;

[0019] Obtain the gas consumption penalty term according to the current gas consumption and the historical optimal gas consumption;

[0020] Among them, the gas consumption penalty term is used to trigger a penalty when the current gas consumption exceeds the historical optimal gas consumption.

[0021] Preferably, the specific method for obtaining the carbon monoxide concentration penalty term according to the carbon monoxide concentration information includes:

[0022] Obtain the carbon monoxide concentration values at multiple time points;

[0023] According to the cumulative mean between the multiple carbon monoxide concentration values and the preset carbon monoxide concentration safety threshold;

[0024] Obtain the carbon monoxide concentration penalty term according to the cumulative mean.

[0025] Preferably, the specific method for obtaining the oxygen concentration penalty term according to the oxygen concentration information includes:

[0026] Obtain the oxygen concentration values and the set temperature values in multiple temperature zones;

[0027] Calculate the change rate of the oxygen concentration between adjacent temperatures according to the oxygen concentration value and the set temperature value;

[0028] Sum the squares of the absolute values of the change rates to obtain the oxygen concentration penalty term;

[0029] Among them, the change rate represents the difference in oxygen concentration caused by a unit temperature change.

[0030] Preferably, the loss function

[0031] L = α·MSE T + β·Var p + γ|ΔF| + λ·E gas + η·C CO + μ·C O2 ;

[0032] Among them, MSE T represents the temperature mean square error term, Var p represents the pressure fluctuation variance term, |ΔF| represents the air volume adjustment regularization term, E gas represents the gas consumption penalty term, C CO represents the carbon monoxide concentration penalty term, C O2 represents the oxygen concentration penalty term, and α, β, γ, λ, η, μ respectively represent the weight coefficients of the temperature mean square error term, the pressure fluctuation variance term, the air volume adjustment regularization term, the gas consumption penalty term, the carbon monoxide concentration penalty term, and the oxygen concentration penalty term.

[0033] Preferably, the carbon monoxide concentration penalty term The oxygen concentration penalty term

[0034] Among them, n represents the number of time points, e represents the natural constant, t represents the index variable of the time point, and CO t represents the carbon monoxide concentration value at the t-th time point, and CO max represents the carbon monoxide concentration safety threshold, k represents the penalty intensity adjustment coefficient, m represents the number of temperature zones divided in the target area, and O2 i+1 、O2 i represent the oxygen concentration values in the (i + 1)-th and i-th temperature zones respectively, and T i+1 、T i represent the set temperature values in the (i + 1)-th and i-th temperature zones respectively.

[0035] A kiln atmosphere intelligent control system is used to implement the described kiln atmosphere intelligent control method, and it includes:

[0036] An information acquisition module is used to acquire temperature information, pressure information, air volume information, gas consumption information, carbon monoxide concentration information, and oxygen concentration information of the target area;

[0037] A loss function term acquisition module is used to acquire the temperature mean square error term, pressure fluctuation variance term, air volume adjustment regularization term, gas consumption penalty term, carbon monoxide concentration penalty term, and oxygen concentration penalty term respectively according to the temperature information, pressure information, air volume information, gas consumption information, carbon monoxide concentration information, and oxygen concentration information;

[0038] A construction module is used to construct an LSTM-CNN hybrid neural network model, and construct the loss function of the LSTM-CNN hybrid neural network model according to the temperature mean square error term, pressure fluctuation variance term, air volume adjustment regularization term, gas consumption penalty term, carbon monoxide concentration penalty term, and oxygen concentration penalty term;

[0039] A control parameter acquisition module is used to train the LSTM-CNN hybrid neural network model, and acquire an optimized control parameter set according to the trained LSTM-CNN hybrid neural network model;

[0040] Among them, the control parameter set at least includes the air volume control parameter of the target area.

[0041] Preferably, the construction module constructs the loss function according to the formula

[0042] L = α·MSE T + β·Var p + γ|ΔF| + λ·E gas + η·C CO + μ·C O2 to construct the loss function;

[0043] Among them, MSE T represents the temperature mean square error term, and Varp represents the pressure fluctuation variance term, |ΔF| represents the air volume adjustment regularization term, E gas represents the gas consumption penalty term, C CO represents the carbon monoxide concentration penalty term, C O2 The oxygen concentration penalty term, and α, β, γ, λ, η, μ respectively represent the weight coefficients of the temperature mean square error term, the pressure fluctuation variance term, the air volume adjustment regularization term, the gas consumption penalty term, the carbon monoxide concentration penalty term, and the oxygen concentration penalty term.

[0044] An intelligent control device for kiln atmosphere, comprising:

[0045] A memory storing executable instructions;

[0046] A controller for executing the executable instructions and implementing the intelligent control method for kiln atmosphere described above. Brief Description of the Drawings

[0047] The present invention can be further understood from the following description in conjunction with the drawings. The components in the drawings are not necessarily drawn to scale, but the emphasis is placed on showing the principles of the embodiments. In different views, the same reference numerals designate corresponding parts.

[0048] Figure 1 is the overall flowchart of an intelligent control method for kiln atmosphere in an embodiment of the present invention;

[0049] Figure 2 is the flowchart of the specific method for obtaining the air volume adjustment regularization term in an embodiment of the present invention;

[0050] Figure 3 is the flowchart of the specific method for obtaining the gas consumption penalty term in an embodiment of the present invention;

[0051] Figure 4 is the flowchart of the specific method for obtaining the carbon monoxide concentration penalty term in an embodiment of the present invention;

[0052] Figure 5 is the flowchart of the specific method for obtaining the oxygen concentration penalty term in an embodiment of the present invention;

[0053] Figure 6 is the overall structural diagram of an intelligent control system for kiln atmosphere in an embodiment of the present invention. Detailed Description of the Embodiments

[0054] In order to make the objectives, technical solutions and advantages of the present invention clearer and more understandable, the present invention will be further described in detail below in conjunction with its embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and do not limit the protection scope of the present invention.

[0055] It should be noted that when an element is referred to as "fixed to" another element, it can be directly on the other element or there may also be an intermediate element. When an element is considered to be "connected" to another element, it can be directly connected to the other element or there may be an intermediate element at the same time. The terms "vertical", "horizontal", "left", "right" and similar expressions used herein are only for the purpose of illustration and do not represent the only implementation.

[0056] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the technical field to which this invention belongs. The terms used in the description of this invention herein are only for the purpose of describing specific implementations and are not intended to limit this invention. The term "and / or" used herein includes any and all combinations of one or more of the related listed items.

[0057] The "first" and "second" mentioned in this invention do not represent specific quantities and orders, but are only used for name distinction.

[0058] Before elaborating on the embodiments of this invention, a brief introduction to the prior art will be given first.

[0059] The inside of the kiln is a complex and changeable fluid mechanics system. When the state inside the kiln deviates (such as temperature), during manual correction operations, it will affect the change of other parameters. When the change is very small, it cannot be detected by the naked eye (and may not have an obvious impact on the product firing), so it naturally cannot be corrected; when the change is obvious (such as too long or too short flame, turbid atmosphere in the interval, zero pressure surface drift, etc.), even when the product comes out of the kiln, color and size deviations are found, and then the deviation is corrected through manual operations. Therefore, this manual firing control method has an obvious feedback lag and is one of the important factors causing product quality fluctuations and instability.

[0060] The roller hearth kiln is an essential equipment for tile production. There are many factors affecting the product firing, among which the three major environmental factors of temperature field, pressure field and atmosphere field are the most important. They are interrelated and interact with each other, and the temperature field is the most core element among the key points. Different intervals have different requirements for the temperature field, thus forming a longitudinal distribution curve. For the same interval cross-section of the roller hearth kiln, the less difference in temperature, pressure and atmosphere, the better, so as to reduce or avoid differences in the calcination effect (size, deformation, color, etc.) of the product, which is also one of the important indicators to measure the technical level of the kiln.

[0061] When the kiln is adjusted to an optimal state and can remain stable, it is the ideal pursuit of kiln control. However, in actual production, affected by external factors such as weather changes, fuel pressure or calorific value fluctuations, and production scheduling, the temperature field inside the kiln changes constantly. At this time, it is necessary to adjust the fuel flow (pressure), combustion air volume (pressure), flue gas extraction force, etc. in a timely manner to make up for the state deviation caused by external factors and maintain the stability of the temperature field inside the kiln.

[0062] Currently, the industry generally automatically controls the "temperature in the heating / holding interval" by adjusting the fuel flow of the burner, but cannot automatically control the atmosphere (CO, O2 concentration). Although some kilns are equipped with instruments such as flue gas composition analyzers, they can only be used for display and cannot achieve automatic control, and manual operation is required.

[0063] To achieve the automatic control of the kiln atmosphere, an embodiment of the present invention provides an intelligent control method for the kiln atmosphere, as Figure 1 shown, which includes the following steps:

[0064] S1, obtaining temperature information, pressure information, air volume information, gas consumption information, carbon monoxide concentration information, and oxygen concentration information of the target area.

[0065] For the target area, such as the heating / holding interval, sensors such as thermocouples, pressure sensors, gas flow meters, and flue gas composition analyzers can be arranged every 2 meters to synchronously collect data including temperature information, pressure information, air volume information, gas consumption information, carbon monoxide concentration information, and oxygen concentration information. An infrared thermal imager can also be installed in the high-temperature section to monitor the temperature field distribution on the surface of the brick blank.

[0066] For the proportion allocation of the flame-retardant air volume in the target area such as the heating / holding interval, the initial air volume value can be set in a ratio of 3:7 (heating: holding) or 4:6 (heating: holding), etc., and then dynamically adjusted by the hybrid neural network model.

[0067] S2, respectively obtaining the temperature mean square error term, pressure fluctuation variance term, air volume adjustment regularization term, gas consumption penalty term, carbon monoxide concentration penalty term, and oxygen concentration penalty term according to the temperature information, pressure information, air volume information, gas consumption information, carbon monoxide concentration information, and oxygen concentration information.

[0068] Preferably, the method for obtaining the temperature mean square error term according to the temperature information includes: obtaining the predicted temperature curve and the target temperature curve of the target area, and obtaining the temperature mean square error term according to the predicted temperature curve and the target temperature curve where N' represents the number of temperature sampling points, T pred,i 、T target,iThey respectively represent the predicted temperature and the target temperature of the i-th sampling point. This temperature mean square error term is used as the main optimization objective to ensure that the temperatures in each temperature zone of the roller hearth kiln strictly follow the preset process curve, and to ensure the heat treatment effect of the brick blank in a specific temperature range (such as 1050 - 1150 °C for glaze formation). For example, if the target temperature curve requires an accuracy of ±8 °C in the high temperature section (>1000 °C), the temperature mean square error term directly measures the deviation between the actual temperature and the target. The weight coefficient α of the temperature mean square error term is usually given a relatively high weight (such as α = 0.5, etc.), reflecting the core position of temperature control. During the glaze formation stage (1050 - 1150 °C), α may be dynamically increased to 0.7 to enhance the glaze layer quality.

[0069] The specific method for obtaining the pressure fluctuation variance term based on pressure information includes: obtaining the interval average pressure value P' in the target area and the real-time pressure values P of N'' sampling points i , and obtaining the pressure fluctuation variance term based on the interval average pressure value in the target area and the real-time pressure values of N'' sampling points The pressure fluctuation variance term is used to measure the pressure stability of the target area of the kiln, suppress abnormal pressure fluctuations (such as sudden changes >10 Pa or 20 Pa), ensure that the pressure is stable within the process allowable range (such as ±5 Pa), and avoid brick blank cracking or glaze defects caused by air flow disturbances. The weight coefficient β of the pressure fluctuation variance term is usually set to 0.3, and the fan adjustment strategy is indirectly optimized by penalizing the variance term to avoid chain reactions caused by pressure mutations.

[0070] S3. Construct an LSTM-CNN hybrid neural network model, and construct a loss function of the LSTM-CNN hybrid neural network model based on the temperature mean square error term, the pressure fluctuation variance term, the air volume adjustment regularization term, the gas consumption penalty term, the carbon monoxide concentration penalty term, and the oxygen concentration penalty term.

[0071] For the LSTM-CNN hybrid neural network model, it includes an LSTM layer for capturing time series features (such as the temperature gradient change rate) and a CNN layer for extracting spatial distribution features (such as the temperature difference across the cross-section of the kiln). The LSTM layer may include 128 nodes, and the CNN layer includes 3X1 convolutional kernels. The output layer of the LSTM-CNN hybrid neural network model can be selected as 2 nodes, corresponding to the correction coefficients of the air volume of the flame-retardant fans in the heating interval and the heat preservation interval respectively. Of course, according to the number of intervals divided in the target area, the number of output layer nodes can be adjusted accordingly. Or rather, according to the number of intervals divided in the target area and the types of control parameters included in the preferred control parameter group obtained based on the trained LSTM-CNN hybrid neural network model, the number of output layer nodes is adjusted accordingly.

[0072] S4. Train the LSTM-CNN hybrid neural network model and obtain an optimized control parameter set according to the trained LSTM-CNN hybrid neural network model. Among them, the control parameter set at least includes the air volume control parameter of the target area.

[0073] For the training of the LSTM-CNN hybrid neural network model, collect historical production data including temperature information, pressure information, air volume information, gas consumption information, carbon monoxide concentration information, oxygen concentration information, etc. (such as historical production data covering normal and abnormal working conditions for 3 months or 5 months), and construct a training data set with multiple groups of sample data such as 10,000 groups, 100,000 groups, or 200,000 groups. For the sample data, preprocessing such as normalization, handling missing values, adding Gaussian noise, and time series translation can be performed. Since preprocessing the sample data and label annotation are conventional technical means in this field, they will not be elaborated here.

[0074] The input features of the LSTM-CNN hybrid neural network model include but are not limited to temperature sequences, pressure gradients, oxygen concentrations, and carbon monoxide concentrations, and the output labels include but are not limited to the target air volume control parameter (such as the optimal air volume ratio), target temperature value, target air pressure value, and gas flow rate. In this way, based on the loss function and the LSTM-CNN hybrid neural network model, multiple targets, specifically the control parameters of the target area of the kiln, can be optimized collaboratively, avoiding the overfitting problem caused by optimizing a single target and improving the overall production efficiency of the target area of the kiln.

[0075] The loss function comprehensively considers multi-dimensional information including temperature information, pressure information, air volume information, gas consumption information, carbon monoxide concentration information, and oxygen concentration information. For the existing kiln control technology that only automatically controls the temperature, it can perform automated and intelligent control on parameters such as kiln temperature, air pressure, and atmosphere, and further optimize multiple control parameters including temperature, air pressure, and atmosphere that affect the quality of kiln products collaboratively, improving the overall production efficiency of the kiln.

[0076] In summary, in the intelligent control method of the kiln furnace atmosphere, the loss function of the LSTM-CNN hybrid neural network model is constructed through the temperature mean square error term, the pressure fluctuation variance term, the air volume adjustment regularization term, the gas consumption penalty term, the carbon monoxide concentration penalty term, and the oxygen concentration penalty term. By integrating multi-dimensional information such as the temperature, pressure, atmosphere, and gas consumption of the kiln furnace, it can reduce the differences in temperature, pressure, and atmosphere during the calcination process of the kiln furnace, improve the calcination quality of the products and the overall production efficiency. By creatively adding the carbon monoxide concentration penalty term and the oxygen concentration penalty term to the loss function to suppress and constrain the carbon monoxide concentration and oxygen concentration during the product calcination process, the stability of the atmosphere control can be ensured and the product quality can be guaranteed. Based on the trained hybrid neural network model, an optimized control parameter group is obtained, and the control parameter group at least includes the air volume control parameter of the target area, which can automatically control the kiln furnace atmosphere.

[0077] As a preferred technical solution, as Figure 2 shown, in step S2, the specific method for obtaining the air volume adjustment regularization term according to the air volume information includes:

[0078] S21, obtain the real-time air volume distribution value Ft t at the t-th time step and the air volume distribution value Ft-1 t-1 at the (t - 1)-th time step, and calculate the absolute value of the air volume change between adjacent time steps |Ft t - Ft-1 t-1 |;

[0079] S22, obtain the preset continuous monitoring time window length T (such as 30 minutes or 45 minutes, etc.), and calculate the cumulative adjusted air volume according to the time window length and the absolute value of the air volume change between adjacent time steps;

[0080] S23, obtain the air volume adjustment L1 regularization term according to the cumulative adjusted air volume

[0081] wherein, the unit of the real-time air volume distribution value Ft t at the t-th time step is m 3 / h. According to the actual situation, it can include the air volume in the low-temperature section (600 - 900 °C) and the air volume in the high-temperature section (900 - 1290 °C). The absolute value of the air volume change between adjacent time steps |Ft t - Ft-1 t-1 | represents the absolute value of the air volume change between adjacent time steps, reflecting the intensity of the control action.

[0082] By setting this air volume adjustment L1 regularization term it has the following functions:

[0083] 1. Suppress high-frequency oscillations. The L1 regularization term restricts the sum of the absolute values (instead of the sum of squares) of the adjustment amplitudes, which can more effectively limit high-frequency small fluctuations. For example, when the air volume correction coefficient in the low-temperature section of the neural network output suddenly changes from 1.05 to 0.95, the L1 term directly constrains this step change of 0.1. Compared with the L2 regularization term (penalizing the squared difference), the L1 term has a stronger suppression effect on continuous small fluctuations (such as ±0.02).

[0084] 2. Protect the actuator. For the regulating valve that controls the air volume of the combustion-supporting fan in the target area, frequent adjustment will lead to an increase in the gear wear rate. By setting the weight coefficient γ of the L1 regularization term for air volume adjustment (for example, setting γ to 0.1), the average daily operation times of the valve can be reduced from 1200 times to less than 400 times.

[0085] 3. Energy consumption optimization mechanism. After introducing the L1 regularization term for air volume adjustment, the system can automatically suppress unnecessary air volume redistribution and reduce the energy consumption per ton of porcelain.

[0086] Furthermore, for the weight coefficient γ of the L1 regularization term for air volume adjustment, it can be adjusted to 0.05 during the glaze formation stage (1050 - 1150 °C). The length T of the continuous monitoring time window can be shortened to 10 minutes during the empty kiln transition period. The L1 regularization term for air volume adjustment and the temperature mean square error term act synergistically to extend the service life of the fan while ensuring the temperature control accuracy.

[0087] As a preferred technical solution, as Figure 3 shown, in step S2, the specific method for obtaining the gas consumption penalty term according to the gas consumption information includes:

[0088] S24, obtain the current gas consumption Q and the historical optimal gas consumption Q'.

[0089] The current gas consumption Q can be understood as the actual gas energy value consumed by the kiln per unit time (such as per hour), and the unit is usually kJ / ton (energy consumption per ton of ceramic tile production). It can be monitored in real time by a gas flow meter and converted into an equivalent energy value in combination with the calorific value of the gas. The historical optimal gas consumption Q' can be understood as the lowest gas consumption benchmark value recorded by the system under the same type of working conditions, representing the limit state of the equipment operation efficiency, and it can be set by technicians according to the actual production data.

[0090] S25, obtain the gas consumption penalty term Egas = max(0, Q - Q') according to the current gas consumption and the historical optimal gas volume.

[0091] Among them, the gas consumption penalty term is used to trigger a penalty when the current gas consumption exceeds the historical optimal gas consumption. It can adjust the neural network weights through backpropagation to suppress inefficient combustion behavior. The weight coefficient λ of the gas consumption penalty term can be set to 0.1 and automatically reduced to 0.05 during the glaze formation stage to avoid overconstraint. Through this gas consumption penalty term, the energy efficiency and temperature / pressure control objectives can be balanced, and the gas consumption per ton of porcelain can be reduced.

[0092] As a preferred technical solution, as Figure 4 shown, the specific method for obtaining the carbon monoxide concentration penalty term according to the carbon monoxide concentration information includes:

[0093] S26, obtaining carbon monoxide concentration values at multiple time points;

[0094] S27, according to the cumulative mean between multiple carbon monoxide concentration values and a preset carbon monoxide concentration safety threshold;

[0095] S28, obtaining the carbon monoxide concentration penalty term according to the cumulative mean The carbon monoxide concentration penalty term is used to reflect the cumulative effect of CO concentration exceeding the standard. The larger its value, the more serious the exceeding standard phenomenon indicates.

[0096] Among them, n represents the number of time points, e represents the natural constant, t represents the index variable of the time point, CO t represents the carbon monoxide concentration value at the t-th time point, CO max represents the carbon monoxide concentration safety threshold, and k represents the penalty intensity adjustment coefficient. The carbon monoxide concentration safety threshold can be set according to production process safety standards or environmental protection requirements, such as set to 200 ppm. The penalty intensity adjustment coefficient is used to control the growth rate of the exponential function. The larger its value, the stronger the penalty for the same amount of exceeding the standard. Generally speaking, k is set to 0.5.

[0097] The carbon monoxide concentration penalty term has an exponential growth characteristic. When CO t ≤CO max , the exponential term has a weak penalty effect; when CO t >CO max , the exponential term rapidly increases, imposing a non-linear penalty on the exceeding standard concentration and effectively suppressing abnormal working conditions. In addition, through the mean calculation, the penalty term not only pays attention to the exceeding standard at a single time point but also reflects the overall control stability. For example, if there are multiple slight exceedings (such as 210 ppm) during a certain period, its penalty value will be significantly higher than the case of occasional serious exceeding (such as 300 ppm) but meeting the standard at other times.

[0098] Adjust the weight coefficient η of the carbon monoxide concentration penalty term through dynamic allocation. For example, set η to 0.2 in the high-temperature section, which can strengthen the CO control accuracy in the key temperature zone and reduce the average value of glaze color difference.

[0099] Through this carbon monoxide concentration penalty and integrating the index penalty mechanism, the CO concentration can be effectively controlled, and the process parameters for kiln atmosphere protection can be optimized.

[0100] As a preferred technical solution, as Figure 5 shown, in step S2, the specific method for obtaining the oxygen concentration penalty term according to the oxygen concentration information includes:

[0101] S29, obtain the oxygen concentration values and set temperature values of multiple temperature zones;

[0102] S210, calculate the change rate of oxygen concentration between adjacent temperatures according to the oxygen concentration value and the set temperature value. Among them, the change rate represents the oxygen concentration difference caused by a unit temperature change.

[0103] S211, sum the squares of the absolute values of the change rates to obtain the oxygen concentration penalty term

[0104]

[0105] where m represents the number of temperature zones divided in the target area, O2 i+1 、O2 i respectively represent the oxygen concentration values of the (i + 1)-th and i-th temperature zones, and T i+1 、T i respectively represent the set temperature values of the (i + 1)-th and i-th temperature zones.

[0106] The oxygen concentration value is used to quantify the oxygen concentration difference between adjacent temperature zones and reflect the stability of the oxidation reaction. The set temperature value characterizes the temperature change gradient between temperature zones and can be understood as a key parameter for driving gas diffusion. m represents the number of temperature zones divided in the target area. If the target area is divided into 2 atmosphere control areas, namely the low-temperature section and the high-temperature section, then m = 2.

[0107] Reflect the oxygen concentration difference caused by a unit temperature change, and the square operation amplifies the influence of abnormal fluctuations. By summing the squared gradient values of all temperature zones, the oxygen concentration penalty term imposes penalties on the following situations: 1. The sudden drop of oxygen concentration in the high-temperature zone: may cause Fe 3 + to be reduced to Fe 2 +, resulting in glaze color difference; 2. The sudden rise of oxygen concentration in the low-temperature zone: may disrupt the carbon oxidation balance and produce excessive carbon monoxide.

[0108] Here, the oxygen concentration penalty term can reduce the glaze layer bubbles and lower the average color difference by restricting the oxygen concentration gradient.

[0109] The loss function L = α·MSE T + β·Var p + γ|ΔF| + λ·E gas + η·C CO + μ·C O2 .

[0110] Among them, MSE T represents the temperature mean square error term, Var p represents the pressure fluctuation variance term, |ΔF| represents the air volume adjustment regularization term, E gas represents the gas consumption penalty term, C CO represents the carbon monoxide concentration penalty term, C O2 represents the oxygen concentration penalty term, and α, β, γ, λ, η, μ respectively represent the weight coefficients of the temperature mean square error term, the pressure fluctuation variance term, the air volume adjustment regularization term, the gas consumption penalty term, the carbon monoxide concentration penalty term, and the oxygen concentration penalty term.

[0111] Specifically, α, β, γ, λ, η, μ can be set to 0.5, 0.3, 0.1, 0.05, and 0.05 respectively. Of course, according to different actual situations and production requirements, the values of α, β, γ, λ, η, μ can also be adjusted accordingly.

[0112] In the said loss function, it includes the temperature mean square error term, the pressure fluctuation variance term, the air volume adjustment regularization term, the gas consumption penalty term, the carbon monoxide concentration penalty term, and the oxygen concentration penalty term. By jointly optimizing the temperature, pressure, air volume, gas energy consumption, and atmosphere parameters (including carbon monoxide concentration and oxygen concentration), it can achieve multi-objective collaborative optimization, avoid overfitting of a single objective, and improve the overall production efficiency.

[0113] An embodiment of the present invention also provides a kiln furnace atmosphere intelligent control system for implementing the said kiln furnace atmosphere intelligent control method, as Figure 6 shown, which includes an information acquisition module, a loss function term acquisition module, a construction module, and a control parameter acquisition module.

[0114] The information acquisition module is used to acquire the temperature information, pressure information, air volume information, gas consumption information, carbon monoxide concentration information, and oxygen concentration information of the target area; the loss function term acquisition module is used to acquire the temperature mean square error term, the pressure fluctuation variance term, the air volume adjustment regularization term, the gas consumption penalty term, the carbon monoxide concentration penalty term, and the oxygen concentration penalty term respectively according to the temperature information, pressure information, air volume information, gas consumption information, carbon monoxide concentration information, and oxygen concentration information.

[0115] The construction module is used to construct an LSTM-CNN hybrid neural network model, and construct a loss function of the LSTM-CNN hybrid neural network model according to the temperature mean square error term, the pressure fluctuation variance term, the air volume adjustment regularization term, the gas consumption penalty term, the carbon monoxide concentration penalty term, and the oxygen concentration penalty term; the control parameter acquisition module is used to train the LSTM-CNN hybrid neural network model and obtain an optimized control parameter set according to the trained LSTM-CNN hybrid neural network model; wherein, the control parameter set at least includes the air volume control parameter of the target area.

[0116] For the target area, it can be the low-temperature section and the high-temperature section in the furnace heating / insulation interval, and multiple flue gas component information collection points are set in the low-temperature section and the high-temperature section to collect the carbon monoxide concentration and the oxygen concentration, and the opening degree of the automatic valve of the main air duct of the combustion-supporting fan is controlled to automatically adjust the proportional distribution of the combustion-supporting air volume in the low-temperature section and the high-temperature section.

[0117] The construction module constructs the loss function according to the formula L = α·MSE T +β·Var p +γ|ΔF|+λ·E gas +η·C CO +μ·C O2 Construct the loss function.

[0118] Wherein, MSE T represents the temperature mean square error term, Var p represents the pressure fluctuation variance term, |ΔF| represents the air volume adjustment regularization term, E gas represents the gas consumption penalty term, C CO represents the carbon monoxide concentration penalty term, C O2 represents the oxygen concentration penalty term, and α, β, γ, λ, η, μ respectively represent the weight coefficients of the temperature mean square error term, the pressure fluctuation variance term, the air volume adjustment regularization term, the gas consumption penalty term, the carbon monoxide concentration penalty term, and the oxygen concentration penalty term.

[0119] Temperature mean square error term Wherein, N' represents the number of temperature sampling points, and T pred,i , T target,i respectively represent the predicted temperature and the target temperature at the i-th sampling point.

[0120] Air volume adjustment regularization term Wherein, the real-time air volume distribution value F t at the t-th time step has the unit of m 3 / h. According to the actual situation, it can include the air volume in the low-temperature section (600 - 900 °C) and the air volume in the high-temperature section (900 - 1290 °C). The absolute value of the change in air volume between adjacent time steps |F t -Ft-1 | represents the absolute value of the air volume change at adjacent time steps, reflecting the intensity of the control action.

[0121] Carbon monoxide concentration penalty term Where n represents the number of time points, e represents the natural constant, t represents the index variable of the time point, and CO t represents the carbon monoxide concentration value at the t-th time point, and CO max represents the carbon monoxide concentration safety threshold, and k represents the penalty intensity adjustment coefficient.

[0122] Oxygen concentration penalty term Where m represents the number of temperature zones divided in the target area, and O2 i+1 and O2 i represent the oxygen concentration values in the (i + 1)-th and i-th temperature zones respectively,

[0123] T i+1 and T i represent the set temperature values in the (i + 1)-th and i-th temperature zones respectively.

[0124] In the intelligent control system of the kiln furnace atmosphere, a loss function of the LSTM-CNN hybrid neural network model is constructed through the temperature mean square error term, the pressure fluctuation variance term, the air volume adjustment regularization term, the gas consumption penalty term, the carbon monoxide concentration penalty term, and the oxygen concentration penalty term, which integrates multi-dimensional information such as the kiln furnace temperature, pressure, atmosphere, and gas consumption, realizes multi-objective collaborative optimization, avoids overfitting of a single objective, can reduce the differences in temperature, pressure, and atmosphere during the kiln furnace calcination process, improve the calcination quality of products and the overall production efficiency; by creatively adding the carbon monoxide concentration penalty term and the oxygen concentration penalty term to the loss function, the carbon monoxide concentration and oxygen concentration during the product calcination process are inhibited and constrained, which can ensure the stability of the atmosphere control and ensure the product quality; based on the trained hybrid neural network model, an optimal control parameter group is obtained, and the control parameter group at least includes the air volume control parameter of the target area, which can automatically control the kiln furnace atmosphere.

[0125] An embodiment of the present invention further provides an intelligent control device for a kiln furnace atmosphere, including: a memory storing executable instructions; a controller for executing the executable instructions and implementing the described intelligent control method for the kiln furnace atmosphere.

[0126] The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity of description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope recorded in this specification.

[0127] The above-described embodiments merely represent several implementation manners of the present invention. The description thereof is relatively specific and detailed, but it should not be construed as a limitation to the scope of the invention patent. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present invention, several modifications and improvements can still be made, and these all fall within the protection scope of the present invention. Therefore, the protection scope of the present invention patent shall be subject to the appended claims.

Claims

1. A kiln atmosphere intelligent control method, characterized in that: The control method comprises: Obtain temperature information, pressure information, air volume information, gas consumption information, carbon monoxide concentration information, and oxygen concentration information of the target area; According to the temperature information, pressure information, air volume information, gas consumption information, carbon monoxide concentration information and oxygen concentration information, a temperature mean square error term, a pressure fluctuation variance term, an air volume adjustment regularization term, a gas consumption penalty term, a carbon monoxide concentration penalty term and an oxygen concentration penalty term are obtained respectively; Construct an LSTM-CNN hybrid neural network model, and construct the loss function of the LSTM-CNN hybrid neural network model based on the temperature mean square error term, pressure fluctuation variance term, air volume adjustment regularization term, gas consumption penalty term, carbon monoxide concentration penalty term, and oxygen concentration penalty term; Train the LSTM-CNN hybrid neural network model, and obtain the optimal control parameter group according to the trained LSTM-CNN hybrid neural network model; The control parameter group at least includes the air volume control parameter of the target area.

2. The method for intelligently controlling kiln atmosphere according to claim 1, characterized in that: The specific method of obtaining the air volume adjustment regularization term according to the air volume information includes: Obtain the real-time air volume distribution value of the t-th time step and the air volume distribution value of the t-1-th time step, and calculate the absolute value of the air volume change in adjacent time steps; Obtain the preset continuous monitoring time window length, and calculate the cumulative adjusted air volume according to the time window length and the absolute value of the air volume change in adjacent time steps; The air volume adjustment L1 regularization term is obtained according to the cumulative adjusted air volume.

3. A furnace atmosphere intelligent control method as claimed in claim 2, characterized in that: The specific method of obtaining the gas consumption penalty item according to the gas consumption information includes: Get the current gas consumption and the historical optimal gas consumption; Obtain gas consumption penalty items based on current gas consumption and historical optimal gas consumption; Among them, the gas consumption penalty item is used to trigger a penalty when the current gas consumption exceeds the historical optimal gas consumption.

4. A furnace atmosphere intelligent control method as claimed in claim 3, characterized in that: The specific method of obtaining the carbon monoxide concentration penalty item according to the carbon monoxide concentration information includes: Obtain carbon monoxide concentration values ​​at multiple time points; According to the cumulative average value between the plurality of carbon monoxide concentration values ​​and the preset carbon monoxide concentration safety threshold value; Get the carbon monoxide concentration penalty term based on the cumulative mean.

5. The method for intelligently controlling kiln atmosphere according to claim 4, characterized in that: The specific method of obtaining the oxygen concentration penalty item according to the oxygen concentration information includes: Obtain oxygen concentration values ​​and set temperature values ​​in multiple temperature zones; Calculate the change rate of oxygen concentration at adjacent temperatures according to the oxygen concentration value and the set temperature value; Sum the squares of the absolute values ​​of the change rates to obtain the oxygen concentration penalty term; The change rate represents the difference in oxygen concentration caused by a unit temperature change.

6. A furnace atmosphere intelligent control method as claimed in claim 5, characterized in that: loss functionL=α·MSE T +β·Var p +γ|ΔF|+λ·E gas +η·C CO +μ·C O2 ; Among them, MSE T represents the temperature mean square error term, Var p represents the pressure fluctuation variance term, |ΔF| represents the air volume adjustment regularization term, and E gas represents the gas consumption penalty term, C CO represents the carbon monoxide concentration penalty term, C O2 Oxygen concentration penalty term, α, β, γ, λ, η, μ respectively represent the weight coefficients of temperature mean square error term, pressure fluctuation variance term, air volume adjustment regularization term, gas consumption penalty term, carbon monoxide concentration penalty term and oxygen concentration penalty term.

7. A furnace atmosphere intelligent control method as claimed in claim 6, characterized in that: Carbon monoxide concentration penalty Oxygen concentration penalty Where n represents the number of time points, e represents the natural constant, t represents the index variable of the time point, CO t Indicates the carbon monoxide concentration at the tth time point, CO max represents the safety threshold of carbon monoxide concentration, k represents the penalty intensity adjustment coefficient, m represents the number of temperature zones divided into the target area, O2 i+1 、O2 i Respectively represent the oxygen concentration values ​​of the i+1th and ith temperature zones, T i+1 , T i They represent the set temperature values ​​of the i+1th and ith temperature zones respectively.

8. A kiln atmosphere intelligent control system, used to implement the kiln atmosphere intelligent control method according to any one of claims 1 to 7, characterized in that: The control system comprises: An information acquisition module is used to acquire temperature information, pressure information, air volume information, gas consumption information, carbon monoxide concentration information, and oxygen concentration information of a target area; A loss function item acquisition module is used to obtain a temperature mean square error item, a pressure fluctuation variance item, an air volume adjustment regularization item, a gas consumption penalty item, a carbon monoxide concentration penalty item, and an oxygen concentration penalty item according to temperature information, pressure information, air volume information, gas consumption information, carbon monoxide concentration information, and oxygen concentration information, respectively; A construction module is used to construct an LSTM-CNN hybrid neural network model, and a loss function of the LSTM-CNN hybrid neural network model is constructed according to the temperature mean square error term, the pressure fluctuation variance term, the air volume adjustment regularization term, the gas consumption penalty term, the carbon monoxide concentration penalty term, and the oxygen concentration penalty term; A control parameter acquisition module is used to train the LSTM-CNN hybrid neural network model and obtain an optimal control parameter group according to the trained LSTM-CNN hybrid neural network model; The control parameter group at least includes the air volume control parameter of the target area.

9. The intelligent kiln atmosphere control system according to claim 8, characterized in that: The building block is based on the formula L = α·MSE T +β·Var p +γ|ΔF|+λ·E gas +η·C CO +μ·C O2 Construct loss function; Among them, MSE T represents the temperature mean square error term, Var p represents the pressure fluctuation variance term, |ΔF| represents the air volume adjustment regularization term, and E gas represents the gas consumption penalty term, C CO represents the carbon monoxide concentration penalty term, C O2 Oxygen concentration penalty term, α, β, γ, λ, η, μ respectively represent the weight coefficients of temperature mean square error term, pressure fluctuation variance term, air volume adjustment regularization term, gas consumption penalty term, carbon monoxide concentration penalty term and oxygen concentration penalty term.

10. A kiln atmosphere intelligent control device, characterized in that: The kiln atmosphere intelligent control equipment includes: A memory storing executable instructions; A controller is used to execute the executable instructions and implement the kiln atmosphere intelligent control method as described in any one of claims 1-7.

Images